xref: /linux/kernel/panic.c (revision f79e4d5f92a129a1159c973735007d4ddc8541f3)
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 	IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
39 static int pause_on_oops;
40 static int pause_on_oops_flag;
41 static DEFINE_SPINLOCK(pause_on_oops_lock);
42 bool crash_kexec_post_notifiers;
43 int panic_on_warn __read_mostly;
44 
45 int panic_timeout = CONFIG_PANIC_TIMEOUT;
46 EXPORT_SYMBOL_GPL(panic_timeout);
47 
48 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
49 
50 EXPORT_SYMBOL(panic_notifier_list);
51 
52 static long no_blink(int state)
53 {
54 	return 0;
55 }
56 
57 /* Returns how long it waited in ms */
58 long (*panic_blink)(int state);
59 EXPORT_SYMBOL(panic_blink);
60 
61 /*
62  * Stop ourself in panic -- architecture code may override this
63  */
64 void __weak panic_smp_self_stop(void)
65 {
66 	while (1)
67 		cpu_relax();
68 }
69 
70 /*
71  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
72  * may override this to prepare for crash dumping, e.g. save regs info.
73  */
74 void __weak nmi_panic_self_stop(struct pt_regs *regs)
75 {
76 	panic_smp_self_stop();
77 }
78 
79 /*
80  * Stop other CPUs in panic.  Architecture dependent code may override this
81  * with more suitable version.  For example, if the architecture supports
82  * crash dump, it should save registers of each stopped CPU and disable
83  * per-CPU features such as virtualization extensions.
84  */
85 void __weak crash_smp_send_stop(void)
86 {
87 	static int cpus_stopped;
88 
89 	/*
90 	 * This function can be called twice in panic path, but obviously
91 	 * we execute this only once.
92 	 */
93 	if (cpus_stopped)
94 		return;
95 
96 	/*
97 	 * Note smp_send_stop is the usual smp shutdown function, which
98 	 * unfortunately means it may not be hardened to work in a panic
99 	 * situation.
100 	 */
101 	smp_send_stop();
102 	cpus_stopped = 1;
103 }
104 
105 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
106 
107 /*
108  * A variant of panic() called from NMI context. We return if we've already
109  * panicked on this CPU. If another CPU already panicked, loop in
110  * nmi_panic_self_stop() which can provide architecture dependent code such
111  * as saving register state for crash dump.
112  */
113 void nmi_panic(struct pt_regs *regs, const char *msg)
114 {
115 	int old_cpu, cpu;
116 
117 	cpu = raw_smp_processor_id();
118 	old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
119 
120 	if (old_cpu == PANIC_CPU_INVALID)
121 		panic("%s", msg);
122 	else if (old_cpu != cpu)
123 		nmi_panic_self_stop(regs);
124 }
125 EXPORT_SYMBOL(nmi_panic);
126 
127 /**
128  *	panic - halt the system
129  *	@fmt: The text string to print
130  *
131  *	Display a message, then perform cleanups.
132  *
133  *	This function never returns.
134  */
135 void panic(const char *fmt, ...)
136 {
137 	static char buf[1024];
138 	va_list args;
139 	long i, i_next = 0;
140 	int state = 0;
141 	int old_cpu, this_cpu;
142 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
143 
144 	/*
145 	 * Disable local interrupts. This will prevent panic_smp_self_stop
146 	 * from deadlocking the first cpu that invokes the panic, since
147 	 * there is nothing to prevent an interrupt handler (that runs
148 	 * after setting panic_cpu) from invoking panic() again.
149 	 */
150 	local_irq_disable();
151 
152 	/*
153 	 * It's possible to come here directly from a panic-assertion and
154 	 * not have preempt disabled. Some functions called from here want
155 	 * preempt to be disabled. No point enabling it later though...
156 	 *
157 	 * Only one CPU is allowed to execute the panic code from here. For
158 	 * multiple parallel invocations of panic, all other CPUs either
159 	 * stop themself or will wait until they are stopped by the 1st CPU
160 	 * with smp_send_stop().
161 	 *
162 	 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
163 	 * comes here, so go ahead.
