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/interrupt.h>
13 #include <linux/kmsg_dump.h>
14 #include <linux/kallsyms.h>
15 #include <linux/notifier.h>
16 #include <linux/module.h>
17 #include <linux/random.h>
18 #include <linux/ftrace.h>
19 #include <linux/reboot.h>
20 #include <linux/delay.h>
21 #include <linux/kexec.h>
22 #include <linux/sched.h>
23 #include <linux/sysrq.h>
24 #include <linux/init.h>
25 #include <linux/nmi.h>
26 #include <linux/console.h>
27 #include <linux/bug.h>
29 #define PANIC_TIMER_STEP 100
30 #define PANIC_BLINK_SPD 18
32 int panic_on_oops
= CONFIG_PANIC_ON_OOPS_VALUE
;
33 static unsigned long tainted_mask
;
34 static int pause_on_oops
;
35 static int pause_on_oops_flag
;
36 static DEFINE_SPINLOCK(pause_on_oops_lock
);
37 bool crash_kexec_post_notifiers
;
38 int panic_on_warn __read_mostly
;
40 int panic_timeout
= CONFIG_PANIC_TIMEOUT
;
41 EXPORT_SYMBOL_GPL(panic_timeout
);
43 ATOMIC_NOTIFIER_HEAD(panic_notifier_list
);
45 EXPORT_SYMBOL(panic_notifier_list
);
47 static long no_blink(int state
)
52 /* Returns how long it waited in ms */
53 long (*panic_blink
)(int state
);
54 EXPORT_SYMBOL(panic_blink
);
57 * Stop ourself in panic -- architecture code may override this
59 void __weak
panic_smp_self_stop(void)
66 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
67 * may override this to prepare for crash dumping, e.g. save regs info.
69 void __weak
nmi_panic_self_stop(struct pt_regs
*regs
)
71 panic_smp_self_stop();
75 * Stop other CPUs in panic. Architecture dependent code may override this
76 * with more suitable version. For example, if the architecture supports
77 * crash dump, it should save registers of each stopped CPU and disable
78 * per-CPU features such as virtualization extensions.
80 void __weak
crash_smp_send_stop(void)
82 static int cpus_stopped
;
85 * This function can be called twice in panic path, but obviously
86 * we execute this only once.
92 * Note smp_send_stop is the usual smp shutdown function, which
93 * unfortunately means it may not be hardened to work in a panic
100 atomic_t panic_cpu
= ATOMIC_INIT(PANIC_CPU_INVALID
);
103 * A variant of panic() called from NMI context. We return if we've already
104 * panicked on this CPU. If another CPU already panicked, loop in
105 * nmi_panic_self_stop() which can provide architecture dependent code such
106 * as saving register state for crash dump.
108 void nmi_panic(struct pt_regs
*regs
, const char *msg
)
112 cpu
= raw_smp_processor_id();
113 old_cpu
= atomic_cmpxchg(&panic_cpu
, PANIC_CPU_INVALID
, cpu
);
115 if (old_cpu
== PANIC_CPU_INVALID
)
117 else if (old_cpu
!= cpu
)
118 nmi_panic_self_stop(regs
);
120 EXPORT_SYMBOL(nmi_panic
);
123 * panic - halt the system
124 * @fmt: The text string to print
126 * Display a message, then perform cleanups.
128 * This function never returns.
130 void panic(const char *fmt
, ...)
132 static char buf
[1024];
136 int old_cpu
, this_cpu
;
137 bool _crash_kexec_post_notifiers
= crash_kexec_post_notifiers
;
140 * Disable local interrupts. This will prevent panic_smp_self_stop
141 * from deadlocking the first cpu that invokes the panic, since
142 * there is nothing to prevent an interrupt handler (that runs
143 * after setting panic_cpu) from invoking panic() again.
148 * It's possible to come here directly from a panic-assertion and
149 * not have preempt disabled. Some functions called from here want
150 * preempt to be disabled. No point enabling it later though...
152 * Only one CPU is allowed to execute the panic code from here. For
153 * multiple parallel invocations of panic, all other CPUs either
154 * stop themself or will wait until they are stopped by the 1st CPU
155 * with smp_send_stop().
157 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
158 * comes here, so go ahead.
