xref: /openbmc/linux/kernel/panic.c (revision f5b06569)
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 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
112 
113 	/*
114 	 * Disable local interrupts. This will prevent panic_smp_self_stop
115 	 * from deadlocking the first cpu that invokes the panic, since
116 	 * there is nothing to prevent an interrupt handler (that runs
117 	 * after setting panic_cpu) from invoking panic() again.
118 	 */
119 	local_irq_disable();
120 
121 	/*
122 	 * It's possible to come here directly from a panic-assertion and
123 	 * not have preempt disabled. Some functions called from here want
124 	 * preempt to be disabled. No point enabling it later though...
125 	 *
126 	 * Only one CPU is allowed to execute the panic code from here. For
127 	 * multiple parallel invocations of panic, all other CPUs either
128 	 * stop themself or will wait until they are stopped by the 1st CPU
129 	 * with smp_send_stop().
130 	 *
131 	 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
132 	 * comes here, so go ahead.
133 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
134 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
135 	 */
136 	this_cpu = raw_smp_processor_id();
137 	old_cpu  = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
138 
139 	if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
140 		panic_smp_self_stop();
141 
142 	console_verbose();
143 	bust_spinlocks(1);
144 	va_start(args, fmt);
145 	vsnprintf(buf, sizeof(buf), fmt, args);
146 	va_end(args);
147 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
148 #ifdef CONFIG_DEBUG_BUGVERBOSE
149 	/*
150 	 * Avoid nested stack-dumping if a panic occurs during oops processing
151 	 */
152 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
153 		dump_stack();
154 #endif
155 
156 	/*
157 	 * If we have crashed and we have a crash kernel loaded let it handle
158 	 * everything else.
159 	 * If we want to run this after calling panic_notifiers, pass
160 	 * the "crash_kexec_post_notifiers" option to the kernel.
161 	 *
162 	 * Bypass the panic_cpu check and call __crash_kexec directly.
163 	 */
164 	if (!_crash_kexec_post_notifiers) {
165 		printk_nmi_flush_on_panic();
166 		__crash_kexec(NULL);
167 	}
168 
169 	/*
170 	 * Note smp_send_stop is the usual smp shutdown function, which
171 	 * unfortunately means it may not be hardened to work in a panic
172 	 * situation.
173 	 */
174 	smp_send_stop();
175 
176 	/*
177 	 * Run any panic handlers, including those that might need to
178 	 * add information to the kmsg dump output.
179 	 */
180 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
181 
182 	/* Call flush even twice. It tries harder with a single online CPU */
183 	printk_nmi_flush_on_panic();
184 	kmsg_dump(KMSG_DUMP_PANIC);
185 
186 	/*
187 	 * If you doubt kdump always works fine in any situation,
188 	 * "crash_kexec_post_notifiers" offers you a chance to run
189 	 * panic_notifiers and dumping kmsg before kdump.
190 	 * Note: since some panic_notifiers can make crashed kernel
191 	 * more unstable, it can increase risks of the kdump failure too.
192 	 *
193 	 * Bypass the panic_cpu check and call __crash_kexec directly.
194 	 */
195 	if (_crash_kexec_post_notifiers)
196 		__crash_kexec(NULL);
197 
198 	bust_spinlocks(0);
199 
200 	/*
201 	 * We may have ended up stopping the CPU holding the lock (in
202 	 * smp_send_stop()) while still having some valuable data in the console
203 	 * buffer.  Try to acquire the lock then release it regardless of the
204 	 * result.  The release will also print the buffers out.  Locks debug
205 	 * should be disabled to avoid reporting bad unlock balance when
206 	 * panic() is not being callled from OOPS.
207 	 */
208 	debug_locks_off();
209 	console_flush_on_panic();
210 
211 	if (!panic_blink)
212 		panic_blink = no_blink;
213 
214 	if (panic_timeout > 0) {
215 		/*
216 		 * Delay timeout seconds before rebooting the machine.
