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 <linux/sysfs.h>
37 #include <trace/events/error_report.h>
38 #include <asm/sections.h>
40 #define PANIC_TIMER_STEP 100
41 #define PANIC_BLINK_SPD 18
45 * Should we dump all CPUs backtraces in an oops event?
46 * Defaults to 0, can be changed via sysctl.
48 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
50 #define sysctl_oops_all_cpu_backtrace 0
51 #endif /* CONFIG_SMP */
53 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
54 static unsigned long tainted_mask =
55 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
56 static int pause_on_oops;
57 static int pause_on_oops_flag;
58 static DEFINE_SPINLOCK(pause_on_oops_lock);
59 bool crash_kexec_post_notifiers;
60 int panic_on_warn __read_mostly;
61 unsigned long panic_on_taint;
62 bool panic_on_taint_nousertaint = false;
63 static unsigned int warn_limit __read_mostly;
65 int panic_timeout = CONFIG_PANIC_TIMEOUT;
66 EXPORT_SYMBOL_GPL(panic_timeout);
68 #define PANIC_PRINT_TASK_INFO 0x00000001
69 #define PANIC_PRINT_MEM_INFO 0x00000002
70 #define PANIC_PRINT_TIMER_INFO 0x00000004
71 #define PANIC_PRINT_LOCK_INFO 0x00000008
72 #define PANIC_PRINT_FTRACE_INFO 0x00000010
73 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
74 #define PANIC_PRINT_ALL_CPU_BT 0x00000040
75 unsigned long panic_print;
77 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
79 EXPORT_SYMBOL(panic_notifier_list);
82 static struct ctl_table kern_panic_table[] = {
85 .procname = "oops_all_cpu_backtrace",
86 .data = &sysctl_oops_all_cpu_backtrace,
87 .maxlen = sizeof(int),
89 .proc_handler = proc_dointvec_minmax,
90 .extra1 = SYSCTL_ZERO,
95 .procname = "warn_limit",
97 .maxlen = sizeof(warn_limit),
99 .proc_handler = proc_douintvec,
104 static __init int kernel_panic_sysctls_init(void)
106 register_sysctl_init("kernel", kern_panic_table);
109 late_initcall(kernel_panic_sysctls_init);
112 static atomic_t warn_count = ATOMIC_INIT(0);
115 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
118 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
121 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
123 static __init int kernel_panic_sysfs_init(void)
125 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
128 late_initcall(kernel_panic_sysfs_init);
131 static long no_blink(int state)
136 /* Returns how long it waited in ms */
137 long (*panic_blink)(int state);
138 EXPORT_SYMBOL(panic_blink);
141 * Stop ourself in panic -- architecture code may override this
143 void __weak panic_smp_self_stop(void)
150 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
151 * may override this to prepare for crash dumping, e.g. save regs info.
153 void __weak nmi_panic_self_stop(struct pt_regs *regs)
155 panic_smp_self_stop();
159 * Stop other CPUs in panic. Architecture dependent code may override this
160 * with more suitable version. For example, if the architecture supports
161 * crash dump, it should save registers of each stopped CPU and disable
162 * per-CPU features such as virtualization extensions.
164 void __weak crash_smp_send_stop(void)
166 static int cpus_stopped;
169 * This function can be called twice in panic path, but obviously
170 * we execute this only once.
176 * Note smp_send_stop is the usual smp shutdown function, which
177 * unfortunately means it may not be hardened to work in a panic
184 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
187 * A variant of panic() called from NMI context. We return if we've already
188 * panicked on this CPU. If another CPU already panicked, loop in
189 * nmi_panic_self_stop() which can provide architecture dependent code such
190 * as saving register state for crash dump.
