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/module.h>
18 #include <linux/random.h>
19 #include <linux/ftrace.h>
20 #include <linux/reboot.h>
21 #include <linux/delay.h>
22 #include <linux/kexec.h>
23 #include <linux/sched.h>
24 #include <linux/sysrq.h>
25 #include <linux/init.h>
26 #include <linux/nmi.h>
27 #include <linux/console.h>
28 #include <linux/bug.h>
29 #include <linux/ratelimit.h>
31 #define PANIC_TIMER_STEP 100
32 #define PANIC_BLINK_SPD 18
34 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
35 static unsigned long tainted_mask;
36 static int pause_on_oops;
37 static int pause_on_oops_flag;
38 static DEFINE_SPINLOCK(pause_on_oops_lock);
39 bool crash_kexec_post_notifiers;
40 int panic_on_warn __read_mostly;
42 int panic_timeout = CONFIG_PANIC_TIMEOUT;
43 EXPORT_SYMBOL_GPL(panic_timeout);
45 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
47 EXPORT_SYMBOL(panic_notifier_list);
49 static long no_blink(int state)
54 /* Returns how long it waited in ms */
55 long (*panic_blink)(int state);
56 EXPORT_SYMBOL(panic_blink);
59 * Stop ourself in panic -- architecture code may override this
61 void __weak panic_smp_self_stop(void)
68 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
69 * may override this to prepare for crash dumping, e.g. save regs info.
71 void __weak nmi_panic_self_stop(struct pt_regs *regs)
73 panic_smp_self_stop();
77 * Stop other CPUs in panic. Architecture dependent code may override this
78 * with more suitable version. For example, if the architecture supports
79 * crash dump, it should save registers of each stopped CPU and disable
80 * per-CPU features such as virtualization extensions.
82 void __weak crash_smp_send_stop(void)
84 static int cpus_stopped;
87 * This function can be called twice in panic path, but obviously
88 * we execute this only once.
94 * Note smp_send_stop is the usual smp shutdown function, which
95 * unfortunately means it may not be hardened to work in a panic
102 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
105 * A variant of panic() called from NMI context. We return if we've already
106 * panicked on this CPU. If another CPU already panicked, loop in
107 * nmi_panic_self_stop() which can provide architecture dependent code such
108 * as saving register state for crash dump.
110 void nmi_panic(struct pt_regs *regs, const char *msg)
114 cpu = raw_smp_processor_id();
115 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
117 if (old_cpu == PANIC_CPU_INVALID)
119 else if (old_cpu != cpu)
120 nmi_panic_self_stop(regs);
122 EXPORT_SYMBOL(nmi_panic);
125 * panic - halt the system
126 * @fmt: The text string to print
128 * Display a message, then perform cleanups.
130 * This function never returns.
132 void panic(const char *fmt, ...)
134 static char buf[1024];
138 int old_cpu, this_cpu;
139 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
142 * Disable local interrupts. This will prevent panic_smp_self_stop
143 * from deadlocking the first cpu that invokes the panic, since
144 * there is nothing to prevent an interrupt handler (that runs
145 * after setting panic_cpu) from invoking panic() again.
150 * It's possible to come here directly from a panic-assertion and
151 * not have preempt disabled. Some functions called from here want
152 * preempt to be disabled. No point enabling it later though...
154 * Only one CPU is allowed to execute the panic code from here. For
155 * multiple parallel invocations of panic, all other CPUs either
156 * stop themself or will wait until they are stopped by the 1st CPU
157 * with smp_send_stop().
159 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
160 * comes here, so go ahead.
161 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
162 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
164 this_cpu = raw_smp_processor_id();
165 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
167 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
168 panic_smp_self_stop();
173 vsnprintf(buf, sizeof(buf), fmt, args);
175 pr_emerg("Kernel panic - not syncing: %s\n", buf);
176 #ifdef CONFIG_DEBUG_BUGVERBOSE
178 * Avoid nested stack-dumping if a panic occurs during oops processing
180 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
185 * If we have crashed and we have a crash kernel loaded let it handle
187 * If we want to run this after calling panic_notifiers, pass
188 * the "crash_kexec_post_notifiers" option to the kernel.
190 * Bypass the panic_cpu check and call __crash_kexec directly.
192 if (!_crash_kexec_post_notifiers) {
193 printk_safe_flush_on_panic();
197 * Note smp_send_stop is the usual smp shutdown function, which
198 * unfortunately means it may not be hardened to work in a
204 * If we want to do crash dump after notifier calls and
205 * kmsg_dump, we will need architecture dependent extra
206 * works in addition to stopping other CPUs.
208 crash_smp_send_stop();
212 * Run any panic handlers, including those that might need to
213 * add information to the kmsg dump output.
215 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
217 /* Call flush even twice. It tries harder with a single online CPU */
218 printk_safe_flush_on_panic();
219 kmsg_dump(KMSG_DUMP_PANIC);
222 * If you doubt kdump always works fine in any situation,
223 * "crash_kexec_post_notifiers" offers you a chance to run
224 * panic_notifiers and dumping kmsg before kdump.
