xref: /linux/kernel/panic.c (revision 11a92190)
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 <trace/events/error_report.h>
39 #include <asm/sections.h>
40 
41 #define PANIC_TIMER_STEP 100
42 #define PANIC_BLINK_SPD 18
43 
44 #ifdef CONFIG_SMP
45 /*
46  * Should we dump all CPUs backtraces in an oops event?
47  * Defaults to 0, can be changed via sysctl.
48  */
49 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
50 #else
51 #define sysctl_oops_all_cpu_backtrace 0
52 #endif /* CONFIG_SMP */
53 
54 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
55 static unsigned long tainted_mask =
56 	IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
57 static int pause_on_oops;
58 static int pause_on_oops_flag;
59 static DEFINE_SPINLOCK(pause_on_oops_lock);
60 bool crash_kexec_post_notifiers;
61 int panic_on_warn __read_mostly;
62 unsigned long panic_on_taint;
63 bool panic_on_taint_nousertaint = false;
64 static unsigned int warn_limit __read_mostly;
65 
66 int panic_timeout = CONFIG_PANIC_TIMEOUT;
67 EXPORT_SYMBOL_GPL(panic_timeout);
68 
69 #define PANIC_PRINT_TASK_INFO		0x00000001
70 #define PANIC_PRINT_MEM_INFO		0x00000002
71 #define PANIC_PRINT_TIMER_INFO		0x00000004
72 #define PANIC_PRINT_LOCK_INFO		0x00000008
73 #define PANIC_PRINT_FTRACE_INFO		0x00000010
74 #define PANIC_PRINT_ALL_PRINTK_MSG	0x00000020
75 #define PANIC_PRINT_ALL_CPU_BT		0x00000040
76 #define PANIC_PRINT_BLOCKED_TASKS	0x00000080
77 unsigned long panic_print;
78 
79 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
80 
81 EXPORT_SYMBOL(panic_notifier_list);
82 
83 #ifdef CONFIG_SYSCTL
84 static struct ctl_table kern_panic_table[] = {
85 #ifdef CONFIG_SMP
86 	{
87 		.procname       = "oops_all_cpu_backtrace",
88 		.data           = &sysctl_oops_all_cpu_backtrace,
89 		.maxlen         = sizeof(int),
90 		.mode           = 0644,
91 		.proc_handler   = proc_dointvec_minmax,
92 		.extra1         = SYSCTL_ZERO,
93 		.extra2         = SYSCTL_ONE,
94 	},
95 #endif
96 	{
97 		.procname       = "warn_limit",
98 		.data           = &warn_limit,
99 		.maxlen         = sizeof(warn_limit),
100 		.mode           = 0644,
101 		.proc_handler   = proc_douintvec,
102 	},
103 };
104 
kernel_panic_sysctls_init(void)105 static __init int kernel_panic_sysctls_init(void)
106 {
107 	register_sysctl_init("kernel", kern_panic_table);
108 	return 0;
109 }
110 late_initcall(kernel_panic_sysctls_init);
111 #endif
112 
113 static atomic_t warn_count = ATOMIC_INIT(0);
114 
115 #ifdef CONFIG_SYSFS
warn_count_show(struct kobject * kobj,struct kobj_attribute * attr,char * page)116 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
117 			       char *page)
118 {
119 	return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
120 }
121 
122 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
123 
kernel_panic_sysfs_init(void)124 static __init int kernel_panic_sysfs_init(void)
125 {
126 	sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
127 	return 0;
128 }
129 late_initcall(kernel_panic_sysfs_init);
130 #endif
131 
no_blink(int state)132 static long no_blink(int state)
133 {
134 	return 0;
135 }
136 
137 /* Returns how long it waited in ms */
138 long (*panic_blink)(int state);
139 EXPORT_SYMBOL(panic_blink);
140 
141 /*
142  * Stop ourself in panic -- architecture code may override this
143  */
panic_smp_self_stop(void)144 void __weak __noreturn panic_smp_self_stop(void)
145 {
146 	while (1)
147 		cpu_relax();
148 }
149 
150 /*
151  * Stop ourselves in NMI context if another CPU has already panicked. Arch code
152  * may override this to prepare for crash dumping, e.g. save regs info.
