projects/linux·blob v5.17-rc6 · refs

719 lines · 18.1 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/sysrq.h>
29#include <linux/init.h>
30#include <linux/nmi.h>
31#include <linux/console.h>
32#include <linux/bug.h>
33#include <linux/ratelimit.h>
34#include <linux/debugfs.h>
35#include <trace/events/error_report.h>
36#include <asm/sections.h>
37
38#define PANIC_TIMER_STEP 100
39#define PANIC_BLINK_SPD 18
40
41#ifdef CONFIG_SMP
42/*
43 * Should we dump all CPUs backtraces in an oops event?
44 * Defaults to 0, can be changed via sysctl.
45 */
46unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
47#endif /* CONFIG_SMP */
48
49int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
50static unsigned long tainted_mask =
51 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
52static int pause_on_oops;
53static int pause_on_oops_flag;
54static DEFINE_SPINLOCK(pause_on_oops_lock);
55bool crash_kexec_post_notifiers;
56int panic_on_warn __read_mostly;
57unsigned long panic_on_taint;
58bool panic_on_taint_nousertaint = false;
59
60int panic_timeout = CONFIG_PANIC_TIMEOUT;
61EXPORT_SYMBOL_GPL(panic_timeout);
62
63#define PANIC_PRINT_TASK_INFO 0x00000001
64#define PANIC_PRINT_MEM_INFO 0x00000002
65#define PANIC_PRINT_TIMER_INFO 0x00000004
66#define PANIC_PRINT_LOCK_INFO 0x00000008
67#define PANIC_PRINT_FTRACE_INFO 0x00000010
68#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
69unsigned long panic_print;
70
71ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
72
73EXPORT_SYMBOL(panic_notifier_list);
74
75static long no_blink(int state)
76{
77 return 0;
78}
79
80/* Returns how long it waited in ms */
81long (*panic_blink)(int state);
82EXPORT_SYMBOL(panic_blink);
83
84/*
85 * Stop ourself in panic -- architecture code may override this
86 */
87void __weak panic_smp_self_stop(void)
88{
89 while (1)
90 cpu_relax();
91}
92
93/*
94 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
95 * may override this to prepare for crash dumping, e.g. save regs info.
96 */
97void __weak nmi_panic_self_stop(struct pt_regs *regs)
98{
99 panic_smp_self_stop();
100}
101
102/*
103 * Stop other CPUs in panic. Architecture dependent code may override this
104 * with more suitable version. For example, if the architecture supports
105 * crash dump, it should save registers of each stopped CPU and disable
106 * per-CPU features such as virtualization extensions.
107 */
108void __weak crash_smp_send_stop(void)
109{
110 static int cpus_stopped;
111
112 /*
113 * This function can be called twice in panic path, but obviously
114 * we execute this only once.
115 */
116 if (cpus_stopped)
117 return;
118
119 /*
120 * Note smp_send_stop is the usual smp shutdown function, which
121 * unfortunately means it may not be hardened to work in a panic
122 * situation.
123 */
124 smp_send_stop();
125 cpus_stopped = 1;
126}
127
128atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
129
130/*
131 * A variant of panic() called from NMI context. We return if we've already
132 * panicked on this CPU. If another CPU already panicked, loop in
133 * nmi_panic_self_stop() which can provide architecture dependent code such
134 * as saving register state for crash dump.
135 */
136void nmi_panic(struct pt_regs *regs, const char *msg)
137{
138 int old_cpu, cpu;
139
140 cpu = raw_smp_processor_id();
141 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
142
143 if (old_cpu == PANIC_CPU_INVALID)
144 panic("%s", msg);
145 else if (old_cpu != cpu)
146 nmi_panic_self_stop(regs);
147}
148EXPORT_SYMBOL(nmi_panic);
149
150static void panic_print_sys_info(void)
151{
152 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
153 console_flush_on_panic(CONSOLE_REPLAY_ALL);
154
155 if (panic_print & PANIC_PRINT_TASK_INFO)
156 show_state();
157
158 if (panic_print & PANIC_PRINT_MEM_INFO)
159 show_mem(0, NULL);
160
161 if (panic_print & PANIC_PRINT_TIMER_INFO)
162 sysrq_timer_list_show();
163
164 if (panic_print & PANIC_PRINT_LOCK_INFO)
165 debug_show_all_locks();
166
167 if (panic_print & PANIC_PRINT_FTRACE_INFO)
168 ftrace_dump(DUMP_ALL);
169}
170
171/**
172 * panic - halt the system
173 * @fmt: The text string to print
174 *
175 * Display a message, then perform cleanups.