192 void nmi_panic(struct pt_regs *regs, const char *msg)
196 cpu = raw_smp_processor_id();
197 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
199 if (old_cpu == PANIC_CPU_INVALID)
201 else if (old_cpu != cpu)
202 nmi_panic_self_stop(regs);
204 EXPORT_SYMBOL(nmi_panic);
206 static void panic_print_sys_info(bool console_flush)
209 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
210 console_flush_on_panic(CONSOLE_REPLAY_ALL);
214 if (panic_print & PANIC_PRINT_ALL_CPU_BT)
215 trigger_all_cpu_backtrace();
217 if (panic_print & PANIC_PRINT_TASK_INFO)
220 if (panic_print & PANIC_PRINT_MEM_INFO)
223 if (panic_print & PANIC_PRINT_TIMER_INFO)
224 sysrq_timer_list_show();
226 if (panic_print & PANIC_PRINT_LOCK_INFO)
227 debug_show_all_locks();
229 if (panic_print & PANIC_PRINT_FTRACE_INFO)
230 ftrace_dump(DUMP_ALL);
233 void check_panic_on_warn(const char *origin)
236 panic("%s: panic_on_warn set ...\n", origin);
238 if (atomic_inc_return(&warn_count) >= READ_ONCE(warn_limit) && warn_limit)
239 panic("%s: system warned too often (kernel.warn_limit is %d)",
244 * panic - halt the system
245 * @fmt: The text string to print
247 * Display a message, then perform cleanups.
249 * This function never returns.
251 void panic(const char *fmt, ...)
253 static char buf[1024];
255 long i, i_next = 0, len;
257 int old_cpu, this_cpu;
258 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
262 * This thread may hit another WARN() in the panic path.
263 * Resetting this prevents additional WARN() from panicking the
264 * system on this thread. Other threads are blocked by the
265 * panic_mutex in panic().
271 * Disable local interrupts. This will prevent panic_smp_self_stop
272 * from deadlocking the first cpu that invokes the panic, since
273 * there is nothing to prevent an interrupt handler (that runs
274 * after setting panic_cpu) from invoking panic() again.
277 preempt_disable_notrace();
280 * It's possible to come here directly from a panic-assertion and
281 * not have preempt disabled. Some functions called from here want
282 * preempt to be disabled. No point enabling it later though...
284 * Only one CPU is allowed to execute the panic code from here. For
285 * multiple parallel invocations of panic, all other CPUs either
286 * stop themself or will wait until they are stopped by the 1st CPU
287 * with smp_send_stop().
289 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
290 * comes here, so go ahead.
291 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
292 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
294 this_cpu = raw_smp_processor_id();
295 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
297 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
298 panic_smp_self_stop();
303 len = vscnprintf(buf, sizeof(buf), fmt, args);
306 if (len && buf[len - 1] == '\n')
309 pr_emerg("Kernel panic - not syncing: %s\n", buf);
310 #ifdef CONFIG_DEBUG_BUGVERBOSE
312 * Avoid nested stack-dumping if a panic occurs during oops processing
314 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
319 * If kgdb is enabled, give it a chance to run before we stop all
320 * the other CPUs or else we won't be able to debug processes left
326 * If we have crashed and we have a crash kernel loaded let it handle
328 * If we want to run this after calling panic_notifiers, pass
329 * the "crash_kexec_post_notifiers" option to the kernel.
331 * Bypass the panic_cpu check and call __crash_kexec directly.
333 if (!_crash_kexec_post_notifiers) {
337 * Note smp_send_stop is the usual smp shutdown function, which
338 * unfortunately means it may not be hardened to work in a
344 * If we want to do crash dump after notifier calls and
345 * kmsg_dump, we will need architecture dependent extra
346 * works in addition to stopping other CPUs.
348 crash_smp_send_stop();
352 * Run any panic handlers, including those that might need to
353 * add information to the kmsg dump output.
355 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
357 panic_print_sys_info(false);
359 kmsg_dump(KMSG_DUMP_PANIC);
362 * If you doubt kdump always works fine in any situation,
363 * "crash_kexec_post_notifiers" offers you a chance to run
364 * panic_notifiers and dumping kmsg before kdump.
365 * Note: since some panic_notifiers can make crashed kernel
366 * more unstable, it can increase risks of the kdump failure too.
368 * Bypass the panic_cpu check and call __crash_kexec directly.
370 if (_crash_kexec_post_notifiers)
376 * We may have ended up stopping the CPU holding the lock (in
377 * smp_send_stop()) while still having some valuable data in the console
378 * buffer. Try to acquire the lock then release it regardless of the
379 * result. The release will also print the buffers out. Locks debug
380 * should be disabled to avoid reporting bad unlock balance when
381 * panic() is not being callled from OOPS.
