projects/linux·blob v5.0-rc3 · refs

701 lines · 17.4 KiB

1/*
2 * linux/kernel/panic.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * This function is used through-out the kernel (including mm and fs)
9 * to indicate a major problem.
10 */
11#include <linux/debug_locks.h>
12#include <linux/sched/debug.h>
13#include <linux/interrupt.h>
14#include <linux/kmsg_dump.h>
15#include <linux/kallsyms.h>
16#include <linux/notifier.h>
17#include <linux/vt_kern.h>
18#include <linux/module.h>
19#include <linux/random.h>
20#include <linux/ftrace.h>
21#include <linux/reboot.h>
22#include <linux/delay.h>
23#include <linux/kexec.h>
24#include <linux/sched.h>
25#include <linux/sysrq.h>
26#include <linux/init.h>
27#include <linux/nmi.h>
28#include <linux/console.h>
29#include <linux/bug.h>
30#include <linux/ratelimit.h>
31#include <linux/debugfs.h>
32#include <asm/sections.h>
33
34#define PANIC_TIMER_STEP 100
35#define PANIC_BLINK_SPD 18
36
37int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
38static unsigned long tainted_mask =
39 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
40static int pause_on_oops;
41static int pause_on_oops_flag;
42static DEFINE_SPINLOCK(pause_on_oops_lock);
43bool crash_kexec_post_notifiers;
44int panic_on_warn __read_mostly;
45
46int panic_timeout = CONFIG_PANIC_TIMEOUT;
47EXPORT_SYMBOL_GPL(panic_timeout);
48
49#define PANIC_PRINT_TASK_INFO 0x00000001
50#define PANIC_PRINT_MEM_INFO 0x00000002
51#define PANIC_PRINT_TIMER_INFO 0x00000004
52#define PANIC_PRINT_LOCK_INFO 0x00000008
53#define PANIC_PRINT_FTRACE_INFO 0x00000010
54unsigned long panic_print;
55
56ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
57
58EXPORT_SYMBOL(panic_notifier_list);
59
60static long no_blink(int state)
61{
62 return 0;
63}
64
65/* Returns how long it waited in ms */
66long (*panic_blink)(int state);
67EXPORT_SYMBOL(panic_blink);
68
69/*
70 * Stop ourself in panic -- architecture code may override this
71 */
72void __weak panic_smp_self_stop(void)
73{
74 while (1)
75 cpu_relax();
76}
77
78/*
79 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
80 * may override this to prepare for crash dumping, e.g. save regs info.
81 */
82void __weak nmi_panic_self_stop(struct pt_regs *regs)
83{
84 panic_smp_self_stop();
85}
86
87/*
88 * Stop other CPUs in panic. Architecture dependent code may override this
89 * with more suitable version. For example, if the architecture supports
90 * crash dump, it should save registers of each stopped CPU and disable
91 * per-CPU features such as virtualization extensions.
92 */
93void __weak crash_smp_send_stop(void)
94{
95 static int cpus_stopped;
96
97 /*
98 * This function can be called twice in panic path, but obviously
99 * we execute this only once.
100 */
101 if (cpus_stopped)
102 return;
103
104 /*
105 * Note smp_send_stop is the usual smp shutdown function, which
106 * unfortunately means it may not be hardened to work in a panic
107 * situation.
108 */
109 smp_send_stop();
110 cpus_stopped = 1;
111}
112
113atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
114
115/*
116 * A variant of panic() called from NMI context. We return if we've already
117 * panicked on this CPU. If another CPU already panicked, loop in
118 * nmi_panic_self_stop() which can provide architecture dependent code such
119 * as saving register state for crash dump.