153  */
nmi_panic_self_stop(struct pt_regs * regs)154 void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
155 {
156 	panic_smp_self_stop();
157 }
158 
159 /*
160  * Stop other CPUs in panic.  Architecture dependent code may override this
161  * with more suitable version.  For example, if the architecture supports
162  * crash dump, it should save registers of each stopped CPU and disable
163  * per-CPU features such as virtualization extensions.
164  */
crash_smp_send_stop(void)165 void __weak crash_smp_send_stop(void)
166 {
167 	static int cpus_stopped;
168 
169 	/*
170 	 * This function can be called twice in panic path, but obviously
171 	 * we execute this only once.
172 	 */
173 	if (cpus_stopped)
174 		return;
175 
176 	/*
177 	 * Note smp_send_stop is the usual smp shutdown function, which
178 	 * unfortunately means it may not be hardened to work in a panic
179 	 * situation.
180 	 */
181 	smp_send_stop();
182 	cpus_stopped = 1;
183 }
184 
185 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
186 
187 /*
188  * A variant of panic() called from NMI context. We return if we've already
189  * panicked on this CPU. If another CPU already panicked, loop in
190  * nmi_panic_self_stop() which can provide architecture dependent code such
191  * as saving register state for crash dump.
192  */
nmi_panic(struct pt_regs * regs,const char * msg)193 void nmi_panic(struct pt_regs *regs, const char *msg)
194 {
195 	int old_cpu, this_cpu;
196 
197 	old_cpu = PANIC_CPU_INVALID;
198 	this_cpu = raw_smp_processor_id();
199 
200 	/* atomic_try_cmpxchg updates old_cpu on failure */
201 	if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu))
202 		panic("%s", msg);
203 	else if (old_cpu != this_cpu)
204 		nmi_panic_self_stop(regs);
205 }
206 EXPORT_SYMBOL(nmi_panic);
207 
panic_print_sys_info(bool console_flush)208 static void panic_print_sys_info(bool console_flush)
209 {
210 	if (console_flush) {
211 		if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
212 			console_flush_on_panic(CONSOLE_REPLAY_ALL);
213 		return;
214 	}
215 
216 	if (panic_print & PANIC_PRINT_TASK_INFO)
217 		show_state();
218 
219 	if (panic_print & PANIC_PRINT_MEM_INFO)
220 		show_mem();
221 
222 	if (panic_print & PANIC_PRINT_TIMER_INFO)
223 		sysrq_timer_list_show();
224 
225 	if (panic_print & PANIC_PRINT_LOCK_INFO)
226 		debug_show_all_locks();
227 
228 	if (panic_print & PANIC_PRINT_FTRACE_INFO)
229 		ftrace_dump(DUMP_ALL);
230 
231 	if (panic_print & PANIC_PRINT_BLOCKED_TASKS)
232 		show_state_filter(TASK_UNINTERRUPTIBLE);
233 }
234 
check_panic_on_warn(const char * origin)235 void check_panic_on_warn(const char *origin)
236 {
237 	unsigned int limit;
238 
239 	if (panic_on_warn)
240 		panic("%s: panic_on_warn set ...\n", origin);
241 
242 	limit = READ_ONCE(warn_limit);
243 	if (atomic_inc_return(&warn_count) >= limit && limit)
244 		panic("%s: system warned too often (kernel.warn_limit is %d)",
245 		      origin, limit);
246 }
247 
248 /*
249  * Helper that triggers the NMI backtrace (if set in panic_print)
250  * and then performs the secondary CPUs shutdown - we cannot have
251  * the NMI backtrace after the CPUs are off!
