xref: /openbmc/linux/kernel/panic.c (revision ff4a7481c3898ffc3cc271d6aca431d190c37247)
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/vt_kern.h>
18 #include <linux/module.h>
19 #include <linux/random.h>
20 #include <linux/ftrace.h>
21 #include <linux/reboot.h>
22 #include <linux/delay.h>
23 #include <linux/kexec.h>
24 #include <linux/sched.h>
25 #include <linux/sysrq.h>
26 #include <linux/init.h>
27 #include <linux/nmi.h>
28 #include <linux/console.h>
29 #include <linux/bug.h>
30 #include <linux/ratelimit.h>
31 #include <linux/debugfs.h>
32 #include <asm/sections.h>
33 
34 #define PANIC_TIMER_STEP 100
35 #define PANIC_BLINK_SPD 18
36 
37 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
38 static unsigned long tainted_mask =
39 	IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
40 static int pause_on_oops;
41 static int pause_on_oops_flag;
42 static DEFINE_SPINLOCK(pause_on_oops_lock);
43 bool crash_kexec_post_notifiers;
44 int panic_on_warn __read_mostly;
45 
46 int panic_timeout = CONFIG_PANIC_TIMEOUT;
47 EXPORT_SYMBOL_GPL(panic_timeout);
48 
49 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
50 
51 EXPORT_SYMBOL(panic_notifier_list);
52 
53 static long no_blink(int state)
54 {
55 	return 0;
56 }
57 
58 /* Returns how long it waited in ms */
59 long (*panic_blink)(int state);
60 EXPORT_SYMBOL(panic_blink);
61 
62 /*
63  * Stop ourself in panic -- architecture code may override this
64  */
65 void __weak panic_smp_self_stop(void)
66 {
67 	while (1)
68 		cpu_relax();
69 }
70 
71 /*
72  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
73  * may override this to prepare for crash dumping, e.g. save regs info.
74  */
75 void __weak nmi_panic_self_stop(struct pt_regs *regs)
76 {
77 	panic_smp_self_stop();
78 }
79 
80 /*
81  * Stop other CPUs in panic.  Architecture dependent code may override this
82  * with more suitable version.  For example, if the architecture supports
83  * crash dump, it should save registers of each stopped CPU and disable
84  * per-CPU features such as virtualization extensions.
85  */
86 void __weak crash_smp_send_stop(void)
87 {
88 	static int cpus_stopped;
89 
90 	/*
91 	 * This function can be called twice in panic path, but obviously
92 	 * we execute this only once.
93 	 */
94 	if (cpus_stopped)
95 		return;
96 
97 	/*
98 	 * Note smp_send_stop is the usual smp shutdown function, which
99 	 * unfortunately means it may not be hardened to work in a panic
100 	 * situation.
101 	 */
102 	smp_send_stop();
103 	cpus_stopped = 1;
104 }
105 
106 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
107 
108 /*
109  * A variant of panic() called from NMI context. We return if we've already
110  * panicked on this CPU. If another CPU already panicked, loop in
111  * nmi_panic_self_stop() which can provide architecture dependent code such
112  * as saving register state for crash dump.
113  */
114 void nmi_panic(struct pt_regs *regs, const char *msg)
115 {
116 	int old_cpu, cpu;
117 
118 	cpu = raw_smp_processor_id();
119 	old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
120 
121 	if (old_cpu == PANIC_CPU_INVALID)
122 		panic("%s", msg);
123 	else if (old_cpu != cpu)
124 		nmi_panic_self_stop(regs);
125 }
126 EXPORT_SYMBOL(nmi_panic);
127 
128 /**
129  *	panic - halt the system
130  *	@fmt: The text string to print
131  *
132  *	Display a message, then perform cleanups.
133  *
134  *	This function never returns.
135  */
136 void panic(const char *fmt, ...)
137 {
138 	static char buf[1024];
139 	va_list args;
140 	long i, i_next = 0, len;
141 	int state = 0;
142 	int old_cpu, this_cpu;
143 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
144 
145 	/*
146 	 * Disable local interrupts. This will prevent panic_smp_self_stop
147 	 * from deadlocking the first cpu that invokes the panic, since
148 	 * there is nothing to prevent an interrupt handler (that runs
149 	 * after setting panic_cpu) from invoking panic() again.
