1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/panic.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8 /*
9 * This function is used through-out the kernel (including mm and fs)
10 * to indicate a major problem.
11 */
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 <linux/seq_buf.h>
39 #include <linux/sys_info.h>
40 #include <trace/events/error_report.h>
41 #include <asm/sections.h>
42
43 #define PANIC_TIMER_STEP 100
44 #define PANIC_BLINK_SPD 18
45 #define PANIC_MSG_BUFSZ 1024
46
47 #ifdef CONFIG_SMP
48 /*
49 * Should we dump all CPUs backtraces in an oops event?
50 * Defaults to 0, can be changed via sysctl.
51 */
52 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
53 #else
54 #define sysctl_oops_all_cpu_backtrace 0
55 #endif /* CONFIG_SMP */
56
57 int panic_on_oops = IS_ENABLED(CONFIG_PANIC_ON_OOPS);
58 static unsigned long tainted_mask =
59 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
60 static int pause_on_oops;
61 static int pause_on_oops_flag;
62 static DEFINE_SPINLOCK(pause_on_oops_lock);
63 bool crash_kexec_post_notifiers;
64 int panic_on_warn __read_mostly;
65 unsigned long panic_on_taint;
66 bool panic_on_taint_nousertaint = false;
67 static unsigned int warn_limit __read_mostly;
68 static bool panic_console_replay;
69
70 bool panic_triggering_all_cpu_backtrace;
71 static bool panic_this_cpu_backtrace_printed;
72
73 int panic_timeout = CONFIG_PANIC_TIMEOUT;
74 EXPORT_SYMBOL_GPL(panic_timeout);
75
76 unsigned long panic_print;
77
78 static int panic_force_cpu = -1;
79
80 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
81
82 EXPORT_SYMBOL(panic_notifier_list);
83
panic_print_deprecated(void)84 static void panic_print_deprecated(void)
85 {
86 pr_info_once("Kernel: The 'panic_print' parameter is now deprecated. Please use 'panic_sys_info' and 'panic_console_replay' instead.\n");
87 }
88
89 #ifdef CONFIG_SYSCTL
90
91 /*
92 * Taint values can only be increased
93 * This means we can safely use a temporary.
94 */
proc_taint(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)95 static int proc_taint(const struct ctl_table *table, int write,
96 void *buffer, size_t *lenp, loff_t *ppos)
97 {
98 struct ctl_table t;
99 unsigned long tmptaint = get_taint();
100 int err;
101
102 if (write && !capable(CAP_SYS_ADMIN))
103 return -EPERM;
104
105 t = *table;
106 t.data = &tmptaint;
107 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
108 if (err < 0)
109 return err;
110
111 if (write) {
112 int i;
113
114 /*
115 * If we are relying on panic_on_taint not producing
116 * false positives due to userspace input, bail out
117 * before setting the requested taint flags.
118 */
119 if (panic_on_taint_nousertaint && (tmptaint & panic_on_taint))
120 return -EINVAL;
121
122 /*
123 * Poor man's atomic or. Not worth adding a primitive
124 * to everyone's atomic.h for this
125 */
126 for (i = 0; i < TAINT_FLAGS_COUNT; i++)
127 if ((1UL << i) & tmptaint)
128 add_taint(i, LOCKDEP_STILL_OK);
129 }
130
131 return err;
132 }
133
sysctl_panic_print_handler(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)134 static int sysctl_panic_print_handler(const struct ctl_table *table, int write,
135 void *buffer, size_t *lenp, loff_t *ppos)
136 {
137 if (write)
138 panic_print_deprecated();
139 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
140 }
141
142 static const struct ctl_table kern_panic_table[] = {
143 #ifdef CONFIG_SMP
144 {
145 .procname = "oops_all_cpu_backtrace",
146 .data = &sysctl_oops_all_cpu_backtrace,
147 .maxlen = sizeof(int),
148 .mode = 0644,
149 .proc_handler = proc_dointvec_minmax,
150 .extra1 = SYSCTL_ZERO,
151 .extra2 = SYSCTL_ONE,
152 },
153 #endif
154 {
155 .procname = "tainted",
156 .maxlen = sizeof(long),
157 .mode = 0644,
158 .proc_handler = proc_taint,
159 },
160 {
161 .procname = "panic",
162 .data = &panic_timeout,
163 .maxlen = sizeof(int),
164 .mode = 0644,
165 .proc_handler = proc_dointvec,
166 },
167 {
168 .procname = "panic_on_oops",
169 .data = &panic_on_oops,
170 .maxlen = sizeof(int),
171 .mode = 0644,
172 .proc_handler = proc_dointvec,
173 },
174 {
175 .procname = "panic_print",
176 .data = &panic_print,
177 .maxlen = sizeof(unsigned long),
178 .mode = 0644,
179 .proc_handler = sysctl_panic_print_handler,
180 },
181 {
182 .procname = "panic_on_warn",
183 .data = &panic_on_warn,
184 .maxlen = sizeof(int),
185 .mode = 0644,
186 .proc_handler = proc_dointvec_minmax,
187 .extra1 = SYSCTL_ZERO,
188 .extra2 = SYSCTL_ONE,
189 },
190 {
191 .procname = "warn_limit",
192 .data = &warn_limit,