252  */
panic_other_cpus_shutdown(bool crash_kexec)253 static void panic_other_cpus_shutdown(bool crash_kexec)
254 {
255 	if (panic_print & PANIC_PRINT_ALL_CPU_BT)
256 		trigger_all_cpu_backtrace();
257 
258 	/*
259 	 * Note that smp_send_stop() is the usual SMP shutdown function,
260 	 * which unfortunately may not be hardened to work in a panic
261 	 * situation. If we want to do crash dump after notifier calls
262 	 * and kmsg_dump, we will need architecture dependent extra
263 	 * bits in addition to stopping other CPUs, hence we rely on
264 	 * crash_smp_send_stop() for that.
265 	 */
266 	if (!crash_kexec)
267 		smp_send_stop();
268 	else
269 		crash_smp_send_stop();
270 }
271 
272 /**
273  *	panic - halt the system
274  *	@fmt: The text string to print
275  *
276  *	Display a message, then perform cleanups.
277  *
278  *	This function never returns.
279  */
panic(const char * fmt,...)280 void panic(const char *fmt, ...)
281 {
282 	static char buf[1024];
283 	va_list args;
284 	long i, i_next = 0, len;
285 	int state = 0;
286 	int old_cpu, this_cpu;
287 	bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
288 
289 	if (panic_on_warn) {
290 		/*
291 		 * This thread may hit another WARN() in the panic path.
292 		 * Resetting this prevents additional WARN() from panicking the
293 		 * system on this thread.  Other threads are blocked by the
294 		 * panic_mutex in panic().
295 		 */
296 		panic_on_warn = 0;
297 	}
298 
299 	/*
300 	 * Disable local interrupts. This will prevent panic_smp_self_stop
301 	 * from deadlocking the first cpu that invokes the panic, since
302 	 * there is nothing to prevent an interrupt handler (that runs
303 	 * after setting panic_cpu) from invoking panic() again.
304 	 */
305 	local_irq_disable();
306 	preempt_disable_notrace();
307 
308 	/*
309 	 * It's possible to come here directly from a panic-assertion and
310 	 * not have preempt disabled. Some functions called from here want
311 	 * preempt to be disabled. No point enabling it later though...
312 	 *
313 	 * Only one CPU is allowed to execute the panic code from here. For
314 	 * multiple parallel invocations of panic, all other CPUs either
315 	 * stop themself or will wait until they are stopped by the 1st CPU
316 	 * with smp_send_stop().
317 	 *
318 	 * cmpxchg success means this is the 1st CPU which comes here,
319 	 * so go ahead.
320 	 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
321 	 * panic_cpu to this CPU.  In this case, this is also the 1st CPU.
322 	 */
323 	old_cpu = PANIC_CPU_INVALID;
324 	this_cpu = raw_smp_processor_id();
325 
326 	/* atomic_try_cmpxchg updates old_cpu on failure */
327 	if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
328 		/* go ahead */
329 	} else if (old_cpu != this_cpu)
330 		panic_smp_self_stop();
331 
332 	console_verbose();
333 	bust_spinlocks(1);
334 	va_start(args, fmt);
335 	len = vscnprintf(buf, sizeof(buf), fmt, args);
336 	va_end(args);
337 
338 	if (len && buf[len - 1] == '\n')
339 		buf[len - 1] = '\0';
340 
341 	pr_emerg("Kernel panic - not syncing: %s\n", buf);
342 #ifdef CONFIG_DEBUG_BUGVERBOSE
343 	/*
344 	 * Avoid nested stack-dumping if a panic occurs during oops processing
345 	 */
346 	if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
347 		dump_stack();
348 #endif
349 
350 	/*
351 	 * If kgdb is enabled, give it a chance to run before we stop all
352 	 * the other CPUs or else we won't be able to debug processes left
353 	 * running on them.
354 	 */
355 	kgdb_panic(buf);
356 
357 	/*
358 	 * If we have crashed and we have a crash kernel loaded let it handle
359 	 * everything else.
360 	 * If we want to run this after calling panic_notifiers, pass
361 	 * the "crash_kexec_post_notifiers" option to the kernel.
362 	 *
363 	 * Bypass the panic_cpu check and call __crash_kexec directly.
