xref: /openbmc/linux/kernel/panic.c (revision 9cfc5c90)
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 
28 #define PANIC_TIMER_STEP 100
29 #define PANIC_BLINK_SPD 18
30 
31 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
32 static unsigned long tainted_mask;
33 static int pause_on_oops;
34 static int pause_on_oops_flag;
35 static DEFINE_SPINLOCK(pause_on_oops_lock);
36 bool crash_kexec_post_notifiers;
37 int panic_on_warn __read_mostly;
38 
39 int panic_timeout = CONFIG_PANIC_TIMEOUT;
40 EXPORT_SYMBOL_GPL(panic_timeout);
41 
42 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
43 
44 EXPORT_SYMBOL(panic_notifier_list);
45 
46 static long no_blink(int state)
47 {
48 	return 0;
49 }
50 
51 /* Returns how long it waited in ms */
52 long (*panic_blink)(int state);
53 EXPORT_SYMBOL(panic_blink);
54 
55 /*
56  * Stop ourself in panic -- architecture code may override this
57  */
58 void __weak panic_smp_self_stop(void)
59 {
60 	while (1)
61 		cpu_relax();
62 }
63 
64 /**
65  *	panic - halt the system
66  *	@fmt: The text string to print
67  *
68  *	Display a message, then perform cleanups.
69  *
70  *	This function never returns.
71  */
72 void panic(const char *fmt, ...)
73 {
74 	static DEFINE_SPINLOCK(panic_lock);
75 	static char buf[1024];
76 	va_list args;
77 	long i, i_next = 0;
78 	int state = 0;
79 
80 	/*
81 	 * Disable local interrupts. This will prevent panic_smp_self_stop
82 	 * from deadlocking the first cpu that invokes the panic, since
83 	 * there is nothing to prevent an interrupt handler (that runs
84 	 * after the panic_lock is acquired) from invoking panic again.
85 	 */
86 	local_irq_disable();
87 
88 	/*
89 	 * It's possible to come here directly from a panic-assertion and
90 	 * not have preempt disabled. Some functions called from here want
91 	 * preempt to be disabled. No point enabling it later though...
92 	 *
93 	 * Only one CPU is allowed to execute the panic code from here. For
94 	 * multiple parallel invocations of panic, all other CPUs either
95 	 * stop themself or will wait until they are stopped by the 1st CPU
96 	 * with smp_send_stop().
97 	 */
98 	if (!spin_trylock(&panic_lock))
99 		panic_smp_self_stop();
100 
101 	console_verbose();
102 	bust_spinlocks(1);
103 	va_start(args, fmt);
104 	vsnprintf(buf, sizeof(buf), fmt, args);
105 	va_end(args);
106 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
107 #ifdef CONFIG_DEBUG_BUGVERBOSE
108 	/*
109 	 * Avoid nested stack-dumping if a panic occurs during oops processing
110 	 */
111 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
112 		dump_stack();
113 #endif
114 
115 	/*
116 	 * If we have crashed and we have a crash kernel loaded let it handle
117 	 * everything else.
118 	 * If we want to run this after calling panic_notifiers, pass
119 	 * the "crash_kexec_post_notifiers" option to the kernel.
120 	 */
121 	if (!crash_kexec_post_notifiers)
122 		crash_kexec(NULL);
123 
124 	/*
125 	 * Note smp_send_stop is the usual smp shutdown function, which
126 	 * unfortunately means it may not be hardened to work in a panic
127 	 * situation.
128 	 */
129 	smp_send_stop();
130 
131 	/*
132 	 * Run any panic handlers, including those that might need to
133 	 * add information to the kmsg dump output.
134 	 */
135 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
136 
137 	kmsg_dump(KMSG_DUMP_PANIC);
138 
139 	/*
140 	 * If you doubt kdump always works fine in any situation,
141 	 * "crash_kexec_post_notifiers" offers you a chance to run
142 	 * panic_notifiers and dumping kmsg before kdump.
143 	 * Note: since some panic_notifiers can make crashed kernel
144 	 * more unstable, it can increase risks of the kdump failure too.
