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/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 <linux/sysfs.h>
36 #include <asm/sections.h>
38 #define PANIC_TIMER_STEP 100
39 #define PANIC_BLINK_SPD 18
43 * Should we dump all CPUs backtraces in an oops event?
44 * Defaults to 0, can be changed via sysctl.
46 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
48 #define sysctl_oops_all_cpu_backtrace 0
49 #endif /* CONFIG_SMP */
51 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
52 static unsigned long tainted_mask =
53 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
54 static int pause_on_oops;
55 static int pause_on_oops_flag;
56 static DEFINE_SPINLOCK(pause_on_oops_lock);
57 bool crash_kexec_post_notifiers;
58 int panic_on_warn __read_mostly;
59 unsigned long panic_on_taint;
60 bool panic_on_taint_nousertaint = false;
61 static unsigned int warn_limit __read_mostly;
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 unsigned long panic_print;
74 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
76 EXPORT_SYMBOL(panic_notifier_list);
79 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,
92 .procname = "warn_limit",
94 .maxlen = sizeof(warn_limit),
96 .proc_handler = proc_douintvec,
101 static __init int kernel_panic_sysctls_init(void)
103 register_sysctl_init("kernel", kern_panic_table);
106 late_initcall(kernel_panic_sysctls_init);
109 static atomic_t warn_count = ATOMIC_INIT(0);
112 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
115 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
118 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
120 static __init int kernel_panic_sysfs_init(void)
122 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
125 late_initcall(kernel_panic_sysfs_init);
128 static long no_blink(int state)
133 /* Returns how long it waited in ms */
134 long (*panic_blink)(int state);
135 EXPORT_SYMBOL(panic_blink);
138 * Stop ourself in panic -- architecture code may override this
140 void __weak panic_smp_self_stop(void)
147 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
148 * may override this to prepare for crash dumping, e.g. save regs info.
150 void __weak nmi_panic_self_stop(struct pt_regs *regs)
152 panic_smp_self_stop();
156 * Stop other CPUs in panic. Architecture dependent code may override this
157 * with more suitable version. For example, if the architecture supports
158 * crash dump, it should save registers of each stopped CPU and disable
159 * per-CPU features such as virtualization extensions.
161 void __weak crash_smp_send_stop(void)
163 static int cpus_stopped;
166 * This function can be called twice in panic path, but obviously
167 * we execute this only once.
173 * Note smp_send_stop is the usual smp shutdown function, which
174 * unfortunately means it may not be hardened to work in a panic
181 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
184 * A variant of panic() called from NMI context. We return if we've already
185 * panicked on this CPU. If another CPU already panicked, loop in
186 * nmi_panic_self_stop() which can provide architecture dependent code such
187 * as saving register state for crash dump.
189 void nmi_panic(struct pt_regs *regs, const char *msg)
193 cpu = raw_smp_processor_id();
194 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
196 if (old_cpu == PANIC_CPU_INVALID)
198 else if (old_cpu != cpu)
199 nmi_panic_self_stop(regs);
201 EXPORT_SYMBOL(nmi_panic);
203 static void panic_print_sys_info(void)
205 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
206 console_flush_on_panic(CONSOLE_REPLAY_ALL);
208 if (panic_print & PANIC_PRINT_TASK_INFO)
211 if (panic_print & PANIC_PRINT_MEM_INFO)
214 if (panic_print & PANIC_PRINT_TIMER_INFO)
215 sysrq_timer_list_show();
217 if (panic_print & PANIC_PRINT_LOCK_INFO)
218 debug_show_all_locks();
220 if (panic_print & PANIC_PRINT_FTRACE_INFO)
221 ftrace_dump(DUMP_ALL);
224 void check_panic_on_warn(const char *origin)
229 panic("%s: panic_on_warn set ...\n", origin);
231 limit = READ_ONCE(warn_limit);
232 if (atomic_inc_return(&warn_count) >= limit && limit)
233 panic("%s: system warned too often (kernel.warn_limit is %d)",
238 * panic - halt the system
239 * @fmt: The text string to print
241 * Display a message, then perform cleanups.
