projects/linux·blob v5.15-rc3 · refs

734 lines · 18.2 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 <asm/sections.h>
36
37#define PANIC_TIMER_STEP 100
38#define PANIC_BLINK_SPD 18
39
40#ifdef CONFIG_SMP
41/*
42 * Should we dump all CPUs backtraces in an oops event?
43 * Defaults to 0, can be changed via sysctl.
44 */
45unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
46#endif /* CONFIG_SMP */
47
48int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
49static unsigned long tainted_mask =
50 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
51static int pause_on_oops;
52static int pause_on_oops_flag;
53static DEFINE_SPINLOCK(pause_on_oops_lock);
54bool crash_kexec_post_notifiers;
55int panic_on_warn __read_mostly;
56unsigned long panic_on_taint;
57bool panic_on_taint_nousertaint = false;
58
59int panic_timeout = CONFIG_PANIC_TIMEOUT;
60EXPORT_SYMBOL_GPL(panic_timeout);
61
62#define PANIC_PRINT_TASK_INFO 0x00000001
63#define PANIC_PRINT_MEM_INFO 0x00000002
64#define PANIC_PRINT_TIMER_INFO 0x00000004
65#define PANIC_PRINT_LOCK_INFO 0x00000008
66#define PANIC_PRINT_FTRACE_INFO 0x00000010
67#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
68unsigned long panic_print;
69
70ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
71
72EXPORT_SYMBOL(panic_notifier_list);
73
74static long no_blink(int state)
75{
76 return 0;
77}
78
79/* Returns how long it waited in ms */
80long (*panic_blink)(int state);
81EXPORT_SYMBOL(panic_blink);
82
83/*
84 * Stop ourself in panic -- architecture code may override this
85 */
86void __weak panic_smp_self_stop(void)
87{
88 while (1)
89 cpu_relax();
90}
91
92/*
93 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
94 * may override this to prepare for crash dumping, e.g. save regs info.
95 */
96void __weak nmi_panic_self_stop(struct pt_regs *regs)
97{
98 panic_smp_self_stop();
99}
100
101/*
102 * Stop other CPUs in panic. Architecture dependent code may override this
103 * with more suitable version. For example, if the architecture supports
104 * crash dump, it should save registers of each stopped CPU and disable
105 * per-CPU features such as virtualization extensions.
106 */
107void __weak crash_smp_send_stop(void)
108{
109 static int cpus_stopped;
110
111 /*
112 * This function can be called twice in panic path, but obviously
113 * we execute this only once.
114 */
115 if (cpus_stopped)
116 return;
117
118 /*
119 * Note smp_send_stop is the usual smp shutdown function, which
120 * unfortunately means it may not be hardened to work in a panic
121 * situation.
122 */
123 smp_send_stop();
124 cpus_stopped = 1;
125}
126
127atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
128
129/*
130 * A variant of panic() called from NMI context. We return if we've already
131 * panicked on this CPU. If another CPU already panicked, loop in
132 * nmi_panic_self_stop() which can provide architecture dependent code such
133 * as saving register state for crash dump.
134 */
135void nmi_panic(struct pt_regs *regs, const char *msg)
136{
137 int old_cpu, cpu;
138
139 cpu = raw_smp_processor_id();
140 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
141
142 if (old_cpu == PANIC_CPU_INVALID)
143 panic("%s", msg);
144 else if (old_cpu != cpu)
145 nmi_panic_self_stop(regs);
146}
147EXPORT_SYMBOL(nmi_panic);
148
149static void panic_print_sys_info(void)
150{
151 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
152 console_flush_on_panic(CONSOLE_REPLAY_ALL);
153
154 if (panic_print & PANIC_PRINT_TASK_INFO)
155 show_state();
156
157 if (panic_print & PANIC_PRINT_MEM_INFO)
158 show_mem(0, NULL);
159
160 if (panic_print & PANIC_PRINT_TIMER_INFO)
161 sysrq_timer_list_show();
162
163 if (panic_print & PANIC_PRINT_LOCK_INFO)
164 debug_show_all_locks();
165
166 if (panic_print & PANIC_PRINT_FTRACE_INFO)
167 ftrace_dump(DUMP_ALL);
168}
169
170/**
171 * panic - halt the system
172 * @fmt: The text string to print
173 *
174 * Display a message, then perform cleanups.
