xref: /openbmc/linux/kernel/panic.c (revision 034f90b3)
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 	{ TAINT_LIVEPATCH,		'K', ' ' },
230 };
231 
232 /**
233  *	print_tainted - return a string to represent the kernel taint state.
234  *
235  *  'P' - Proprietary module has been loaded.
236  *  'F' - Module has been forcibly loaded.
237  *  'S' - SMP with CPUs not designed for SMP.
238  *  'R' - User forced a module unload.
239  *  'M' - System experienced a machine check exception.
240  *  'B' - System has hit bad_page.
241  *  'U' - Userspace-defined naughtiness.
242  *  'D' - Kernel has oopsed before
243  *  'A' - ACPI table overridden.
244  *  'W' - Taint on warning.
245  *  'C' - modules from drivers/staging are loaded.
246  *  'I' - Working around severe firmware bug.
247  *  'O' - Out-of-tree module has been loaded.
248  *  'E' - Unsigned module has been loaded.
249  *  'L' - A soft lockup has previously occurred.
250  *  'K' - Kernel has been live patched.
251  *
252  *	The string is overwritten by the next call to print_tainted().
253  */
254 const char *print_tainted(void)
255 {
256 	static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
257 
258 	if (tainted_mask) {
259 		char *s;
260 		int i;
261 
262 		s = buf + sprintf(buf, "Tainted: ");
263 		for (i = 0; i < ARRAY_SIZE(tnts); i++) {
264 			const struct tnt *t = &tnts[i];
265 			*s++ = test_bit(t->bit, &tainted_mask) ?
266 					t->true : t->false;
267 		}
268 		*s = 0;
269 	} else
270 		snprintf(buf, sizeof(buf), "Not tainted");
271 
272 	return buf;
273 }
274 
275 int test_taint(unsigned flag)
276 {
277 	return test_bit(flag, &tainted_mask);
278 }
279 EXPORT_SYMBOL(test_taint);
280 
281 unsigned long get_taint(void)
282 {
283 	return tainted_mask;
284 }
285 
286 /**
287  * add_taint: add a taint flag if not already set.
288  * @flag: one of the TAINT_* constants.
289  * @lockdep_ok: whether lock debugging is still OK.
290  *
291  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
292  * some notewortht-but-not-corrupting cases, it can be set to true.
293  */
294 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
295 {
296 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
297 		pr_warn("Disabling lock debugging due to kernel taint\n");
298 
299 	set_bit(flag, &tainted_mask);
300 }
301 EXPORT_SYMBOL(add_taint);
302 
303 static void spin_msec(int msecs)
304 {
305 	int i;
306 
307 	for (i = 0; i < msecs; i++) {
308 		touch_nmi_watchdog();
309 		mdelay(1);
310 	}
311 }
312 
313 /*
314  * It just happens that oops_enter() and oops_exit() are identically
315  * implemented...
316  */
317 static void do_oops_enter_exit(void)
318 {
319 	unsigned long flags;
320 	static int spin_counter;
321 
322 	if (!pause_on_oops)
323 		return;
324 
325 	spin_lock_irqsave(&pause_on_oops_lock, flags);
326 	if (pause_on_oops_flag == 0) {
327 		/* This CPU may now print the oops message */
328 		pause_on_oops_flag = 1;
329 	} else {
330 		/* We need to stall this CPU */
331 		if (!spin_counter) {
332 			/* This CPU gets to do the counting */
333 			spin_counter = pause_on_oops;
334 			do {
335 				spin_unlock(&pause_on_oops_lock);
336 				spin_msec(MSEC_PER_SEC);
337 				spin_lock(&pause_on_oops_lock);
338 			} while (--spin_counter);
339 			pause_on_oops_flag = 0;
340 		} else {
341 			/* This CPU waits for a different one */
342 			while (spin_counter) {
343 				spin_unlock(&pause_on_oops_lock);
344 				spin_msec(1);
345 				spin_lock(&pause_on_oops_lock);
346 			}
347 		}
348 	}
349 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
350 }
351 
352 /*
353  * Return true if the calling CPU is allowed to print oops-related info.
354  * This is a bit racy..
355  */
356 int oops_may_print(void)
357 {
358 	return pause_on_oops_flag == 0;
359 }
360 
361 /*
362  * Called when the architecture enters its oops handler, before it prints
363  * anything.  If this is the first CPU to oops, and it's oopsing the first
364  * time then let it proceed.
365  *
366  * This is all enabled by the pause_on_oops kernel boot option.  We do all
367  * this to ensure that oopses don't scroll off the screen.  It has the
368  * side-effect of preventing later-oopsing CPUs from mucking up the display,
369  * too.
