xref: /openbmc/linux/kernel/panic.c (revision 8684014d)
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 
27 #define PANIC_TIMER_STEP 100
28 #define PANIC_BLINK_SPD 18
29 
30 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
31 static unsigned long tainted_mask;
32 static int pause_on_oops;
33 static int pause_on_oops_flag;
34 static DEFINE_SPINLOCK(pause_on_oops_lock);
35 static bool crash_kexec_post_notifiers;
36 int panic_on_warn __read_mostly;
37 
38 int panic_timeout = CONFIG_PANIC_TIMEOUT;
39 EXPORT_SYMBOL_GPL(panic_timeout);
40 
41 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
42 
43 EXPORT_SYMBOL(panic_notifier_list);
44 
45 static long no_blink(int state)
46 {
47 	return 0;
48 }
49 
50 /* Returns how long it waited in ms */
51 long (*panic_blink)(int state);
52 EXPORT_SYMBOL(panic_blink);
53 
54 /*
55  * Stop ourself in panic -- architecture code may override this
56  */
57 void __weak panic_smp_self_stop(void)
58 {
59 	while (1)
60 		cpu_relax();
61 }
62 
63 /**
64  *	panic - halt the system
65  *	@fmt: The text string to print
66  *
67  *	Display a message, then perform cleanups.
68  *
69  *	This function never returns.
70  */
71 void panic(const char *fmt, ...)
72 {
73 	static DEFINE_SPINLOCK(panic_lock);
74 	static char buf[1024];
75 	va_list args;
76 	long i, i_next = 0;
77 	int state = 0;
78 
79 	/*
80 	 * Disable local interrupts. This will prevent panic_smp_self_stop
81 	 * from deadlocking the first cpu that invokes the panic, since
82 	 * there is nothing to prevent an interrupt handler (that runs
83 	 * after the panic_lock is acquired) from invoking panic again.
84 	 */
85 	local_irq_disable();
86 
87 	/*
88 	 * It's possible to come here directly from a panic-assertion and
89 	 * not have preempt disabled. Some functions called from here want
90 	 * preempt to be disabled. No point enabling it later though...
91 	 *
92 	 * Only one CPU is allowed to execute the panic code from here. For
93 	 * multiple parallel invocations of panic, all other CPUs either
94 	 * stop themself or will wait until they are stopped by the 1st CPU
95 	 * with smp_send_stop().
96 	 */
97 	if (!spin_trylock(&panic_lock))
98 		panic_smp_self_stop();
99 
100 	console_verbose();
101 	bust_spinlocks(1);
102 	va_start(args, fmt);
103 	vsnprintf(buf, sizeof(buf), fmt, args);
104 	va_end(args);
105 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
106 #ifdef CONFIG_DEBUG_BUGVERBOSE
107 	/*
108 	 * Avoid nested stack-dumping if a panic occurs during oops processing
109 	 */
110 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
111 		dump_stack();
112 #endif
113 
114 	/*
115 	 * If we have crashed and we have a crash kernel loaded let it handle
116 	 * everything else.
117 	 * If we want to run this after calling panic_notifiers, pass
118 	 * the "crash_kexec_post_notifiers" option to the kernel.
119 	 */
120 	if (!crash_kexec_post_notifiers)
121 		crash_kexec(NULL);
122 
123 	/*
124 	 * Note smp_send_stop is the usual smp shutdown function, which
125 	 * unfortunately means it may not be hardened to work in a panic
126 	 * situation.
127 	 */
128 	smp_send_stop();
129 
130 	/*
131 	 * Run any panic handlers, including those that might need to
132 	 * add information to the kmsg dump output.
133 	 */
134 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
135 
136 	kmsg_dump(KMSG_DUMP_PANIC);
137 
138 	/*
139 	 * If you doubt kdump always works fine in any situation,
140 	 * "crash_kexec_post_notifiers" offers you a chance to run
141 	 * panic_notifiers and dumping kmsg before kdump.
142 	 * Note: since some panic_notifiers can make crashed kernel
143 	 * more unstable, it can increase risks of the kdump failure too.