145 	 */
146 	if (crash_kexec_post_notifiers)
147 		crash_kexec(NULL);
148 
149 	bust_spinlocks(0);
150 
151 	/*
152 	 * We may have ended up stopping the CPU holding the lock (in
153 	 * smp_send_stop()) while still having some valuable data in the console
154 	 * buffer.  Try to acquire the lock then release it regardless of the
155 	 * result.  The release will also print the buffers out.
156 	 */
157 	console_trylock();
158 	console_unlock();
159 
160 	if (!panic_blink)
161 		panic_blink = no_blink;
162 
163 	if (panic_timeout > 0) {
164 		/*
165 		 * Delay timeout seconds before rebooting the machine.
166 		 * We can't use the "normal" timers since we just panicked.
167 		 */
168 		pr_emerg("Rebooting in %d seconds..", panic_timeout);
169 
170 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
171 			touch_nmi_watchdog();
172 			if (i >= i_next) {
173 				i += panic_blink(state ^= 1);
174 				i_next = i + 3600 / PANIC_BLINK_SPD;
175 			}
176 			mdelay(PANIC_TIMER_STEP);
177 		}
178 	}
179 	if (panic_timeout != 0) {
180 		/*
181 		 * This will not be a clean reboot, with everything
182 		 * shutting down.  But if there is a chance of
183 		 * rebooting the system it will be rebooted.
184 		 */
185 		emergency_restart();
186 	}
187 #ifdef __sparc__
188 	{
189 		extern int stop_a_enabled;
190 		/* Make sure the user can actually press Stop-A (L1-A) */
191 		stop_a_enabled = 1;
192 		pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
193 	}
194 #endif
195 #if defined(CONFIG_S390)
196 	{
197 		unsigned long caller;
198 
199 		caller = (unsigned long)__builtin_return_address(0);
200 		disabled_wait(caller);
201 	}
202 #endif
203 	pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
204 	local_irq_enable();
205 	for (i = 0; ; i += PANIC_TIMER_STEP) {
206 		touch_softlockup_watchdog();
207 		if (i >= i_next) {
208 			i += panic_blink(state ^= 1);
209 			i_next = i + 3600 / PANIC_BLINK_SPD;
210 		}
211 		mdelay(PANIC_TIMER_STEP);
212 	}
213 }
214 
215 EXPORT_SYMBOL(panic);
216 
217 
218 struct tnt {
219 	u8	bit;
220 	char	true;
221 	char	false;
222 };
223 
224 static const struct tnt tnts[] = {
225 	{ TAINT_PROPRIETARY_MODULE,	'P', 'G' },
226 	{ TAINT_FORCED_MODULE,		'F', ' ' },
227 	{ TAINT_CPU_OUT_OF_SPEC,	'S', ' ' },
228 	{ TAINT_FORCED_RMMOD,		'R', ' ' },
229 	{ TAINT_MACHINE_CHECK,		'M', ' ' },
230 	{ TAINT_BAD_PAGE,		'B', ' ' },
231 	{ TAINT_USER,			'U', ' ' },
232 	{ TAINT_DIE,			'D', ' ' },
233 	{ TAINT_OVERRIDDEN_ACPI_TABLE,	'A', ' ' },
234 	{ TAINT_WARN,			'W', ' ' },
235 	{ TAINT_CRAP,			'C', ' ' },
236 	{ TAINT_FIRMWARE_WORKAROUND,	'I', ' ' },
237 	{ TAINT_OOT_MODULE,		'O', ' ' },
238 	{ TAINT_UNSIGNED_MODULE,	'E', ' ' },
239 	{ TAINT_SOFTLOCKUP,		'L', ' ' },
240 	{ TAINT_LIVEPATCH,		'K', ' ' },
241 };
242 
243 /**
244  *	print_tainted - return a string to represent the kernel taint state.
245  *
246  *  'P' - Proprietary module has been loaded.
247  *  'F' - Module has been forcibly loaded.
248  *  'S' - SMP with CPUs not designed for SMP.
249  *  'R' - User forced a module unload.
250  *  'M' - System experienced a machine check exception.
251  *  'B' - System has hit bad_page.
252  *  'U' - Userspace-defined naughtiness.
253  *  'D' - Kernel has oopsed before
254  *  'A' - ACPI table overridden.
