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 <linux/context_tracking.h>
38 #include <trace/events/error_report.h>
39 #include <asm/sections.h>
41 #define PANIC_TIMER_STEP 100
42 #define PANIC_BLINK_SPD 18
46 * Should we dump all CPUs backtraces in an oops event?
47 * Defaults to 0, can be changed via sysctl.
49 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
51 #define sysctl_oops_all_cpu_backtrace 0
52 #endif /* CONFIG_SMP */
54 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
55 static unsigned long tainted_mask =
56 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
57 static int pause_on_oops;
58 static int pause_on_oops_flag;
59 static DEFINE_SPINLOCK(pause_on_oops_lock);
60 bool crash_kexec_post_notifiers;
61 int panic_on_warn __read_mostly;
62 unsigned long panic_on_taint;
63 bool panic_on_taint_nousertaint = false;
64 static unsigned int warn_limit __read_mostly;
66 int panic_timeout = CONFIG_PANIC_TIMEOUT;
67 EXPORT_SYMBOL_GPL(panic_timeout);
69 #define PANIC_PRINT_TASK_INFO 0x00000001
70 #define PANIC_PRINT_MEM_INFO 0x00000002
71 #define PANIC_PRINT_TIMER_INFO 0x00000004
72 #define PANIC_PRINT_LOCK_INFO 0x00000008
73 #define PANIC_PRINT_FTRACE_INFO 0x00000010
74 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
75 #define PANIC_PRINT_ALL_CPU_BT 0x00000040
76 unsigned long panic_print;
78 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
80 EXPORT_SYMBOL(panic_notifier_list);
83 static struct ctl_table kern_panic_table[] = {
86 .procname = "oops_all_cpu_backtrace",
87 .data = &sysctl_oops_all_cpu_backtrace,
88 .maxlen = sizeof(int),
90 .proc_handler = proc_dointvec_minmax,
91 .extra1 = SYSCTL_ZERO,
96 .procname = "warn_limit",
98 .maxlen = sizeof(warn_limit),
100 .proc_handler = proc_douintvec,
105 static __init int kernel_panic_sysctls_init(void)
107 register_sysctl_init("kernel", kern_panic_table);
110 late_initcall(kernel_panic_sysctls_init);
113 static atomic_t warn_count = ATOMIC_INIT(0);
116 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
119 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
122 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
124 static __init int kernel_panic_sysfs_init(void)
126 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
129 late_initcall(kernel_panic_sysfs_init);
132 static long no_blink(int state)
137 /* Returns how long it waited in ms */
138 long (*panic_blink)(int state);
139 EXPORT_SYMBOL(panic_blink);
142 * Stop ourself in panic -- architecture code may override this
144 void __weak panic_smp_self_stop(void)
151 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
152 * may override this to prepare for crash dumping, e.g. save regs info.
154 void __weak nmi_panic_self_stop(struct pt_regs *regs)
156 panic_smp_self_stop();
160 * Stop other CPUs in panic. Architecture dependent code may override this
161 * with more suitable version. For example, if the architecture supports
162 * crash dump, it should save registers of each stopped CPU and disable
163 * per-CPU features such as virtualization extensions.
165 void __weak crash_smp_send_stop(void)
167 static int cpus_stopped;
170 * This function can be called twice in panic path, but obviously
171 * we execute this only once.
177 * Note smp_send_stop is the usual smp shutdown function, which
178 * unfortunately means it may not be hardened to work in a panic
185 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
188 * A variant of panic() called from NMI context. We return if we've already
189 * panicked on this CPU. If another CPU already panicked, loop in
190 * nmi_panic_self_stop() which can provide architecture dependent code such
191 * as saving register state for crash dump.
