1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * This function is used through-out the kernel (including mm and fs)
10 * to indicate a major problem.
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 <trace/events/error_report.h>
37 #include <asm/sections.h>
39 #define PANIC_TIMER_STEP 100
40 #define PANIC_BLINK_SPD 18
44 * Should we dump all CPUs backtraces in an oops event?
45 * Defaults to 0, can be changed via sysctl.
47 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
49 #define sysctl_oops_all_cpu_backtrace 0
50 #endif /* CONFIG_SMP */
52 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
53 static unsigned long tainted_mask =
54 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
55 static int pause_on_oops;
56 static int pause_on_oops_flag;
57 static DEFINE_SPINLOCK(pause_on_oops_lock);
58 bool crash_kexec_post_notifiers;
59 int panic_on_warn __read_mostly;
60 unsigned long panic_on_taint;
61 bool panic_on_taint_nousertaint = false;
63 int panic_timeout = CONFIG_PANIC_TIMEOUT;
64 EXPORT_SYMBOL_GPL(panic_timeout);
66 #define PANIC_PRINT_TASK_INFO 0x00000001
67 #define PANIC_PRINT_MEM_INFO 0x00000002
68 #define PANIC_PRINT_TIMER_INFO 0x00000004
69 #define PANIC_PRINT_LOCK_INFO 0x00000008
70 #define PANIC_PRINT_FTRACE_INFO 0x00000010
71 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
72 #define PANIC_PRINT_ALL_CPU_BT 0x00000040
73 unsigned long panic_print;
75 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
77 EXPORT_SYMBOL(panic_notifier_list);
79 #if defined(CONFIG_SMP) && defined(CONFIG_SYSCTL)
80 static struct ctl_table kern_panic_table[] = {
82 .procname = "oops_all_cpu_backtrace",
83 .data = &sysctl_oops_all_cpu_backtrace,
84 .maxlen = sizeof(int),
86 .proc_handler = proc_dointvec_minmax,
87 .extra1 = SYSCTL_ZERO,
93 static __init int kernel_panic_sysctls_init(void)
95 register_sysctl_init("kernel", kern_panic_table);
98 late_initcall(kernel_panic_sysctls_init);
101 static long no_blink(int state)
106 /* Returns how long it waited in ms */
107 long (*panic_blink)(int state);
108 EXPORT_SYMBOL(panic_blink);
111 * Stop ourself in panic -- architecture code may override this
113 void __weak panic_smp_self_stop(void)
120 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
121 * may override this to prepare for crash dumping, e.g. save regs info.
123 void __weak nmi_panic_self_stop(struct pt_regs *regs)
125 panic_smp_self_stop();
129 * Stop other CPUs in panic. Architecture dependent code may override this
130 * with more suitable version. For example, if the architecture supports
131 * crash dump, it should save registers of each stopped CPU and disable
132 * per-CPU features such as virtualization extensions.
134 void __weak crash_smp_send_stop(void)
136 static int cpus_stopped;
139 * This function can be called twice in panic path, but obviously
140 * we execute this only once.
146 * Note smp_send_stop is the usual smp shutdown function, which
147 * unfortunately means it may not be hardened to work in a panic
154 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
157 * A variant of panic() called from NMI context. We return if we've already
158 * panicked on this CPU. If another CPU already panicked, loop in
159 * nmi_panic_self_stop() which can provide architecture dependent code such
160 * as saving register state for crash dump.
162 void nmi_panic(struct pt_regs *regs, const char *msg)
166 cpu = raw_smp_processor_id();
167 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
169 if (old_cpu == PANIC_CPU_INVALID)
171 else if (old_cpu != cpu)
172 nmi_panic_self_stop(regs);
174 EXPORT_SYMBOL(nmi_panic);
176 static void panic_print_sys_info(bool console_flush)
179 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
180 console_flush_on_panic(CONSOLE_REPLAY_ALL);
184 if (panic_print & PANIC_PRINT_ALL_CPU_BT)
185 trigger_all_cpu_backtrace();
187 if (panic_print & PANIC_PRINT_TASK_INFO)
190 if (panic_print & PANIC_PRINT_MEM_INFO)
193 if (panic_print & PANIC_PRINT_TIMER_INFO)
194 sysrq_timer_list_show();
196 if (panic_print & PANIC_PRINT_LOCK_INFO)
197 debug_show_all_locks();
199 if (panic_print & PANIC_PRINT_FTRACE_INFO)
200 ftrace_dump(DUMP_ALL);
204 * panic - halt the system
205 * @fmt: The text string to print
207 * Display a message, then perform cleanups.