364 	 */
365 	if (!_crash_kexec_post_notifiers)
366 		__crash_kexec(NULL);
367 
368 	panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
369 
370 	/*
371 	 * Run any panic handlers, including those that might need to
372 	 * add information to the kmsg dump output.
373 	 */
374 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
375 
376 	panic_print_sys_info(false);
377 
378 	kmsg_dump(KMSG_DUMP_PANIC);
379 
380 	/*
381 	 * If you doubt kdump always works fine in any situation,
382 	 * "crash_kexec_post_notifiers" offers you a chance to run
383 	 * panic_notifiers and dumping kmsg before kdump.
384 	 * Note: since some panic_notifiers can make crashed kernel
385 	 * more unstable, it can increase risks of the kdump failure too.
386 	 *
387 	 * Bypass the panic_cpu check and call __crash_kexec directly.
388 	 */
389 	if (_crash_kexec_post_notifiers)
390 		__crash_kexec(NULL);
391 
392 	console_unblank();
393 
394 	/*
395 	 * We may have ended up stopping the CPU holding the lock (in
396 	 * smp_send_stop()) while still having some valuable data in the console
397 	 * buffer.  Try to acquire the lock then release it regardless of the
398 	 * result.  The release will also print the buffers out.  Locks debug
399 	 * should be disabled to avoid reporting bad unlock balance when
400 	 * panic() is not being callled from OOPS.
401 	 */
402 	debug_locks_off();
403 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
404 
405 	panic_print_sys_info(true);
406 
407 	if (!panic_blink)
408 		panic_blink = no_blink;
409 
410 	if (panic_timeout > 0) {
411 		/*
412 		 * Delay timeout seconds before rebooting the machine.
413 		 * We can't use the "normal" timers since we just panicked.
414 		 */
415 		pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
416 
417 		for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
418 			touch_nmi_watchdog();
419 			if (i >= i_next) {
420 				i += panic_blink(state ^= 1);
421 				i_next = i + 3600 / PANIC_BLINK_SPD;
422 			}
423 			mdelay(PANIC_TIMER_STEP);
424 		}
425 	}
426 	if (panic_timeout != 0) {
427 		/*
428 		 * This will not be a clean reboot, with everything
429 		 * shutting down.  But if there is a chance of
430 		 * rebooting the system it will be rebooted.
431 		 */
432 		if (panic_reboot_mode != REBOOT_UNDEFINED)
433 			reboot_mode = panic_reboot_mode;
434 		emergency_restart();
435 	}
436 #ifdef __sparc__
437 	{
438 		extern int stop_a_enabled;
439 		/* Make sure the user can actually press Stop-A (L1-A) */
440 		stop_a_enabled = 1;
441 		pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
442 			 "twice on console to return to the boot prom\n");
443 	}
444 #endif
445 #if defined(CONFIG_S390)
446 	disabled_wait();
447 #endif
448 	pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
449 
450 	/* Do not scroll important messages printed above */
451 	suppress_printk = 1;
452 
453 	/*
454 	 * The final messages may not have been printed if in a context that
455 	 * defers printing (such as NMI) and irq_work is not available.
456 	 * Explicitly flush the kernel log buffer one last time.
457 	 */
458 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
459 
460 	local_irq_enable();
461 	for (i = 0; ; i += PANIC_TIMER_STEP) {
462 		touch_softlockup_watchdog();
463 		if (i >= i_next) {
464 			i += panic_blink(state ^= 1);
465 			i_next = i + 3600 / PANIC_BLINK_SPD;
466 		}
467 		mdelay(PANIC_TIMER_STEP);
468 	}
469 }
470 
471 EXPORT_SYMBOL(panic);
472 
473 /*
474  * TAINT_FORCED_RMMOD could be a per-module flag but the module
475  * is being removed anyway.