159 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
160 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
162 this_cpu
= raw_smp_processor_id();
163 old_cpu
= atomic_cmpxchg(&panic_cpu
, PANIC_CPU_INVALID
, this_cpu
);
165 if (old_cpu
!= PANIC_CPU_INVALID
&& old_cpu
!= this_cpu
)
166 panic_smp_self_stop();
171 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
173 pr_emerg("Kernel panic - not syncing: %s\n", buf
);
174 #ifdef CONFIG_DEBUG_BUGVERBOSE
176 * Avoid nested stack-dumping if a panic occurs during oops processing
178 if (!test_taint(TAINT_DIE
) && oops_in_progress
<= 1)
183 * If we have crashed and we have a crash kernel loaded let it handle
185 * If we want to run this after calling panic_notifiers, pass
186 * the "crash_kexec_post_notifiers" option to the kernel.
188 * Bypass the panic_cpu check and call __crash_kexec directly.
190 if (!_crash_kexec_post_notifiers
) {
191 printk_nmi_flush_on_panic();
195 * Note smp_send_stop is the usual smp shutdown function, which
196 * unfortunately means it may not be hardened to work in a
202 * If we want to do crash dump after notifier calls and
203 * kmsg_dump, we will need architecture dependent extra
204 * works in addition to stopping other CPUs.
206 crash_smp_send_stop();
210 * Run any panic handlers, including those that might need to
211 * add information to the kmsg dump output.
213 atomic_notifier_call_chain(&panic_notifier_list
, 0, buf
);
215 /* Call flush even twice. It tries harder with a single online CPU */
216 printk_nmi_flush_on_panic();
217 kmsg_dump(KMSG_DUMP_PANIC
);
220 * If you doubt kdump always works fine in any situation,
221 * "crash_kexec_post_notifiers" offers you a chance to run
222 * panic_notifiers and dumping kmsg before kdump.
223 * Note: since some panic_notifiers can make crashed kernel
224 * more unstable, it can increase risks of the kdump failure too.
226 * Bypass the panic_cpu check and call __crash_kexec directly.
228 if (_crash_kexec_post_notifiers
)
234 * We may have ended up stopping the CPU holding the lock (in
235 * smp_send_stop()) while still having some valuable data in the console
236 * buffer. Try to acquire the lock then release it regardless of the
237 * result. The release will also print the buffers out. Locks debug
238 * should be disabled to avoid reporting bad unlock balance when
239 * panic() is not being callled from OOPS.
242 console_flush_on_panic();
245 panic_blink
= no_blink
;
247 if (panic_timeout
> 0) {
249 * Delay timeout seconds before rebooting the machine.
250 * We can't use the "normal" timers since we just panicked.
252 pr_emerg("Rebooting in %d seconds..\n", panic_timeout
);
254 for (i
= 0; i
< panic_timeout
* 1000; i
+= PANIC_TIMER_STEP
) {
255 touch_nmi_watchdog();
257 i
+= panic_blink(state
^= 1);
258 i_next
= i
+ 3600 / PANIC_BLINK_SPD
;
260 mdelay(PANIC_TIMER_STEP
);
263 if (panic_timeout
!= 0) {
265 * This will not be a clean reboot, with everything
266 * shutting down. But if there is a chance of
267 * rebooting the system it will be rebooted.
273 extern int stop_a_enabled
;
274 /* Make sure the user can actually press Stop-A (L1-A) */
276 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
279 #if defined(CONFIG_S390)
281 unsigned long caller
;
283 caller
= (unsigned long)__builtin_return_address(0);
284 disabled_wait(caller
);
287 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf
);
289 for (i
= 0; ; i
+= PANIC_TIMER_STEP
) {
290 touch_softlockup_watchdog();
292 i
+= panic_blink(state
^= 1);
293 i_next
= i
+ 3600 / PANIC_BLINK_SPD
;
295 mdelay(PANIC_TIMER_STEP
);
299 EXPORT_SYMBOL(panic
);
302 * TAINT_FORCED_RMMOD could be a per-module flag but the module
303 * is being removed anyway.
