4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * This function is used through-out the kernel (including mm and fs)
9 * to indicate a major problem.
11 #include <linux/debug_locks.h>
12 #include <linux/sched/debug.h>
13 #include <linux/interrupt.h>
14 #include <linux/kmsg_dump.h>
15 #include <linux/kallsyms.h>
16 #include <linux/notifier.h>
17 #include <linux/vt_kern.h>
18 #include <linux/module.h>
19 #include <linux/random.h>
20 #include <linux/ftrace.h>
21 #include <linux/reboot.h>
22 #include <linux/delay.h>
23 #include <linux/kexec.h>
24 #include <linux/sched.h>
25 #include <linux/sysrq.h>
26 #include <linux/init.h>
27 #include <linux/nmi.h>
28 #include <linux/console.h>
29 #include <linux/bug.h>
30 #include <linux/ratelimit.h>
31 #include <linux/sysfs.h>
33 #define PANIC_TIMER_STEP 100
34 #define PANIC_BLINK_SPD 18
36 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
37 static unsigned long tainted_mask;
38 static int pause_on_oops;
39 static int pause_on_oops_flag;
40 static DEFINE_SPINLOCK(pause_on_oops_lock);
41 bool crash_kexec_post_notifiers;
42 int panic_on_warn __read_mostly;
43 static unsigned int warn_limit __read_mostly;
45 int panic_timeout = CONFIG_PANIC_TIMEOUT;
46 EXPORT_SYMBOL_GPL(panic_timeout);
48 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
50 EXPORT_SYMBOL(panic_notifier_list);
53 static struct ctl_table kern_panic_table[] = {
55 .procname = "warn_limit",
57 .maxlen = sizeof(warn_limit),
59 .proc_handler = proc_douintvec,
64 static __init int kernel_panic_sysctls_init(void)
66 register_sysctl_init("kernel", kern_panic_table);
69 late_initcall(kernel_panic_sysctls_init);
72 static atomic_t warn_count = ATOMIC_INIT(0);
75 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
78 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
81 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
83 static __init int kernel_panic_sysfs_init(void)
85 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
88 late_initcall(kernel_panic_sysfs_init);
91 static long no_blink(int state)
96 /* Returns how long it waited in ms */
97 long (*panic_blink)(int state);
98 EXPORT_SYMBOL(panic_blink);
101 * Stop ourself in panic -- architecture code may override this
103 void __weak panic_smp_self_stop(void)
110 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
111 * may override this to prepare for crash dumping, e.g. save regs info.
113 void __weak nmi_panic_self_stop(struct pt_regs *regs)
115 panic_smp_self_stop();
119 * Stop other CPUs in panic. Architecture dependent code may override this
120 * with more suitable version. For example, if the architecture supports
121 * crash dump, it should save registers of each stopped CPU and disable
122 * per-CPU features such as virtualization extensions.
124 void __weak crash_smp_send_stop(void)
126 static int cpus_stopped;
129 * This function can be called twice in panic path, but obviously
130 * we execute this only once.
136 * Note smp_send_stop is the usual smp shutdown function, which
137 * unfortunately means it may not be hardened to work in a panic
144 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
147 * A variant of panic() called from NMI context. We return if we've already
148 * panicked on this CPU. If another CPU already panicked, loop in
149 * nmi_panic_self_stop() which can provide architecture dependent code such
150 * as saving register state for crash dump.
152 void nmi_panic(struct pt_regs *regs, const char *msg)
156 cpu = raw_smp_processor_id();
157 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
159 if (old_cpu == PANIC_CPU_INVALID)
161 else if (old_cpu != cpu)
162 nmi_panic_self_stop(regs);
164 EXPORT_SYMBOL(nmi_panic);
166 void check_panic_on_warn(const char *origin)
171 panic("%s: panic_on_warn set ...\n", origin);
173 limit = READ_ONCE(warn_limit);
174 if (atomic_inc_return(&warn_count) >= limit && limit)
175 panic("%s: system warned too often (kernel.warn_limit is %d)",
180 * panic - halt the system
181 * @fmt: The text string to print
183 * Display a message, then perform cleanups.
