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/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 <linux/sysfs.h>
35 #include <asm/sections.h>
37 #define PANIC_TIMER_STEP 100
38 #define PANIC_BLINK_SPD 18
42 * Should we dump all CPUs backtraces in an oops event?
43 * Defaults to 0, can be changed via sysctl.
45 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
47 #define sysctl_oops_all_cpu_backtrace 0
48 #endif /* CONFIG_SMP */
50 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
51 static unsigned long tainted_mask =
52 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
53 static int pause_on_oops;
54 static int pause_on_oops_flag;
55 static DEFINE_SPINLOCK(pause_on_oops_lock);
56 bool crash_kexec_post_notifiers;
57 int panic_on_warn __read_mostly;
58 unsigned long panic_on_taint;
59 bool panic_on_taint_nousertaint = false;
60 static unsigned int warn_limit __read_mostly;
62 int panic_timeout = CONFIG_PANIC_TIMEOUT;
63 EXPORT_SYMBOL_GPL(panic_timeout);
65 #define PANIC_PRINT_TASK_INFO 0x00000001
66 #define PANIC_PRINT_MEM_INFO 0x00000002
67 #define PANIC_PRINT_TIMER_INFO 0x00000004
68 #define PANIC_PRINT_LOCK_INFO 0x00000008
69 #define PANIC_PRINT_FTRACE_INFO 0x00000010
70 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
71 unsigned long panic_print;
73 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
75 EXPORT_SYMBOL(panic_notifier_list);
78 static struct ctl_table kern_panic_table[] = {
81 .procname = "oops_all_cpu_backtrace",
82 .data = &sysctl_oops_all_cpu_backtrace,
83 .maxlen = sizeof(int),
85 .proc_handler = proc_dointvec_minmax,
86 .extra1 = SYSCTL_ZERO,
91 .procname = "warn_limit",
93 .maxlen = sizeof(warn_limit),
95 .proc_handler = proc_douintvec,
100 static __init int kernel_panic_sysctls_init(void)
102 register_sysctl_init("kernel", kern_panic_table);
105 late_initcall(kernel_panic_sysctls_init);
108 static atomic_t warn_count = ATOMIC_INIT(0);
111 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
114 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
117 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
119 static __init int kernel_panic_sysfs_init(void)
121 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
124 late_initcall(kernel_panic_sysfs_init);
127 static long no_blink(int state)
132 /* Returns how long it waited in ms */
133 long (*panic_blink)(int state);
134 EXPORT_SYMBOL(panic_blink);
137 * Stop ourself in panic -- architecture code may override this
139 void __weak panic_smp_self_stop(void)
146 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
147 * may override this to prepare for crash dumping, e.g. save regs info.
149 void __weak nmi_panic_self_stop(struct pt_regs *regs)
151 panic_smp_self_stop();
155 * Stop other CPUs in panic. Architecture dependent code may override this
156 * with more suitable version. For example, if the architecture supports
157 * crash dump, it should save registers of each stopped CPU and disable
158 * per-CPU features such as virtualization extensions.
160 void __weak crash_smp_send_stop(void)
162 static int cpus_stopped;
165 * This function can be called twice in panic path, but obviously
166 * we execute this only once.
172 * Note smp_send_stop is the usual smp shutdown function, which
173 * unfortunately means it may not be hardened to work in a panic
180 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
183 * A variant of panic() called from NMI context. We return if we've already
184 * panicked on this CPU. If another CPU already panicked, loop in
185 * nmi_panic_self_stop() which can provide architecture dependent code such
186 * as saving register state for crash dump.
