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/sched.h>
27 #include <linux/sysrq.h>
28 #include <linux/init.h>
29 #include <linux/nmi.h>
30 #include <linux/console.h>
31 #include <linux/bug.h>
32 #include <linux/ratelimit.h>
33 #include <linux/debugfs.h>
34 #include <asm/sections.h>
36 #define PANIC_TIMER_STEP 100
37 #define PANIC_BLINK_SPD 18
39 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
40 static unsigned long tainted_mask =
41 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
42 static int pause_on_oops;
43 static int pause_on_oops_flag;
44 static DEFINE_SPINLOCK(pause_on_oops_lock);
45 bool crash_kexec_post_notifiers;
46 int panic_on_warn __read_mostly;
48 int panic_timeout = CONFIG_PANIC_TIMEOUT;
49 EXPORT_SYMBOL_GPL(panic_timeout);
51 #define PANIC_PRINT_TASK_INFO 0x00000001
52 #define PANIC_PRINT_MEM_INFO 0x00000002
53 #define PANIC_PRINT_TIMER_INFO 0x00000004
54 #define PANIC_PRINT_LOCK_INFO 0x00000008
55 #define PANIC_PRINT_FTRACE_INFO 0x00000010
56 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
57 unsigned long panic_print;
59 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
61 EXPORT_SYMBOL(panic_notifier_list);
63 static long no_blink(int state)
68 /* Returns how long it waited in ms */
69 long (*panic_blink)(int state);
70 EXPORT_SYMBOL(panic_blink);
73 * Stop ourself in panic -- architecture code may override this
75 void __weak panic_smp_self_stop(void)
82 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
83 * may override this to prepare for crash dumping, e.g. save regs info.
85 void __weak nmi_panic_self_stop(struct pt_regs *regs)
87 panic_smp_self_stop();
91 * Stop other CPUs in panic. Architecture dependent code may override this
92 * with more suitable version. For example, if the architecture supports
93 * crash dump, it should save registers of each stopped CPU and disable
94 * per-CPU features such as virtualization extensions.
96 void __weak crash_smp_send_stop(void)
98 static int cpus_stopped;
101 * This function can be called twice in panic path, but obviously
102 * we execute this only once.
108 * Note smp_send_stop is the usual smp shutdown function, which
109 * unfortunately means it may not be hardened to work in a panic
116 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
119 * A variant of panic() called from NMI context. We return if we've already
120 * panicked on this CPU. If another CPU already panicked, loop in
121 * nmi_panic_self_stop() which can provide architecture dependent code such
122 * as saving register state for crash dump.
124 void nmi_panic(struct pt_regs *regs, const char *msg)
128 cpu = raw_smp_processor_id();
129 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
131 if (old_cpu == PANIC_CPU_INVALID)
133 else if (old_cpu != cpu)
134 nmi_panic_self_stop(regs);
136 EXPORT_SYMBOL(nmi_panic);
138 static void panic_print_sys_info(void)
140 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
141 console_flush_on_panic(CONSOLE_REPLAY_ALL);
143 if (panic_print & PANIC_PRINT_TASK_INFO)
146 if (panic_print & PANIC_PRINT_MEM_INFO)
149 if (panic_print & PANIC_PRINT_TIMER_INFO)
150 sysrq_timer_list_show();
152 if (panic_print & PANIC_PRINT_LOCK_INFO)
153 debug_show_all_locks();
155 if (panic_print & PANIC_PRINT_FTRACE_INFO)
156 ftrace_dump(DUMP_ALL);
160 * panic - halt the system
161 * @fmt: The text string to print
163 * Display a message, then perform cleanups.
165 * This function never returns.
167 void panic(const char *fmt, ...)
169 static char buf[1024];
171 long i, i_next = 0, len;
173 int old_cpu, this_cpu;
174 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
177 * Disable local interrupts. This will prevent panic_smp_self_stop
178 * from deadlocking the first cpu that invokes the panic, since
179 * there is nothing to prevent an interrupt handler (that runs
180 * after setting panic_cpu) from invoking panic() again.
185 * It's possible to come here directly from a panic-assertion and
186 * not have preempt disabled. Some functions called from here want
187 * preempt to be disabled. No point enabling it later though...
189 * Only one CPU is allowed to execute the panic code from here. For
190 * multiple parallel invocations of panic, all other CPUs either
191 * stop themself or will wait until they are stopped by the 1st CPU
192 * with smp_send_stop().
194 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
195 * comes here, so go ahead.
196 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
197 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
199 this_cpu = raw_smp_processor_id();
200 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
202 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
203 panic_smp_self_stop();
208 len = vscnprintf(buf, sizeof(buf), fmt, args);
211 if (len && buf[len - 1] == '\n')
214 pr_emerg("Kernel panic - not syncing: %s\n", buf);
215 #ifdef CONFIG_DEBUG_BUGVERBOSE
217 * Avoid nested stack-dumping if a panic occurs during oops processing
219 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
224 * If kgdb is enabled, give it a chance to run before we stop all
225 * the other CPUs or else we won't be able to debug processes left
231 * If we have crashed and we have a crash kernel loaded let it handle
233 * If we want to run this after calling panic_notifiers, pass
234 * the "crash_kexec_post_notifiers" option to the kernel.
