]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/panic.c
Merge tag 'for-4.15-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
[mirror_ubuntu-bionic-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/sched/debug.h>
13 #include <linux/interrupt.h>
14 #include <linux/kmsg_dump.h>
15 #include <linux/kallsyms.h>
16 #include <linux/notifier.h>
17 #include <linux/module.h>
18 #include <linux/random.h>
19 #include <linux/ftrace.h>
20 #include <linux/reboot.h>
21 #include <linux/delay.h>
22 #include <linux/kexec.h>
23 #include <linux/sched.h>
24 #include <linux/sysrq.h>
25 #include <linux/init.h>
26 #include <linux/nmi.h>
27 #include <linux/console.h>
28 #include <linux/bug.h>
29 #include <linux/ratelimit.h>
30 #include <linux/debugfs.h>
31 #include <asm/sections.h>
32
33 #define PANIC_TIMER_STEP 100
34 #define PANIC_BLINK_SPD 18
35
36 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
37 static unsigned long tainted_mask;
38 static int pause_on_oops;
39 static int pause_on_oops_flag;
40 static DEFINE_SPINLOCK(pause_on_oops_lock);
41 bool crash_kexec_post_notifiers;
42 int panic_on_warn __read_mostly;
43
44 int panic_timeout = CONFIG_PANIC_TIMEOUT;
45 EXPORT_SYMBOL_GPL(panic_timeout);
46
47 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
48
49 EXPORT_SYMBOL(panic_notifier_list);
50
51 static long no_blink(int state)
52 {
53 return 0;
54 }
55
56 /* Returns how long it waited in ms */
57 long (*panic_blink)(int state);
58 EXPORT_SYMBOL(panic_blink);
59
60 /*
61 * Stop ourself in panic -- architecture code may override this
62 */
63 void __weak panic_smp_self_stop(void)
64 {
65 while (1)
66 cpu_relax();
67 }
68
69 /*
70 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
71 * may override this to prepare for crash dumping, e.g. save regs info.
72 */
73 void __weak nmi_panic_self_stop(struct pt_regs *regs)
74 {
75 panic_smp_self_stop();
76 }
77
78 /*
79 * Stop other CPUs in panic. Architecture dependent code may override this
80 * with more suitable version. For example, if the architecture supports
81 * crash dump, it should save registers of each stopped CPU and disable
82 * per-CPU features such as virtualization extensions.
83 */
84 void __weak crash_smp_send_stop(void)
85 {
86 static int cpus_stopped;
87
88 /*
89 * This function can be called twice in panic path, but obviously
90 * we execute this only once.
91 */
92 if (cpus_stopped)
93 return;
94
95 /*
96 * Note smp_send_stop is the usual smp shutdown function, which
97 * unfortunately means it may not be hardened to work in a panic
98 * situation.
99 */
100 smp_send_stop();
101 cpus_stopped = 1;
102 }
103
104 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
105
106 /*
107 * A variant of panic() called from NMI context. We return if we've already
108 * panicked on this CPU. If another CPU already panicked, loop in
109 * nmi_panic_self_stop() which can provide architecture dependent code such
110 * as saving register state for crash dump.
111 */
112 void nmi_panic(struct pt_regs *regs, const char *msg)
113 {
114 int old_cpu, cpu;
115
116 cpu = raw_smp_processor_id();
117 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
118
119 if (old_cpu == PANIC_CPU_INVALID)
120 panic("%s", msg);
121 else if (old_cpu != cpu)
122 nmi_panic_self_stop(regs);
123 }
124 EXPORT_SYMBOL(nmi_panic);
125
126 /**
127 * panic - halt the system
128 * @fmt: The text string to print
129 *
130 * Display a message, then perform cleanups.
131 *
132 * This function never returns.
133 */
134 void panic(const char *fmt, ...)
135 {
136 static char buf[1024];
137 va_list args;
138 long i, i_next = 0;
139 int state = 0;
140 int old_cpu, this_cpu;
141 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
142
143 /*
144 * Disable local interrupts. This will prevent panic_smp_self_stop
145 * from deadlocking the first cpu that invokes the panic, since
146 * there is nothing to prevent an interrupt handler (that runs
147 * after setting panic_cpu) from invoking panic() again.