217 		 * We can't use the "normal" timers since we just panicked.
218 		 */
219 		pr_emerg("Rebooting in %d seconds..", panic_timeout);
220 
221 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
222 			touch_nmi_watchdog();
223 			if (i >= i_next) {
224 				i += panic_blink(state ^= 1);
225 				i_next = i + 3600 / PANIC_BLINK_SPD;
226 			}
227 			mdelay(PANIC_TIMER_STEP);
228 		}
229 	}
230 	if (panic_timeout != 0) {
231 		/*
232 		 * This will not be a clean reboot, with everything
233 		 * shutting down.  But if there is a chance of
234 		 * rebooting the system it will be rebooted.
235 		 */
236 		emergency_restart();
237 	}
238 #ifdef __sparc__
239 	{
240 		extern int stop_a_enabled;
241 		/* Make sure the user can actually press Stop-A (L1-A) */
242 		stop_a_enabled = 1;
243 		pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
244 	}
245 #endif
246 #if defined(CONFIG_S390)
247 	{
248 		unsigned long caller;
249 
250 		caller = (unsigned long)__builtin_return_address(0);
251 		disabled_wait(caller);
252 	}
253 #endif
254 	pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
255 	local_irq_enable();
256 	for (i = 0; ; i += PANIC_TIMER_STEP) {
257 		touch_softlockup_watchdog();
258 		if (i >= i_next) {
259 			i += panic_blink(state ^= 1);
260 			i_next = i + 3600 / PANIC_BLINK_SPD;
261 		}
262 		mdelay(PANIC_TIMER_STEP);
263 	}
264 }
265 
266 EXPORT_SYMBOL(panic);
267 
268 
269 struct tnt {
270 	u8	bit;
271 	char	true;
272 	char	false;
273 };
274 
275 static const struct tnt tnts[] = {
276 	{ TAINT_PROPRIETARY_MODULE,	'P', 'G' },
277 	{ TAINT_FORCED_MODULE,		'F', ' ' },
278 	{ TAINT_CPU_OUT_OF_SPEC,	'S', ' ' },
279 	{ TAINT_FORCED_RMMOD,		'R', ' ' },
280 	{ TAINT_MACHINE_CHECK,		'M', ' ' },
281 	{ TAINT_BAD_PAGE,		'B', ' ' },
282 	{ TAINT_USER,			'U', ' ' },
283 	{ TAINT_DIE,			'D', ' ' },
284 	{ TAINT_OVERRIDDEN_ACPI_TABLE,	'A', ' ' },
285 	{ TAINT_WARN,			'W', ' ' },
286 	{ TAINT_CRAP,			'C', ' ' },
287 	{ TAINT_FIRMWARE_WORKAROUND,	'I', ' ' },
288 	{ TAINT_OOT_MODULE,		'O', ' ' },
289 	{ TAINT_UNSIGNED_MODULE,	'E', ' ' },
290 	{ TAINT_SOFTLOCKUP,		'L', ' ' },
291 	{ TAINT_LIVEPATCH,		'K', ' ' },
292 };
293 
294 /**
295  *	print_tainted - return a string to represent the kernel taint state.
296  *
297  *  'P' - Proprietary module has been loaded.
298  *  'F' - Module has been forcibly loaded.
299  *  'S' - SMP with CPUs not designed for SMP.
300  *  'R' - User forced a module unload.
301  *  'M' - System experienced a machine check exception.
302  *  'B' - System has hit bad_page.
303  *  'U' - Userspace-defined naughtiness.
304  *  'D' - Kernel has oopsed before
305  *  'A' - ACPI table overridden.
306  *  'W' - Taint on warning.
307  *  'C' - modules from drivers/staging are loaded.
308  *  'I' - Working around severe firmware bug.
309  *  'O' - Out-of-tree module has been loaded.
310  *  'E' - Unsigned module has been loaded.
311  *  'L' - A soft lockup has previously occurred.
312  *  'K' - Kernel has been live patched.
