xref: /linux/kernel/panic.c (revision 4f372263ef92ed2af55a8c226750b72021ff8d0f)
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 <trace/events/error_report.h>
40 #include <asm/sections.h>
41 
42 #define PANIC_TIMER_STEP 100
43 #define PANIC_BLINK_SPD 18
44 
45 #ifdef CONFIG_SMP
46 /*
47  * Should we dump all CPUs backtraces in an oops event?
48  * Defaults to 0, can be changed via sysctl.
49  */
50 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
51 #else
52 #define sysctl_oops_all_cpu_backtrace 0
53 #endif /* CONFIG_SMP */
54 
55 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
56 static unsigned long tainted_mask =
57 	IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
58 static int pause_on_oops;
59 static int pause_on_oops_flag;
60 static DEFINE_SPINLOCK(pause_on_oops_lock);
61 bool crash_kexec_post_notifiers;
62 int panic_on_warn __read_mostly;
63 unsigned long panic_on_taint;
64 bool panic_on_taint_nousertaint = false;
65 static unsigned int warn_limit __read_mostly;
66 
67 bool panic_triggering_all_cpu_backtrace;
68 
69 int panic_timeout = CONFIG_PANIC_TIMEOUT;
70 EXPORT_SYMBOL_GPL(panic_timeout);
71 
72 #define PANIC_PRINT_TASK_INFO		0x00000001
73 #define PANIC_PRINT_MEM_INFO		0x00000002
74 #define PANIC_PRINT_TIMER_INFO		0x00000004
75 #define PANIC_PRINT_LOCK_INFO		0x00000008
76 #define PANIC_PRINT_FTRACE_INFO		0x00000010
77 #define PANIC_PRINT_ALL_PRINTK_MSG	0x00000020
78 #define PANIC_PRINT_ALL_CPU_BT		0x00000040
79 #define PANIC_PRINT_BLOCKED_TASKS	0x00000080
80 unsigned long panic_print;
81 
82 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
83 
84 EXPORT_SYMBOL(panic_notifier_list);
85 
86 #ifdef CONFIG_SYSCTL
87 static const struct ctl_table kern_panic_table[] = {
88 #ifdef CONFIG_SMP
89 	{
90 		.procname       = "oops_all_cpu_backtrace",
91 		.data           = &sysctl_oops_all_cpu_backtrace,
92 		.maxlen         = sizeof(int),
93 		.mode           = 0644,
94 		.proc_handler   = proc_dointvec_minmax,
95 		.extra1         = SYSCTL_ZERO,
96 		.extra2         = SYSCTL_ONE,
97 	},
98 #endif
99 	{
100 		.procname	= "panic",
101 		.data		= &panic_timeout,
102 		.maxlen		= sizeof(int),
103 		.mode		= 0644,
104 		.proc_handler	= proc_dointvec,
105 	},
106 	{
107 		.procname	= "panic_on_oops",
108 		.data		= &panic_on_oops,
109 		.maxlen		= sizeof(int),
110 		.mode		= 0644,
111 		.proc_handler	= proc_dointvec,
112 	},
113 	{
114 		.procname	= "panic_print",
115 		.data		= &panic_print,
116 		.maxlen		= sizeof(unsigned long),
117 		.mode		= 0644,
118 		.proc_handler	= proc_doulongvec_minmax,
119 	},
120 	{
121 		.procname	= "panic_on_warn",
122 		.data		= &panic_on_warn,
123 		.maxlen		= sizeof(int),
124 		.mode		= 0644,
125 		.proc_handler	= proc_dointvec_minmax,
126 		.extra1		= SYSCTL_ZERO,
127 		.extra2		= SYSCTL_ONE,
128 	},
129 	{
130 		.procname       = "warn_limit",
131 		.data           = &warn_limit,
132 		.maxlen         = sizeof(warn_limit),
133 		.mode           = 0644,
134 		.proc_handler   = proc_douintvec,
135 	},
136 };
137 
138 static __init int kernel_panic_sysctls_init(void)
139 {
140 	register_sysctl_init("kernel", kern_panic_table);
141 	return 0;
142 }
143 late_initcall(kernel_panic_sysctls_init);
144 #endif
145 
146 static atomic_t warn_count = ATOMIC_INIT(0);
147 
148 #ifdef CONFIG_SYSFS
149 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
150 			       char *page)
151 {
152 	return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
153 }
154 
155 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
156 
157 static __init int kernel_panic_sysfs_init(void)
158 {
159 	sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
160 	return 0;
161 }
162 late_initcall(kernel_panic_sysfs_init);
163 #endif
164 
165 static long no_blink(int state)
166 {
167 	return 0;
168 }
169 
170 /* Returns how long it waited in ms */
171 long (*panic_blink)(int state);
172 EXPORT_SYMBOL(panic_blink);
173 
174 /*
175  * Stop ourself in panic -- architecture code may override this
176  */
177 void __weak __noreturn panic_smp_self_stop(void)
178 {
179 	while (1)
180 		cpu_relax();
181 }
182 
183 /*
184  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
185  * may override this to prepare for crash dumping, e.g. save regs info.
