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 <trace/events/error_report.h>
40 #include <asm/sections.h>
41
42 #define PANIC_TIMER_STEP 100
43 #define PANIC_BLINK_SPD 18
44
45 #ifdef CONFIG_SMP
46 /*
47 * Should we dump all CPUs backtraces in an oops event?
48 * Defaults to 0, can be changed via sysctl.
49 */
50 static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
51 #else
52 #define sysctl_oops_all_cpu_backtrace 0
53 #endif /* CONFIG_SMP */
54
55 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
56 static unsigned long tainted_mask =
57 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
58 static int pause_on_oops;
59 static int pause_on_oops_flag;
60 static DEFINE_SPINLOCK(pause_on_oops_lock);
61 bool crash_kexec_post_notifiers;
62 int panic_on_warn __read_mostly;
63 unsigned long panic_on_taint;
64 bool panic_on_taint_nousertaint = false;
65 static unsigned int warn_limit __read_mostly;
66
67 bool panic_triggering_all_cpu_backtrace;
68
69 int panic_timeout = CONFIG_PANIC_TIMEOUT;
70 EXPORT_SYMBOL_GPL(panic_timeout);
71
72 #define PANIC_PRINT_TASK_INFO 0x00000001
73 #define PANIC_PRINT_MEM_INFO 0x00000002
74 #define PANIC_PRINT_TIMER_INFO 0x00000004
75 #define PANIC_PRINT_LOCK_INFO 0x00000008
76 #define PANIC_PRINT_FTRACE_INFO 0x00000010
77 #define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
78 #define PANIC_PRINT_ALL_CPU_BT 0x00000040
79 #define PANIC_PRINT_BLOCKED_TASKS 0x00000080
80 unsigned long panic_print;
81
82 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
83
84 EXPORT_SYMBOL(panic_notifier_list);
85
86 #ifdef CONFIG_SYSCTL
87 static const struct ctl_table kern_panic_table[] = {
88 #ifdef CONFIG_SMP
89 {
90 .procname = "oops_all_cpu_backtrace",
91 .data = &sysctl_oops_all_cpu_backtrace,
92 .maxlen = sizeof(int),
93 .mode = 0644,
94 .proc_handler = proc_dointvec_minmax,
95 .extra1 = SYSCTL_ZERO,
96 .extra2 = SYSCTL_ONE,
97 },
98 #endif
99 {
100 .procname = "warn_limit",
101 .data = &warn_limit,
102 .maxlen = sizeof(warn_limit),
103 .mode = 0644,
104 .proc_handler = proc_douintvec,
105 },
106 };
107
kernel_panic_sysctls_init(void)108 static __init int kernel_panic_sysctls_init(void)
109 {
110 register_sysctl_init("kernel", kern_panic_table);
111 return 0;
112 }
113 late_initcall(kernel_panic_sysctls_init);
114 #endif
115
116 static atomic_t warn_count = ATOMIC_INIT(0);
117
118 #ifdef CONFIG_SYSFS
warn_count_show(struct kobject * kobj,struct kobj_attribute * attr,char * page)119 static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
120 char *page)
121 {
122 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
123 }
124
125 static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
126
kernel_panic_sysfs_init(void)127 static __init int kernel_panic_sysfs_init(void)
128 {
129 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
130 return 0;
131 }
132 late_initcall(kernel_panic_sysfs_init);
133 #endif
134
no_blink(int state)135 static long no_blink(int state)
136 {
137 return 0;
138 }
139
140 /* Returns how long it waited in ms */
141 long (*panic_blink)(int state);
142 EXPORT_SYMBOL(panic_blink);
143
144 /*
145 * Stop ourself in panic -- architecture code may override this
146 */
panic_smp_self_stop(void)147 void __weak __noreturn panic_smp_self_stop(void)
148 {
149 while (1)
150 cpu_relax();
151 }
152
153 /*
154 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
155 * may override this to prepare for crash dumping, e.g. save regs info.