255  *  'W' - Taint on warning.
256  *  'C' - modules from drivers/staging are loaded.
257  *  'I' - Working around severe firmware bug.
258  *  'O' - Out-of-tree module has been loaded.
259  *  'E' - Unsigned module has been loaded.
260  *  'L' - A soft lockup has previously occurred.
261  *  'K' - Kernel has been live patched.
262  *
263  *	The string is overwritten by the next call to print_tainted().
264  */
265 const char *print_tainted(void)
266 {
267 	static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
268 
269 	if (tainted_mask) {
270 		char *s;
271 		int i;
272 
273 		s = buf + sprintf(buf, "Tainted: ");
274 		for (i = 0; i < ARRAY_SIZE(tnts); i++) {
275 			const struct tnt *t = &tnts[i];
276 			*s++ = test_bit(t->bit, &tainted_mask) ?
277 					t->true : t->false;
278 		}
279 		*s = 0;
280 	} else
281 		snprintf(buf, sizeof(buf), "Not tainted");
282 
283 	return buf;
284 }
285 
286 int test_taint(unsigned flag)
287 {
288 	return test_bit(flag, &tainted_mask);
289 }
290 EXPORT_SYMBOL(test_taint);
291 
292 unsigned long get_taint(void)
293 {
294 	return tainted_mask;
295 }
296 
297 /**
298  * add_taint: add a taint flag if not already set.
299  * @flag: one of the TAINT_* constants.
300  * @lockdep_ok: whether lock debugging is still OK.
301  *
302  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
303  * some notewortht-but-not-corrupting cases, it can be set to true.
304  */
305 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
306 {
307 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
308 		pr_warn("Disabling lock debugging due to kernel taint\n");
309 
310 	set_bit(flag, &tainted_mask);
311 }
312 EXPORT_SYMBOL(add_taint);
313 
314 static void spin_msec(int msecs)
315 {
316 	int i;
317 
318 	for (i = 0; i < msecs; i++) {
319 		touch_nmi_watchdog();
320 		mdelay(1);
321 	}
322 }
323 
324 /*
325  * It just happens that oops_enter() and oops_exit() are identically
326  * implemented...
327  */
328 static void do_oops_enter_exit(void)
329 {
330 	unsigned long flags;
331 	static int spin_counter;
332 
333 	if (!pause_on_oops)
334 		return;
335 
336 	spin_lock_irqsave(&pause_on_oops_lock, flags);
337 	if (pause_on_oops_flag == 0) {
338 		/* This CPU may now print the oops message */
339 		pause_on_oops_flag = 1;
340 	} else {
341 		/* We need to stall this CPU */
342 		if (!spin_counter) {
343 			/* This CPU gets to do the counting */
344 			spin_counter = pause_on_oops;
345 			do {
346 				spin_unlock(&pause_on_oops_lock);
347 				spin_msec(MSEC_PER_SEC);
348 				spin_lock(&pause_on_oops_lock);
349 			} while (--spin_counter);
350 			pause_on_oops_flag = 0;
351 		} else {
352 			/* This CPU waits for a different one */
353 			while (spin_counter) {
354 				spin_unlock(&pause_on_oops_lock);
355 				spin_msec(1);
356 				spin_lock(&pause_on_oops_lock);
357 			}
358 		}
359 	}
360 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
361 }
362 
363 /*
364  * Return true if the calling CPU is allowed to print oops-related info.
365  * This is a bit racy..
366  */
367 int oops_may_print(void)
368 {
369 	return pause_on_oops_flag == 0;
370 }
371 
372 /*
373  * Called when the architecture enters its oops handler, before it prints
374  * anything.  If this is the first CPU to oops, and it's oopsing the first
375  * time then let it proceed.
376  *
377  * This is all enabled by the pause_on_oops kernel boot option.  We do all
378  * this to ensure that oopses don't scroll off the screen.  It has the
379  * side-effect of preventing later-oopsing CPUs from mucking up the display,
380  * too.
381  *
382  * It turns out that the CPU which is allowed to print ends up pausing for
383  * the right duration, whereas all the other CPUs pause for twice as long:
384  * once in oops_enter(), once in oops_exit().