150 	 */
151 	local_irq_disable();
152 
153 	/*
154 	 * It's possible to come here directly from a panic-assertion and
155 	 * not have preempt disabled. Some functions called from here want
156 	 * preempt to be disabled. No point enabling it later though...
157 	 *
158 	 * Only one CPU is allowed to execute the panic code from here. For
159 	 * multiple parallel invocations of panic, all other CPUs either
160 	 * stop themself or will wait until they are stopped by the 1st CPU
161 	 * with smp_send_stop().
162 	 *
163 	 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
164 	 * comes here, so go ahead.
165 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
166 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
167 	 */
168 	this_cpu = raw_smp_processor_id();
169 	old_cpu  = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
170 
171 	if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
172 		panic_smp_self_stop();
173 
174 	console_verbose();
175 	bust_spinlocks(1);
176 	va_start(args, fmt);
177 	len = vscnprintf(buf, sizeof(buf), fmt, args);
178 	va_end(args);
179 
180 	if (len && buf[len - 1] == '\n')
181 		buf[len - 1] = '\0';
182 
183 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
184 #ifdef CONFIG_DEBUG_BUGVERBOSE
185 	/*
186 	 * Avoid nested stack-dumping if a panic occurs during oops processing
187 	 */
188 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
189 		dump_stack();
190 #endif
191 
192 	/*
193 	 * If we have crashed and we have a crash kernel loaded let it handle
194 	 * everything else.
195 	 * If we want to run this after calling panic_notifiers, pass
196 	 * the "crash_kexec_post_notifiers" option to the kernel.
197 	 *
198 	 * Bypass the panic_cpu check and call __crash_kexec directly.
199 	 */
200 	if (!_crash_kexec_post_notifiers) {
201 		printk_safe_flush_on_panic();
202 		__crash_kexec(NULL);
203 
204 		/*
205 		 * Note smp_send_stop is the usual smp shutdown function, which
206 		 * unfortunately means it may not be hardened to work in a
207 		 * panic situation.
208 		 */
209 		smp_send_stop();
210 	} else {
211 		/*
212 		 * If we want to do crash dump after notifier calls and
213 		 * kmsg_dump, we will need architecture dependent extra
214 		 * works in addition to stopping other CPUs.
215 		 */
216 		crash_smp_send_stop();
217 	}
218 
219 	/*
220 	 * Run any panic handlers, including those that might need to
221 	 * add information to the kmsg dump output.
222 	 */
223 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
224 
225 	/* Call flush even twice. It tries harder with a single online CPU */
226 	printk_safe_flush_on_panic();
227 	kmsg_dump(KMSG_DUMP_PANIC);
228 
229 	/*
230 	 * If you doubt kdump always works fine in any situation,
231 	 * "crash_kexec_post_notifiers" offers you a chance to run
232 	 * panic_notifiers and dumping kmsg before kdump.
233 	 * Note: since some panic_notifiers can make crashed kernel
234 	 * more unstable, it can increase risks of the kdump failure too.
235 	 *
236 	 * Bypass the panic_cpu check and call __crash_kexec directly.
237 	 */
238 	if (_crash_kexec_post_notifiers)
239 		__crash_kexec(NULL);
240 
241 #ifdef CONFIG_VT
242 	unblank_screen();
243 #endif
244 	console_unblank();
245 
246 	/*
247 	 * We may have ended up stopping the CPU holding the lock (in
248 	 * smp_send_stop()) while still having some valuable data in the console
249 	 * buffer.  Try to acquire the lock then release it regardless of the
250 	 * result.  The release will also print the buffers out.  Locks debug
251 	 * should be disabled to avoid reporting bad unlock balance when
252 	 * panic() is not being callled from OOPS.
253 	 */
254 	debug_locks_off();
255 	console_flush_on_panic();
256 
257 	if (!panic_blink)
258 		panic_blink = no_blink;
259 
260 	if (panic_timeout > 0) {
261 		/*
262 		 * Delay timeout seconds before rebooting the machine.
263 		 * We can't use the "normal" timers since we just panicked.