186  */
187 void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
188 {
189 	panic_smp_self_stop();
190 }
191 
192 /*
193  * Stop other CPUs in panic.  Architecture dependent code may override this
194  * with more suitable version.  For example, if the architecture supports
195  * crash dump, it should save registers of each stopped CPU and disable
196  * per-CPU features such as virtualization extensions.
197  */
198 void __weak crash_smp_send_stop(void)
199 {
200 	static int cpus_stopped;
201 
202 	/*
203 	 * This function can be called twice in panic path, but obviously
204 	 * we execute this only once.
205 	 */
206 	if (cpus_stopped)
207 		return;
208 
209 	/*
210 	 * Note smp_send_stop is the usual smp shutdown function, which
211 	 * unfortunately means it may not be hardened to work in a panic
212 	 * situation.
213 	 */
214 	smp_send_stop();
215 	cpus_stopped = 1;
216 }
217 
218 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
219 
220 /*
221  * A variant of panic() called from NMI context. We return if we've already
222  * panicked on this CPU. If another CPU already panicked, loop in
223  * nmi_panic_self_stop() which can provide architecture dependent code such
224  * as saving register state for crash dump.
225  */
226 void nmi_panic(struct pt_regs *regs, const char *msg)
227 {
228 	int old_cpu, this_cpu;
229 
230 	old_cpu = PANIC_CPU_INVALID;
231 	this_cpu = raw_smp_processor_id();
232 
233 	/* atomic_try_cmpxchg updates old_cpu on failure */
234 	if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu))
235 		panic("%s", msg);
236 	else if (old_cpu != this_cpu)
237 		nmi_panic_self_stop(regs);
238 }
239 EXPORT_SYMBOL(nmi_panic);
240 
241 static void panic_print_sys_info(bool console_flush)
242 {
243 	if (console_flush) {
244 		if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
245 			console_flush_on_panic(CONSOLE_REPLAY_ALL);
246 		return;
247 	}
248 
249 	if (panic_print & PANIC_PRINT_TASK_INFO)
250 		show_state();
251 
252 	if (panic_print & PANIC_PRINT_MEM_INFO)
253 		show_mem();
254 
255 	if (panic_print & PANIC_PRINT_TIMER_INFO)
256 		sysrq_timer_list_show();
257 
258 	if (panic_print & PANIC_PRINT_LOCK_INFO)
259 		debug_show_all_locks();
260 
261 	if (panic_print & PANIC_PRINT_FTRACE_INFO)
262 		ftrace_dump(DUMP_ALL);
263 
264 	if (panic_print & PANIC_PRINT_BLOCKED_TASKS)
265 		show_state_filter(TASK_UNINTERRUPTIBLE);
266 }
267 
268 void check_panic_on_warn(const char *origin)
269 {
270 	unsigned int limit;
271 
272 	if (panic_on_warn)
273 		panic("%s: panic_on_warn set ...\n", origin);
274 
275 	limit = READ_ONCE(warn_limit);
276 	if (atomic_inc_return(&warn_count) >= limit && limit)
277 		panic("%s: system warned too often (kernel.warn_limit is %d)",
278 		      origin, limit);
279 }
280 
281 /*
282  * Helper that triggers the NMI backtrace (if set in panic_print)
283  * and then performs the secondary CPUs shutdown - we cannot have
284  * the NMI backtrace after the CPUs are off!
