xref: /openbmc/linux/kernel/panic.c (revision 41e4b7dc)
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/sched/debug.h>
13 #include <linux/interrupt.h>
14 #include <linux/kmsg_dump.h>
15 #include <linux/kallsyms.h>
16 #include <linux/notifier.h>
17 #include <linux/module.h>
18 #include <linux/random.h>
19 #include <linux/ftrace.h>
20 #include <linux/reboot.h>
21 #include <linux/delay.h>
22 #include <linux/kexec.h>
23 #include <linux/sched.h>
24 #include <linux/sysrq.h>
25 #include <linux/init.h>
26 #include <linux/nmi.h>
27 #include <linux/console.h>
28 #include <linux/bug.h>
29 #include <linux/ratelimit.h>
30 
31 #define PANIC_TIMER_STEP 100
32 #define PANIC_BLINK_SPD 18
33 
34 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
35 static unsigned long tainted_mask;
36 static int pause_on_oops;
37 static int pause_on_oops_flag;
38 static DEFINE_SPINLOCK(pause_on_oops_lock);
39 bool crash_kexec_post_notifiers;
40 int panic_on_warn __read_mostly;
41 
42 int panic_timeout = CONFIG_PANIC_TIMEOUT;
43 EXPORT_SYMBOL_GPL(panic_timeout);
44 
45 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
46 
47 EXPORT_SYMBOL(panic_notifier_list);
48 
49 static long no_blink(int state)
50 {
51 	return 0;
52 }
53 
54 /* Returns how long it waited in ms */
55 long (*panic_blink)(int state);
56 EXPORT_SYMBOL(panic_blink);
57 
58 /*
59  * Stop ourself in panic -- architecture code may override this
60  */
61 void __weak panic_smp_self_stop(void)
62 {
63 	while (1)
64 		cpu_relax();
65 }
66 
67 /*
68  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
69  * may override this to prepare for crash dumping, e.g. save regs info.
70  */
71 void __weak nmi_panic_self_stop(struct pt_regs *regs)
72 {
73 	panic_smp_self_stop();
74 }
75 
76 /*
77  * Stop other CPUs in panic.  Architecture dependent code may override this
78  * with more suitable version.  For example, if the architecture supports
79  * crash dump, it should save registers of each stopped CPU and disable
80  * per-CPU features such as virtualization extensions.
81  */
82 void __weak crash_smp_send_stop(void)
83 {
84 	static int cpus_stopped;
85 
86 	/*
87 	 * This function can be called twice in panic path, but obviously
88 	 * we execute this only once.
89 	 */
90 	if (cpus_stopped)
91 		return;
92 
93 	/*
94 	 * Note smp_send_stop is the usual smp shutdown function, which
95 	 * unfortunately means it may not be hardened to work in a panic
96 	 * situation.
97 	 */
98 	smp_send_stop();
99 	cpus_stopped = 1;
100 }
101 
102 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
103 
104 /*
105  * A variant of panic() called from NMI context. We return if we've already
106  * panicked on this CPU. If another CPU already panicked, loop in
107  * nmi_panic_self_stop() which can provide architecture dependent code such
108  * as saving register state for crash dump.
109  */
110 void nmi_panic(struct pt_regs *regs, const char *msg)
111 {
112 	int old_cpu, cpu;
113 
114 	cpu = raw_smp_processor_id();
115 	old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
116 
117 	if (old_cpu == PANIC_CPU_INVALID)
118 		panic("%s", msg);
119 	else if (old_cpu != cpu)
120 		nmi_panic_self_stop(regs);
121 }
122 EXPORT_SYMBOL(nmi_panic);
123 
124 /**
125  *	panic - halt the system
126  *	@fmt: The text string to print
127  *
128  *	Display a message, then perform cleanups.
129  *
130  *	This function never returns.
131  */
132 void panic(const char *fmt, ...)
