xref: /linux/kernel/panic.c (revision 056e065a6b6e01ab54bb9770c0d5a15350e571e2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/panic.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 /*
9  * This function is used through-out the kernel (including mm and fs)
10  * to indicate a major problem.
11  */
12 #include <linux/debug_locks.h>
13 #include <linux/sched/debug.h>
14 #include <linux/interrupt.h>
15 #include <linux/kgdb.h>
16 #include <linux/kmsg_dump.h>
17 #include <linux/kallsyms.h>
18 #include <linux/notifier.h>
19 #include <linux/vt_kern.h>
20 #include <linux/module.h>
21 #include <linux/random.h>
22 #include <linux/ftrace.h>
23 #include <linux/reboot.h>
24 #include <linux/delay.h>
25 #include <linux/kexec.h>
26 #include <linux/panic_notifier.h>
27 #include <linux/sched.h>
28 #include <linux/string_helpers.h>
29 #include <linux/sysrq.h>
30 #include <linux/init.h>
31 #include <linux/nmi.h>
32 #include <linux/console.h>
33 #include <linux/bug.h>
34 #include <linux/ratelimit.h>
35 #include <linux/debugfs.h>
36 #include <linux/sysfs.h>
37 #include <linux/context_tracking.h>
38 #include <linux/seq_buf.h>
39 #include <linux/sys_info.h>
40 #include <trace/events/error_report.h>
41 #include <asm/sections.h>
42 #include <kunit/test-bug.h>
43 
44 #define PANIC_TIMER_STEP 100
45 #define PANIC_BLINK_SPD 18
46 #define PANIC_MSG_BUFSZ 1024
47 
48 #ifdef CONFIG_SMP
49 /*
50  * Should we dump all CPUs backtraces in an oops event?
51  * Defaults to 0, can be changed via sysctl.
52  */
53 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
54 #else
55 #define sysctl_oops_all_cpu_backtrace 0
56 #endif /* CONFIG_SMP */
57 
58 int panic_on_oops = IS_ENABLED(CONFIG_PANIC_ON_OOPS);
59 static unsigned long tainted_mask =
60 	IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
61 static int pause_on_oops;
62 static int pause_on_oops_flag;
63 static DEFINE_SPINLOCK(pause_on_oops_lock);
64 bool crash_kexec_post_notifiers;
65 int panic_on_warn __read_mostly;
66 unsigned long panic_on_taint;
67 bool panic_on_taint_nousertaint = false;
68 static unsigned int warn_limit __read_mostly;
69 static bool panic_console_replay;
70 
71 bool panic_triggering_all_cpu_backtrace;
72 static bool panic_this_cpu_backtrace_printed;
73 
74 int panic_timeout = CONFIG_PANIC_TIMEOUT;
75 EXPORT_SYMBOL_GPL(panic_timeout);
76 
77 unsigned long panic_print;
78 
79 static int panic_force_cpu = -1;
80 
81 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
82 
83 EXPORT_SYMBOL(panic_notifier_list);
84 
85 static void panic_print_deprecated(void)
86 {
87 	pr_info_once("Kernel: The 'panic_print' parameter is now deprecated. Please use 'panic_sys_info' and 'panic_console_replay' instead.\n");
88 }
89 
90 #ifdef CONFIG_SYSCTL
91 
92 /*
93  * Taint values can only be increased
94  * This means we can safely use a temporary.
95  */
96 static int proc_taint(const struct ctl_table *table, int write,
97 			       void *buffer, size_t *lenp, loff_t *ppos)
98 {
99 	struct ctl_table t;
100 	unsigned long tmptaint = get_taint();
101 	int err;
102 
103 	if (write && !capable(CAP_SYS_ADMIN))
104 		return -EPERM;
105 
106 	t = *table;
107 	t.data = &tmptaint;
108 	err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
109 	if (err < 0)
110 		return err;
111 
112 	if (write) {
113 		int i;
114 
115 		/*
116 		 * If we are relying on panic_on_taint not producing
117 		 * false positives due to userspace input, bail out
118 		 * before setting the requested taint flags.
