Loading...
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 */
49static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
50#else
51#define sysctl_oops_all_cpu_backtrace 0
52#endif /* CONFIG_SMP */
53
54int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
55static unsigned long tainted_mask =
56 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
57static int pause_on_oops;
58static int pause_on_oops_flag;
59static DEFINE_SPINLOCK(pause_on_oops_lock);
60bool crash_kexec_post_notifiers;
61int panic_on_warn __read_mostly;
62unsigned long panic_on_taint;
63bool panic_on_taint_nousertaint = false;
64static unsigned int warn_limit __read_mostly;
65
66int panic_timeout = CONFIG_PANIC_TIMEOUT;
67EXPORT_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
77unsigned long panic_print;
78
79ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
80
81EXPORT_SYMBOL(panic_notifier_list);
82
83#ifdef CONFIG_SYSCTL
84static 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};
105
106static __init int kernel_panic_sysctls_init(void)
107{
108 register_sysctl_init("kernel", kern_panic_table);
109 return 0;
110}
111late_initcall(kernel_panic_sysctls_init);
112#endif
113
114static atomic_t warn_count = ATOMIC_INIT(0);
115
116#ifdef CONFIG_SYSFS
117static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
118 char *page)
119{
120 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
121}
122
123static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
124
125static __init int kernel_panic_sysfs_init(void)
126{
127 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
128 return 0;
129}
130late_initcall(kernel_panic_sysfs_init);
131#endif
132
133static long no_blink(int state)
134{
135 return 0;
136}
137
138/* Returns how long it waited in ms */
139long (*panic_blink)(int state);
140EXPORT_SYMBOL(panic_blink);
141
142/*
143 * Stop ourself in panic -- architecture code may override this
144 */
145void __weak __noreturn panic_smp_self_stop(void)
146{
147 while (1)
148 cpu_relax();
149}
150
151/*
152 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
153 * may override this to prepare for crash dumping, e.g. save regs info.
154 */
155void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
156{
157 panic_smp_self_stop();
158}
159
160/*
161 * Stop other CPUs in panic. Architecture dependent code may override this
162 * with more suitable version. For example, if the architecture supports
163 * crash dump, it should save registers of each stopped CPU and disable
164 * per-CPU features such as virtualization extensions.
165 */
166void __weak crash_smp_send_stop(void)
167{
168 static int cpus_stopped;
169
170 /*
171 * This function can be called twice in panic path, but obviously
172 * we execute this only once.
173 */
174 if (cpus_stopped)
175 return;
176
177 /*
178 * Note smp_send_stop is the usual smp shutdown function, which
179 * unfortunately means it may not be hardened to work in a panic
180 * situation.
181 */
182 smp_send_stop();
183 cpus_stopped = 1;
184}
185
186atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
187
188/*
189 * A variant of panic() called from NMI context. We return if we've already
190 * panicked on this CPU. If another CPU already panicked, loop in
191 * nmi_panic_self_stop() which can provide architecture dependent code such
192 * as saving register state for crash dump.
193 */
194void nmi_panic(struct pt_regs *regs, const char *msg)
195{
196 int old_cpu, this_cpu;
197
198 old_cpu = PANIC_CPU_INVALID;
199 this_cpu = raw_smp_processor_id();
200
201 /* atomic_try_cmpxchg updates old_cpu on failure */
202 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu))
203 panic("%s", msg);
204 else if (old_cpu != this_cpu)
205 nmi_panic_self_stop(regs);
206}
207EXPORT_SYMBOL(nmi_panic);
208
209static void panic_print_sys_info(bool console_flush)
210{
211 if (console_flush) {
212 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
213 console_flush_on_panic(CONSOLE_REPLAY_ALL);
214 return;
215 }
216
217 if (panic_print & PANIC_PRINT_TASK_INFO)
218 show_state();
219
220 if (panic_print & PANIC_PRINT_MEM_INFO)
221 show_mem();
222
223 if (panic_print & PANIC_PRINT_TIMER_INFO)
224 sysrq_timer_list_show();
225
226 if (panic_print & PANIC_PRINT_LOCK_INFO)
227 debug_show_all_locks();
228
229 if (panic_print & PANIC_PRINT_FTRACE_INFO)
230 ftrace_dump(DUMP_ALL);
231
232 if (panic_print & PANIC_PRINT_BLOCKED_TASKS)
233 show_state_filter(TASK_UNINTERRUPTIBLE);
234}
235
236void check_panic_on_warn(const char *origin)
237{
238 unsigned int limit;
239
240 if (panic_on_warn)
241 panic("%s: panic_on_warn set ...\n", origin);
242
243 limit = READ_ONCE(warn_limit);
244 if (atomic_inc_return(&warn_count) >= limit && limit)
245 panic("%s: system warned too often (kernel.warn_limit is %d)",
246 origin, limit);
247}
248
249/*
250 * Helper that triggers the NMI backtrace (if set in panic_print)
251 * and then performs the secondary CPUs shutdown - we cannot have
252 * the NMI backtrace after the CPUs are off!
