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/sysrq.h>
29#include <linux/init.h>
30#include <linux/nmi.h>
31#include <linux/console.h>
32#include <linux/bug.h>
33#include <linux/ratelimit.h>
34#include <linux/debugfs.h>
35#include <asm/sections.h>
36
37#define PANIC_TIMER_STEP 100
38#define PANIC_BLINK_SPD 18
39
40#ifdef CONFIG_SMP
41/*
42 * Should we dump all CPUs backtraces in an oops event?
43 * Defaults to 0, can be changed via sysctl.
44 */
45unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
46#endif /* CONFIG_SMP */
47
48int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
49static unsigned long tainted_mask =
50 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
51static int pause_on_oops;
52static int pause_on_oops_flag;
53static DEFINE_SPINLOCK(pause_on_oops_lock);
54bool crash_kexec_post_notifiers;
55int panic_on_warn __read_mostly;
56unsigned long panic_on_taint;
57bool panic_on_taint_nousertaint = false;
58
59int panic_timeout = CONFIG_PANIC_TIMEOUT;
60EXPORT_SYMBOL_GPL(panic_timeout);
61
62#define PANIC_PRINT_TASK_INFO 0x00000001
63#define PANIC_PRINT_MEM_INFO 0x00000002
64#define PANIC_PRINT_TIMER_INFO 0x00000004
65#define PANIC_PRINT_LOCK_INFO 0x00000008
66#define PANIC_PRINT_FTRACE_INFO 0x00000010
67#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
68unsigned long panic_print;
69
70ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
71
72EXPORT_SYMBOL(panic_notifier_list);
73
74static long no_blink(int state)
75{
76 return 0;
77}
78
79/* Returns how long it waited in ms */
80long (*panic_blink)(int state);
81EXPORT_SYMBOL(panic_blink);
82
83/*
84 * Stop ourself in panic -- architecture code may override this
85 */
86void __weak panic_smp_self_stop(void)
87{
88 while (1)
89 cpu_relax();
90}
91
92/*
93 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
94 * may override this to prepare for crash dumping, e.g. save regs info.
95 */
96void __weak nmi_panic_self_stop(struct pt_regs *regs)
97{
98 panic_smp_self_stop();
99}
100
101/*
102 * Stop other CPUs in panic. Architecture dependent code may override this
103 * with more suitable version. For example, if the architecture supports
104 * crash dump, it should save registers of each stopped CPU and disable
105 * per-CPU features such as virtualization extensions.
106 */
107void __weak crash_smp_send_stop(void)
108{
109 static int cpus_stopped;
110
111 /*
112 * This function can be called twice in panic path, but obviously
113 * we execute this only once.
114 */
115 if (cpus_stopped)
116 return;
117
118 /*
119 * Note smp_send_stop is the usual smp shutdown function, which
120 * unfortunately means it may not be hardened to work in a panic
121 * situation.
122 */
123 smp_send_stop();
124 cpus_stopped = 1;
125}
126
127atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
128
129/*
130 * A variant of panic() called from NMI context. We return if we've already
131 * panicked on this CPU. If another CPU already panicked, loop in
132 * nmi_panic_self_stop() which can provide architecture dependent code such
133 * as saving register state for crash dump.
134 */
135void nmi_panic(struct pt_regs *regs, const char *msg)
136{
137 int old_cpu, cpu;
138
139 cpu = raw_smp_processor_id();
140 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
141
142 if (old_cpu == PANIC_CPU_INVALID)
143 panic("%s", msg);
144 else if (old_cpu != cpu)
145 nmi_panic_self_stop(regs);
146}
147EXPORT_SYMBOL(nmi_panic);
148
149static void panic_print_sys_info(void)
150{
151 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
152 console_flush_on_panic(CONSOLE_REPLAY_ALL);
153
154 if (panic_print & PANIC_PRINT_TASK_INFO)
155 show_state();
156
157 if (panic_print & PANIC_PRINT_MEM_INFO)
158 show_mem(0, NULL);
159
160 if (panic_print & PANIC_PRINT_TIMER_INFO)
161 sysrq_timer_list_show();
162
163 if (panic_print & PANIC_PRINT_LOCK_INFO)
164 debug_show_all_locks();
165
166 if (panic_print & PANIC_PRINT_FTRACE_INFO)
167 ftrace_dump(DUMP_ALL);
168}
169
170/**
171 * panic - halt the system
172 * @fmt: The text string to print
173 *
174 * Display a message, then perform cleanups.
175 *
176 * This function never returns.
177 */
178void panic(const char *fmt, ...)
179{
180 static char buf[1024];
181 va_list args;
182 long i, i_next = 0, len;
183 int state = 0;
184 int old_cpu, this_cpu;
185 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
186
187 /*
188 * Disable local interrupts. This will prevent panic_smp_self_stop
189 * from deadlocking the first cpu that invokes the panic, since
190 * there is nothing to prevent an interrupt handler (that runs
191 * after setting panic_cpu) from invoking panic() again.