313  *
314  *	The string is overwritten by the next call to print_tainted().
315  */
316 const char *print_tainted(void)
317 {
318 	static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
319 
320 	if (tainted_mask) {
321 		char *s;
322 		int i;
323 
324 		s = buf + sprintf(buf, "Tainted: ");
325 		for (i = 0; i < ARRAY_SIZE(tnts); i++) {
326 			const struct tnt *t = &tnts[i];
327 			*s++ = test_bit(t->bit, &tainted_mask) ?
328 					t->true : t->false;
329 		}
330 		*s = 0;
331 	} else
332 		snprintf(buf, sizeof(buf), "Not tainted");
333 
334 	return buf;
335 }
336 
337 int test_taint(unsigned flag)
338 {
339 	return test_bit(flag, &tainted_mask);
340 }
341 EXPORT_SYMBOL(test_taint);
342 
343 unsigned long get_taint(void)
344 {
345 	return tainted_mask;
346 }
347 
348 /**
349  * add_taint: add a taint flag if not already set.
350  * @flag: one of the TAINT_* constants.
351  * @lockdep_ok: whether lock debugging is still OK.
352  *
353  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
354  * some notewortht-but-not-corrupting cases, it can be set to true.
355  */
356 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
357 {
358 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
359 		pr_warn("Disabling lock debugging due to kernel taint\n");
360 
361 	set_bit(flag, &tainted_mask);
362 }
363 EXPORT_SYMBOL(add_taint);
364 
365 static void spin_msec(int msecs)
366 {
367 	int i;
368 
369 	for (i = 0; i < msecs; i++) {
370 		touch_nmi_watchdog();
371 		mdelay(1);
372 	}
373 }
374 
375 /*
376  * It just happens that oops_enter() and oops_exit() are identically
377  * implemented...
378  */
379 static void do_oops_enter_exit(void)
380 {
381 	unsigned long flags;
382 	static int spin_counter;
383 
384 	if (!pause_on_oops)
385 		return;
386 
387 	spin_lock_irqsave(&pause_on_oops_lock, flags);
388 	if (pause_on_oops_flag == 0) {
389 		/* This CPU may now print the oops message */
390 		pause_on_oops_flag = 1;
391 	} else {
392 		/* We need to stall this CPU */
393 		if (!spin_counter) {
394 			/* This CPU gets to do the counting */
395 			spin_counter = pause_on_oops;
396 			do {
397 				spin_unlock(&pause_on_oops_lock);
398 				spin_msec(MSEC_PER_SEC);
399 				spin_lock(&pause_on_oops_lock);
400 			} while (--spin_counter);
401 			pause_on_oops_flag = 0;
402 		} else {
403 			/* This CPU waits for a different one */
404 			while (spin_counter) {
405 				spin_unlock(&pause_on_oops_lock);
406 				spin_msec(1);
407 				spin_lock(&pause_on_oops_lock);
408 			}
409 		}
410 	}
411 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
412 }
413 
414 /*
415  * Return true if the calling CPU is allowed to print oops-related info.
416  * This is a bit racy..
417  */
418 int oops_may_print(void)
419 {
420 	return pause_on_oops_flag == 0;
421 }
422 
423 /*
424  * Called when the architecture enters its oops handler, before it prints
425  * anything.  If this is the first CPU to oops, and it's oopsing the first
426  * time then let it proceed.
427  *
428  * This is all enabled by the pause_on_oops kernel boot option.  We do all
429  * this to ensure that oopses don't scroll off the screen.  It has the
430  * side-effect of preventing later-oopsing CPUs from mucking up the display,
431  * too.
432  *
433  * It turns out that the CPU which is allowed to print ends up pausing for
434  * the right duration, whereas all the other CPUs pause for twice as long:
435  * once in oops_enter(), once in oops_exit().