188 void nmi_panic(struct pt_regs *regs, const char *msg)
192 cpu = raw_smp_processor_id();
193 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
195 if (old_cpu == PANIC_CPU_INVALID)
197 else if (old_cpu != cpu)
198 nmi_panic_self_stop(regs);
200 EXPORT_SYMBOL(nmi_panic);
202 static void panic_print_sys_info(void)
204 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
205 console_flush_on_panic(CONSOLE_REPLAY_ALL);
207 if (panic_print & PANIC_PRINT_TASK_INFO)
210 if (panic_print & PANIC_PRINT_MEM_INFO)
213 if (panic_print & PANIC_PRINT_TIMER_INFO)
214 sysrq_timer_list_show();
216 if (panic_print & PANIC_PRINT_LOCK_INFO)
217 debug_show_all_locks();
219 if (panic_print & PANIC_PRINT_FTRACE_INFO)
220 ftrace_dump(DUMP_ALL);
223 void check_panic_on_warn(const char *origin)
228 panic("%s: panic_on_warn set ...\n", origin);
230 limit = READ_ONCE(warn_limit);
231 if (atomic_inc_return(&warn_count) >= limit && limit)
232 panic("%s: system warned too often (kernel.warn_limit is %d)",
237 * panic - halt the system
238 * @fmt: The text string to print
240 * Display a message, then perform cleanups.
242 * This function never returns.
244 void panic(const char *fmt, ...)
246 static char buf[1024];
248 long i, i_next = 0, len;
250 int old_cpu, this_cpu;
251 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
255 * This thread may hit another WARN() in the panic path.
256 * Resetting this prevents additional WARN() from panicking the
257 * system on this thread. Other threads are blocked by the
258 * panic_mutex in panic().
264 * Disable local interrupts. This will prevent panic_smp_self_stop
265 * from deadlocking the first cpu that invokes the panic, since
266 * there is nothing to prevent an interrupt handler (that runs
267 * after setting panic_cpu) from invoking panic() again.
270 preempt_disable_notrace();
273 * It's possible to come here directly from a panic-assertion and
274 * not have preempt disabled. Some functions called from here want
275 * preempt to be disabled. No point enabling it later though...
277 * Only one CPU is allowed to execute the panic code from here. For
278 * multiple parallel invocations of panic, all other CPUs either
279 * stop themself or will wait until they are stopped by the 1st CPU
280 * with smp_send_stop().
282 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
283 * comes here, so go ahead.
284 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
285 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
287 this_cpu = raw_smp_processor_id();
288 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
290 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
291 panic_smp_self_stop();
296 len = vscnprintf(buf, sizeof(buf), fmt, args);
299 if (len && buf[len - 1] == '\n')
302 pr_emerg("Kernel panic - not syncing: %s\n", buf);
303 #ifdef CONFIG_DEBUG_BUGVERBOSE
305 * Avoid nested stack-dumping if a panic occurs during oops processing
307 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
312 * If kgdb is enabled, give it a chance to run before we stop all
313 * the other CPUs or else we won't be able to debug processes left
319 * If we have crashed and we have a crash kernel loaded let it handle
321 * If we want to run this after calling panic_notifiers, pass
322 * the "crash_kexec_post_notifiers" option to the kernel.
324 * Bypass the panic_cpu check and call __crash_kexec directly.
326 if (!_crash_kexec_post_notifiers) {
327 printk_safe_flush_on_panic();
331 * Note smp_send_stop is the usual smp shutdown function, which
332 * unfortunately means it may not be hardened to work in a
338 * If we want to do crash dump after notifier calls and
339 * kmsg_dump, we will need architecture dependent extra
340 * works in addition to stopping other CPUs.
342 crash_smp_send_stop();
346 * Run any panic handlers, including those that might need to
347 * add information to the kmsg dump output.
349 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
351 /* Call flush even twice. It tries harder with a single online CPU */
352 printk_safe_flush_on_panic();
353 kmsg_dump(KMSG_DUMP_PANIC);
356 * If you doubt kdump always works fine in any situation,
357 * "crash_kexec_post_notifiers" offers you a chance to run
358 * panic_notifiers and dumping kmsg before kdump.
359 * Note: since some panic_notifiers can make crashed kernel
360 * more unstable, it can increase risks of the kdump failure too.
362 * Bypass the panic_cpu check and call __crash_kexec directly.
364 if (_crash_kexec_post_notifiers)
373 * We may have ended up stopping the CPU holding the lock (in
374 * smp_send_stop()) while still having some valuable data in the console
375 * buffer. Try to acquire the lock then release it regardless of the
376 * result. The release will also print the buffers out. Locks debug
377 * should be disabled to avoid reporting bad unlock balance when
378 * panic() is not being callled from OOPS.