185 * This function never returns.
187 void panic(const char *fmt, ...)
189 static char buf[1024];
193 int old_cpu, this_cpu;
194 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
198 * This thread may hit another WARN() in the panic path.
199 * Resetting this prevents additional WARN() from panicking the
200 * system on this thread. Other threads are blocked by the
201 * panic_mutex in panic().
207 * Disable local interrupts. This will prevent panic_smp_self_stop
208 * from deadlocking the first cpu that invokes the panic, since
209 * there is nothing to prevent an interrupt handler (that runs
210 * after setting panic_cpu) from invoking panic() again.
213 preempt_disable_notrace();
216 * It's possible to come here directly from a panic-assertion and
217 * not have preempt disabled. Some functions called from here want
218 * preempt to be disabled. No point enabling it later though...
220 * Only one CPU is allowed to execute the panic code from here. For
221 * multiple parallel invocations of panic, all other CPUs either
222 * stop themself or will wait until they are stopped by the 1st CPU
223 * with smp_send_stop().
225 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
226 * comes here, so go ahead.
227 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
228 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
230 this_cpu = raw_smp_processor_id();
231 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
233 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
234 panic_smp_self_stop();
239 vsnprintf(buf, sizeof(buf), fmt, args);
241 pr_emerg("Kernel panic - not syncing: %s\n", buf);
242 #ifdef CONFIG_DEBUG_BUGVERBOSE
244 * Avoid nested stack-dumping if a panic occurs during oops processing
246 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
251 * If we have crashed and we have a crash kernel loaded let it handle
253 * If we want to run this after calling panic_notifiers, pass
254 * the "crash_kexec_post_notifiers" option to the kernel.
256 * Bypass the panic_cpu check and call __crash_kexec directly.
258 if (!_crash_kexec_post_notifiers) {
259 printk_safe_flush_on_panic();
263 * Note smp_send_stop is the usual smp shutdown function, which
264 * unfortunately means it may not be hardened to work in a
270 * If we want to do crash dump after notifier calls and
271 * kmsg_dump, we will need architecture dependent extra
272 * works in addition to stopping other CPUs.
274 crash_smp_send_stop();
278 * Run any panic handlers, including those that might need to
279 * add information to the kmsg dump output.
281 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
283 /* Call flush even twice. It tries harder with a single online CPU */
284 printk_safe_flush_on_panic();
285 kmsg_dump(KMSG_DUMP_PANIC);
288 * If you doubt kdump always works fine in any situation,
289 * "crash_kexec_post_notifiers" offers you a chance to run
290 * panic_notifiers and dumping kmsg before kdump.
291 * Note: since some panic_notifiers can make crashed kernel
292 * more unstable, it can increase risks of the kdump failure too.
294 * Bypass the panic_cpu check and call __crash_kexec directly.
296 if (_crash_kexec_post_notifiers)
305 * We may have ended up stopping the CPU holding the lock (in
306 * smp_send_stop()) while still having some valuable data in the console
307 * buffer. Try to acquire the lock then release it regardless of the
308 * result. The release will also print the buffers out. Locks debug
309 * should be disabled to avoid reporting bad unlock balance when
310 * panic() is not being callled from OOPS.
313 console_flush_on_panic();
316 panic_blink = no_blink;
318 if (panic_timeout > 0) {
320 * Delay timeout seconds before rebooting the machine.
321 * We can't use the "normal" timers since we just panicked.
323 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
325 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
326 touch_nmi_watchdog();
328 i += panic_blink(state ^= 1);
329 i_next = i + 3600 / PANIC_BLINK_SPD;
331 mdelay(PANIC_TIMER_STEP);
334 if (panic_timeout != 0) {
336 * This will not be a clean reboot, with everything
337 * shutting down. But if there is a chance of
338 * rebooting the system it will be rebooted.