193 .maxlen = sizeof(warn_limit),
194 .mode = 0644,
195 .proc_handler = proc_douintvec,
196 },
197 #if (defined(CONFIG_X86_32) || defined(CONFIG_PARISC)) && \
198 defined(CONFIG_DEBUG_STACKOVERFLOW)
199 {
200 .procname = "panic_on_stackoverflow",
201 .data = &sysctl_panic_on_stackoverflow,
202 .maxlen = sizeof(int),
203 .mode = 0644,
204 .proc_handler = proc_dointvec,
205 },
206 #endif
207 {
208 .procname = "panic_sys_info",
209 .data = &panic_print,
210 .maxlen = sizeof(panic_print),
211 .mode = 0644,
212 .proc_handler = sysctl_sys_info_handler,
213 },
214 };
215
kernel_panic_sysctls_init(void)216 static __init int kernel_panic_sysctls_init(void)
217 {
218 register_sysctl_init("kernel", kern_panic_table);
219 return 0;
220 }
221 late_initcall(kernel_panic_sysctls_init);
222 #endif
223
224 /* The format is "panic_sys_info=tasks,mem,locks,ftrace,..." */
setup_panic_sys_info(char * buf)225 static int __init setup_panic_sys_info(char *buf)
226 {
227 /* There is no risk of race in kernel boot phase */
228 panic_print = sys_info_parse_param(buf);
229 return 1;
230 }
231 __setup("panic_sys_info=", setup_panic_sys_info);
232
233 static atomic_t warn_count = ATOMIC_INIT(0);
234
235 #ifdef CONFIG_SYSFS
warn_count_show(struct kobject * kobj,struct kobj_attribute * attr,char * page)236 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
237 char *page)
238 {
239 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
240 }
241
242 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
243
kernel_panic_sysfs_init(void)244 static __init int kernel_panic_sysfs_init(void)
245 {
246 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
247 return 0;
248 }
249 late_initcall(kernel_panic_sysfs_init);
250 #endif
251
no_blink(int state)252 static long no_blink(int state)
253 {
254 return 0;
255 }
256
257 /* Returns how long it waited in ms */
258 long (*panic_blink)(int state);
259 EXPORT_SYMBOL(panic_blink);
260
261 /*
262 * Stop ourself in panic -- architecture code may override this
263 */
panic_smp_self_stop(void)264 void __weak __noreturn panic_smp_self_stop(void)
265 {
266 while (1)
267 cpu_relax();
268 }
269
270 /*
271 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
272 * may override this to prepare for crash dumping, e.g. save regs info.
273 */
nmi_panic_self_stop(struct pt_regs * regs)274 void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
275 {
276 panic_smp_self_stop();
277 }
278
279 /*
280 * Stop other CPUs in panic. Architecture dependent code may override this
281 * with more suitable version. For example, if the architecture supports
282 * crash dump, it should save registers of each stopped CPU and disable
283 * per-CPU features such as virtualization extensions.
284 */
crash_smp_send_stop(void)285 void __weak crash_smp_send_stop(void)
286 {
287 static int cpus_stopped;
288
289 /*
290 * This function can be called twice in panic path, but obviously
291 * we execute this only once.
292 */
293 if (cpus_stopped)
294 return;
295
296 /*
297 * Note smp_send_stop is the usual smp shutdown function, which
298 * unfortunately means it may not be hardened to work in a panic
299 * situation.
300 */
301 smp_send_stop();
302 cpus_stopped = 1;
303 }
304
305 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
306 atomic_t panic_redirect_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
307
308 #if defined(CONFIG_SMP) && defined(CONFIG_CRASH_DUMP)
309 static char *panic_force_buf;
310
panic_force_cpu_setup(char * str)311 static int __init panic_force_cpu_setup(char *str)
312 {
313 int cpu;
314
315 if (!str)
316 return -EINVAL;
317
318 if (kstrtoint(str, 0, &cpu) || cpu < 0 || cpu >= nr_cpu_ids) {
319 pr_warn("panic_force_cpu: invalid value '%s'\n", str);
320 return -EINVAL;
321 }
322
323 panic_force_cpu = cpu;
324 return 0;
325 }
326 early_param("panic_force_cpu", panic_force_cpu_setup);
327
panic_force_cpu_late_init(void)328 static int __init panic_force_cpu_late_init(void)
329 {
330 if (panic_force_cpu < 0)
331 return 0;
332
333 panic_force_buf = kmalloc(PANIC_MSG_BUFSZ, GFP_KERNEL);
334
335 return 0;
336 }
337 late_initcall(panic_force_cpu_late_init);
338
do_panic_on_target_cpu(void * info)339 static void do_panic_on_target_cpu(void *info)
340 {
341 panic("%s", (char *)info);
342 }
343
344 /**
345 * panic_smp_redirect_cpu - Redirect panic to target CPU
346 * @target_cpu: CPU that should handle the panic
347 * @msg: formatted panic message
348 *
349 * Default implementation uses IPI. Architectures with NMI support
350 * can override this for more reliable delivery.