436  */
437 void oops_enter(void)
438 {
439 	tracing_off();
440 	/* can't trust the integrity of the kernel anymore: */
441 	debug_locks_off();
442 	do_oops_enter_exit();
443 }
444 
445 /*
446  * 64-bit random ID for oopses:
447  */
448 static u64 oops_id;
449 
450 static int init_oops_id(void)
451 {
452 	if (!oops_id)
453 		get_random_bytes(&oops_id, sizeof(oops_id));
454 	else
455 		oops_id++;
456 
457 	return 0;
458 }
459 late_initcall(init_oops_id);
460 
461 void print_oops_end_marker(void)
462 {
463 	init_oops_id();
464 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
465 }
466 
467 /*
468  * Called when the architecture exits its oops handler, after printing
469  * everything.
470  */
471 void oops_exit(void)
472 {
473 	do_oops_enter_exit();
474 	print_oops_end_marker();
475 	kmsg_dump(KMSG_DUMP_OOPS);
476 }
477 
478 struct warn_args {
479 	const char *fmt;
480 	va_list args;
481 };
482 
483 void __warn(const char *file, int line, void *caller, unsigned taint,
484 	    struct pt_regs *regs, struct warn_args *args)
485 {
486 	disable_trace_on_warning();
487 
488 	pr_warn("------------[ cut here ]------------\n");
489 
490 	if (file)
491 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
492 			raw_smp_processor_id(), current->pid, file, line,
493 			caller);
494 	else
495 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
496 			raw_smp_processor_id(), current->pid, caller);
497 
498 	if (args)
499 		vprintk(args->fmt, args->args);
500 
501 	if (panic_on_warn) {
502 		/*
503 		 * This thread may hit another WARN() in the panic path.
504 		 * Resetting this prevents additional WARN() from panicking the
505 		 * system on this thread.  Other threads are blocked by the
506 		 * panic_mutex in panic().
507 		 */
508 		panic_on_warn = 0;
509 		panic("panic_on_warn set ...\n");
510 	}
511 
512 	print_modules();
513 
514 	if (regs)
515 		show_regs(regs);
516 	else
517 		dump_stack();
518 
519 	print_oops_end_marker();
520 
521 	/* Just a warning, don't kill lockdep. */
522 	add_taint(taint, LOCKDEP_STILL_OK);
523 }
524 
525 #ifdef WANT_WARN_ON_SLOWPATH
526 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
527 {
528 	struct warn_args args;
529 
530 	args.fmt = fmt;
531 	va_start(args.args, fmt);
532 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
533 	       &args);
534 	va_end(args.args);
535 }
536 EXPORT_SYMBOL(warn_slowpath_fmt);
537 
538 void warn_slowpath_fmt_taint(const char *file, int line,
539 			     unsigned taint, const char *fmt, ...)
540 {
541 	struct warn_args args;
542 
543 	args.fmt = fmt;
544 	va_start(args.args, fmt);
545 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
546 	va_end(args.args);
547 }
548 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
549 
550 void warn_slowpath_null(const char *file, int line)
551 {
552 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
553 }
554 EXPORT_SYMBOL(warn_slowpath_null);
555 #endif
556 
557 #ifdef CONFIG_CC_STACKPROTECTOR
558 
559 /*
560  * Called when gcc's -fstack-protector feature is used, and
561  * gcc detects corruption of the on-stack canary value
562  */
563 __visible void __stack_chk_fail(void)
564 {
565 	panic("stack-protector: Kernel stack is corrupted in: %p\n",
566 		__builtin_return_address(0));
567 }
568 EXPORT_SYMBOL(__stack_chk_fail);
569 
570 #endif
571 
572 core_param(panic, panic_timeout, int, 0644);
573 core_param(pause_on_oops, pause_on_oops, int, 0644);
574 core_param(panic_on_warn, panic_on_warn, int, 0644);
575 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
576 
577 static int __init oops_setup(char *s)
578 {
579 	if (!s)
580 		return -EINVAL;
581 	if (!strcmp(s, "panic"))
582 		panic_on_oops = 1;
583 	return 0;
584 }
585 early_param("oops", oops_setup);
586