119 		 */
120 		if (panic_on_taint_nousertaint && (tmptaint & panic_on_taint))
121 			return -EINVAL;
122 
123 		/*
124 		 * Poor man's atomic or. Not worth adding a primitive
125 		 * to everyone's atomic.h for this
126 		 */
127 		for (i = 0; i < TAINT_FLAGS_COUNT; i++)
128 			if ((1UL << i) & tmptaint)
129 				add_taint(i, LOCKDEP_STILL_OK);
130 	}
131 
132 	return err;
133 }
134 
135 static int sysctl_panic_print_handler(const struct ctl_table *table, int write,
136 			   void *buffer, size_t *lenp, loff_t *ppos)
137 {
138 	if (write)
139 		panic_print_deprecated();
140 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
141 }
142 
143 static const struct ctl_table kern_panic_table[] = {
144 #ifdef CONFIG_SMP
145 	{
146 		.procname       = "oops_all_cpu_backtrace",
147 		.data           = &sysctl_oops_all_cpu_backtrace,
148 		.maxlen         = sizeof(int),
149 		.mode           = 0644,
150 		.proc_handler   = proc_dointvec_minmax,
151 		.extra1         = SYSCTL_ZERO,
152 		.extra2         = SYSCTL_ONE,
153 	},
154 #endif
155 	{
156 		.procname	= "tainted",
157 		.maxlen		= sizeof(long),
158 		.mode		= 0644,
159 		.proc_handler	= proc_taint,
160 	},
161 	{
162 		.procname	= "panic",
163 		.data		= &panic_timeout,
164 		.maxlen		= sizeof(int),
165 		.mode		= 0644,
166 		.proc_handler	= proc_dointvec,
167 	},
168 	{
169 		.procname	= "panic_on_oops",
170 		.data		= &panic_on_oops,
171 		.maxlen		= sizeof(int),
172 		.mode		= 0644,
173 		.proc_handler	= proc_dointvec,
174 	},
175 	{
176 		.procname	= "panic_print",
177 		.data		= &panic_print,
178 		.maxlen		= sizeof(unsigned long),
179 		.mode		= 0644,
180 		.proc_handler	= sysctl_panic_print_handler,
181 	},
182 	{
183 		.procname	= "panic_on_warn",
184 		.data		= &panic_on_warn,
185 		.maxlen		= sizeof(int),
186 		.mode		= 0644,
187 		.proc_handler	= proc_dointvec_minmax,
188 		.extra1		= SYSCTL_ZERO,
189 		.extra2		= SYSCTL_ONE,
190 	},
191 	{
192 		.procname       = "warn_limit",
193 		.data           = &warn_limit,
194 		.maxlen         = sizeof(warn_limit),
195 		.mode           = 0644,
196 		.proc_handler   = proc_douintvec,
197 	},
198 #if (defined(CONFIG_X86_32) || defined(CONFIG_PARISC)) && \
199 	defined(CONFIG_DEBUG_STACKOVERFLOW)
200 	{
201 		.procname	= "panic_on_stackoverflow",
202 		.data		= &sysctl_panic_on_stackoverflow,
203 		.maxlen		= sizeof(int),
204 		.mode		= 0644,
205 		.proc_handler	= proc_dointvec,
206 	},
207 #endif
208 	{
209 		.procname	= "panic_sys_info",
210 		.data		= &panic_print,
211 		.maxlen         = sizeof(panic_print),
212 		.mode		= 0644,
213 		.proc_handler	= sysctl_sys_info_handler,
214 	},
215 };
216 
217 static __init int kernel_panic_sysctls_init(void)
218 {
219 	register_sysctl_init("kernel", kern_panic_table);
220 	return 0;
221 }
222 late_initcall(kernel_panic_sysctls_init);
223 #endif
224 
225 /* The format is "panic_sys_info=tasks,mem,locks,ftrace,..." */
226 static int __init setup_panic_sys_info(char *buf)
227 {
228 	/* There is no risk of race in kernel boot phase */
229 	panic_print = sys_info_parse_param(buf);
230 	return 1;
231 }
232 __setup("panic_sys_info=", setup_panic_sys_info);
233 
234 static atomic_t warn_count = ATOMIC_INIT(0);
235 
236 #ifdef CONFIG_SYSFS
237 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
238 			       char *page)
239 {
240 	return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
241 }
242 
243 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
244 
245 static __init int kernel_panic_sysfs_init(void)
246 {
247 	sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
248 	return 0;
249 }
250 late_initcall(kernel_panic_sysfs_init);
251 #endif
252 
253 static long no_blink(int state)
254 {
255 	return 0;
256 }
257 
258 /* Returns how long it waited in ms */
259 long (*panic_blink)(int state);
260 EXPORT_SYMBOL(panic_blink);
261 
262 /*
263  * Stop ourself in panic -- architecture code may override this
264  */
265 void __weak __noreturn panic_smp_self_stop(void)
266 {
267 	while (1)
268 		cpu_relax();
269 }
270 
271 /*
272  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
273  * may override this to prepare for crash dumping, e.g. save regs info.
274  */
275 void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
276 {
277 	panic_smp_self_stop();
278 }
279 
280 /*
281  * Stop other CPUs in panic.  Architecture dependent code may override this
282  * with more suitable version.  For example, if the architecture supports
283  * crash dump, it should save registers of each stopped CPU and disable
284  * per-CPU features such as virtualization extensions.
285  */
286 void __weak crash_smp_send_stop(void)
287 {
288 	static int cpus_stopped;
289 
290 	/*
291 	 * This function can be called twice in panic path, but obviously
292 	 * we execute this only once.
293 	 */
294 	if (cpus_stopped)
295 		return;
296 
297 	/*
298 	 * Note smp_send_stop is the usual smp shutdown function, which
299 	 * unfortunately means it may not be hardened to work in a panic
300 	 * situation.
301 	 */
302 	smp_send_stop();
303 	cpus_stopped = 1;
304 }
305 
306 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
307 atomic_t panic_redirect_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
308 
309 #if defined(CONFIG_SMP) && defined(CONFIG_CRASH_DUMP)
310 static char *panic_force_buf;
311 
312 static int __init panic_force_cpu_setup(char *str)
313 {
314 	int cpu;
315 
316 	if (!str)
317 		return -EINVAL;
318 
319 	if (kstrtoint(str, 0, &cpu) || cpu < 0 || cpu >= nr_cpu_ids) {
320 		pr_warn("panic_force_cpu: invalid value '%s'\n", str);
321 		return -EINVAL;
322 	}
323 
324 	panic_force_cpu = cpu;
325 	return 0;
326 }
327 early_param("panic_force_cpu", panic_force_cpu_setup);
328 
329 static int __init panic_force_cpu_late_init(void)
330 {
331 	if (panic_force_cpu < 0)
332 		return 0;
333 
334 	panic_force_buf = kmalloc(PANIC_MSG_BUFSZ, GFP_KERNEL);
335 
336 	return 0;
337 }
338 late_initcall(panic_force_cpu_late_init);
339 
340 static void do_panic_on_target_cpu(void *info)
341 {
342 	panic("%s", (char *)info);
343 }
344 
345 /**
346  * panic_smp_redirect_cpu - Redirect panic to target CPU
347  * @target_cpu: CPU that should handle the panic
348  * @msg: formatted panic message
349  *
350  * Default implementation uses IPI. Architectures with NMI support
351  * can override this for more reliable delivery.
