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
40 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
41 static unsigned long tainted_mask =
42 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
43 static int pause_on_oops;
44 static int pause_on_oops_flag;
45 static DEFINE_SPINLOCK(pause_on_oops_lock);
46 bool crash_kexec_post_notifiers;
47 int panic_on_warn __read_mostly;
48 static unsigned int warn_limit __read_mostly;
50 int panic_timeout = CONFIG_PANIC_TIMEOUT;
51 EXPORT_SYMBOL_GPL(panic_timeout);
53 #define PANIC_PRINT_TASK_INFO 0x00000001
54 #define PANIC_PRINT_MEM_INFO 0x00000002
55 #define PANIC_PRINT_TIMER_INFO 0x00000004
56 #define PANIC_PRINT_LOCK_INFO 0x00000008
57 #define PANIC_PRINT_FTRACE_INFO 0x00000010
58 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
59 unsigned long panic_print;
61 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
63 EXPORT_SYMBOL(panic_notifier_list);
66 static struct ctl_table kern_panic_table[] = {
68 .procname = "warn_limit",
70 .maxlen = sizeof(warn_limit),
72 .proc_handler = proc_douintvec,
77 static __init int kernel_panic_sysctls_init(void)
79 register_sysctl_init("kernel", kern_panic_table);
82 late_initcall(kernel_panic_sysctls_init);
85 static atomic_t warn_count = ATOMIC_INIT(0);
88 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
91 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
94 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
96 static __init int kernel_panic_sysfs_init(void)
98 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
101 late_initcall(kernel_panic_sysfs_init);
104 static long no_blink(int state)
109 /* Returns how long it waited in ms */
110 long (*panic_blink)(int state);
111 EXPORT_SYMBOL(panic_blink);
114 * Stop ourself in panic -- architecture code may override this
116 void __weak panic_smp_self_stop(void)
123 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
124 * may override this to prepare for crash dumping, e.g. save regs info.
126 void __weak nmi_panic_self_stop(struct pt_regs *regs)
128 panic_smp_self_stop();
132 * Stop other CPUs in panic. Architecture dependent code may override this
133 * with more suitable version. For example, if the architecture supports
134 * crash dump, it should save registers of each stopped CPU and disable
135 * per-CPU features such as virtualization extensions.
137 void __weak crash_smp_send_stop(void)
139 static int cpus_stopped;
142 * This function can be called twice in panic path, but obviously
143 * we execute this only once.
149 * Note smp_send_stop is the usual smp shutdown function, which
150 * unfortunately means it may not be hardened to work in a panic
157 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
160 * A variant of panic() called from NMI context. We return if we've already
161 * panicked on this CPU. If another CPU already panicked, loop in
162 * nmi_panic_self_stop() which can provide architecture dependent code such
163 * as saving register state for crash dump.
165 void nmi_panic(struct pt_regs *regs, const char *msg)
169 cpu = raw_smp_processor_id();
170 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
172 if (old_cpu == PANIC_CPU_INVALID)
174 else if (old_cpu != cpu)
175 nmi_panic_self_stop(regs);
177 EXPORT_SYMBOL(nmi_panic);
179 static void panic_print_sys_info(void)
181 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
182 console_flush_on_panic(CONSOLE_REPLAY_ALL);
184 if (panic_print & PANIC_PRINT_TASK_INFO)
187 if (panic_print & PANIC_PRINT_MEM_INFO)
190 if (panic_print & PANIC_PRINT_TIMER_INFO)
191 sysrq_timer_list_show();
193 if (panic_print & PANIC_PRINT_LOCK_INFO)
194 debug_show_all_locks();
196 if (panic_print & PANIC_PRINT_FTRACE_INFO)
197 ftrace_dump(DUMP_ALL);
200 void check_panic_on_warn(const char *origin)
205 panic("%s: panic_on_warn set ...\n", origin);
207 limit = READ_ONCE(warn_limit);
208 if (atomic_inc_return(&warn_count) >= limit && limit)
209 panic("%s: system warned too often (kernel.warn_limit is %d)",
214 * panic - halt the system
215 * @fmt: The text string to print
217 * Display a message, then perform cleanups.
219 * This function never returns.
221 void panic(const char *fmt, ...)
223 static char buf[1024];
225 long i, i_next = 0, len;
227 int old_cpu, this_cpu;
228 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
232 * This thread may hit another WARN() in the panic path.
233 * Resetting this prevents additional WARN() from panicking the
234 * system on this thread. Other threads are blocked by the
235 * panic_mutex in panic().
241 * Disable local interrupts. This will prevent panic_smp_self_stop
242 * from deadlocking the first cpu that invokes the panic, since
243 * there is nothing to prevent an interrupt handler (that runs
244 * after setting panic_cpu) from invoking panic() again.
247 preempt_disable_notrace();
250 * It's possible to come here directly from a panic-assertion and
251 * not have preempt disabled. Some functions called from here want
252 * preempt to be disabled. No point enabling it later though...
254 * Only one CPU is allowed to execute the panic code from here. For
255 * multiple parallel invocations of panic, all other CPUs either
256 * stop themself or will wait until they are stopped by the 1st CPU
257 * with smp_send_stop().