193 void nmi_panic(struct pt_regs *regs, const char *msg)
197 cpu = raw_smp_processor_id();
198 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
200 if (old_cpu == PANIC_CPU_INVALID)
202 else if (old_cpu != cpu)
203 nmi_panic_self_stop(regs);
205 EXPORT_SYMBOL(nmi_panic);
207 static void panic_print_sys_info(bool console_flush)
210 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
211 console_flush_on_panic(CONSOLE_REPLAY_ALL);
215 if (panic_print & PANIC_PRINT_TASK_INFO)
218 if (panic_print & PANIC_PRINT_MEM_INFO)
221 if (panic_print & PANIC_PRINT_TIMER_INFO)
222 sysrq_timer_list_show();
224 if (panic_print & PANIC_PRINT_LOCK_INFO)
225 debug_show_all_locks();
227 if (panic_print & PANIC_PRINT_FTRACE_INFO)
228 ftrace_dump(DUMP_ALL);
231 void check_panic_on_warn(const char *origin)
236 panic("%s: panic_on_warn set ...\n", origin);
238 limit = READ_ONCE(warn_limit);
239 if (atomic_inc_return(&warn_count) >= limit && limit)
240 panic("%s: system warned too often (kernel.warn_limit is %d)",
245 * Helper that triggers the NMI backtrace (if set in panic_print)
246 * and then performs the secondary CPUs shutdown - we cannot have
247 * the NMI backtrace after the CPUs are off!
249 static void panic_other_cpus_shutdown(bool crash_kexec)
251 if (panic_print & PANIC_PRINT_ALL_CPU_BT)
252 trigger_all_cpu_backtrace();
255 * Note that smp_send_stop() is the usual SMP shutdown function,
256 * which unfortunately may not be hardened to work in a panic
257 * situation. If we want to do crash dump after notifier calls
258 * and kmsg_dump, we will need architecture dependent extra
259 * bits in addition to stopping other CPUs, hence we rely on
260 * crash_smp_send_stop() for that.
265 crash_smp_send_stop();
269 * panic - halt the system
270 * @fmt: The text string to print
272 * Display a message, then perform cleanups.
274 * This function never returns.
276 void panic(const char *fmt, ...)
278 static char buf[1024];
280 long i, i_next = 0, len;
282 int old_cpu, this_cpu;
283 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
287 * This thread may hit another WARN() in the panic path.
288 * Resetting this prevents additional WARN() from panicking the
289 * system on this thread. Other threads are blocked by the
290 * panic_mutex in panic().
296 * Disable local interrupts. This will prevent panic_smp_self_stop
297 * from deadlocking the first cpu that invokes the panic, since
298 * there is nothing to prevent an interrupt handler (that runs
299 * after setting panic_cpu) from invoking panic() again.
302 preempt_disable_notrace();
305 * It's possible to come here directly from a panic-assertion and
306 * not have preempt disabled. Some functions called from here want
307 * preempt to be disabled. No point enabling it later though...
309 * Only one CPU is allowed to execute the panic code from here. For
310 * multiple parallel invocations of panic, all other CPUs either
311 * stop themself or will wait until they are stopped by the 1st CPU
312 * with smp_send_stop().
314 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
315 * comes here, so go ahead.
316 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
317 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
319 this_cpu = raw_smp_processor_id();
320 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
322 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
323 panic_smp_self_stop();
328 len = vscnprintf(buf, sizeof(buf), fmt, args);
331 if (len && buf[len - 1] == '\n')
334 pr_emerg("Kernel panic - not syncing: %s\n", buf);
335 #ifdef CONFIG_DEBUG_BUGVERBOSE
337 * Avoid nested stack-dumping if a panic occurs during oops processing
339 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
344 * If kgdb is enabled, give it a chance to run before we stop all
345 * the other CPUs or else we won't be able to debug processes left
351 * If we have crashed and we have a crash kernel loaded let it handle
353 * If we want to run this after calling panic_notifiers, pass
354 * the "crash_kexec_post_notifiers" option to the kernel.
356 * Bypass the panic_cpu check and call __crash_kexec directly.
358 if (!_crash_kexec_post_notifiers)
361 panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
364 * Run any panic handlers, including those that might need to
365 * add information to the kmsg dump output.
367 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
369 panic_print_sys_info(false);
371 kmsg_dump(KMSG_DUMP_PANIC);
374 * If you doubt kdump always works fine in any situation,
375 * "crash_kexec_post_notifiers" offers you a chance to run
376 * panic_notifiers and dumping kmsg before kdump.
377 * Note: since some panic_notifiers can make crashed kernel
378 * more unstable, it can increase risks of the kdump failure too.
380 * Bypass the panic_cpu check and call __crash_kexec directly.