385  */
386 void oops_enter(void)
387 {
388 	tracing_off();
389 	/* can't trust the integrity of the kernel anymore: */
390 	debug_locks_off();
391 	do_oops_enter_exit();
392 }
393 
394 /*
395  * 64-bit random ID for oopses:
396  */
397 static u64 oops_id;
398 
399 static int init_oops_id(void)
400 {
401 	if (!oops_id)
402 		get_random_bytes(&oops_id, sizeof(oops_id));
403 	else
404 		oops_id++;
405 
406 	return 0;
407 }
408 late_initcall(init_oops_id);
409 
410 void print_oops_end_marker(void)
411 {
412 	init_oops_id();
413 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
414 }
415 
416 /*
417  * Called when the architecture exits its oops handler, after printing
418  * everything.
419  */
420 void oops_exit(void)
421 {
422 	do_oops_enter_exit();
423 	print_oops_end_marker();
424 	kmsg_dump(KMSG_DUMP_OOPS);
425 }
426 
427 #ifdef WANT_WARN_ON_SLOWPATH
428 struct slowpath_args {
429 	const char *fmt;
430 	va_list args;
431 };
432 
433 static void warn_slowpath_common(const char *file, int line, void *caller,
434 				 unsigned taint, struct slowpath_args *args)
435 {
436 	disable_trace_on_warning();
437 
438 	pr_warn("------------[ cut here ]------------\n");
439 	pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS()\n",
440 		raw_smp_processor_id(), current->pid, file, line, caller);
441 
442 	if (args)
443 		vprintk(args->fmt, args->args);
444 
445 	if (panic_on_warn) {
446 		/*
447 		 * This thread may hit another WARN() in the panic path.
448 		 * Resetting this prevents additional WARN() from panicking the
449 		 * system on this thread.  Other threads are blocked by the
450 		 * panic_mutex in panic().
451 		 */
452 		panic_on_warn = 0;
453 		panic("panic_on_warn set ...\n");
454 	}
455 
456 	print_modules();
457 	dump_stack();
458 	print_oops_end_marker();
459 	/* Just a warning, don't kill lockdep. */
460 	add_taint(taint, LOCKDEP_STILL_OK);
461 }
462 
463 void warn_slowpath_fmt(const char *file, int line, 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_WARN, &args);
471 	va_end(args.args);
472 }
473 EXPORT_SYMBOL(warn_slowpath_fmt);
474 
475 void warn_slowpath_fmt_taint(const char *file, int line,
476 			     unsigned taint, const char *fmt, ...)
477 {
478 	struct slowpath_args args;
479 
480 	args.fmt = fmt;
481 	va_start(args.args, fmt);
482 	warn_slowpath_common(file, line, __builtin_return_address(0),
483 			     taint, &args);
484 	va_end(args.args);
485 }
486 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
487 
488 void warn_slowpath_null(const char *file, int line)
489 {
490 	warn_slowpath_common(file, line, __builtin_return_address(0),
491 			     TAINT_WARN, NULL);
492 }
493 EXPORT_SYMBOL(warn_slowpath_null);
494 #endif
495 
496 #ifdef CONFIG_CC_STACKPROTECTOR
497 
498 /*
499  * Called when gcc's -fstack-protector feature is used, and
500  * gcc detects corruption of the on-stack canary value
501  */
502 __visible void __stack_chk_fail(void)
503 {
504 	panic("stack-protector: Kernel stack is corrupted in: %p\n",
505 		__builtin_return_address(0));
506 }
507 EXPORT_SYMBOL(__stack_chk_fail);
508 
509 #endif
510 
511 core_param(panic, panic_timeout, int, 0644);
512 core_param(pause_on_oops, pause_on_oops, int, 0644);
513 core_param(panic_on_warn, panic_on_warn, int, 0644);
514 
515 static int __init setup_crash_kexec_post_notifiers(char *s)
516 {
517 	crash_kexec_post_notifiers = true;
518 	return 0;
519 }
520 early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
521 
522 static int __init oops_setup(char *s)
523 {
524 	if (!s)
525 		return -EINVAL;
526 	if (!strcmp(s, "panic"))
527 		panic_on_oops = 1;
528 	return 0;
529 }
530 early_param("oops", oops_setup);
531