156 */
nmi_panic_self_stop(struct pt_regs * regs)157 void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
158 {
159 panic_smp_self_stop();
160 }
161
162 /*
163 * Stop other CPUs in panic. Architecture dependent code may override this
164 * with more suitable version. For example, if the architecture supports
165 * crash dump, it should save registers of each stopped CPU and disable
166 * per-CPU features such as virtualization extensions.
167 */
crash_smp_send_stop(void)168 void __weak crash_smp_send_stop(void)
169 {
170 static int cpus_stopped;
171
172 /*
173 * This function can be called twice in panic path, but obviously
174 * we execute this only once.
175 */
176 if (cpus_stopped)
177 return;
178
179 /*
180 * Note smp_send_stop is the usual smp shutdown function, which
181 * unfortunately means it may not be hardened to work in a panic
182 * situation.
183 */
184 smp_send_stop();
185 cpus_stopped = 1;
186 }
187
188 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
189
190 /*
191 * A variant of panic() called from NMI context. We return if we've already
192 * panicked on this CPU. If another CPU already panicked, loop in
193 * nmi_panic_self_stop() which can provide architecture dependent code such
194 * as saving register state for crash dump.
195 */
nmi_panic(struct pt_regs * regs,const char * msg)196 void nmi_panic(struct pt_regs *regs, const char *msg)
197 {
198 int old_cpu, this_cpu;
199
200 old_cpu = PANIC_CPU_INVALID;
201 this_cpu = raw_smp_processor_id();
202
203 /* atomic_try_cmpxchg updates old_cpu on failure */
204 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu))
205 panic("%s", msg);
206 else if (old_cpu != this_cpu)
207 nmi_panic_self_stop(regs);
208 }
209 EXPORT_SYMBOL(nmi_panic);
210
panic_print_sys_info(bool console_flush)211 static void panic_print_sys_info(bool console_flush)
212 {
213 if (console_flush) {
214 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
215 console_flush_on_panic(CONSOLE_REPLAY_ALL);
216 return;
217 }
218
219 if (panic_print & PANIC_PRINT_TASK_INFO)
220 show_state();
221
222 if (panic_print & PANIC_PRINT_MEM_INFO)
223 show_mem();
224
225 if (panic_print & PANIC_PRINT_TIMER_INFO)
226 sysrq_timer_list_show();
227
228 if (panic_print & PANIC_PRINT_LOCK_INFO)
229 debug_show_all_locks();
230
231 if (panic_print & PANIC_PRINT_FTRACE_INFO)
232 ftrace_dump(DUMP_ALL);
233
234 if (panic_print & PANIC_PRINT_BLOCKED_TASKS)
235 show_state_filter(TASK_UNINTERRUPTIBLE);
236 }
237
check_panic_on_warn(const char * origin)238 void check_panic_on_warn(const char *origin)
239 {
240 unsigned int limit;
241
242 if (panic_on_warn)
243 panic("%s: panic_on_warn set ...\n", origin);
244
245 limit = READ_ONCE(warn_limit);
246 if (atomic_inc_return(&warn_count) >= limit && limit)
247 panic("%s: system warned too often (kernel.warn_limit is %d)",
248 origin, limit);
249 }
250
251 /*
252 * Helper that triggers the NMI backtrace (if set in panic_print)
253 * and then performs the secondary CPUs shutdown - we cannot have
254 * the NMI backtrace after the CPUs are off!
255 */
panic_other_cpus_shutdown(bool crash_kexec)256 static void panic_other_cpus_shutdown(bool crash_kexec)
257 {
258 if (panic_print & PANIC_PRINT_ALL_CPU_BT) {
259 /* Temporary allow non-panic CPUs to write their backtraces. */
260 panic_triggering_all_cpu_backtrace = true;
261 trigger_all_cpu_backtrace();
262 panic_triggering_all_cpu_backtrace = false;
263 }
264
265 /*
266 * Note that smp_send_stop() is the usual SMP shutdown function,
267 * which unfortunately may not be hardened to work in a panic
268 * situation. If we want to do crash dump after notifier calls
269 * and kmsg_dump, we will need architecture dependent extra
270 * bits in addition to stopping other CPUs, hence we rely on
271 * crash_smp_send_stop() for that.