209 * This function never returns.
211 void panic(const char *fmt, ...)
213 static char buf[1024];
215 long i, i_next = 0, len;
217 int old_cpu, this_cpu;
218 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
222 * This thread may hit another WARN() in the panic path.
223 * Resetting this prevents additional WARN() from panicking the
224 * system on this thread. Other threads are blocked by the
225 * panic_mutex in panic().
231 * Disable local interrupts. This will prevent panic_smp_self_stop
232 * from deadlocking the first cpu that invokes the panic, since
233 * there is nothing to prevent an interrupt handler (that runs
234 * after setting panic_cpu) from invoking panic() again.
237 preempt_disable_notrace();
240 * It's possible to come here directly from a panic-assertion and
241 * not have preempt disabled. Some functions called from here want
242 * preempt to be disabled. No point enabling it later though...
244 * Only one CPU is allowed to execute the panic code from here. For
245 * multiple parallel invocations of panic, all other CPUs either
246 * stop themself or will wait until they are stopped by the 1st CPU
247 * with smp_send_stop().
249 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
250 * comes here, so go ahead.
251 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
252 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
254 this_cpu = raw_smp_processor_id();
255 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
257 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
258 panic_smp_self_stop();
263 len = vscnprintf(buf, sizeof(buf), fmt, args);
266 if (len && buf[len - 1] == '\n')
269 pr_emerg("Kernel panic - not syncing: %s\n", buf);
270 #ifdef CONFIG_DEBUG_BUGVERBOSE
272 * Avoid nested stack-dumping if a panic occurs during oops processing
274 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
279 * If kgdb is enabled, give it a chance to run before we stop all
280 * the other CPUs or else we won't be able to debug processes left
286 * If we have crashed and we have a crash kernel loaded let it handle
288 * If we want to run this after calling panic_notifiers, pass
289 * the "crash_kexec_post_notifiers" option to the kernel.
291 * Bypass the panic_cpu check and call __crash_kexec directly.
293 if (!_crash_kexec_post_notifiers) {
297 * Note smp_send_stop is the usual smp shutdown function, which
298 * unfortunately means it may not be hardened to work in a
304 * If we want to do crash dump after notifier calls and
305 * kmsg_dump, we will need architecture dependent extra
306 * works in addition to stopping other CPUs.
308 crash_smp_send_stop();
312 * Run any panic handlers, including those that might need to
313 * add information to the kmsg dump output.
315 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
317 panic_print_sys_info(false);
319 kmsg_dump(KMSG_DUMP_PANIC);
322 * If you doubt kdump always works fine in any situation,
323 * "crash_kexec_post_notifiers" offers you a chance to run
324 * panic_notifiers and dumping kmsg before kdump.
325 * Note: since some panic_notifiers can make crashed kernel
326 * more unstable, it can increase risks of the kdump failure too.
328 * Bypass the panic_cpu check and call __crash_kexec directly.
330 if (_crash_kexec_post_notifiers)
336 * We may have ended up stopping the CPU holding the lock (in
337 * smp_send_stop()) while still having some valuable data in the console
338 * buffer. Try to acquire the lock then release it regardless of the
339 * result. The release will also print the buffers out. Locks debug
340 * should be disabled to avoid reporting bad unlock balance when
341 * panic() is not being callled from OOPS.
344 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
346 panic_print_sys_info(true);
349 panic_blink = no_blink;
351 if (panic_timeout > 0) {
353 * Delay timeout seconds before rebooting the machine.