133 {
134 	static char buf[1024];
135 	va_list args;
136 	long i, i_next = 0;
137 	int state = 0;
138 	int old_cpu, this_cpu;
139 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
140 
141 	/*
142 	 * Disable local interrupts. This will prevent panic_smp_self_stop
143 	 * from deadlocking the first cpu that invokes the panic, since
144 	 * there is nothing to prevent an interrupt handler (that runs
145 	 * after setting panic_cpu) from invoking panic() again.
146 	 */
147 	local_irq_disable();
148 
149 	/*
150 	 * It's possible to come here directly from a panic-assertion and
151 	 * not have preempt disabled. Some functions called from here want
152 	 * preempt to be disabled. No point enabling it later though...
153 	 *
154 	 * Only one CPU is allowed to execute the panic code from here. For
155 	 * multiple parallel invocations of panic, all other CPUs either
156 	 * stop themself or will wait until they are stopped by the 1st CPU
157 	 * with smp_send_stop().
158 	 *
159 	 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
160 	 * comes here, so go ahead.
161 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
162 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
163 	 */
164 	this_cpu = raw_smp_processor_id();
165 	old_cpu  = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
166 
167 	if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
168 		panic_smp_self_stop();
169 
170 	console_verbose();
171 	bust_spinlocks(1);
172 	va_start(args, fmt);
173 	vsnprintf(buf, sizeof(buf), fmt, args);
174 	va_end(args);
175 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
176 #ifdef CONFIG_DEBUG_BUGVERBOSE
177 	/*
178 	 * Avoid nested stack-dumping if a panic occurs during oops processing
179 	 */
180 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
181 		dump_stack();
182 #endif
183 
184 	/*
185 	 * If we have crashed and we have a crash kernel loaded let it handle
186 	 * everything else.
187 	 * If we want to run this after calling panic_notifiers, pass
188 	 * the "crash_kexec_post_notifiers" option to the kernel.
189 	 *
190 	 * Bypass the panic_cpu check and call __crash_kexec directly.
191 	 */
192 	if (!_crash_kexec_post_notifiers) {
193 		printk_safe_flush_on_panic();
194 		__crash_kexec(NULL);
195 
196 		/*
197 		 * Note smp_send_stop is the usual smp shutdown function, which
198 		 * unfortunately means it may not be hardened to work in a
199 		 * panic situation.
200 		 */
201 		smp_send_stop();
202 	} else {
203 		/*
204 		 * If we want to do crash dump after notifier calls and
205 		 * kmsg_dump, we will need architecture dependent extra
206 		 * works in addition to stopping other CPUs.
207 		 */
208 		crash_smp_send_stop();
209 	}
210 
211 	/*
212 	 * Run any panic handlers, including those that might need to
213 	 * add information to the kmsg dump output.
214 	 */
215 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
216 
217 	/* Call flush even twice. It tries harder with a single online CPU */
218 	printk_safe_flush_on_panic();
219 	kmsg_dump(KMSG_DUMP_PANIC);
220 
221 	/*
222 	 * If you doubt kdump always works fine in any situation,
223 	 * "crash_kexec_post_notifiers" offers you a chance to run
224 	 * panic_notifiers and dumping kmsg before kdump.
225 	 * Note: since some panic_notifiers can make crashed kernel
226 	 * more unstable, it can increase risks of the kdump failure too.
227 	 *
228 	 * Bypass the panic_cpu check and call __crash_kexec directly.
229 	 */
230 	if (_crash_kexec_post_notifiers)
231 		__crash_kexec(NULL);
232 
233 	bust_spinlocks(0);
234 
235 	/*
236 	 * We may have ended up stopping the CPU holding the lock (in
237 	 * smp_send_stop()) while still having some valuable data in the console
238 	 * buffer.  Try to acquire the lock then release it regardless of the
239 	 * result.  The release will also print the buffers out.  Locks debug
240 	 * should be disabled to avoid reporting bad unlock balance when
241 	 * panic() is not being callled from OOPS.
242 	 */
243 	debug_locks_off();
244 	console_flush_on_panic();
245 
246 	if (!panic_blink)
247 		panic_blink = no_blink;
248 
249 	if (panic_timeout > 0) {
250 		/*
251 		 * Delay timeout seconds before rebooting the machine.