144 	 */
145 	crash_kexec(NULL);
146 
147 	bust_spinlocks(0);
148 
149 	if (!panic_blink)
150 		panic_blink = no_blink;
151 
152 	if (panic_timeout > 0) {
153 		/*
154 		 * Delay timeout seconds before rebooting the machine.
155 		 * We can't use the "normal" timers since we just panicked.
156 		 */
157 		pr_emerg("Rebooting in %d seconds..", panic_timeout);
158 
159 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
160 			touch_nmi_watchdog();
161 			if (i >= i_next) {
162 				i += panic_blink(state ^= 1);
163 				i_next = i + 3600 / PANIC_BLINK_SPD;
164 			}
165 			mdelay(PANIC_TIMER_STEP);
166 		}
167 	}
168 	if (panic_timeout != 0) {
169 		/*
170 		 * This will not be a clean reboot, with everything
171 		 * shutting down.  But if there is a chance of
172 		 * rebooting the system it will be rebooted.
173 		 */
174 		emergency_restart();
175 	}
176 #ifdef __sparc__
177 	{
178 		extern int stop_a_enabled;
179 		/* Make sure the user can actually press Stop-A (L1-A) */
180 		stop_a_enabled = 1;
181 		pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
182 	}
183 #endif
184 #if defined(CONFIG_S390)
185 	{
186 		unsigned long caller;
187 
188 		caller = (unsigned long)__builtin_return_address(0);
189 		disabled_wait(caller);
190 	}
191 #endif
192 	pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
193 	local_irq_enable();
194 	for (i = 0; ; i += PANIC_TIMER_STEP) {
195 		touch_softlockup_watchdog();
196 		if (i >= i_next) {
197 			i += panic_blink(state ^= 1);
198 			i_next = i + 3600 / PANIC_BLINK_SPD;
199 		}
200 		mdelay(PANIC_TIMER_STEP);
201 	}
202 }
203 
204 EXPORT_SYMBOL(panic);
205 
206 
207 struct tnt {
208 	u8	bit;
209 	char	true;
210 	char	false;
211 };
212 
213 static const struct tnt tnts[] = {
214 	{ TAINT_PROPRIETARY_MODULE,	'P', 'G' },
215 	{ TAINT_FORCED_MODULE,		'F', ' ' },
216 	{ TAINT_CPU_OUT_OF_SPEC,	'S', ' ' },
217 	{ TAINT_FORCED_RMMOD,		'R', ' ' },
218 	{ TAINT_MACHINE_CHECK,		'M', ' ' },
219 	{ TAINT_BAD_PAGE,		'B', ' ' },
220 	{ TAINT_USER,			'U', ' ' },
221 	{ TAINT_DIE,			'D', ' ' },
222 	{ TAINT_OVERRIDDEN_ACPI_TABLE,	'A', ' ' },
223 	{ TAINT_WARN,			'W', ' ' },
224 	{ TAINT_CRAP,			'C', ' ' },
225 	{ TAINT_FIRMWARE_WORKAROUND,	'I', ' ' },
226 	{ TAINT_OOT_MODULE,		'O', ' ' },
227 	{ TAINT_UNSIGNED_MODULE,	'E', ' ' },
228 	{ TAINT_SOFTLOCKUP,		'L', ' ' },
229 };
230 
231 /**
232  *	print_tainted - return a string to represent the kernel taint state.
233  *
234  *  'P' - Proprietary module has been loaded.
235  *  'F' - Module has been forcibly loaded.
236  *  'S' - SMP with CPUs not designed for SMP.
237  *  'R' - User forced a module unload.
238  *  'M' - System experienced a machine check exception.
239  *  'B' - System has hit bad_page.
240  *  'U' - Userspace-defined naughtiness.
241  *  'D' - Kernel has oopsed before
242  *  'A' - ACPI table overridden.
243  *  'W' - Taint on warning.
244  *  'C' - modules from drivers/staging are loaded.
245  *  'I' - Working around severe firmware bug.
246  *  'O' - Out-of-tree module has been loaded.
247  *  'E' - Unsigned module has been loaded.
248  *  'L' - A soft lockup has previously occurred.