243 * This function never returns.
245 void panic(const char *fmt, ...)
247 static char buf[1024];
250 long i, i_next = 0, len;
252 int old_cpu, this_cpu;
253 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
256 pr_emerg("Kernel panic - not syncing:\n");
257 va_start(args2, fmt);
258 va_copy(args, args2);
261 #ifdef CONFIG_DEBUG_BUGVERBOSE
263 * Avoid nested stack-dumping if a panic occurs during oops processing
265 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
268 pr_flush(1000, true);
271 * This thread may hit another WARN() in the panic path.
272 * Resetting this prevents additional WARN() from panicking the
273 * system on this thread. Other threads are blocked by the
274 * panic_mutex in panic().
280 * Disable local interrupts. This will prevent panic_smp_self_stop
281 * from deadlocking the first cpu that invokes the panic, since
282 * there is nothing to prevent an interrupt handler (that runs
283 * after setting panic_cpu) from invoking panic() again.
286 preempt_disable_notrace();
289 * It's possible to come here directly from a panic-assertion and
290 * not have preempt disabled. Some functions called from here want
291 * preempt to be disabled. No point enabling it later though...
293 * Only one CPU is allowed to execute the panic code from here. For
294 * multiple parallel invocations of panic, all other CPUs either
295 * stop themself or will wait until they are stopped by the 1st CPU
296 * with smp_send_stop().
298 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
299 * comes here, so go ahead.
300 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
301 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
303 this_cpu = raw_smp_processor_id();
304 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
306 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
307 panic_smp_self_stop();
310 len = vscnprintf(buf, sizeof(buf), fmt, args);
313 if (len && buf[len - 1] == '\n')
317 * If kgdb is enabled, give it a chance to run before we stop all
318 * the other CPUs or else we won't be able to debug processes left
324 * If we have crashed and we have a crash kernel loaded let it handle
326 * If we want to run this after calling panic_notifiers, pass
327 * the "crash_kexec_post_notifiers" option to the kernel.
329 * Bypass the panic_cpu check and call __crash_kexec directly.
331 if (!_crash_kexec_post_notifiers) {
335 * Note smp_send_stop is the usual smp shutdown function, which
336 * unfortunately means it may not be hardened to work in a
342 * If we want to do crash dump after notifier calls and
343 * kmsg_dump, we will need architecture dependent extra
344 * works in addition to stopping other CPUs.
346 crash_smp_send_stop();
350 * Run any panic handlers, including those that might need to
351 * add information to the kmsg dump output.
353 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
355 kmsg_dump(KMSG_DUMP_PANIC);
358 * If you doubt kdump always works fine in any situation,
359 * "crash_kexec_post_notifiers" offers you a chance to run
360 * panic_notifiers and dumping kmsg before kdump.
361 * Note: since some panic_notifiers can make crashed kernel
362 * more unstable, it can increase risks of the kdump failure too.
364 * Bypass the panic_cpu check and call __crash_kexec directly.
366 if (_crash_kexec_post_notifiers)
375 * We may have ended up stopping the CPU holding the lock (in
376 * smp_send_stop()) while still having some valuable data in the console
377 * buffer. Try to acquire the lock then release it regardless of the
378 * result. The release will also print the buffers out. Locks debug
379 * should be disabled to avoid reporting bad unlock balance when
380 * panic() is not being callled from OOPS.
383 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
385 panic_print_sys_info();
388 panic_blink = no_blink;
390 if (panic_timeout > 0) {
392 * Delay timeout seconds before rebooting the machine.
393 * We can't use the "normal" timers since we just panicked.
395 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
397 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
398 touch_nmi_watchdog();
400 i += panic_blink(state ^= 1);
401 i_next = i + 3600 / PANIC_BLINK_SPD;
403 mdelay(PANIC_TIMER_STEP);
406 if (panic_timeout != 0) {
408 * This will not be a clean reboot, with everything
409 * shutting down. But if there is a chance of
410 * rebooting the system it will be rebooted.