148 */
149 local_irq_disable();
150
151 /*
152 * It's possible to come here directly from a panic-assertion and
153 * not have preempt disabled. Some functions called from here want
154 * preempt to be disabled. No point enabling it later though...
155 *
156 * Only one CPU is allowed to execute the panic code from here. For
157 * multiple parallel invocations of panic, all other CPUs either
158 * stop themself or will wait until they are stopped by the 1st CPU
159 * with smp_send_stop().
160 *
161 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
162 * comes here, so go ahead.
163 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
164 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
165 */
166 this_cpu = raw_smp_processor_id();
167 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
168
169 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
170 panic_smp_self_stop();
171
172 console_verbose();
173 bust_spinlocks(1);
174 va_start(args, fmt);
175 vsnprintf(buf, sizeof(buf), fmt, args);
176 va_end(args);
177 pr_emerg("Kernel panic - not syncing: %s\n", buf);
178 #ifdef CONFIG_DEBUG_BUGVERBOSE
179 /*
180 * Avoid nested stack-dumping if a panic occurs during oops processing
181 */
182 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
183 dump_stack();
184 #endif
185
186 /*
187 * If we have crashed and we have a crash kernel loaded let it handle
188 * everything else.
189 * If we want to run this after calling panic_notifiers, pass
190 * the "crash_kexec_post_notifiers" option to the kernel.
191 *
192 * Bypass the panic_cpu check and call __crash_kexec directly.
193 */
194 if (!_crash_kexec_post_notifiers) {
195 printk_safe_flush_on_panic();
196 __crash_kexec(NULL);
197
198 /*
199 * Note smp_send_stop is the usual smp shutdown function, which
200 * unfortunately means it may not be hardened to work in a
201 * panic situation.
202 */
203 smp_send_stop();
204 } else {
205 /*
206 * If we want to do crash dump after notifier calls and
207 * kmsg_dump, we will need architecture dependent extra
208 * works in addition to stopping other CPUs.
209 */
210 crash_smp_send_stop();
211 }
212
213 /*
214 * Run any panic handlers, including those that might need to
215 * add information to the kmsg dump output.
216 */
217 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
218
219 /* Call flush even twice. It tries harder with a single online CPU */
220 printk_safe_flush_on_panic();
221 kmsg_dump(KMSG_DUMP_PANIC);
222
223 /*
224 * If you doubt kdump always works fine in any situation,
225 * "crash_kexec_post_notifiers" offers you a chance to run
226 * panic_notifiers and dumping kmsg before kdump.
227 * Note: since some panic_notifiers can make crashed kernel
228 * more unstable, it can increase risks of the kdump failure too.
229 *
230 * Bypass the panic_cpu check and call __crash_kexec directly.
231 */
232 if (_crash_kexec_post_notifiers)
233 __crash_kexec(NULL);
234
235 bust_spinlocks(0);
236
237 /*
238 * We may have ended up stopping the CPU holding the lock (in
239 * smp_send_stop()) while still having some valuable data in the console
240 * buffer. Try to acquire the lock then release it regardless of the
241 * result. The release will also print the buffers out. Locks debug
242 * should be disabled to avoid reporting bad unlock balance when
243 * panic() is not being callled from OOPS.
244 */
245 debug_locks_off();
246 console_flush_on_panic();
247
248 if (!panic_blink)
249 panic_blink = no_blink;
250
251 if (panic_timeout > 0) {
252 /*
253 * Delay timeout seconds before rebooting the machine.
254 * We can't use the "normal" timers since we just panicked.
255 */
256 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
257
258 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
259 touch_nmi_watchdog();
260 if (i >= i_next) {
261 i += panic_blink(state ^= 1);
262 i_next = i + 3600 / PANIC_BLINK_SPD;
263 }
264 mdelay(PANIC_TIMER_STEP);
265 }
266 }
267 if (panic_timeout != 0) {
268 /*
269 * This will not be a clean reboot, with everything
270 * shutting down. But if there is a chance of
271 * rebooting the system it will be rebooted.