192 */
193 local_irq_disable();
194 preempt_disable_notrace();
195
196 /*
197 * It's possible to come here directly from a panic-assertion and
198 * not have preempt disabled. Some functions called from here want
199 * preempt to be disabled. No point enabling it later though...
200 *
201 * Only one CPU is allowed to execute the panic code from here. For
202 * multiple parallel invocations of panic, all other CPUs either
203 * stop themself or will wait until they are stopped by the 1st CPU
204 * with smp_send_stop().
205 *
206 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
207 * comes here, so go ahead.
208 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
209 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
210 */
211 this_cpu = raw_smp_processor_id();
212 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
213
214 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
215 panic_smp_self_stop();
216
217 console_verbose();
218 bust_spinlocks(1);
219 va_start(args, fmt);
220 len = vscnprintf(buf, sizeof(buf), fmt, args);
221 va_end(args);
222
223 if (len && buf[len - 1] == '\n')
224 buf[len - 1] = '\0';
225
226 pr_emerg("Kernel panic - not syncing: %s\n", buf);
227#ifdef CONFIG_DEBUG_BUGVERBOSE
228 /*
229 * Avoid nested stack-dumping if a panic occurs during oops processing
230 */
231 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
232 dump_stack();
233#endif
234
235 /*
236 * If kgdb is enabled, give it a chance to run before we stop all
237 * the other CPUs or else we won't be able to debug processes left
238 * running on them.
239 */
240 kgdb_panic(buf);
241
242 /*
243 * If we have crashed and we have a crash kernel loaded let it handle
244 * everything else.
245 * If we want to run this after calling panic_notifiers, pass
246 * the "crash_kexec_post_notifiers" option to the kernel.
247 *
248 * Bypass the panic_cpu check and call __crash_kexec directly.
249 */
250 if (!_crash_kexec_post_notifiers) {
251 printk_safe_flush_on_panic();
252 __crash_kexec(NULL);
253
254 /*
255 * Note smp_send_stop is the usual smp shutdown function, which
256 * unfortunately means it may not be hardened to work in a
257 * panic situation.
258 */
259 smp_send_stop();
260 } else {
261 /*
262 * If we want to do crash dump after notifier calls and
263 * kmsg_dump, we will need architecture dependent extra
264 * works in addition to stopping other CPUs.
265 */
266 crash_smp_send_stop();
267 }
268
269 /*
270 * Run any panic handlers, including those that might need to
271 * add information to the kmsg dump output.
272 */
273 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
274
275 /* Call flush even twice. It tries harder with a single online CPU */
276 printk_safe_flush_on_panic();
277 kmsg_dump(KMSG_DUMP_PANIC);
278
279 /*
280 * If you doubt kdump always works fine in any situation,
281 * "crash_kexec_post_notifiers" offers you a chance to run
282 * panic_notifiers and dumping kmsg before kdump.
283 * Note: since some panic_notifiers can make crashed kernel
284 * more unstable, it can increase risks of the kdump failure too.
285 *
286 * Bypass the panic_cpu check and call __crash_kexec directly.
287 */
288 if (_crash_kexec_post_notifiers)
289 __crash_kexec(NULL);
290
291#ifdef CONFIG_VT
292 unblank_screen();
293#endif
294 console_unblank();
295
296 /*
297 * We may have ended up stopping the CPU holding the lock (in
298 * smp_send_stop()) while still having some valuable data in the console
299 * buffer. Try to acquire the lock then release it regardless of the
300 * result. The release will also print the buffers out. Locks debug
301 * should be disabled to avoid reporting bad unlock balance when
302 * panic() is not being callled from OOPS.
303 */
304 debug_locks_off();
305 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
306
307 panic_print_sys_info();
308
309 if (!panic_blink)
310 panic_blink = no_blink;
311
312 if (panic_timeout > 0) {
313 /*
314 * Delay timeout seconds before rebooting the machine.
315 * We can't use the "normal" timers since we just panicked.
316 */
317 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
318
319 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
320 touch_nmi_watchdog();
321 if (i >= i_next) {
322 i += panic_blink(state ^= 1);
323 i_next = i + 3600 / PANIC_BLINK_SPD;
324 }
325 mdelay(PANIC_TIMER_STEP);
326 }
327 }
328 if (panic_timeout != 0) {
329 /*
330 * This will not be a clean reboot, with everything
331 * shutting down. But if there is a chance of
332 * rebooting the system it will be rebooted.
333 */
334 if (panic_reboot_mode != REBOOT_UNDEFINED)
335 reboot_mode = panic_reboot_mode;
336 emergency_restart();
337 }
338#ifdef __sparc__
339 {
340 extern int stop_a_enabled;
341 /* Make sure the user can actually press Stop-A (L1-A) */
342 stop_a_enabled = 1;
343 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
344 "twice on console to return to the boot prom\n");
345 }
346#endif
347#if defined(CONFIG_S390)
348 disabled_wait();
349#endif
350 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
351
352 /* Do not scroll important messages printed above */
353 suppress_printk = 1;
354 local_irq_enable();
355 for (i = 0; ; i += PANIC_TIMER_STEP) {
356 touch_softlockup_watchdog();
357 if (i >= i_next) {
358 i += panic_blink(state ^= 1);
359 i_next = i + 3600 / PANIC_BLINK_SPD;
360 }
361 mdelay(PANIC_TIMER_STEP);
362 }
363}
364
365EXPORT_SYMBOL(panic);
366
367/*
368 * TAINT_FORCED_RMMOD could be a per-module flag but the module
369 * is being removed anyway.