382 if (_crash_kexec_post_notifiers)
388 * We may have ended up stopping the CPU holding the lock (in
389 * smp_send_stop()) while still having some valuable data in the console
390 * buffer. Try to acquire the lock then release it regardless of the
391 * result. The release will also print the buffers out. Locks debug
392 * should be disabled to avoid reporting bad unlock balance when
393 * panic() is not being callled from OOPS.
396 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
398 panic_print_sys_info(true);
401 panic_blink = no_blink;
403 if (panic_timeout > 0) {
405 * Delay timeout seconds before rebooting the machine.
406 * We can't use the "normal" timers since we just panicked.
408 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
410 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
411 touch_nmi_watchdog();
413 i += panic_blink(state ^= 1);
414 i_next = i + 3600 / PANIC_BLINK_SPD;
416 mdelay(PANIC_TIMER_STEP);
419 if (panic_timeout != 0) {
421 * This will not be a clean reboot, with everything
422 * shutting down. But if there is a chance of
423 * rebooting the system it will be rebooted.
425 if (panic_reboot_mode != REBOOT_UNDEFINED)
426 reboot_mode = panic_reboot_mode;
431 extern int stop_a_enabled;
432 /* Make sure the user can actually press Stop-A (L1-A) */
434 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
435 "twice on console to return to the boot prom\n");
438 #if defined(CONFIG_S390)
441 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
443 /* Do not scroll important messages printed above */
446 for (i = 0; ; i += PANIC_TIMER_STEP) {
447 touch_softlockup_watchdog();
449 i += panic_blink(state ^= 1);
450 i_next = i + 3600 / PANIC_BLINK_SPD;
452 mdelay(PANIC_TIMER_STEP);
456 EXPORT_SYMBOL(panic);
459 * TAINT_FORCED_RMMOD could be a per-module flag but the module
460 * is being removed anyway.
462 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
463 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
464 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
465 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
466 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
467 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
468 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
469 [ TAINT_USER ] = { 'U', ' ', false },
470 [ TAINT_DIE ] = { 'D', ' ', false },
471 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
472 [ TAINT_WARN ] = { 'W', ' ', false },
473 [ TAINT_CRAP ] = { 'C', ' ', true },
474 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
475 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
476 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
477 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
478 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
479 [ TAINT_AUX ] = { 'X', ' ', true },
480 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
481 [ TAINT_TEST ] = { 'N', ' ', true },
485 * print_tainted - return a string to represent the kernel taint state.
487 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
489 * The string is overwritten by the next call to print_tainted(),
490 * but is always NULL terminated.
492 const char *print_tainted(void)
494 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
496 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
502 s = buf + sprintf(buf, "Tainted: ");
503 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
504 const struct taint_flag *t = &taint_flags[i];
505 *s++ = test_bit(i, &tainted_mask) ?
506 t->c_true : t->c_false;
510 snprintf(buf, sizeof(buf), "Not tainted");
515 int test_taint(unsigned flag)
517 return test_bit(flag, &tainted_mask);
519 EXPORT_SYMBOL(test_taint);
521 unsigned long get_taint(void)
527 * add_taint: add a taint flag if not already set.
528 * @flag: one of the TAINT_* constants.
529 * @lockdep_ok: whether lock debugging is still OK.
531 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
532 * some notewortht-but-not-corrupting cases, it can be set to true.
534 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
536 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
537 pr_warn("Disabling lock debugging due to kernel taint\n");
539 set_bit(flag, &tainted_mask);
541 if (tainted_mask & panic_on_taint) {
543 panic("panic_on_taint set ...");
546 EXPORT_SYMBOL(add_taint);
548 static void spin_msec(int msecs)
552 for (i = 0; i < msecs; i++) {
553 touch_nmi_watchdog();
559 * It just happens that oops_enter() and oops_exit() are identically
562 static void do_oops_enter_exit(void)
565 static int spin_counter;
570 spin_lock_irqsave(&pause_on_oops_lock, flags);
571 if (pause_on_oops_flag == 0) {
572 /* This CPU may now print the oops message */
573 pause_on_oops_flag = 1;
575 /* We need to stall this CPU */
577 /* This CPU gets to do the counting */
578 spin_counter = pause_on_oops;
580 spin_unlock(&pause_on_oops_lock);
581 spin_msec(MSEC_PER_SEC);
582 spin_lock(&pause_on_oops_lock);
583 } while (--spin_counter);
584 pause_on_oops_flag = 0;
586 /* This CPU waits for a different one */
587 while (spin_counter) {
588 spin_unlock(&pause_on_oops_lock);
590 spin_lock(&pause_on_oops_lock);
594 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
598 * Return true if the calling CPU is allowed to print oops-related info.