344 extern int stop_a_enabled;
345 /* Make sure the user can actually press Stop-A (L1-A) */
347 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
348 "twice on console to return to the boot prom\n");
351 #if defined(CONFIG_S390)
353 unsigned long caller;
355 caller = (unsigned long)__builtin_return_address(0);
356 disabled_wait(caller);
359 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
361 for (i = 0; ; i += PANIC_TIMER_STEP) {
362 touch_softlockup_watchdog();
364 i += panic_blink(state ^= 1);
365 i_next = i + 3600 / PANIC_BLINK_SPD;
367 mdelay(PANIC_TIMER_STEP);
371 EXPORT_SYMBOL(panic);
374 * TAINT_FORCED_RMMOD could be a per-module flag but the module
375 * is being removed anyway.
377 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
378 { 'P', 'G', true }, /* TAINT_PROPRIETARY_MODULE */
379 { 'F', ' ', true }, /* TAINT_FORCED_MODULE */
380 { 'S', ' ', false }, /* TAINT_CPU_OUT_OF_SPEC */
381 { 'R', ' ', false }, /* TAINT_FORCED_RMMOD */
382 { 'M', ' ', false }, /* TAINT_MACHINE_CHECK */
383 { 'B', ' ', false }, /* TAINT_BAD_PAGE */
384 { 'U', ' ', false }, /* TAINT_USER */
385 { 'D', ' ', false }, /* TAINT_DIE */
386 { 'A', ' ', false }, /* TAINT_OVERRIDDEN_ACPI_TABLE */
387 { 'W', ' ', false }, /* TAINT_WARN */
388 { 'C', ' ', true }, /* TAINT_CRAP */
389 { 'I', ' ', false }, /* TAINT_FIRMWARE_WORKAROUND */
390 { 'O', ' ', true }, /* TAINT_OOT_MODULE */
391 { 'E', ' ', true }, /* TAINT_UNSIGNED_MODULE */
392 { 'L', ' ', false }, /* TAINT_SOFTLOCKUP */
393 { 'K', ' ', true }, /* TAINT_LIVEPATCH */
397 * print_tainted - return a string to represent the kernel taint state.
399 * 'P' - Proprietary module has been loaded.
400 * 'F' - Module has been forcibly loaded.
401 * 'S' - SMP with CPUs not designed for SMP.
402 * 'R' - User forced a module unload.
403 * 'M' - System experienced a machine check exception.
404 * 'B' - System has hit bad_page.
405 * 'U' - Userspace-defined naughtiness.
406 * 'D' - Kernel has oopsed before
407 * 'A' - ACPI table overridden.
408 * 'W' - Taint on warning.
409 * 'C' - modules from drivers/staging are loaded.
410 * 'I' - Working around severe firmware bug.
411 * 'O' - Out-of-tree module has been loaded.
412 * 'E' - Unsigned module has been loaded.
413 * 'L' - A soft lockup has previously occurred.
414 * 'K' - Kernel has been live patched.
416 * The string is overwritten by the next call to print_tainted().
418 const char *print_tainted(void)
420 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
426 s = buf + sprintf(buf, "Tainted: ");
427 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
428 const struct taint_flag *t = &taint_flags[i];
429 *s++ = test_bit(i, &tainted_mask) ?
430 t->c_true : t->c_false;
434 snprintf(buf, sizeof(buf), "Not tainted");
439 int test_taint(unsigned flag)
441 return test_bit(flag, &tainted_mask);
443 EXPORT_SYMBOL(test_taint);
445 unsigned long get_taint(void)
451 * add_taint: add a taint flag if not already set.
452 * @flag: one of the TAINT_* constants.
453 * @lockdep_ok: whether lock debugging is still OK.
455 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
456 * some notewortht-but-not-corrupting cases, it can be set to true.
458 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
460 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
461 pr_warn("Disabling lock debugging due to kernel taint\n");
463 set_bit(flag, &tainted_mask);
465 EXPORT_SYMBOL(add_taint);
467 static void spin_msec(int msecs)
471 for (i = 0; i < msecs; i++) {
472 touch_nmi_watchdog();
478 * It just happens that oops_enter() and oops_exit() are identically
481 static void do_oops_enter_exit(void)
484 static int spin_counter;
489 spin_lock_irqsave(&pause_on_oops_lock, flags);
490 if (pause_on_oops_flag == 0) {
491 /* This CPU may now print the oops message */
492 pause_on_oops_flag = 1;
494 /* We need to stall this CPU */
496 /* This CPU gets to do the counting */
497 spin_counter = pause_on_oops;
499 spin_unlock(&pause_on_oops_lock);
500 spin_msec(MSEC_PER_SEC);
501 spin_lock(&pause_on_oops_lock);
502 } while (--spin_counter);
503 pause_on_oops_flag = 0;
505 /* This CPU waits for a different one */
506 while (spin_counter) {
507 spin_unlock(&pause_on_oops_lock);
509 spin_lock(&pause_on_oops_lock);
513 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
517 * Return true if the calling CPU is allowed to print oops-related info.