370 */
371const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
372 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
373 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
374 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
375 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
376 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
377 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
378 [ TAINT_USER ] = { 'U', ' ', false },
379 [ TAINT_DIE ] = { 'D', ' ', false },
380 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
381 [ TAINT_WARN ] = { 'W', ' ', false },
382 [ TAINT_CRAP ] = { 'C', ' ', true },
383 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
384 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
385 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
386 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
387 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
388 [ TAINT_AUX ] = { 'X', ' ', true },
389 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
390};
391
392/**
393 * print_tainted - return a string to represent the kernel taint state.
394 *
395 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
396 *
397 * The string is overwritten by the next call to print_tainted(),
398 * but is always NULL terminated.
399 */
400const char *print_tainted(void)
401{
402 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
403
404 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
405
406 if (tainted_mask) {
407 char *s;
408 int i;
409
410 s = buf + sprintf(buf, "Tainted: ");
411 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
412 const struct taint_flag *t = &taint_flags[i];
413 *s++ = test_bit(i, &tainted_mask) ?
414 t->c_true : t->c_false;
415 }
416 *s = 0;
417 } else
418 snprintf(buf, sizeof(buf), "Not tainted");
419
420 return buf;
421}
422
423int test_taint(unsigned flag)
424{
425 return test_bit(flag, &tainted_mask);
426}
427EXPORT_SYMBOL(test_taint);
428
429unsigned long get_taint(void)
430{
431 return tainted_mask;
432}
433
434/**
435 * add_taint: add a taint flag if not already set.
436 * @flag: one of the TAINT_* constants.
437 * @lockdep_ok: whether lock debugging is still OK.
438 *
439 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
440 * some notewortht-but-not-corrupting cases, it can be set to true.
441 */
442void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
443{
444 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
445 pr_warn("Disabling lock debugging due to kernel taint\n");
446
447 set_bit(flag, &tainted_mask);
448
449 if (tainted_mask & panic_on_taint) {
450 panic_on_taint = 0;
451 panic("panic_on_taint set ...");
452 }
453}
454EXPORT_SYMBOL(add_taint);
455
456static void spin_msec(int msecs)
457{
458 int i;
459
460 for (i = 0; i < msecs; i++) {
461 touch_nmi_watchdog();
462 mdelay(1);
463 }
464}
465
466/*
467 * It just happens that oops_enter() and oops_exit() are identically
468 * implemented...
469 */
470static void do_oops_enter_exit(void)
471{
472 unsigned long flags;
473 static int spin_counter;
474
475 if (!pause_on_oops)
476 return;
477
478 spin_lock_irqsave(&pause_on_oops_lock, flags);
479 if (pause_on_oops_flag == 0) {
480 /* This CPU may now print the oops message */
481 pause_on_oops_flag = 1;
482 } else {
483 /* We need to stall this CPU */
484 if (!spin_counter) {
485 /* This CPU gets to do the counting */
486 spin_counter = pause_on_oops;
487 do {
488 spin_unlock(&pause_on_oops_lock);
489 spin_msec(MSEC_PER_SEC);
490 spin_lock(&pause_on_oops_lock);
491 } while (--spin_counter);
492 pause_on_oops_flag = 0;
493 } else {
494 /* This CPU waits for a different one */
495 while (spin_counter) {
496 spin_unlock(&pause_on_oops_lock);
497 spin_msec(1);
498 spin_lock(&pause_on_oops_lock);
499 }
500 }
501 }
502 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
503}
504
505/*
506 * Return true if the calling CPU is allowed to print oops-related info.
507 * This is a bit racy..
508 */
509bool oops_may_print(void)
510{
511 return pause_on_oops_flag == 0;
512}
513
514/*
515 * Called when the architecture enters its oops handler, before it prints
516 * anything. If this is the first CPU to oops, and it's oopsing the first
517 * time then let it proceed.
518 *
519 * This is all enabled by the pause_on_oops kernel boot option. We do all
520 * this to ensure that oopses don't scroll off the screen. It has the
521 * side-effect of preventing later-oopsing CPUs from mucking up the display,
522 * too.
523 *
524 * It turns out that the CPU which is allowed to print ends up pausing for
525 * the right duration, whereas all the other CPUs pause for twice as long:
526 * once in oops_enter(), once in oops_exit().