285  */
286 static void panic_other_cpus_shutdown(bool crash_kexec)
287 {
288 	if (panic_print & PANIC_PRINT_ALL_CPU_BT) {
289 		/* Temporary allow non-panic CPUs to write their backtraces. */
290 		panic_triggering_all_cpu_backtrace = true;
291 		trigger_all_cpu_backtrace();
292 		panic_triggering_all_cpu_backtrace = false;
293 	}
294 
295 	/*
296 	 * Note that smp_send_stop() is the usual SMP shutdown function,
297 	 * which unfortunately may not be hardened to work in a panic
298 	 * situation. If we want to do crash dump after notifier calls
299 	 * and kmsg_dump, we will need architecture dependent extra
300 	 * bits in addition to stopping other CPUs, hence we rely on
301 	 * crash_smp_send_stop() for that.
302 	 */
303 	if (!crash_kexec)
304 		smp_send_stop();
305 	else
306 		crash_smp_send_stop();
307 }
308 
309 /**
310  *	panic - halt the system
311  *	@fmt: The text string to print
312  *
313  *	Display a message, then perform cleanups.
314  *
315  *	This function never returns.
316  */
317 void panic(const char *fmt, ...)
318 {
319 	static char buf[1024];
320 	va_list args;
321 	long i, i_next = 0, len;
322 	int state = 0;
323 	int old_cpu, this_cpu;
324 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
325 
326 	if (panic_on_warn) {
327 		/*
328 		 * This thread may hit another WARN() in the panic path.
329 		 * Resetting this prevents additional WARN() from panicking the
330 		 * system on this thread.  Other threads are blocked by the
331 		 * panic_mutex in panic().
332 		 */
333 		panic_on_warn = 0;
334 	}
335 
336 	/*
337 	 * Disable local interrupts. This will prevent panic_smp_self_stop
338 	 * from deadlocking the first cpu that invokes the panic, since
339 	 * there is nothing to prevent an interrupt handler (that runs
340 	 * after setting panic_cpu) from invoking panic() again.
341 	 */
342 	local_irq_disable();
343 	preempt_disable_notrace();
344 
345 	/*
346 	 * It's possible to come here directly from a panic-assertion and
347 	 * not have preempt disabled. Some functions called from here want
348 	 * preempt to be disabled. No point enabling it later though...
349 	 *
350 	 * Only one CPU is allowed to execute the panic code from here. For
351 	 * multiple parallel invocations of panic, all other CPUs either
352 	 * stop themself or will wait until they are stopped by the 1st CPU
353 	 * with smp_send_stop().
354 	 *
355 	 * cmpxchg success means this is the 1st CPU which comes here,
356 	 * so go ahead.
357 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
358 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
359 	 */
360 	old_cpu = PANIC_CPU_INVALID;
361 	this_cpu = raw_smp_processor_id();
362 
363 	/* atomic_try_cmpxchg updates old_cpu on failure */
364 	if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
365 		/* go ahead */
366 	} else if (old_cpu != this_cpu)
367 		panic_smp_self_stop();
368 
369 	console_verbose();
370 	bust_spinlocks(1);
371 	va_start(args, fmt);
372 	len = vscnprintf(buf, sizeof(buf), fmt, args);
373 	va_end(args);
374 
375 	if (len && buf[len - 1] == '\n')
376 		buf[len - 1] = '\0';
377 
378 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
379 #ifdef CONFIG_DEBUG_BUGVERBOSE
380 	/*
381 	 * Avoid nested stack-dumping if a panic occurs during oops processing
382 	 */
383 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
384 		dump_stack();
385 #endif
386 
387 	/*
388 	 * If kgdb is enabled, give it a chance to run before we stop all
389 	 * the other CPUs or else we won't be able to debug processes left
390 	 * running on them.
391 	 */
392 	kgdb_panic(buf);
393 
394 	/*
395 	 * If we have crashed and we have a crash kernel loaded let it handle
396 	 * everything else.
397 	 * If we want to run this after calling panic_notifiers, pass
398 	 * the "crash_kexec_post_notifiers" option to the kernel.
399 	 *
400 	 * Bypass the panic_cpu check and call __crash_kexec directly.