249  *
250  *	The string is overwritten by the next call to print_tainted().
251  */
252 const char *print_tainted(void)
253 {
254 	static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
255 
256 	if (tainted_mask) {
257 		char *s;
258 		int i;
259 
260 		s = buf + sprintf(buf, "Tainted: ");
261 		for (i = 0; i < ARRAY_SIZE(tnts); i++) {
262 			const struct tnt *t = &tnts[i];
263 			*s++ = test_bit(t->bit, &tainted_mask) ?
264 					t->true : t->false;
265 		}
266 		*s = 0;
267 	} else
268 		snprintf(buf, sizeof(buf), "Not tainted");
269 
270 	return buf;
271 }
272 
273 int test_taint(unsigned flag)
274 {
275 	return test_bit(flag, &tainted_mask);
276 }
277 EXPORT_SYMBOL(test_taint);
278 
279 unsigned long get_taint(void)
280 {
281 	return tainted_mask;
282 }
283 
284 /**
285  * add_taint: add a taint flag if not already set.
286  * @flag: one of the TAINT_* constants.
287  * @lockdep_ok: whether lock debugging is still OK.
288  *
289  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
290  * some notewortht-but-not-corrupting cases, it can be set to true.
291  */
292 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
293 {
294 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
295 		pr_warn("Disabling lock debugging due to kernel taint\n");
296 
297 	set_bit(flag, &tainted_mask);
298 }
299 EXPORT_SYMBOL(add_taint);
300 
301 static void spin_msec(int msecs)
302 {
303 	int i;
304 
305 	for (i = 0; i < msecs; i++) {
306 		touch_nmi_watchdog();
307 		mdelay(1);
308 	}
309 }
310 
311 /*
312  * It just happens that oops_enter() and oops_exit() are identically
313  * implemented...
314  */
315 static void do_oops_enter_exit(void)
316 {
317 	unsigned long flags;
318 	static int spin_counter;
319 
320 	if (!pause_on_oops)
321 		return;
322 
323 	spin_lock_irqsave(&pause_on_oops_lock, flags);
324 	if (pause_on_oops_flag == 0) {
325 		/* This CPU may now print the oops message */
326 		pause_on_oops_flag = 1;
327 	} else {
328 		/* We need to stall this CPU */
329 		if (!spin_counter) {
330 			/* This CPU gets to do the counting */
331 			spin_counter = pause_on_oops;
332 			do {
333 				spin_unlock(&pause_on_oops_lock);
334 				spin_msec(MSEC_PER_SEC);
335 				spin_lock(&pause_on_oops_lock);
336 			} while (--spin_counter);
337 			pause_on_oops_flag = 0;
338 		} else {
339 			/* This CPU waits for a different one */
340 			while (spin_counter) {
341 				spin_unlock(&pause_on_oops_lock);
342 				spin_msec(1);
343 				spin_lock(&pause_on_oops_lock);
344 			}
345 		}
346 	}
347 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
348 }
349 
350 /*
351  * Return true if the calling CPU is allowed to print oops-related info.
352  * This is a bit racy..
353  */
354 int oops_may_print(void)
355 {
356 	return pause_on_oops_flag == 0;
357 }
358 
359 /*
360  * Called when the architecture enters its oops handler, before it prints
361  * anything.  If this is the first CPU to oops, and it's oopsing the first
362  * time then let it proceed.
363  *
364  * This is all enabled by the pause_on_oops kernel boot option.  We do all
365  * this to ensure that oopses don't scroll off the screen.  It has the
366  * side-effect of preventing later-oopsing CPUs from mucking up the display,
367  * too.
368  *
369  * It turns out that the CPU which is allowed to print ends up pausing for
370  * the right duration, whereas all the other CPUs pause for twice as long:
371  * once in oops_enter(), once in oops_exit().