120 */
121void nmi_panic(struct pt_regs *regs, const char *msg)
122{
123 int old_cpu, cpu;
124
125 cpu = raw_smp_processor_id();
126 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
127
128 if (old_cpu == PANIC_CPU_INVALID)
129 panic("%s", msg);
130 else if (old_cpu != cpu)
131 nmi_panic_self_stop(regs);
132}
133EXPORT_SYMBOL(nmi_panic);
134
135static void panic_print_sys_info(void)
136{
137 if (panic_print & PANIC_PRINT_TASK_INFO)
138 show_state();
139
140 if (panic_print & PANIC_PRINT_MEM_INFO)
141 show_mem(0, NULL);
142
143 if (panic_print & PANIC_PRINT_TIMER_INFO)
144 sysrq_timer_list_show();
145
146 if (panic_print & PANIC_PRINT_LOCK_INFO)
147 debug_show_all_locks();
148
149 if (panic_print & PANIC_PRINT_FTRACE_INFO)
150 ftrace_dump(DUMP_ALL);
151}
152
153/**
154 * panic - halt the system
155 * @fmt: The text string to print
156 *
157 * Display a message, then perform cleanups.
158 *
159 * This function never returns.
160 */
161void panic(const char *fmt, ...)
162{
163 static char buf[1024];
164 va_list args;
165 long i, i_next = 0, len;
166 int state = 0;
167 int old_cpu, this_cpu;
168 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
169
170 /*
171 * Disable local interrupts. This will prevent panic_smp_self_stop
172 * from deadlocking the first cpu that invokes the panic, since
173 * there is nothing to prevent an interrupt handler (that runs
174 * after setting panic_cpu) from invoking panic() again.
175 */
176 local_irq_disable();
177
178 /*
179 * It's possible to come here directly from a panic-assertion and
180 * not have preempt disabled. Some functions called from here want
181 * preempt to be disabled. No point enabling it later though...
182 *
183 * Only one CPU is allowed to execute the panic code from here. For
184 * multiple parallel invocations of panic, all other CPUs either
185 * stop themself or will wait until they are stopped by the 1st CPU
186 * with smp_send_stop().
187 *
188 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
189 * comes here, so go ahead.
190 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
191 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
192 */
193 this_cpu = raw_smp_processor_id();
194 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
195
196 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
197 panic_smp_self_stop();
198
199 console_verbose();
200 bust_spinlocks(1);
201 va_start(args, fmt);
202 len = vscnprintf(buf, sizeof(buf), fmt, args);
203 va_end(args);
204
205 if (len && buf[len - 1] == '\n')
206 buf[len - 1] = '\0';
207
208 pr_emerg("Kernel panic - not syncing: %s\n", buf);
209#ifdef CONFIG_DEBUG_BUGVERBOSE
210 /*
211 * Avoid nested stack-dumping if a panic occurs during oops processing
212 */
213 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
214 dump_stack();
215#endif
216
217 /*
218 * If we have crashed and we have a crash kernel loaded let it handle
219 * everything else.
220 * If we want to run this after calling panic_notifiers, pass
221 * the "crash_kexec_post_notifiers" option to the kernel.
222 *
223 * Bypass the panic_cpu check and call __crash_kexec directly.
224 */
225 if (!_crash_kexec_post_notifiers) {
226 printk_safe_flush_on_panic();
227 __crash_kexec(NULL);
228
229 /*
230 * Note smp_send_stop is the usual smp shutdown function, which
231 * unfortunately means it may not be hardened to work in a
232 * panic situation.
233 */
234 smp_send_stop();
235 } else {
236 /*
237 * If we want to do crash dump after notifier calls and
238 * kmsg_dump, we will need architecture dependent extra
239 * works in addition to stopping other CPUs.
240 */
241 crash_smp_send_stop();
242 }
243
244 /*
245 * Run any panic handlers, including those that might need to
246 * add information to the kmsg dump output.