164 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
165 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
166 	 */
167 	this_cpu = raw_smp_processor_id();
168 	old_cpu  = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
169 
170 	if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
171 		panic_smp_self_stop();
172 
173 	console_verbose();
174 	bust_spinlocks(1);
175 	va_start(args, fmt);
176 	vsnprintf(buf, sizeof(buf), fmt, args);
177 	va_end(args);
178 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
179 #ifdef CONFIG_DEBUG_BUGVERBOSE
180 	/*
181 	 * Avoid nested stack-dumping if a panic occurs during oops processing
182 	 */
183 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
184 		dump_stack();
185 #endif
186 
187 	/*
188 	 * If we have crashed and we have a crash kernel loaded let it handle
189 	 * everything else.
190 	 * If we want to run this after calling panic_notifiers, pass
191 	 * the "crash_kexec_post_notifiers" option to the kernel.
192 	 *
193 	 * Bypass the panic_cpu check and call __crash_kexec directly.
194 	 */
195 	if (!_crash_kexec_post_notifiers) {
196 		printk_safe_flush_on_panic();
197 		__crash_kexec(NULL);
198 
199 		/*
200 		 * Note smp_send_stop is the usual smp shutdown function, which
201 		 * unfortunately means it may not be hardened to work in a
202 		 * panic situation.
203 		 */
204 		smp_send_stop();
205 	} else {
206 		/*
207 		 * If we want to do crash dump after notifier calls and
208 		 * kmsg_dump, we will need architecture dependent extra
209 		 * works in addition to stopping other CPUs.
210 		 */
211 		crash_smp_send_stop();
212 	}
213 
214 	/*
215 	 * Run any panic handlers, including those that might need to
216 	 * add information to the kmsg dump output.
217 	 */
218 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
219 
220 	/* Call flush even twice. It tries harder with a single online CPU */
221 	printk_safe_flush_on_panic();
222 	kmsg_dump(KMSG_DUMP_PANIC);
223 
224 	/*
225 	 * If you doubt kdump always works fine in any situation,
226 	 * "crash_kexec_post_notifiers" offers you a chance to run
227 	 * panic_notifiers and dumping kmsg before kdump.
228 	 * Note: since some panic_notifiers can make crashed kernel
229 	 * more unstable, it can increase risks of the kdump failure too.
230 	 *
231 	 * Bypass the panic_cpu check and call __crash_kexec directly.
232 	 */
233 	if (_crash_kexec_post_notifiers)
234 		__crash_kexec(NULL);
235 
236 	bust_spinlocks(0);
237 
238 	/*
239 	 * We may have ended up stopping the CPU holding the lock (in
240 	 * smp_send_stop()) while still having some valuable data in the console
241 	 * buffer.  Try to acquire the lock then release it regardless of the
242 	 * result.  The release will also print the buffers out.  Locks debug
243 	 * should be disabled to avoid reporting bad unlock balance when
244 	 * panic() is not being callled from OOPS.
245 	 */
246 	debug_locks_off();
247 	console_flush_on_panic();
248 
249 	if (!panic_blink)
250 		panic_blink = no_blink;
251 
252 	if (panic_timeout > 0) {
253 		/*
254 		 * Delay timeout seconds before rebooting the machine.
255 		 * We can't use the "normal" timers since we just panicked.
256 		 */
257 		pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
258 
259 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
260 			touch_nmi_watchdog();
261 			if (i >= i_next) {
262 				i += panic_blink(state ^= 1);
263 				i_next = i + 3600 / PANIC_BLINK_SPD;
264 			}
265 			mdelay(PANIC_TIMER_STEP);
266 		}
267 	}
268 	if (panic_timeout != 0) {
269 		/*
270 		 * This will not be a clean reboot, with everything
271 		 * shutting down.  But if there is a chance of
272 		 * rebooting the system it will be rebooted.