252 		 * We can't use the "normal" timers since we just panicked.
253 		 */
254 		pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
255 
256 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
257 			touch_nmi_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 	if (panic_timeout != 0) {
266 		/*
267 		 * This will not be a clean reboot, with everything
268 		 * shutting down.  But if there is a chance of
269 		 * rebooting the system it will be rebooted.
270 		 */
271 		emergency_restart();
272 	}
273 #ifdef __sparc__
274 	{
275 		extern int stop_a_enabled;
276 		/* Make sure the user can actually press Stop-A (L1-A) */
277 		stop_a_enabled = 1;
278 		pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
279 			 "twice on console to return to the boot prom\n");
280 	}
281 #endif
282 #if defined(CONFIG_S390)
283 	{
284 		unsigned long caller;
285 
286 		caller = (unsigned long)__builtin_return_address(0);
287 		disabled_wait(caller);
288 	}
289 #endif
290 	pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
291 	local_irq_enable();
292 	for (i = 0; ; i += PANIC_TIMER_STEP) {
293 		touch_softlockup_watchdog();
294 		if (i >= i_next) {
295 			i += panic_blink(state ^= 1);
296 			i_next = i + 3600 / PANIC_BLINK_SPD;
297 		}
298 		mdelay(PANIC_TIMER_STEP);
299 	}
300 }
301 
302 EXPORT_SYMBOL(panic);
303 
304 /*
305  * TAINT_FORCED_RMMOD could be a per-module flag but the module
306  * is being removed anyway.
307  */
308 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
309 	{ 'P', 'G', true },	/* TAINT_PROPRIETARY_MODULE */
310 	{ 'F', ' ', true },	/* TAINT_FORCED_MODULE */
311 	{ 'S', ' ', false },	/* TAINT_CPU_OUT_OF_SPEC */
312 	{ 'R', ' ', false },	/* TAINT_FORCED_RMMOD */
313 	{ 'M', ' ', false },	/* TAINT_MACHINE_CHECK */
314 	{ 'B', ' ', false },	/* TAINT_BAD_PAGE */
315 	{ 'U', ' ', false },	/* TAINT_USER */
316 	{ 'D', ' ', false },	/* TAINT_DIE */
317 	{ 'A', ' ', false },	/* TAINT_OVERRIDDEN_ACPI_TABLE */
318 	{ 'W', ' ', false },	/* TAINT_WARN */
319 	{ 'C', ' ', true },	/* TAINT_CRAP */
320 	{ 'I', ' ', false },	/* TAINT_FIRMWARE_WORKAROUND */
321 	{ 'O', ' ', true },	/* TAINT_OOT_MODULE */
322 	{ 'E', ' ', true },	/* TAINT_UNSIGNED_MODULE */
323 	{ 'L', ' ', false },	/* TAINT_SOFTLOCKUP */
324 	{ 'K', ' ', true },	/* TAINT_LIVEPATCH */
325 };
326 
327 /**
328  *	print_tainted - return a string to represent the kernel taint state.
329  *
330  *  'P' - Proprietary module has been loaded.
331  *  'F' - Module has been forcibly loaded.
332  *  'S' - SMP with CPUs not designed for SMP.
333  *  'R' - User forced a module unload.
334  *  'M' - System experienced a machine check exception.
335  *  'B' - System has hit bad_page.
336  *  'U' - Userspace-defined naughtiness.
337  *  'D' - Kernel has oopsed before
338  *  'A' - ACPI table overridden.
339  *  'W' - Taint on warning.
340  *  'C' - modules from drivers/staging are loaded.
341  *  'I' - Working around severe firmware bug.
342  *  'O' - Out-of-tree module has been loaded.
343  *  'E' - Unsigned module has been loaded.
344  *  'L' - A soft lockup has previously occurred.
345  *  'K' - Kernel has been live patched.
346  *
347  *	The string is overwritten by the next call to print_tainted().