236 * Bypass the panic_cpu check and call __crash_kexec directly.
238 if (!_crash_kexec_post_notifiers) {
239 printk_safe_flush_on_panic();
243 * Note smp_send_stop is the usual smp shutdown function, which
244 * unfortunately means it may not be hardened to work in a
250 * If we want to do crash dump after notifier calls and
251 * kmsg_dump, we will need architecture dependent extra
252 * works in addition to stopping other CPUs.
254 crash_smp_send_stop();
258 * Run any panic handlers, including those that might need to
259 * add information to the kmsg dump output.
261 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
263 /* Call flush even twice. It tries harder with a single online CPU */
264 printk_safe_flush_on_panic();
265 kmsg_dump(KMSG_DUMP_PANIC);
268 * If you doubt kdump always works fine in any situation,
269 * "crash_kexec_post_notifiers" offers you a chance to run
270 * panic_notifiers and dumping kmsg before kdump.
271 * Note: since some panic_notifiers can make crashed kernel
272 * more unstable, it can increase risks of the kdump failure too.
274 * Bypass the panic_cpu check and call __crash_kexec directly.
276 if (_crash_kexec_post_notifiers)
285 * We may have ended up stopping the CPU holding the lock (in
286 * smp_send_stop()) while still having some valuable data in the console
287 * buffer. Try to acquire the lock then release it regardless of the
288 * result. The release will also print the buffers out. Locks debug
289 * should be disabled to avoid reporting bad unlock balance when
290 * panic() is not being callled from OOPS.
293 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
295 panic_print_sys_info();
298 panic_blink = no_blink;
300 if (panic_timeout > 0) {
302 * Delay timeout seconds before rebooting the machine.
303 * We can't use the "normal" timers since we just panicked.
305 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
307 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
308 touch_nmi_watchdog();
310 i += panic_blink(state ^= 1);
311 i_next = i + 3600 / PANIC_BLINK_SPD;
313 mdelay(PANIC_TIMER_STEP);
316 if (panic_timeout != 0) {
318 * This will not be a clean reboot, with everything
319 * shutting down. But if there is a chance of
320 * rebooting the system it will be rebooted.
322 if (panic_reboot_mode != REBOOT_UNDEFINED)
323 reboot_mode = panic_reboot_mode;
328 extern int stop_a_enabled;
329 /* Make sure the user can actually press Stop-A (L1-A) */
331 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
332 "twice on console to return to the boot prom\n");
335 #if defined(CONFIG_S390)
338 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
340 /* Do not scroll important messages printed above */
343 for (i = 0; ; i += PANIC_TIMER_STEP) {
344 touch_softlockup_watchdog();
346 i += panic_blink(state ^= 1);
347 i_next = i + 3600 / PANIC_BLINK_SPD;
349 mdelay(PANIC_TIMER_STEP);
353 EXPORT_SYMBOL(panic);
356 * TAINT_FORCED_RMMOD could be a per-module flag but the module
357 * is being removed anyway.
359 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
360 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
361 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
362 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
363 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
364 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
365 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
366 [ TAINT_USER ] = { 'U', ' ', false },
367 [ TAINT_DIE ] = { 'D', ' ', false },
368 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
369 [ TAINT_WARN ] = { 'W', ' ', false },
370 [ TAINT_CRAP ] = { 'C', ' ', true },
371 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
372 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
373 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
374 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
375 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
376 [ TAINT_AUX ] = { 'X', ' ', true },
377 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
381 * print_tainted - return a string to represent the kernel taint state.
383 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
385 * The string is overwritten by the next call to print_tainted(),
386 * but is always NULL terminated.
388 const char *print_tainted(void)
390 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
392 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
398 s = buf + sprintf(buf, "Tainted: ");
399 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
400 const struct taint_flag *t = &taint_flags[i];
401 *s++ = test_bit(i, &tainted_mask) ?
402 t->c_true : t->c_false;
406 snprintf(buf, sizeof(buf), "Not tainted");
411 int test_taint(unsigned flag)
413 return test_bit(flag, &tainted_mask);
415 EXPORT_SYMBOL(test_taint);
417 unsigned long get_taint(void)
423 * add_taint: add a taint flag if not already set.
424 * @flag: one of the TAINT_* constants.
425 * @lockdep_ok: whether lock debugging is still OK.
427 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
428 * some notewortht-but-not-corrupting cases, it can be set to true.