412 if (panic_reboot_mode != REBOOT_UNDEFINED)
413 reboot_mode = panic_reboot_mode;
418 extern int stop_a_enabled;
419 /* Make sure the user can actually press Stop-A (L1-A) */
421 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
422 "twice on console to return to the boot prom\n");
425 #if defined(CONFIG_S390)
428 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
430 /* Do not scroll important messages printed above */
433 for (i = 0; ; i += PANIC_TIMER_STEP) {
434 touch_softlockup_watchdog();
436 i += panic_blink(state ^= 1);
437 i_next = i + 3600 / PANIC_BLINK_SPD;
439 mdelay(PANIC_TIMER_STEP);
443 EXPORT_SYMBOL(panic);
446 * TAINT_FORCED_RMMOD could be a per-module flag but the module
447 * is being removed anyway.
449 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
450 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
451 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
452 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
453 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
454 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
455 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
456 [ TAINT_USER ] = { 'U', ' ', false },
457 [ TAINT_DIE ] = { 'D', ' ', false },
458 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
459 [ TAINT_WARN ] = { 'W', ' ', false },
460 [ TAINT_CRAP ] = { 'C', ' ', true },
461 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
462 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
463 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
464 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
465 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
466 [ TAINT_AUX ] = { 'X', ' ', true },
467 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
471 * print_tainted - return a string to represent the kernel taint state.
473 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
475 * The string is overwritten by the next call to print_tainted(),
476 * but is always NULL terminated.
478 const char *print_tainted(void)
480 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
482 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
488 s = buf + sprintf(buf, "Tainted: ");
489 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
490 const struct taint_flag *t = &taint_flags[i];
491 *s++ = test_bit(i, &tainted_mask) ?
492 t->c_true : t->c_false;
496 snprintf(buf, sizeof(buf), "Not tainted");
501 int test_taint(unsigned flag)
503 return test_bit(flag, &tainted_mask);
505 EXPORT_SYMBOL(test_taint);
507 unsigned long get_taint(void)
513 * add_taint: add a taint flag if not already set.
514 * @flag: one of the TAINT_* constants.
515 * @lockdep_ok: whether lock debugging is still OK.
517 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
518 * some notewortht-but-not-corrupting cases, it can be set to true.
520 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
522 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
523 pr_warn("Disabling lock debugging due to kernel taint\n");
525 set_bit(flag, &tainted_mask);
527 if (tainted_mask & panic_on_taint) {
529 panic("panic_on_taint set ...");
532 EXPORT_SYMBOL(add_taint);
534 static void spin_msec(int msecs)
538 for (i = 0; i < msecs; i++) {
539 touch_nmi_watchdog();
545 * It just happens that oops_enter() and oops_exit() are identically
548 static void do_oops_enter_exit(void)
551 static int spin_counter;
556 spin_lock_irqsave(&pause_on_oops_lock, flags);
557 if (pause_on_oops_flag == 0) {
558 /* This CPU may now print the oops message */
559 pause_on_oops_flag = 1;
561 /* We need to stall this CPU */
563 /* This CPU gets to do the counting */
564 spin_counter = pause_on_oops;
566 spin_unlock(&pause_on_oops_lock);
567 spin_msec(MSEC_PER_SEC);
568 spin_lock(&pause_on_oops_lock);
569 } while (--spin_counter);
570 pause_on_oops_flag = 0;
572 /* This CPU waits for a different one */
573 while (spin_counter) {
574 spin_unlock(&pause_on_oops_lock);
576 spin_lock(&pause_on_oops_lock);
580 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
584 * Return true if the calling CPU is allowed to print oops-related info.
585 * This is a bit racy..