381 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
383 panic_print_sys_info();
386 panic_blink = no_blink;
388 if (panic_timeout > 0) {
390 * Delay timeout seconds before rebooting the machine.
391 * We can't use the "normal" timers since we just panicked.
393 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
395 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
396 touch_nmi_watchdog();
398 i += panic_blink(state ^= 1);
399 i_next = i + 3600 / PANIC_BLINK_SPD;
401 mdelay(PANIC_TIMER_STEP);
404 if (panic_timeout != 0) {
406 * This will not be a clean reboot, with everything
407 * shutting down. But if there is a chance of
408 * rebooting the system it will be rebooted.
410 if (panic_reboot_mode != REBOOT_UNDEFINED)
411 reboot_mode = panic_reboot_mode;
416 extern int stop_a_enabled;
417 /* Make sure the user can actually press Stop-A (L1-A) */
419 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
420 "twice on console to return to the boot prom\n");
423 #if defined(CONFIG_S390)
426 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
428 /* Do not scroll important messages printed above */
432 * The final messages may not have been printed if in a context that
433 * defers printing (such as NMI) and irq_work is not available.
434 * Explicitly flush the kernel log buffer one last time.
436 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
439 for (i = 0; ; i += PANIC_TIMER_STEP) {
440 touch_softlockup_watchdog();
442 i += panic_blink(state ^= 1);
443 i_next = i + 3600 / PANIC_BLINK_SPD;
445 mdelay(PANIC_TIMER_STEP);
449 EXPORT_SYMBOL(panic);
452 * TAINT_FORCED_RMMOD could be a per-module flag but the module
453 * is being removed anyway.
455 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
456 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
457 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
458 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
459 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
460 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
461 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
462 [ TAINT_USER ] = { 'U', ' ', false },
463 [ TAINT_DIE ] = { 'D', ' ', false },
464 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
465 [ TAINT_WARN ] = { 'W', ' ', false },
466 [ TAINT_CRAP ] = { 'C', ' ', true },
467 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
468 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
469 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
470 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
471 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
472 [ TAINT_AUX ] = { 'X', ' ', true },
473 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
477 * print_tainted - return a string to represent the kernel taint state.
479 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
481 * The string is overwritten by the next call to print_tainted(),
482 * but is always NULL terminated.
484 const char *print_tainted(void)
486 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
488 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
494 s = buf + sprintf(buf, "Tainted: ");
495 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
496 const struct taint_flag *t = &taint_flags[i];
497 *s++ = test_bit(i, &tainted_mask) ?
498 t->c_true : t->c_false;
502 snprintf(buf, sizeof(buf), "Not tainted");
507 int test_taint(unsigned flag)
509 return test_bit(flag, &tainted_mask);
511 EXPORT_SYMBOL(test_taint);
513 unsigned long get_taint(void)
519 * add_taint: add a taint flag if not already set.
520 * @flag: one of the TAINT_* constants.
521 * @lockdep_ok: whether lock debugging is still OK.
523 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
524 * some notewortht-but-not-corrupting cases, it can be set to true.
526 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
528 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
529 pr_warn("Disabling lock debugging due to kernel taint\n");
531 set_bit(flag, &tainted_mask);
533 if (tainted_mask & panic_on_taint) {
535 panic("panic_on_taint set ...");
538 EXPORT_SYMBOL(add_taint);
540 static void spin_msec(int msecs)
544 for (i = 0; i < msecs; i++) {
545 touch_nmi_watchdog();
551 * It just happens that oops_enter() and oops_exit() are identically
554 static void do_oops_enter_exit(void)
557 static int spin_counter;
562 spin_lock_irqsave(&pause_on_oops_lock, flags);
563 if (pause_on_oops_flag == 0) {
564 /* This CPU may now print the oops message */
565 pause_on_oops_flag = 1;
567 /* We need to stall this CPU */
569 /* This CPU gets to do the counting */
570 spin_counter = pause_on_oops;
572 spin_unlock(&pause_on_oops_lock);
573 spin_msec(MSEC_PER_SEC);
574 spin_lock(&pause_on_oops_lock);
575 } while (--spin_counter);
576 pause_on_oops_flag = 0;
578 /* This CPU waits for a different one */
579 while (spin_counter) {
580 spin_unlock(&pause_on_oops_lock);
582 spin_lock(&pause_on_oops_lock);
586 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
590 * Return true if the calling CPU is allowed to print oops-related info.
