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 #define PANIC_PRINT_BLOCKED_TASKS 0x00000080
77 unsigned long panic_print
;
79 ATOMIC_NOTIFIER_HEAD(panic_notifier_list
);
81 EXPORT_SYMBOL(panic_notifier_list
);
84 static struct ctl_table kern_panic_table
[] = {
87 .procname
= "oops_all_cpu_backtrace",
88 .data
= &sysctl_oops_all_cpu_backtrace
,
89 .maxlen
= sizeof(int),
91 .proc_handler
= proc_dointvec_minmax
,
92 .extra1
= SYSCTL_ZERO
,
97 .procname
= "warn_limit",
99 .maxlen
= sizeof(warn_limit
),
101 .proc_handler
= proc_douintvec
,
106 static __init
int kernel_panic_sysctls_init(void)
108 register_sysctl_init("kernel", kern_panic_table
);
111 late_initcall(kernel_panic_sysctls_init
);
114 static atomic_t warn_count
= ATOMIC_INIT(0);
117 static ssize_t
warn_count_show(struct kobject
*kobj
, struct kobj_attribute
*attr
,
120 return sysfs_emit(page
, "%d\n", atomic_read(&warn_count
));
123 static struct kobj_attribute warn_count_attr
= __ATTR_RO(warn_count
);
125 static __init
int kernel_panic_sysfs_init(void)
127 sysfs_add_file_to_group(kernel_kobj
, &warn_count_attr
.attr
, NULL
);
130 late_initcall(kernel_panic_sysfs_init
);
133 static long no_blink(int state
)
138 /* Returns how long it waited in ms */
139 long (*panic_blink
)(int state
);
140 EXPORT_SYMBOL(panic_blink
);
143 * Stop ourself in panic -- architecture code may override this
145 void __weak __noreturn
panic_smp_self_stop(void)
152 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
153 * may override this to prepare for crash dumping, e.g. save regs info.
155 void __weak __noreturn
nmi_panic_self_stop(struct pt_regs
*regs
)
157 panic_smp_self_stop();
161 * Stop other CPUs in panic. Architecture dependent code may override this
162 * with more suitable version. For example, if the architecture supports
163 * crash dump, it should save registers of each stopped CPU and disable
164 * per-CPU features such as virtualization extensions.
166 void __weak
crash_smp_send_stop(void)
168 static int cpus_stopped
;
171 * This function can be called twice in panic path, but obviously
172 * we execute this only once.
178 * Note smp_send_stop is the usual smp shutdown function, which
179 * unfortunately means it may not be hardened to work in a panic
186 atomic_t panic_cpu
= ATOMIC_INIT(PANIC_CPU_INVALID
);
189 * A variant of panic() called from NMI context. We return if we've already
190 * panicked on this CPU. If another CPU already panicked, loop in
191 * nmi_panic_self_stop() which can provide architecture dependent code such
192 * as saving register state for crash dump.
194 void nmi_panic(struct pt_regs
*regs
, const char *msg
)
196 int old_cpu
, this_cpu
;
198 old_cpu
= PANIC_CPU_INVALID
;
199 this_cpu
= raw_smp_processor_id();
201 /* atomic_try_cmpxchg updates old_cpu on failure */
202 if (atomic_try_cmpxchg(&panic_cpu
, &old_cpu
, this_cpu
))
204 else if (old_cpu
!= this_cpu
)
205 nmi_panic_self_stop(regs
);
207 EXPORT_SYMBOL(nmi_panic
);
209 static void panic_print_sys_info(bool console_flush
)
212 if (panic_print
& PANIC_PRINT_ALL_PRINTK_MSG
)
213 console_flush_on_panic(CONSOLE_REPLAY_ALL
);
217 if (panic_print
& PANIC_PRINT_TASK_INFO
)
220 if (panic_print
& PANIC_PRINT_MEM_INFO
)
223 if (panic_print
& PANIC_PRINT_TIMER_INFO
)
224 sysrq_timer_list_show();
226 if (panic_print
& PANIC_PRINT_LOCK_INFO
)
227 debug_show_all_locks();
229 if (panic_print
& PANIC_PRINT_FTRACE_INFO
)
230 ftrace_dump(DUMP_ALL
);
232 if (panic_print
& PANIC_PRINT_BLOCKED_TASKS
)
233 show_state_filter(TASK_UNINTERRUPTIBLE
);
236 void check_panic_on_warn(const char *origin
)
241 panic("%s: panic_on_warn set ...\n", origin
);
243 limit
= READ_ONCE(warn_limit
);
244 if (atomic_inc_return(&warn_count
) >= limit
&& limit
)
245 panic("%s: system warned too often (kernel.warn_limit is %d)",
250 * Helper that triggers the NMI backtrace (if set in panic_print)
251 * and then performs the secondary CPUs shutdown - we cannot have
252 * the NMI backtrace after the CPUs are off!
