xref: /openbmc/linux/kernel/panic.c (revision 8730046c)
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 /*
75  * Stop other CPUs in panic.  Architecture dependent code may override this
76  * with more suitable version.  For example, if the architecture supports
77  * crash dump, it should save registers of each stopped CPU and disable
78  * per-CPU features such as virtualization extensions.
79  */
80 void __weak crash_smp_send_stop(void)
81 {
82 	static int cpus_stopped;
83 
84 	/*
85 	 * This function can be called twice in panic path, but obviously
86 	 * we execute this only once.
87 	 */
88 	if (cpus_stopped)
89 		return;
90 
91 	/*
92 	 * Note smp_send_stop is the usual smp shutdown function, which
93 	 * unfortunately means it may not be hardened to work in a panic
94 	 * situation.
95 	 */
96 	smp_send_stop();
97 	cpus_stopped = 1;
98 }
99 
100 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
101 
102 /*
103  * A variant of panic() called from NMI context. We return if we've already
104  * panicked on this CPU. If another CPU already panicked, loop in
105  * nmi_panic_self_stop() which can provide architecture dependent code such
106  * as saving register state for crash dump.
107  */
108 void nmi_panic(struct pt_regs *regs, const char *msg)
109 {
110 	int old_cpu, cpu;
111 
112 	cpu = raw_smp_processor_id();
113 	old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
114 
115 	if (old_cpu == PANIC_CPU_INVALID)
116 		panic("%s", msg);
117 	else if (old_cpu != cpu)
118 		nmi_panic_self_stop(regs);
119 }
120 EXPORT_SYMBOL(nmi_panic);
121 
122 /**
123  *	panic - halt the system
124  *	@fmt: The text string to print
125  *
126  *	Display a message, then perform cleanups.
127  *
128  *	This function never returns.
129  */
130 void panic(const char *fmt, ...)
131 {
132 	static char buf[1024];
133 	va_list args;
134 	long i, i_next = 0;
135 	int state = 0;
136 	int old_cpu, this_cpu;
137 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
138 
139 	/*
140 	 * Disable local interrupts. This will prevent panic_smp_self_stop
141 	 * from deadlocking the first cpu that invokes the panic, since
142 	 * there is nothing to prevent an interrupt handler (that runs
143 	 * after setting panic_cpu) from invoking panic() again.
144 	 */
145 	local_irq_disable();
146 
147 	/*
148 	 * It's possible to come here directly from a panic-assertion and
149 	 * not have preempt disabled. Some functions called from here want
150 	 * preempt to be disabled. No point enabling it later though...
151 	 *
152 	 * Only one CPU is allowed to execute the panic code from here. For
153 	 * multiple parallel invocations of panic, all other CPUs either
154 	 * stop themself or will wait until they are stopped by the 1st CPU
155 	 * with smp_send_stop().
156 	 *
157 	 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
158 	 * comes here, so go ahead.
159 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
160 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
161 	 */
162 	this_cpu = raw_smp_processor_id();
163 	old_cpu  = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
164 
165 	if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
166 		panic_smp_self_stop();
167 
168 	console_verbose();
169 	bust_spinlocks(1);
170 	va_start(args, fmt);
171 	vsnprintf(buf, sizeof(buf), fmt, args);
172 	va_end(args);
173 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
174 #ifdef CONFIG_DEBUG_BUGVERBOSE
175 	/*
176 	 * Avoid nested stack-dumping if a panic occurs during oops processing
177 	 */
178 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
179 		dump_stack();
180 #endif
181 
182 	/*
183 	 * If we have crashed and we have a crash kernel loaded let it handle
184 	 * everything else.
185 	 * If we want to run this after calling panic_notifiers, pass
186 	 * the "crash_kexec_post_notifiers" option to the kernel.
187 	 *
188 	 * Bypass the panic_cpu check and call __crash_kexec directly.
189 	 */
190 	if (!_crash_kexec_post_notifiers) {
191 		printk_nmi_flush_on_panic();
192 		__crash_kexec(NULL);
193 
194 		/*
195 		 * Note smp_send_stop is the usual smp shutdown function, which
196 		 * unfortunately means it may not be hardened to work in a
197 		 * panic situation.
198 		 */
199 		smp_send_stop();
200 	} else {
201 		/*
202 		 * If we want to do crash dump after notifier calls and
203 		 * kmsg_dump, we will need architecture dependent extra
204 		 * works in addition to stopping other CPUs.