401 	 */
402 	if (!_crash_kexec_post_notifiers)
403 		__crash_kexec(NULL);
404 
405 	panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
406 
407 	printk_legacy_allow_panic_sync();
408 
409 	/*
410 	 * Run any panic handlers, including those that might need to
411 	 * add information to the kmsg dump output.
412 	 */
413 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
414 
415 	panic_print_sys_info(false);
416 
417 	kmsg_dump_desc(KMSG_DUMP_PANIC, buf);
418 
419 	/*
420 	 * If you doubt kdump always works fine in any situation,
421 	 * "crash_kexec_post_notifiers" offers you a chance to run
422 	 * panic_notifiers and dumping kmsg before kdump.
423 	 * Note: since some panic_notifiers can make crashed kernel
424 	 * more unstable, it can increase risks of the kdump failure too.
425 	 *
426 	 * Bypass the panic_cpu check and call __crash_kexec directly.
427 	 */
428 	if (_crash_kexec_post_notifiers)
429 		__crash_kexec(NULL);
430 
431 	console_unblank();
432 
433 	/*
434 	 * We may have ended up stopping the CPU holding the lock (in
435 	 * smp_send_stop()) while still having some valuable data in the console
436 	 * buffer.  Try to acquire the lock then release it regardless of the
437 	 * result.  The release will also print the buffers out.  Locks debug
438 	 * should be disabled to avoid reporting bad unlock balance when
439 	 * panic() is not being callled from OOPS.
440 	 */
441 	debug_locks_off();
442 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
443 
444 	panic_print_sys_info(true);
445 
446 	if (!panic_blink)
447 		panic_blink = no_blink;
448 
449 	if (panic_timeout > 0) {
450 		/*
451 		 * Delay timeout seconds before rebooting the machine.
452 		 * We can't use the "normal" timers since we just panicked.
453 		 */
454 		pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
455 
456 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
457 			touch_nmi_watchdog();
458 			if (i >= i_next) {
459 				i += panic_blink(state ^= 1);
460 				i_next = i + 3600 / PANIC_BLINK_SPD;
461 			}
462 			mdelay(PANIC_TIMER_STEP);
463 		}
464 	}
465 	if (panic_timeout != 0) {
466 		/*
467 		 * This will not be a clean reboot, with everything
468 		 * shutting down.  But if there is a chance of
469 		 * rebooting the system it will be rebooted.
470 		 */
471 		if (panic_reboot_mode != REBOOT_UNDEFINED)
472 			reboot_mode = panic_reboot_mode;
473 		emergency_restart();
474 	}
475 #ifdef __sparc__
476 	{
477 		extern int stop_a_enabled;
478 		/* Make sure the user can actually press Stop-A (L1-A) */
479 		stop_a_enabled = 1;
480 		pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
481 			 "twice on console to return to the boot prom\n");
482 	}
483 #endif
484 #if defined(CONFIG_S390)
485 	disabled_wait();
486 #endif
487 	pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
488 
489 	/* Do not scroll important messages printed above */
490 	suppress_printk = 1;
491 
492 	/*
493 	 * The final messages may not have been printed if in a context that
494 	 * defers printing (such as NMI) and irq_work is not available.
495 	 * Explicitly flush the kernel log buffer one last time.
496 	 */
497 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
498 	nbcon_atomic_flush_unsafe();
499 
500 	local_irq_enable();
501 	for (i = 0; ; i += PANIC_TIMER_STEP) {
502 		touch_softlockup_watchdog();
503 		if (i >= i_next) {
504 			i += panic_blink(state ^= 1);
505 			i_next = i + 3600 / PANIC_BLINK_SPD;
506 		}
507 		mdelay(PANIC_TIMER_STEP);
508 	}
509 }
510 
511 EXPORT_SYMBOL(panic);
512 
513 #define TAINT_FLAG(taint, _c_true, _c_false, _module)			\
514 	[ TAINT_##taint ] = {						\
515 		.c_true = _c_true, .c_false = _c_false,			\
516 		.module = _module,					\
517 		.desc = #taint,						\
518 	}
519 
520 /*
521  * TAINT_FORCED_RMMOD could be a per-module flag but the module
522  * is being removed anyway.