253 */
254static void panic_other_cpus_shutdown(bool crash_kexec)
255{
256 if (panic_print & PANIC_PRINT_ALL_CPU_BT)
257 trigger_all_cpu_backtrace();
258
259 /*
260 * Note that smp_send_stop() is the usual SMP shutdown function,
261 * which unfortunately may not be hardened to work in a panic
262 * situation. If we want to do crash dump after notifier calls
263 * and kmsg_dump, we will need architecture dependent extra
264 * bits in addition to stopping other CPUs, hence we rely on
265 * crash_smp_send_stop() for that.
266 */
267 if (!crash_kexec)
268 smp_send_stop();
269 else
270 crash_smp_send_stop();
271}
272
273/**
274 * panic - halt the system
275 * @fmt: The text string to print
276 *
277 * Display a message, then perform cleanups.
278 *
279 * This function never returns.
280 */
281void panic(const char *fmt, ...)
282{
283 static char buf[1024];
284 va_list args;
285 long i, i_next = 0, len;
286 int state = 0;
287 int old_cpu, this_cpu;
288 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
289
290 if (panic_on_warn) {
291 /*
292 * This thread may hit another WARN() in the panic path.
293 * Resetting this prevents additional WARN() from panicking the
294 * system on this thread. Other threads are blocked by the
295 * panic_mutex in panic().
296 */
297 panic_on_warn = 0;
298 }
299
300 /*
301 * Disable local interrupts. This will prevent panic_smp_self_stop
302 * from deadlocking the first cpu that invokes the panic, since
303 * there is nothing to prevent an interrupt handler (that runs
304 * after setting panic_cpu) from invoking panic() again.
305 */
306 local_irq_disable();
307 preempt_disable_notrace();
308
309 /*
310 * It's possible to come here directly from a panic-assertion and
311 * not have preempt disabled. Some functions called from here want
312 * preempt to be disabled. No point enabling it later though...
313 *
314 * Only one CPU is allowed to execute the panic code from here. For
315 * multiple parallel invocations of panic, all other CPUs either
316 * stop themself or will wait until they are stopped by the 1st CPU
317 * with smp_send_stop().
318 *
319 * cmpxchg success means this is the 1st CPU which comes here,
320 * so go ahead.
321 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
322 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
323 */
324 old_cpu = PANIC_CPU_INVALID;
325 this_cpu = raw_smp_processor_id();
326
327 /* atomic_try_cmpxchg updates old_cpu on failure */
328 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
329 /* go ahead */
330 } else if (old_cpu != this_cpu)
331 panic_smp_self_stop();
332
333 console_verbose();
334 bust_spinlocks(1);
335 va_start(args, fmt);
336 len = vscnprintf(buf, sizeof(buf), fmt, args);
337 va_end(args);
338
339 if (len && buf[len - 1] == '\n')
340 buf[len - 1] = '\0';
341
342 pr_emerg("Kernel panic - not syncing: %s\n", buf);
343#ifdef CONFIG_DEBUG_BUGVERBOSE
344 /*
345 * Avoid nested stack-dumping if a panic occurs during oops processing
346 */
347 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
348 dump_stack();
349#endif
350
351 /*
352 * If kgdb is enabled, give it a chance to run before we stop all
353 * the other CPUs or else we won't be able to debug processes left
354 * running on them.
355 */
356 kgdb_panic(buf);
357
358 /*
359 * If we have crashed and we have a crash kernel loaded let it handle
360 * everything else.
361 * If we want to run this after calling panic_notifiers, pass
362 * the "crash_kexec_post_notifiers" option to the kernel.
363 *
364 * Bypass the panic_cpu check and call __crash_kexec directly.
365 */
366 if (!_crash_kexec_post_notifiers)
367 __crash_kexec(NULL);
368
369 panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
370
371 /*
372 * Run any panic handlers, including those that might need to
373 * add information to the kmsg dump output.
374 */
375 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
376
377 panic_print_sys_info(false);
378
379 kmsg_dump(KMSG_DUMP_PANIC);
380
381 /*
382 * If you doubt kdump always works fine in any situation,
383 * "crash_kexec_post_notifiers" offers you a chance to run
384 * panic_notifiers and dumping kmsg before kdump.
385 * Note: since some panic_notifiers can make crashed kernel
386 * more unstable, it can increase risks of the kdump failure too.
387 *
388 * Bypass the panic_cpu check and call __crash_kexec directly.
389 */
390 if (_crash_kexec_post_notifiers)
391 __crash_kexec(NULL);
392
393 console_unblank();
394
395 /*
396 * We may have ended up stopping the CPU holding the lock (in
397 * smp_send_stop()) while still having some valuable data in the console
398 * buffer. Try to acquire the lock then release it regardless of the
399 * result. The release will also print the buffers out. Locks debug
400 * should be disabled to avoid reporting bad unlock balance when
401 * panic() is not being callled from OOPS.
