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 <trace/events/error_report.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_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;
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 #define PANIC_PRINT_ALL_CPU_BT 0x00000040
72 unsigned long panic_print;
74 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
76 EXPORT_SYMBOL(panic_notifier_list);
78 #if defined(CONFIG_SMP) && defined(CONFIG_SYSCTL)
79 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,
92 static __init int kernel_panic_sysctls_init(void)
94 register_sysctl_init("kernel", kern_panic_table);
97 late_initcall(kernel_panic_sysctls_init);
100 static long no_blink(int state)
105 /* Returns how long it waited in ms */
106 long (*panic_blink)(int state);
107 EXPORT_SYMBOL(panic_blink);
110 * Stop ourself in panic -- architecture code may override this
112 void __weak panic_smp_self_stop(void)
119 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
120 * may override this to prepare for crash dumping, e.g. save regs info.
122 void __weak nmi_panic_self_stop(struct pt_regs *regs)
124 panic_smp_self_stop();
128 * Stop other CPUs in panic. Architecture dependent code may override this
129 * with more suitable version. For example, if the architecture supports
130 * crash dump, it should save registers of each stopped CPU and disable
131 * per-CPU features such as virtualization extensions.
133 void __weak crash_smp_send_stop(void)
135 static int cpus_stopped;
138 * This function can be called twice in panic path, but obviously
139 * we execute this only once.
145 * Note smp_send_stop is the usual smp shutdown function, which
146 * unfortunately means it may not be hardened to work in a panic
153 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
156 * A variant of panic() called from NMI context. We return if we've already
157 * panicked on this CPU. If another CPU already panicked, loop in
158 * nmi_panic_self_stop() which can provide architecture dependent code such
159 * as saving register state for crash dump.
161 void nmi_panic(struct pt_regs *regs, const char *msg)
165 cpu = raw_smp_processor_id();
166 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
168 if (old_cpu == PANIC_CPU_INVALID)
170 else if (old_cpu != cpu)
171 nmi_panic_self_stop(regs);
173 EXPORT_SYMBOL(nmi_panic);
175 static void panic_print_sys_info(bool console_flush)
178 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
179 console_flush_on_panic(CONSOLE_REPLAY_ALL);
183 if (panic_print & PANIC_PRINT_ALL_CPU_BT)
184 trigger_all_cpu_backtrace();
186 if (panic_print & PANIC_PRINT_TASK_INFO)
189 if (panic_print & PANIC_PRINT_MEM_INFO)
192 if (panic_print & PANIC_PRINT_TIMER_INFO)
193 sysrq_timer_list_show();
195 if (panic_print & PANIC_PRINT_LOCK_INFO)
196 debug_show_all_locks();
198 if (panic_print & PANIC_PRINT_FTRACE_INFO)
199 ftrace_dump(DUMP_ALL);
203 * panic - halt the system
204 * @fmt: The text string to print
206 * Display a message, then perform cleanups.
208 * This function never returns.
210 void panic(const char *fmt, ...)
212 static char buf[1024];
214 long i, i_next = 0, len;
216 int old_cpu, this_cpu;
217 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
221 * This thread may hit another WARN() in the panic path.
222 * Resetting this prevents additional WARN() from panicking the
223 * system on this thread. Other threads are blocked by the
224 * panic_mutex in panic().
230 * Disable local interrupts. This will prevent panic_smp_self_stop
231 * from deadlocking the first cpu that invokes the panic, since
232 * there is nothing to prevent an interrupt handler (that runs
233 * after setting panic_cpu) from invoking panic() again.
236 preempt_disable_notrace();
239 * It's possible to come here directly from a panic-assertion and
240 * not have preempt disabled. Some functions called from here want
241 * preempt to be disabled. No point enabling it later though...
243 * Only one CPU is allowed to execute the panic code from here. For
244 * multiple parallel invocations of panic, all other CPUs either
245 * stop themself or will wait until they are stopped by the 1st CPU
246 * with smp_send_stop().
248 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
249 * comes here, so go ahead.
250 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
251 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
253 this_cpu = raw_smp_processor_id();
254 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
256 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
257 panic_smp_self_stop();
262 len = vscnprintf(buf, sizeof(buf), fmt, args);
265 if (len && buf[len - 1] == '\n')
268 pr_emerg("Kernel panic - not syncing: %s\n", buf);
269 #ifdef CONFIG_DEBUG_BUGVERBOSE
271 * Avoid nested stack-dumping if a panic occurs during oops processing
273 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
278 * If kgdb is enabled, give it a chance to run before we stop all
279 * the other CPUs or else we won't be able to debug processes left
285 * If we have crashed and we have a crash kernel loaded let it handle
287 * If we want to run this after calling panic_notifiers, pass
288 * the "crash_kexec_post_notifiers" option to the kernel.