352  *
353  * Return: 0 on success, negative errno on failure
354  */
355 int __weak panic_smp_redirect_cpu(int target_cpu, void *msg)
356 {
357 	static call_single_data_t panic_csd;
358 
359 	panic_csd.func = do_panic_on_target_cpu;
360 	panic_csd.info = msg;
361 
362 	return smp_call_function_single_async(target_cpu, &panic_csd);
363 }
364 
365 /**
366  * panic_try_force_cpu - Redirect panic to a specific CPU for crash kernel
367  * @fmt: panic message format string
368  * @args: arguments for format string
369  *
370  * Some platforms require panic handling to occur on a specific CPU
371  * for the crash kernel to function correctly. This function redirects
372  * panic handling to the CPU specified via the panic_force_cpu= boot parameter.
373  *
374  * Returns false if panic should proceed on current CPU.
375  * Returns true if panic was redirected.
376  */
377 __printf(1, 0)
378 static bool panic_try_force_cpu(const char *fmt, va_list args)
379 {
380 	int this_cpu = raw_smp_processor_id();
381 	int old_cpu = PANIC_CPU_INVALID;
382 	const char *msg;
383 
384 	/* Feature not enabled via boot parameter */
385 	if (panic_force_cpu < 0)
386 		return false;
387 
388 	/* Already on target CPU - proceed normally */
389 	if (this_cpu == panic_force_cpu)
390 		return false;
391 
392 	/* Target CPU is offline, can't redirect */
393 	if (!cpu_online(panic_force_cpu)) {
394 		pr_warn("panic: target CPU %d is offline, continuing on CPU %d\n",
395 			panic_force_cpu, this_cpu);
396 		return false;
397 	}
398 
399 	/* Another panic already in progress */
400 	if (panic_in_progress())
401 		return false;
402 
403 	/*
404 	 * Only one CPU can do the redirect. Use atomic cmpxchg to ensure
405 	 * we don't race with another CPU also trying to redirect.
406 	 */
407 	if (!atomic_try_cmpxchg(&panic_redirect_cpu, &old_cpu, this_cpu))
408 		return false;
409 
410 	/*
411 	 * Use dynamically allocated buffer if available, otherwise
412 	 * fall back to static message for early boot panics or allocation failure.
413 	 */
414 	if (panic_force_buf) {
415 		vsnprintf(panic_force_buf, PANIC_MSG_BUFSZ, fmt, args);
416 		msg = panic_force_buf;
417 	} else {
418 		msg = "Redirected panic (buffer unavailable)";
419 	}
420 
421 	console_verbose();
422 	bust_spinlocks(1);
423 
424 	pr_emerg("panic: Redirecting from CPU %d to CPU %d for crash kernel.\n",
425 		 this_cpu, panic_force_cpu);
426 
427 	/* Dump original CPU before redirecting */
428 	if (!test_taint(TAINT_DIE) &&
429 	    oops_in_progress <= 1 &&
430 	    IS_ENABLED(CONFIG_DEBUG_BUGVERBOSE)) {
431 		dump_stack();
432 	}
433 
434 	if (panic_smp_redirect_cpu(panic_force_cpu, (void *)msg) != 0) {
435 		atomic_set(&panic_redirect_cpu, PANIC_CPU_INVALID);
436 		pr_warn("panic: failed to redirect to CPU %d, continuing on CPU %d\n",
437 			panic_force_cpu, this_cpu);
438 		return false;
439 	}
440 
441 	/* IPI/NMI sent, this CPU should stop */
442 	return true;
443 }
444 #else
445 __printf(1, 0)
446 static inline bool panic_try_force_cpu(const char *fmt, va_list args)
447 {
448 	return false;
449 }
450 #endif /* CONFIG_SMP && CONFIG_CRASH_DUMP */
451 
452 bool panic_try_start(void)
453 {
454 	int old_cpu, this_cpu;
455 
456 	/*
457 	 * Only one CPU is allowed to execute the crash_kexec() code as with
458 	 * panic().  Otherwise parallel calls of panic() and crash_kexec()
459 	 * may stop each other.  To exclude them, we use panic_cpu here too.
460 	 */
461 	old_cpu = PANIC_CPU_INVALID;
462 	this_cpu = raw_smp_processor_id();
463 
464 	return atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu);
465 }
466 EXPORT_SYMBOL(panic_try_start);
467 
468 void panic_reset(void)
469 {
470 	atomic_set(&panic_cpu, PANIC_CPU_INVALID);
471 }
472 EXPORT_SYMBOL(panic_reset);
473 
474 bool panic_in_progress(void)
475 {
476 	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
477 }
478 EXPORT_SYMBOL(panic_in_progress);
479 
480 /* Return true if a panic is in progress on the current CPU. */
481 bool panic_on_this_cpu(void)
482 {
483 	/*
484 	 * We can use raw_smp_processor_id() here because it is impossible for
485 	 * the task to be migrated to the panic_cpu, or away from it. If
486 	 * panic_cpu has already been set, and we're not currently executing on
487 	 * that CPU, then we never will be.