527 */
528void oops_enter(void)
529{
530 tracing_off();
531 /* can't trust the integrity of the kernel anymore: */
532 debug_locks_off();
533 do_oops_enter_exit();
534
535 if (sysctl_oops_all_cpu_backtrace)
536 trigger_all_cpu_backtrace();
537}
538
539/*
540 * 64-bit random ID for oopses:
541 */
542static u64 oops_id;
543
544static int init_oops_id(void)
545{
546 if (!oops_id)
547 get_random_bytes(&oops_id, sizeof(oops_id));
548 else
549 oops_id++;
550
551 return 0;
552}
553late_initcall(init_oops_id);
554
555static void print_oops_end_marker(void)
556{
557 init_oops_id();
558 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
559}
560
561/*
562 * Called when the architecture exits its oops handler, after printing
563 * everything.
564 */
565void oops_exit(void)
566{
567 do_oops_enter_exit();
568 print_oops_end_marker();
569 kmsg_dump(KMSG_DUMP_OOPS);
570}
571
572struct warn_args {
573 const char *fmt;
574 va_list args;
575};
576
577void __warn(const char *file, int line, void *caller, unsigned taint,
578 struct pt_regs *regs, struct warn_args *args)
579{
580 disable_trace_on_warning();
581
582 if (file)
583 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
584 raw_smp_processor_id(), current->pid, file, line,
585 caller);
586 else
587 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
588 raw_smp_processor_id(), current->pid, caller);
589
590 if (args)
591 vprintk(args->fmt, args->args);
592
593 print_modules();
594
595 if (regs)
596 show_regs(regs);
597
598 if (panic_on_warn) {
599 /*
600 * This thread may hit another WARN() in the panic path.
601 * Resetting this prevents additional WARN() from panicking the
602 * system on this thread. Other threads are blocked by the
603 * panic_mutex in panic().
604 */
605 panic_on_warn = 0;
606 panic("panic_on_warn set ...\n");
607 }
608
609 if (!regs)
610 dump_stack();
611
612 print_irqtrace_events(current);
613
614 print_oops_end_marker();
615
616 /* Just a warning, don't kill lockdep. */
617 add_taint(taint, LOCKDEP_STILL_OK);
618}
619
620#ifndef __WARN_FLAGS
621void warn_slowpath_fmt(const char *file, int line, unsigned taint,
622 const char *fmt, ...)
623{
624 struct warn_args args;
625
626 pr_warn(CUT_HERE);
627
628 if (!fmt) {
629 __warn(file, line, __builtin_return_address(0), taint,
630 NULL, NULL);
631 return;
632 }
633
634 args.fmt = fmt;
635 va_start(args.args, fmt);
636 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
637 va_end(args.args);
638}
639EXPORT_SYMBOL(warn_slowpath_fmt);
640#else
641void __warn_printk(const char *fmt, ...)
642{
643 va_list args;
644
645 pr_warn(CUT_HERE);
646
647 va_start(args, fmt);
648 vprintk(fmt, args);
649 va_end(args);
650}
651EXPORT_SYMBOL(__warn_printk);
652#endif
653
654#ifdef CONFIG_BUG
655
656/* Support resetting WARN*_ONCE state */
657
658static int clear_warn_once_set(void *data, u64 val)
659{
660 generic_bug_clear_once();
661 memset(__start_once, 0, __end_once - __start_once);
662 return 0;
663}
664
665DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
666 "%lld\n");
667
668static __init int register_warn_debugfs(void)
669{
670 /* Don't care about failure */
671 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
672 &clear_warn_once_fops);
673 return 0;
674}
675
676device_initcall(register_warn_debugfs);
677#endif
678
679#ifdef CONFIG_STACKPROTECTOR
680
681/*
682 * Called when gcc's -fstack-protector feature is used, and
683 * gcc detects corruption of the on-stack canary value
684 */
685__visible noinstr void __stack_chk_fail(void)
686{
687 instrumentation_begin();
688 panic("stack-protector: Kernel stack is corrupted in: %pB",
689 __builtin_return_address(0));
690 instrumentation_end();
691}
692EXPORT_SYMBOL(__stack_chk_fail);
693
694#endif
695
696core_param(panic, panic_timeout, int, 0644);
697core_param(panic_print, panic_print, ulong, 0644);
698core_param(pause_on_oops, pause_on_oops, int, 0644);
699core_param(panic_on_warn, panic_on_warn, int, 0644);
700core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
701
702static int __init oops_setup(char *s)
703{
704 if (!s)
705 return -EINVAL;
706 if (!strcmp(s, "panic"))
707 panic_on_oops = 1;
708 return 0;
709}
710early_param("oops", oops_setup);
711
712static int __init panic_on_taint_setup(char *s)
713{
714 char *taint_str;
715
716 if (!s)
717 return -EINVAL;
718
719 taint_str = strsep(&s, ",");
720 if (kstrtoul(taint_str, 16, &panic_on_taint))
721 return -EINVAL;
722
723 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
724 panic_on_taint &= TAINT_FLAGS_MAX;
725
726 if (!panic_on_taint)
727 return -EINVAL;
728
729 if (s && !strcmp(s, "nousertaint"))
730 panic_on_taint_nousertaint = true;
731
732 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
733 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
734
735 return 0;
736}
737early_param("panic_on_taint", panic_on_taint_setup);
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/sched.h>
27#include <linux/sysrq.h>
28#include <linux/init.h>
29#include <linux/nmi.h>
30#include <linux/console.h>
31#include <linux/bug.h>
32#include <linux/ratelimit.h>
33#include <linux/debugfs.h>
34#include <asm/sections.h>
35
36#define PANIC_TIMER_STEP 100
37#define PANIC_BLINK_SPD 18
38
39#ifdef CONFIG_SMP
40/*
41 * Should we dump all CPUs backtraces in an oops event?