354 * We can't use the "normal" timers since we just panicked.
356 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
358 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
359 touch_nmi_watchdog();
361 i += panic_blink(state ^= 1);
362 i_next = i + 3600 / PANIC_BLINK_SPD;
364 mdelay(PANIC_TIMER_STEP);
367 if (panic_timeout != 0) {
369 * This will not be a clean reboot, with everything
370 * shutting down. But if there is a chance of
371 * rebooting the system it will be rebooted.
373 if (panic_reboot_mode != REBOOT_UNDEFINED)
374 reboot_mode = panic_reboot_mode;
379 extern int stop_a_enabled;
380 /* Make sure the user can actually press Stop-A (L1-A) */
382 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
383 "twice on console to return to the boot prom\n");
386 #if defined(CONFIG_S390)
389 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
391 /* Do not scroll important messages printed above */
394 for (i = 0; ; i += PANIC_TIMER_STEP) {
395 touch_softlockup_watchdog();
397 i += panic_blink(state ^= 1);
398 i_next = i + 3600 / PANIC_BLINK_SPD;
400 mdelay(PANIC_TIMER_STEP);
404 EXPORT_SYMBOL(panic);
407 * TAINT_FORCED_RMMOD could be a per-module flag but the module
408 * is being removed anyway.
410 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
411 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
412 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
413 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
414 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
415 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
416 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
417 [ TAINT_USER ] = { 'U', ' ', false },
418 [ TAINT_DIE ] = { 'D', ' ', false },
419 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
420 [ TAINT_WARN ] = { 'W', ' ', false },
421 [ TAINT_CRAP ] = { 'C', ' ', true },
422 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
423 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
424 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
425 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
426 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
427 [ TAINT_AUX ] = { 'X', ' ', true },
428 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
429 [ TAINT_TEST ] = { 'N', ' ', true },
433 * print_tainted - return a string to represent the kernel taint state.
435 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
437 * The string is overwritten by the next call to print_tainted(),
438 * but is always NULL terminated.
440 const char *print_tainted(void)
442 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
444 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
450 s = buf + sprintf(buf, "Tainted: ");
451 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
452 const struct taint_flag *t = &taint_flags[i];
453 *s++ = test_bit(i, &tainted_mask) ?
454 t->c_true : t->c_false;
458 snprintf(buf, sizeof(buf), "Not tainted");
463 int test_taint(unsigned flag)
465 return test_bit(flag, &tainted_mask);
467 EXPORT_SYMBOL(test_taint);
469 unsigned long get_taint(void)
475 * add_taint: add a taint flag if not already set.
476 * @flag: one of the TAINT_* constants.
477 * @lockdep_ok: whether lock debugging is still OK.
479 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
480 * some notewortht-but-not-corrupting cases, it can be set to true.
482 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
484 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
485 pr_warn("Disabling lock debugging due to kernel taint\n");
487 set_bit(flag, &tainted_mask);
489 if (tainted_mask & panic_on_taint) {
491 panic("panic_on_taint set ...");
494 EXPORT_SYMBOL(add_taint);
496 static void spin_msec(int msecs)
500 for (i = 0; i < msecs; i++) {
501 touch_nmi_watchdog();
507 * It just happens that oops_enter() and oops_exit() are identically
510 static void do_oops_enter_exit(void)
513 static int spin_counter;
518 spin_lock_irqsave(&pause_on_oops_lock, flags);
519 if (pause_on_oops_flag == 0) {
520 /* This CPU may now print the oops message */
521 pause_on_oops_flag = 1;
523 /* We need to stall this CPU */
525 /* This CPU gets to do the counting */
526 spin_counter = pause_on_oops;
528 spin_unlock(&pause_on_oops_lock);
529 spin_msec(MSEC_PER_SEC);
530 spin_lock(&pause_on_oops_lock);
531 } while (--spin_counter);
532 pause_on_oops_flag = 0;
534 /* This CPU waits for a different one */
535 while (spin_counter) {
536 spin_unlock(&pause_on_oops_lock);
538 spin_lock(&pause_on_oops_lock);
542 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
546 * Return true if the calling CPU is allowed to print oops-related info.