272 */
273 if (!crash_kexec)
274 smp_send_stop();
275 else
276 crash_smp_send_stop();
277 }
278
279 /**
280 * panic - halt the system
281 * @fmt: The text string to print
282 *
283 * Display a message, then perform cleanups.
284 *
285 * This function never returns.
286 */
panic(const char * fmt,...)287 void panic(const char *fmt, ...)
288 {
289 static char buf[1024];
290 va_list args;
291 long i, i_next = 0, len;
292 int state = 0;
293 int old_cpu, this_cpu;
294 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
295
296 if (panic_on_warn) {
297 /*
298 * This thread may hit another WARN() in the panic path.
299 * Resetting this prevents additional WARN() from panicking the
300 * system on this thread. Other threads are blocked by the
301 * panic_mutex in panic().
302 */
303 panic_on_warn = 0;
304 }
305
306 /*
307 * Disable local interrupts. This will prevent panic_smp_self_stop
308 * from deadlocking the first cpu that invokes the panic, since
309 * there is nothing to prevent an interrupt handler (that runs
310 * after setting panic_cpu) from invoking panic() again.
311 */
312 local_irq_disable();
313 preempt_disable_notrace();
314
315 /*
316 * It's possible to come here directly from a panic-assertion and
317 * not have preempt disabled. Some functions called from here want
318 * preempt to be disabled. No point enabling it later though...
319 *
320 * Only one CPU is allowed to execute the panic code from here. For
321 * multiple parallel invocations of panic, all other CPUs either
322 * stop themself or will wait until they are stopped by the 1st CPU
323 * with smp_send_stop().
324 *
325 * cmpxchg success means this is the 1st CPU which comes here,
326 * so go ahead.
327 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
328 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
329 */
330 old_cpu = PANIC_CPU_INVALID;
331 this_cpu = raw_smp_processor_id();
332
333 /* atomic_try_cmpxchg updates old_cpu on failure */
334 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
335 /* go ahead */
336 } else if (old_cpu != this_cpu)
337 panic_smp_self_stop();
338
339 console_verbose();
340 bust_spinlocks(1);
341 va_start(args, fmt);
342 len = vscnprintf(buf, sizeof(buf), fmt, args);
343 va_end(args);
344
345 if (len && buf[len - 1] == '\n')
346 buf[len - 1] = '\0';
347
348 pr_emerg("Kernel panic - not syncing: %s\n", buf);
349 #ifdef CONFIG_DEBUG_BUGVERBOSE
350 /*
351 * Avoid nested stack-dumping if a panic occurs during oops processing
352 */
353 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
354 dump_stack();
355 #endif
356
357 /*
358 * If kgdb is enabled, give it a chance to run before we stop all
359 * the other CPUs or else we won't be able to debug processes left
360 * running on them.
361 */
362 kgdb_panic(buf);
363
364 /*
365 * If we have crashed and we have a crash kernel loaded let it handle
366 * everything else.
367 * If we want to run this after calling panic_notifiers, pass
368 * the "crash_kexec_post_notifiers" option to the kernel.
369 *
370 * Bypass the panic_cpu check and call __crash_kexec directly.
371 */
372 if (!_crash_kexec_post_notifiers)
373 __crash_kexec(NULL);
374
375 panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
376
377 printk_legacy_allow_panic_sync();
378
379 /*
380 * Run any panic handlers, including those that might need to
381 * add information to the kmsg dump output.
382 */
383 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
384
385 panic_print_sys_info(false);
386
387 kmsg_dump_desc(KMSG_DUMP_PANIC, buf);
388
389 /*
390 * If you doubt kdump always works fine in any situation,
391 * "crash_kexec_post_notifiers" offers you a chance to run
392 * panic_notifiers and dumping kmsg before kdump.