254 static void panic_other_cpus_shutdown(bool crash_kexec
)
256 if (panic_print
& PANIC_PRINT_ALL_CPU_BT
)
257 trigger_all_cpu_backtrace();
260 * Note that smp_send_stop() is the usual SMP shutdown function,
261 * which unfortunately may not be hardened to work in a panic
262 * situation. If we want to do crash dump after notifier calls
263 * and kmsg_dump, we will need architecture dependent extra
264 * bits in addition to stopping other CPUs, hence we rely on
265 * crash_smp_send_stop() for that.
270 crash_smp_send_stop();
274 * panic - halt the system
275 * @fmt: The text string to print
277 * Display a message, then perform cleanups.
279 * This function never returns.
281 void panic(const char *fmt
, ...)
283 static char buf
[1024];
285 long i
, i_next
= 0, len
;
287 int old_cpu
, this_cpu
;
288 bool _crash_kexec_post_notifiers
= crash_kexec_post_notifiers
;
292 * This thread may hit another WARN() in the panic path.
293 * Resetting this prevents additional WARN() from panicking the
294 * system on this thread. Other threads are blocked by the
295 * panic_mutex in panic().
301 * Disable local interrupts. This will prevent panic_smp_self_stop
302 * from deadlocking the first cpu that invokes the panic, since
303 * there is nothing to prevent an interrupt handler (that runs
304 * after setting panic_cpu) from invoking panic() again.
307 preempt_disable_notrace();
310 * It's possible to come here directly from a panic-assertion and
311 * not have preempt disabled. Some functions called from here want
312 * preempt to be disabled. No point enabling it later though...
314 * Only one CPU is allowed to execute the panic code from here. For
315 * multiple parallel invocations of panic, all other CPUs either
316 * stop themself or will wait until they are stopped by the 1st CPU
317 * with smp_send_stop().
319 * cmpxchg success means this is the 1st CPU which comes here,
321 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
322 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
324 old_cpu
= PANIC_CPU_INVALID
;
325 this_cpu
= raw_smp_processor_id();
327 /* atomic_try_cmpxchg updates old_cpu on failure */
328 if (atomic_try_cmpxchg(&panic_cpu
, &old_cpu
, this_cpu
)) {
330 } else if (old_cpu
!= this_cpu
)
331 panic_smp_self_stop();
336 len
= vscnprintf(buf
, sizeof(buf
), fmt
, args
);
339 if (len
&& buf
[len
- 1] == '\n')
342 pr_emerg("Kernel panic - not syncing: %s\n", buf
);
343 #ifdef CONFIG_DEBUG_BUGVERBOSE
345 * Avoid nested stack-dumping if a panic occurs during oops processing
347 if (!test_taint(TAINT_DIE
) && oops_in_progress
<= 1)
352 * If kgdb is enabled, give it a chance to run before we stop all
353 * the other CPUs or else we won't be able to debug processes left
359 * If we have crashed and we have a crash kernel loaded let it handle
361 * If we want to run this after calling panic_notifiers, pass
362 * the "crash_kexec_post_notifiers" option to the kernel.
364 * Bypass the panic_cpu check and call __crash_kexec directly.
366 if (!_crash_kexec_post_notifiers
)
369 panic_other_cpus_shutdown(_crash_kexec_post_notifiers
);
372 * Run any panic handlers, including those that might need to
373 * add information to the kmsg dump output.