205 		 */
206 		crash_smp_send_stop();
207 	}
208 
209 	/*
210 	 * Run any panic handlers, including those that might need to
211 	 * add information to the kmsg dump output.
212 	 */
213 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
214 
215 	/* Call flush even twice. It tries harder with a single online CPU */
216 	printk_nmi_flush_on_panic();
217 	kmsg_dump(KMSG_DUMP_PANIC);
218 
219 	/*
220 	 * If you doubt kdump always works fine in any situation,
221 	 * "crash_kexec_post_notifiers" offers you a chance to run
222 	 * panic_notifiers and dumping kmsg before kdump.
223 	 * Note: since some panic_notifiers can make crashed kernel
224 	 * more unstable, it can increase risks of the kdump failure too.
225 	 *
226 	 * Bypass the panic_cpu check and call __crash_kexec directly.
227 	 */
228 	if (_crash_kexec_post_notifiers)
229 		__crash_kexec(NULL);
230 
231 	bust_spinlocks(0);
232 
233 	/*
234 	 * We may have ended up stopping the CPU holding the lock (in
235 	 * smp_send_stop()) while still having some valuable data in the console
236 	 * buffer.  Try to acquire the lock then release it regardless of the
237 	 * result.  The release will also print the buffers out.  Locks debug
238 	 * should be disabled to avoid reporting bad unlock balance when
239 	 * panic() is not being callled from OOPS.
240 	 */
241 	debug_locks_off();
242 	console_flush_on_panic();
243 
244 	if (!panic_blink)
245 		panic_blink = no_blink;
246 
247 	if (panic_timeout > 0) {
248 		/*
249 		 * Delay timeout seconds before rebooting the machine.
250 		 * We can't use the "normal" timers since we just panicked.
251 		 */
252 		pr_emerg("Rebooting in %d seconds..", panic_timeout);
253 
254 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
255 			touch_nmi_watchdog();
256 			if (i >= i_next) {
257 				i += panic_blink(state ^= 1);
258 				i_next = i + 3600 / PANIC_BLINK_SPD;
259 			}
260 			mdelay(PANIC_TIMER_STEP);
261 		}
262 	}
263 	if (panic_timeout != 0) {
264 		/*
265 		 * This will not be a clean reboot, with everything
266 		 * shutting down.  But if there is a chance of
267 		 * rebooting the system it will be rebooted.
268 		 */
269 		emergency_restart();
270 	}
271 #ifdef __sparc__
272 	{
273 		extern int stop_a_enabled;
274 		/* Make sure the user can actually press Stop-A (L1-A) */
275 		stop_a_enabled = 1;
276 		pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
277 	}
278 #endif
279 #if defined(CONFIG_S390)
280 	{
281 		unsigned long caller;
282 
283 		caller = (unsigned long)__builtin_return_address(0);
284 		disabled_wait(caller);
285 	}
286 #endif
287 	pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
288 	local_irq_enable();
289 	for (i = 0; ; i += PANIC_TIMER_STEP) {
290 		touch_softlockup_watchdog();
291 		if (i >= i_next) {
292 			i += panic_blink(state ^= 1);
293 			i_next = i + 3600 / PANIC_BLINK_SPD;
294 		}
295 		mdelay(PANIC_TIMER_STEP);
296 	}
297 }
298 
299 EXPORT_SYMBOL(panic);
300 
301 /*
302  * TAINT_FORCED_RMMOD could be a per-module flag but the module
303  * is being removed anyway.
304  */
305 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
306 	{ 'P', 'G', true },	/* TAINT_PROPRIETARY_MODULE */
307 	{ 'F', ' ', true },	/* TAINT_FORCED_MODULE */
308 	{ 'S', ' ', false },	/* TAINT_CPU_OUT_OF_SPEC */
309 	{ 'R', ' ', false },	/* TAINT_FORCED_RMMOD */
310 	{ 'M', ' ', false },	/* TAINT_MACHINE_CHECK */
311 	{ 'B', ' ', false },	/* TAINT_BAD_PAGE */
312 	{ 'U', ' ', false },	/* TAINT_USER */
313 	{ 'D', ' ', false },	/* TAINT_DIE */
314 	{ 'A', ' ', false },	/* TAINT_OVERRIDDEN_ACPI_TABLE */
315 	{ 'W', ' ', false },	/* TAINT_WARN */
316 	{ 'C', ' ', true },	/* TAINT_CRAP */
317 	{ 'I', ' ', false },	/* TAINT_FIRMWARE_WORKAROUND */
318 	{ 'O', ' ', true },	/* TAINT_OOT_MODULE */
319 	{ 'E', ' ', true },	/* TAINT_UNSIGNED_MODULE */
320 	{ 'L', ' ', false },	/* TAINT_SOFTLOCKUP */
321 	{ 'K', ' ', true },	/* TAINT_LIVEPATCH */
322 };
323 
324 /**
325  *	print_tainted - return a string to represent the kernel taint state.