176 *
177 * This function never returns.
178 */
179void panic(const char *fmt, ...)
180{
181 static char buf[1024];
182 va_list args;
183 long i, i_next = 0, len;
184 int state = 0;
185 int old_cpu, this_cpu;
186 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
187
188 /*
189 * Disable local interrupts. This will prevent panic_smp_self_stop
190 * from deadlocking the first cpu that invokes the panic, since
191 * there is nothing to prevent an interrupt handler (that runs
192 * after setting panic_cpu) from invoking panic() again.
193 */
194 local_irq_disable();
195 preempt_disable_notrace();
196
197 /*
198 * It's possible to come here directly from a panic-assertion and
199 * not have preempt disabled. Some functions called from here want
200 * preempt to be disabled. No point enabling it later though...
201 *
202 * Only one CPU is allowed to execute the panic code from here. For
203 * multiple parallel invocations of panic, all other CPUs either
204 * stop themself or will wait until they are stopped by the 1st CPU
205 * with smp_send_stop().
206 *
207 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
208 * comes here, so go ahead.
209 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
210 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
211 */
212 this_cpu = raw_smp_processor_id();
213 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
214
215 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
216 panic_smp_self_stop();
217
218 console_verbose();
219 bust_spinlocks(1);
220 va_start(args, fmt);
221 len = vscnprintf(buf, sizeof(buf), fmt, args);
222 va_end(args);
223
224 if (len && buf[len - 1] == '\n')
225 buf[len - 1] = '\0';
226
227 pr_emerg("Kernel panic - not syncing: %s\n", buf);
228#ifdef CONFIG_DEBUG_BUGVERBOSE
229 /*
230 * Avoid nested stack-dumping if a panic occurs during oops processing
231 */
232 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
233 dump_stack();
234#endif
235
236 /*
237 * If kgdb is enabled, give it a chance to run before we stop all
238 * the other CPUs or else we won't be able to debug processes left
239 * running on them.
240 */
241 kgdb_panic(buf);
242
243 /*
244 * If we have crashed and we have a crash kernel loaded let it handle
245 * everything else.
246 * If we want to run this after calling panic_notifiers, pass
247 * the "crash_kexec_post_notifiers" option to the kernel.
248 *
249 * Bypass the panic_cpu check and call __crash_kexec directly.
250 */
251 if (!_crash_kexec_post_notifiers) {
252 __crash_kexec(NULL);
253
254 /*
255 * Note smp_send_stop is the usual smp shutdown function, which
256 * unfortunately means it may not be hardened to work in a
257 * panic situation.
258 */
259 smp_send_stop();
260 } else {
261 /*
262 * If we want to do crash dump after notifier calls and
263 * kmsg_dump, we will need architecture dependent extra
264 * works in addition to stopping other CPUs.
265 */
266 crash_smp_send_stop();
267 }
268
269 /*
270 * Run any panic handlers, including those that might need to
271 * add information to the kmsg dump output.
272 */
273 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
274
275 kmsg_dump(KMSG_DUMP_PANIC);
276
277 /*
278 * If you doubt kdump always works fine in any situation,
279 * "crash_kexec_post_notifiers" offers you a chance to run
280 * panic_notifiers and dumping kmsg before kdump.
281 * Note: since some panic_notifiers can make crashed kernel
282 * more unstable, it can increase risks of the kdump failure too.