476  */
477 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
478 	[ TAINT_PROPRIETARY_MODULE ]	= { 'P', 'G', true },
479 	[ TAINT_FORCED_MODULE ]		= { 'F', ' ', true },
480 	[ TAINT_CPU_OUT_OF_SPEC ]	= { 'S', ' ', false },
481 	[ TAINT_FORCED_RMMOD ]		= { 'R', ' ', false },
482 	[ TAINT_MACHINE_CHECK ]		= { 'M', ' ', false },
483 	[ TAINT_BAD_PAGE ]		= { 'B', ' ', false },
484 	[ TAINT_USER ]			= { 'U', ' ', false },
485 	[ TAINT_DIE ]			= { 'D', ' ', false },
486 	[ TAINT_OVERRIDDEN_ACPI_TABLE ]	= { 'A', ' ', false },
487 	[ TAINT_WARN ]			= { 'W', ' ', false },
488 	[ TAINT_CRAP ]			= { 'C', ' ', true },
489 	[ TAINT_FIRMWARE_WORKAROUND ]	= { 'I', ' ', false },
490 	[ TAINT_OOT_MODULE ]		= { 'O', ' ', true },
491 	[ TAINT_UNSIGNED_MODULE ]	= { 'E', ' ', true },
492 	[ TAINT_SOFTLOCKUP ]		= { 'L', ' ', false },
493 	[ TAINT_LIVEPATCH ]		= { 'K', ' ', true },
494 	[ TAINT_AUX ]			= { 'X', ' ', true },
495 	[ TAINT_RANDSTRUCT ]		= { 'T', ' ', true },
496 	[ TAINT_TEST ]			= { 'N', ' ', true },
497 };
498 
499 /**
500  * print_tainted - return a string to represent the kernel taint state.
501  *
502  * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
503  *
504  * The string is overwritten by the next call to print_tainted(),
505  * but is always NULL terminated.
506  */
print_tainted(void)507 const char *print_tainted(void)
508 {
509 	static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
510 
511 	BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
512 
513 	if (tainted_mask) {
514 		char *s;
515 		int i;
516 
517 		s = buf + sprintf(buf, "Tainted: ");
518 		for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
519 			const struct taint_flag *t = &taint_flags[i];
520 			*s++ = test_bit(i, &tainted_mask) ?
521 					t->c_true : t->c_false;
522 		}
523 		*s = 0;
524 	} else
525 		snprintf(buf, sizeof(buf), "Not tainted");
526 
527 	return buf;
528 }
529 
test_taint(unsigned flag)530 int test_taint(unsigned flag)
531 {
532 	return test_bit(flag, &tainted_mask);
533 }
534 EXPORT_SYMBOL(test_taint);
535 
get_taint(void)536 unsigned long get_taint(void)
537 {
538 	return tainted_mask;
539 }
540 
541 /**
542  * add_taint: add a taint flag if not already set.
543  * @flag: one of the TAINT_* constants.
544  * @lockdep_ok: whether lock debugging is still OK.
545  *
546  * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
547  * some notewortht-but-not-corrupting cases, it can be set to true.
548  */
add_taint(unsigned flag,enum lockdep_ok lockdep_ok)549 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
550 {
551 	if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
552 		pr_warn("Disabling lock debugging due to kernel taint\n");
553 
554 	set_bit(flag, &tainted_mask);
555 
556 	if (tainted_mask & panic_on_taint) {
557 		panic_on_taint = 0;
558 		panic("panic_on_taint set ...");
559 	}
560 }
561 EXPORT_SYMBOL(add_taint);
562 
spin_msec(int msecs)563 static void spin_msec(int msecs)
564 {
565 	int i;
566 
567 	for (i = 0; i < msecs; i++) {
568 		touch_nmi_watchdog();
569 		mdelay(1);
570 	}
571 }
572 
573 /*
574  * It just happens that oops_enter() and oops_exit() are identically
575  * implemented...