290 * Bypass the panic_cpu check and call __crash_kexec directly.
292 if (!_crash_kexec_post_notifiers) {
296 * Note smp_send_stop is the usual smp shutdown function, which
297 * unfortunately means it may not be hardened to work in a
303 * If we want to do crash dump after notifier calls and
304 * kmsg_dump, we will need architecture dependent extra
305 * works in addition to stopping other CPUs.
307 crash_smp_send_stop();
311 * Run any panic handlers, including those that might need to
312 * add information to the kmsg dump output.
314 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
316 panic_print_sys_info(false);
318 kmsg_dump(KMSG_DUMP_PANIC);
321 * If you doubt kdump always works fine in any situation,
322 * "crash_kexec_post_notifiers" offers you a chance to run
323 * panic_notifiers and dumping kmsg before kdump.
324 * Note: since some panic_notifiers can make crashed kernel
325 * more unstable, it can increase risks of the kdump failure too.
327 * Bypass the panic_cpu check and call __crash_kexec directly.
329 if (_crash_kexec_post_notifiers)
338 * We may have ended up stopping the CPU holding the lock (in
339 * smp_send_stop()) while still having some valuable data in the console
340 * buffer. Try to acquire the lock then release it regardless of the
341 * result. The release will also print the buffers out. Locks debug
342 * should be disabled to avoid reporting bad unlock balance when
343 * panic() is not being callled from OOPS.
346 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
348 panic_print_sys_info(true);
351 panic_blink = no_blink;
353 if (panic_timeout > 0) {
355 * Delay timeout seconds before rebooting the machine.
356 * We can't use the "normal" timers since we just panicked.
358 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
360 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
361 touch_nmi_watchdog();
363 i += panic_blink(state ^= 1);
364 i_next = i + 3600 / PANIC_BLINK_SPD;
366 mdelay(PANIC_TIMER_STEP);
369 if (panic_timeout != 0) {
371 * This will not be a clean reboot, with everything
372 * shutting down. But if there is a chance of
373 * rebooting the system it will be rebooted.
375 if (panic_reboot_mode != REBOOT_UNDEFINED)
376 reboot_mode = panic_reboot_mode;
381 extern int stop_a_enabled;
382 /* Make sure the user can actually press Stop-A (L1-A) */
384 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
385 "twice on console to return to the boot prom\n");
388 #if defined(CONFIG_S390)
391 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
393 /* Do not scroll important messages printed above */
396 for (i = 0; ; i += PANIC_TIMER_STEP) {
397 touch_softlockup_watchdog();
399 i += panic_blink(state ^= 1);
400 i_next = i + 3600 / PANIC_BLINK_SPD;
402 mdelay(PANIC_TIMER_STEP);
406 EXPORT_SYMBOL(panic);
409 * TAINT_FORCED_RMMOD could be a per-module flag but the module
410 * is being removed anyway.
412 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
413 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
414 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
415 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
416 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
417 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
418 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
419 [ TAINT_USER ] = { 'U', ' ', false },
420 [ TAINT_DIE ] = { 'D', ' ', false },
421 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
422 [ TAINT_WARN ] = { 'W', ' ', false },
423 [ TAINT_CRAP ] = { 'C', ' ', true },
424 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
425 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
426 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
427 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
428 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
429 [ TAINT_AUX ] = { 'X', ' ', true },
430 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
434 * print_tainted - return a string to represent the kernel taint state.
436 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
438 * The string is overwritten by the next call to print_tainted(),
439 * but is always NULL terminated.
441 const char *print_tainted(void)
443 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
445 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
451 s = buf + sprintf(buf, "Tainted: ");
452 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
453 const struct taint_flag *t = &taint_flags[i];
454 *s++ = test_bit(i, &tainted_mask) ?
455 t->c_true : t->c_false;
459 snprintf(buf, sizeof(buf), "Not tainted");
464 int test_taint(unsigned flag)
466 return test_bit(flag, &tainted_mask);
468 EXPORT_SYMBOL(test_taint);
470 unsigned long get_taint(void)
476 * add_taint: add a taint flag if not already set.
477 * @flag: one of the TAINT_* constants.
478 * @lockdep_ok: whether lock debugging is still OK.
480 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
481 * some notewortht-but-not-corrupting cases, it can be set to true.