283 *
284 * Bypass the panic_cpu check and call __crash_kexec directly.
285 */
286 if (_crash_kexec_post_notifiers)
287 __crash_kexec(NULL);
288
289#ifdef CONFIG_VT
290 unblank_screen();
291#endif
292 console_unblank();
293
294 /*
295 * We may have ended up stopping the CPU holding the lock (in
296 * smp_send_stop()) while still having some valuable data in the console
297 * buffer. Try to acquire the lock then release it regardless of the
298 * result. The release will also print the buffers out. Locks debug
299 * should be disabled to avoid reporting bad unlock balance when
300 * panic() is not being callled from OOPS.
301 */
302 debug_locks_off();
303 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
304
305 panic_print_sys_info();
306
307 if (!panic_blink)
308 panic_blink = no_blink;
309
310 if (panic_timeout > 0) {
311 /*
312 * Delay timeout seconds before rebooting the machine.
313 * We can't use the "normal" timers since we just panicked.
314 */
315 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
316
317 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
318 touch_nmi_watchdog();
319 if (i >= i_next) {
320 i += panic_blink(state ^= 1);
321 i_next = i + 3600 / PANIC_BLINK_SPD;
322 }
323 mdelay(PANIC_TIMER_STEP);
324 }
325 }
326 if (panic_timeout != 0) {
327 /*
328 * This will not be a clean reboot, with everything
329 * shutting down. But if there is a chance of
330 * rebooting the system it will be rebooted.
331 */
332 if (panic_reboot_mode != REBOOT_UNDEFINED)
333 reboot_mode = panic_reboot_mode;
334 emergency_restart();
335 }
336#ifdef __sparc__
337 {
338 extern int stop_a_enabled;
339 /* Make sure the user can actually press Stop-A (L1-A) */
340 stop_a_enabled = 1;
341 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
342 "twice on console to return to the boot prom\n");
343 }
344#endif
345#if defined(CONFIG_S390)
346 disabled_wait();
347#endif
348 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
349
350 /* Do not scroll important messages printed above */
351 suppress_printk = 1;
352 local_irq_enable();
353 for (i = 0; ; i += PANIC_TIMER_STEP) {
354 touch_softlockup_watchdog();
355 if (i >= i_next) {
356 i += panic_blink(state ^= 1);
357 i_next = i + 3600 / PANIC_BLINK_SPD;
358 }
359 mdelay(PANIC_TIMER_STEP);
360 }
361}
362
363EXPORT_SYMBOL(panic);
364
365/*
366 * TAINT_FORCED_RMMOD could be a per-module flag but the module
367 * is being removed anyway.
368 */
369const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
370 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
371 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
372 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
373 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
374 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
375 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
376 [ TAINT_USER ] = { 'U', ' ', false },
377 [ TAINT_DIE ] = { 'D', ' ', false },
378 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
379 [ TAINT_WARN ] = { 'W', ' ', false },
380 [ TAINT_CRAP ] = { 'C', ' ', true },
381 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
382 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
383 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
384 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
385 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
386 [ TAINT_AUX ] = { 'X', ' ', true },
387 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
388};
389
390/**
391 * print_tainted - return a string to represent the kernel taint state.
392 *
393 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
394 *
395 * The string is overwritten by the next call to print_tainted(),
396 * but is always NULL terminated.
397 */
398const char *print_tainted(void)
399{
400 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
401
402 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
403
404 if (tainted_mask) {
405 char *s;
406 int i;
407
408 s = buf + sprintf(buf, "Tainted: ");
409 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
410 const struct taint_flag *t = &taint_flags[i];
411 *s++ = test_bit(i, &tainted_mask) ?
412 t->c_true : t->c_false;
413 }
414 *s = 0;
415 } else
416 snprintf(buf, sizeof(buf), "Not tainted");
417
418 return buf;
419}
420
421int test_taint(unsigned flag)
422{
423 return test_bit(flag, &tainted_mask);
424}
425EXPORT_SYMBOL(test_taint);
426
427unsigned long get_taint(void)
428{
429 return tainted_mask;
430}
431
432/**
433 * add_taint: add a taint flag if not already set.