272 */
273 emergency_restart();
274 }
275 #ifdef __sparc__
276 {
277 extern int stop_a_enabled;
278 /* Make sure the user can actually press Stop-A (L1-A) */
279 stop_a_enabled = 1;
280 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
281 "twice on console to return to the boot prom\n");
282 }
283 #endif
284 #if defined(CONFIG_S390)
285 {
286 unsigned long caller;
287
288 caller = (unsigned long)__builtin_return_address(0);
289 disabled_wait(caller);
290 }
291 #endif
292 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
293 local_irq_enable();
294 for (i = 0; ; i += PANIC_TIMER_STEP) {
295 touch_softlockup_watchdog();
296 if (i >= i_next) {
297 i += panic_blink(state ^= 1);
298 i_next = i + 3600 / PANIC_BLINK_SPD;
299 }
300 mdelay(PANIC_TIMER_STEP);
301 }
302 }
303
304 EXPORT_SYMBOL(panic);
305
306 /*
307 * TAINT_FORCED_RMMOD could be a per-module flag but the module
308 * is being removed anyway.
309 */
310 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
311 { 'P', 'G', true }, /* TAINT_PROPRIETARY_MODULE */
312 { 'F', ' ', true }, /* TAINT_FORCED_MODULE */
313 { 'S', ' ', false }, /* TAINT_CPU_OUT_OF_SPEC */
314 { 'R', ' ', false }, /* TAINT_FORCED_RMMOD */
315 { 'M', ' ', false }, /* TAINT_MACHINE_CHECK */
316 { 'B', ' ', false }, /* TAINT_BAD_PAGE */
317 { 'U', ' ', false }, /* TAINT_USER */
318 { 'D', ' ', false }, /* TAINT_DIE */
319 { 'A', ' ', false }, /* TAINT_OVERRIDDEN_ACPI_TABLE */
320 { 'W', ' ', false }, /* TAINT_WARN */
321 { 'C', ' ', true }, /* TAINT_CRAP */
322 { 'I', ' ', false }, /* TAINT_FIRMWARE_WORKAROUND */
323 { 'O', ' ', true }, /* TAINT_OOT_MODULE */
324 { 'E', ' ', true }, /* TAINT_UNSIGNED_MODULE */
325 { 'L', ' ', false }, /* TAINT_SOFTLOCKUP */
326 { 'K', ' ', true }, /* TAINT_LIVEPATCH */
327 { 'X', ' ', true }, /* TAINT_AUX */
328 };
329
330 /**
331 * print_tainted - return a string to represent the kernel taint state.
332 *
333 * 'P' - Proprietary module has been loaded.
334 * 'F' - Module has been forcibly loaded.
335 * 'S' - SMP with CPUs not designed for SMP.
336 * 'R' - User forced a module unload.
337 * 'M' - System experienced a machine check exception.
338 * 'B' - System has hit bad_page.
339 * 'U' - Userspace-defined naughtiness.
340 * 'D' - Kernel has oopsed before
341 * 'A' - ACPI table overridden.
342 * 'W' - Taint on warning.
343 * 'C' - modules from drivers/staging are loaded.
344 * 'I' - Working around severe firmware bug.
345 * 'O' - Out-of-tree module has been loaded.
346 * 'E' - Unsigned module has been loaded.
347 * 'L' - A soft lockup has previously occurred.
348 * 'K' - Kernel has been live patched.
349 * 'X' - Auxiliary taint, for distros' use.
350 *
351 * The string is overwritten by the next call to print_tainted().
352 */
353 const char *print_tainted(void)
354 {
355 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
356
357 if (tainted_mask) {
358 char *s;
359 int i;
360
361 s = buf + sprintf(buf, "Tainted: ");
362 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
363 const struct taint_flag *t = &taint_flags[i];
364 *s++ = test_bit(i, &tainted_mask) ?
365 t->c_true : t->c_false;
366 }
367 *s = 0;
368 } else
369 snprintf(buf, sizeof(buf), "Not tainted");
370
371 return buf;
372 }
373
374 int test_taint(unsigned flag)
375 {
376 return test_bit(flag, &tainted_mask);
377 }
378 EXPORT_SYMBOL(test_taint);
379
380 unsigned long get_taint(void)
381 {
382 return tainted_mask;
383 }
384
385 /**
386 * add_taint: add a taint flag if not already set.