402 */
403 debug_locks_off();
404 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
405
406 panic_print_sys_info(true);
407
408 if (!panic_blink)
409 panic_blink = no_blink;
410
411 if (panic_timeout > 0) {
412 /*
413 * Delay timeout seconds before rebooting the machine.
414 * We can't use the "normal" timers since we just panicked.
415 */
416 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
417
418 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
419 touch_nmi_watchdog();
420 if (i >= i_next) {
421 i += panic_blink(state ^= 1);
422 i_next = i + 3600 / PANIC_BLINK_SPD;
423 }
424 mdelay(PANIC_TIMER_STEP);
425 }
426 }
427 if (panic_timeout != 0) {
428 /*
429 * This will not be a clean reboot, with everything
430 * shutting down. But if there is a chance of
431 * rebooting the system it will be rebooted.
432 */
433 if (panic_reboot_mode != REBOOT_UNDEFINED)
434 reboot_mode = panic_reboot_mode;
435 emergency_restart();
436 }
437#ifdef __sparc__
438 {
439 extern int stop_a_enabled;
440 /* Make sure the user can actually press Stop-A (L1-A) */
441 stop_a_enabled = 1;
442 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
443 "twice on console to return to the boot prom\n");
444 }
445#endif
446#if defined(CONFIG_S390)
447 disabled_wait();
448#endif
449 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
450
451 /* Do not scroll important messages printed above */
452 suppress_printk = 1;
453
454 /*
455 * The final messages may not have been printed if in a context that
456 * defers printing (such as NMI) and irq_work is not available.
457 * Explicitly flush the kernel log buffer one last time.
458 */
459 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
460
461 local_irq_enable();
462 for (i = 0; ; i += PANIC_TIMER_STEP) {
463 touch_softlockup_watchdog();
464 if (i >= i_next) {
465 i += panic_blink(state ^= 1);
466 i_next = i + 3600 / PANIC_BLINK_SPD;
467 }
468 mdelay(PANIC_TIMER_STEP);
469 }
470}
471
472EXPORT_SYMBOL(panic);
473
474/*
475 * TAINT_FORCED_RMMOD could be a per-module flag but the module
476 * is being removed anyway.
477 */
478const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
479 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
480 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
481 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
482 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
483 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
484 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
485 [ TAINT_USER ] = { 'U', ' ', false },
486 [ TAINT_DIE ] = { 'D', ' ', false },
487 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
488 [ TAINT_WARN ] = { 'W', ' ', false },
489 [ TAINT_CRAP ] = { 'C', ' ', true },
490 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
491 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
492 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
493 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
494 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
495 [ TAINT_AUX ] = { 'X', ' ', true },
496 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
497 [ TAINT_TEST ] = { 'N', ' ', true },
498};
499
500/**
501 * print_tainted - return a string to represent the kernel taint state.
502 *
503 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
504 *
505 * The string is overwritten by the next call to print_tainted(),
506 * but is always NULL terminated.
507 */
508const char *print_tainted(void)
509{
510 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
511
512 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
513
514 if (tainted_mask) {
515 char *s;
516 int i;
517
518 s = buf + sprintf(buf, "Tainted: ");
519 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
520 const struct taint_flag *t = &taint_flags[i];
521 *s++ = test_bit(i, &tainted_mask) ?
522 t->c_true : t->c_false;
523 }
524 *s = 0;
525 } else
526 snprintf(buf, sizeof(buf), "Not tainted");
527
528 return buf;
529}
530
531int test_taint(unsigned flag)
532{
533 return test_bit(flag, &tainted_mask);
534}
535EXPORT_SYMBOL(test_taint);
536
537unsigned long get_taint(void)
538{
539 return tainted_mask;
540}
541
542/**
543 * add_taint: add a taint flag if not already set.
544 * @flag: one of the TAINT_* constants.
545 * @lockdep_ok: whether lock debugging is still OK.
546 *
547 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
548 * some notewortht-but-not-corrupting cases, it can be set to true.
549 */
550void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
551{
552 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
553 pr_warn("Disabling lock debugging due to kernel taint\n");
554
555 set_bit(flag, &tainted_mask);
556
557 if (tainted_mask & panic_on_taint) {
558 panic_on_taint = 0;
559 panic("panic_on_taint set ...");
560 }
561}
562EXPORT_SYMBOL(add_taint);
563
564static void spin_msec(int msecs)
565{
566 int i;
567
568 for (i = 0; i < msecs; i++) {
569 touch_nmi_watchdog();
570 mdelay(1);
571 }
572}
573
574/*
575 * It just happens that oops_enter() and oops_exit() are identically
576 * implemented...