393 * Note: since some panic_notifiers can make crashed kernel
394 * more unstable, it can increase risks of the kdump failure too.
395 *
396 * Bypass the panic_cpu check and call __crash_kexec directly.
397 */
398 if (_crash_kexec_post_notifiers)
399 __crash_kexec(NULL);
400
401 console_unblank();
402
403 /*
404 * We may have ended up stopping the CPU holding the lock (in
405 * smp_send_stop()) while still having some valuable data in the console
406 * buffer. Try to acquire the lock then release it regardless of the
407 * result. The release will also print the buffers out. Locks debug
408 * should be disabled to avoid reporting bad unlock balance when
409 * panic() is not being callled from OOPS.
410 */
411 debug_locks_off();
412 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
413
414 panic_print_sys_info(true);
415
416 if (!panic_blink)
417 panic_blink = no_blink;
418
419 if (panic_timeout > 0) {
420 /*
421 * Delay timeout seconds before rebooting the machine.
422 * We can't use the "normal" timers since we just panicked.
423 */
424 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
425
426 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
427 touch_nmi_watchdog();
428 if (i >= i_next) {
429 i += panic_blink(state ^= 1);
430 i_next = i + 3600 / PANIC_BLINK_SPD;
431 }
432 mdelay(PANIC_TIMER_STEP);
433 }
434 }
435 if (panic_timeout != 0) {
436 /*
437 * This will not be a clean reboot, with everything
438 * shutting down. But if there is a chance of
439 * rebooting the system it will be rebooted.
440 */
441 if (panic_reboot_mode != REBOOT_UNDEFINED)
442 reboot_mode = panic_reboot_mode;
443 emergency_restart();
444 }
445 #ifdef __sparc__
446 {
447 extern int stop_a_enabled;
448 /* Make sure the user can actually press Stop-A (L1-A) */
449 stop_a_enabled = 1;
450 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
451 "twice on console to return to the boot prom\n");
452 }
453 #endif
454 #if defined(CONFIG_S390)
455 disabled_wait();
456 #endif
457 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
458
459 /* Do not scroll important messages printed above */
460 suppress_printk = 1;
461
462 /*
463 * The final messages may not have been printed if in a context that
464 * defers printing (such as NMI) and irq_work is not available.
465 * Explicitly flush the kernel log buffer one last time.
466 */
467 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
468 nbcon_atomic_flush_unsafe();
469
470 local_irq_enable();
471 for (i = 0; ; i += PANIC_TIMER_STEP) {
472 touch_softlockup_watchdog();
473 if (i >= i_next) {
474 i += panic_blink(state ^= 1);
475 i_next = i + 3600 / PANIC_BLINK_SPD;
476 }
477 mdelay(PANIC_TIMER_STEP);
478 }
479 }
480
481 EXPORT_SYMBOL(panic);
482
483 #define TAINT_FLAG(taint, _c_true, _c_false, _module) \
484 [ TAINT_##taint ] = { \
485 .c_true = _c_true, .c_false = _c_false, \
486 .module = _module, \
487 .desc = #taint, \
488 }
489
490 /*
491 * TAINT_FORCED_RMMOD could be a per-module flag but the module
492 * is being removed anyway.
