projects/linux·blob v5.6 · refs

688 lines · 17.2 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/sched.h>
27#include <linux/sysrq.h>
28#include <linux/init.h>
29#include <linux/nmi.h>
30#include <linux/console.h>
31#include <linux/bug.h>
32#include <linux/ratelimit.h>
33#include <linux/debugfs.h>
34#include <asm/sections.h>
35
36#define PANIC_TIMER_STEP 100
37#define PANIC_BLINK_SPD 18
38
39int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
40static unsigned long tainted_mask =
41 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
42static int pause_on_oops;
43static int pause_on_oops_flag;
44static DEFINE_SPINLOCK(pause_on_oops_lock);
45bool crash_kexec_post_notifiers;
46int panic_on_warn __read_mostly;
47
48int panic_timeout = CONFIG_PANIC_TIMEOUT;
49EXPORT_SYMBOL_GPL(panic_timeout);
50
51#define PANIC_PRINT_TASK_INFO 0x00000001
52#define PANIC_PRINT_MEM_INFO 0x00000002
53#define PANIC_PRINT_TIMER_INFO 0x00000004
54#define PANIC_PRINT_LOCK_INFO 0x00000008
55#define PANIC_PRINT_FTRACE_INFO 0x00000010
56#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
57unsigned long panic_print;
58
59ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
60
61EXPORT_SYMBOL(panic_notifier_list);
62
63static long no_blink(int state)
64{
65 return 0;
66}
67
68/* Returns how long it waited in ms */
69long (*panic_blink)(int state);
70EXPORT_SYMBOL(panic_blink);
71
72/*
73 * Stop ourself in panic -- architecture code may override this
74 */
75void __weak panic_smp_self_stop(void)
76{
77 while (1)
78 cpu_relax();
79}
80
81/*
82 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
83 * may override this to prepare for crash dumping, e.g. save regs info.
84 */
85void __weak nmi_panic_self_stop(struct pt_regs *regs)
86{
87 panic_smp_self_stop();
88}
89
90/*
91 * Stop other CPUs in panic. Architecture dependent code may override this
92 * with more suitable version. For example, if the architecture supports
93 * crash dump, it should save registers of each stopped CPU and disable
94 * per-CPU features such as virtualization extensions.
95 */
96void __weak crash_smp_send_stop(void)
97{
98 static int cpus_stopped;
99
100 /*
101 * This function can be called twice in panic path, but obviously
102 * we execute this only once.
103 */
104 if (cpus_stopped)
105 return;
106
107 /*
108 * Note smp_send_stop is the usual smp shutdown function, which
109 * unfortunately means it may not be hardened to work in a panic
110 * situation.
111 */
112 smp_send_stop();
113 cpus_stopped = 1;
114}
115
116atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
117
118/*
119 * A variant of panic() called from NMI context. We return if we've already
120 * panicked on this CPU. If another CPU already panicked, loop in
121 * nmi_panic_self_stop() which can provide architecture dependent code such
122 * as saving register state for crash dump.
123 */
124void nmi_panic(struct pt_regs *regs, const char *msg)
125{
126 int old_cpu, cpu;
127
128 cpu = raw_smp_processor_id();
129 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
130
131 if (old_cpu == PANIC_CPU_INVALID)
132 panic("%s", msg);
133 else if (old_cpu != cpu)
134 nmi_panic_self_stop(regs);
135}
136EXPORT_SYMBOL(nmi_panic);
137
138static void panic_print_sys_info(void)
139{
140 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
141 console_flush_on_panic(CONSOLE_REPLAY_ALL);
142
143 if (panic_print & PANIC_PRINT_TASK_INFO)
144 show_state();
145
146 if (panic_print & PANIC_PRINT_MEM_INFO)
147 show_mem(0, NULL);
148
149 if (panic_print & PANIC_PRINT_TIMER_INFO)
150 sysrq_timer_list_show();
151
152 if (panic_print & PANIC_PRINT_LOCK_INFO)
153 debug_show_all_locks();
154
155 if (panic_print & PANIC_PRINT_FTRACE_INFO)
156 ftrace_dump(DUMP_ALL);
157}
158
159/**
160 * panic - halt the system
161 * @fmt: The text string to print
162 *
163 * Display a message, then perform cleanups.
164 *
165 * This function never returns.
166 */
167void panic(const char *fmt, ...)