175 *
176 * This function never returns.
177 */
178void panic(const char *fmt, ...)
179{
180 static char buf[1024];
181 va_list args;
182 long i, i_next = 0, len;
183 int state = 0;
184 int old_cpu, this_cpu;
185 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
186
187 /*
188 * Disable local interrupts. This will prevent panic_smp_self_stop
189 * from deadlocking the first cpu that invokes the panic, since
190 * there is nothing to prevent an interrupt handler (that runs
191 * after setting panic_cpu) from invoking panic() again.
192 */
193 local_irq_disable();
194 preempt_disable_notrace();
195
196 /*
197 * It's possible to come here directly from a panic-assertion and
198 * not have preempt disabled. Some functions called from here want
199 * preempt to be disabled. No point enabling it later though...
200 *
201 * Only one CPU is allowed to execute the panic code from here. For
202 * multiple parallel invocations of panic, all other CPUs either
203 * stop themself or will wait until they are stopped by the 1st CPU
204 * with smp_send_stop().
205 *
206 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
207 * comes here, so go ahead.
208 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
209 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
210 */
211 this_cpu = raw_smp_processor_id();
212 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
213
214 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
215 panic_smp_self_stop();
216
217 console_verbose();
218 bust_spinlocks(1);
219 va_start(args, fmt);
220 len = vscnprintf(buf, sizeof(buf), fmt, args);
221 va_end(args);
222
223 if (len && buf[len - 1] == '\n')
224 buf[len - 1] = '\0';
225
226 pr_emerg("Kernel panic - not syncing: %s\n", buf);
227#ifdef CONFIG_DEBUG_BUGVERBOSE
228 /*
229 * Avoid nested stack-dumping if a panic occurs during oops processing
230 */
231 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
232 dump_stack();
233#endif
234
235 /*
236 * If kgdb is enabled, give it a chance to run before we stop all
237 * the other CPUs or else we won't be able to debug processes left
238 * running on them.
239 */
240 kgdb_panic(buf);
241
242 /*
243 * If we have crashed and we have a crash kernel loaded let it handle
244 * everything else.
245 * If we want to run this after calling panic_notifiers, pass
246 * the "crash_kexec_post_notifiers" option to the kernel.
247 *
248 * Bypass the panic_cpu check and call __crash_kexec directly.
249 */
250 if (!_crash_kexec_post_notifiers) {
251 __crash_kexec(NULL);
252
253 /*
254 * Note smp_send_stop is the usual smp shutdown function, which
255 * unfortunately means it may not be hardened to work in a
256 * panic situation.
257 */
258 smp_send_stop();
259 } else {
260 /*
261 * If we want to do crash dump after notifier calls and
262 * kmsg_dump, we will need architecture dependent extra
263 * works in addition to stopping other CPUs.
264 */
265 crash_smp_send_stop();
266 }
267
268 /*
269 * Run any panic handlers, including those that might need to
270 * add information to the kmsg dump output.
271 */
272 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
273
274 kmsg_dump(KMSG_DUMP_PANIC);
275
276 /*
277 * If you doubt kdump always works fine in any situation,
278 * "crash_kexec_post_notifiers" offers you a chance to run
279 * panic_notifiers and dumping kmsg before kdump.
280 * Note: since some panic_notifiers can make crashed kernel
281 * more unstable, it can increase risks of the kdump failure too.
282 *
283 * Bypass the panic_cpu check and call __crash_kexec directly.