375 atomic_notifier_call_chain(&panic_notifier_list
, 0, buf
);
377 panic_print_sys_info(false);
379 kmsg_dump(KMSG_DUMP_PANIC
);
382 * If you doubt kdump always works fine in any situation,
383 * "crash_kexec_post_notifiers" offers you a chance to run
384 * panic_notifiers and dumping kmsg before kdump.
385 * Note: since some panic_notifiers can make crashed kernel
386 * more unstable, it can increase risks of the kdump failure too.
388 * Bypass the panic_cpu check and call __crash_kexec directly.
390 if (_crash_kexec_post_notifiers
)
396 * We may have ended up stopping the CPU holding the lock (in
397 * smp_send_stop()) while still having some valuable data in the console
398 * buffer. Try to acquire the lock then release it regardless of the
399 * result. The release will also print the buffers out. Locks debug
400 * should be disabled to avoid reporting bad unlock balance when
401 * panic() is not being callled from OOPS.
404 console_flush_on_panic(CONSOLE_FLUSH_PENDING
);
406 panic_print_sys_info(true);
409 panic_blink
= no_blink
;
411 if (panic_timeout
> 0) {
413 * Delay timeout seconds before rebooting the machine.
414 * We can't use the "normal" timers since we just panicked.
416 pr_emerg("Rebooting in %d seconds..\n", panic_timeout
);
418 for (i
= 0; i
< panic_timeout
* 1000; i
+= PANIC_TIMER_STEP
) {
419 touch_nmi_watchdog();
421 i
+= panic_blink(state
^= 1);
422 i_next
= i
+ 3600 / PANIC_BLINK_SPD
;
424 mdelay(PANIC_TIMER_STEP
);
427 if (panic_timeout
!= 0) {
429 * This will not be a clean reboot, with everything
430 * shutting down. But if there is a chance of
431 * rebooting the system it will be rebooted.
433 if (panic_reboot_mode
!= REBOOT_UNDEFINED
)
434 reboot_mode
= panic_reboot_mode
;
439 extern int stop_a_enabled
;
440 /* Make sure the user can actually press Stop-A (L1-A) */
442 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
443 "twice on console to return to the boot prom\n");
446 #if defined(CONFIG_S390)
449 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf
);
451 /* Do not scroll important messages printed above */
455 * The final messages may not have been printed if in a context that
456 * defers printing (such as NMI) and irq_work is not available.
457 * Explicitly flush the kernel log buffer one last time.
459 console_flush_on_panic(CONSOLE_FLUSH_PENDING
);
462 for (i
= 0; ; i
+= PANIC_TIMER_STEP
) {
463 touch_softlockup_watchdog();
465 i
+= panic_blink(state
^= 1);
466 i_next
= i
+ 3600 / PANIC_BLINK_SPD
;
468 mdelay(PANIC_TIMER_STEP
);
472 EXPORT_SYMBOL(panic
);
475 * TAINT_FORCED_RMMOD could be a per-module flag but the module
476 * is being removed anyway.
478 const struct taint_flag taint_flags
[TAINT_FLAGS_COUNT
] = {
479 [ TAINT_PROPRIETARY_MODULE
] = { 'P', 'G', true },
480 [ TAINT_FORCED_MODULE
] = { 'F', ' ', true },
481 [ TAINT_CPU_OUT_OF_SPEC
] = { 'S', ' ', false },
482 [ TAINT_FORCED_RMMOD
] = { 'R', ' ', false },
483 [ TAINT_MACHINE_CHECK
] = { 'M', ' ', false },
484 [ TAINT_BAD_PAGE
] = { 'B', ' ', false },
485 [ TAINT_USER
] = { 'U', ' ', false },
486 [ TAINT_DIE
] = { 'D', ' ', false },
487 [ TAINT_OVERRIDDEN_ACPI_TABLE
] = { 'A', ' ', false },
488 [ TAINT_WARN
] = { 'W', ' ', false },
489 [ TAINT_CRAP
] = { 'C', ' ', true },
490 [ TAINT_FIRMWARE_WORKAROUND
] = { 'I', ' ', false },
491 [ TAINT_OOT_MODULE
] = { 'O', ' ', true },
492 [ TAINT_UNSIGNED_MODULE
] = { 'E', ' ', true },
493 [ TAINT_SOFTLOCKUP
] = { 'L', ' ', false },
494 [ TAINT_LIVEPATCH
] = { 'K', ' ', true },
495 [ TAINT_AUX
] = { 'X', ' ', true },
496 [ TAINT_RANDSTRUCT
] = { 'T', ' ', true },
497 [ TAINT_TEST
] = { 'N', ' ', true },
501 * print_tainted - return a string to represent the kernel taint state.