576  */
do_oops_enter_exit(void)577 static void do_oops_enter_exit(void)
578 {
579 	unsigned long flags;
580 	static int spin_counter;
581 
582 	if (!pause_on_oops)
583 		return;
584 
585 	spin_lock_irqsave(&pause_on_oops_lock, flags);
586 	if (pause_on_oops_flag == 0) {
587 		/* This CPU may now print the oops message */
588 		pause_on_oops_flag = 1;
589 	} else {
590 		/* We need to stall this CPU */
591 		if (!spin_counter) {
592 			/* This CPU gets to do the counting */
593 			spin_counter = pause_on_oops;
594 			do {
595 				spin_unlock(&pause_on_oops_lock);
596 				spin_msec(MSEC_PER_SEC);
597 				spin_lock(&pause_on_oops_lock);
598 			} while (--spin_counter);
599 			pause_on_oops_flag = 0;
600 		} else {
601 			/* This CPU waits for a different one */
602 			while (spin_counter) {
603 				spin_unlock(&pause_on_oops_lock);
604 				spin_msec(1);
605 				spin_lock(&pause_on_oops_lock);
606 			}
607 		}
608 	}
609 	spin_unlock_irqrestore(&pause_on_oops_lock, flags);
610 }
611 
612 /*
613  * Return true if the calling CPU is allowed to print oops-related info.
614  * This is a bit racy..
615  */
oops_may_print(void)616 bool oops_may_print(void)
617 {
618 	return pause_on_oops_flag == 0;
619 }
620 
621 /*
622  * Called when the architecture enters its oops handler, before it prints
623  * anything.  If this is the first CPU to oops, and it's oopsing the first
624  * time then let it proceed.
625  *
626  * This is all enabled by the pause_on_oops kernel boot option.  We do all
627  * this to ensure that oopses don't scroll off the screen.  It has the
628  * side-effect of preventing later-oopsing CPUs from mucking up the display,
629  * too.
630  *
631  * It turns out that the CPU which is allowed to print ends up pausing for
632  * the right duration, whereas all the other CPUs pause for twice as long:
633  * once in oops_enter(), once in oops_exit().
634  */
oops_enter(void)635 void oops_enter(void)
636 {
637 	tracing_off();
638 	/* can't trust the integrity of the kernel anymore: */
639 	debug_locks_off();
640 	do_oops_enter_exit();
641 
642 	if (sysctl_oops_all_cpu_backtrace)
643 		trigger_all_cpu_backtrace();
644 }
645 
print_oops_end_marker(void)646 static void print_oops_end_marker(void)
647 {
648 	pr_warn("---[ end trace %016llx ]---\n", 0ULL);
649 }
650 
651 /*
652  * Called when the architecture exits its oops handler, after printing
653  * everything.
654  */
oops_exit(void)655 void oops_exit(void)
656 {
657 	do_oops_enter_exit();
658 	print_oops_end_marker();
659 	kmsg_dump(KMSG_DUMP_OOPS);
660 }
661 
662 struct warn_args {
663 	const char *fmt;
664 	va_list args;
665 };
666 
__warn(const char * file,int line,void * caller,unsigned taint,struct pt_regs * regs,struct warn_args * args)667 void __warn(const char *file, int line, void *caller, unsigned taint,
668 	    struct pt_regs *regs, struct warn_args *args)
669 {
670 	disable_trace_on_warning();
671 
672 	if (file)
673 		pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
674 			raw_smp_processor_id(), current->pid, file, line,
675 			caller);
676 	else
677 		pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
678 			raw_smp_processor_id(), current->pid, caller);
679 
680 #pragma GCC diagnostic push
681 #ifndef __clang__
682 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
683 #endif
684 	if (args)
685 		vprintk(args->fmt, args->args);
686 #pragma GCC diagnostic pop
687 
688 	print_modules();
689 
690 	if (regs)
691 		show_regs(regs);
692 
693 	check_panic_on_warn("kernel");
694 
695 	if (!regs)
696 		dump_stack();
697 
698 	print_irqtrace_events(current);
699 
700 	print_oops_end_marker();
701 	trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
702 
703 	/* Just a warning, don't kill lockdep. */
704 	add_taint(taint, LOCKDEP_STILL_OK);
705 }
706 
707 #ifdef CONFIG_BUG
708 #ifndef __WARN_FLAGS
warn_slowpath_fmt(const char * file,int line,unsigned taint,const char * fmt,...)709 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
710 		       const char *fmt, ...)