384 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
386 panic_print_sys_info(true);
389 panic_blink = no_blink;
391 if (panic_timeout > 0) {
393 * Delay timeout seconds before rebooting the machine.
394 * We can't use the "normal" timers since we just panicked.
396 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
398 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
399 touch_nmi_watchdog();
401 i += panic_blink(state ^= 1);
402 i_next = i + 3600 / PANIC_BLINK_SPD;
404 mdelay(PANIC_TIMER_STEP);
407 if (panic_timeout != 0) {
409 * This will not be a clean reboot, with everything
410 * shutting down. But if there is a chance of
411 * rebooting the system it will be rebooted.
413 if (panic_reboot_mode != REBOOT_UNDEFINED)
414 reboot_mode = panic_reboot_mode;
419 extern int stop_a_enabled;
420 /* Make sure the user can actually press Stop-A (L1-A) */
422 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
423 "twice on console to return to the boot prom\n");
426 #if defined(CONFIG_S390)
429 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
431 /* Do not scroll important messages printed above */
434 for (i = 0; ; i += PANIC_TIMER_STEP) {
435 touch_softlockup_watchdog();
437 i += panic_blink(state ^= 1);
438 i_next = i + 3600 / PANIC_BLINK_SPD;
440 mdelay(PANIC_TIMER_STEP);
444 EXPORT_SYMBOL(panic);
447 * TAINT_FORCED_RMMOD could be a per-module flag but the module
448 * is being removed anyway.
450 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
451 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
452 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
453 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
454 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
455 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
456 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
457 [ TAINT_USER ] = { 'U', ' ', false },
458 [ TAINT_DIE ] = { 'D', ' ', false },
459 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
460 [ TAINT_WARN ] = { 'W', ' ', false },
461 [ TAINT_CRAP ] = { 'C', ' ', true },
462 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
463 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
464 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
465 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
466 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
467 [ TAINT_AUX ] = { 'X', ' ', true },
468 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
469 [ TAINT_TEST ] = { 'N', ' ', true },
473 * print_tainted - return a string to represent the kernel taint state.
475 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
477 * The string is overwritten by the next call to print_tainted(),
478 * but is always NULL terminated.
480 const char *print_tainted(void)
482 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
484 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
490 s = buf + sprintf(buf, "Tainted: ");
491 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
492 const struct taint_flag *t = &taint_flags[i];
493 *s++ = test_bit(i, &tainted_mask) ?
494 t->c_true : t->c_false;
498 snprintf(buf, sizeof(buf), "Not tainted");
503 int test_taint(unsigned flag)
505 return test_bit(flag, &tainted_mask);
507 EXPORT_SYMBOL(test_taint);
509 unsigned long get_taint(void)
515 * add_taint: add a taint flag if not already set.
516 * @flag: one of the TAINT_* constants.
517 * @lockdep_ok: whether lock debugging is still OK.
519 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
520 * some notewortht-but-not-corrupting cases, it can be set to true.
522 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
524 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
525 pr_warn("Disabling lock debugging due to kernel taint\n");
527 set_bit(flag, &tainted_mask);
529 if (tainted_mask & panic_on_taint) {
531 panic("panic_on_taint set ...");
534 EXPORT_SYMBOL(add_taint);
536 static void spin_msec(int msecs)
540 for (i = 0; i < msecs; i++) {
541 touch_nmi_watchdog();
547 * It just happens that oops_enter() and oops_exit() are identically
550 static void do_oops_enter_exit(void)
553 static int spin_counter;
558 spin_lock_irqsave(&pause_on_oops_lock, flags);
559 if (pause_on_oops_flag == 0) {
560 /* This CPU may now print the oops message */
561 pause_on_oops_flag = 1;
563 /* We need to stall this CPU */
565 /* This CPU gets to do the counting */
566 spin_counter = pause_on_oops;
568 spin_unlock(&pause_on_oops_lock);
569 spin_msec(MSEC_PER_SEC);
570 spin_lock(&pause_on_oops_lock);
571 } while (--spin_counter);
572 pause_on_oops_flag = 0;
574 /* This CPU waits for a different one */
575 while (spin_counter) {
576 spin_unlock(&pause_on_oops_lock);
578 spin_lock(&pause_on_oops_lock);
582 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
586 * Return true if the calling CPU is allowed to print oops-related info.