488 	 */
489 	return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
490 }
491 EXPORT_SYMBOL(panic_on_this_cpu);
492 
493 /*
494  * Return true if a panic is in progress on a remote CPU.
495  *
496  * On true, the local CPU should immediately release any printing resources
497  * that may be needed by the panic CPU.
498  */
499 bool panic_on_other_cpu(void)
500 {
501 	return (panic_in_progress() && !panic_on_this_cpu());
502 }
503 EXPORT_SYMBOL(panic_on_other_cpu);
504 
505 /*
506  * A variant of panic() called from NMI context. We return if we've already
507  * panicked on this CPU. If another CPU already panicked, loop in
508  * nmi_panic_self_stop() which can provide architecture dependent code such
509  * as saving register state for crash dump.
510  */
511 void nmi_panic(struct pt_regs *regs, const char *msg)
512 {
513 	if (panic_try_start())
514 		panic("%s", msg);
515 	else if (panic_on_other_cpu())
516 		nmi_panic_self_stop(regs);
517 }
518 EXPORT_SYMBOL(nmi_panic);
519 
520 void check_panic_on_warn(const char *origin)
521 {
522 	unsigned int limit;
523 
524 	if (panic_on_warn)
525 		panic("%s: panic_on_warn set ...\n", origin);
526 
527 	limit = READ_ONCE(warn_limit);
528 	if (atomic_inc_return(&warn_count) >= limit && limit)
529 		panic("%s: system warned too often (kernel.warn_limit is %d)",
530 		      origin, limit);
531 }
532 
533 static void panic_trigger_all_cpu_backtrace(void)
534 {
535 	/* Temporary allow non-panic CPUs to write their backtraces. */
536 	panic_triggering_all_cpu_backtrace = true;
537 
538 	if (panic_this_cpu_backtrace_printed)
539 		trigger_allbutcpu_cpu_backtrace(raw_smp_processor_id());
540 	else
541 		trigger_all_cpu_backtrace();
542 
543 	panic_triggering_all_cpu_backtrace = false;
544 }
545 
546 /*
547  * Helper that triggers the NMI backtrace (if set in panic_print)
548  * and then performs the secondary CPUs shutdown - we cannot have
549  * the NMI backtrace after the CPUs are off!
550  */
551 static void panic_other_cpus_shutdown(bool crash_kexec)
552 {
553 	if (panic_print & SYS_INFO_ALL_BT)
554 		panic_trigger_all_cpu_backtrace();
555 
556 	/*
557 	 * Note that smp_send_stop() is the usual SMP shutdown function,
558 	 * which unfortunately may not be hardened to work in a panic
559 	 * situation. If we want to do crash dump after notifier calls
560 	 * and kmsg_dump, we will need architecture dependent extra
561 	 * bits in addition to stopping other CPUs, hence we rely on
562 	 * crash_smp_send_stop() for that.
563 	 */
564 	if (!crash_kexec)
565 		smp_send_stop();
566 	else
567 		crash_smp_send_stop();
568 }
569 
570 /**
571  * vpanic - halt the system
572  * @fmt: The text string to print
573  * @args: Arguments for the format string
574  *
575  * Display a message, then perform cleanups. This function never returns.
576  */
577 void vpanic(const char *fmt, va_list args)
578 {
579 	static char buf[PANIC_MSG_BUFSZ];
580 	long i, i_next = 0, len;
581 	int state = 0;
582 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
583 
584 	if (panic_on_warn) {
585 		/*
586 		 * This thread may hit another WARN() in the panic path.
587 		 * Resetting this prevents additional WARN() from panicking the
588 		 * system on this thread.  Other threads are blocked by the
589 		 * panic_mutex in panic().
590 		 */
591 		panic_on_warn = 0;
592 	}
593 
594 	/*
595 	 * Disable local interrupts. This will prevent panic_smp_self_stop
596 	 * from deadlocking the first cpu that invokes the panic, since
597 	 * there is nothing to prevent an interrupt handler (that runs
598 	 * after setting panic_cpu) from invoking panic() again.
599 	 */
600 	local_irq_disable();
601 	preempt_disable_notrace();
602 
603 	/* Redirect panic to target CPU if configured via panic_force_cpu=. */
604 	if (panic_try_force_cpu(fmt, args)) {
605 		/*
606 		 * Mark ourselves offline so panic_other_cpus_shutdown() won't wait
607 		 * for us on architectures that check num_online_cpus().
608 		 */
609 		set_cpu_online(smp_processor_id(), false);
610 		panic_smp_self_stop();
611 	}
612 	/*
613 	 * It's possible to come here directly from a panic-assertion and
614 	 * not have preempt disabled. Some functions called from here want
615 	 * preempt to be disabled. No point enabling it later though...
616 	 *
617 	 * Only one CPU is allowed to execute the panic code from here. For
618 	 * multiple parallel invocations of panic, all other CPUs either
619 	 * stop themself or will wait until they are stopped by the 1st CPU
620 	 * with smp_send_stop().
621 	 *
622 	 * cmpxchg success means this is the 1st CPU which comes here,
623 	 * so go ahead.