370  *
371  * It turns out that the CPU which is allowed to print ends up pausing for
372  * the right duration, whereas all the other CPUs pause for twice as long:
373  * once in oops_enter(), once in oops_exit().
374  */
375 void oops_enter(void)
376 {
377 	tracing_off();
378 	/* can't trust the integrity of the kernel anymore: */
379 	debug_locks_off();
380 	do_oops_enter_exit();
381 }
382 
383 /*
384  * 64-bit random ID for oopses:
385  */
386 static u64 oops_id;
387 
388 static int init_oops_id(void)
389 {
390 	if (!oops_id)
391 		get_random_bytes(&oops_id, sizeof(oops_id));
392 	else
393 		oops_id++;
394 
395 	return 0;
396 }
397 late_initcall(init_oops_id);
398 
399 void print_oops_end_marker(void)
400 {
401 	init_oops_id();
402 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
403 }
404 
405 /*
406  * Called when the architecture exits its oops handler, after printing
407  * everything.
408  */
409 void oops_exit(void)
410 {
411 	do_oops_enter_exit();
412 	print_oops_end_marker();
413 	kmsg_dump(KMSG_DUMP_OOPS);
414 }
415 
416 #ifdef WANT_WARN_ON_SLOWPATH
417 struct slowpath_args {
418 	const char *fmt;
419 	va_list args;
420 };
421 
422 static void warn_slowpath_common(const char *file, int line, void *caller,
423 				 unsigned taint, struct slowpath_args *args)
424 {
425 	disable_trace_on_warning();
426 
427 	pr_warn("------------[ cut here ]------------\n");
428 	pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS()\n",
429 		raw_smp_processor_id(), current->pid, file, line, caller);
430 
431 	if (args)
432 		vprintk(args->fmt, args->args);
433 
434 	if (panic_on_warn) {
435 		/*
436 		 * This thread may hit another WARN() in the panic path.
437 		 * Resetting this prevents additional WARN() from panicking the
438 		 * system on this thread.  Other threads are blocked by the
439 		 * panic_mutex in panic().
440 		 */
441 		panic_on_warn = 0;
442 		panic("panic_on_warn set ...\n");
443 	}
444 
445 	print_modules();
446 	dump_stack();
447 	print_oops_end_marker();
448 	/* Just a warning, don't kill lockdep. */
449 	add_taint(taint, LOCKDEP_STILL_OK);
450 }
451 
452 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
453 {
454 	struct slowpath_args args;
455 
456 	args.fmt = fmt;
457 	va_start(args.args, fmt);
458 	warn_slowpath_common(file, line, __builtin_return_address(0),
459 			     TAINT_WARN, &args);
460 	va_end(args.args);
461 }
462 EXPORT_SYMBOL(warn_slowpath_fmt);
463 
464 void warn_slowpath_fmt_taint(const char *file, int line,
465 			     unsigned taint, const char *fmt, ...)
466 {
467 	struct slowpath_args args;
468 
469 	args.fmt = fmt;
470 	va_start(args.args, fmt);
471 	warn_slowpath_common(file, line, __builtin_return_address(0),
472 			     taint, &args);
473 	va_end(args.args);
474 }
475 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
476 
477 void warn_slowpath_null(const char *file, int line)
478 {
479 	warn_slowpath_common(file, line, __builtin_return_address(0),
480 			     TAINT_WARN, NULL);
481 }
482 EXPORT_SYMBOL(warn_slowpath_null);
483 #endif
484 
485 #ifdef CONFIG_CC_STACKPROTECTOR
486 
487 /*
488  * Called when gcc's -fstack-protector feature is used, and
489  * gcc detects corruption of the on-stack canary value
490  */
491 __visible void __stack_chk_fail(void)
492 {
493 	panic("stack-protector: Kernel stack is corrupted in: %p\n",
494 		__builtin_return_address(0));
495 }
496 EXPORT_SYMBOL(__stack_chk_fail);
497 
498 #endif
499 
500 core_param(panic, panic_timeout, int, 0644);
501 core_param(pause_on_oops, pause_on_oops, int, 0644);
502 core_param(panic_on_warn, panic_on_warn, int, 0644);
503 
504 static int __init setup_crash_kexec_post_notifiers(char *s)
505 {
506 	crash_kexec_post_notifiers = true;
507 	return 0;
508 }
509 early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
510 
511 static int __init oops_setup(char *s)
512 {
513 	if (!s)
514 		return -EINVAL;
515 	if (!strcmp(s, "panic"))
516 		panic_on_oops = 1;
517 	return 0;
518 }
519 early_param("oops", oops_setup);
520