518 * This is a bit racy..
520 int oops_may_print(void)
522 return pause_on_oops_flag == 0;
526 * Called when the architecture enters its oops handler, before it prints
527 * anything. If this is the first CPU to oops, and it's oopsing the first
528 * time then let it proceed.
530 * This is all enabled by the pause_on_oops kernel boot option. We do all
531 * this to ensure that oopses don't scroll off the screen. It has the
532 * side-effect of preventing later-oopsing CPUs from mucking up the display,
535 * It turns out that the CPU which is allowed to print ends up pausing for
536 * the right duration, whereas all the other CPUs pause for twice as long:
537 * once in oops_enter(), once in oops_exit().
539 void oops_enter(void)
542 /* can't trust the integrity of the kernel anymore: */
544 do_oops_enter_exit();
548 * 64-bit random ID for oopses:
552 static int init_oops_id(void)
555 get_random_bytes(&oops_id, sizeof(oops_id));
561 late_initcall(init_oops_id);
563 void print_oops_end_marker(void)
566 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
570 * Called when the architecture exits its oops handler, after printing
575 do_oops_enter_exit();
576 print_oops_end_marker();
577 kmsg_dump(KMSG_DUMP_OOPS);
585 void __warn(const char *file, int line, void *caller, unsigned taint,
586 struct pt_regs *regs, struct warn_args *args)
588 disable_trace_on_warning();
590 pr_warn("------------[ cut here ]------------\n");
593 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
594 raw_smp_processor_id(), current->pid, file, line,
597 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
598 raw_smp_processor_id(), current->pid, caller);
601 vprintk(args->fmt, args->args);
603 check_panic_on_warn("kernel");
612 print_oops_end_marker();
614 /* Just a warning, don't kill lockdep. */
615 add_taint(taint, LOCKDEP_STILL_OK);
618 #ifdef WANT_WARN_ON_SLOWPATH
619 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
621 struct warn_args args;
624 va_start(args.args, fmt);
625 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
629 EXPORT_SYMBOL(warn_slowpath_fmt);
631 void warn_slowpath_fmt_taint(const char *file, int line,
632 unsigned taint, const char *fmt, ...)
634 struct warn_args args;
637 va_start(args.args, fmt);
638 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
641 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
643 void warn_slowpath_null(const char *file, int line)
645 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
647 EXPORT_SYMBOL(warn_slowpath_null);
650 #ifdef CONFIG_CC_STACKPROTECTOR
653 * Called when gcc's -fstack-protector feature is used, and
654 * gcc detects corruption of the on-stack canary value
656 __visible void __stack_chk_fail(void)
658 panic("stack-protector: Kernel stack is corrupted in: %p\n",
659 __builtin_return_address(0));
661 EXPORT_SYMBOL(__stack_chk_fail);
665 #ifdef CONFIG_ARCH_HAS_REFCOUNT
666 void refcount_error_report(struct pt_regs *regs, const char *err)
668 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
669 err, (void *)instruction_pointer(regs),
670 current->comm, task_pid_nr(current),
671 from_kuid_munged(&init_user_ns, current_uid()),
672 from_kuid_munged(&init_user_ns, current_euid()));
676 core_param(panic, panic_timeout, int, 0644);
677 core_param(pause_on_oops, pause_on_oops, int, 0644);
678 core_param(panic_on_warn, panic_on_warn, int, 0644);
679 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
681 static int __init oops_setup(char *s)
685 if (!strcmp(s, "panic"))
689 early_param("oops", oops_setup);