348  */
349 const char *print_tainted(void)
350 {
351 	static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
352 
353 	if (tainted_mask) {
354 		char *s;
355 		int i;
356 
357 		s = buf + sprintf(buf, "Tainted: ");
358 		for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
359 			const struct taint_flag *t = &taint_flags[i];
360 			*s++ = test_bit(i, &tainted_mask) ?
361 					t->c_true : t->c_false;
362 		}
363 		*s = 0;
364 	} else
365 		snprintf(buf, sizeof(buf), "Not tainted");
366 
367 	return buf;
368 }
369 
370 int test_taint(unsigned flag)
371 {
372 	return test_bit(flag, &tainted_mask);
373 }
374 EXPORT_SYMBOL(test_taint);
375 
376 unsigned long get_taint(void)
377 {
378 	return tainted_mask;
379 }
380 
381 /**
382  * add_taint: add a taint flag if not already set.
383  * @flag: one of the TAINT_* constants.
384  * @lockdep_ok: whether lock debugging is still OK.
385  *
386  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
387  * some notewortht-but-not-corrupting cases, it can be set to true.
388  */
389 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
390 {
391 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
392 		pr_warn("Disabling lock debugging due to kernel taint\n");
393 
394 	set_bit(flag, &tainted_mask);
395 }
396 EXPORT_SYMBOL(add_taint);
397 
398 static void spin_msec(int msecs)
399 {
400 	int i;
401 
402 	for (i = 0; i < msecs; i++) {
403 		touch_nmi_watchdog();
404 		mdelay(1);
405 	}
406 }
407 
408 /*
409  * It just happens that oops_enter() and oops_exit() are identically
410  * implemented...
411  */
412 static void do_oops_enter_exit(void)
413 {
414 	unsigned long flags;
415 	static int spin_counter;
416 
417 	if (!pause_on_oops)
418 		return;
419 
420 	spin_lock_irqsave(&pause_on_oops_lock, flags);
421 	if (pause_on_oops_flag == 0) {
422 		/* This CPU may now print the oops message */
423 		pause_on_oops_flag = 1;
424 	} else {
425 		/* We need to stall this CPU */
426 		if (!spin_counter) {
427 			/* This CPU gets to do the counting */
428 			spin_counter = pause_on_oops;
429 			do {
430 				spin_unlock(&pause_on_oops_lock);
431 				spin_msec(MSEC_PER_SEC);
432 				spin_lock(&pause_on_oops_lock);
433 			} while (--spin_counter);
434 			pause_on_oops_flag = 0;
435 		} else {
436 			/* This CPU waits for a different one */
437 			while (spin_counter) {
438 				spin_unlock(&pause_on_oops_lock);
439 				spin_msec(1);
440 				spin_lock(&pause_on_oops_lock);
441 			}
442 		}
443 	}
444 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
445 }
446 
447 /*
448  * Return true if the calling CPU is allowed to print oops-related info.
449  * This is a bit racy..
450  */
451 int oops_may_print(void)
452 {
453 	return pause_on_oops_flag == 0;
454 }
455 
456 /*
457  * Called when the architecture enters its oops handler, before it prints
458  * anything.  If this is the first CPU to oops, and it's oopsing the first
459  * time then let it proceed.
460  *
461  * This is all enabled by the pause_on_oops kernel boot option.  We do all
462  * this to ensure that oopses don't scroll off the screen.  It has the
463  * side-effect of preventing later-oopsing CPUs from mucking up the display,
464  * too.
465  *
466  * It turns out that the CPU which is allowed to print ends up pausing for
467  * the right duration, whereas all the other CPUs pause for twice as long:
468  * once in oops_enter(), once in oops_exit().
469  */
470 void oops_enter(void)
471 {
472 	tracing_off();
473 	/* can't trust the integrity of the kernel anymore: */
474 	debug_locks_off();
475 	do_oops_enter_exit();
476 }
477 
478 /*
479  * 64-bit random ID for oopses:
480  */
481 static u64 oops_id;
482 
483 static int init_oops_id(void)
484 {
485 	if (!oops_id)
486 		get_random_bytes(&oops_id, sizeof(oops_id));
487 	else
488 		oops_id++;
489 
490 	return 0;
491 }
492 late_initcall(init_oops_id);
493 
494 void print_oops_end_marker(void)
495 {
496 	init_oops_id();
497 	pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
498 }
499 
500 /*
501  * Called when the architecture exits its oops handler, after printing
502  * everything.