503 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
505 * The string is overwritten by the next call to print_tainted(),
506 * but is always NULL terminated.
508 const char *print_tainted(void)
510 static char buf
[TAINT_FLAGS_COUNT
+ sizeof("Tainted: ")];
512 BUILD_BUG_ON(ARRAY_SIZE(taint_flags
) != TAINT_FLAGS_COUNT
);
518 s
= buf
+ sprintf(buf
, "Tainted: ");
519 for (i
= 0; i
< TAINT_FLAGS_COUNT
; i
++) {
520 const struct taint_flag
*t
= &taint_flags
[i
];
521 *s
++ = test_bit(i
, &tainted_mask
) ?
522 t
->c_true
: t
->c_false
;
526 snprintf(buf
, sizeof(buf
), "Not tainted");
531 int test_taint(unsigned flag
)
533 return test_bit(flag
, &tainted_mask
);
535 EXPORT_SYMBOL(test_taint
);
537 unsigned long get_taint(void)
543 * add_taint: add a taint flag if not already set.
544 * @flag: one of the TAINT_* constants.
545 * @lockdep_ok: whether lock debugging is still OK.
547 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
548 * some notewortht-but-not-corrupting cases, it can be set to true.
550 void add_taint(unsigned flag
, enum lockdep_ok lockdep_ok
)
552 if (lockdep_ok
== LOCKDEP_NOW_UNRELIABLE
&& __debug_locks_off())
553 pr_warn("Disabling lock debugging due to kernel taint\n");
555 set_bit(flag
, &tainted_mask
);
557 if (tainted_mask
& panic_on_taint
) {
559 panic("panic_on_taint set ...");
562 EXPORT_SYMBOL(add_taint
);
564 static void spin_msec(int msecs
)
568 for (i
= 0; i
< msecs
; i
++) {
569 touch_nmi_watchdog();
575 * It just happens that oops_enter() and oops_exit() are identically
578 static void do_oops_enter_exit(void)
581 static int spin_counter
;
586 spin_lock_irqsave(&pause_on_oops_lock
, flags
);
587 if (pause_on_oops_flag
== 0) {
588 /* This CPU may now print the oops message */
589 pause_on_oops_flag
= 1;
591 /* We need to stall this CPU */
593 /* This CPU gets to do the counting */
594 spin_counter
= pause_on_oops
;
596 spin_unlock(&pause_on_oops_lock
);
597 spin_msec(MSEC_PER_SEC
);
598 spin_lock(&pause_on_oops_lock
);
599 } while (--spin_counter
);
600 pause_on_oops_flag
= 0;
602 /* This CPU waits for a different one */
603 while (spin_counter
) {
604 spin_unlock(&pause_on_oops_lock
);
606 spin_lock(&pause_on_oops_lock
);
610 spin_unlock_irqrestore(&pause_on_oops_lock
, flags
);
614 * Return true if the calling CPU is allowed to print oops-related info.
615 * This is a bit racy..
617 bool oops_may_print(void)
619 return pause_on_oops_flag
== 0;
623 * Called when the architecture enters its oops handler, before it prints
624 * anything. If this is the first CPU to oops, and it's oopsing the first
625 * time then let it proceed.
627 * This is all enabled by the pause_on_oops kernel boot option. We do all
628 * this to ensure that oopses don't scroll off the screen. It has the
629 * side-effect of preventing later-oopsing CPUs from mucking up the display,
632 * It turns out that the CPU which is allowed to print ends up pausing for
633 * the right duration, whereas all the other CPUs pause for twice as long:
634 * once in oops_enter(), once in oops_exit().