577 */
578static void do_oops_enter_exit(void)
579{
580 unsigned long flags;
581 static int spin_counter;
582
583 if (!pause_on_oops)
584 return;
585
586 spin_lock_irqsave(&pause_on_oops_lock, flags);
587 if (pause_on_oops_flag == 0) {
588 /* This CPU may now print the oops message */
589 pause_on_oops_flag = 1;
590 } else {
591 /* We need to stall this CPU */
592 if (!spin_counter) {
593 /* This CPU gets to do the counting */
594 spin_counter = pause_on_oops;
595 do {
596 spin_unlock(&pause_on_oops_lock);
597 spin_msec(MSEC_PER_SEC);
598 spin_lock(&pause_on_oops_lock);
599 } while (--spin_counter);
600 pause_on_oops_flag = 0;
601 } else {
602 /* This CPU waits for a different one */
603 while (spin_counter) {
604 spin_unlock(&pause_on_oops_lock);
605 spin_msec(1);
606 spin_lock(&pause_on_oops_lock);
607 }
608 }
609 }
610 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
611}
612
613/*
614 * Return true if the calling CPU is allowed to print oops-related info.
615 * This is a bit racy..
616 */
617bool oops_may_print(void)
618{
619 return pause_on_oops_flag == 0;
620}
621
622/*
623 * Called when the architecture enters its oops handler, before it prints
624 * anything. If this is the first CPU to oops, and it's oopsing the first
625 * time then let it proceed.
626 *
627 * This is all enabled by the pause_on_oops kernel boot option. We do all
628 * this to ensure that oopses don't scroll off the screen. It has the
629 * side-effect of preventing later-oopsing CPUs from mucking up the display,
630 * too.
631 *
632 * It turns out that the CPU which is allowed to print ends up pausing for
633 * the right duration, whereas all the other CPUs pause for twice as long:
634 * once in oops_enter(), once in oops_exit().
635 */
636void oops_enter(void)
637{
638 tracing_off();
639 /* can't trust the integrity of the kernel anymore: */
640 debug_locks_off();
641 do_oops_enter_exit();
642
643 if (sysctl_oops_all_cpu_backtrace)
644 trigger_all_cpu_backtrace();
645}
646
647static void print_oops_end_marker(void)
648{
649 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
650}
651
652/*
653 * Called when the architecture exits its oops handler, after printing
654 * everything.
655 */
656void oops_exit(void)
657{
658 do_oops_enter_exit();
659 print_oops_end_marker();
660 kmsg_dump(KMSG_DUMP_OOPS);
661}
662
663struct warn_args {
664 const char *fmt;
665 va_list args;
666};
667
668void __warn(const char *file, int line, void *caller, unsigned taint,
669 struct pt_regs *regs, struct warn_args *args)
670{
671 disable_trace_on_warning();
672
673 if (file)
674 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
675 raw_smp_processor_id(), current->pid, file, line,
676 caller);
677 else
678 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
679 raw_smp_processor_id(), current->pid, caller);
680
681#pragma GCC diagnostic push
682#ifndef __clang__
683#pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
684#endif
685 if (args)
686 vprintk(args->fmt, args->args);
687#pragma GCC diagnostic pop
688
689 print_modules();
690
691 if (regs)
692 show_regs(regs);
693
694 check_panic_on_warn("kernel");
695
696 if (!regs)
697 dump_stack();
698
699 print_irqtrace_events(current);
700
701 print_oops_end_marker();
702 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
703
704 /* Just a warning, don't kill lockdep. */
705 add_taint(taint, LOCKDEP_STILL_OK);
706}
707
708#ifdef CONFIG_BUG
709#ifndef __WARN_FLAGS
710void warn_slowpath_fmt(const char *file, int line, unsigned taint,
711 const char *fmt, ...)
712{
713 bool rcu = warn_rcu_enter();
714 struct warn_args args;
715
716 pr_warn(CUT_HERE);
717
718 if (!fmt) {
719 __warn(file, line, __builtin_return_address(0), taint,
720 NULL, NULL);
721 warn_rcu_exit(rcu);
722 return;
723 }
724
725 args.fmt = fmt;
726 va_start(args.args, fmt);
727 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
728 va_end(args.args);
729 warn_rcu_exit(rcu);
730}
731EXPORT_SYMBOL(warn_slowpath_fmt);
732#else
733void __warn_printk(const char *fmt, ...)