387 * @flag: one of the TAINT_* constants.
388 * @lockdep_ok: whether lock debugging is still OK.
389 *
390 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
391 * some notewortht-but-not-corrupting cases, it can be set to true.
392 */
393 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
394 {
395 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
396 pr_warn("Disabling lock debugging due to kernel taint\n");
397
398 set_bit(flag, &tainted_mask);
399 }
400 EXPORT_SYMBOL(add_taint);
401
402 static void spin_msec(int msecs)
403 {
404 int i;
405
406 for (i = 0; i < msecs; i++) {
407 touch_nmi_watchdog();
408 mdelay(1);
409 }
410 }
411
412 /*
413 * It just happens that oops_enter() and oops_exit() are identically
414 * implemented...
415 */
416 static void do_oops_enter_exit(void)
417 {
418 unsigned long flags;
419 static int spin_counter;
420
421 if (!pause_on_oops)
422 return;
423
424 spin_lock_irqsave(&pause_on_oops_lock, flags);
425 if (pause_on_oops_flag == 0) {
426 /* This CPU may now print the oops message */
427 pause_on_oops_flag = 1;
428 } else {
429 /* We need to stall this CPU */
430 if (!spin_counter) {
431 /* This CPU gets to do the counting */
432 spin_counter = pause_on_oops;
433 do {
434 spin_unlock(&pause_on_oops_lock);
435 spin_msec(MSEC_PER_SEC);
436 spin_lock(&pause_on_oops_lock);
437 } while (--spin_counter);
438 pause_on_oops_flag = 0;
439 } else {
440 /* This CPU waits for a different one */
441 while (spin_counter) {
442 spin_unlock(&pause_on_oops_lock);
443 spin_msec(1);
444 spin_lock(&pause_on_oops_lock);
445 }
446 }
447 }
448 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
449 }
450
451 /*
452 * Return true if the calling CPU is allowed to print oops-related info.
453 * This is a bit racy..
454 */
455 int oops_may_print(void)
456 {
457 return pause_on_oops_flag == 0;
458 }
459
460 /*
461 * Called when the architecture enters its oops handler, before it prints
462 * anything. If this is the first CPU to oops, and it's oopsing the first
463 * time then let it proceed.
464 *
465 * This is all enabled by the pause_on_oops kernel boot option. We do all
466 * this to ensure that oopses don't scroll off the screen. It has the
467 * side-effect of preventing later-oopsing CPUs from mucking up the display,
468 * too.
469 *
470 * It turns out that the CPU which is allowed to print ends up pausing for
471 * the right duration, whereas all the other CPUs pause for twice as long:
472 * once in oops_enter(), once in oops_exit().
473 */
474 void oops_enter(void)
475 {
476 tracing_off();
477 /* can't trust the integrity of the kernel anymore: */
478 debug_locks_off();
479 do_oops_enter_exit();
480 }
481
482 /*
483 * 64-bit random ID for oopses:
484 */
485 static u64 oops_id;
486
487 static int init_oops_id(void)
488 {
489 if (!oops_id)
490 get_random_bytes(&oops_id, sizeof(oops_id));
491 else
492 oops_id++;
493
494 return 0;
495 }
496 late_initcall(init_oops_id);
497
498 void print_oops_end_marker(void)
499 {
500 init_oops_id();
501 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
502 }
503
504 /*
505 * Called when the architecture exits its oops handler, after printing
506 * everything.
507 */
508 void oops_exit(void)
509 {
510 do_oops_enter_exit();
511 print_oops_end_marker();
512 kmsg_dump(KMSG_DUMP_OOPS);
513 }
514
515 struct warn_args {
516 const char *fmt;
517 va_list args;
518 };
519
520 void __warn(const char *file, int line, void *caller, unsigned taint,
521 struct pt_regs *regs, struct warn_args *args)
522 {
523 disable_trace_on_warning();
524
525 if (args)
526 pr_warn(CUT_HERE);
527
528 if (file)
529 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
530 raw_smp_processor_id(), current->pid, file, line,
531 caller);
532 else
533 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
534 raw_smp_processor_id(), current->pid, caller);
535
536 if (args)
537 vprintk(args->fmt, args->args);
538
539 if (panic_on_warn) {
540 /*
541 * This thread may hit another WARN() in the panic path.