493 */
494 const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
495 TAINT_FLAG(PROPRIETARY_MODULE, 'P', 'G', true),
496 TAINT_FLAG(FORCED_MODULE, 'F', ' ', true),
497 TAINT_FLAG(CPU_OUT_OF_SPEC, 'S', ' ', false),
498 TAINT_FLAG(FORCED_RMMOD, 'R', ' ', false),
499 TAINT_FLAG(MACHINE_CHECK, 'M', ' ', false),
500 TAINT_FLAG(BAD_PAGE, 'B', ' ', false),
501 TAINT_FLAG(USER, 'U', ' ', false),
502 TAINT_FLAG(DIE, 'D', ' ', false),
503 TAINT_FLAG(OVERRIDDEN_ACPI_TABLE, 'A', ' ', false),
504 TAINT_FLAG(WARN, 'W', ' ', false),
505 TAINT_FLAG(CRAP, 'C', ' ', true),
506 TAINT_FLAG(FIRMWARE_WORKAROUND, 'I', ' ', false),
507 TAINT_FLAG(OOT_MODULE, 'O', ' ', true),
508 TAINT_FLAG(UNSIGNED_MODULE, 'E', ' ', true),
509 TAINT_FLAG(SOFTLOCKUP, 'L', ' ', false),
510 TAINT_FLAG(LIVEPATCH, 'K', ' ', true),
511 TAINT_FLAG(AUX, 'X', ' ', true),
512 TAINT_FLAG(RANDSTRUCT, 'T', ' ', true),
513 TAINT_FLAG(TEST, 'N', ' ', true),
514 TAINT_FLAG(FWCTL, 'J', ' ', true),
515 };
516
517 #undef TAINT_FLAG
518
print_tainted_seq(struct seq_buf * s,bool verbose)519 static void print_tainted_seq(struct seq_buf *s, bool verbose)
520 {
521 const char *sep = "";
522 int i;
523
524 if (!tainted_mask) {
525 seq_buf_puts(s, "Not tainted");
526 return;
527 }
528
529 seq_buf_printf(s, "Tainted: ");
530 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
531 const struct taint_flag *t = &taint_flags[i];
532 bool is_set = test_bit(i, &tainted_mask);
533 char c = is_set ? t->c_true : t->c_false;
534
535 if (verbose) {
536 if (is_set) {
537 seq_buf_printf(s, "%s[%c]=%s", sep, c, t->desc);
538 sep = ", ";
539 }
540 } else {
541 seq_buf_putc(s, c);
542 }
543 }
544 }
545
_print_tainted(bool verbose)546 static const char *_print_tainted(bool verbose)
547 {
548 /* FIXME: what should the size be? */
549 static char buf[sizeof(taint_flags)];
550 struct seq_buf s;
551
552 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
553
554 seq_buf_init(&s, buf, sizeof(buf));
555
556 print_tainted_seq(&s, verbose);
557
558 return seq_buf_str(&s);
559 }
560
561 /**
562 * print_tainted - return a string to represent the kernel taint state.
563 *
564 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
565 *
566 * The string is overwritten by the next call to print_tainted(),
567 * but is always NULL terminated.
568 */
print_tainted(void)569 const char *print_tainted(void)
570 {
571 return _print_tainted(false);
572 }
573
574 /**
575 * print_tainted_verbose - A more verbose version of print_tainted()
576 */
print_tainted_verbose(void)577 const char *print_tainted_verbose(void)
578 {
579 return _print_tainted(true);
580 }
581
test_taint(unsigned flag)582 int test_taint(unsigned flag)
583 {
584 return test_bit(flag, &tainted_mask);
585 }
586 EXPORT_SYMBOL(test_taint);
587
get_taint(void)588 unsigned long get_taint(void)
589 {
590 return tainted_mask;
591 }
592
593 /**
594 * add_taint: add a taint flag if not already set.
595 * @flag: one of the TAINT_* constants.
596 * @lockdep_ok: whether lock debugging is still OK.
597 *
598 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
599 * some notewortht-but-not-corrupting cases, it can be set to true.
600 */
add_taint(unsigned flag,enum lockdep_ok lockdep_ok)601 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
602 {
603 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
604 pr_warn("Disabling lock debugging due to kernel taint\n");
605
606 set_bit(flag, &tainted_mask);
607
608 if (tainted_mask & panic_on_taint) {
609 panic_on_taint = 0;
610 panic("panic_on_taint set ...");
611 }
612 }
613 EXPORT_SYMBOL(add_taint);
614
spin_msec(int msecs)615 static void spin_msec(int msecs)
616 {
617 int i;
618
619 for (i = 0; i < msecs; i++) {
620 touch_nmi_watchdog();
621 mdelay(1);
622 }
623 }
624
625 /*
626 * It just happens that oops_enter() and oops_exit() are identically
627 * implemented...