734{
735 bool rcu = warn_rcu_enter();
736 va_list args;
737
738 pr_warn(CUT_HERE);
739
740 va_start(args, fmt);
741 vprintk(fmt, args);
742 va_end(args);
743 warn_rcu_exit(rcu);
744}
745EXPORT_SYMBOL(__warn_printk);
746#endif
747
748/* Support resetting WARN*_ONCE state */
749
750static int clear_warn_once_set(void *data, u64 val)
751{
752 generic_bug_clear_once();
753 memset(__start_once, 0, __end_once - __start_once);
754 return 0;
755}
756
757DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
758 "%lld\n");
759
760static __init int register_warn_debugfs(void)
761{
762 /* Don't care about failure */
763 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
764 &clear_warn_once_fops);
765 return 0;
766}
767
768device_initcall(register_warn_debugfs);
769#endif
770
771#ifdef CONFIG_STACKPROTECTOR
772
773/*
774 * Called when gcc's -fstack-protector feature is used, and
775 * gcc detects corruption of the on-stack canary value
776 */
777__visible noinstr void __stack_chk_fail(void)
778{
779 instrumentation_begin();
780 panic("stack-protector: Kernel stack is corrupted in: %pB",
781 __builtin_return_address(0));
782 instrumentation_end();
783}
784EXPORT_SYMBOL(__stack_chk_fail);
785
786#endif
787
788core_param(panic, panic_timeout, int, 0644);
789core_param(panic_print, panic_print, ulong, 0644);
790core_param(pause_on_oops, pause_on_oops, int, 0644);
791core_param(panic_on_warn, panic_on_warn, int, 0644);
792core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
793
794static int __init oops_setup(char *s)
795{
796 if (!s)
797 return -EINVAL;
798 if (!strcmp(s, "panic"))
799 panic_on_oops = 1;
800 return 0;
801}
802early_param("oops", oops_setup);
803
804static int __init panic_on_taint_setup(char *s)
805{
806 char *taint_str;
807
808 if (!s)
809 return -EINVAL;
810
811 taint_str = strsep(&s, ",");
812 if (kstrtoul(taint_str, 16, &panic_on_taint))
813 return -EINVAL;
814
815 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
816 panic_on_taint &= TAINT_FLAGS_MAX;
817
818 if (!panic_on_taint)
819 return -EINVAL;
820
821 if (s && !strcmp(s, "nousertaint"))
822 panic_on_taint_nousertaint = true;
823
824 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
825 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
826
827 return 0;
828}
829early_param("panic_on_taint", panic_on_taint_setup);
1/*
2 * linux/kernel/panic.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * This function is used through-out the kernel (including mm and fs)
9 * to indicate a major problem.
10 */
11#include <linux/debug_locks.h>
12#include <linux/interrupt.h>
13#include <linux/kmsg_dump.h>
14#include <linux/kallsyms.h>
15#include <linux/notifier.h>
16#include <linux/module.h>
17#include <linux/random.h>
18#include <linux/ftrace.h>
19#include <linux/reboot.h>
20#include <linux/delay.h>
21#include <linux/kexec.h>
22#include <linux/sched.h>
23#include <linux/sysrq.h>
24#include <linux/init.h>
25#include <linux/nmi.h>
26#include <linux/console.h>
27#include <linux/bug.h>
28
29#define PANIC_TIMER_STEP 100
30#define PANIC_BLINK_SPD 18
31
32int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
33static unsigned long tainted_mask;
34static int pause_on_oops;
35static int pause_on_oops_flag;
36static DEFINE_SPINLOCK(pause_on_oops_lock);
37bool crash_kexec_post_notifiers;
38int panic_on_warn __read_mostly;
39
40int panic_timeout = CONFIG_PANIC_TIMEOUT;
41EXPORT_SYMBOL_GPL(panic_timeout);
42
43ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
44
45EXPORT_SYMBOL(panic_notifier_list);
46
47static long no_blink(int state)
48{
49 return 0;
50}
51
52/* Returns how long it waited in ms */
53long (*panic_blink)(int state);
54EXPORT_SYMBOL(panic_blink);
55
56/*
57 * Stop ourself in panic -- architecture code may override this
58 */
59void __weak panic_smp_self_stop(void)
60{
61 while (1)
62 cpu_relax();
63}
64
65/*
66 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
67 * may override this to prepare for crash dumping, e.g. save regs info.
68 */
69void __weak nmi_panic_self_stop(struct pt_regs *regs)
70{
71 panic_smp_self_stop();
72}
73
74atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
75
76/*
77 * A variant of panic() called from NMI context. We return if we've already
78 * panicked on this CPU. If another CPU already panicked, loop in
79 * nmi_panic_self_stop() which can provide architecture dependent code such
80 * as saving register state for crash dump.
81 */
82void nmi_panic(struct pt_regs *regs, const char *msg)
83{
84 int old_cpu, cpu;
85
86 cpu = raw_smp_processor_id();
87 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
88
89 if (old_cpu == PANIC_CPU_INVALID)
90 panic("%s", msg);
91 else if (old_cpu != cpu)
92 nmi_panic_self_stop(regs);
93}
94EXPORT_SYMBOL(nmi_panic);
95
96/**
97 * panic - halt the system
98 * @fmt: The text string to print
99 *
100 * Display a message, then perform cleanups.
101 *
102 * This function never returns.
103 */
104void panic(const char *fmt, ...)