42 * Defaults to 0, can be changed via sysctl.
43 */
44unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
45#endif /* CONFIG_SMP */
46
47int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
48static unsigned long tainted_mask =
49 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
50static int pause_on_oops;
51static int pause_on_oops_flag;
52static DEFINE_SPINLOCK(pause_on_oops_lock);
53bool crash_kexec_post_notifiers;
54int panic_on_warn __read_mostly;
55unsigned long panic_on_taint;
56bool panic_on_taint_nousertaint = false;
57
58int panic_timeout = CONFIG_PANIC_TIMEOUT;
59EXPORT_SYMBOL_GPL(panic_timeout);
60
61#define PANIC_PRINT_TASK_INFO 0x00000001
62#define PANIC_PRINT_MEM_INFO 0x00000002
63#define PANIC_PRINT_TIMER_INFO 0x00000004
64#define PANIC_PRINT_LOCK_INFO 0x00000008
65#define PANIC_PRINT_FTRACE_INFO 0x00000010
66#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
67unsigned long panic_print;
68
69ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
70
71EXPORT_SYMBOL(panic_notifier_list);
72
73static long no_blink(int state)
74{
75 return 0;
76}
77
78/* Returns how long it waited in ms */
79long (*panic_blink)(int state);
80EXPORT_SYMBOL(panic_blink);
81
82/*
83 * Stop ourself in panic -- architecture code may override this
84 */
85void __weak panic_smp_self_stop(void)
86{
87 while (1)
88 cpu_relax();
89}
90
91/*
92 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
93 * may override this to prepare for crash dumping, e.g. save regs info.
94 */
95void __weak nmi_panic_self_stop(struct pt_regs *regs)
96{
97 panic_smp_self_stop();
98}
99
100/*
101 * Stop other CPUs in panic. Architecture dependent code may override this
102 * with more suitable version. For example, if the architecture supports
103 * crash dump, it should save registers of each stopped CPU and disable
104 * per-CPU features such as virtualization extensions.
105 */
106void __weak crash_smp_send_stop(void)
107{
108 static int cpus_stopped;
109
110 /*
111 * This function can be called twice in panic path, but obviously
112 * we execute this only once.
113 */
114 if (cpus_stopped)
115 return;
116
117 /*
118 * Note smp_send_stop is the usual smp shutdown function, which
119 * unfortunately means it may not be hardened to work in a panic
120 * situation.
121 */
122 smp_send_stop();
123 cpus_stopped = 1;
124}
125
126atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
127
128/*
129 * A variant of panic() called from NMI context. We return if we've already
130 * panicked on this CPU. If another CPU already panicked, loop in
131 * nmi_panic_self_stop() which can provide architecture dependent code such
132 * as saving register state for crash dump.
133 */
134void nmi_panic(struct pt_regs *regs, const char *msg)
135{
136 int old_cpu, cpu;
137
138 cpu = raw_smp_processor_id();
139 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
140
141 if (old_cpu == PANIC_CPU_INVALID)
142 panic("%s", msg);
143 else if (old_cpu != cpu)
144 nmi_panic_self_stop(regs);
145}
146EXPORT_SYMBOL(nmi_panic);
147
148static void panic_print_sys_info(void)
149{
150 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
151 console_flush_on_panic(CONSOLE_REPLAY_ALL);
152
153 if (panic_print & PANIC_PRINT_TASK_INFO)
154 show_state();
155
156 if (panic_print & PANIC_PRINT_MEM_INFO)
157 show_mem(0, NULL);
158
159 if (panic_print & PANIC_PRINT_TIMER_INFO)
160 sysrq_timer_list_show();
161
162 if (panic_print & PANIC_PRINT_LOCK_INFO)
163 debug_show_all_locks();
164
165 if (panic_print & PANIC_PRINT_FTRACE_INFO)
166 ftrace_dump(DUMP_ALL);
167}
168
169/**
170 * panic - halt the system
171 * @fmt: The text string to print
172 *
173 * Display a message, then perform cleanups.
174 *
175 * This function never returns.
176 */
177void panic(const char *fmt, ...)
178{
179 static char buf[1024];
180 va_list args;
181 long i, i_next = 0, len;
182 int state = 0;
183 int old_cpu, this_cpu;
184 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
185
186 /*
187 * Disable local interrupts. This will prevent panic_smp_self_stop
188 * from deadlocking the first cpu that invokes the panic, since
189 * there is nothing to prevent an interrupt handler (that runs
190 * after setting panic_cpu) from invoking panic() again.