273 		 */
274 		emergency_restart();
275 	}
276 #ifdef __sparc__
277 	{
278 		extern int stop_a_enabled;
279 		/* Make sure the user can actually press Stop-A (L1-A) */
280 		stop_a_enabled = 1;
281 		pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
282 			 "twice on console to return to the boot prom\n");
283 	}
284 #endif
285 #if defined(CONFIG_S390)
286 	{
287 		unsigned long caller;
288 
289 		caller = (unsigned long)__builtin_return_address(0);
290 		disabled_wait(caller);
291 	}
292 #endif
293 	pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
294 	local_irq_enable();
295 	for (i = 0; ; i += PANIC_TIMER_STEP) {
296 		touch_softlockup_watchdog();
297 		if (i >= i_next) {
298 			i += panic_blink(state ^= 1);
299 			i_next = i + 3600 / PANIC_BLINK_SPD;
300 		}
301 		mdelay(PANIC_TIMER_STEP);
302 	}
303 }
304 
305 EXPORT_SYMBOL(panic);
306 
307 /*
308  * TAINT_FORCED_RMMOD could be a per-module flag but the module
309  * is being removed anyway.
310  */
311 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
312 	[ TAINT_PROPRIETARY_MODULE ]	= { 'P', 'G', true },
313 	[ TAINT_FORCED_MODULE ]		= { 'F', ' ', true },
314 	[ TAINT_CPU_OUT_OF_SPEC ]	= { 'S', ' ', false },
315 	[ TAINT_FORCED_RMMOD ]		= { 'R', ' ', false },
316 	[ TAINT_MACHINE_CHECK ]		= { 'M', ' ', false },
317 	[ TAINT_BAD_PAGE ]		= { 'B', ' ', false },
318 	[ TAINT_USER ]			= { 'U', ' ', false },
319 	[ TAINT_DIE ]			= { 'D', ' ', false },
320 	[ TAINT_OVERRIDDEN_ACPI_TABLE ]	= { 'A', ' ', false },
321 	[ TAINT_WARN ]			= { 'W', ' ', false },
322 	[ TAINT_CRAP ]			= { 'C', ' ', true },
323 	[ TAINT_FIRMWARE_WORKAROUND ]	= { 'I', ' ', false },
324 	[ TAINT_OOT_MODULE ]		= { 'O', ' ', true },
325 	[ TAINT_UNSIGNED_MODULE ]	= { 'E', ' ', true },
326 	[ TAINT_SOFTLOCKUP ]		= { 'L', ' ', false },
327 	[ TAINT_LIVEPATCH ]		= { 'K', ' ', true },
328 	[ TAINT_AUX ]			= { 'X', ' ', true },
329 	[ TAINT_RANDSTRUCT ]		= { 'T', ' ', true },
330 };
331 
332 /**
333  * print_tainted - return a string to represent the kernel taint state.
334  *
335  * For individual taint flag meanings, see Documentation/sysctl/kernel.txt
336  *
337  * The string is overwritten by the next call to print_tainted(),
338  * but is always NULL terminated.
339  */
340 const char *print_tainted(void)
341 {
342 	static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
343 
344 	BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
345 
346 	if (tainted_mask) {
347 		char *s;
348 		int i;
349 
350 		s = buf + sprintf(buf, "Tainted: ");
351 		for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
352 			const struct taint_flag *t = &taint_flags[i];
353 			*s++ = test_bit(i, &tainted_mask) ?
354 					t->c_true : t->c_false;
355 		}
356 		*s = 0;
357 	} else
358 		snprintf(buf, sizeof(buf), "Not tainted");
359 
360 	return buf;
361 }
362 
363 int test_taint(unsigned flag)
364 {
365 	return test_bit(flag, &tainted_mask);
366 }
367 EXPORT_SYMBOL(test_taint);
368 
369 unsigned long get_taint(void)
370 {
371 	return tainted_mask;
372 }
373 
374 /**
375  * add_taint: add a taint flag if not already set.
376  * @flag: one of the TAINT_* constants.
377  * @lockdep_ok: whether lock debugging is still OK.
378  *
379  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
380  * some notewortht-but-not-corrupting cases, it can be set to true.