284 */
285 if (_crash_kexec_post_notifiers)
286 __crash_kexec(NULL);
287
288#ifdef CONFIG_VT
289 unblank_screen();
290#endif
291 console_unblank();
292
293 /*
294 * We may have ended up stopping the CPU holding the lock (in
295 * smp_send_stop()) while still having some valuable data in the console
296 * buffer. Try to acquire the lock then release it regardless of the
297 * result. The release will also print the buffers out. Locks debug
298 * should be disabled to avoid reporting bad unlock balance when
299 * panic() is not being callled from OOPS.
300 */
301 debug_locks_off();
302 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
303
304 panic_print_sys_info();
305
306 if (!panic_blink)
307 panic_blink = no_blink;
308
309 if (panic_timeout > 0) {
310 /*
311 * Delay timeout seconds before rebooting the machine.
312 * We can't use the "normal" timers since we just panicked.
313 */
314 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
315
316 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
317 touch_nmi_watchdog();
318 if (i >= i_next) {
319 i += panic_blink(state ^= 1);
320 i_next = i + 3600 / PANIC_BLINK_SPD;
321 }
322 mdelay(PANIC_TIMER_STEP);
323 }
324 }
325 if (panic_timeout != 0) {
326 /*
327 * This will not be a clean reboot, with everything
328 * shutting down. But if there is a chance of
329 * rebooting the system it will be rebooted.
330 */
331 if (panic_reboot_mode != REBOOT_UNDEFINED)
332 reboot_mode = panic_reboot_mode;
333 emergency_restart();
334 }
335#ifdef __sparc__
336 {
337 extern int stop_a_enabled;
338 /* Make sure the user can actually press Stop-A (L1-A) */
339 stop_a_enabled = 1;
340 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
341 "twice on console to return to the boot prom\n");
342 }
343#endif
344#if defined(CONFIG_S390)
345 disabled_wait();
346#endif
347 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
348
349 /* Do not scroll important messages printed above */
350 suppress_printk = 1;
351 local_irq_enable();
352 for (i = 0; ; i += PANIC_TIMER_STEP) {
353 touch_softlockup_watchdog();
354 if (i >= i_next) {
355 i += panic_blink(state ^= 1);
356 i_next = i + 3600 / PANIC_BLINK_SPD;
357 }
358 mdelay(PANIC_TIMER_STEP);
359 }
360}
361
362EXPORT_SYMBOL(panic);
363
364/*
365 * TAINT_FORCED_RMMOD could be a per-module flag but the module
366 * is being removed anyway.
367 */
368const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
369 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
370 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
371 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
372 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
373 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
374 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
375 [ TAINT_USER ] = { 'U', ' ', false },
376 [ TAINT_DIE ] = { 'D', ' ', false },
377 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
378 [ TAINT_WARN ] = { 'W', ' ', false },
379 [ TAINT_CRAP ] = { 'C', ' ', true },
380 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
381 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
382 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
383 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
384 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
385 [ TAINT_AUX ] = { 'X', ' ', true },
386 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
387};
388
389/**
390 * print_tainted - return a string to represent the kernel taint state.
391 *
392 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
393 *
394 * The string is overwritten by the next call to print_tainted(),
395 * but is always NULL terminated.
396 */
397const char *print_tainted(void)
398{
399 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
400
401 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
402
403 if (tainted_mask) {
404 char *s;
405 int i;
406
407 s = buf + sprintf(buf, "Tainted: ");
408 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
409 const struct taint_flag *t = &taint_flags[i];
410 *s++ = test_bit(i, &tainted_mask) ?
411 t->c_true : t->c_false;
412 }
413 *s = 0;
414 } else
415 snprintf(buf, sizeof(buf), "Not tainted");
416
417 return buf;
418}
419
420int test_taint(unsigned flag)
421{
422 return test_bit(flag, &tainted_mask);
423}
424EXPORT_SYMBOL(test_taint);
425
426unsigned long get_taint(void)
427{
428 return tainted_mask;
429}
430
431/**
432 * add_taint: add a taint flag if not already set.
433 * @flag: one of the TAINT_* constants.
434 * @lockdep_ok: whether lock debugging is still OK.
435 *
436 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
437 * some notewortht-but-not-corrupting cases, it can be set to true.