326  *
327  *  'P' - Proprietary module has been loaded.
328  *  'F' - Module has been forcibly loaded.
329  *  'S' - SMP with CPUs not designed for SMP.
330  *  'R' - User forced a module unload.
331  *  'M' - System experienced a machine check exception.
332  *  'B' - System has hit bad_page.
333  *  'U' - Userspace-defined naughtiness.
334  *  'D' - Kernel has oopsed before
335  *  'A' - ACPI table overridden.
336  *  'W' - Taint on warning.
337  *  'C' - modules from drivers/staging are loaded.
338  *  'I' - Working around severe firmware bug.
339  *  'O' - Out-of-tree module has been loaded.
340  *  'E' - Unsigned module has been loaded.
341  *  'L' - A soft lockup has previously occurred.
342  *  'K' - Kernel has been live patched.
343  *
344  *	The string is overwritten by the next call to print_tainted().
345  */
346 const char *print_tainted(void)
347 {
348 	static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
349 
350 	if (tainted_mask) {
351 		char *s;
352 		int i;
353 
354 		s = buf + sprintf(buf, "Tainted: ");
355 		for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
356 			const struct taint_flag *t = &taint_flags[i];
357 			*s++ = test_bit(i, &tainted_mask) ?
358 					t->true : t->false;
359 		}
360 		*s = 0;
361 	} else
362 		snprintf(buf, sizeof(buf), "Not tainted");
363 
364 	return buf;
365 }
366 
367 int test_taint(unsigned flag)
368 {
369 	return test_bit(flag, &tainted_mask);
370 }
371 EXPORT_SYMBOL(test_taint);
372 
373 unsigned long get_taint(void)
374 {
375 	return tainted_mask;
376 }
377 
378 /**
379  * add_taint: add a taint flag if not already set.
380  * @flag: one of the TAINT_* constants.
381  * @lockdep_ok: whether lock debugging is still OK.
382  *
383  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
384  * some notewortht-but-not-corrupting cases, it can be set to true.
385  */
386 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
387 {
388 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
389 		pr_warn("Disabling lock debugging due to kernel taint\n");
390 
391 	set_bit(flag, &tainted_mask);
392 }
393 EXPORT_SYMBOL(add_taint);
394 
395 static void spin_msec(int msecs)
396 {
397 	int i;
398 
399 	for (i = 0; i < msecs; i++) {
400 		touch_nmi_watchdog();
401 		mdelay(1);
402 	}
403 }
404 
405 /*
406  * It just happens that oops_enter() and oops_exit() are identically
407  * implemented...
408  */
409 static void do_oops_enter_exit(void)
410 {
411 	unsigned long flags;
412 	static int spin_counter;
413 
414 	if (!pause_on_oops)
415 		return;
416 
417 	spin_lock_irqsave(&pause_on_oops_lock, flags);
418 	if (pause_on_oops_flag == 0) {
419 		/* This CPU may now print the oops message */
420 		pause_on_oops_flag = 1;
421 	} else {
422 		/* We need to stall this CPU */
423 		if (!spin_counter) {
424 			/* This CPU gets to do the counting */
425 			spin_counter = pause_on_oops;
426 			do {
427 				spin_unlock(&pause_on_oops_lock);
428 				spin_msec(MSEC_PER_SEC);
429 				spin_lock(&pause_on_oops_lock);
430 			} while (--spin_counter);
431 			pause_on_oops_flag = 0;
432 		} else {
433 			/* This CPU waits for a different one */
434 			while (spin_counter) {
435 				spin_unlock(&pause_on_oops_lock);
436 				spin_msec(1);
437 				spin_lock(&pause_on_oops_lock);
438 			}
439 		}
440 	}
441 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
442 }
443 
444 /*
445  * Return true if the calling CPU is allowed to print oops-related info.
446  * This is a bit racy..
447  */
448 int oops_may_print(void)
449 {
450 	return pause_on_oops_flag == 0;
451 }
452 
453 /*
454  * Called when the architecture enters its oops handler, before it prints
455  * anything.  If this is the first CPU to oops, and it's oopsing the first
456  * time then let it proceed.