381  */
382 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
383 {
384 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
385 		pr_warn("Disabling lock debugging due to kernel taint\n");
386 
387 	set_bit(flag, &tainted_mask);
388 }
389 EXPORT_SYMBOL(add_taint);
390 
391 static void spin_msec(int msecs)
392 {
393 	int i;
394 
395 	for (i = 0; i < msecs; i++) {
396 		touch_nmi_watchdog();
397 		mdelay(1);
398 	}
399 }
400 
401 /*
402  * It just happens that oops_enter() and oops_exit() are identically
403  * implemented...
404  */
405 static void do_oops_enter_exit(void)
406 {
407 	unsigned long flags;
408 	static int spin_counter;
409 
410 	if (!pause_on_oops)
411 		return;
412 
413 	spin_lock_irqsave(&pause_on_oops_lock, flags);
414 	if (pause_on_oops_flag == 0) {
415 		/* This CPU may now print the oops message */
416 		pause_on_oops_flag = 1;
417 	} else {
418 		/* We need to stall this CPU */
419 		if (!spin_counter) {
420 			/* This CPU gets to do the counting */
421 			spin_counter = pause_on_oops;
422 			do {
423 				spin_unlock(&pause_on_oops_lock);
424 				spin_msec(MSEC_PER_SEC);
425 				spin_lock(&pause_on_oops_lock);
426 			} while (--spin_counter);
427 			pause_on_oops_flag = 0;
428 		} else {
429 			/* This CPU waits for a different one */
430 			while (spin_counter) {
431 				spin_unlock(&pause_on_oops_lock);
432 				spin_msec(1);
433 				spin_lock(&pause_on_oops_lock);
434 			}
435 		}
436 	}
437 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
438 }
439 
440 /*
441  * Return true if the calling CPU is allowed to print oops-related info.
442  * This is a bit racy..
443  */
444 int oops_may_print(void)
445 {
446 	return pause_on_oops_flag == 0;
447 }
448 
449 /*
450  * Called when the architecture enters its oops handler, before it prints
451  * anything.  If this is the first CPU to oops, and it's oopsing the first
452  * time then let it proceed.
453  *
454  * This is all enabled by the pause_on_oops kernel boot option.  We do all
455  * this to ensure that oopses don't scroll off the screen.  It has the
456  * side-effect of preventing later-oopsing CPUs from mucking up the display,
457  * too.
458  *
459  * It turns out that the CPU which is allowed to print ends up pausing for
460  * the right duration, whereas all the other CPUs pause for twice as long:
461  * once in oops_enter(), once in oops_exit().
462  */
463 void oops_enter(void)
464 {
465 	tracing_off();
466 	/* can't trust the integrity of the kernel anymore: */
467 	debug_locks_off();
468 	do_oops_enter_exit();
469 }
470 
471 /*
472  * 64-bit random ID for oopses:
473  */
474 static u64 oops_id;
475 
476 static int init_oops_id(void)
477 {
478 	if (!oops_id)
479 		get_random_bytes(&oops_id, sizeof(oops_id));
480 	else
481 		oops_id++;
482 
483 	return 0;
484 }
485 late_initcall(init_oops_id);
486 
487 void print_oops_end_marker(void)
488 {
489 	init_oops_id();
490 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
491 }
492 
493 /*
494  * Called when the architecture exits its oops handler, after printing
495  * everything.
496  */
497 void oops_exit(void)
498 {
499 	do_oops_enter_exit();
500 	print_oops_end_marker();
501 	kmsg_dump(KMSG_DUMP_OOPS);
502 }
503 
504 struct warn_args {
505 	const char *fmt;
506 	va_list args;
507 };
508 
509 void __warn(const char *file, int line, void *caller, unsigned taint,
510 	    struct pt_regs *regs, struct warn_args *args)
511 {
512 	disable_trace_on_warning();
513 
514 	if (args)
515 		pr_warn(CUT_HERE);
516 
517 	if (file)
518 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
519 			raw_smp_processor_id(), current->pid, file, line,
520 			caller);
521 	else
522 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
523 			raw_smp_processor_id(), current->pid, caller);
524 
525 	if (args)
526 		vprintk(args->fmt, args->args);
527 
528 	if (panic_on_warn) {
529 		/*
530 		 * This thread may hit another WARN() in the panic path.