105{
106 static char buf[1024];
107 va_list args;
108 long i, i_next = 0;
109 int state = 0;
110 int old_cpu, this_cpu;
111
112 /*
113 * Disable local interrupts. This will prevent panic_smp_self_stop
114 * from deadlocking the first cpu that invokes the panic, since
115 * there is nothing to prevent an interrupt handler (that runs
116 * after setting panic_cpu) from invoking panic() again.
117 */
118 local_irq_disable();
119
120 /*
121 * It's possible to come here directly from a panic-assertion and
122 * not have preempt disabled. Some functions called from here want
123 * preempt to be disabled. No point enabling it later though...
124 *
125 * Only one CPU is allowed to execute the panic code from here. For
126 * multiple parallel invocations of panic, all other CPUs either
127 * stop themself or will wait until they are stopped by the 1st CPU
128 * with smp_send_stop().
129 *
130 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
131 * comes here, so go ahead.
132 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
133 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
134 */
135 this_cpu = raw_smp_processor_id();
136 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
137
138 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
139 panic_smp_self_stop();
140
141 console_verbose();
142 bust_spinlocks(1);
143 va_start(args, fmt);
144 vsnprintf(buf, sizeof(buf), fmt, args);
145 va_end(args);
146 pr_emerg("Kernel panic - not syncing: %s\n", buf);
147#ifdef CONFIG_DEBUG_BUGVERBOSE
148 /*
149 * Avoid nested stack-dumping if a panic occurs during oops processing
150 */
151 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
152 dump_stack();
153#endif
154
155 /*
156 * If we have crashed and we have a crash kernel loaded let it handle
157 * everything else.
158 * If we want to run this after calling panic_notifiers, pass
159 * the "crash_kexec_post_notifiers" option to the kernel.
160 *
161 * Bypass the panic_cpu check and call __crash_kexec directly.
162 */
163 if (!crash_kexec_post_notifiers)
164 __crash_kexec(NULL);
165
166 /*
167 * Note smp_send_stop is the usual smp shutdown function, which
168 * unfortunately means it may not be hardened to work in a panic
169 * situation.
170 */
171 smp_send_stop();
172
173 /*
174 * Run any panic handlers, including those that might need to
175 * add information to the kmsg dump output.
176 */
177 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
178
179 kmsg_dump(KMSG_DUMP_PANIC);
180
181 /*
182 * If you doubt kdump always works fine in any situation,
183 * "crash_kexec_post_notifiers" offers you a chance to run
184 * panic_notifiers and dumping kmsg before kdump.
185 * Note: since some panic_notifiers can make crashed kernel
186 * more unstable, it can increase risks of the kdump failure too.
187 *
188 * Bypass the panic_cpu check and call __crash_kexec directly.
189 */
190 if (crash_kexec_post_notifiers)
191 __crash_kexec(NULL);
192
193 bust_spinlocks(0);
194
195 /*
196 * We may have ended up stopping the CPU holding the lock (in
197 * smp_send_stop()) while still having some valuable data in the console
198 * buffer. Try to acquire the lock then release it regardless of the
199 * result. The release will also print the buffers out. Locks debug
200 * should be disabled to avoid reporting bad unlock balance when
201 * panic() is not being callled from OOPS.
202 */
203 debug_locks_off();
204 console_flush_on_panic();
205
206 if (!panic_blink)
207 panic_blink = no_blink;
208
209 if (panic_timeout > 0) {
210 /*
211 * Delay timeout seconds before rebooting the machine.
212 * We can't use the "normal" timers since we just panicked.
213 */
214 pr_emerg("Rebooting in %d seconds..", panic_timeout);
215
216 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
217 touch_nmi_watchdog();
218 if (i >= i_next) {
219 i += panic_blink(state ^= 1);
220 i_next = i + 3600 / PANIC_BLINK_SPD;
221 }
222 mdelay(PANIC_TIMER_STEP);
223 }
224 }
225 if (panic_timeout != 0) {
226 /*
227 * This will not be a clean reboot, with everything
228 * shutting down. But if there is a chance of
229 * rebooting the system it will be rebooted.