351 *
352 * Return: 0 on success, negative errno on failure
353 */
panic_smp_redirect_cpu(int target_cpu,void * msg)354 int __weak panic_smp_redirect_cpu(int target_cpu, void *msg)
355 {
356 static call_single_data_t panic_csd;
357
358 panic_csd.func = do_panic_on_target_cpu;
359 panic_csd.info = msg;
360
361 return smp_call_function_single_async(target_cpu, &panic_csd);
362 }
363
364 /**
365 * panic_try_force_cpu - Redirect panic to a specific CPU for crash kernel
366 * @fmt: panic message format string
367 * @args: arguments for format string
368 *
369 * Some platforms require panic handling to occur on a specific CPU
370 * for the crash kernel to function correctly. This function redirects
371 * panic handling to the CPU specified via the panic_force_cpu= boot parameter.
372 *
373 * Returns false if panic should proceed on current CPU.
374 * Returns true if panic was redirected.
375 */
376 __printf(1, 0)
panic_try_force_cpu(const char * fmt,va_list args)377 static bool panic_try_force_cpu(const char *fmt, va_list args)
378 {
379 int this_cpu = raw_smp_processor_id();
380 int old_cpu = PANIC_CPU_INVALID;
381 const char *msg;
382
383 /* Feature not enabled via boot parameter */
384 if (panic_force_cpu < 0)
385 return false;
386
387 /* Already on target CPU - proceed normally */
388 if (this_cpu == panic_force_cpu)
389 return false;
390
391 /* Target CPU is offline, can't redirect */
392 if (!cpu_online(panic_force_cpu)) {
393 pr_warn("panic: target CPU %d is offline, continuing on CPU %d\n",
394 panic_force_cpu, this_cpu);
395 return false;
396 }
397
398 /* Another panic already in progress */
399 if (panic_in_progress())
400 return false;
401
402 /*
403 * Only one CPU can do the redirect. Use atomic cmpxchg to ensure
404 * we don't race with another CPU also trying to redirect.
405 */
406 if (!atomic_try_cmpxchg(&panic_redirect_cpu, &old_cpu, this_cpu))
407 return false;
408
409 /*
410 * Use dynamically allocated buffer if available, otherwise
411 * fall back to static message for early boot panics or allocation failure.
412 */
413 if (panic_force_buf) {
414 vsnprintf(panic_force_buf, PANIC_MSG_BUFSZ, fmt, args);
415 msg = panic_force_buf;
416 } else {
417 msg = "Redirected panic (buffer unavailable)";
418 }
419
420 console_verbose();
421 bust_spinlocks(1);
422
423 pr_emerg("panic: Redirecting from CPU %d to CPU %d for crash kernel.\n",
424 this_cpu, panic_force_cpu);
425
426 /* Dump original CPU before redirecting */
427 if (!test_taint(TAINT_DIE) &&
428 oops_in_progress <= 1 &&
429 IS_ENABLED(CONFIG_DEBUG_BUGVERBOSE)) {
430 dump_stack();
431 }
432
433 if (panic_smp_redirect_cpu(panic_force_cpu, (void *)msg) != 0) {
434 atomic_set(&panic_redirect_cpu, PANIC_CPU_INVALID);
435 pr_warn("panic: failed to redirect to CPU %d, continuing on CPU %d\n",
436 panic_force_cpu, this_cpu);
437 return false;
438 }
439
440 /* IPI/NMI sent, this CPU should stop */
441 return true;
442 }
443 #else
444 __printf(1, 0)
panic_try_force_cpu(const char * fmt,va_list args)445 static inline bool panic_try_force_cpu(const char *fmt, va_list args)
446 {
447 return false;
448 }
449 #endif /* CONFIG_SMP && CONFIG_CRASH_DUMP */
450
panic_try_start(void)451 bool panic_try_start(void)
452 {
453 int old_cpu, this_cpu;
454
455 /*
456 * Only one CPU is allowed to execute the crash_kexec() code as with
457 * panic(). Otherwise parallel calls of panic() and crash_kexec()
458 * may stop each other. To exclude them, we use panic_cpu here too.
459 */
460 old_cpu = PANIC_CPU_INVALID;
461 this_cpu = raw_smp_processor_id();
462
463 return atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu);
464 }
465 EXPORT_SYMBOL(panic_try_start);
466
panic_reset(void)467 void panic_reset(void)
468 {
469 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
470 }
471 EXPORT_SYMBOL(panic_reset);
472
panic_in_progress(void)473 bool panic_in_progress(void)
474 {
475 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
476 }
477 EXPORT_SYMBOL(panic_in_progress);
478
479 /* Return true if a panic is in progress on the current CPU. */
panic_on_this_cpu(void)480 bool panic_on_this_cpu(void)
481 {
482 /*
483 * We can use raw_smp_processor_id() here because it is impossible for
484 * the task to be migrated to the panic_cpu, or away from it. If
485 * panic_cpu has already been set, and we're not currently executing on
486 * that CPU, then we never will be.