624 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
625 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
626 	 */
627 	/* atomic_try_cmpxchg updates old_cpu on failure */
628 	if (panic_try_start()) {
629 		/* go ahead */
630 	} else if (panic_on_other_cpu())
631 		panic_smp_self_stop();
632 
633 	console_verbose();
634 	bust_spinlocks(1);
635 	len = vscnprintf(buf, sizeof(buf), fmt, args);
636 
637 	if (len && buf[len - 1] == '\n')
638 		buf[len - 1] = '\0';
639 
640 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
641 	/*
642 	 * Avoid nested stack-dumping if a panic occurs during oops processing
643 	 */
644 	if (atomic_read(&panic_redirect_cpu) != PANIC_CPU_INVALID &&
645 	    panic_force_cpu == raw_smp_processor_id()) {
646 		pr_emerg("panic: Redirected from CPU %d, skipping stack dump.\n",
647 			 atomic_read(&panic_redirect_cpu));
648 	} else if (test_taint(TAINT_DIE) || oops_in_progress > 1) {
649 		panic_this_cpu_backtrace_printed = true;
650 	} else if (IS_ENABLED(CONFIG_DEBUG_BUGVERBOSE)) {
651 		dump_stack();
652 		panic_this_cpu_backtrace_printed = true;
653 	}
654 
655 	/*
656 	 * If kgdb is enabled, give it a chance to run before we stop all
657 	 * the other CPUs or else we won't be able to debug processes left
658 	 * running on them.
659 	 */
660 	kgdb_panic(buf);
661 
662 	/*
663 	 * If we have crashed and we have a crash kernel loaded let it handle
664 	 * everything else.
665 	 * If we want to run this after calling panic_notifiers, pass
666 	 * the "crash_kexec_post_notifiers" option to the kernel.
667 	 *
668 	 * Bypass the panic_cpu check and call __crash_kexec directly.
669 	 */
670 	if (!_crash_kexec_post_notifiers)
671 		__crash_kexec(NULL);
672 
673 	panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
674 
675 	printk_legacy_allow_panic_sync();
676 
677 	/*
678 	 * Run any panic handlers, including those that might need to
679 	 * add information to the kmsg dump output.
680 	 */
681 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
682 
683 	sys_info(panic_print);
684 
685 	kmsg_dump_desc(KMSG_DUMP_PANIC, buf);
686 
687 	/*
688 	 * If you doubt kdump always works fine in any situation,
689 	 * "crash_kexec_post_notifiers" offers you a chance to run
690 	 * panic_notifiers and dumping kmsg before kdump.
691 	 * Note: since some panic_notifiers can make crashed kernel
692 	 * more unstable, it can increase risks of the kdump failure too.
693 	 *
694 	 * Bypass the panic_cpu check and call __crash_kexec directly.
695 	 */
696 	if (_crash_kexec_post_notifiers)
697 		__crash_kexec(NULL);
698 
699 	console_unblank();
700 
701 	/*
702 	 * We may have ended up stopping the CPU holding the lock (in
703 	 * smp_send_stop()) while still having some valuable data in the console
704 	 * buffer.  Try to acquire the lock then release it regardless of the
705 	 * result.  The release will also print the buffers out.  Locks debug
706 	 * should be disabled to avoid reporting bad unlock balance when
707 	 * panic() is not being callled from OOPS.
708 	 */
709 	debug_locks_off();
710 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
711 
712 	if ((panic_print & SYS_INFO_PANIC_CONSOLE_REPLAY) ||
713 		panic_console_replay)
714 		console_flush_on_panic(CONSOLE_REPLAY_ALL);
715 
716 	if (!panic_blink)
717 		panic_blink = no_blink;
718 
719 	if (panic_timeout > 0) {
720 		/*
721 		 * Delay timeout seconds before rebooting the machine.
722 		 * We can't use the "normal" timers since we just panicked.
723 		 */
724 		pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
725 
726 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
727 			touch_nmi_watchdog();
728 			if (i >= i_next) {
729 				i += panic_blink(state ^= 1);
730 				i_next = i + 3600 / PANIC_BLINK_SPD;
731 			}
732 			mdelay(PANIC_TIMER_STEP);
733 		}
734 	}
735 	if (panic_timeout != 0) {
736 		/*
737 		 * This will not be a clean reboot, with everything
738 		 * shutting down.  But if there is a chance of
739 		 * rebooting the system it will be rebooted.
740 		 */
741 		if (panic_reboot_mode != REBOOT_UNDEFINED)
742 			reboot_mode = panic_reboot_mode;
743 		emergency_restart();
744 	}
745 #ifdef __sparc__
746 	{
747 		extern int stop_a_enabled;
748 		/* Make sure the user can actually press Stop-A (L1-A) */
749 		stop_a_enabled = 1;
750 		pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
751 			 "twice on console to return to the boot prom\n");
752 	}
753 #endif
754 #if defined(CONFIG_S390)
755 	disabled_wait();
756 #endif
757 	pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
758 
759 	/* Do not scroll important messages printed above */
760 	suppress_printk = 1;
761 
762 	/*
763 	 * The final messages may not have been printed if in a context that
764 	 * defers printing (such as NMI) and irq_work is not available.
765 	 * Explicitly flush the kernel log buffer one last time.
766 	 */
767 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
768 	nbcon_atomic_flush_unsafe();
769 
770 	local_irq_enable();
771 	for (i = 0; ; i += PANIC_TIMER_STEP) {
772 		touch_softlockup_watchdog();
773 		if (i >= i_next) {
774 			i += panic_blink(state ^= 1);
775 			i_next = i + 3600 / PANIC_BLINK_SPD;
776 		}
777 		mdelay(PANIC_TIMER_STEP);
778 	}
779 }
780 EXPORT_SYMBOL(vpanic);
781 
782 /* Identical to vpanic(), except it takes variadic arguments instead of va_list */
783 void panic(const char *fmt, ...)