587 * This is a bit racy..
589 bool oops_may_print(void)
591 return pause_on_oops_flag == 0;
595 * Called when the architecture enters its oops handler, before it prints
596 * anything. If this is the first CPU to oops, and it's oopsing the first
597 * time then let it proceed.
599 * This is all enabled by the pause_on_oops kernel boot option. We do all
600 * this to ensure that oopses don't scroll off the screen. It has the
601 * side-effect of preventing later-oopsing CPUs from mucking up the display,
604 * It turns out that the CPU which is allowed to print ends up pausing for
605 * the right duration, whereas all the other CPUs pause for twice as long:
606 * once in oops_enter(), once in oops_exit().
608 void oops_enter(void)
611 /* can't trust the integrity of the kernel anymore: */
613 do_oops_enter_exit();
615 if (sysctl_oops_all_cpu_backtrace)
616 trigger_all_cpu_backtrace();
619 static void print_oops_end_marker(void)
621 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
625 * Called when the architecture exits its oops handler, after printing
630 do_oops_enter_exit();
631 print_oops_end_marker();
632 kmsg_dump(KMSG_DUMP_OOPS);
640 void __warn(const char *file, int line, void *caller, unsigned taint,
641 struct pt_regs *regs, struct warn_args *args)
643 disable_trace_on_warning();
646 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
647 raw_smp_processor_id(), current->pid, file, line,
650 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
651 raw_smp_processor_id(), current->pid, caller);
654 vprintk(args->fmt, args->args);
661 check_panic_on_warn("kernel");
666 print_irqtrace_events(current);
668 print_oops_end_marker();
669 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
671 /* Just a warning, don't kill lockdep. */
672 add_taint(taint, LOCKDEP_STILL_OK);
676 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
677 const char *fmt, ...)
679 struct warn_args args;
684 __warn(file, line, __builtin_return_address(0), taint,
690 va_start(args.args, fmt);
691 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
694 EXPORT_SYMBOL(warn_slowpath_fmt);
696 void __warn_printk(const char *fmt, ...)
706 EXPORT_SYMBOL(__warn_printk);
711 /* Support resetting WARN*_ONCE state */
713 static int clear_warn_once_set(void *data, u64 val)
715 generic_bug_clear_once();
716 memset(__start_once, 0, __end_once - __start_once);
720 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
723 static __init int register_warn_debugfs(void)
725 /* Don't care about failure */
726 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
727 &clear_warn_once_fops);
731 device_initcall(register_warn_debugfs);
734 #ifdef CONFIG_STACKPROTECTOR
737 * Called when gcc's -fstack-protector feature is used, and
738 * gcc detects corruption of the on-stack canary value
740 __visible noinstr void __stack_chk_fail(void)
742 instrumentation_begin();
743 panic("stack-protector: Kernel stack is corrupted in: %pB",
744 __builtin_return_address(0));
745 instrumentation_end();
747 EXPORT_SYMBOL(__stack_chk_fail);
751 core_param(panic, panic_timeout, int, 0644);
752 core_param(panic_print, panic_print, ulong, 0644);
753 core_param(pause_on_oops, pause_on_oops, int, 0644);
754 core_param(panic_on_warn, panic_on_warn, int, 0644);
755 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
757 static int __init oops_setup(char *s)
761 if (!strcmp(s, "panic"))
765 early_param("oops", oops_setup);
767 static int __init panic_on_taint_setup(char *s)
774 taint_str = strsep(&s, ",");
775 if (kstrtoul(taint_str, 16, &panic_on_taint))
778 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
779 panic_on_taint &= TAINT_FLAGS_MAX;
784 if (s && !strcmp(s, "nousertaint"))
785 panic_on_taint_nousertaint = true;
787 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
788 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
792 early_param("panic_on_taint", panic_on_taint_setup);