225 * Note: since some panic_notifiers can make crashed kernel
226 * more unstable, it can increase risks of the kdump failure too.
228 * Bypass the panic_cpu check and call __crash_kexec directly.
230 if (_crash_kexec_post_notifiers)
236 * We may have ended up stopping the CPU holding the lock (in
237 * smp_send_stop()) while still having some valuable data in the console
238 * buffer. Try to acquire the lock then release it regardless of the
239 * result. The release will also print the buffers out. Locks debug
240 * should be disabled to avoid reporting bad unlock balance when
241 * panic() is not being callled from OOPS.
244 console_flush_on_panic();
247 panic_blink = no_blink;
249 if (panic_timeout > 0) {
251 * Delay timeout seconds before rebooting the machine.
252 * We can't use the "normal" timers since we just panicked.
254 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
256 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
257 touch_nmi_watchdog();
259 i += panic_blink(state ^= 1);
260 i_next = i + 3600 / PANIC_BLINK_SPD;
262 mdelay(PANIC_TIMER_STEP);
265 if (panic_timeout != 0) {
267 * This will not be a clean reboot, with everything
268 * shutting down. But if there is a chance of
269 * rebooting the system it will be rebooted.
275 extern int stop_a_enabled;
276 /* Make sure the user can actually press Stop-A (L1-A) */
278 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
279 "twice on console to return to the boot prom\n");
282 #if defined(CONFIG_S390)
284 unsigned long caller;
286 caller = (unsigned long)__builtin_return_address(0);
287 disabled_wait(caller);
290 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
292 for (i = 0; ; i += PANIC_TIMER_STEP) {
293 touch_softlockup_watchdog();
295 i += panic_blink(state ^= 1);
296 i_next = i + 3600 / PANIC_BLINK_SPD;
298 mdelay(PANIC_TIMER_STEP);
302 EXPORT_SYMBOL(panic);
305 * TAINT_FORCED_RMMOD could be a per-module flag but the module
306 * is being removed anyway.
308 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
309 { 'P', 'G', true }, /* TAINT_PROPRIETARY_MODULE */
310 { 'F', ' ', true }, /* TAINT_FORCED_MODULE */
311 { 'S', ' ', false }, /* TAINT_CPU_OUT_OF_SPEC */
312 { 'R', ' ', false }, /* TAINT_FORCED_RMMOD */
313 { 'M', ' ', false }, /* TAINT_MACHINE_CHECK */
314 { 'B', ' ', false }, /* TAINT_BAD_PAGE */
315 { 'U', ' ', false }, /* TAINT_USER */
316 { 'D', ' ', false }, /* TAINT_DIE */
317 { 'A', ' ', false }, /* TAINT_OVERRIDDEN_ACPI_TABLE */
318 { 'W', ' ', false }, /* TAINT_WARN */
319 { 'C', ' ', true }, /* TAINT_CRAP */
320 { 'I', ' ', false }, /* TAINT_FIRMWARE_WORKAROUND */
321 { 'O', ' ', true }, /* TAINT_OOT_MODULE */
322 { 'E', ' ', true }, /* TAINT_UNSIGNED_MODULE */
323 { 'L', ' ', false }, /* TAINT_SOFTLOCKUP */
324 { 'K', ' ', true }, /* TAINT_LIVEPATCH */
328 * print_tainted - return a string to represent the kernel taint state.
330 * 'P' - Proprietary module has been loaded.
331 * 'F' - Module has been forcibly loaded.
332 * 'S' - SMP with CPUs not designed for SMP.
333 * 'R' - User forced a module unload.
334 * 'M' - System experienced a machine check exception.
335 * 'B' - System has hit bad_page.
336 * 'U' - Userspace-defined naughtiness.
337 * 'D' - Kernel has oopsed before
338 * 'A' - ACPI table overridden.
339 * 'W' - Taint on warning.
340 * 'C' - modules from drivers/staging are loaded.
341 * 'I' - Working around severe firmware bug.
342 * 'O' - Out-of-tree module has been loaded.
343 * 'E' - Unsigned module has been loaded.
344 * 'L' - A soft lockup has previously occurred.
345 * 'K' - Kernel has been live patched.
347 * The string is overwritten by the next call to print_tainted().
349 const char *print_tainted(void)
351 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
357 s = buf + sprintf(buf, "Tainted: ");
358 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
359 const struct taint_flag *t = &taint_flags[i];
360 *s++ = test_bit(i, &tainted_mask) ?
361 t->c_true : t->c_false;
365 snprintf(buf, sizeof(buf), "Not tainted");
370 int test_taint(unsigned flag)
372 return test_bit(flag, &tainted_mask);
374 EXPORT_SYMBOL(test_taint);
376 unsigned long get_taint(void)
382 * add_taint: add a taint flag if not already set.
383 * @flag: one of the TAINT_* constants.
384 * @lockdep_ok: whether lock debugging is still OK.