591 * This is a bit racy..
593 bool oops_may_print(void)
595 return pause_on_oops_flag == 0;
599 * Called when the architecture enters its oops handler, before it prints
600 * anything. If this is the first CPU to oops, and it's oopsing the first
601 * time then let it proceed.
603 * This is all enabled by the pause_on_oops kernel boot option. We do all
604 * this to ensure that oopses don't scroll off the screen. It has the
605 * side-effect of preventing later-oopsing CPUs from mucking up the display,
608 * It turns out that the CPU which is allowed to print ends up pausing for
609 * the right duration, whereas all the other CPUs pause for twice as long:
610 * once in oops_enter(), once in oops_exit().
612 void oops_enter(void)
615 /* can't trust the integrity of the kernel anymore: */
617 do_oops_enter_exit();
619 if (sysctl_oops_all_cpu_backtrace)
620 trigger_all_cpu_backtrace();
624 * 64-bit random ID for oopses:
628 static int init_oops_id(void)
631 get_random_bytes(&oops_id, sizeof(oops_id));
637 late_initcall(init_oops_id);
639 static void print_oops_end_marker(void)
642 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
646 * Called when the architecture exits its oops handler, after printing
651 do_oops_enter_exit();
652 print_oops_end_marker();
653 kmsg_dump(KMSG_DUMP_OOPS);
661 void __warn(const char *file, int line, void *caller, unsigned taint,
662 struct pt_regs *regs, struct warn_args *args)
664 disable_trace_on_warning();
667 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
668 raw_smp_processor_id(), current->pid, file, line,
671 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
672 raw_smp_processor_id(), current->pid, caller);
675 vprintk(args->fmt, args->args);
682 check_panic_on_warn("kernel");
687 print_irqtrace_events(current);
689 print_oops_end_marker();
691 /* Just a warning, don't kill lockdep. */
692 add_taint(taint, LOCKDEP_STILL_OK);
696 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
697 const char *fmt, ...)
699 struct warn_args args;
704 __warn(file, line, __builtin_return_address(0), taint,
710 va_start(args.args, fmt);
711 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
714 EXPORT_SYMBOL(warn_slowpath_fmt);
716 void __warn_printk(const char *fmt, ...)
726 EXPORT_SYMBOL(__warn_printk);
731 /* Support resetting WARN*_ONCE state */
733 static int clear_warn_once_set(void *data, u64 val)
735 generic_bug_clear_once();
736 memset(__start_once, 0, __end_once - __start_once);
740 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
743 static __init int register_warn_debugfs(void)
745 /* Don't care about failure */
746 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
747 &clear_warn_once_fops);
751 device_initcall(register_warn_debugfs);
754 #ifdef CONFIG_STACKPROTECTOR
757 * Called when gcc's -fstack-protector feature is used, and
758 * gcc detects corruption of the on-stack canary value
760 __visible noinstr void __stack_chk_fail(void)
762 instrumentation_begin();
763 panic("stack-protector: Kernel stack is corrupted in: %pB",
764 __builtin_return_address(0));
765 instrumentation_end();
767 EXPORT_SYMBOL(__stack_chk_fail);
771 core_param(panic, panic_timeout, int, 0644);
772 core_param(panic_print, panic_print, ulong, 0644);
773 core_param(pause_on_oops, pause_on_oops, int, 0644);
774 core_param(panic_on_warn, panic_on_warn, int, 0644);
775 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
777 static int __init oops_setup(char *s)
781 if (!strcmp(s, "panic"))
785 early_param("oops", oops_setup);
787 static int __init panic_on_taint_setup(char *s)
794 taint_str = strsep(&s, ",");
795 if (kstrtoul(taint_str, 16, &panic_on_taint))
798 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
799 panic_on_taint &= TAINT_FLAGS_MAX;
804 if (s && !strcmp(s, "nousertaint"))
805 panic_on_taint_nousertaint = true;
807 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
808 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
812 early_param("panic_on_taint", panic_on_taint_setup);