305 const struct taint_flag taint_flags
[TAINT_FLAGS_COUNT
] = {
306 { 'P', 'G', true }, /* TAINT_PROPRIETARY_MODULE */
307 { 'F', ' ', true }, /* TAINT_FORCED_MODULE */
308 { 'S', ' ', false }, /* TAINT_CPU_OUT_OF_SPEC */
309 { 'R', ' ', false }, /* TAINT_FORCED_RMMOD */
310 { 'M', ' ', false }, /* TAINT_MACHINE_CHECK */
311 { 'B', ' ', false }, /* TAINT_BAD_PAGE */
312 { 'U', ' ', false }, /* TAINT_USER */
313 { 'D', ' ', false }, /* TAINT_DIE */
314 { 'A', ' ', false }, /* TAINT_OVERRIDDEN_ACPI_TABLE */
315 { 'W', ' ', false }, /* TAINT_WARN */
316 { 'C', ' ', true }, /* TAINT_CRAP */
317 { 'I', ' ', false }, /* TAINT_FIRMWARE_WORKAROUND */
318 { 'O', ' ', true }, /* TAINT_OOT_MODULE */
319 { 'E', ' ', true }, /* TAINT_UNSIGNED_MODULE */
320 { 'L', ' ', false }, /* TAINT_SOFTLOCKUP */
321 { 'K', ' ', true }, /* TAINT_LIVEPATCH */
325 * print_tainted - return a string to represent the kernel taint state.
327 * 'P' - Proprietary module has been loaded.
328 * 'F' - Module has been forcibly loaded.
329 * 'S' - SMP with CPUs not designed for SMP.
330 * 'R' - User forced a module unload.
331 * 'M' - System experienced a machine check exception.
332 * 'B' - System has hit bad_page.
333 * 'U' - Userspace-defined naughtiness.
334 * 'D' - Kernel has oopsed before
335 * 'A' - ACPI table overridden.
336 * 'W' - Taint on warning.
337 * 'C' - modules from drivers/staging are loaded.
338 * 'I' - Working around severe firmware bug.
339 * 'O' - Out-of-tree module has been loaded.
340 * 'E' - Unsigned module has been loaded.
341 * 'L' - A soft lockup has previously occurred.
342 * 'K' - Kernel has been live patched.
344 * The string is overwritten by the next call to print_tainted().
346 const char *print_tainted(void)
348 static char buf
[TAINT_FLAGS_COUNT
+ sizeof("Tainted: ")];
354 s
= buf
+ sprintf(buf
, "Tainted: ");
355 for (i
= 0; i
< TAINT_FLAGS_COUNT
; i
++) {
356 const struct taint_flag
*t
= &taint_flags
[i
];
357 *s
++ = test_bit(i
, &tainted_mask
) ?
358 t
->c_true
: t
->c_false
;
362 snprintf(buf
, sizeof(buf
), "Not tainted");
367 int test_taint(unsigned flag
)
369 return test_bit(flag
, &tainted_mask
);
371 EXPORT_SYMBOL(test_taint
);
373 unsigned long get_taint(void)
379 * add_taint: add a taint flag if not already set.
380 * @flag: one of the TAINT_* constants.
381 * @lockdep_ok: whether lock debugging is still OK.
383 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
384 * some notewortht-but-not-corrupting cases, it can be set to true.
386 void add_taint(unsigned flag
, enum lockdep_ok lockdep_ok
)
388 if (lockdep_ok
== LOCKDEP_NOW_UNRELIABLE
&& __debug_locks_off())
389 pr_warn("Disabling lock debugging due to kernel taint\n");
391 set_bit(flag
, &tainted_mask
);
393 EXPORT_SYMBOL(add_taint
);
395 static void spin_msec(int msecs
)
399 for (i
= 0; i
< msecs
; i
++) {
400 touch_nmi_watchdog();
406 * It just happens that oops_enter() and oops_exit() are identically
409 static void do_oops_enter_exit(void)
412 static int spin_counter
;
417 spin_lock_irqsave(&pause_on_oops_lock
, flags
);
418 if (pause_on_oops_flag
== 0) {
419 /* This CPU may now print the oops message */
420 pause_on_oops_flag
= 1;
422 /* We need to stall this CPU */
424 /* This CPU gets to do the counting */
425 spin_counter
= pause_on_oops
;
427 spin_unlock(&pause_on_oops_lock
);
428 spin_msec(MSEC_PER_SEC
);
429 spin_lock(&pause_on_oops_lock
);
430 } while (--spin_counter
);
431 pause_on_oops_flag
= 0;
433 /* This CPU waits for a different one */
434 while (spin_counter
) {
435 spin_unlock(&pause_on_oops_lock
);
437 spin_lock(&pause_on_oops_lock
);
441 spin_unlock_irqrestore(&pause_on_oops_lock
, flags
);
445 * Return true if the calling CPU is allowed to print oops-related info.