784 {
785 	va_list args;
786 
787 	va_start(args, fmt);
788 	vpanic(fmt, args);
789 	va_end(args);
790 }
791 EXPORT_SYMBOL(panic);
792 
793 #define TAINT_FLAG(taint, _c_true, _c_false)				\
794 	[ TAINT_##taint ] = {						\
795 		.c_true = _c_true, .c_false = _c_false,			\
796 		.desc = #taint,						\
797 	}
798 
799 /*
800  * NOTE: if you modify the taint_flags or TAINT_FLAGS_COUNT,
801  * please also modify tools/debugging/kernel-chktaint and
802  * Documentation/admin-guide/tainted-kernels.rst, including its
803  * small shell script that prints the TAINT_FLAGS_COUNT bits of
804  * /proc/sys/kernel/tainted.
805  *
806  * Also, update INIT_TAINT_BUF_MAX below.
807  */
808 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
809 	TAINT_FLAG(PROPRIETARY_MODULE,		'P', 'G'),
810 	TAINT_FLAG(FORCED_MODULE,		'F', ' '),
811 	TAINT_FLAG(CPU_OUT_OF_SPEC,		'S', ' '),
812 	TAINT_FLAG(FORCED_RMMOD,		'R', ' '),
813 	TAINT_FLAG(MACHINE_CHECK,		'M', ' '),
814 	TAINT_FLAG(BAD_PAGE,			'B', ' '),
815 	TAINT_FLAG(USER,			'U', ' '),
816 	TAINT_FLAG(DIE,				'D', ' '),
817 	TAINT_FLAG(OVERRIDDEN_ACPI_TABLE,	'A', ' '),
818 	TAINT_FLAG(WARN,			'W', ' '),
819 	TAINT_FLAG(CRAP,			'C', ' '),
820 	TAINT_FLAG(FIRMWARE_WORKAROUND,		'I', ' '),
821 	TAINT_FLAG(OOT_MODULE,			'O', ' '),
822 	TAINT_FLAG(UNSIGNED_MODULE,		'E', ' '),
823 	TAINT_FLAG(SOFTLOCKUP,			'L', ' '),
824 	TAINT_FLAG(LIVEPATCH,			'K', ' '),
825 	TAINT_FLAG(AUX,				'X', ' '),
826 	TAINT_FLAG(RANDSTRUCT,			'T', ' '),
827 	TAINT_FLAG(TEST,			'N', ' '),
828 	TAINT_FLAG(FWCTL,			'J', ' '),
829 };
830 
831 #undef TAINT_FLAG
832 
833 static void print_tainted_seq(struct seq_buf *s, bool verbose)
834 {
835 	const char *sep = "";
836 	int i;
837 
838 	if (!tainted_mask) {
839 		seq_buf_puts(s, "Not tainted");
840 		return;
841 	}
842 
843 	seq_buf_printf(s, "Tainted: ");
844 	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
845 		const struct taint_flag *t = &taint_flags[i];
846 		bool is_set = test_bit(i, &tainted_mask);
847 		char c = is_set ? t->c_true : t->c_false;
848 
849 		if (verbose) {
850 			if (is_set) {
851 				seq_buf_printf(s, "%s[%c]=%s", sep, c, t->desc);
852 				sep = ", ";
853 			}
854 		} else {
855 			seq_buf_putc(s, c);
856 		}
857 	}
858 }
859 
860 /* The initial buffer can accommodate all taint flags in verbose
861  * mode, with some headroom. Once the allocator is available, the
862  * exact size is allocated dynamically; the initial buffer remains
863  * as a fallback if allocation fails.
864  *
865  * The verbose taint string currently requires up to 327 characters.
866  */
867 #define INIT_TAINT_BUF_MAX 350
868 
869 static char init_taint_buf[INIT_TAINT_BUF_MAX] __initdata;
870 static char *taint_buf __refdata = init_taint_buf;
871 static size_t taint_buf_size = INIT_TAINT_BUF_MAX;
872 
873 static __init int alloc_taint_buf(void)
874 {
875 	int i;
876 	char *buf;
877 	size_t size = 0;
878 
879 	size += sizeof("Tainted: ") - 1;
880 	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
881 		size += 2; /* For ", " */
882 		size += 4; /* For "[%c]=" */
883 		size += strlen(taint_flags[i].desc);
884 	}
885 
886 	size += 1; /* For NULL terminator */
887 
888 	buf = kmalloc(size, GFP_KERNEL);
889 
890 	if (!buf) {
891 		panic("Failed to allocate taint string buffer");
892 	}
893 
894 	taint_buf = buf;
895 	taint_buf_size = size;
896 
897 	return 0;
898 }
899 postcore_initcall(alloc_taint_buf);
900 
901 static const char *_print_tainted(bool verbose)
902 {
903 	struct seq_buf s;
904 
905 	BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
906 
907 	seq_buf_init(&s, taint_buf, taint_buf_size);
908 
909 	print_tainted_seq(&s, verbose);
910 
911 	return seq_buf_str(&s);
912 }
913 
914 /**
915  * print_tainted - return a string to represent the kernel taint state.
916  *
917  * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
918  *
919  * The string is overwritten by the next call to print_tainted(),
920  * but is always NULL terminated.