191 */
192 local_irq_disable();
193 preempt_disable_notrace();
194
195 /*
196 * It's possible to come here directly from a panic-assertion and
197 * not have preempt disabled. Some functions called from here want
198 * preempt to be disabled. No point enabling it later though...
199 *
200 * Only one CPU is allowed to execute the panic code from here. For
201 * multiple parallel invocations of panic, all other CPUs either
202 * stop themself or will wait until they are stopped by the 1st CPU
203 * with smp_send_stop().
204 *
205 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
206 * comes here, so go ahead.
207 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
208 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
209 */
210 this_cpu = raw_smp_processor_id();
211 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
212
213 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
214 panic_smp_self_stop();
215
216 console_verbose();
217 bust_spinlocks(1);
218 va_start(args, fmt);
219 len = vscnprintf(buf, sizeof(buf), fmt, args);
220 va_end(args);
221
222 if (len && buf[len - 1] == '\n')
223 buf[len - 1] = '\0';
224
225 pr_emerg("Kernel panic - not syncing: %s\n", buf);
226#ifdef CONFIG_DEBUG_BUGVERBOSE
227 /*
228 * Avoid nested stack-dumping if a panic occurs during oops processing
229 */
230 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
231 dump_stack();
232#endif
233
234 /*
235 * If kgdb is enabled, give it a chance to run before we stop all
236 * the other CPUs or else we won't be able to debug processes left
237 * running on them.
238 */
239 kgdb_panic(buf);
240
241 /*
242 * If we have crashed and we have a crash kernel loaded let it handle
243 * everything else.
244 * If we want to run this after calling panic_notifiers, pass
245 * the "crash_kexec_post_notifiers" option to the kernel.
246 *
247 * Bypass the panic_cpu check and call __crash_kexec directly.
248 */
249 if (!_crash_kexec_post_notifiers) {
250 printk_safe_flush_on_panic();
251 __crash_kexec(NULL);
252
253 /*
254 * Note smp_send_stop is the usual smp shutdown function, which
255 * unfortunately means it may not be hardened to work in a
256 * panic situation.
257 */
258 smp_send_stop();
259 } else {
260 /*
261 * If we want to do crash dump after notifier calls and
262 * kmsg_dump, we will need architecture dependent extra
263 * works in addition to stopping other CPUs.
264 */
265 crash_smp_send_stop();
266 }
267
268 /*
269 * Run any panic handlers, including those that might need to
270 * add information to the kmsg dump output.
271 */
272 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
273
274 /* Call flush even twice. It tries harder with a single online CPU */
275 printk_safe_flush_on_panic();
276 kmsg_dump(KMSG_DUMP_PANIC);
277
278 /*
279 * If you doubt kdump always works fine in any situation,
280 * "crash_kexec_post_notifiers" offers you a chance to run
281 * panic_notifiers and dumping kmsg before kdump.
282 * Note: since some panic_notifiers can make crashed kernel
283 * more unstable, it can increase risks of the kdump failure too.
284 *
285 * Bypass the panic_cpu check and call __crash_kexec directly.
286 */
287 if (_crash_kexec_post_notifiers)
288 __crash_kexec(NULL);
289
290#ifdef CONFIG_VT
291 unblank_screen();
292#endif
293 console_unblank();
294
295 /*
296 * We may have ended up stopping the CPU holding the lock (in
297 * smp_send_stop()) while still having some valuable data in the console
298 * buffer. Try to acquire the lock then release it regardless of the
299 * result. The release will also print the buffers out. Locks debug
300 * should be disabled to avoid reporting bad unlock balance when
301 * panic() is not being callled from OOPS.
302 */
303 debug_locks_off();
304 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
305
306 panic_print_sys_info();
307
308 if (!panic_blink)
309 panic_blink = no_blink;
310
311 if (panic_timeout > 0) {
312 /*
313 * Delay timeout seconds before rebooting the machine.
314 * We can't use the "normal" timers since we just panicked.
315 */
316 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
317
318 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
319 touch_nmi_watchdog();
320 if (i >= i_next) {
321 i += panic_blink(state ^= 1);
322 i_next = i + 3600 / PANIC_BLINK_SPD;
323 }
324 mdelay(PANIC_TIMER_STEP);
325 }
326 }
327 if (panic_timeout != 0) {
328 /*
329 * This will not be a clean reboot, with everything
330 * shutting down. But if there is a chance of
331 * rebooting the system it will be rebooted.
332 */
333 if (panic_reboot_mode != REBOOT_UNDEFINED)
334 reboot_mode = panic_reboot_mode;
335 emergency_restart();
336 }
337#ifdef __sparc__
338 {
339 extern int stop_a_enabled;
340 /* Make sure the user can actually press Stop-A (L1-A) */
341 stop_a_enabled = 1;
342 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
343 "twice on console to return to the boot prom\n");
344 }
345#endif
346#if defined(CONFIG_S390)
347 disabled_wait();
348#endif
349 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
350
351 /* Do not scroll important messages printed above */
352 suppress_printk = 1;
353 local_irq_enable();
354 for (i = 0; ; i += PANIC_TIMER_STEP) {
355 touch_softlockup_watchdog();
356 if (i >= i_next) {
357 i += panic_blink(state ^= 1);
358 i_next = i + 3600 / PANIC_BLINK_SPD;
359 }
360 mdelay(PANIC_TIMER_STEP);
361 }
362}
363
364EXPORT_SYMBOL(panic);
365
366/*
367 * TAINT_FORCED_RMMOD could be a per-module flag but the module
368 * is being removed anyway.