587 bool oops_may_print(void)
589 return pause_on_oops_flag == 0;
593 * Called when the architecture enters its oops handler, before it prints
594 * anything. If this is the first CPU to oops, and it's oopsing the first
595 * time then let it proceed.
597 * This is all enabled by the pause_on_oops kernel boot option. We do all
598 * this to ensure that oopses don't scroll off the screen. It has the
599 * side-effect of preventing later-oopsing CPUs from mucking up the display,
602 * It turns out that the CPU which is allowed to print ends up pausing for
603 * the right duration, whereas all the other CPUs pause for twice as long:
604 * once in oops_enter(), once in oops_exit().
606 void oops_enter(void)
609 /* can't trust the integrity of the kernel anymore: */
611 do_oops_enter_exit();
613 if (sysctl_oops_all_cpu_backtrace)
614 trigger_all_cpu_backtrace();
618 * 64-bit random ID for oopses:
622 static int init_oops_id(void)
624 #ifndef CONFIG_PREEMPT_RT
626 get_random_bytes(&oops_id, sizeof(oops_id));
633 late_initcall(init_oops_id);
635 static void print_oops_end_marker(void)
638 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
639 pr_flush(1000, true);
643 * Called when the architecture exits its oops handler, after printing
648 do_oops_enter_exit();
649 print_oops_end_marker();
650 kmsg_dump(KMSG_DUMP_OOPS);
658 void __warn(const char *file, int line, void *caller, unsigned taint,
659 struct pt_regs *regs, struct warn_args *args)
661 disable_trace_on_warning();
664 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
665 raw_smp_processor_id(), current->pid, file, line,
668 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
669 raw_smp_processor_id(), current->pid, caller);
672 vprintk(args->fmt, args->args);
679 check_panic_on_warn("kernel");
684 print_irqtrace_events(current);
686 print_oops_end_marker();
688 /* Just a warning, don't kill lockdep. */
689 add_taint(taint, LOCKDEP_STILL_OK);
693 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
694 const char *fmt, ...)
696 struct warn_args args;
701 __warn(file, line, __builtin_return_address(0), taint,
707 va_start(args.args, fmt);
708 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
711 EXPORT_SYMBOL(warn_slowpath_fmt);
713 void __warn_printk(const char *fmt, ...)
723 EXPORT_SYMBOL(__warn_printk);
728 /* Support resetting WARN*_ONCE state */
730 static int clear_warn_once_set(void *data, u64 val)
732 generic_bug_clear_once();
733 memset(__start_once, 0, __end_once - __start_once);
737 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
740 static __init int register_warn_debugfs(void)
742 /* Don't care about failure */
743 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
744 &clear_warn_once_fops);
748 device_initcall(register_warn_debugfs);
751 #ifdef CONFIG_STACKPROTECTOR
754 * Called when gcc's -fstack-protector feature is used, and
755 * gcc detects corruption of the on-stack canary value
757 __visible noinstr void __stack_chk_fail(void)
759 instrumentation_begin();
760 panic("stack-protector: Kernel stack is corrupted in: %pB",
761 __builtin_return_address(0));
762 instrumentation_end();
764 EXPORT_SYMBOL(__stack_chk_fail);
768 core_param(panic, panic_timeout, int, 0644);
769 core_param(panic_print, panic_print, ulong, 0644);
770 core_param(pause_on_oops, pause_on_oops, int, 0644);
771 core_param(panic_on_warn, panic_on_warn, int, 0644);
772 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
774 static int __init oops_setup(char *s)
778 if (!strcmp(s, "panic"))
782 early_param("oops", oops_setup);
784 static int __init panic_on_taint_setup(char *s)
791 taint_str = strsep(&s, ",");
792 if (kstrtoul(taint_str, 16, &panic_on_taint))
795 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
796 panic_on_taint &= TAINT_FLAGS_MAX;
801 if (s && !strcmp(s, "nousertaint"))
802 panic_on_taint_nousertaint = true;
804 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
805 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
809 early_param("panic_on_taint", panic_on_taint_setup);