457  *
458  * This is all enabled by the pause_on_oops kernel boot option.  We do all
459  * this to ensure that oopses don't scroll off the screen.  It has the
460  * side-effect of preventing later-oopsing CPUs from mucking up the display,
461  * too.
462  *
463  * It turns out that the CPU which is allowed to print ends up pausing for
464  * the right duration, whereas all the other CPUs pause for twice as long:
465  * once in oops_enter(), once in oops_exit().
466  */
467 void oops_enter(void)
468 {
469 	tracing_off();
470 	/* can't trust the integrity of the kernel anymore: */
471 	debug_locks_off();
472 	do_oops_enter_exit();
473 }
474 
475 /*
476  * 64-bit random ID for oopses:
477  */
478 static u64 oops_id;
479 
480 static int init_oops_id(void)
481 {
482 	if (!oops_id)
483 		get_random_bytes(&oops_id, sizeof(oops_id));
484 	else
485 		oops_id++;
486 
487 	return 0;
488 }
489 late_initcall(init_oops_id);
490 
491 void print_oops_end_marker(void)
492 {
493 	init_oops_id();
494 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
495 }
496 
497 /*
498  * Called when the architecture exits its oops handler, after printing
499  * everything.
500  */
501 void oops_exit(void)
502 {
503 	do_oops_enter_exit();
504 	print_oops_end_marker();
505 	kmsg_dump(KMSG_DUMP_OOPS);
506 }
507 
508 struct warn_args {
509 	const char *fmt;
510 	va_list args;
511 };
512 
513 void __warn(const char *file, int line, void *caller, unsigned taint,
514 	    struct pt_regs *regs, struct warn_args *args)
515 {
516 	disable_trace_on_warning();
517 
518 	pr_warn("------------[ cut here ]------------\n");
519 
520 	if (file)
521 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
522 			raw_smp_processor_id(), current->pid, file, line,
523 			caller);
524 	else
525 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
526 			raw_smp_processor_id(), current->pid, caller);
527 
528 	if (args)
529 		vprintk(args->fmt, args->args);
530 
531 	if (panic_on_warn) {
532 		/*
533 		 * This thread may hit another WARN() in the panic path.
534 		 * Resetting this prevents additional WARN() from panicking the
535 		 * system on this thread.  Other threads are blocked by the
536 		 * panic_mutex in panic().
537 		 */
538 		panic_on_warn = 0;
539 		panic("panic_on_warn set ...\n");
540 	}
541 
542 	print_modules();
543 
544 	if (regs)
545 		show_regs(regs);
546 	else
547 		dump_stack();
548 
549 	print_oops_end_marker();
550 
551 	/* Just a warning, don't kill lockdep. */
552 	add_taint(taint, LOCKDEP_STILL_OK);
553 }
554 
555 #ifdef WANT_WARN_ON_SLOWPATH
556 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
557 {
558 	struct warn_args args;
559 
560 	args.fmt = fmt;
561 	va_start(args.args, fmt);
562 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
563 	       &args);
564 	va_end(args.args);
565 }
566 EXPORT_SYMBOL(warn_slowpath_fmt);
567 
568 void warn_slowpath_fmt_taint(const char *file, int line,
569 			     unsigned taint, const char *fmt, ...)
570 {
571 	struct warn_args args;
572 
573 	args.fmt = fmt;
574 	va_start(args.args, fmt);
575 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
576 	va_end(args.args);
577 }
578 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
579 
580 void warn_slowpath_null(const char *file, int line)
581 {
582 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
583 }
584 EXPORT_SYMBOL(warn_slowpath_null);
585 #endif
586 
587 #ifdef CONFIG_CC_STACKPROTECTOR
588 
589 /*
590  * Called when gcc's -fstack-protector feature is used, and
591  * gcc detects corruption of the on-stack canary value
592  */
593 __visible void __stack_chk_fail(void)
594 {
595 	panic("stack-protector: Kernel stack is corrupted in: %p\n",
596 		__builtin_return_address(0));
597 }
598 EXPORT_SYMBOL(__stack_chk_fail);
599 
600 #endif
601 
602 core_param(panic, panic_timeout, int, 0644);
603 core_param(pause_on_oops, pause_on_oops, int, 0644);
604 core_param(panic_on_warn, panic_on_warn, int, 0644);
605 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
606 
607 static int __init oops_setup(char *s)
608 {
609 	if (!s)
610 		return -EINVAL;
611 	if (!strcmp(s, "panic"))
612 		panic_on_oops = 1;
613 	return 0;
614 }
615 early_param("oops", oops_setup);
616