599 * This is a bit racy..
601 bool oops_may_print(void)
603 return pause_on_oops_flag == 0;
607 * Called when the architecture enters its oops handler, before it prints
608 * anything. If this is the first CPU to oops, and it's oopsing the first
609 * time then let it proceed.
611 * This is all enabled by the pause_on_oops kernel boot option. We do all
612 * this to ensure that oopses don't scroll off the screen. It has the
613 * side-effect of preventing later-oopsing CPUs from mucking up the display,
616 * It turns out that the CPU which is allowed to print ends up pausing for
617 * the right duration, whereas all the other CPUs pause for twice as long:
618 * once in oops_enter(), once in oops_exit().
620 void oops_enter(void)
623 /* can't trust the integrity of the kernel anymore: */
625 do_oops_enter_exit();
627 if (sysctl_oops_all_cpu_backtrace)
628 trigger_all_cpu_backtrace();
631 static void print_oops_end_marker(void)
633 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
637 * Called when the architecture exits its oops handler, after printing
642 do_oops_enter_exit();
643 print_oops_end_marker();
644 kmsg_dump(KMSG_DUMP_OOPS);
652 void __warn(const char *file, int line, void *caller, unsigned taint,
653 struct pt_regs *regs, struct warn_args *args)
655 disable_trace_on_warning();
658 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
659 raw_smp_processor_id(), current->pid, file, line,
662 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
663 raw_smp_processor_id(), current->pid, caller);
666 vprintk(args->fmt, args->args);
673 check_panic_on_warn("kernel");
678 print_irqtrace_events(current);
680 print_oops_end_marker();
681 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
683 /* Just a warning, don't kill lockdep. */
684 add_taint(taint, LOCKDEP_STILL_OK);
688 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
689 const char *fmt, ...)
691 bool rcu = warn_rcu_enter();
692 struct warn_args args;
697 __warn(file, line, __builtin_return_address(0), taint,
703 va_start(args.args, fmt);
704 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
708 EXPORT_SYMBOL(warn_slowpath_fmt);
710 void __warn_printk(const char *fmt, ...)
712 bool rcu = warn_rcu_enter();
722 EXPORT_SYMBOL(__warn_printk);
727 /* Support resetting WARN*_ONCE state */
729 static int clear_warn_once_set(void *data, u64 val)
731 generic_bug_clear_once();
732 memset(__start_once, 0, __end_once - __start_once);
736 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
739 static __init int register_warn_debugfs(void)
741 /* Don't care about failure */
742 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
743 &clear_warn_once_fops);
747 device_initcall(register_warn_debugfs);
750 #ifdef CONFIG_STACKPROTECTOR
753 * Called when gcc's -fstack-protector feature is used, and
754 * gcc detects corruption of the on-stack canary value
756 __visible noinstr void __stack_chk_fail(void)
758 instrumentation_begin();
759 panic("stack-protector: Kernel stack is corrupted in: %pB",
760 __builtin_return_address(0));
761 instrumentation_end();
763 EXPORT_SYMBOL(__stack_chk_fail);
767 core_param(panic, panic_timeout, int, 0644);
768 core_param(panic_print, panic_print, ulong, 0644);
769 core_param(pause_on_oops, pause_on_oops, int, 0644);
770 core_param(panic_on_warn, panic_on_warn, int, 0644);
771 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
773 static int __init oops_setup(char *s)
777 if (!strcmp(s, "panic"))
781 early_param("oops", oops_setup);
783 static int __init panic_on_taint_setup(char *s)
790 taint_str = strsep(&s, ",");
791 if (kstrtoul(taint_str, 16, &panic_on_taint))
794 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
795 panic_on_taint &= TAINT_FLAGS_MAX;
800 if (s && !strcmp(s, "nousertaint"))
801 panic_on_taint_nousertaint = true;
803 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
804 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
808 early_param("panic_on_taint", panic_on_taint_setup);