628 */
do_oops_enter_exit(void)629 static void do_oops_enter_exit(void)
630 {
631 unsigned long flags;
632 static int spin_counter;
633
634 if (!pause_on_oops)
635 return;
636
637 spin_lock_irqsave(&pause_on_oops_lock, flags);
638 if (pause_on_oops_flag == 0) {
639 /* This CPU may now print the oops message */
640 pause_on_oops_flag = 1;
641 } else {
642 /* We need to stall this CPU */
643 if (!spin_counter) {
644 /* This CPU gets to do the counting */
645 spin_counter = pause_on_oops;
646 do {
647 spin_unlock(&pause_on_oops_lock);
648 spin_msec(MSEC_PER_SEC);
649 spin_lock(&pause_on_oops_lock);
650 } while (--spin_counter);
651 pause_on_oops_flag = 0;
652 } else {
653 /* This CPU waits for a different one */
654 while (spin_counter) {
655 spin_unlock(&pause_on_oops_lock);
656 spin_msec(1);
657 spin_lock(&pause_on_oops_lock);
658 }
659 }
660 }
661 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
662 }
663
664 /*
665 * Return true if the calling CPU is allowed to print oops-related info.
666 * This is a bit racy..
667 */
oops_may_print(void)668 bool oops_may_print(void)
669 {
670 return pause_on_oops_flag == 0;
671 }
672
673 /*
674 * Called when the architecture enters its oops handler, before it prints
675 * anything. If this is the first CPU to oops, and it's oopsing the first
676 * time then let it proceed.
677 *
678 * This is all enabled by the pause_on_oops kernel boot option. We do all
679 * this to ensure that oopses don't scroll off the screen. It has the
680 * side-effect of preventing later-oopsing CPUs from mucking up the display,
681 * too.
682 *
683 * It turns out that the CPU which is allowed to print ends up pausing for
684 * the right duration, whereas all the other CPUs pause for twice as long:
685 * once in oops_enter(), once in oops_exit().
686 */
oops_enter(void)687 void oops_enter(void)
688 {
689 nbcon_cpu_emergency_enter();
690 tracing_off();
691 /* can't trust the integrity of the kernel anymore: */
692 debug_locks_off();
693 do_oops_enter_exit();
694
695 if (sysctl_oops_all_cpu_backtrace)
696 trigger_all_cpu_backtrace();
697 }
698
print_oops_end_marker(void)699 static void print_oops_end_marker(void)
700 {
701 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
702 }
703
704 /*
705 * Called when the architecture exits its oops handler, after printing
706 * everything.
707 */
oops_exit(void)708 void oops_exit(void)
709 {
710 do_oops_enter_exit();
711 print_oops_end_marker();
712 nbcon_cpu_emergency_exit();
713 kmsg_dump(KMSG_DUMP_OOPS);
714 }
715
716 struct warn_args {
717 const char *fmt;
718 va_list args;
719 };
720
__warn(const char * file,int line,void * caller,unsigned taint,struct pt_regs * regs,struct warn_args * args)721 void __warn(const char *file, int line, void *caller, unsigned taint,
722 struct pt_regs *regs, struct warn_args *args)
723 {
724 nbcon_cpu_emergency_enter();
725
726 disable_trace_on_warning();
727
728 if (file)
729 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
730 raw_smp_processor_id(), current->pid, file, line,
731 caller);
732 else
733 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
734 raw_smp_processor_id(), current->pid, caller);
735
736 #pragma GCC diagnostic push
737 #ifndef __clang__
738 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
739 #endif
740 if (args)
741 vprintk(args->fmt, args->args);
742 #pragma GCC diagnostic pop
743
744 print_modules();
745
746 if (regs)
747 show_regs(regs);
748
749 check_panic_on_warn("kernel");
750
751 if (!regs)
752 dump_stack();
753
754 print_irqtrace_events(current);
755
756 print_oops_end_marker();
757 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
758
759 /* Just a warning, don't kill lockdep. */
760 add_taint(taint, LOCKDEP_STILL_OK);
761
762 nbcon_cpu_emergency_exit();
763 }
764
765 #ifdef CONFIG_BUG
766 #ifndef __WARN_FLAGS
warn_slowpath_fmt(const char * file,int line,unsigned taint,const char * fmt,...)767 void warn_slowpath_fmt(const char *file, int line, unsigned taint,
768 const char *fmt, ...)