259 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
260 * comes here, so go ahead.
261 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
262 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
264 this_cpu = raw_smp_processor_id();
265 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
267 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
268 panic_smp_self_stop();
273 len = vscnprintf(buf, sizeof(buf), fmt, args);
276 if (len && buf[len - 1] == '\n')
279 pr_emerg("Kernel panic - not syncing: %s\n", buf);
280 #ifdef CONFIG_DEBUG_BUGVERBOSE
282 * Avoid nested stack-dumping if a panic occurs during oops processing
284 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
289 * If kgdb is enabled, give it a chance to run before we stop all
290 * the other CPUs or else we won't be able to debug processes left
296 * If we have crashed and we have a crash kernel loaded let it handle
298 * If we want to run this after calling panic_notifiers, pass
299 * the "crash_kexec_post_notifiers" option to the kernel.
301 * Bypass the panic_cpu check and call __crash_kexec directly.
303 if (!_crash_kexec_post_notifiers) {
304 printk_safe_flush_on_panic();
308 * Note smp_send_stop is the usual smp shutdown function, which
309 * unfortunately means it may not be hardened to work in a
315 * If we want to do crash dump after notifier calls and
316 * kmsg_dump, we will need architecture dependent extra
317 * works in addition to stopping other CPUs.
319 crash_smp_send_stop();
323 * Run any panic handlers, including those that might need to
324 * add information to the kmsg dump output.
326 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
328 /* Call flush even twice. It tries harder with a single online CPU */
329 printk_safe_flush_on_panic();
330 kmsg_dump(KMSG_DUMP_PANIC);
333 * If you doubt kdump always works fine in any situation,
334 * "crash_kexec_post_notifiers" offers you a chance to run
335 * panic_notifiers and dumping kmsg before kdump.
336 * Note: since some panic_notifiers can make crashed kernel
337 * more unstable, it can increase risks of the kdump failure too.
339 * Bypass the panic_cpu check and call __crash_kexec directly.
341 if (_crash_kexec_post_notifiers)
350 * We may have ended up stopping the CPU holding the lock (in
351 * smp_send_stop()) while still having some valuable data in the console
352 * buffer. Try to acquire the lock then release it regardless of the
353 * result. The release will also print the buffers out. Locks debug
354 * should be disabled to avoid reporting bad unlock balance when
355 * panic() is not being callled from OOPS.
358 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
360 panic_print_sys_info();
363 panic_blink = no_blink;
365 if (panic_timeout > 0) {
367 * Delay timeout seconds before rebooting the machine.
368 * We can't use the "normal" timers since we just panicked.
370 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
372 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
373 touch_nmi_watchdog();
375 i += panic_blink(state ^= 1);
376 i_next = i + 3600 / PANIC_BLINK_SPD;
378 mdelay(PANIC_TIMER_STEP);
381 if (panic_timeout != 0) {
383 * This will not be a clean reboot, with everything
384 * shutting down. But if there is a chance of
385 * rebooting the system it will be rebooted.
387 if (panic_reboot_mode != REBOOT_UNDEFINED)
388 reboot_mode = panic_reboot_mode;
393 extern int stop_a_enabled;
394 /* Make sure the user can actually press Stop-A (L1-A) */
396 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
397 "twice on console to return to the boot prom\n");
400 #if defined(CONFIG_S390)
403 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
405 /* Do not scroll important messages printed above */
408 for (i = 0; ; i += PANIC_TIMER_STEP) {
409 touch_softlockup_watchdog();
411 i += panic_blink(state ^= 1);
412 i_next = i + 3600 / PANIC_BLINK_SPD;
414 mdelay(PANIC_TIMER_STEP);
418 EXPORT_SYMBOL(panic);
421 * TAINT_FORCED_RMMOD could be a per-module flag but the module
422 * is being removed anyway.
424 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
425 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
426 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
427 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
428 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
429 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
430 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
431 [ TAINT_USER ] = { 'U', ' ', false },
432 [ TAINT_DIE ] = { 'D', ' ', false },
433 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
434 [ TAINT_WARN ] = { 'W', ' ', false },
435 [ TAINT_CRAP ] = { 'C', ' ', true },
436 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
437 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
438 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
439 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
440 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
441 [ TAINT_AUX ] = { 'X', ' ', true },
442 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
446 * print_tainted - return a string to represent the kernel taint state.
448 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
450 * The string is overwritten by the next call to print_tainted(),
451 * but is always NULL terminated.
453 const char *print_tainted(void)
455 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
457 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
463 s = buf + sprintf(buf, "Tainted: ");
464 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
465 const struct taint_flag *t = &taint_flags[i];
466 *s++ = test_bit(i, &tainted_mask) ?
467 t->c_true : t->c_false;
471 snprintf(buf, sizeof(buf), "Not tainted");
476 int test_taint(unsigned flag)
478 return test_bit(flag, &tainted_mask);
480 EXPORT_SYMBOL(test_taint);
482 unsigned long get_taint(void)
488 * add_taint: add a taint flag if not already set.
489 * @flag: one of the TAINT_* constants.
