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 __noreturn
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 __noreturn
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
)
195 int old_cpu
, this_cpu
;
197 old_cpu
= PANIC_CPU_INVALID
;
198 this_cpu
= raw_smp_processor_id();
200 /* atomic_try_cmpxchg updates old_cpu on failure */
201 if (atomic_try_cmpxchg(&panic_cpu
, &old_cpu
, this_cpu
))
203 else if (old_cpu
!= this_cpu
)
204 nmi_panic_self_stop(regs
);
206 EXPORT_SYMBOL(nmi_panic
);
208 static void panic_print_sys_info(bool console_flush
)
211 if (panic_print
& PANIC_PRINT_ALL_PRINTK_MSG
)
212 console_flush_on_panic(CONSOLE_REPLAY_ALL
);
216 if (panic_print
& PANIC_PRINT_TASK_INFO
)
219 if (panic_print
& PANIC_PRINT_MEM_INFO
)
222 if (panic_print
& PANIC_PRINT_TIMER_INFO
)
223 sysrq_timer_list_show();
225 if (panic_print
& PANIC_PRINT_LOCK_INFO
)
226 debug_show_all_locks();
228 if (panic_print
& PANIC_PRINT_FTRACE_INFO
)
229 ftrace_dump(DUMP_ALL
);
232 void check_panic_on_warn(const char *origin
)
237 panic("%s: panic_on_warn set ...\n", origin
);
239 limit
= READ_ONCE(warn_limit
);
240 if (atomic_inc_return(&warn_count
) >= limit
&& limit
)
241 panic("%s: system warned too often (kernel.warn_limit is %d)",
246 * Helper that triggers the NMI backtrace (if set in panic_print)
247 * and then performs the secondary CPUs shutdown - we cannot have
248 * the NMI backtrace after the CPUs are off!
250 static void panic_other_cpus_shutdown(bool crash_kexec
)
252 if (panic_print
& PANIC_PRINT_ALL_CPU_BT
)
253 trigger_all_cpu_backtrace();
256 * Note that smp_send_stop() is the usual SMP shutdown function,
257 * which unfortunately may not be hardened to work in a panic
258 * situation. If we want to do crash dump after notifier calls
259 * and kmsg_dump, we will need architecture dependent extra
260 * bits in addition to stopping other CPUs, hence we rely on
261 * crash_smp_send_stop() for that.
266 crash_smp_send_stop();
270 * panic - halt the system
271 * @fmt: The text string to print
273 * Display a message, then perform cleanups.
275 * This function never returns.
277 void panic(const char *fmt
, ...)
279 static char buf
[1024];
281 long i
, i_next
= 0, len
;
283 int old_cpu
, this_cpu
;
284 bool _crash_kexec_post_notifiers
= crash_kexec_post_notifiers
;
288 * This thread may hit another WARN() in the panic path.
289 * Resetting this prevents additional WARN() from panicking the
290 * system on this thread. Other threads are blocked by the
291 * panic_mutex in panic().
297 * Disable local interrupts. This will prevent panic_smp_self_stop
298 * from deadlocking the first cpu that invokes the panic, since
299 * there is nothing to prevent an interrupt handler (that runs
300 * after setting panic_cpu) from invoking panic() again.
303 preempt_disable_notrace();
306 * It's possible to come here directly from a panic-assertion and
307 * not have preempt disabled. Some functions called from here want
308 * preempt to be disabled. No point enabling it later though...
310 * Only one CPU is allowed to execute the panic code from here. For
311 * multiple parallel invocations of panic, all other CPUs either
312 * stop themself or will wait until they are stopped by the 1st CPU
313 * with smp_send_stop().
315 * cmpxchg success means this is the 1st CPU which comes here,
317 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
318 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
320 old_cpu
= PANIC_CPU_INVALID
;
321 this_cpu
= raw_smp_processor_id();
323 /* atomic_try_cmpxchg updates old_cpu on failure */
324 if (atomic_try_cmpxchg(&panic_cpu
, &old_cpu
, this_cpu
)) {
326 } else if (old_cpu
!= this_cpu
)
327 panic_smp_self_stop();
332 len
= vscnprintf(buf
, sizeof(buf
), fmt
, args
);
335 if (len
&& buf
[len
- 1] == '\n')
338 pr_emerg("Kernel panic - not syncing: %s\n", buf
);
339 #ifdef CONFIG_DEBUG_BUGVERBOSE
341 * Avoid nested stack-dumping if a panic occurs during oops processing
343 if (!test_taint(TAINT_DIE
) && oops_in_progress
<= 1)
348 * If kgdb is enabled, give it a chance to run before we stop all
349 * the other CPUs or else we won't be able to debug processes left
355 * If we have crashed and we have a crash kernel loaded let it handle
357 * If we want to run this after calling panic_notifiers, pass
358 * the "crash_kexec_post_notifiers" option to the kernel.