264 		 */
265 		pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
266 
267 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
268 			touch_nmi_watchdog();
269 			if (i >= i_next) {
270 				i += panic_blink(state ^= 1);
271 				i_next = i + 3600 / PANIC_BLINK_SPD;
272 			}
273 			mdelay(PANIC_TIMER_STEP);
274 		}
275 	}
276 	if (panic_timeout != 0) {
277 		/*
278 		 * This will not be a clean reboot, with everything
279 		 * shutting down.  But if there is a chance of
280 		 * rebooting the system it will be rebooted.
281 		 */
282 		emergency_restart();
283 	}
284 #ifdef __sparc__
285 	{
286 		extern int stop_a_enabled;
287 		/* Make sure the user can actually press Stop-A (L1-A) */
288 		stop_a_enabled = 1;
289 		pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
290 			 "twice on console to return to the boot prom\n");
291 	}
292 #endif
293 #if defined(CONFIG_S390)
294 	{
295 		unsigned long caller;
296 
297 		caller = (unsigned long)__builtin_return_address(0);
298 		disabled_wait(caller);
299 	}
300 #endif
301 	pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
302 	local_irq_enable();
303 	for (i = 0; ; i += PANIC_TIMER_STEP) {
304 		touch_softlockup_watchdog();
305 		if (i >= i_next) {
306 			i += panic_blink(state ^= 1);
307 			i_next = i + 3600 / PANIC_BLINK_SPD;
308 		}
309 		mdelay(PANIC_TIMER_STEP);
310 	}
311 }
312 
313 EXPORT_SYMBOL(panic);
314 
315 /*
316  * TAINT_FORCED_RMMOD could be a per-module flag but the module
317  * is being removed anyway.
318  */
319 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
320 	[ TAINT_PROPRIETARY_MODULE ]	= { 'P', 'G', true },
321 	[ TAINT_FORCED_MODULE ]		= { 'F', ' ', true },
322 	[ TAINT_CPU_OUT_OF_SPEC ]	= { 'S', ' ', false },
323 	[ TAINT_FORCED_RMMOD ]		= { 'R', ' ', false },
324 	[ TAINT_MACHINE_CHECK ]		= { 'M', ' ', false },
325 	[ TAINT_BAD_PAGE ]		= { 'B', ' ', false },
326 	[ TAINT_USER ]			= { 'U', ' ', false },
327 	[ TAINT_DIE ]			= { 'D', ' ', false },
328 	[ TAINT_OVERRIDDEN_ACPI_TABLE ]	= { 'A', ' ', false },
329 	[ TAINT_WARN ]			= { 'W', ' ', false },
330 	[ TAINT_CRAP ]			= { 'C', ' ', true },
331 	[ TAINT_FIRMWARE_WORKAROUND ]	= { 'I', ' ', false },
332 	[ TAINT_OOT_MODULE ]		= { 'O', ' ', true },
333 	[ TAINT_UNSIGNED_MODULE ]	= { 'E', ' ', true },
334 	[ TAINT_SOFTLOCKUP ]		= { 'L', ' ', false },
335 	[ TAINT_LIVEPATCH ]		= { 'K', ' ', true },
336 	[ TAINT_AUX ]			= { 'X', ' ', true },
337 	[ TAINT_RANDSTRUCT ]		= { 'T', ' ', true },
338 };
339 
340 /**
341  * print_tainted - return a string to represent the kernel taint state.
342  *
343  * For individual taint flag meanings, see Documentation/sysctl/kernel.txt
344  *
345  * The string is overwritten by the next call to print_tainted(),
346  * but is always NULL terminated.
347  */
348 const char *print_tainted(void)
349 {
350 	static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
351 
352 	BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
353 
354 	if (tainted_mask) {
355 		char *s;
356 		int i;
357 
358 		s = buf + sprintf(buf, "Tainted: ");
359 		for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
360 			const struct taint_flag *t = &taint_flags[i];
361 			*s++ = test_bit(i, &tainted_mask) ?
362 					t->c_true : t->c_false;
363 		}
364 		*s = 0;
365 	} else
366 		snprintf(buf, sizeof(buf), "Not tainted");
367 
368 	return buf;
369 }
370 
371 int test_taint(unsigned flag)
372 {
373 	return test_bit(flag, &tainted_mask);
374 }
375 EXPORT_SYMBOL(test_taint);
376 
377 unsigned long get_taint(void)
378 {
379 	return tainted_mask;
380 }
381 
382 /**
383  * add_taint: add a taint flag if not already set.