921  */
922 const char *print_tainted(void)
923 {
924 	return _print_tainted(false);
925 }
926 
927 /**
928  * print_tainted_verbose - A more verbose version of print_tainted()
929  */
930 const char *print_tainted_verbose(void)
931 {
932 	return _print_tainted(true);
933 }
934 
935 int test_taint(unsigned flag)
936 {
937 	return test_bit(flag, &tainted_mask);
938 }
939 EXPORT_SYMBOL(test_taint);
940 
941 unsigned long get_taint(void)
942 {
943 	return tainted_mask;
944 }
945 
946 /**
947  * add_taint: add a taint flag if not already set.
948  * @flag: one of the TAINT_* constants.
949  * @lockdep_ok: whether lock debugging is still OK.
950  *
951  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
952  * some notewortht-but-not-corrupting cases, it can be set to true.
953  */
954 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
955 {
956 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
957 		pr_warn("Disabling lock debugging due to kernel taint\n");
958 
959 	set_bit(flag, &tainted_mask);
960 
961 	if (tainted_mask & panic_on_taint) {
962 		panic_on_taint = 0;
963 		panic("panic_on_taint set ...");
964 	}
965 }
966 EXPORT_SYMBOL(add_taint);
967 
968 static void spin_msec(int msecs)
969 {
970 	int i;
971 
972 	for (i = 0; i < msecs; i++) {
973 		touch_nmi_watchdog();
974 		mdelay(1);
975 	}
976 }
977 
978 /*
979  * It just happens that oops_enter() and oops_exit() are identically
980  * implemented...
981  */
982 static void do_oops_enter_exit(void)
983 {
984 	unsigned long flags;
985 	static int spin_counter;
986 
987 	if (!pause_on_oops)
988 		return;
989 
990 	spin_lock_irqsave(&pause_on_oops_lock, flags);
991 	if (pause_on_oops_flag == 0) {
992 		/* This CPU may now print the oops message */
993 		pause_on_oops_flag = 1;
994 	} else {
995 		/* We need to stall this CPU */
996 		if (!spin_counter) {
997 			/* This CPU gets to do the counting */
998 			spin_counter = pause_on_oops;
999 			do {
1000 				spin_unlock(&pause_on_oops_lock);
1001 				spin_msec(MSEC_PER_SEC);
1002 				spin_lock(&pause_on_oops_lock);
1003 			} while (--spin_counter);
1004 			pause_on_oops_flag = 0;
1005 		} else {
1006 			/* This CPU waits for a different one */
1007 			while (spin_counter) {
1008 				spin_unlock(&pause_on_oops_lock);
1009 				spin_msec(1);
1010 				spin_lock(&pause_on_oops_lock);
1011 			}
1012 		}
1013 	}
1014 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
1015 }
1016 
1017 /*
1018  * Return true if the calling CPU is allowed to print oops-related info.
1019  * This is a bit racy..
1020  */
1021 bool oops_may_print(void)
1022 {
1023 	return pause_on_oops_flag == 0;
1024 }
1025 
1026 /*
1027  * Called when the architecture enters its oops handler, before it prints
1028  * anything.  If this is the first CPU to oops, and it's oopsing the first
1029  * time then let it proceed.
1030  *
1031  * This is all enabled by the pause_on_oops kernel boot option.  We do all
1032  * this to ensure that oopses don't scroll off the screen.  It has the
1033  * side-effect of preventing later-oopsing CPUs from mucking up the display,
1034  * too.
1035  *
1036  * It turns out that the CPU which is allowed to print ends up pausing for
1037  * the right duration, whereas all the other CPUs pause for twice as long:
1038  * once in oops_enter(), once in oops_exit().
1039  */
1040 void oops_enter(void)
1041 {
1042 	nbcon_cpu_emergency_enter();
1043 	tracing_off();
1044 	/* can't trust the integrity of the kernel anymore: */
1045 	debug_locks_off();
1046 	do_oops_enter_exit();
1047 
1048 	if (sysctl_oops_all_cpu_backtrace)
1049 		trigger_all_cpu_backtrace();
1050 }
1051 
1052 static void print_oops_end_marker(void)
1053 {
1054 	pr_warn("---[ end trace %016llx ]---\n", 0ULL);
1055 }
1056 
1057 /*
1058  * Called when the architecture exits its oops handler, after printing
1059  * everything.
1060  */
1061 void oops_exit(void)
1062 {
1063 	do_oops_enter_exit();
1064 	print_oops_end_marker();
1065 	nbcon_cpu_emergency_exit();
1066 	kmsg_dump(KMSG_DUMP_OOPS);
1067 }
1068 
1069 struct warn_args {
1070 	const char *fmt;
1071 	va_list args;
1072 };
1073 
1074 void __warn(const char *file, int line, void *caller, unsigned taint,
1075 	    struct pt_regs *regs, struct warn_args *args)
1076 {
1077 	nbcon_cpu_emergency_enter();
1078 
1079 	disable_trace_on_warning();
1080 
1081 	if (file) {
1082 		pr_warn("WARNING: %s:%d at %pS, CPU#%d: %s/%d\n",
1083 			file, line, caller,
1084 			raw_smp_processor_id(), current->comm, current->pid);
1085 	} else {
1086 		pr_warn("WARNING: at %pS, CPU#%d: %s/%d\n",
1087 			caller,
1088 			raw_smp_processor_id(), current->comm, current->pid);
1089 	}
1090 
1091 #pragma GCC diagnostic push
1092 #ifndef __clang__
1093 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
1094 #endif
1095 	if (args)
1096 		vprintk(args->fmt, args->args);
1097 #pragma GCC diagnostic pop
1098 
1099 	print_modules();
1100 
1101 	if (regs)
1102 		show_regs(regs);
1103 
1104 	check_panic_on_warn("kernel");
1105 
1106 	if (!regs)
1107 		dump_stack();
1108 
1109 	print_irqtrace_events(current);
1110 
1111 	print_oops_end_marker();
1112 	trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
1113 
1114 	/* Just a warning, don't kill lockdep. */
1115 	add_taint(taint, LOCKDEP_STILL_OK);
1116 
1117 	nbcon_cpu_emergency_exit();
1118 }
1119 
1120 #ifdef CONFIG_BUG
1121 #ifndef __WARN_FLAGS
1122 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
1123 		       const char *fmt, ...)