523  */
524 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
525 	TAINT_FLAG(PROPRIETARY_MODULE,		'P', 'G', true),
526 	TAINT_FLAG(FORCED_MODULE,		'F', ' ', true),
527 	TAINT_FLAG(CPU_OUT_OF_SPEC,		'S', ' ', false),
528 	TAINT_FLAG(FORCED_RMMOD,		'R', ' ', false),
529 	TAINT_FLAG(MACHINE_CHECK,		'M', ' ', false),
530 	TAINT_FLAG(BAD_PAGE,			'B', ' ', false),
531 	TAINT_FLAG(USER,			'U', ' ', false),
532 	TAINT_FLAG(DIE,				'D', ' ', false),
533 	TAINT_FLAG(OVERRIDDEN_ACPI_TABLE,	'A', ' ', false),
534 	TAINT_FLAG(WARN,			'W', ' ', false),
535 	TAINT_FLAG(CRAP,			'C', ' ', true),
536 	TAINT_FLAG(FIRMWARE_WORKAROUND,		'I', ' ', false),
537 	TAINT_FLAG(OOT_MODULE,			'O', ' ', true),
538 	TAINT_FLAG(UNSIGNED_MODULE,		'E', ' ', true),
539 	TAINT_FLAG(SOFTLOCKUP,			'L', ' ', false),
540 	TAINT_FLAG(LIVEPATCH,			'K', ' ', true),
541 	TAINT_FLAG(AUX,				'X', ' ', true),
542 	TAINT_FLAG(RANDSTRUCT,			'T', ' ', true),
543 	TAINT_FLAG(TEST,			'N', ' ', true),
544 	TAINT_FLAG(FWCTL,			'J', ' ', true),
545 };
546 
547 #undef TAINT_FLAG
548 
549 static void print_tainted_seq(struct seq_buf *s, bool verbose)
550 {
551 	const char *sep = "";
552 	int i;
553 
554 	if (!tainted_mask) {
555 		seq_buf_puts(s, "Not tainted");
556 		return;
557 	}
558 
559 	seq_buf_printf(s, "Tainted: ");
560 	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
561 		const struct taint_flag *t = &taint_flags[i];
562 		bool is_set = test_bit(i, &tainted_mask);
563 		char c = is_set ? t->c_true : t->c_false;
564 
565 		if (verbose) {
566 			if (is_set) {
567 				seq_buf_printf(s, "%s[%c]=%s", sep, c, t->desc);
568 				sep = ", ";
569 			}
570 		} else {
571 			seq_buf_putc(s, c);
572 		}
573 	}
574 }
575 
576 static const char *_print_tainted(bool verbose)
577 {
578 	/* FIXME: what should the size be? */
579 	static char buf[sizeof(taint_flags)];
580 	struct seq_buf s;
581 
582 	BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
583 
584 	seq_buf_init(&s, buf, sizeof(buf));
585 
586 	print_tainted_seq(&s, verbose);
587 
588 	return seq_buf_str(&s);
589 }
590 
591 /**
592  * print_tainted - return a string to represent the kernel taint state.
593  *
594  * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
595  *
596  * The string is overwritten by the next call to print_tainted(),
597  * but is always NULL terminated.
598  */
599 const char *print_tainted(void)
600 {
601 	return _print_tainted(false);
602 }
603 
604 /**
605  * print_tainted_verbose - A more verbose version of print_tainted()
606  */
607 const char *print_tainted_verbose(void)
608 {
609 	return _print_tainted(true);
610 }
611 
612 int test_taint(unsigned flag)
613 {
614 	return test_bit(flag, &tainted_mask);
615 }
616 EXPORT_SYMBOL(test_taint);
617 
618 unsigned long get_taint(void)
619 {
620 	return tainted_mask;
621 }
622 
623 /**
624  * add_taint: add a taint flag if not already set.
625  * @flag: one of the TAINT_* constants.
626  * @lockdep_ok: whether lock debugging is still OK.
627  *
628  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
629  * some notewortht-but-not-corrupting cases, it can be set to true.
