projects/linux·blob v6.19-rc1 · refs

1062 lines · 26.4 KiB

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