503  */
504 void oops_exit(void)
505 {
506 	do_oops_enter_exit();
507 	print_oops_end_marker();
508 	kmsg_dump(KMSG_DUMP_OOPS);
509 }
510 
511 struct warn_args {
512 	const char *fmt;
513 	va_list args;
514 };
515 
516 void __warn(const char *file, int line, void *caller, unsigned taint,
517 	    struct pt_regs *regs, struct warn_args *args)
518 {
519 	disable_trace_on_warning();
520 
521 	pr_warn("------------[ cut here ]------------\n");
522 
523 	if (file)
524 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
525 			raw_smp_processor_id(), current->pid, file, line,
526 			caller);
527 	else
528 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
529 			raw_smp_processor_id(), current->pid, caller);
530 
531 	if (args)
532 		vprintk(args->fmt, args->args);
533 
534 	if (panic_on_warn) {
535 		/*
536 		 * This thread may hit another WARN() in the panic path.
537 		 * Resetting this prevents additional WARN() from panicking the
538 		 * system on this thread.  Other threads are blocked by the
539 		 * panic_mutex in panic().
540 		 */
541 		panic_on_warn = 0;
542 		panic("panic_on_warn set ...\n");
543 	}
544 
545 	print_modules();
546 
547 	if (regs)
548 		show_regs(regs);
549 	else
550 		dump_stack();
551 
552 	print_oops_end_marker();
553 
554 	/* Just a warning, don't kill lockdep. */
555 	add_taint(taint, LOCKDEP_STILL_OK);
556 }
557 
558 #ifdef WANT_WARN_ON_SLOWPATH
559 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
560 {
561 	struct warn_args args;
562 
563 	args.fmt = fmt;
564 	va_start(args.args, fmt);
565 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
566 	       &args);
567 	va_end(args.args);
568 }
569 EXPORT_SYMBOL(warn_slowpath_fmt);
570 
571 void warn_slowpath_fmt_taint(const char *file, int line,
572 			     unsigned taint, const char *fmt, ...)
573 {
574 	struct warn_args args;
575 
576 	args.fmt = fmt;
577 	va_start(args.args, fmt);
578 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
579 	va_end(args.args);
580 }
581 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
582 
583 void warn_slowpath_null(const char *file, int line)
584 {
585 	__warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
586 }
587 EXPORT_SYMBOL(warn_slowpath_null);
588 #endif
589 
590 #ifdef CONFIG_CC_STACKPROTECTOR
591 
592 /*
593  * Called when gcc's -fstack-protector feature is used, and
594  * gcc detects corruption of the on-stack canary value
595  */
596 __visible void __stack_chk_fail(void)
597 {
598 	panic("stack-protector: Kernel stack is corrupted in: %p\n",
599 		__builtin_return_address(0));
600 }
601 EXPORT_SYMBOL(__stack_chk_fail);
602 
603 #endif
604 
605 #ifdef CONFIG_ARCH_HAS_REFCOUNT
606 void refcount_error_report(struct pt_regs *regs, const char *err)
607 {
608 	WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
609 		err, (void *)instruction_pointer(regs),
610 		current->comm, task_pid_nr(current),
611 		from_kuid_munged(&init_user_ns, current_uid()),
612 		from_kuid_munged(&init_user_ns, current_euid()));
613 }
614 #endif
615 
616 core_param(panic, panic_timeout, int, 0644);
617 core_param(pause_on_oops, pause_on_oops, int, 0644);
618 core_param(panic_on_warn, panic_on_warn, int, 0644);
619 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
620 
621 static int __init oops_setup(char *s)
622 {
623 	if (!s)
624 		return -EINVAL;
625 	if (!strcmp(s, "panic"))
626 		panic_on_oops = 1;
627 	return 0;
628 }
629 early_param("oops", oops_setup);
630