372  */
373 void oops_enter(void)
374 {
375 	tracing_off();
376 	/* can't trust the integrity of the kernel anymore: */
377 	debug_locks_off();
378 	do_oops_enter_exit();
379 }
380 
381 /*
382  * 64-bit random ID for oopses:
383  */
384 static u64 oops_id;
385 
386 static int init_oops_id(void)
387 {
388 	if (!oops_id)
389 		get_random_bytes(&oops_id, sizeof(oops_id));
390 	else
391 		oops_id++;
392 
393 	return 0;
394 }
395 late_initcall(init_oops_id);
396 
397 void print_oops_end_marker(void)
398 {
399 	init_oops_id();
400 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
401 }
402 
403 /*
404  * Called when the architecture exits its oops handler, after printing
405  * everything.
406  */
407 void oops_exit(void)
408 {
409 	do_oops_enter_exit();
410 	print_oops_end_marker();
411 	kmsg_dump(KMSG_DUMP_OOPS);
412 }
413 
414 #ifdef WANT_WARN_ON_SLOWPATH
415 struct slowpath_args {
416 	const char *fmt;
417 	va_list args;
418 };
419 
420 static void warn_slowpath_common(const char *file, int line, void *caller,
421 				 unsigned taint, struct slowpath_args *args)
422 {
423 	disable_trace_on_warning();
424 
425 	pr_warn("------------[ cut here ]------------\n");
426 	pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS()\n",
427 		raw_smp_processor_id(), current->pid, file, line, caller);
428 
429 	if (args)
430 		vprintk(args->fmt, args->args);
431 
432 	if (panic_on_warn) {
433 		/*
434 		 * This thread may hit another WARN() in the panic path.
435 		 * Resetting this prevents additional WARN() from panicking the
436 		 * system on this thread.  Other threads are blocked by the
437 		 * panic_mutex in panic().
438 		 */
439 		panic_on_warn = 0;
440 		panic("panic_on_warn set ...\n");
441 	}
442 
443 	print_modules();
444 	dump_stack();
445 	print_oops_end_marker();
446 	/* Just a warning, don't kill lockdep. */
447 	add_taint(taint, LOCKDEP_STILL_OK);
448 }
449 
450 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
451 {
452 	struct slowpath_args args;
453 
454 	args.fmt = fmt;
455 	va_start(args.args, fmt);
456 	warn_slowpath_common(file, line, __builtin_return_address(0),
457 			     TAINT_WARN, &args);
458 	va_end(args.args);
459 }
460 EXPORT_SYMBOL(warn_slowpath_fmt);
461 
462 void warn_slowpath_fmt_taint(const char *file, int line,
463 			     unsigned taint, const char *fmt, ...)
464 {
465 	struct slowpath_args args;
466 
467 	args.fmt = fmt;
468 	va_start(args.args, fmt);
469 	warn_slowpath_common(file, line, __builtin_return_address(0),
470 			     taint, &args);
471 	va_end(args.args);
472 }
473 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
474 
475 void warn_slowpath_null(const char *file, int line)
476 {
477 	warn_slowpath_common(file, line, __builtin_return_address(0),
478 			     TAINT_WARN, NULL);
479 }
480 EXPORT_SYMBOL(warn_slowpath_null);
481 #endif
482 
483 #ifdef CONFIG_CC_STACKPROTECTOR
484 
485 /*
486  * Called when gcc's -fstack-protector feature is used, and
487  * gcc detects corruption of the on-stack canary value
488  */
489 __visible void __stack_chk_fail(void)
490 {
491 	panic("stack-protector: Kernel stack is corrupted in: %p\n",
492 		__builtin_return_address(0));
493 }
494 EXPORT_SYMBOL(__stack_chk_fail);
495 
496 #endif
497 
498 core_param(panic, panic_timeout, int, 0644);
499 core_param(pause_on_oops, pause_on_oops, int, 0644);
500 core_param(panic_on_warn, panic_on_warn, int, 0644);
501 
502 static int __init setup_crash_kexec_post_notifiers(char *s)
503 {
504 	crash_kexec_post_notifiers = true;
505 	return 0;
506 }
507 early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
508 
509 static int __init oops_setup(char *s)
510 {
511 	if (!s)
512 		return -EINVAL;
513 	if (!strcmp(s, "panic"))
514 		panic_on_oops = 1;
515 	return 0;
516 }
517 early_param("oops", oops_setup);
518