490 * @lockdep_ok: whether lock debugging is still OK.
492 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
493 * some notewortht-but-not-corrupting cases, it can be set to true.
495 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
497 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
498 pr_warn("Disabling lock debugging due to kernel taint\n");
500 set_bit(flag, &tainted_mask);
502 EXPORT_SYMBOL(add_taint);
504 static void spin_msec(int msecs)
508 for (i = 0; i < msecs; i++) {
509 touch_nmi_watchdog();
515 * It just happens that oops_enter() and oops_exit() are identically
518 static void do_oops_enter_exit(void)
521 static int spin_counter;
526 spin_lock_irqsave(&pause_on_oops_lock, flags);
527 if (pause_on_oops_flag == 0) {
528 /* This CPU may now print the oops message */
529 pause_on_oops_flag = 1;
531 /* We need to stall this CPU */
533 /* This CPU gets to do the counting */
534 spin_counter = pause_on_oops;
536 spin_unlock(&pause_on_oops_lock);
537 spin_msec(MSEC_PER_SEC);
538 spin_lock(&pause_on_oops_lock);
539 } while (--spin_counter);
540 pause_on_oops_flag = 0;
542 /* This CPU waits for a different one */
543 while (spin_counter) {
544 spin_unlock(&pause_on_oops_lock);
546 spin_lock(&pause_on_oops_lock);
550 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
554 * Return true if the calling CPU is allowed to print oops-related info.
555 * This is a bit racy..
557 int oops_may_print(void)
559 return pause_on_oops_flag == 0;
563 * Called when the architecture enters its oops handler, before it prints
564 * anything. If this is the first CPU to oops, and it's oopsing the first
565 * time then let it proceed.
567 * This is all enabled by the pause_on_oops kernel boot option. We do all
568 * this to ensure that oopses don't scroll off the screen. It has the
569 * side-effect of preventing later-oopsing CPUs from mucking up the display,
572 * It turns out that the CPU which is allowed to print ends up pausing for
573 * the right duration, whereas all the other CPUs pause for twice as long:
574 * once in oops_enter(), once in oops_exit().
576 void oops_enter(void)
579 /* can't trust the integrity of the kernel anymore: */
581 do_oops_enter_exit();
585 * 64-bit random ID for oopses:
589 static int init_oops_id(void)
592 get_random_bytes(&oops_id, sizeof(oops_id));
598 late_initcall(init_oops_id);
600 void print_oops_end_marker(void)
603 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
607 * Called when the architecture exits its oops handler, after printing
612 do_oops_enter_exit();
613 print_oops_end_marker();
614 kmsg_dump(KMSG_DUMP_OOPS);
622 void __warn(const char *file, int line, void *caller, unsigned taint,
623 struct pt_regs *regs, struct warn_args *args)
625 disable_trace_on_warning();
628 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
629 raw_smp_processor_id(), current->pid, file, line,
632 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
633 raw_smp_processor_id(), current->pid, caller);
636 vprintk(args->fmt, args->args);
638 check_panic_on_warn("kernel");
647 print_irqtrace_events(current);
649 print_oops_end_marker();
651 /* Just a warning, don't kill lockdep. */
652 add_taint(taint, LOCKDEP_STILL_OK);
656 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
657 const char *fmt, ...)
659 struct warn_args args;
664 __warn(file, line, __builtin_return_address(0), taint,
670 va_start(args.args, fmt);
671 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
674 EXPORT_SYMBOL(warn_slowpath_fmt);
676 void __warn_printk(const char *fmt, ...)
686 EXPORT_SYMBOL(__warn_printk);
691 /* Support resetting WARN*_ONCE state */
693 static int clear_warn_once_set(void *data, u64 val)
695 generic_bug_clear_once();
696 memset(__start_once, 0, __end_once - __start_once);
700 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
703 static __init int register_warn_debugfs(void)
705 /* Don't care about failure */
706 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
707 &clear_warn_once_fops);
711 device_initcall(register_warn_debugfs);
714 #ifdef CONFIG_STACKPROTECTOR
717 * Called when gcc's -fstack-protector feature is used, and
718 * gcc detects corruption of the on-stack canary value
720 __visible void __stack_chk_fail(void)
722 panic("stack-protector: Kernel stack is corrupted in: %pB",
723 __builtin_return_address(0));
725 EXPORT_SYMBOL(__stack_chk_fail);
729 #ifdef CONFIG_ARCH_HAS_REFCOUNT
730 void refcount_error_report(struct pt_regs *regs, const char *err)
732 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
733 err, (void *)instruction_pointer(regs),
734 current->comm, task_pid_nr(current),
735 from_kuid_munged(&init_user_ns, current_uid()),
736 from_kuid_munged(&init_user_ns, current_euid()));
740 core_param(panic, panic_timeout, int, 0644);
741 core_param(panic_print, panic_print, ulong, 0644);
742 core_param(pause_on_oops, pause_on_oops, int, 0644);
743 core_param(panic_on_warn, panic_on_warn, int, 0644);
744 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
746 static int __init oops_setup(char *s)
750 if (!strcmp(s, "panic"))
754 early_param("oops", oops_setup);