630  */
631 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
632 {
633 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
634 		pr_warn("Disabling lock debugging due to kernel taint\n");
635 
636 	set_bit(flag, &tainted_mask);
637 
638 	if (tainted_mask & panic_on_taint) {
639 		panic_on_taint = 0;
640 		panic("panic_on_taint set ...");
641 	}
642 }
643 EXPORT_SYMBOL(add_taint);
644 
645 static void spin_msec(int msecs)
646 {
647 	int i;
648 
649 	for (i = 0; i < msecs; i++) {
650 		touch_nmi_watchdog();
651 		mdelay(1);
652 	}
653 }
654 
655 /*
656  * It just happens that oops_enter() and oops_exit() are identically
657  * implemented...
658  */
659 static void do_oops_enter_exit(void)
660 {
661 	unsigned long flags;
662 	static int spin_counter;
663 
664 	if (!pause_on_oops)
665 		return;
666 
667 	spin_lock_irqsave(&pause_on_oops_lock, flags);
668 	if (pause_on_oops_flag == 0) {
669 		/* This CPU may now print the oops message */
670 		pause_on_oops_flag = 1;
671 	} else {
672 		/* We need to stall this CPU */
673 		if (!spin_counter) {
674 			/* This CPU gets to do the counting */
675 			spin_counter = pause_on_oops;
676 			do {
677 				spin_unlock(&pause_on_oops_lock);
678 				spin_msec(MSEC_PER_SEC);
679 				spin_lock(&pause_on_oops_lock);
680 			} while (--spin_counter);
681 			pause_on_oops_flag = 0;
682 		} else {
683 			/* This CPU waits for a different one */
684 			while (spin_counter) {
685 				spin_unlock(&pause_on_oops_lock);
686 				spin_msec(1);
687 				spin_lock(&pause_on_oops_lock);
688 			}
689 		}
690 	}
691 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
692 }
693 
694 /*
695  * Return true if the calling CPU is allowed to print oops-related info.
696  * This is a bit racy..
697  */
698 bool oops_may_print(void)
699 {
700 	return pause_on_oops_flag == 0;
701 }
702 
703 /*
704  * Called when the architecture enters its oops handler, before it prints
705  * anything.  If this is the first CPU to oops, and it's oopsing the first
706  * time then let it proceed.
707  *
708  * This is all enabled by the pause_on_oops kernel boot option.  We do all
709  * this to ensure that oopses don't scroll off the screen.  It has the
710  * side-effect of preventing later-oopsing CPUs from mucking up the display,
711  * too.
712  *
713  * It turns out that the CPU which is allowed to print ends up pausing for
714  * the right duration, whereas all the other CPUs pause for twice as long:
715  * once in oops_enter(), once in oops_exit().
716  */
717 void oops_enter(void)
718 {
719 	nbcon_cpu_emergency_enter();
720 	tracing_off();
721 	/* can't trust the integrity of the kernel anymore: */
722 	debug_locks_off();
723 	do_oops_enter_exit();
724 
725 	if (sysctl_oops_all_cpu_backtrace)
726 		trigger_all_cpu_backtrace();
727 }
728 
729 static void print_oops_end_marker(void)
730 {
731 	pr_warn("---[ end trace %016llx ]---\n", 0ULL);
732 }
733 
734 /*
735  * Called when the architecture exits its oops handler, after printing
736  * everything.
737  */
738 void oops_exit(void)
739 {
740 	do_oops_enter_exit();
741 	print_oops_end_marker();
742 	nbcon_cpu_emergency_exit();
743 	kmsg_dump(KMSG_DUMP_OOPS);
744 }
745 
746 struct warn_args {
747 	const char *fmt;
748 	va_list args;
749 };
750 
751 void __warn(const char *file, int line, void *caller, unsigned taint,
752 	    struct pt_regs *regs, struct warn_args *args)
753 {
754 	nbcon_cpu_emergency_enter();
755 
756 	disable_trace_on_warning();
757 
758 	if (file)
759 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
760 			raw_smp_processor_id(), current->pid, file, line,
761 			caller);
762 	else
763 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
764 			raw_smp_processor_id(), current->pid, caller);
765 
766 #pragma GCC diagnostic push
767 #ifndef __clang__
768 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
769 #endif
770 	if (args)
771 		vprintk(args->fmt, args->args);
772 #pragma GCC diagnostic pop
773 
774 	print_modules();
775 
776 	if (regs)
777 		show_regs(regs);
778 
779 	check_panic_on_warn("kernel");
780 
781 	if (!regs)
782 		dump_stack();
783 
784 	print_irqtrace_events(current);
785 
786 	print_oops_end_marker();
787 	trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
788 
789 	/* Just a warning, don't kill lockdep. */
790 	add_taint(taint, LOCKDEP_STILL_OK);
791 
792 	nbcon_cpu_emergency_exit();
793 }
794 
795 #ifdef CONFIG_BUG
796 #ifndef __WARN_FLAGS
797 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
798 		       const char *fmt, ...)