438 */
439void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
440{
441 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
442 pr_warn("Disabling lock debugging due to kernel taint\n");
443
444 set_bit(flag, &tainted_mask);
445
446 if (tainted_mask & panic_on_taint) {
447 panic_on_taint = 0;
448 panic("panic_on_taint set ...");
449 }
450}
451EXPORT_SYMBOL(add_taint);
452
453static void spin_msec(int msecs)
454{
455 int i;
456
457 for (i = 0; i < msecs; i++) {
458 touch_nmi_watchdog();
459 mdelay(1);
460 }
461}
462
463/*
464 * It just happens that oops_enter() and oops_exit() are identically
465 * implemented...
466 */
467static void do_oops_enter_exit(void)
468{
469 unsigned long flags;
470 static int spin_counter;
471
472 if (!pause_on_oops)
473 return;
474
475 spin_lock_irqsave(&pause_on_oops_lock, flags);
476 if (pause_on_oops_flag == 0) {
477 /* This CPU may now print the oops message */
478 pause_on_oops_flag = 1;
479 } else {
480 /* We need to stall this CPU */
481 if (!spin_counter) {
482 /* This CPU gets to do the counting */
483 spin_counter = pause_on_oops;
484 do {
485 spin_unlock(&pause_on_oops_lock);
486 spin_msec(MSEC_PER_SEC);
487 spin_lock(&pause_on_oops_lock);
488 } while (--spin_counter);
489 pause_on_oops_flag = 0;
490 } else {
491 /* This CPU waits for a different one */
492 while (spin_counter) {
493 spin_unlock(&pause_on_oops_lock);
494 spin_msec(1);
495 spin_lock(&pause_on_oops_lock);
496 }
497 }
498 }
499 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
500}
501
502/*
503 * Return true if the calling CPU is allowed to print oops-related info.
504 * This is a bit racy..
505 */
506bool oops_may_print(void)
507{
508 return pause_on_oops_flag == 0;
509}
510
511/*
512 * Called when the architecture enters its oops handler, before it prints
513 * anything. If this is the first CPU to oops, and it's oopsing the first
514 * time then let it proceed.
515 *
516 * This is all enabled by the pause_on_oops kernel boot option. We do all
517 * this to ensure that oopses don't scroll off the screen. It has the
518 * side-effect of preventing later-oopsing CPUs from mucking up the display,
519 * too.
520 *
521 * It turns out that the CPU which is allowed to print ends up pausing for
522 * the right duration, whereas all the other CPUs pause for twice as long:
523 * once in oops_enter(), once in oops_exit().
524 */
525void oops_enter(void)
526{
527 tracing_off();
528 /* can't trust the integrity of the kernel anymore: */
529 debug_locks_off();
530 do_oops_enter_exit();
531
532 if (sysctl_oops_all_cpu_backtrace)
533 trigger_all_cpu_backtrace();
534}
535
536/*
537 * 64-bit random ID for oopses:
538 */
539static u64 oops_id;
540
541static int init_oops_id(void)
542{
543 if (!oops_id)
544 get_random_bytes(&oops_id, sizeof(oops_id));
545 else
546 oops_id++;
547
548 return 0;
549}
550late_initcall(init_oops_id);
551
552static void print_oops_end_marker(void)
553{
554 init_oops_id();
555 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
556}
557
558/*
559 * Called when the architecture exits its oops handler, after printing
560 * everything.
561 */
562void oops_exit(void)
563{
564 do_oops_enter_exit();
565 print_oops_end_marker();
566 kmsg_dump(KMSG_DUMP_OOPS);
567}
568
569struct warn_args {
570 const char *fmt;
571 va_list args;
572};
573
574void __warn(const char *file, int line, void *caller, unsigned taint,
575 struct pt_regs *regs, struct warn_args *args)
576{
577 disable_trace_on_warning();
578
579 if (file)
580 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
581 raw_smp_processor_id(), current->pid, file, line,
582 caller);
583 else
584 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
585 raw_smp_processor_id(), current->pid, caller);
586
587 if (args)
588 vprintk(args->fmt, args->args);
589
590 print_modules();
591
592 if (regs)
593 show_regs(regs);
594
595 if (panic_on_warn) {
596 /*
597 * This thread may hit another WARN() in the panic path.