430 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
432 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
433 pr_warn("Disabling lock debugging due to kernel taint\n");
435 set_bit(flag, &tainted_mask);
437 EXPORT_SYMBOL(add_taint);
439 static void spin_msec(int msecs)
443 for (i = 0; i < msecs; i++) {
444 touch_nmi_watchdog();
450 * It just happens that oops_enter() and oops_exit() are identically
453 static void do_oops_enter_exit(void)
456 static int spin_counter;
461 spin_lock_irqsave(&pause_on_oops_lock, flags);
462 if (pause_on_oops_flag == 0) {
463 /* This CPU may now print the oops message */
464 pause_on_oops_flag = 1;
466 /* We need to stall this CPU */
468 /* This CPU gets to do the counting */
469 spin_counter = pause_on_oops;
471 spin_unlock(&pause_on_oops_lock);
472 spin_msec(MSEC_PER_SEC);
473 spin_lock(&pause_on_oops_lock);
474 } while (--spin_counter);
475 pause_on_oops_flag = 0;
477 /* This CPU waits for a different one */
478 while (spin_counter) {
479 spin_unlock(&pause_on_oops_lock);
481 spin_lock(&pause_on_oops_lock);
485 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
489 * Return true if the calling CPU is allowed to print oops-related info.
490 * This is a bit racy..
492 int oops_may_print(void)
494 return pause_on_oops_flag == 0;
498 * Called when the architecture enters its oops handler, before it prints
499 * anything. If this is the first CPU to oops, and it's oopsing the first
500 * time then let it proceed.
502 * This is all enabled by the pause_on_oops kernel boot option. We do all
503 * this to ensure that oopses don't scroll off the screen. It has the
504 * side-effect of preventing later-oopsing CPUs from mucking up the display,
507 * It turns out that the CPU which is allowed to print ends up pausing for
508 * the right duration, whereas all the other CPUs pause for twice as long:
509 * once in oops_enter(), once in oops_exit().
511 void oops_enter(void)
514 /* can't trust the integrity of the kernel anymore: */
516 do_oops_enter_exit();
520 * 64-bit random ID for oopses:
524 static int init_oops_id(void)
527 get_random_bytes(&oops_id, sizeof(oops_id));
533 late_initcall(init_oops_id);
535 void print_oops_end_marker(void)
538 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
542 * Called when the architecture exits its oops handler, after printing
547 do_oops_enter_exit();
548 print_oops_end_marker();
549 kmsg_dump(KMSG_DUMP_OOPS);
557 void __warn(const char *file, int line, void *caller, unsigned taint,
558 struct pt_regs *regs, struct warn_args *args)
560 disable_trace_on_warning();
563 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
564 raw_smp_processor_id(), current->pid, file, line,
567 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
568 raw_smp_processor_id(), current->pid, caller);
571 vprintk(args->fmt, args->args);
575 * This thread may hit another WARN() in the panic path.
576 * Resetting this prevents additional WARN() from panicking the
577 * system on this thread. Other threads are blocked by the
578 * panic_mutex in panic().
581 panic("panic_on_warn set ...\n");
591 print_irqtrace_events(current);
593 print_oops_end_marker();
595 /* Just a warning, don't kill lockdep. */
596 add_taint(taint, LOCKDEP_STILL_OK);
600 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
601 const char *fmt, ...)
603 struct warn_args args;
608 __warn(file, line, __builtin_return_address(0), taint,
614 va_start(args.args, fmt);
615 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
618 EXPORT_SYMBOL(warn_slowpath_fmt);
620 void __warn_printk(const char *fmt, ...)
630 EXPORT_SYMBOL(__warn_printk);
635 /* Support resetting WARN*_ONCE state */
637 static int clear_warn_once_set(void *data, u64 val)
639 generic_bug_clear_once();
640 memset(__start_once, 0, __end_once - __start_once);
644 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
647 static __init int register_warn_debugfs(void)
649 /* Don't care about failure */
650 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
651 &clear_warn_once_fops);
655 device_initcall(register_warn_debugfs);
658 #ifdef CONFIG_STACKPROTECTOR
661 * Called when gcc's -fstack-protector feature is used, and
662 * gcc detects corruption of the on-stack canary value
664 __visible void __stack_chk_fail(void)
666 panic("stack-protector: Kernel stack is corrupted in: %pB",
667 __builtin_return_address(0));
669 EXPORT_SYMBOL(__stack_chk_fail);
673 #ifdef CONFIG_ARCH_HAS_REFCOUNT
674 void refcount_error_report(struct pt_regs *regs, const char *err)
676 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
677 err, (void *)instruction_pointer(regs),
678 current->comm, task_pid_nr(current),
679 from_kuid_munged(&init_user_ns, current_uid()),
680 from_kuid_munged(&init_user_ns, current_euid()));
684 core_param(panic, panic_timeout, int, 0644);
685 core_param(panic_print, panic_print, ulong, 0644);
686 core_param(pause_on_oops, pause_on_oops, int, 0644);
687 core_param(panic_on_warn, panic_on_warn, int, 0644);
688 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
690 static int __init oops_setup(char *s)
694 if (!strcmp(s, "panic"))
698 early_param("oops", oops_setup);