769 {
770 bool rcu = warn_rcu_enter();
771 struct warn_args args;
772
773 pr_warn(CUT_HERE);
774
775 if (!fmt) {
776 __warn(file, line, __builtin_return_address(0), taint,
777 NULL, NULL);
778 warn_rcu_exit(rcu);
779 return;
780 }
781
782 args.fmt = fmt;
783 va_start(args.args, fmt);
784 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
785 va_end(args.args);
786 warn_rcu_exit(rcu);
787 }
788 EXPORT_SYMBOL(warn_slowpath_fmt);
789 #else
__warn_printk(const char * fmt,...)790 void __warn_printk(const char *fmt, ...)
791 {
792 bool rcu = warn_rcu_enter();
793 va_list args;
794
795 pr_warn(CUT_HERE);
796
797 va_start(args, fmt);
798 vprintk(fmt, args);
799 va_end(args);
800 warn_rcu_exit(rcu);
801 }
802 EXPORT_SYMBOL(__warn_printk);
803 #endif
804
805 /* Support resetting WARN*_ONCE state */
806
clear_warn_once_set(void * data,u64 val)807 static int clear_warn_once_set(void *data, u64 val)
808 {
809 generic_bug_clear_once();
810 memset(__start_once, 0, __end_once - __start_once);
811 return 0;
812 }
813
814 DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
815 "%lld\n");
816
register_warn_debugfs(void)817 static __init int register_warn_debugfs(void)
818 {
819 /* Don't care about failure */
820 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
821 &clear_warn_once_fops);
822 return 0;
823 }
824
825 device_initcall(register_warn_debugfs);
826 #endif
827
828 #ifdef CONFIG_STACKPROTECTOR
829
830 /*
831 * Called when gcc's -fstack-protector feature is used, and
832 * gcc detects corruption of the on-stack canary value
833 */
__stack_chk_fail(void)834 __visible noinstr void __stack_chk_fail(void)
835 {
836 unsigned long flags;
837
838 instrumentation_begin();
839 flags = user_access_save();
840
841 panic("stack-protector: Kernel stack is corrupted in: %pB",
842 __builtin_return_address(0));
843
844 user_access_restore(flags);
845 instrumentation_end();
846 }
847 EXPORT_SYMBOL(__stack_chk_fail);
848
849 #endif
850
851 core_param(panic, panic_timeout, int, 0644);
852 core_param(panic_print, panic_print, ulong, 0644);
853 core_param(pause_on_oops, pause_on_oops, int, 0644);
854 core_param(panic_on_warn, panic_on_warn, int, 0644);
855 core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
856
oops_setup(char * s)857 static int __init oops_setup(char *s)
858 {
859 if (!s)
860 return -EINVAL;
861 if (!strcmp(s, "panic"))
862 panic_on_oops = 1;
863 return 0;
864 }
865 early_param("oops", oops_setup);
866
panic_on_taint_setup(char * s)867 static int __init panic_on_taint_setup(char *s)
868 {
869 char *taint_str;
870
871 if (!s)
872 return -EINVAL;
873
874 taint_str = strsep(&s, ",");
875 if (kstrtoul(taint_str, 16, &panic_on_taint))
876 return -EINVAL;
877
878 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
879 panic_on_taint &= TAINT_FLAGS_MAX;
880
881 if (!panic_on_taint)
882 return -EINVAL;
883
884 if (s && !strcmp(s, "nousertaint"))
885 panic_on_taint_nousertaint = true;
886
887 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
888 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
889
890 return 0;
891 }
892 early_param("panic_on_taint", panic_on_taint_setup);
893