434 * @flag: one of the TAINT_* constants.
435 * @lockdep_ok: whether lock debugging is still OK.
436 *
437 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
438 * some notewortht-but-not-corrupting cases, it can be set to true.
439 */
440void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
441{
442 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
443 pr_warn("Disabling lock debugging due to kernel taint\n");
444
445 set_bit(flag, &tainted_mask);
446
447 if (tainted_mask & panic_on_taint) {
448 panic_on_taint = 0;
449 panic("panic_on_taint set ...");
450 }
451}
452EXPORT_SYMBOL(add_taint);
453
454static void spin_msec(int msecs)
455{
456 int i;
457
458 for (i = 0; i < msecs; i++) {
459 touch_nmi_watchdog();
460 mdelay(1);
461 }
462}
463
464/*
465 * It just happens that oops_enter() and oops_exit() are identically
466 * implemented...
467 */
468static void do_oops_enter_exit(void)
469{
470 unsigned long flags;
471 static int spin_counter;
472
473 if (!pause_on_oops)
474 return;
475
476 spin_lock_irqsave(&pause_on_oops_lock, flags);
477 if (pause_on_oops_flag == 0) {
478 /* This CPU may now print the oops message */
479 pause_on_oops_flag = 1;
480 } else {
481 /* We need to stall this CPU */
482 if (!spin_counter) {
483 /* This CPU gets to do the counting */
484 spin_counter = pause_on_oops;
485 do {
486 spin_unlock(&pause_on_oops_lock);
487 spin_msec(MSEC_PER_SEC);
488 spin_lock(&pause_on_oops_lock);
489 } while (--spin_counter);
490 pause_on_oops_flag = 0;
491 } else {
492 /* This CPU waits for a different one */
493 while (spin_counter) {
494 spin_unlock(&pause_on_oops_lock);
495 spin_msec(1);
496 spin_lock(&pause_on_oops_lock);
497 }
498 }
499 }
500 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
501}
502
503/*
504 * Return true if the calling CPU is allowed to print oops-related info.
505 * This is a bit racy..
506 */
507bool oops_may_print(void)
508{
509 return pause_on_oops_flag == 0;
510}
511
512/*
513 * Called when the architecture enters its oops handler, before it prints
514 * anything. If this is the first CPU to oops, and it's oopsing the first
515 * time then let it proceed.
516 *
517 * This is all enabled by the pause_on_oops kernel boot option. We do all
518 * this to ensure that oopses don't scroll off the screen. It has the
519 * side-effect of preventing later-oopsing CPUs from mucking up the display,
520 * too.
521 *
522 * It turns out that the CPU which is allowed to print ends up pausing for
523 * the right duration, whereas all the other CPUs pause for twice as long:
524 * once in oops_enter(), once in oops_exit().
525 */
526void oops_enter(void)
527{
528 tracing_off();
529 /* can't trust the integrity of the kernel anymore: */
530 debug_locks_off();
531 do_oops_enter_exit();
532
533 if (sysctl_oops_all_cpu_backtrace)
534 trigger_all_cpu_backtrace();
535}
536
537static void print_oops_end_marker(void)
538{
539 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
540}
541
542/*
543 * Called when the architecture exits its oops handler, after printing
544 * everything.
545 */
546void oops_exit(void)
547{
548 do_oops_enter_exit();
549 print_oops_end_marker();
550 kmsg_dump(KMSG_DUMP_OOPS);
551}
552
553struct warn_args {
554 const char *fmt;
555 va_list args;
556};
557
558void __warn(const char *file, int line, void *caller, unsigned taint,
559 struct pt_regs *regs, struct warn_args *args)
560{
561 disable_trace_on_warning();
562
563 if (file)
564 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
565 raw_smp_processor_id(), current->pid, file, line,
566 caller);
567 else
568 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
569 raw_smp_processor_id(), current->pid, caller);
570
571 if (args)
572 vprintk(args->fmt, args->args);
573
574 print_modules();
575
576 if (regs)
577 show_regs(regs);
578
579 if (panic_on_warn) {
580 /*
581 * This thread may hit another WARN() in the panic path.