384  * @flag: one of the TAINT_* constants.
385  * @lockdep_ok: whether lock debugging is still OK.
386  *
387  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
388  * some notewortht-but-not-corrupting cases, it can be set to true.
389  */
390 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
391 {
392 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
393 		pr_warn("Disabling lock debugging due to kernel taint\n");
394 
395 	set_bit(flag, &tainted_mask);
396 }
397 EXPORT_SYMBOL(add_taint);
398 
399 static void spin_msec(int msecs)
400 {
401 	int i;
402 
403 	for (i = 0; i < msecs; i++) {
404 		touch_nmi_watchdog();
405 		mdelay(1);
406 	}
407 }
408 
409 /*
410  * It just happens that oops_enter() and oops_exit() are identically
411  * implemented...
412  */
413 static void do_oops_enter_exit(void)
414 {
415 	unsigned long flags;
416 	static int spin_counter;
417 
418 	if (!pause_on_oops)
419 		return;
420 
421 	spin_lock_irqsave(&pause_on_oops_lock, flags);
422 	if (pause_on_oops_flag == 0) {
423 		/* This CPU may now print the oops message */
424 		pause_on_oops_flag = 1;
425 	} else {
426 		/* We need to stall this CPU */
427 		if (!spin_counter) {
428 			/* This CPU gets to do the counting */
429 			spin_counter = pause_on_oops;
430 			do {
431 				spin_unlock(&pause_on_oops_lock);
432 				spin_msec(MSEC_PER_SEC);
433 				spin_lock(&pause_on_oops_lock);
434 			} while (--spin_counter);
435 			pause_on_oops_flag = 0;
436 		} else {
437 			/* This CPU waits for a different one */
438 			while (spin_counter) {
439 				spin_unlock(&pause_on_oops_lock);
440 				spin_msec(1);
441 				spin_lock(&pause_on_oops_lock);
442 			}
443 		}
444 	}
445 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
446 }
447 
448 /*
449  * Return true if the calling CPU is allowed to print oops-related info.
450  * This is a bit racy..
451  */
452 int oops_may_print(void)
453 {
454 	return pause_on_oops_flag == 0;
455 }
456 
457 /*
458  * Called when the architecture enters its oops handler, before it prints
459  * anything.  If this is the first CPU to oops, and it's oopsing the first
460  * time then let it proceed.
461  *
462  * This is all enabled by the pause_on_oops kernel boot option.  We do all
463  * this to ensure that oopses don't scroll off the screen.  It has the
464  * side-effect of preventing later-oopsing CPUs from mucking up the display,
465  * too.
466  *
467  * It turns out that the CPU which is allowed to print ends up pausing for
468  * the right duration, whereas all the other CPUs pause for twice as long:
469  * once in oops_enter(), once in oops_exit().
470  */
471 void oops_enter(void)
472 {
473 	tracing_off();
474 	/* can't trust the integrity of the kernel anymore: */
475 	debug_locks_off();
476 	do_oops_enter_exit();
477 }
478 
479 /*
480  * 64-bit random ID for oopses:
481  */
482 static u64 oops_id;
483 
484 static int init_oops_id(void)
485 {
486 	if (!oops_id)
487 		get_random_bytes(&oops_id, sizeof(oops_id));
488 	else
489 		oops_id++;
490 
491 	return 0;
492 }
493 late_initcall(init_oops_id);
494 
495 void print_oops_end_marker(void)
496 {
497 	init_oops_id();
498 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
499 }
500 
501 /*
502  * Called when the architecture exits its oops handler, after printing
503  * everything.
504  */
505 void oops_exit(void)
506 {
507 	do_oops_enter_exit();
508 	print_oops_end_marker();
509 	kmsg_dump(KMSG_DUMP_OOPS);
510 }
511 
512 struct warn_args {
513 	const char *fmt;
514 	va_list args;
515 };
516 
517 void __warn(const char *file, int line, void *caller, unsigned taint,
518 	    struct pt_regs *regs, struct warn_args *args)
519 {
520 	disable_trace_on_warning();
521 
522 	if (args)
523 		pr_warn(CUT_HERE);
524 
525 	if (file)
526 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
527 			raw_smp_processor_id(), current->pid, file, line,
528 			caller);
529 	else
530 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
531 			raw_smp_processor_id(), current->pid, caller);
532 
533 	if (args)
534 		vprintk(args->fmt, args->args);
535 
536 	if (panic_on_warn) {
537 		/*
538 		 * This thread may hit another WARN() in the panic path.