487 */
488 return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
489 }
490 EXPORT_SYMBOL(panic_on_this_cpu);
491
492 /*
493 * Return true if a panic is in progress on a remote CPU.
494 *
495 * On true, the local CPU should immediately release any printing resources
496 * that may be needed by the panic CPU.
497 */
panic_on_other_cpu(void)498 bool panic_on_other_cpu(void)
499 {
500 return (panic_in_progress() && !panic_on_this_cpu());
501 }
502 EXPORT_SYMBOL(panic_on_other_cpu);
503
504 /*
505 * A variant of panic() called from NMI context. We return if we've already
506 * panicked on this CPU. If another CPU already panicked, loop in
507 * nmi_panic_self_stop() which can provide architecture dependent code such
508 * as saving register state for crash dump.
509 */
nmi_panic(struct pt_regs * regs,const char * msg)510 void nmi_panic(struct pt_regs *regs, const char *msg)
511 {
512 if (panic_try_start())
513 panic("%s", msg);
514 else if (panic_on_other_cpu())
515 nmi_panic_self_stop(regs);
516 }
517 EXPORT_SYMBOL(nmi_panic);
518
check_panic_on_warn(const char * origin)519 void check_panic_on_warn(const char *origin)
520 {
521 unsigned int limit;
522
523 if (panic_on_warn)
524 panic("%s: panic_on_warn set ...\n", origin);
525
526 limit = READ_ONCE(warn_limit);
527 if (atomic_inc_return(&warn_count) >= limit && limit)
528 panic("%s: system warned too often (kernel.warn_limit is %d)",
529 origin, limit);
530 }
531
panic_trigger_all_cpu_backtrace(void)532 static void panic_trigger_all_cpu_backtrace(void)
533 {
534 /* Temporary allow non-panic CPUs to write their backtraces. */
535 panic_triggering_all_cpu_backtrace = true;
536
537 if (panic_this_cpu_backtrace_printed)
538 trigger_allbutcpu_cpu_backtrace(raw_smp_processor_id());
539 else
540 trigger_all_cpu_backtrace();
541
542 panic_triggering_all_cpu_backtrace = false;
543 }
544
545 /*
546 * Helper that triggers the NMI backtrace (if set in panic_print)
547 * and then performs the secondary CPUs shutdown - we cannot have
548 * the NMI backtrace after the CPUs are off!
549 */
panic_other_cpus_shutdown(bool crash_kexec)550 static void panic_other_cpus_shutdown(bool crash_kexec)
551 {
552 if (panic_print & SYS_INFO_ALL_BT)
553 panic_trigger_all_cpu_backtrace();
554
555 /*
556 * Note that smp_send_stop() is the usual SMP shutdown function,
557 * which unfortunately may not be hardened to work in a panic
558 * situation. If we want to do crash dump after notifier calls
559 * and kmsg_dump, we will need architecture dependent extra
560 * bits in addition to stopping other CPUs, hence we rely on
561 * crash_smp_send_stop() for that.
562 */
563 if (!crash_kexec)
564 smp_send_stop();
565 else
566 crash_smp_send_stop();
567 }
568
569 /**
570 * vpanic - halt the system
571 * @fmt: The text string to print
572 * @args: Arguments for the format string
573 *
574 * Display a message, then perform cleanups. This function never returns.
575 */
vpanic(const char * fmt,va_list args)576 void vpanic(const char *fmt, va_list args)
577 {
578 static char buf[PANIC_MSG_BUFSZ];
579 long i, i_next = 0, len;
580 int state = 0;
581 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
582
583 if (panic_on_warn) {
584 /*
585 * This thread may hit another WARN() in the panic path.
586 * Resetting this prevents additional WARN() from panicking the
587 * system on this thread. Other threads are blocked by the
588 * panic_mutex in panic().
589 */
590 panic_on_warn = 0;
591 }
592
593 /*
594 * Disable local interrupts. This will prevent panic_smp_self_stop
595 * from deadlocking the first cpu that invokes the panic, since
596 * there is nothing to prevent an interrupt handler (that runs
597 * after setting panic_cpu) from invoking panic() again.
598 */
599 local_irq_disable();
600 preempt_disable_notrace();
601
602 /* Redirect panic to target CPU if configured via panic_force_cpu=. */
603 if (panic_try_force_cpu(fmt, args)) {
604 /*
605 * Mark ourselves offline so panic_other_cpus_shutdown() won't wait
606 * for us on architectures that check num_online_cpus().
607 */
608 set_cpu_online(smp_processor_id(), false);
609 panic_smp_self_stop();
610 }
611 /*
612 * It's possible to come here directly from a panic-assertion and
613 * not have preempt disabled. Some functions called from here want
614 * preempt to be disabled. No point enabling it later though...
615 *
616 * Only one CPU is allowed to execute the panic code from here. For
617 * multiple parallel invocations of panic, all other CPUs either
618 * stop themself or will wait until they are stopped by the 1st CPU
619 * with smp_send_stop().
620 *
621 * cmpxchg success means this is the 1st CPU which comes here,
622 * so go ahead.