582 * Resetting this prevents additional WARN() from panicking the
583 * system on this thread. Other threads are blocked by the
584 * panic_mutex in panic().
585 */
586 panic_on_warn = 0;
587 panic("panic_on_warn set ...\n");
588 }
589
590 if (!regs)
591 dump_stack();
592
593 print_irqtrace_events(current);
594
595 print_oops_end_marker();
596 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
597
598 /* Just a warning, don't kill lockdep. */
599 add_taint(taint, LOCKDEP_STILL_OK);
600}
601
602#ifndef __WARN_FLAGS
603void warn_slowpath_fmt(const char *file, int line, unsigned taint,
604 const char *fmt, ...)
605{
606 struct warn_args args;
607
608 pr_warn(CUT_HERE);
609
610 if (!fmt) {
611 __warn(file, line, __builtin_return_address(0), taint,
612 NULL, NULL);
613 return;
614 }
615
616 args.fmt = fmt;
617 va_start(args.args, fmt);
618 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
619 va_end(args.args);
620}
621EXPORT_SYMBOL(warn_slowpath_fmt);
622#else
623void __warn_printk(const char *fmt, ...)
624{
625 va_list args;
626
627 pr_warn(CUT_HERE);
628
629 va_start(args, fmt);
630 vprintk(fmt, args);
631 va_end(args);
632}
633EXPORT_SYMBOL(__warn_printk);
634#endif
635
636#ifdef CONFIG_BUG
637
638/* Support resetting WARN*_ONCE state */
639
640static int clear_warn_once_set(void *data, u64 val)
641{
642 generic_bug_clear_once();
643 memset(__start_once, 0, __end_once - __start_once);
644 return 0;
645}
646
647DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
648 "%lld\n");
649
650static __init int register_warn_debugfs(void)
651{
652 /* Don't care about failure */
653 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
654 &clear_warn_once_fops);
655 return 0;
656}
657
658device_initcall(register_warn_debugfs);
659#endif
660
661#ifdef CONFIG_STACKPROTECTOR
662
663/*
664 * Called when gcc's -fstack-protector feature is used, and
665 * gcc detects corruption of the on-stack canary value
666 */
667__visible noinstr void __stack_chk_fail(void)
668{
669 instrumentation_begin();
670 panic("stack-protector: Kernel stack is corrupted in: %pB",
671 __builtin_return_address(0));
672 instrumentation_end();
673}
674EXPORT_SYMBOL(__stack_chk_fail);
675
676#endif
677
678core_param(panic, panic_timeout, int, 0644);
679core_param(panic_print, panic_print, ulong, 0644);
680core_param(pause_on_oops, pause_on_oops, int, 0644);
681core_param(panic_on_warn, panic_on_warn, int, 0644);
682core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
683
684static int __init oops_setup(char *s)
685{
686 if (!s)
687 return -EINVAL;
688 if (!strcmp(s, "panic"))
689 panic_on_oops = 1;
690 return 0;
691}
692early_param("oops", oops_setup);
693
694static int __init panic_on_taint_setup(char *s)
695{
696 char *taint_str;
697
698 if (!s)
699 return -EINVAL;
700
701 taint_str = strsep(&s, ",");
702 if (kstrtoul(taint_str, 16, &panic_on_taint))
703 return -EINVAL;
704
705 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
706 panic_on_taint &= TAINT_FLAGS_MAX;
707
708 if (!panic_on_taint)
709 return -EINVAL;
710
711 if (s && !strcmp(s, "nousertaint"))
712 panic_on_taint_nousertaint = true;
713
714 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
715 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
716
717 return 0;
718}
719early_param("panic_on_taint", panic_on_taint_setup);