168{
169 static char buf[1024];
170 va_list args;
171 long i, i_next = 0, len;
172 int state = 0;
173 int old_cpu, this_cpu;
174 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
175
176 /*
177 * Disable local interrupts. This will prevent panic_smp_self_stop
178 * from deadlocking the first cpu that invokes the panic, since
179 * there is nothing to prevent an interrupt handler (that runs
180 * after setting panic_cpu) from invoking panic() again.
181 */
182 local_irq_disable();
183 preempt_disable_notrace();
184
185 /*
186 * It's possible to come here directly from a panic-assertion and
187 * not have preempt disabled. Some functions called from here want
188 * preempt to be disabled. No point enabling it later though...
189 *
190 * Only one CPU is allowed to execute the panic code from here. For
191 * multiple parallel invocations of panic, all other CPUs either
192 * stop themself or will wait until they are stopped by the 1st CPU
193 * with smp_send_stop().
194 *
195 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
196 * comes here, so go ahead.
197 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
198 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
199 */
200 this_cpu = raw_smp_processor_id();
201 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
202
203 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
204 panic_smp_self_stop();
205
206 console_verbose();
207 bust_spinlocks(1);
208 va_start(args, fmt);
209 len = vscnprintf(buf, sizeof(buf), fmt, args);
210 va_end(args);
211
212 if (len && buf[len - 1] == '\n')
213 buf[len - 1] = '\0';
214
215 pr_emerg("Kernel panic - not syncing: %s\n", buf);
216#ifdef CONFIG_DEBUG_BUGVERBOSE
217 /*
218 * Avoid nested stack-dumping if a panic occurs during oops processing
219 */
220 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
221 dump_stack();
222#endif
223
224 /*
225 * If kgdb is enabled, give it a chance to run before we stop all
226 * the other CPUs or else we won't be able to debug processes left
227 * running on them.
228 */
229 kgdb_panic(buf);
230
231 /*
232 * If we have crashed and we have a crash kernel loaded let it handle
233 * everything else.
234 * If we want to run this after calling panic_notifiers, pass
235 * the "crash_kexec_post_notifiers" option to the kernel.
236 *
237 * Bypass the panic_cpu check and call __crash_kexec directly.
238 */
239 if (!_crash_kexec_post_notifiers) {
240 printk_safe_flush_on_panic();
241 __crash_kexec(NULL);
242
243 /*
244 * Note smp_send_stop is the usual smp shutdown function, which
245 * unfortunately means it may not be hardened to work in a
246 * panic situation.
247 */
248 smp_send_stop();
249 } else {
250 /*
251 * If we want to do crash dump after notifier calls and
252 * kmsg_dump, we will need architecture dependent extra
253 * works in addition to stopping other CPUs.
254 */
255 crash_smp_send_stop();
256 }
257
258 /*
259 * Run any panic handlers, including those that might need to
260 * add information to the kmsg dump output.
261 */
262 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
263
264 /* Call flush even twice. It tries harder with a single online CPU */
265 printk_safe_flush_on_panic();
266 kmsg_dump(KMSG_DUMP_PANIC);
267
268 /*
269 * If you doubt kdump always works fine in any situation,
270 * "crash_kexec_post_notifiers" offers you a chance to run
271 * panic_notifiers and dumping kmsg before kdump.
272 * Note: since some panic_notifiers can make crashed kernel
273 * more unstable, it can increase risks of the kdump failure too.
274 *
275 * Bypass the panic_cpu check and call __crash_kexec directly.
276 */
277 if (_crash_kexec_post_notifiers)
278 __crash_kexec(NULL);
279
280#ifdef CONFIG_VT
281 unblank_screen();
282#endif
283 console_unblank();
284
285 /*
286 * We may have ended up stopping the CPU holding the lock (in
287 * smp_send_stop()) while still having some valuable data in the console
288 * buffer. Try to acquire the lock then release it regardless of the
289 * result. The release will also print the buffers out. Locks debug
290 * should be disabled to avoid reporting bad unlock balance when
291 * panic() is not being callled from OOPS.
292 */
293 debug_locks_off();
294 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
295
296 panic_print_sys_info();
297
298 if (!panic_blink)
299 panic_blink = no_blink;
300
301 if (panic_timeout > 0) {
302 /*
303 * Delay timeout seconds before rebooting the machine.
304 * We can't use the "normal" timers since we just panicked.
305 */
306 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
307
308 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
309 touch_nmi_watchdog();
310 if (i >= i_next) {
311 i += panic_blink(state ^= 1);
312 i_next = i + 3600 / PANIC_BLINK_SPD;
313 }
314 mdelay(PANIC_TIMER_STEP);
315 }
316 }
317 if (panic_timeout != 0) {
318 /*
319 * This will not be a clean reboot, with everything
320 * shutting down. But if there is a chance of
321 * rebooting the system it will be rebooted.