623 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
624 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
625 */
626 /* atomic_try_cmpxchg updates old_cpu on failure */
627 if (panic_try_start()) {
628 /* go ahead */
629 } else if (panic_on_other_cpu())
630 panic_smp_self_stop();
631
632 console_verbose();
633 bust_spinlocks(1);
634 len = vscnprintf(buf, sizeof(buf), fmt, args);
635
636 if (len && buf[len - 1] == '\n')
637 buf[len - 1] = '\0';
638
639 pr_emerg("Kernel panic - not syncing: %s\n", buf);
640 /*
641 * Avoid nested stack-dumping if a panic occurs during oops processing
642 */
643 if (atomic_read(&panic_redirect_cpu) != PANIC_CPU_INVALID &&
644 panic_force_cpu == raw_smp_processor_id()) {
645 pr_emerg("panic: Redirected from CPU %d, skipping stack dump.\n",
646 atomic_read(&panic_redirect_cpu));
647 } else if (test_taint(TAINT_DIE) || oops_in_progress > 1) {
648 panic_this_cpu_backtrace_printed = true;
649 } else if (IS_ENABLED(CONFIG_DEBUG_BUGVERBOSE)) {
650 dump_stack();
651 panic_this_cpu_backtrace_printed = true;
652 }
653
654 /*
655 * If kgdb is enabled, give it a chance to run before we stop all
656 * the other CPUs or else we won't be able to debug processes left
657 * running on them.
658 */
659 kgdb_panic(buf);
660
661 /*
662 * If we have crashed and we have a crash kernel loaded let it handle
663 * everything else.
664 * If we want to run this after calling panic_notifiers, pass
665 * the "crash_kexec_post_notifiers" option to the kernel.
666 *
667 * Bypass the panic_cpu check and call __crash_kexec directly.
668 */
669 if (!_crash_kexec_post_notifiers)
670 __crash_kexec(NULL);
671
672 panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
673
674 printk_legacy_allow_panic_sync();
675
676 /*
677 * Run any panic handlers, including those that might need to
678 * add information to the kmsg dump output.
679 */
680 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
681
682 sys_info(panic_print);
683
684 kmsg_dump_desc(KMSG_DUMP_PANIC, buf);
685
686 /*
687 * If you doubt kdump always works fine in any situation,
688 * "crash_kexec_post_notifiers" offers you a chance to run
689 * panic_notifiers and dumping kmsg before kdump.
690 * Note: since some panic_notifiers can make crashed kernel
691 * more unstable, it can increase risks of the kdump failure too.
692 *
693 * Bypass the panic_cpu check and call __crash_kexec directly.
694 */
695 if (_crash_kexec_post_notifiers)
696 __crash_kexec(NULL);
697
698 console_unblank();
699
700 /*
701 * We may have ended up stopping the CPU holding the lock (in
702 * smp_send_stop()) while still having some valuable data in the console
703 * buffer. Try to acquire the lock then release it regardless of the
704 * result. The release will also print the buffers out. Locks debug
705 * should be disabled to avoid reporting bad unlock balance when
706 * panic() is not being callled from OOPS.
707 */
708 debug_locks_off();
709 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
710
711 if ((panic_print & SYS_INFO_PANIC_CONSOLE_REPLAY) ||
712 panic_console_replay)
713 console_flush_on_panic(CONSOLE_REPLAY_ALL);
714
715 if (!panic_blink)
716 panic_blink = no_blink;
717
718 if (panic_timeout > 0) {
719 /*
720 * Delay timeout seconds before rebooting the machine.
721 * We can't use the "normal" timers since we just panicked.
722 */
723 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
724
725 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
726 touch_nmi_watchdog();
727 if (i >= i_next) {
728 i += panic_blink(state ^= 1);
729 i_next = i + 3600 / PANIC_BLINK_SPD;
730 }
731 mdelay(PANIC_TIMER_STEP);
732 }
733 }
734 if (panic_timeout != 0) {
735 /*
736 * This will not be a clean reboot, with everything
737 * shutting down. But if there is a chance of
738 * rebooting the system it will be rebooted.
739 */
740 if (panic_reboot_mode != REBOOT_UNDEFINED)
741 reboot_mode = panic_reboot_mode;
742 emergency_restart();
743 }
744 #ifdef __sparc__
745 {
746 extern int stop_a_enabled;
747 /* Make sure the user can actually press Stop-A (L1-A) */
748 stop_a_enabled = 1;
749 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
750 "twice on console to return to the boot prom\n");
751 }
752 #endif
753 #if defined(CONFIG_S390)
754 disabled_wait();
755 #endif
756 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
757
758 /* Do not scroll important messages printed above */
759 suppress_printk = 1;
760
761 /*
762 * The final messages may not have been printed if in a context that
763 * defers printing (such as NMI) and irq_work is not available.
764 * Explicitly flush the kernel log buffer one last time.