711 {
712 	bool rcu = warn_rcu_enter();
713 	struct warn_args args;
714 
715 	pr_warn(CUT_HERE);
716 
717 	if (!fmt) {
718 		__warn(file, line, __builtin_return_address(0), taint,
719 		       NULL, NULL);
720 		warn_rcu_exit(rcu);
721 		return;
722 	}
723 
724 	args.fmt = fmt;
725 	va_start(args.args, fmt);
726 	__warn(file, line, __builtin_return_address(0), taint, NULL, &args);
727 	va_end(args.args);
728 	warn_rcu_exit(rcu);
729 }
730 EXPORT_SYMBOL(warn_slowpath_fmt);
731 #else
__warn_printk(const char * fmt,...)732 void __warn_printk(const char *fmt, ...)
733 {
734 	bool rcu = warn_rcu_enter();
735 	va_list args;
736 
737 	pr_warn(CUT_HERE);
738 
739 	va_start(args, fmt);
740 	vprintk(fmt, args);
741 	va_end(args);
742 	warn_rcu_exit(rcu);
743 }
744 EXPORT_SYMBOL(__warn_printk);
745 #endif
746 
747 /* Support resetting WARN*_ONCE state */
748 
clear_warn_once_set(void * data,u64 val)749 static int clear_warn_once_set(void *data, u64 val)
750 {
751 	generic_bug_clear_once();
752 	memset(__start_once, 0, __end_once - __start_once);
753 	return 0;
754 }
755 
756 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
757 			 "%lld\n");
758 
register_warn_debugfs(void)759 static __init int register_warn_debugfs(void)
760 {
761 	/* Don't care about failure */
762 	debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
763 				   &clear_warn_once_fops);
764 	return 0;
765 }
766 
767 device_initcall(register_warn_debugfs);
768 #endif
769 
770 #ifdef CONFIG_STACKPROTECTOR
771 
772 /*
773  * Called when gcc's -fstack-protector feature is used, and
774  * gcc detects corruption of the on-stack canary value
775  */
__stack_chk_fail(void)776 __visible noinstr void __stack_chk_fail(void)
777 {
778 	instrumentation_begin();
779 	panic("stack-protector: Kernel stack is corrupted in: %pB",
780 		__builtin_return_address(0));
781 	instrumentation_end();
782 }
783 EXPORT_SYMBOL(__stack_chk_fail);
784 
785 #endif
786 
787 core_param(panic, panic_timeout, int, 0644);
788 core_param(panic_print, panic_print, ulong, 0644);
789 core_param(pause_on_oops, pause_on_oops, int, 0644);
790 core_param(panic_on_warn, panic_on_warn, int, 0644);
791 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
792 
oops_setup(char * s)793 static int __init oops_setup(char *s)
794 {
795 	if (!s)
796 		return -EINVAL;
797 	if (!strcmp(s, "panic"))
798 		panic_on_oops = 1;
799 	return 0;
800 }
801 early_param("oops", oops_setup);
802 
panic_on_taint_setup(char * s)803 static int __init panic_on_taint_setup(char *s)
804 {
805 	char *taint_str;
806 
807 	if (!s)
808 		return -EINVAL;
809 
810 	taint_str = strsep(&s, ",");
811 	if (kstrtoul(taint_str, 16, &panic_on_taint))
812 		return -EINVAL;
813 
814 	/* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
815 	panic_on_taint &= TAINT_FLAGS_MAX;
816 
817 	if (!panic_on_taint)
818 		return -EINVAL;
819 
820 	if (s && !strcmp(s, "nousertaint"))
821 		panic_on_taint_nousertaint = true;
822 
823 	pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
824 		panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
825 
826 	return 0;
827 }
828 early_param("panic_on_taint", panic_on_taint_setup);
829