598 * Resetting this prevents additional WARN() from panicking the
599 * system on this thread. Other threads are blocked by the
600 * panic_mutex in panic().
601 */
602 panic_on_warn = 0;
603 panic("panic_on_warn set ...\n");
604 }
605
606 if (!regs)
607 dump_stack();
608
609 print_irqtrace_events(current);
610
611 print_oops_end_marker();
612
613 /* Just a warning, don't kill lockdep. */
614 add_taint(taint, LOCKDEP_STILL_OK);
615}
616
617#ifndef __WARN_FLAGS
618void warn_slowpath_fmt(const char *file, int line, unsigned taint,
619 const char *fmt, ...)
620{
621 struct warn_args args;
622
623 pr_warn(CUT_HERE);
624
625 if (!fmt) {
626 __warn(file, line, __builtin_return_address(0), taint,
627 NULL, NULL);
628 return;
629 }
630
631 args.fmt = fmt;
632 va_start(args.args, fmt);
633 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
634 va_end(args.args);
635}
636EXPORT_SYMBOL(warn_slowpath_fmt);
637#else
638void __warn_printk(const char *fmt, ...)
639{
640 va_list args;
641
642 pr_warn(CUT_HERE);
643
644 va_start(args, fmt);
645 vprintk(fmt, args);
646 va_end(args);
647}
648EXPORT_SYMBOL(__warn_printk);
649#endif
650
651#ifdef CONFIG_BUG
652
653/* Support resetting WARN*_ONCE state */
654
655static int clear_warn_once_set(void *data, u64 val)
656{
657 generic_bug_clear_once();
658 memset(__start_once, 0, __end_once - __start_once);
659 return 0;
660}
661
662DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
663 "%lld\n");
664
665static __init int register_warn_debugfs(void)
666{
667 /* Don't care about failure */
668 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
669 &clear_warn_once_fops);
670 return 0;
671}
672
673device_initcall(register_warn_debugfs);
674#endif
675
676#ifdef CONFIG_STACKPROTECTOR
677
678/*
679 * Called when gcc's -fstack-protector feature is used, and
680 * gcc detects corruption of the on-stack canary value
681 */
682__visible noinstr void __stack_chk_fail(void)
683{
684 instrumentation_begin();
685 panic("stack-protector: Kernel stack is corrupted in: %pB",
686 __builtin_return_address(0));
687 instrumentation_end();
688}
689EXPORT_SYMBOL(__stack_chk_fail);
690
691#endif
692
693core_param(panic, panic_timeout, int, 0644);
694core_param(panic_print, panic_print, ulong, 0644);
695core_param(pause_on_oops, pause_on_oops, int, 0644);
696core_param(panic_on_warn, panic_on_warn, int, 0644);
697core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
698
699static int __init oops_setup(char *s)
700{
701 if (!s)
702 return -EINVAL;
703 if (!strcmp(s, "panic"))
704 panic_on_oops = 1;
705 return 0;
706}
707early_param("oops", oops_setup);
708
709static int __init panic_on_taint_setup(char *s)
710{
711 char *taint_str;
712
713 if (!s)
714 return -EINVAL;
715
716 taint_str = strsep(&s, ",");
717 if (kstrtoul(taint_str, 16, &panic_on_taint))
718 return -EINVAL;
719
720 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
721 panic_on_taint &= TAINT_FLAGS_MAX;
722
723 if (!panic_on_taint)
724 return -EINVAL;
725
726 if (s && !strcmp(s, "nousertaint"))
727 panic_on_taint_nousertaint = true;
728
729 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%sabled\n",
730 panic_on_taint, panic_on_taint_nousertaint ? "en" : "dis");
731
732 return 0;
733}
734early_param("panic_on_taint", panic_on_taint_setup);