765 */
766 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
767 nbcon_atomic_flush_unsafe();
768
769 local_irq_enable();
770 for (i = 0; ; i += PANIC_TIMER_STEP) {
771 touch_softlockup_watchdog();
772 if (i >= i_next) {
773 i += panic_blink(state ^= 1);
774 i_next = i + 3600 / PANIC_BLINK_SPD;
775 }
776 mdelay(PANIC_TIMER_STEP);
777 }
778 }
779 EXPORT_SYMBOL(vpanic);
780
781 /* Identical to vpanic(), except it takes variadic arguments instead of va_list */
panic(const char * fmt,...)782 void panic(const char *fmt, ...)
783 {
784 va_list args;
785
786 va_start(args, fmt);
787 vpanic(fmt, args);
788 va_end(args);
789 }
790 EXPORT_SYMBOL(panic);
791
792 #define TAINT_FLAG(taint, _c_true, _c_false) \
793 [ TAINT_##taint ] = { \
794 .c_true = _c_true, .c_false = _c_false, \
795 .desc = #taint, \
796 }
797
798 /*
799 * NOTE: if you modify the taint_flags or TAINT_FLAGS_COUNT,
800 * please also modify tools/debugging/kernel-chktaint and
801 * Documentation/admin-guide/tainted-kernels.rst, including its
802 * small shell script that prints the TAINT_FLAGS_COUNT bits of
803 * /proc/sys/kernel/tainted.
804 */
805 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
806 TAINT_FLAG(PROPRIETARY_MODULE, 'P', 'G'),
807 TAINT_FLAG(FORCED_MODULE, 'F', ' '),
808 TAINT_FLAG(CPU_OUT_OF_SPEC, 'S', ' '),
809 TAINT_FLAG(FORCED_RMMOD, 'R', ' '),
810 TAINT_FLAG(MACHINE_CHECK, 'M', ' '),
811 TAINT_FLAG(BAD_PAGE, 'B', ' '),
812 TAINT_FLAG(USER, 'U', ' '),
813 TAINT_FLAG(DIE, 'D', ' '),
814 TAINT_FLAG(OVERRIDDEN_ACPI_TABLE, 'A', ' '),
815 TAINT_FLAG(WARN, 'W', ' '),
816 TAINT_FLAG(CRAP, 'C', ' '),
817 TAINT_FLAG(FIRMWARE_WORKAROUND, 'I', ' '),
818 TAINT_FLAG(OOT_MODULE, 'O', ' '),
819 TAINT_FLAG(UNSIGNED_MODULE, 'E', ' '),
820 TAINT_FLAG(SOFTLOCKUP, 'L', ' '),
821 TAINT_FLAG(LIVEPATCH, 'K', ' '),
822 TAINT_FLAG(AUX, 'X', ' '),
823 TAINT_FLAG(RANDSTRUCT, 'T', ' '),
824 TAINT_FLAG(TEST, 'N', ' '),
825 TAINT_FLAG(FWCTL, 'J', ' '),
826 };
827
828 #undef TAINT_FLAG
829
print_tainted_seq(struct seq_buf * s,bool verbose)830 static void print_tainted_seq(struct seq_buf *s, bool verbose)
831 {
832 const char *sep = "";
833 int i;
834
835 if (!tainted_mask) {
836 seq_buf_puts(s, "Not tainted");
837 return;
838 }
839
840 seq_buf_printf(s, "Tainted: ");
841 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
842 const struct taint_flag *t = &taint_flags[i];
843 bool is_set = test_bit(i, &tainted_mask);
844 char c = is_set ? t->c_true : t->c_false;
845
846 if (verbose) {
847 if (is_set) {
848 seq_buf_printf(s, "%s[%c]=%s", sep, c, t->desc);
849 sep = ", ";
850 }
851 } else {
852 seq_buf_putc(s, c);
853 }
854 }
855 }
856
_print_tainted(bool verbose)857 static const char *_print_tainted(bool verbose)
858 {
859 /* FIXME: what should the size be? */
860 static char buf[sizeof(taint_flags)];
861 struct seq_buf s;
862
863 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
864
865 seq_buf_init(&s, buf, sizeof(buf));
866
867 print_tainted_seq(&s, verbose);
868
869 return seq_buf_str(&s);
870 }
871
872 /**
873 * print_tainted - return a string to represent the kernel taint state.
874 *
875 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
876 *
877 * The string is overwritten by the next call to print_tainted(),
878 * but is always NULL terminated.
879 */
print_tainted(void)880 const char *print_tainted(void)
881 {
882 return _print_tainted(false);
883 }
884
885 /**
886 * print_tainted_verbose - A more verbose version of print_tainted()
887 */
print_tainted_verbose(void)888 const char *print_tainted_verbose(void)
889 {
890 return _print_tainted(true);
891 }
892
test_taint(unsigned flag)893 int test_taint(unsigned flag)
894 {
895 return test_bit(flag, &tainted_mask);
896 }
897 EXPORT_SYMBOL(test_taint);
898
get_taint(void)899 unsigned long get_taint(void)
900 {
901 return tainted_mask;
902 }
903
904 /**
905 * add_taint: add a taint flag if not already set.
906 * @flag: one of the TAINT_* constants.
907 * @lockdep_ok: whether lock debugging is still OK.