446 * This is a bit racy..
448 int oops_may_print(void)
450 return pause_on_oops_flag
== 0;
454 * Called when the architecture enters its oops handler, before it prints
455 * anything. If this is the first CPU to oops, and it's oopsing the first
456 * time then let it proceed.
458 * This is all enabled by the pause_on_oops kernel boot option. We do all
459 * this to ensure that oopses don't scroll off the screen. It has the
460 * side-effect of preventing later-oopsing CPUs from mucking up the display,
463 * It turns out that the CPU which is allowed to print ends up pausing for
464 * the right duration, whereas all the other CPUs pause for twice as long:
465 * once in oops_enter(), once in oops_exit().
467 void oops_enter(void)
470 /* can't trust the integrity of the kernel anymore: */
472 do_oops_enter_exit();
476 * 64-bit random ID for oopses:
480 static int init_oops_id(void)
483 get_random_bytes(&oops_id
, sizeof(oops_id
));
489 late_initcall(init_oops_id
);
491 void print_oops_end_marker(void)
494 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id
);
498 * Called when the architecture exits its oops handler, after printing
503 do_oops_enter_exit();
504 print_oops_end_marker();
505 kmsg_dump(KMSG_DUMP_OOPS
);
513 void __warn(const char *file
, int line
, void *caller
, unsigned taint
,
514 struct pt_regs
*regs
, struct warn_args
*args
)
516 disable_trace_on_warning();
518 pr_warn("------------[ cut here ]------------\n");
521 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
522 raw_smp_processor_id(), current
->pid
, file
, line
,
525 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
526 raw_smp_processor_id(), current
->pid
, caller
);
529 vprintk(args
->fmt
, args
->args
);
533 * This thread may hit another WARN() in the panic path.
534 * Resetting this prevents additional WARN() from panicking the
535 * system on this thread. Other threads are blocked by the
536 * panic_mutex in panic().
539 panic("panic_on_warn set ...\n");
549 print_oops_end_marker();
551 /* Just a warning, don't kill lockdep. */
552 add_taint(taint
, LOCKDEP_STILL_OK
);
555 #ifdef WANT_WARN_ON_SLOWPATH
556 void warn_slowpath_fmt(const char *file
, int line
, const char *fmt
, ...)
558 struct warn_args args
;
561 va_start(args
.args
, fmt
);
562 __warn(file
, line
, __builtin_return_address(0), TAINT_WARN
, NULL
,
566 EXPORT_SYMBOL(warn_slowpath_fmt
);
568 void warn_slowpath_fmt_taint(const char *file
, int line
,
569 unsigned taint
, const char *fmt
, ...)
571 struct warn_args args
;
574 va_start(args
.args
, fmt
);
575 __warn(file
, line
, __builtin_return_address(0), taint
, NULL
, &args
);
578 EXPORT_SYMBOL(warn_slowpath_fmt_taint
);
580 void warn_slowpath_null(const char *file
, int line
)
582 __warn(file
, line
, __builtin_return_address(0), TAINT_WARN
, NULL
, NULL
);
584 EXPORT_SYMBOL(warn_slowpath_null
);
587 #ifdef CONFIG_CC_STACKPROTECTOR
590 * Called when gcc's -fstack-protector feature is used, and
591 * gcc detects corruption of the on-stack canary value
593 __visible
void __stack_chk_fail(void)
595 panic("stack-protector: Kernel stack is corrupted in: %p\n",
596 __builtin_return_address(0));
598 EXPORT_SYMBOL(__stack_chk_fail
);
602 core_param(panic
, panic_timeout
, int, 0644);
603 core_param(pause_on_oops
, pause_on_oops
, int, 0644);
604 core_param(panic_on_warn
, panic_on_warn
, int, 0644);
605 core_param(crash_kexec_post_notifiers
, crash_kexec_post_notifiers
, bool, 0644);
607 static int __init
oops_setup(char *s
)
611 if (!strcmp(s
, "panic"))
615 early_param("oops", oops_setup
);