1124 {
1125 	bool rcu = warn_rcu_enter();
1126 	struct warn_args args;
1127 
1128 	if (kunit_is_suppressed_warning(true)) {
1129 		warn_rcu_exit(rcu);
1130 		return;
1131 	}
1132 
1133 	pr_warn(CUT_HERE);
1134 
1135 	if (!fmt) {
1136 		__warn(file, line, __builtin_return_address(0), taint,
1137 		       NULL, NULL);
1138 		warn_rcu_exit(rcu);
1139 		return;
1140 	}
1141 
1142 	args.fmt = fmt;
1143 	va_start(args.args, fmt);
1144 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
1145 	va_end(args.args);
1146 	warn_rcu_exit(rcu);
1147 }
1148 EXPORT_SYMBOL(warn_slowpath_fmt);
1149 #else
1150 void __warn_printk(const char *fmt, ...)
1151 {
1152 	bool rcu = warn_rcu_enter();
1153 	va_list args;
1154 
1155 	if (kunit_is_suppressed_warning(false)) {
1156 		warn_rcu_exit(rcu);
1157 		return;
1158 	}
1159 
1160 	pr_warn(CUT_HERE);
1161 
1162 	va_start(args, fmt);
1163 	vprintk(fmt, args);
1164 	va_end(args);
1165 	warn_rcu_exit(rcu);
1166 }
1167 EXPORT_SYMBOL(__warn_printk);
1168 #endif
1169 
1170 /* Support resetting WARN*_ONCE state */
1171 
1172 static int clear_warn_once_set(void *data, u64 val)
1173 {
1174 	generic_bug_clear_once();
1175 	memset(__start_once, 0, __end_once - __start_once);
1176 	return 0;
1177 }
1178 
1179 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
1180 			 "%lld\n");
1181 
1182 static __init int register_warn_debugfs(void)
1183 {
1184 	/* Don't care about failure */
1185 	debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
1186 				   &clear_warn_once_fops);
1187 	return 0;
1188 }
1189 
1190 device_initcall(register_warn_debugfs);
1191 #endif
1192 
1193 #ifdef CONFIG_STACKPROTECTOR
1194 
1195 /*
1196  * Called when gcc's -fstack-protector feature is used, and
1197  * gcc detects corruption of the on-stack canary value
1198  */
1199 __visible noinstr void __stack_chk_fail(void)
1200 {
1201 	unsigned long flags;
1202 
1203 	instrumentation_begin();
1204 	flags = user_access_save();
1205 
1206 	panic("stack-protector: Kernel stack is corrupted in: %pB",
1207 		__builtin_return_address(0));
1208 
1209 	user_access_restore(flags);
1210 	instrumentation_end();
1211 }
1212 EXPORT_SYMBOL(__stack_chk_fail);
1213 
1214 #endif
1215 
1216 core_param(panic, panic_timeout, int, 0644);
1217 core_param(pause_on_oops, pause_on_oops, int, 0644);
1218 core_param(panic_on_warn, panic_on_warn, int, 0644);
1219 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
1220 core_param(panic_console_replay, panic_console_replay, bool, 0644);
1221 
1222 static int panic_print_set(const char *val, const struct kernel_param *kp)
1223 {
1224 	panic_print_deprecated();
1225 	return  param_set_ulong(val, kp);
1226 }
1227 
1228 static int panic_print_get(char *val, const struct kernel_param *kp)
1229 {
1230 	return  param_get_ulong(val, kp);
1231 }
1232 
1233 static const struct kernel_param_ops panic_print_ops = {
1234 	.set	= panic_print_set,
1235 	.get	= panic_print_get,
1236 };
1237 __core_param_cb(panic_print, &panic_print_ops, &panic_print, 0644);
1238 
1239 static int __init oops_setup(char *s)
1240 {
1241 	if (!s)
1242 		return -EINVAL;
1243 	if (!strcmp(s, "panic"))
1244 		panic_on_oops = 1;
1245 	return 0;
1246 }
1247 early_param("oops", oops_setup);
1248 
1249 static int __init panic_on_taint_setup(char *s)
1250 {
1251 	char *taint_str;
1252 
1253 	if (!s)
1254 		return -EINVAL;
1255 
1256 	taint_str = strsep(&s, ",");
1257 	if (kstrtoul(taint_str, 16, &panic_on_taint))
1258 		return -EINVAL;
1259 
1260 	/* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
1261 	panic_on_taint &= TAINT_FLAGS_MAX;
1262 
1263 	if (!panic_on_taint)
1264 		return -EINVAL;
1265 
1266 	if (s && !strcmp(s, "nousertaint"))
1267 		panic_on_taint_nousertaint = true;
1268 
1269 	pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
1270 		panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
1271 
1272 	return 0;
1273 }
1274 early_param("panic_on_taint", panic_on_taint_setup);
1275