539 		 * Resetting this prevents additional WARN() from panicking the
540 		 * system on this thread.  Other threads are blocked by the
541 		 * panic_mutex in panic().
542 		 */
543 		panic_on_warn = 0;
544 		panic("panic_on_warn set ...\n");
545 	}
546 
547 	print_modules();
548 
549 	if (regs)
550 		show_regs(regs);
551 	else
552 		dump_stack();
553 
554 	print_irqtrace_events(current);
555 
556 	print_oops_end_marker();
557 
558 	/* Just a warning, don't kill lockdep. */
559 	add_taint(taint, LOCKDEP_STILL_OK);
560 }
561 
562 #ifdef WANT_WARN_ON_SLOWPATH
563 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
564 {
565 	struct warn_args args;
566 
567 	args.fmt = fmt;
568 	va_start(args.args, fmt);
569 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
570 	       &args);
571 	va_end(args.args);
572 }
573 EXPORT_SYMBOL(warn_slowpath_fmt);
574 
575 void warn_slowpath_fmt_taint(const char *file, int line,
576 			     unsigned taint, const char *fmt, ...)
577 {
578 	struct warn_args args;
579 
580 	args.fmt = fmt;
581 	va_start(args.args, fmt);
582 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
583 	va_end(args.args);
584 }
585 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
586 
587 void warn_slowpath_null(const char *file, int line)
588 {
589 	pr_warn(CUT_HERE);
590 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
591 }
592 EXPORT_SYMBOL(warn_slowpath_null);
593 #else
594 void __warn_printk(const char *fmt, ...)
595 {
596 	va_list args;
597 
598 	pr_warn(CUT_HERE);
599 
600 	va_start(args, fmt);
601 	vprintk(fmt, args);
602 	va_end(args);
603 }
604 EXPORT_SYMBOL(__warn_printk);
605 #endif
606 
607 #ifdef CONFIG_BUG
608 
609 /* Support resetting WARN*_ONCE state */
610 
611 static int clear_warn_once_set(void *data, u64 val)
612 {
613 	generic_bug_clear_once();
614 	memset(__start_once, 0, __end_once - __start_once);
615 	return 0;
616 }
617 
618 DEFINE_SIMPLE_ATTRIBUTE(clear_warn_once_fops,
619 			NULL,
620 			clear_warn_once_set,
621 			"%lld\n");
622 
623 static __init int register_warn_debugfs(void)
624 {
625 	/* Don't care about failure */
626 	debugfs_create_file("clear_warn_once", 0200, NULL,
627 			    NULL, &clear_warn_once_fops);
628 	return 0;
629 }
630 
631 device_initcall(register_warn_debugfs);
632 #endif
633 
634 #ifdef CONFIG_STACKPROTECTOR
635 
636 /*
637  * Called when gcc's -fstack-protector feature is used, and
638  * gcc detects corruption of the on-stack canary value
639  */
640 __visible void __stack_chk_fail(void)
641 {
642 	panic("stack-protector: Kernel stack is corrupted in: %pB",
643 		__builtin_return_address(0));
644 }
645 EXPORT_SYMBOL(__stack_chk_fail);
646 
647 #endif
648 
649 #ifdef CONFIG_ARCH_HAS_REFCOUNT
650 void refcount_error_report(struct pt_regs *regs, const char *err)
651 {
652 	WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
653 		err, (void *)instruction_pointer(regs),
654 		current->comm, task_pid_nr(current),
655 		from_kuid_munged(&init_user_ns, current_uid()),
656 		from_kuid_munged(&init_user_ns, current_euid()));
657 }
658 #endif
659 
660 core_param(panic, panic_timeout, int, 0644);
661 core_param(pause_on_oops, pause_on_oops, int, 0644);
662 core_param(panic_on_warn, panic_on_warn, int, 0644);
663 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
664 
665 static int __init oops_setup(char *s)
666 {
667 	if (!s)
668 		return -EINVAL;
669 	if (!strcmp(s, "panic"))
670 		panic_on_oops = 1;
671 	return 0;
672 }
673 early_param("oops", oops_setup);
674