230 */
231 emergency_restart();
232 }
233#ifdef __sparc__
234 {
235 extern int stop_a_enabled;
236 /* Make sure the user can actually press Stop-A (L1-A) */
237 stop_a_enabled = 1;
238 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
239 }
240#endif
241#if defined(CONFIG_S390)
242 {
243 unsigned long caller;
244
245 caller = (unsigned long)__builtin_return_address(0);
246 disabled_wait(caller);
247 }
248#endif
249 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
250 local_irq_enable();
251 for (i = 0; ; i += PANIC_TIMER_STEP) {
252 touch_softlockup_watchdog();
253 if (i >= i_next) {
254 i += panic_blink(state ^= 1);
255 i_next = i + 3600 / PANIC_BLINK_SPD;
256 }
257 mdelay(PANIC_TIMER_STEP);
258 }
259}
260
261EXPORT_SYMBOL(panic);
262
263
264struct tnt {
265 u8 bit;
266 char true;
267 char false;
268};
269
270static const struct tnt tnts[] = {
271 { TAINT_PROPRIETARY_MODULE, 'P', 'G' },
272 { TAINT_FORCED_MODULE, 'F', ' ' },
273 { TAINT_CPU_OUT_OF_SPEC, 'S', ' ' },
274 { TAINT_FORCED_RMMOD, 'R', ' ' },
275 { TAINT_MACHINE_CHECK, 'M', ' ' },
276 { TAINT_BAD_PAGE, 'B', ' ' },
277 { TAINT_USER, 'U', ' ' },
278 { TAINT_DIE, 'D', ' ' },
279 { TAINT_OVERRIDDEN_ACPI_TABLE, 'A', ' ' },
280 { TAINT_WARN, 'W', ' ' },
281 { TAINT_CRAP, 'C', ' ' },
282 { TAINT_FIRMWARE_WORKAROUND, 'I', ' ' },
283 { TAINT_OOT_MODULE, 'O', ' ' },
284 { TAINT_UNSIGNED_MODULE, 'E', ' ' },
285 { TAINT_SOFTLOCKUP, 'L', ' ' },
286 { TAINT_LIVEPATCH, 'K', ' ' },
287};
288
289/**
290 * print_tainted - return a string to represent the kernel taint state.
291 *
292 * 'P' - Proprietary module has been loaded.
293 * 'F' - Module has been forcibly loaded.
294 * 'S' - SMP with CPUs not designed for SMP.
295 * 'R' - User forced a module unload.
296 * 'M' - System experienced a machine check exception.
297 * 'B' - System has hit bad_page.
298 * 'U' - Userspace-defined naughtiness.
299 * 'D' - Kernel has oopsed before
300 * 'A' - ACPI table overridden.
301 * 'W' - Taint on warning.
302 * 'C' - modules from drivers/staging are loaded.
303 * 'I' - Working around severe firmware bug.
304 * 'O' - Out-of-tree module has been loaded.
305 * 'E' - Unsigned module has been loaded.
306 * 'L' - A soft lockup has previously occurred.
307 * 'K' - Kernel has been live patched.
308 *
309 * The string is overwritten by the next call to print_tainted().
310 */
311const char *print_tainted(void)
312{
313 static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
314
315 if (tainted_mask) {
316 char *s;
317 int i;
318
319 s = buf + sprintf(buf, "Tainted: ");
320 for (i = 0; i < ARRAY_SIZE(tnts); i++) {
321 const struct tnt *t = &tnts[i];
322 *s++ = test_bit(t->bit, &tainted_mask) ?
323 t->true : t->false;
324 }
325 *s = 0;
326 } else
327 snprintf(buf, sizeof(buf), "Not tainted");
328
329 return buf;
330}
331
332int test_taint(unsigned flag)
333{
334 return test_bit(flag, &tainted_mask);
335}
336EXPORT_SYMBOL(test_taint);
337
338unsigned long get_taint(void)
339{
340 return tainted_mask;
341}
342
343/**
344 * add_taint: add a taint flag if not already set.
345 * @flag: one of the TAINT_* constants.
346 * @lockdep_ok: whether lock debugging is still OK.
347 *
348 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
349 * some notewortht-but-not-corrupting cases, it can be set to true.
350 */
351void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
352{
353 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
354 pr_warn("Disabling lock debugging due to kernel taint\n");
355
356 set_bit(flag, &tainted_mask);
357}
358EXPORT_SYMBOL(add_taint);
359
360static void spin_msec(int msecs)
361{
362 int i;
363
364 for (i = 0; i < msecs; i++) {
365 touch_nmi_watchdog();
366 mdelay(1);
367 }
368}
369
370/*
371 * It just happens that oops_enter() and oops_exit() are identically
372 * implemented...
373 */
374static void do_oops_enter_exit(void)
375{
376 unsigned long flags;
377 static int spin_counter;
378
379 if (!pause_on_oops)
380 return;
381
382 spin_lock_irqsave(&pause_on_oops_lock, flags);
383 if (pause_on_oops_flag == 0) {
384 /* This CPU may now print the oops message */
385 pause_on_oops_flag = 1;
386 } else {
387 /* We need to stall this CPU */
388 if (!spin_counter) {
389 /* This CPU gets to do the counting */
390 spin_counter = pause_on_oops;
391 do {
392 spin_unlock(&pause_on_oops_lock);
393 spin_msec(MSEC_PER_SEC);
394 spin_lock(&pause_on_oops_lock);
395 } while (--spin_counter);
396 pause_on_oops_flag = 0;
397 } else {
398 /* This CPU waits for a different one */
399 while (spin_counter) {
400 spin_unlock(&pause_on_oops_lock);
401 spin_msec(1);
402 spin_lock(&pause_on_oops_lock);
403 }
404 }
405 }
406 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
407}
408
409/*
410 * Return true if the calling CPU is allowed to print oops-related info.