636 void oops_enter(void)
639 /* can't trust the integrity of the kernel anymore: */
641 do_oops_enter_exit();
643 if (sysctl_oops_all_cpu_backtrace
)
644 trigger_all_cpu_backtrace();
647 static void print_oops_end_marker(void)
649 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
653 * Called when the architecture exits its oops handler, after printing
658 do_oops_enter_exit();
659 print_oops_end_marker();
660 kmsg_dump(KMSG_DUMP_OOPS
);
668 void __warn(const char *file
, int line
, void *caller
, unsigned taint
,
669 struct pt_regs
*regs
, struct warn_args
*args
)
671 disable_trace_on_warning();
674 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
675 raw_smp_processor_id(), current
->pid
, file
, line
,
678 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
679 raw_smp_processor_id(), current
->pid
, caller
);
681 #pragma GCC diagnostic push
683 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
686 vprintk(args
->fmt
, args
->args
);
687 #pragma GCC diagnostic pop
694 check_panic_on_warn("kernel");
699 print_irqtrace_events(current
);
701 print_oops_end_marker();
702 trace_error_report_end(ERROR_DETECTOR_WARN
, (unsigned long)caller
);
704 /* Just a warning, don't kill lockdep. */
705 add_taint(taint
, LOCKDEP_STILL_OK
);
710 void warn_slowpath_fmt(const char *file
, int line
, unsigned taint
,
711 const char *fmt
, ...)
713 bool rcu
= warn_rcu_enter();
714 struct warn_args args
;
719 __warn(file
, line
, __builtin_return_address(0), taint
,
726 va_start(args
.args
, fmt
);
727 __warn(file
, line
, __builtin_return_address(0), taint
, NULL
, &args
);
731 EXPORT_SYMBOL(warn_slowpath_fmt
);
733 void __warn_printk(const char *fmt
, ...)
735 bool rcu
= warn_rcu_enter();
745 EXPORT_SYMBOL(__warn_printk
);
748 /* Support resetting WARN*_ONCE state */
750 static int clear_warn_once_set(void *data
, u64 val
)
752 generic_bug_clear_once();
753 memset(__start_once
, 0, __end_once
- __start_once
);
757 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops
, NULL
, clear_warn_once_set
,
760 static __init
int register_warn_debugfs(void)
762 /* Don't care about failure */
763 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL
, NULL
,
764 &clear_warn_once_fops
);
768 device_initcall(register_warn_debugfs
);
771 #ifdef CONFIG_STACKPROTECTOR
774 * Called when gcc's -fstack-protector feature is used, and
775 * gcc detects corruption of the on-stack canary value
777 __visible noinstr
void __stack_chk_fail(void)
779 instrumentation_begin();
780 panic("stack-protector: Kernel stack is corrupted in: %pB",
781 __builtin_return_address(0));
782 instrumentation_end();
784 EXPORT_SYMBOL(__stack_chk_fail
);
788 core_param(panic
, panic_timeout
, int, 0644);
789 core_param(panic_print
, panic_print
, ulong
, 0644);
790 core_param(pause_on_oops
, pause_on_oops
, int, 0644);
791 core_param(panic_on_warn
, panic_on_warn
, int, 0644);
792 core_param(crash_kexec_post_notifiers
, crash_kexec_post_notifiers
, bool, 0644);
794 static int __init
oops_setup(char *s
)
798 if (!strcmp(s
, "panic"))
802 early_param("oops", oops_setup
);
804 static int __init
panic_on_taint_setup(char *s
)
811 taint_str
= strsep(&s
, ",");
812 if (kstrtoul(taint_str
, 16, &panic_on_taint
))
815 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
816 panic_on_taint
&= TAINT_FLAGS_MAX
;
821 if (s
&& !strcmp(s
, "nousertaint"))
822 panic_on_taint_nousertaint
= true;
824 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
825 panic_on_taint
, str_enabled_disabled(panic_on_taint_nousertaint
));
829 early_param("panic_on_taint", panic_on_taint_setup
);