322 */
323 if (panic_reboot_mode != REBOOT_UNDEFINED)
324 reboot_mode = panic_reboot_mode;
325 emergency_restart();
326 }
327#ifdef __sparc__
328 {
329 extern int stop_a_enabled;
330 /* Make sure the user can actually press Stop-A (L1-A) */
331 stop_a_enabled = 1;
332 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
333 "twice on console to return to the boot prom\n");
334 }
335#endif
336#if defined(CONFIG_S390)
337 disabled_wait();
338#endif
339 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
340
341 /* Do not scroll important messages printed above */
342 suppress_printk = 1;
343 local_irq_enable();
344 for (i = 0; ; i += PANIC_TIMER_STEP) {
345 touch_softlockup_watchdog();
346 if (i >= i_next) {
347 i += panic_blink(state ^= 1);
348 i_next = i + 3600 / PANIC_BLINK_SPD;
349 }
350 mdelay(PANIC_TIMER_STEP);
351 }
352}
353
354EXPORT_SYMBOL(panic);
355
356/*
357 * TAINT_FORCED_RMMOD could be a per-module flag but the module
358 * is being removed anyway.
359 */
360const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
361 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
362 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
363 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
364 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
365 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
366 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
367 [ TAINT_USER ] = { 'U', ' ', false },
368 [ TAINT_DIE ] = { 'D', ' ', false },
369 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
370 [ TAINT_WARN ] = { 'W', ' ', false },
371 [ TAINT_CRAP ] = { 'C', ' ', true },
372 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
373 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
374 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
375 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
376 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
377 [ TAINT_AUX ] = { 'X', ' ', true },
378 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
379};
380
381/**
382 * print_tainted - return a string to represent the kernel taint state.
383 *
384 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
385 *
386 * The string is overwritten by the next call to print_tainted(),
387 * but is always NULL terminated.
388 */
389const char *print_tainted(void)
390{
391 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
392
393 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
394
395 if (tainted_mask) {
396 char *s;
397 int i;
398
399 s = buf + sprintf(buf, "Tainted: ");
400 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
401 const struct taint_flag *t = &taint_flags[i];
402 *s++ = test_bit(i, &tainted_mask) ?
403 t->c_true : t->c_false;
404 }
405 *s = 0;
406 } else
407 snprintf(buf, sizeof(buf), "Not tainted");
408
409 return buf;
410}
411
412int test_taint(unsigned flag)
413{
414 return test_bit(flag, &tainted_mask);
415}
416EXPORT_SYMBOL(test_taint);
417
418unsigned long get_taint(void)
419{
420 return tainted_mask;
421}
422
423/**
424 * add_taint: add a taint flag if not already set.
425 * @flag: one of the TAINT_* constants.
426 * @lockdep_ok: whether lock debugging is still OK.
427 *
428 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
429 * some notewortht-but-not-corrupting cases, it can be set to true.
430 */
431void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
432{
433 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
434 pr_warn("Disabling lock debugging due to kernel taint\n");
435
436 set_bit(flag, &tainted_mask);
437}
438EXPORT_SYMBOL(add_taint);
439
440static void spin_msec(int msecs)
441{
442 int i;
443
444 for (i = 0; i < msecs; i++) {
445 touch_nmi_watchdog();
446 mdelay(1);
447 }
448}
449
450/*
451 * It just happens that oops_enter() and oops_exit() are identically
452 * implemented...
453 */
454static void do_oops_enter_exit(void)
455{
456 unsigned long flags;
457 static int spin_counter;
458
459 if (!pause_on_oops)
460 return;
461
462 spin_lock_irqsave(&pause_on_oops_lock, flags);
463 if (pause_on_oops_flag == 0) {
464 /* This CPU may now print the oops message */
465 pause_on_oops_flag = 1;
466 } else {
467 /* We need to stall this CPU */
468 if (!spin_counter) {
469 /* This CPU gets to do the counting */
470 spin_counter = pause_on_oops;
471 do {
472 spin_unlock(&pause_on_oops_lock);
473 spin_msec(MSEC_PER_SEC);
474 spin_lock(&pause_on_oops_lock);
475 } while (--spin_counter);
476 pause_on_oops_flag = 0;
477 } else {
478 /* This CPU waits for a different one */
479 while (spin_counter) {
480 spin_unlock(&pause_on_oops_lock);
481 spin_msec(1);
482 spin_lock(&pause_on_oops_lock);
483 }
484 }
485 }
486 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
487}
488
489/*
490 * Return true if the calling CPU is allowed to print oops-related info.