483 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
485 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
486 pr_warn("Disabling lock debugging due to kernel taint\n");
488 set_bit(flag, &tainted_mask);
490 if (tainted_mask & panic_on_taint) {
492 panic("panic_on_taint set ...");
495 EXPORT_SYMBOL(add_taint);
497 static void spin_msec(int msecs)
501 for (i = 0; i < msecs; i++) {
502 touch_nmi_watchdog();
508 * It just happens that oops_enter() and oops_exit() are identically
511 static void do_oops_enter_exit(void)
514 static int spin_counter;
519 spin_lock_irqsave(&pause_on_oops_lock, flags);
520 if (pause_on_oops_flag == 0) {
521 /* This CPU may now print the oops message */
522 pause_on_oops_flag = 1;
524 /* We need to stall this CPU */
526 /* This CPU gets to do the counting */
527 spin_counter = pause_on_oops;
529 spin_unlock(&pause_on_oops_lock);
530 spin_msec(MSEC_PER_SEC);
531 spin_lock(&pause_on_oops_lock);
532 } while (--spin_counter);
533 pause_on_oops_flag = 0;
535 /* This CPU waits for a different one */
536 while (spin_counter) {
537 spin_unlock(&pause_on_oops_lock);
539 spin_lock(&pause_on_oops_lock);
543 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
547 * Return true if the calling CPU is allowed to print oops-related info.
548 * This is a bit racy..
550 bool oops_may_print(void)
552 return pause_on_oops_flag == 0;
556 * Called when the architecture enters its oops handler, before it prints
557 * anything. If this is the first CPU to oops, and it's oopsing the first
558 * time then let it proceed.
560 * This is all enabled by the pause_on_oops kernel boot option. We do all
561 * this to ensure that oopses don't scroll off the screen. It has the
562 * side-effect of preventing later-oopsing CPUs from mucking up the display,
565 * It turns out that the CPU which is allowed to print ends up pausing for
566 * the right duration, whereas all the other CPUs pause for twice as long:
567 * once in oops_enter(), once in oops_exit().
569 void oops_enter(void)
572 /* can't trust the integrity of the kernel anymore: */
574 do_oops_enter_exit();
576 if (sysctl_oops_all_cpu_backtrace)
577 trigger_all_cpu_backtrace();
580 static void print_oops_end_marker(void)
582 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
586 * Called when the architecture exits its oops handler, after printing
591 do_oops_enter_exit();
592 print_oops_end_marker();
593 kmsg_dump(KMSG_DUMP_OOPS);
601 void __warn(const char *file, int line, void *caller, unsigned taint,
602 struct pt_regs *regs, struct warn_args *args)
604 disable_trace_on_warning();
606 printk_prefer_direct_enter();
609 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
610 raw_smp_processor_id(), current->pid, file, line,
613 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
614 raw_smp_processor_id(), current->pid, caller);
617 vprintk(args->fmt, args->args);
625 panic("panic_on_warn set ...\n");
630 print_irqtrace_events(current);
632 print_oops_end_marker();
633 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
635 /* Just a warning, don't kill lockdep. */
636 add_taint(taint, LOCKDEP_STILL_OK);
638 printk_prefer_direct_exit();
642 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
643 const char *fmt, ...)
645 struct warn_args args;
650 __warn(file, line, __builtin_return_address(0), taint,
656 va_start(args.args, fmt);
657 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
660 EXPORT_SYMBOL(warn_slowpath_fmt);
662 void __warn_printk(const char *fmt, ...)
672 EXPORT_SYMBOL(__warn_printk);
677 /* Support resetting WARN*_ONCE state */
679 static int clear_warn_once_set(void *data, u64 val)
681 generic_bug_clear_once();
682 memset(__start_once, 0, __end_once - __start_once);
686 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
689 static __init int register_warn_debugfs(void)
691 /* Don't care about failure */
692 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
693 &clear_warn_once_fops);
697 device_initcall(register_warn_debugfs);
700 #ifdef CONFIG_STACKPROTECTOR
703 * Called when gcc's -fstack-protector feature is used, and
704 * gcc detects corruption of the on-stack canary value
706 __visible noinstr void __stack_chk_fail(void)
708 instrumentation_begin();
709 panic("stack-protector: Kernel stack is corrupted in: %pB",
710 __builtin_return_address(0));
711 instrumentation_end();
713 EXPORT_SYMBOL(__stack_chk_fail);
717 core_param(panic, panic_timeout, int, 0644);
718 core_param(panic_print, panic_print, ulong, 0644);
719 core_param(pause_on_oops, pause_on_oops, int, 0644);
720 core_param(panic_on_warn, panic_on_warn, int, 0644);
721 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
723 static int __init oops_setup(char *s)
727 if (!strcmp(s, "panic"))
731 early_param("oops", oops_setup);
733 static int __init panic_on_taint_setup(char *s)
740 taint_str = strsep(&s, ",");
741 if (kstrtoul(taint_str, 16, &panic_on_taint))
744 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
745 panic_on_taint &= TAINT_FLAGS_MAX;
750 if (s && !strcmp(s, "nousertaint"))
751 panic_on_taint_nousertaint = true;
753 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
754 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
758 early_param("panic_on_taint", panic_on_taint_setup);