542 * Resetting this prevents additional WARN() from panicking the
543 * system on this thread. Other threads are blocked by the
544 * panic_mutex in panic().
545 */
546 panic_on_warn = 0;
547 panic("panic_on_warn set ...\n");
548 }
549
550 print_modules();
551
552 if (regs)
553 show_regs(regs);
554 else
555 dump_stack();
556
557 print_oops_end_marker();
558
559 /* Just a warning, don't kill lockdep. */
560 add_taint(taint, LOCKDEP_STILL_OK);
561 }
562
563 #ifdef WANT_WARN_ON_SLOWPATH
564 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
565 {
566 struct warn_args args;
567
568 args.fmt = fmt;
569 va_start(args.args, fmt);
570 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
571 &args);
572 va_end(args.args);
573 }
574 EXPORT_SYMBOL(warn_slowpath_fmt);
575
576 void warn_slowpath_fmt_taint(const char *file, int line,
577 unsigned taint, const char *fmt, ...)
578 {
579 struct warn_args args;
580
581 args.fmt = fmt;
582 va_start(args.args, fmt);
583 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
584 va_end(args.args);
585 }
586 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
587
588 void warn_slowpath_null(const char *file, int line)
589 {
590 pr_warn(CUT_HERE);
591 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
592 }
593 EXPORT_SYMBOL(warn_slowpath_null);
594 #else
595 void __warn_printk(const char *fmt, ...)
596 {
597 va_list args;
598
599 pr_warn(CUT_HERE);
600
601 va_start(args, fmt);
602 vprintk(fmt, args);
603 va_end(args);
604 }
605 EXPORT_SYMBOL(__warn_printk);
606 #endif
607
608 #ifdef CONFIG_BUG
609
610 /* Support resetting WARN*_ONCE state */
611
612 static int clear_warn_once_set(void *data, u64 val)
613 {
614 generic_bug_clear_once();
615 memset(__start_once, 0, __end_once - __start_once);
616 return 0;
617 }
618
619 DEFINE_SIMPLE_ATTRIBUTE(clear_warn_once_fops,
620 NULL,
621 clear_warn_once_set,
622 "%lld\n");
623
624 static __init int register_warn_debugfs(void)
625 {
626 /* Don't care about failure */
627 debugfs_create_file("clear_warn_once", 0200, NULL,
628 NULL, &clear_warn_once_fops);
629 return 0;
630 }
631
632 device_initcall(register_warn_debugfs);
633 #endif
634
635 #ifdef CONFIG_CC_STACKPROTECTOR
636
637 /*
638 * Called when gcc's -fstack-protector feature is used, and
639 * gcc detects corruption of the on-stack canary value
640 */
641 __visible void __stack_chk_fail(void)
642 {
643 panic("stack-protector: Kernel stack is corrupted in: %p\n",
644 __builtin_return_address(0));
645 }
646 EXPORT_SYMBOL(__stack_chk_fail);
647
648 #endif
649
650 #ifdef CONFIG_ARCH_HAS_REFCOUNT
651 void refcount_error_report(struct pt_regs *regs, const char *err)
652 {
653 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
654 err, (void *)instruction_pointer(regs),
655 current->comm, task_pid_nr(current),
656 from_kuid_munged(&init_user_ns, current_uid()),
657 from_kuid_munged(&init_user_ns, current_euid()));
658 }
659 #endif
660
661 core_param(panic, panic_timeout, int, 0644);
662 core_param(pause_on_oops, pause_on_oops, int, 0644);
663 core_param(panic_on_warn, panic_on_warn, int, 0644);
664 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
665
666 static int __init oops_setup(char *s)
667 {
668 if (!s)
669 return -EINVAL;
670 if (!strcmp(s, "panic"))
671 panic_on_oops = 1;
672 return 0;
673 }
674 early_param("oops", oops_setup);