547 * This is a bit racy..
549 bool oops_may_print(void)
551 return pause_on_oops_flag == 0;
555 * Called when the architecture enters its oops handler, before it prints
556 * anything. If this is the first CPU to oops, and it's oopsing the first
557 * time then let it proceed.
559 * This is all enabled by the pause_on_oops kernel boot option. We do all
560 * this to ensure that oopses don't scroll off the screen. It has the
561 * side-effect of preventing later-oopsing CPUs from mucking up the display,
564 * It turns out that the CPU which is allowed to print ends up pausing for
565 * the right duration, whereas all the other CPUs pause for twice as long:
566 * once in oops_enter(), once in oops_exit().
568 void oops_enter(void)
571 /* can't trust the integrity of the kernel anymore: */
573 do_oops_enter_exit();
575 if (sysctl_oops_all_cpu_backtrace)
576 trigger_all_cpu_backtrace();
579 static void print_oops_end_marker(void)
581 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
585 * Called when the architecture exits its oops handler, after printing
590 do_oops_enter_exit();
591 print_oops_end_marker();
592 kmsg_dump(KMSG_DUMP_OOPS);
600 void __warn(const char *file, int line, void *caller, unsigned taint,
601 struct pt_regs *regs, struct warn_args *args)
603 disable_trace_on_warning();
606 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
607 raw_smp_processor_id(), current->pid, file, line,
610 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
611 raw_smp_processor_id(), current->pid, caller);
614 vprintk(args->fmt, args->args);
622 panic("panic_on_warn set ...\n");
627 print_irqtrace_events(current);
629 print_oops_end_marker();
630 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
632 /* Just a warning, don't kill lockdep. */
633 add_taint(taint, LOCKDEP_STILL_OK);
637 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
638 const char *fmt, ...)
640 struct warn_args args;
645 __warn(file, line, __builtin_return_address(0), taint,
651 va_start(args.args, fmt);
652 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
655 EXPORT_SYMBOL(warn_slowpath_fmt);
657 void __warn_printk(const char *fmt, ...)
667 EXPORT_SYMBOL(__warn_printk);
672 /* Support resetting WARN*_ONCE state */
674 static int clear_warn_once_set(void *data, u64 val)
676 generic_bug_clear_once();
677 memset(__start_once, 0, __end_once - __start_once);
681 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
684 static __init int register_warn_debugfs(void)
686 /* Don't care about failure */
687 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
688 &clear_warn_once_fops);
692 device_initcall(register_warn_debugfs);
695 #ifdef CONFIG_STACKPROTECTOR
698 * Called when gcc's -fstack-protector feature is used, and
699 * gcc detects corruption of the on-stack canary value
701 __visible noinstr void __stack_chk_fail(void)
703 instrumentation_begin();
704 panic("stack-protector: Kernel stack is corrupted in: %pB",
705 __builtin_return_address(0));
706 instrumentation_end();
708 EXPORT_SYMBOL(__stack_chk_fail);
712 core_param(panic, panic_timeout, int, 0644);
713 core_param(panic_print, panic_print, ulong, 0644);
714 core_param(pause_on_oops, pause_on_oops, int, 0644);
715 core_param(panic_on_warn, panic_on_warn, int, 0644);
716 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
718 static int __init oops_setup(char *s)
722 if (!strcmp(s, "panic"))
726 early_param("oops", oops_setup);
728 static int __init panic_on_taint_setup(char *s)
735 taint_str = strsep(&s, ",");
736 if (kstrtoul(taint_str, 16, &panic_on_taint))
739 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
740 panic_on_taint &= TAINT_FLAGS_MAX;
745 if (s && !strcmp(s, "nousertaint"))
746 panic_on_taint_nousertaint = true;
748 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
749 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
753 early_param("panic_on_taint", panic_on_taint_setup);