799 {
800 	bool rcu = warn_rcu_enter();
801 	struct warn_args args;
802 
803 	pr_warn(CUT_HERE);
804 
805 	if (!fmt) {
806 		__warn(file, line, __builtin_return_address(0), taint,
807 		       NULL, NULL);
808 		warn_rcu_exit(rcu);
809 		return;
810 	}
811 
812 	args.fmt = fmt;
813 	va_start(args.args, fmt);
814 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
815 	va_end(args.args);
816 	warn_rcu_exit(rcu);
817 }
818 EXPORT_SYMBOL(warn_slowpath_fmt);
819 #else
820 void __warn_printk(const char *fmt, ...)
821 {
822 	bool rcu = warn_rcu_enter();
823 	va_list args;
824 
825 	pr_warn(CUT_HERE);
826 
827 	va_start(args, fmt);
828 	vprintk(fmt, args);
829 	va_end(args);
830 	warn_rcu_exit(rcu);
831 }
832 EXPORT_SYMBOL(__warn_printk);
833 #endif
834 
835 /* Support resetting WARN*_ONCE state */
836 
837 static int clear_warn_once_set(void *data, u64 val)
838 {
839 	generic_bug_clear_once();
840 	memset(__start_once, 0, __end_once - __start_once);
841 	return 0;
842 }
843 
844 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
845 			 "%lld\n");
846 
847 static __init int register_warn_debugfs(void)
848 {
849 	/* Don't care about failure */
850 	debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
851 				   &clear_warn_once_fops);
852 	return 0;
853 }
854 
855 device_initcall(register_warn_debugfs);
856 #endif
857 
858 #ifdef CONFIG_STACKPROTECTOR
859 
860 /*
861  * Called when gcc's -fstack-protector feature is used, and
862  * gcc detects corruption of the on-stack canary value
863  */
864 __visible noinstr void __stack_chk_fail(void)
865 {
866 	unsigned long flags;
867 
868 	instrumentation_begin();
869 	flags = user_access_save();
870 
871 	panic("stack-protector: Kernel stack is corrupted in: %pB",
872 		__builtin_return_address(0));
873 
874 	user_access_restore(flags);
875 	instrumentation_end();
876 }
877 EXPORT_SYMBOL(__stack_chk_fail);
878 
879 #endif
880 
881 core_param(panic, panic_timeout, int, 0644);
882 core_param(panic_print, panic_print, ulong, 0644);
883 core_param(pause_on_oops, pause_on_oops, int, 0644);
884 core_param(panic_on_warn, panic_on_warn, int, 0644);
885 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
886 
887 static int __init oops_setup(char *s)
888 {
889 	if (!s)
890 		return -EINVAL;
891 	if (!strcmp(s, "panic"))
892 		panic_on_oops = 1;
893 	return 0;
894 }
895 early_param("oops", oops_setup);
896 
897 static int __init panic_on_taint_setup(char *s)
898 {
899 	char *taint_str;
900 
901 	if (!s)
902 		return -EINVAL;
903 
904 	taint_str = strsep(&s, ",");
905 	if (kstrtoul(taint_str, 16, &panic_on_taint))
906 		return -EINVAL;
907 
908 	/* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
909 	panic_on_taint &= TAINT_FLAGS_MAX;
910 
911 	if (!panic_on_taint)
912 		return -EINVAL;
913 
914 	if (s && !strcmp(s, "nousertaint"))
915 		panic_on_taint_nousertaint = true;
916 
917 	pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
918 		panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
919 
920 	return 0;
921 }
922 early_param("panic_on_taint", panic_on_taint_setup);
923