386 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
387 * some notewortht-but-not-corrupting cases, it can be set to true.
389 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
391 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
392 pr_warn("Disabling lock debugging due to kernel taint\n");
394 set_bit(flag, &tainted_mask);
396 EXPORT_SYMBOL(add_taint);
398 static void spin_msec(int msecs)
402 for (i = 0; i < msecs; i++) {
403 touch_nmi_watchdog();
409 * It just happens that oops_enter() and oops_exit() are identically
412 static void do_oops_enter_exit(void)
415 static int spin_counter;
420 spin_lock_irqsave(&pause_on_oops_lock, flags);
421 if (pause_on_oops_flag == 0) {
422 /* This CPU may now print the oops message */
423 pause_on_oops_flag = 1;
425 /* We need to stall this CPU */
427 /* This CPU gets to do the counting */
428 spin_counter = pause_on_oops;
430 spin_unlock(&pause_on_oops_lock);
431 spin_msec(MSEC_PER_SEC);
432 spin_lock(&pause_on_oops_lock);
433 } while (--spin_counter);
434 pause_on_oops_flag = 0;
436 /* This CPU waits for a different one */
437 while (spin_counter) {
438 spin_unlock(&pause_on_oops_lock);
440 spin_lock(&pause_on_oops_lock);
444 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
448 * Return true if the calling CPU is allowed to print oops-related info.
449 * This is a bit racy..
451 int oops_may_print(void)
453 return pause_on_oops_flag == 0;
457 * Called when the architecture enters its oops handler, before it prints
458 * anything. If this is the first CPU to oops, and it's oopsing the first
459 * time then let it proceed.
461 * This is all enabled by the pause_on_oops kernel boot option. We do all
462 * this to ensure that oopses don't scroll off the screen. It has the
463 * side-effect of preventing later-oopsing CPUs from mucking up the display,
466 * It turns out that the CPU which is allowed to print ends up pausing for
467 * the right duration, whereas all the other CPUs pause for twice as long:
468 * once in oops_enter(), once in oops_exit().
470 void oops_enter(void)
473 /* can't trust the integrity of the kernel anymore: */
475 do_oops_enter_exit();
479 * 64-bit random ID for oopses:
483 static int init_oops_id(void)
486 get_random_bytes(&oops_id, sizeof(oops_id));
492 late_initcall(init_oops_id);
494 void print_oops_end_marker(void)
497 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
501 * Called when the architecture exits its oops handler, after printing
506 do_oops_enter_exit();
507 print_oops_end_marker();
508 kmsg_dump(KMSG_DUMP_OOPS);
516 void __warn(const char *file, int line, void *caller, unsigned taint,
517 struct pt_regs *regs, struct warn_args *args)
519 disable_trace_on_warning();
521 pr_warn("------------[ cut here ]------------\n");
524 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
525 raw_smp_processor_id(), current->pid, file, line,
528 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
529 raw_smp_processor_id(), current->pid, caller);
532 vprintk(args->fmt, args->args);
536 * This thread may hit another WARN() in the panic path.
537 * Resetting this prevents additional WARN() from panicking the
538 * system on this thread. Other threads are blocked by the
539 * panic_mutex in panic().
542 panic("panic_on_warn set ...\n");
552 print_oops_end_marker();
554 /* Just a warning, don't kill lockdep. */
555 add_taint(taint, LOCKDEP_STILL_OK);
558 #ifdef WANT_WARN_ON_SLOWPATH
559 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
561 struct warn_args args;
564 va_start(args.args, fmt);
565 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
569 EXPORT_SYMBOL(warn_slowpath_fmt);
571 void warn_slowpath_fmt_taint(const char *file, int line,
572 unsigned taint, const char *fmt, ...)
574 struct warn_args args;
577 va_start(args.args, fmt);
578 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
581 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
583 void warn_slowpath_null(const char *file, int line)
585 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
587 EXPORT_SYMBOL(warn_slowpath_null);
590 #ifdef CONFIG_CC_STACKPROTECTOR
593 * Called when gcc's -fstack-protector feature is used, and
594 * gcc detects corruption of the on-stack canary value
596 __visible void __stack_chk_fail(void)
598 panic("stack-protector: Kernel stack is corrupted in: %p\n",
599 __builtin_return_address(0));
601 EXPORT_SYMBOL(__stack_chk_fail);
605 #ifdef CONFIG_ARCH_HAS_REFCOUNT
606 void refcount_error_report(struct pt_regs *regs, const char *err)
608 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
609 err, (void *)instruction_pointer(regs),
610 current->comm, task_pid_nr(current),
611 from_kuid_munged(&init_user_ns, current_uid()),
612 from_kuid_munged(&init_user_ns, current_euid()));
616 core_param(panic, panic_timeout, int, 0644);
617 core_param(pause_on_oops, pause_on_oops, int, 0644);
618 core_param(panic_on_warn, panic_on_warn, int, 0644);
619 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
621 static int __init oops_setup(char *s)
625 if (!strcmp(s, "panic"))
629 early_param("oops", oops_setup);