369 */
370const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
371 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
372 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
373 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
374 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
375 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
376 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
377 [ TAINT_USER ] = { 'U', ' ', false },
378 [ TAINT_DIE ] = { 'D', ' ', false },
379 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
380 [ TAINT_WARN ] = { 'W', ' ', false },
381 [ TAINT_CRAP ] = { 'C', ' ', true },
382 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
383 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
384 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
385 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
386 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
387 [ TAINT_AUX ] = { 'X', ' ', true },
388 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
389};
390
391/**
392 * print_tainted - return a string to represent the kernel taint state.
393 *
394 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
395 *
396 * The string is overwritten by the next call to print_tainted(),
397 * but is always NULL terminated.
398 */
399const char *print_tainted(void)
400{
401 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
402
403 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
404
405 if (tainted_mask) {
406 char *s;
407 int i;
408
409 s = buf + sprintf(buf, "Tainted: ");
410 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
411 const struct taint_flag *t = &taint_flags[i];
412 *s++ = test_bit(i, &tainted_mask) ?
413 t->c_true : t->c_false;
414 }
415 *s = 0;
416 } else
417 snprintf(buf, sizeof(buf), "Not tainted");
418
419 return buf;
420}
421
422int test_taint(unsigned flag)
423{
424 return test_bit(flag, &tainted_mask);
425}
426EXPORT_SYMBOL(test_taint);
427
428unsigned long get_taint(void)
429{
430 return tainted_mask;
431}
432
433/**
434 * add_taint: add a taint flag if not already set.
435 * @flag: one of the TAINT_* constants.
436 * @lockdep_ok: whether lock debugging is still OK.
437 *
438 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
439 * some notewortht-but-not-corrupting cases, it can be set to true.
440 */
441void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
442{
443 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
444 pr_warn("Disabling lock debugging due to kernel taint\n");
445
446 set_bit(flag, &tainted_mask);
447
448 if (tainted_mask & panic_on_taint) {
449 panic_on_taint = 0;
450 panic("panic_on_taint set ...");
451 }
452}
453EXPORT_SYMBOL(add_taint);
454
455static void spin_msec(int msecs)
456{
457 int i;
458
459 for (i = 0; i < msecs; i++) {
460 touch_nmi_watchdog();
461 mdelay(1);
462 }
463}
464
465/*
466 * It just happens that oops_enter() and oops_exit() are identically
467 * implemented...
468 */
469static void do_oops_enter_exit(void)
470{
471 unsigned long flags;
472 static int spin_counter;
473
474 if (!pause_on_oops)
475 return;
476
477 spin_lock_irqsave(&pause_on_oops_lock, flags);
478 if (pause_on_oops_flag == 0) {
479 /* This CPU may now print the oops message */
480 pause_on_oops_flag = 1;
481 } else {
482 /* We need to stall this CPU */
483 if (!spin_counter) {
484 /* This CPU gets to do the counting */
485 spin_counter = pause_on_oops;
486 do {
487 spin_unlock(&pause_on_oops_lock);
488 spin_msec(MSEC_PER_SEC);
489 spin_lock(&pause_on_oops_lock);
490 } while (--spin_counter);
491 pause_on_oops_flag = 0;
492 } else {
493 /* This CPU waits for a different one */
494 while (spin_counter) {
495 spin_unlock(&pause_on_oops_lock);
496 spin_msec(1);
497 spin_lock(&pause_on_oops_lock);
498 }
499 }
500 }
501 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
502}
503
504/*
505 * Return true if the calling CPU is allowed to print oops-related info.
506 * This is a bit racy..
507 */
508bool oops_may_print(void)
509{
510 return pause_on_oops_flag == 0;
511}
512
513/*
514 * Called when the architecture enters its oops handler, before it prints
515 * anything. If this is the first CPU to oops, and it's oopsing the first
516 * time then let it proceed.
517 *
518 * This is all enabled by the pause_on_oops kernel boot option. We do all
519 * this to ensure that oopses don't scroll off the screen. It has the
520 * side-effect of preventing later-oopsing CPUs from mucking up the display,
521 * too.
522 *
523 * It turns out that the CPU which is allowed to print ends up pausing for
524 * the right duration, whereas all the other CPUs pause for twice as long:
525 * once in oops_enter(), once in oops_exit().