531 		 * Resetting this prevents additional WARN() from panicking the
532 		 * system on this thread.  Other threads are blocked by the
533 		 * panic_mutex in panic().
534 		 */
535 		panic_on_warn = 0;
536 		panic("panic_on_warn set ...\n");
537 	}
538 
539 	print_modules();
540 
541 	if (regs)
542 		show_regs(regs);
543 	else
544 		dump_stack();
545 
546 	print_irqtrace_events(current);
547 
548 	print_oops_end_marker();
549 
550 	/* Just a warning, don't kill lockdep. */
551 	add_taint(taint, LOCKDEP_STILL_OK);
552 }
553 
554 #ifdef WANT_WARN_ON_SLOWPATH
555 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
556 {
557 	struct warn_args args;
558 
559 	args.fmt = fmt;
560 	va_start(args.args, fmt);
561 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
562 	       &args);
563 	va_end(args.args);
564 }
565 EXPORT_SYMBOL(warn_slowpath_fmt);
566 
567 void warn_slowpath_fmt_taint(const char *file, int line,
568 			     unsigned taint, const char *fmt, ...)
569 {
570 	struct warn_args args;
571 
572 	args.fmt = fmt;
573 	va_start(args.args, fmt);
574 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
575 	va_end(args.args);
576 }
577 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
578 
579 void warn_slowpath_null(const char *file, int line)
580 {
581 	pr_warn(CUT_HERE);
582 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
583 }
584 EXPORT_SYMBOL(warn_slowpath_null);
585 #else
586 void __warn_printk(const char *fmt, ...)
587 {
588 	va_list args;
589 
590 	pr_warn(CUT_HERE);
591 
592 	va_start(args, fmt);
593 	vprintk(fmt, args);
594 	va_end(args);
595 }
596 EXPORT_SYMBOL(__warn_printk);
597 #endif
598 
599 #ifdef CONFIG_BUG
600 
601 /* Support resetting WARN*_ONCE state */
602 
603 static int clear_warn_once_set(void *data, u64 val)
604 {
605 	generic_bug_clear_once();
606 	memset(__start_once, 0, __end_once - __start_once);
607 	return 0;
608 }
609 
610 DEFINE_SIMPLE_ATTRIBUTE(clear_warn_once_fops,
611 			NULL,
612 			clear_warn_once_set,
613 			"%lld\n");
614 
615 static __init int register_warn_debugfs(void)
616 {
617 	/* Don't care about failure */
618 	debugfs_create_file("clear_warn_once", 0200, NULL,
619 			    NULL, &clear_warn_once_fops);
620 	return 0;
621 }
622 
623 device_initcall(register_warn_debugfs);
624 #endif
625 
626 #ifdef CONFIG_STACKPROTECTOR
627 
628 /*
629  * Called when gcc's -fstack-protector feature is used, and
630  * gcc detects corruption of the on-stack canary value
631  */
632 __visible void __stack_chk_fail(void)
633 {
634 	panic("stack-protector: Kernel stack is corrupted in: %pB\n",
635 		__builtin_return_address(0));
636 }
637 EXPORT_SYMBOL(__stack_chk_fail);
638 
639 #endif
640 
641 #ifdef CONFIG_ARCH_HAS_REFCOUNT
642 void refcount_error_report(struct pt_regs *regs, const char *err)
643 {
644 	WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
645 		err, (void *)instruction_pointer(regs),
646 		current->comm, task_pid_nr(current),
647 		from_kuid_munged(&init_user_ns, current_uid()),
648 		from_kuid_munged(&init_user_ns, current_euid()));
649 }
650 #endif
651 
652 core_param(panic, panic_timeout, int, 0644);
653 core_param(pause_on_oops, pause_on_oops, int, 0644);
654 core_param(panic_on_warn, panic_on_warn, int, 0644);
655 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
656 
657 static int __init oops_setup(char *s)
658 {
659 	if (!s)
660 		return -EINVAL;
661 	if (!strcmp(s, "panic"))
662 		panic_on_oops = 1;
663 	return 0;
664 }
665 early_param("oops", oops_setup);
666