411 * This is a bit racy..
412 */
413int oops_may_print(void)
414{
415 return pause_on_oops_flag == 0;
416}
417
418/*
419 * Called when the architecture enters its oops handler, before it prints
420 * anything. If this is the first CPU to oops, and it's oopsing the first
421 * time then let it proceed.
422 *
423 * This is all enabled by the pause_on_oops kernel boot option. We do all
424 * this to ensure that oopses don't scroll off the screen. It has the
425 * side-effect of preventing later-oopsing CPUs from mucking up the display,
426 * too.
427 *
428 * It turns out that the CPU which is allowed to print ends up pausing for
429 * the right duration, whereas all the other CPUs pause for twice as long:
430 * once in oops_enter(), once in oops_exit().
431 */
432void oops_enter(void)
433{
434 tracing_off();
435 /* can't trust the integrity of the kernel anymore: */
436 debug_locks_off();
437 do_oops_enter_exit();
438}
439
440/*
441 * 64-bit random ID for oopses:
442 */
443static u64 oops_id;
444
445static int init_oops_id(void)
446{
447 if (!oops_id)
448 get_random_bytes(&oops_id, sizeof(oops_id));
449 else
450 oops_id++;
451
452 return 0;
453}
454late_initcall(init_oops_id);
455
456void print_oops_end_marker(void)
457{
458 init_oops_id();
459 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
460}
461
462/*
463 * Called when the architecture exits its oops handler, after printing
464 * everything.
465 */
466void oops_exit(void)
467{
468 do_oops_enter_exit();
469 print_oops_end_marker();
470 kmsg_dump(KMSG_DUMP_OOPS);
471}
472
473struct warn_args {
474 const char *fmt;
475 va_list args;
476};
477
478void __warn(const char *file, int line, void *caller, unsigned taint,
479 struct pt_regs *regs, struct warn_args *args)
480{
481 disable_trace_on_warning();
482
483 pr_warn("------------[ cut here ]------------\n");
484
485 if (file)
486 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
487 raw_smp_processor_id(), current->pid, file, line,
488 caller);
489 else
490 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
491 raw_smp_processor_id(), current->pid, caller);
492
493 if (args)
494 vprintk(args->fmt, args->args);
495
496 if (panic_on_warn) {
497 /*
498 * This thread may hit another WARN() in the panic path.
499 * Resetting this prevents additional WARN() from panicking the
500 * system on this thread. Other threads are blocked by the
501 * panic_mutex in panic().
502 */
503 panic_on_warn = 0;
504 panic("panic_on_warn set ...\n");
505 }
506
507 print_modules();
508
509 if (regs)
510 show_regs(regs);
511 else
512 dump_stack();
513
514 print_oops_end_marker();
515
516 /* Just a warning, don't kill lockdep. */
517 add_taint(taint, LOCKDEP_STILL_OK);
518}
519
520#ifdef WANT_WARN_ON_SLOWPATH
521void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
522{
523 struct warn_args args;
524
525 args.fmt = fmt;
526 va_start(args.args, fmt);
527 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
528 &args);
529 va_end(args.args);
530}
531EXPORT_SYMBOL(warn_slowpath_fmt);
532
533void warn_slowpath_fmt_taint(const char *file, int line,
534 unsigned taint, const char *fmt, ...)
535{
536 struct warn_args args;
537
538 args.fmt = fmt;
539 va_start(args.args, fmt);
540 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
541 va_end(args.args);
542}
543EXPORT_SYMBOL(warn_slowpath_fmt_taint);
544
545void warn_slowpath_null(const char *file, int line)
546{
547 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
548}
549EXPORT_SYMBOL(warn_slowpath_null);
550#endif
551
552#ifdef CONFIG_CC_STACKPROTECTOR
553
554/*
555 * Called when gcc's -fstack-protector feature is used, and
556 * gcc detects corruption of the on-stack canary value
557 */
558__visible void __stack_chk_fail(void)
559{
560 panic("stack-protector: Kernel stack is corrupted in: %p\n",
561 __builtin_return_address(0));
562}
563EXPORT_SYMBOL(__stack_chk_fail);
564
565#endif
566
567core_param(panic, panic_timeout, int, 0644);
568core_param(pause_on_oops, pause_on_oops, int, 0644);
569core_param(panic_on_warn, panic_on_warn, int, 0644);
570
571static int __init setup_crash_kexec_post_notifiers(char *s)
572{
573 crash_kexec_post_notifiers = true;
574 return 0;
575}
576early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
577
578static int __init oops_setup(char *s)
579{
580 if (!s)
581 return -EINVAL;
582 if (!strcmp(s, "panic"))
583 panic_on_oops = 1;
584 return 0;
585}
586early_param("oops", oops_setup);