360 * Bypass the panic_cpu check and call __crash_kexec directly.
362 if (!_crash_kexec_post_notifiers
)
365 panic_other_cpus_shutdown(_crash_kexec_post_notifiers
);
368 * Run any panic handlers, including those that might need to
369 * add information to the kmsg dump output.
371 atomic_notifier_call_chain(&panic_notifier_list
, 0, buf
);
373 panic_print_sys_info(false);
375 kmsg_dump(KMSG_DUMP_PANIC
);
378 * If you doubt kdump always works fine in any situation,
379 * "crash_kexec_post_notifiers" offers you a chance to run
380 * panic_notifiers and dumping kmsg before kdump.
381 * Note: since some panic_notifiers can make crashed kernel
382 * more unstable, it can increase risks of the kdump failure too.
384 * Bypass the panic_cpu check and call __crash_kexec directly.
386 if (_crash_kexec_post_notifiers
)
392 * We may have ended up stopping the CPU holding the lock (in
393 * smp_send_stop()) while still having some valuable data in the console
394 * buffer. Try to acquire the lock then release it regardless of the
395 * result. The release will also print the buffers out. Locks debug
396 * should be disabled to avoid reporting bad unlock balance when
397 * panic() is not being callled from OOPS.
400 console_flush_on_panic(CONSOLE_FLUSH_PENDING
);
402 panic_print_sys_info(true);
405 panic_blink
= no_blink
;
407 if (panic_timeout
> 0) {
409 * Delay timeout seconds before rebooting the machine.
410 * We can't use the "normal" timers since we just panicked.
412 pr_emerg("Rebooting in %d seconds..\n", panic_timeout
);
414 for (i
= 0; i
< panic_timeout
* 1000; i
+= PANIC_TIMER_STEP
) {
415 touch_nmi_watchdog();
417 i
+= panic_blink(state
^= 1);
418 i_next
= i
+ 3600 / PANIC_BLINK_SPD
;
420 mdelay(PANIC_TIMER_STEP
);
423 if (panic_timeout
!= 0) {
425 * This will not be a clean reboot, with everything
426 * shutting down. But if there is a chance of
427 * rebooting the system it will be rebooted.
429 if (panic_reboot_mode
!= REBOOT_UNDEFINED
)
430 reboot_mode
= panic_reboot_mode
;
435 extern int stop_a_enabled
;
436 /* Make sure the user can actually press Stop-A (L1-A) */
438 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
439 "twice on console to return to the boot prom\n");
442 #if defined(CONFIG_S390)
445 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf
);
447 /* Do not scroll important messages printed above */
450 for (i
= 0; ; i
+= PANIC_TIMER_STEP
) {
451 touch_softlockup_watchdog();
453 i
+= panic_blink(state
^= 1);
454 i_next
= i
+ 3600 / PANIC_BLINK_SPD
;
456 mdelay(PANIC_TIMER_STEP
);
460 EXPORT_SYMBOL(panic
);
463 * TAINT_FORCED_RMMOD could be a per-module flag but the module
464 * is being removed anyway.