491 * This is a bit racy..
492 */
493int oops_may_print(void)
494{
495 return pause_on_oops_flag == 0;
496}
497
498/*
499 * Called when the architecture enters its oops handler, before it prints
500 * anything. If this is the first CPU to oops, and it's oopsing the first
501 * time then let it proceed.
502 *
503 * This is all enabled by the pause_on_oops kernel boot option. We do all
504 * this to ensure that oopses don't scroll off the screen. It has the
505 * side-effect of preventing later-oopsing CPUs from mucking up the display,
506 * too.
507 *
508 * It turns out that the CPU which is allowed to print ends up pausing for
509 * the right duration, whereas all the other CPUs pause for twice as long:
510 * once in oops_enter(), once in oops_exit().
511 */
512void oops_enter(void)
513{
514 tracing_off();
515 /* can't trust the integrity of the kernel anymore: */
516 debug_locks_off();
517 do_oops_enter_exit();
518}
519
520/*
521 * 64-bit random ID for oopses:
522 */
523static u64 oops_id;
524
525static int init_oops_id(void)
526{
527 if (!oops_id)
528 get_random_bytes(&oops_id, sizeof(oops_id));
529 else
530 oops_id++;
531
532 return 0;
533}
534late_initcall(init_oops_id);
535
536void print_oops_end_marker(void)
537{
538 init_oops_id();
539 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
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 if (panic_on_warn) {
575 /*
576 * This thread may hit another WARN() in the panic path.
577 * Resetting this prevents additional WARN() from panicking the
578 * system on this thread. Other threads are blocked by the
579 * panic_mutex in panic().
580 */
581 panic_on_warn = 0;
582 panic("panic_on_warn set ...\n");
583 }
584
585 print_modules();
586
587 if (regs)
588 show_regs(regs);
589 else
590 dump_stack();
591
592 print_irqtrace_events(current);
593
594 print_oops_end_marker();
595
596 /* Just a warning, don't kill lockdep. */
597 add_taint(taint, LOCKDEP_STILL_OK);
598}
599
600#ifndef __WARN_FLAGS
601void warn_slowpath_fmt(const char *file, int line, unsigned taint,
602 const char *fmt, ...)
603{
604 struct warn_args args;
605
606 pr_warn(CUT_HERE);
607
608 if (!fmt) {
609 __warn(file, line, __builtin_return_address(0), taint,
610 NULL, NULL);
611 return;
612 }
613
614 args.fmt = fmt;
615 va_start(args.args, fmt);
616 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
617 va_end(args.args);
618}
619EXPORT_SYMBOL(warn_slowpath_fmt);
620#else
621void __warn_printk(const char *fmt, ...)
622{
623 va_list args;
624
625 pr_warn(CUT_HERE);
626
627 va_start(args, fmt);
628 vprintk(fmt, args);
629 va_end(args);
630}
631EXPORT_SYMBOL(__warn_printk);
632#endif
633
634#ifdef CONFIG_BUG
635
636/* Support resetting WARN*_ONCE state */
637
638static int clear_warn_once_set(void *data, u64 val)
639{
640 generic_bug_clear_once();
641 memset(__start_once, 0, __end_once - __start_once);
642 return 0;
643}
644
645DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
646 "%lld\n");
647
648static __init int register_warn_debugfs(void)
649{
650 /* Don't care about failure */
651 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
652 &clear_warn_once_fops);
653 return 0;
654}
655
656device_initcall(register_warn_debugfs);
657#endif
658
659#ifdef CONFIG_STACKPROTECTOR
660
661/*
662 * Called when gcc's -fstack-protector feature is used, and
663 * gcc detects corruption of the on-stack canary value
664 */
665__visible void __stack_chk_fail(void)
666{
667 panic("stack-protector: Kernel stack is corrupted in: %pB",
668 __builtin_return_address(0));
669}
670EXPORT_SYMBOL(__stack_chk_fail);
671
672#endif
673
674core_param(panic, panic_timeout, int, 0644);
675core_param(panic_print, panic_print, ulong, 0644);
676core_param(pause_on_oops, pause_on_oops, int, 0644);
677core_param(panic_on_warn, panic_on_warn, int, 0644);
678core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
679
680static int __init oops_setup(char *s)
681{
682 if (!s)
683 return -EINVAL;
684 if (!strcmp(s, "panic"))
685 panic_on_oops = 1;
686 return 0;
687}
688early_param("oops", oops_setup);