526 */
527void oops_enter(void)
528{
529 tracing_off();
530 /* can't trust the integrity of the kernel anymore: */
531 debug_locks_off();
532 do_oops_enter_exit();
533
534 if (sysctl_oops_all_cpu_backtrace)
535 trigger_all_cpu_backtrace();
536}
537
538/*
539 * 64-bit random ID for oopses:
540 */
541static u64 oops_id;
542
543static int init_oops_id(void)
544{
545 if (!oops_id)
546 get_random_bytes(&oops_id, sizeof(oops_id));
547 else
548 oops_id++;
549
550 return 0;
551}
552late_initcall(init_oops_id);
553
554static void print_oops_end_marker(void)
555{
556 init_oops_id();
557 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
558}
559
560/*
561 * Called when the architecture exits its oops handler, after printing
562 * everything.
563 */
564void oops_exit(void)
565{
566 do_oops_enter_exit();
567 print_oops_end_marker();
568 kmsg_dump(KMSG_DUMP_OOPS);
569}
570
571struct warn_args {
572 const char *fmt;
573 va_list args;
574};
575
576void __warn(const char *file, int line, void *caller, unsigned taint,
577 struct pt_regs *regs, struct warn_args *args)
578{
579 disable_trace_on_warning();
580
581 if (file)
582 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
583 raw_smp_processor_id(), current->pid, file, line,
584 caller);
585 else
586 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
587 raw_smp_processor_id(), current->pid, caller);
588
589 if (args)
590 vprintk(args->fmt, args->args);
591
592 if (panic_on_warn) {
593 /*
594 * This thread may hit another WARN() in the panic path.
595 * Resetting this prevents additional WARN() from panicking the
596 * system on this thread. Other threads are blocked by the
597 * panic_mutex in panic().
598 */
599 panic_on_warn = 0;
600 panic("panic_on_warn set ...\n");
601 }
602
603 print_modules();
604
605 if (regs)
606 show_regs(regs);
607 else
608 dump_stack();
609
610 print_irqtrace_events(current);
611
612 print_oops_end_marker();
613
614 /* Just a warning, don't kill lockdep. */
615 add_taint(taint, LOCKDEP_STILL_OK);
616}
617
618#ifndef __WARN_FLAGS
619void warn_slowpath_fmt(const char *file, int line, unsigned taint,
620 const char *fmt, ...)
621{
622 struct warn_args args;
623
624 pr_warn(CUT_HERE);
625
626 if (!fmt) {
627 __warn(file, line, __builtin_return_address(0), taint,
628 NULL, NULL);
629 return;
630 }
631
632 args.fmt = fmt;
633 va_start(args.args, fmt);
634 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
635 va_end(args.args);
636}
637EXPORT_SYMBOL(warn_slowpath_fmt);
638#else
639void __warn_printk(const char *fmt, ...)
640{
641 va_list args;
642
643 pr_warn(CUT_HERE);
644
645 va_start(args, fmt);
646 vprintk(fmt, args);
647 va_end(args);
648}
649EXPORT_SYMBOL(__warn_printk);
650#endif
651
652#ifdef CONFIG_BUG
653
654/* Support resetting WARN*_ONCE state */
655
656static int clear_warn_once_set(void *data, u64 val)
657{
658 generic_bug_clear_once();
659 memset(__start_once, 0, __end_once - __start_once);
660 return 0;
661}
662
663DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
664 "%lld\n");
665
666static __init int register_warn_debugfs(void)
667{
668 /* Don't care about failure */
669 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
670 &clear_warn_once_fops);
671 return 0;
672}
673
674device_initcall(register_warn_debugfs);
675#endif
676
677#ifdef CONFIG_STACKPROTECTOR
678
679/*
680 * Called when gcc's -fstack-protector feature is used, and
681 * gcc detects corruption of the on-stack canary value
682 */
683__visible noinstr void __stack_chk_fail(void)
684{
685 instrumentation_begin();
686 panic("stack-protector: Kernel stack is corrupted in: %pB",
687 __builtin_return_address(0));
688 instrumentation_end();
689}
690EXPORT_SYMBOL(__stack_chk_fail);
691
692#endif
693
694core_param(panic, panic_timeout, int, 0644);
695core_param(panic_print, panic_print, ulong, 0644);
696core_param(pause_on_oops, pause_on_oops, int, 0644);
697core_param(panic_on_warn, panic_on_warn, int, 0644);
698core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
699
700static int __init oops_setup(char *s)
701{
702 if (!s)
703 return -EINVAL;
704 if (!strcmp(s, "panic"))
705 panic_on_oops = 1;
706 return 0;
707}
708early_param("oops", oops_setup);
709
710static int __init panic_on_taint_setup(char *s)
711{
712 char *taint_str;
713
714 if (!s)
715 return -EINVAL;
716
717 taint_str = strsep(&s, ",");
718 if (kstrtoul(taint_str, 16, &panic_on_taint))
719 return -EINVAL;
720
721 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
722 panic_on_taint &= TAINT_FLAGS_MAX;
723
724 if (!panic_on_taint)
725 return -EINVAL;
726
727 if (s && !strcmp(s, "nousertaint"))
728 panic_on_taint_nousertaint = true;
729
730 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
731 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
732
733 return 0;
734}
735early_param("panic_on_taint", panic_on_taint_setup);