466 const struct taint_flag taint_flags
[TAINT_FLAGS_COUNT
] = {
467 [ TAINT_PROPRIETARY_MODULE
] = { 'P', 'G', true },
468 [ TAINT_FORCED_MODULE
] = { 'F', ' ', true },
469 [ TAINT_CPU_OUT_OF_SPEC
] = { 'S', ' ', false },
470 [ TAINT_FORCED_RMMOD
] = { 'R', ' ', false },
471 [ TAINT_MACHINE_CHECK
] = { 'M', ' ', false },
472 [ TAINT_BAD_PAGE
] = { 'B', ' ', false },
473 [ TAINT_USER
] = { 'U', ' ', false },
474 [ TAINT_DIE
] = { 'D', ' ', false },
475 [ TAINT_OVERRIDDEN_ACPI_TABLE
] = { 'A', ' ', false },
476 [ TAINT_WARN
] = { 'W', ' ', false },
477 [ TAINT_CRAP
] = { 'C', ' ', true },
478 [ TAINT_FIRMWARE_WORKAROUND
] = { 'I', ' ', false },
479 [ TAINT_OOT_MODULE
] = { 'O', ' ', true },
480 [ TAINT_UNSIGNED_MODULE
] = { 'E', ' ', true },
481 [ TAINT_SOFTLOCKUP
] = { 'L', ' ', false },
482 [ TAINT_LIVEPATCH
] = { 'K', ' ', true },
483 [ TAINT_AUX
] = { 'X', ' ', true },
484 [ TAINT_RANDSTRUCT
] = { 'T', ' ', true },
485 [ TAINT_TEST
] = { 'N', ' ', true },
489 * print_tainted - return a string to represent the kernel taint state.
491 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
493 * The string is overwritten by the next call to print_tainted(),
494 * but is always NULL terminated.
496 const char *print_tainted(void)
498 static char buf
[TAINT_FLAGS_COUNT
+ sizeof("Tainted: ")];
500 BUILD_BUG_ON(ARRAY_SIZE(taint_flags
) != TAINT_FLAGS_COUNT
);
506 s
= buf
+ sprintf(buf
, "Tainted: ");
507 for (i
= 0; i
< TAINT_FLAGS_COUNT
; i
++) {
508 const struct taint_flag
*t
= &taint_flags
[i
];
509 *s
++ = test_bit(i
, &tainted_mask
) ?
510 t
->c_true
: t
->c_false
;
514 snprintf(buf
, sizeof(buf
), "Not tainted");
519 int test_taint(unsigned flag
)
521 return test_bit(flag
, &tainted_mask
);
523 EXPORT_SYMBOL(test_taint
);
525 unsigned long get_taint(void)
531 * add_taint: add a taint flag if not already set.
532 * @flag: one of the TAINT_* constants.
533 * @lockdep_ok: whether lock debugging is still OK.
535 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
536 * some notewortht-but-not-corrupting cases, it can be set to true.
538 void add_taint(unsigned flag
, enum lockdep_ok lockdep_ok
)
540 if (lockdep_ok
== LOCKDEP_NOW_UNRELIABLE
&& __debug_locks_off())
541 pr_warn("Disabling lock debugging due to kernel taint\n");
543 set_bit(flag
, &tainted_mask
);
545 if (tainted_mask
& panic_on_taint
) {
547 panic("panic_on_taint set ...");
550 EXPORT_SYMBOL(add_taint
);
552 static void spin_msec(int msecs
)
556 for (i
= 0; i
< msecs
; i
++) {
557 touch_nmi_watchdog();
563 * It just happens that oops_enter() and oops_exit() are identically
566 static void do_oops_enter_exit(void)
569 static int spin_counter
;
574 spin_lock_irqsave(&pause_on_oops_lock
, flags
);
575 if (pause_on_oops_flag
== 0) {
576 /* This CPU may now print the oops message */
577 pause_on_oops_flag
= 1;
579 /* We need to stall this CPU */
581 /* This CPU gets to do the counting */
582 spin_counter
= pause_on_oops
;
584 spin_unlock(&pause_on_oops_lock
);
585 spin_msec(MSEC_PER_SEC
);
586 spin_lock(&pause_on_oops_lock
);
587 } while (--spin_counter
);
588 pause_on_oops_flag
= 0;
590 /* This CPU waits for a different one */
591 while (spin_counter
) {
592 spin_unlock(&pause_on_oops_lock
);
594 spin_lock(&pause_on_oops_lock
);
598 spin_unlock_irqrestore(&pause_on_oops_lock
, flags
);
602 * Return true if the calling CPU is allowed to print oops-related info.
603 * This is a bit racy..
605 bool oops_may_print(void)
607 return pause_on_oops_flag
== 0;
611 * Called when the architecture enters its oops handler, before it prints
612 * anything. If this is the first CPU to oops, and it's oopsing the first
613 * time then let it proceed.