908 *
909 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
910 * some notewortht-but-not-corrupting cases, it can be set to true.
911 */
add_taint(unsigned flag,enum lockdep_ok lockdep_ok)912 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
913 {
914 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
915 pr_warn("Disabling lock debugging due to kernel taint\n");
916
917 set_bit(flag, &tainted_mask);
918
919 if (tainted_mask & panic_on_taint) {
920 panic_on_taint = 0;
921 panic("panic_on_taint set ...");
922 }
923 }
924 EXPORT_SYMBOL(add_taint);
925
spin_msec(int msecs)926 static void spin_msec(int msecs)
927 {
928 int i;
929
930 for (i = 0; i < msecs; i++) {
931 touch_nmi_watchdog();
932 mdelay(1);
933 }
934 }
935
936 /*
937 * It just happens that oops_enter() and oops_exit() are identically
938 * implemented...
939 */
do_oops_enter_exit(void)940 static void do_oops_enter_exit(void)
941 {
942 unsigned long flags;
943 static int spin_counter;
944
945 if (!pause_on_oops)
946 return;
947
948 spin_lock_irqsave(&pause_on_oops_lock, flags);
949 if (pause_on_oops_flag == 0) {
950 /* This CPU may now print the oops message */
951 pause_on_oops_flag = 1;
952 } else {
953 /* We need to stall this CPU */
954 if (!spin_counter) {
955 /* This CPU gets to do the counting */
956 spin_counter = pause_on_oops;
957 do {
958 spin_unlock(&pause_on_oops_lock);
959 spin_msec(MSEC_PER_SEC);
960 spin_lock(&pause_on_oops_lock);
961 } while (--spin_counter);
962 pause_on_oops_flag = 0;
963 } else {
964 /* This CPU waits for a different one */
965 while (spin_counter) {
966 spin_unlock(&pause_on_oops_lock);
967 spin_msec(1);
968 spin_lock(&pause_on_oops_lock);
969 }
970 }
971 }
972 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
973 }
974
975 /*
976 * Return true if the calling CPU is allowed to print oops-related info.
977 * This is a bit racy..
978 */
oops_may_print(void)979 bool oops_may_print(void)
980 {
981 return pause_on_oops_flag == 0;
982 }
983
984 /*
985 * Called when the architecture enters its oops handler, before it prints
986 * anything. If this is the first CPU to oops, and it's oopsing the first
987 * time then let it proceed.
988 *
989 * This is all enabled by the pause_on_oops kernel boot option. We do all
990 * this to ensure that oopses don't scroll off the screen. It has the
991 * side-effect of preventing later-oopsing CPUs from mucking up the display,
992 * too.
993 *
994 * It turns out that the CPU which is allowed to print ends up pausing for
995 * the right duration, whereas all the other CPUs pause for twice as long:
996 * once in oops_enter(), once in oops_exit().
997 */
oops_enter(void)998 void oops_enter(void)
999 {
1000 nbcon_cpu_emergency_enter();
1001 tracing_off();
1002 /* can't trust the integrity of the kernel anymore: */
1003 debug_locks_off();
1004 do_oops_enter_exit();
1005
1006 if (sysctl_oops_all_cpu_backtrace)
1007 trigger_all_cpu_backtrace();
1008 }
1009
print_oops_end_marker(void)1010 static void print_oops_end_marker(void)
1011 {
1012 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
1013 }
1014
1015 /*
1016 * Called when the architecture exits its oops handler, after printing
1017 * everything.
1018 */
oops_exit(void)1019 void oops_exit(void)
1020 {
1021 do_oops_enter_exit();
1022 print_oops_end_marker();
1023 nbcon_cpu_emergency_exit();
1024 kmsg_dump(KMSG_DUMP_OOPS);
1025 }
1026
1027 struct warn_args {
1028 const char *fmt;
1029 va_list args;
1030 };
1031
__warn(const char * file,int line,void * caller,unsigned taint,struct pt_regs * regs,struct warn_args * args)1032 void __warn(const char *file, int line, void *caller, unsigned taint,
1033 struct pt_regs *regs, struct warn_args *args)
1034 {
1035 nbcon_cpu_emergency_enter();
1036
1037 disable_trace_on_warning();
1038
1039 if (file) {
1040 pr_warn("WARNING: %s:%d at %pS, CPU#%d: %s/%d\n",
1041 file, line, caller,
1042 raw_smp_processor_id(), current->comm, current->pid);
1043 } else {
1044 pr_warn("WARNING: at %pS, CPU#%d: %s/%d\n",
1045 caller,
1046 raw_smp_processor_id(), current->comm, current->pid);
1047 }
1048
1049 #pragma GCC diagnostic push
1050 #ifndef __clang__
1051 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
1052 #endif
1053 if (args)
1054 vprintk(args->fmt, args->args);
1055 #pragma GCC diagnostic pop
1056
1057 print_modules();
1058
1059 if (regs)
1060 show_regs(regs);
1061
1062 check_panic_on_warn("kernel");
1063
1064 if (!regs)
1065 dump_stack();
1066
1067 print_irqtrace_events(current);
1068
1069 print_oops_end_marker();
1070 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
1071
1072 /* Just a warning, don't kill lockdep. */
1073 add_taint(taint, LOCKDEP_STILL_OK);
1074
1075 nbcon_cpu_emergency_exit();
1076 }
1077
1078 #ifdef CONFIG_BUG
1079 #ifndef __WARN_FLAGS
warn_slowpath_fmt(const char * file,int line,unsigned taint,const char * fmt,...)1080 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
1081 const char *fmt, ...)