615 * This is all enabled by the pause_on_oops kernel boot option. We do all
616 * this to ensure that oopses don't scroll off the screen. It has the
617 * side-effect of preventing later-oopsing CPUs from mucking up the display,
620 * It turns out that the CPU which is allowed to print ends up pausing for
621 * the right duration, whereas all the other CPUs pause for twice as long:
622 * once in oops_enter(), once in oops_exit().
624 void oops_enter(void)
627 /* can't trust the integrity of the kernel anymore: */
629 do_oops_enter_exit();
631 if (sysctl_oops_all_cpu_backtrace
)
632 trigger_all_cpu_backtrace();
635 static void print_oops_end_marker(void)
637 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
641 * Called when the architecture exits its oops handler, after printing
646 do_oops_enter_exit();
647 print_oops_end_marker();
648 kmsg_dump(KMSG_DUMP_OOPS
);
656 void __warn(const char *file
, int line
, void *caller
, unsigned taint
,
657 struct pt_regs
*regs
, struct warn_args
*args
)
659 disable_trace_on_warning();
662 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
663 raw_smp_processor_id(), current
->pid
, file
, line
,
666 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
667 raw_smp_processor_id(), current
->pid
, caller
);
670 vprintk(args
->fmt
, args
->args
);
677 check_panic_on_warn("kernel");
682 print_irqtrace_events(current
);
684 print_oops_end_marker();
685 trace_error_report_end(ERROR_DETECTOR_WARN
, (unsigned long)caller
);
687 /* Just a warning, don't kill lockdep. */
688 add_taint(taint
, LOCKDEP_STILL_OK
);
693 void warn_slowpath_fmt(const char *file
, int line
, unsigned taint
,
694 const char *fmt
, ...)
696 bool rcu
= warn_rcu_enter();
697 struct warn_args args
;
702 __warn(file
, line
, __builtin_return_address(0), taint
,
709 va_start(args
.args
, fmt
);
710 __warn(file
, line
, __builtin_return_address(0), taint
, NULL
, &args
);
714 EXPORT_SYMBOL(warn_slowpath_fmt
);
716 void __warn_printk(const char *fmt
, ...)
718 bool rcu
= warn_rcu_enter();
728 EXPORT_SYMBOL(__warn_printk
);
731 /* Support resetting WARN*_ONCE state */
733 static int clear_warn_once_set(void *data
, u64 val
)
735 generic_bug_clear_once();
736 memset(__start_once
, 0, __end_once
- __start_once
);
740 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops
, NULL
, clear_warn_once_set
,
743 static __init
int register_warn_debugfs(void)
745 /* Don't care about failure */
746 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL
, NULL
,
747 &clear_warn_once_fops
);
751 device_initcall(register_warn_debugfs
);
754 #ifdef CONFIG_STACKPROTECTOR
757 * Called when gcc's -fstack-protector feature is used, and
758 * gcc detects corruption of the on-stack canary value
760 __visible noinstr
void __stack_chk_fail(void)
762 instrumentation_begin();
763 panic("stack-protector: Kernel stack is corrupted in: %pB",
764 __builtin_return_address(0));
765 instrumentation_end();
767 EXPORT_SYMBOL(__stack_chk_fail
);
771 core_param(panic
, panic_timeout
, int, 0644);
772 core_param(panic_print
, panic_print
, ulong
, 0644);
773 core_param(pause_on_oops
, pause_on_oops
, int, 0644);
774 core_param(panic_on_warn
, panic_on_warn
, int, 0644);
775 core_param(crash_kexec_post_notifiers
, crash_kexec_post_notifiers
, bool, 0644);
777 static int __init
oops_setup(char *s
)
781 if (!strcmp(s
, "panic"))
785 early_param("oops", oops_setup
);
787 static int __init
panic_on_taint_setup(char *s
)
794 taint_str
= strsep(&s
, ",");
795 if (kstrtoul(taint_str
, 16, &panic_on_taint
))
798 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
799 panic_on_taint
&= TAINT_FLAGS_MAX
;
804 if (s
&& !strcmp(s
, "nousertaint"))
805 panic_on_taint_nousertaint
= true;
807 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
808 panic_on_taint
, str_enabled_disabled(panic_on_taint_nousertaint
));
812 early_param("panic_on_taint", panic_on_taint_setup
);