1082 {
1083 bool rcu = warn_rcu_enter();
1084 struct warn_args args;
1085
1086 pr_warn(CUT_HERE);
1087
1088 if (!fmt) {
1089 __warn(file, line, __builtin_return_address(0), taint,
1090 NULL, NULL);
1091 warn_rcu_exit(rcu);
1092 return;
1093 }
1094
1095 args.fmt = fmt;
1096 va_start(args.args, fmt);
1097 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
1098 va_end(args.args);
1099 warn_rcu_exit(rcu);
1100 }
1101 EXPORT_SYMBOL(warn_slowpath_fmt);
1102 #else
__warn_printk(const char * fmt,...)1103 void __warn_printk(const char *fmt, ...)
1104 {
1105 bool rcu = warn_rcu_enter();
1106 va_list args;
1107
1108 pr_warn(CUT_HERE);
1109
1110 va_start(args, fmt);
1111 vprintk(fmt, args);
1112 va_end(args);
1113 warn_rcu_exit(rcu);
1114 }
1115 EXPORT_SYMBOL(__warn_printk);
1116 #endif
1117
1118 /* Support resetting WARN*_ONCE state */
1119
clear_warn_once_set(void * data,u64 val)1120 static int clear_warn_once_set(void *data, u64 val)
1121 {
1122 generic_bug_clear_once();
1123 memset(__start_once, 0, __end_once - __start_once);
1124 return 0;
1125 }
1126
1127 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
1128 "%lld\n");
1129
register_warn_debugfs(void)1130 static __init int register_warn_debugfs(void)
1131 {
1132 /* Don't care about failure */
1133 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
1134 &clear_warn_once_fops);
1135 return 0;
1136 }
1137
1138 device_initcall(register_warn_debugfs);
1139 #endif
1140
1141 #ifdef CONFIG_STACKPROTECTOR
1142
1143 /*
1144 * Called when gcc's -fstack-protector feature is used, and
1145 * gcc detects corruption of the on-stack canary value
1146 */
__stack_chk_fail(void)1147 __visible noinstr void __stack_chk_fail(void)
1148 {
1149 unsigned long flags;
1150
1151 instrumentation_begin();
1152 flags = user_access_save();
1153
1154 panic("stack-protector: Kernel stack is corrupted in: %pB",
1155 __builtin_return_address(0));
1156
1157 user_access_restore(flags);
1158 instrumentation_end();
1159 }
1160 EXPORT_SYMBOL(__stack_chk_fail);
1161
1162 #endif
1163
1164 core_param(panic, panic_timeout, int, 0644);
1165 core_param(pause_on_oops, pause_on_oops, int, 0644);
1166 core_param(panic_on_warn, panic_on_warn, int, 0644);
1167 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
1168 core_param(panic_console_replay, panic_console_replay, bool, 0644);
1169
panic_print_set(const char * val,const struct kernel_param * kp)1170 static int panic_print_set(const char *val, const struct kernel_param *kp)
1171 {
1172 panic_print_deprecated();
1173 return param_set_ulong(val, kp);
1174 }
1175
panic_print_get(char * val,const struct kernel_param * kp)1176 static int panic_print_get(char *val, const struct kernel_param *kp)
1177 {
1178 return param_get_ulong(val, kp);
1179 }
1180
1181 static const struct kernel_param_ops panic_print_ops = {
1182 .set = panic_print_set,
1183 .get = panic_print_get,
1184 };
1185 __core_param_cb(panic_print, &panic_print_ops, &panic_print, 0644);
1186
oops_setup(char * s)1187 static int __init oops_setup(char *s)
1188 {
1189 if (!s)
1190 return -EINVAL;
1191 if (!strcmp(s, "panic"))
1192 panic_on_oops = 1;
1193 return 0;
1194 }
1195 early_param("oops", oops_setup);
1196
panic_on_taint_setup(char * s)1197 static int __init panic_on_taint_setup(char *s)
1198 {
1199 char *taint_str;
1200
1201 if (!s)
1202 return -EINVAL;
1203
1204 taint_str = strsep(&s, ",");
1205 if (kstrtoul(taint_str, 16, &panic_on_taint))
1206 return -EINVAL;
1207
1208 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
1209 panic_on_taint &= TAINT_FLAGS_MAX;
1210
1211 if (!panic_on_taint)
1212 return -EINVAL;
1213
1214 if (s && !strcmp(s, "nousertaint"))
1215 panic_on_taint_nousertaint = true;
1216
1217 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
1218 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
1219
1220 return 0;
1221 }
1222 early_param("panic_on_taint", panic_on_taint_setup);
1223