247 */
248 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
249
250 /* Call flush even twice. It tries harder with a single online CPU */
251 printk_safe_flush_on_panic();
252 kmsg_dump(KMSG_DUMP_PANIC);
253
254 /*
255 * If you doubt kdump always works fine in any situation,
256 * "crash_kexec_post_notifiers" offers you a chance to run
257 * panic_notifiers and dumping kmsg before kdump.
258 * Note: since some panic_notifiers can make crashed kernel
259 * more unstable, it can increase risks of the kdump failure too.
260 *
261 * Bypass the panic_cpu check and call __crash_kexec directly.
262 */
263 if (_crash_kexec_post_notifiers)
264 __crash_kexec(NULL);
265
266#ifdef CONFIG_VT
267 unblank_screen();
268#endif
269 console_unblank();
270
271 /*
272 * We may have ended up stopping the CPU holding the lock (in
273 * smp_send_stop()) while still having some valuable data in the console
274 * buffer. Try to acquire the lock then release it regardless of the
275 * result. The release will also print the buffers out. Locks debug
276 * should be disabled to avoid reporting bad unlock balance when
277 * panic() is not being callled from OOPS.
278 */
279 debug_locks_off();
280 console_flush_on_panic();
281
282 panic_print_sys_info();
283
284 if (!panic_blink)
285 panic_blink = no_blink;
286
287 if (panic_timeout > 0) {
288 /*
289 * Delay timeout seconds before rebooting the machine.
290 * We can't use the "normal" timers since we just panicked.
291 */
292 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
293
294 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
295 touch_nmi_watchdog();
296 if (i >= i_next) {
297 i += panic_blink(state ^= 1);
298 i_next = i + 3600 / PANIC_BLINK_SPD;
299 }
300 mdelay(PANIC_TIMER_STEP);
301 }
302 }
303 if (panic_timeout != 0) {
304 /*
305 * This will not be a clean reboot, with everything
306 * shutting down. But if there is a chance of
307 * rebooting the system it will be rebooted.
308 */
309 emergency_restart();
310 }
311#ifdef __sparc__
312 {
313 extern int stop_a_enabled;
314 /* Make sure the user can actually press Stop-A (L1-A) */
315 stop_a_enabled = 1;
316 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
317 "twice on console to return to the boot prom\n");
318 }
319#endif
320#if defined(CONFIG_S390)
321 {
322 unsigned long caller;
323
324 caller = (unsigned long)__builtin_return_address(0);
325 disabled_wait(caller);
326 }
327#endif
328 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
329 local_irq_enable();
330 for (i = 0; ; i += PANIC_TIMER_STEP) {
331 touch_softlockup_watchdog();
332 if (i >= i_next) {
333 i += panic_blink(state ^= 1);
334 i_next = i + 3600 / PANIC_BLINK_SPD;
335 }
336 mdelay(PANIC_TIMER_STEP);
337 }
338}
339
340EXPORT_SYMBOL(panic);
341
342/*
343 * TAINT_FORCED_RMMOD could be a per-module flag but the module
344 * is being removed anyway.
345 */
346const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
347 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
348 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
349 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
350 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
351 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
352 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
353 [ TAINT_USER ] = { 'U', ' ', false },
354 [ TAINT_DIE ] = { 'D', ' ', false },
355 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
356 [ TAINT_WARN ] = { 'W', ' ', false },
357 [ TAINT_CRAP ] = { 'C', ' ', true },
358 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
359 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
360 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
361 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
362 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
363 [ TAINT_AUX ] = { 'X', ' ', true },
364 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
365};
366
367/**
368 * print_tainted - return a string to represent the kernel taint state.
369 *
370 * For individual taint flag meanings, see Documentation/sysctl/kernel.txt
371 *
372 * The string is overwritten by the next call to print_tainted(),
373 * but is always NULL terminated.
374 */
375const char *print_tainted(void)
376{
377 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
378
379 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
380
381 if (tainted_mask) {
382 char *s;
383 int i;
384
385 s = buf + sprintf(buf, "Tainted: ");
386 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
387 const struct taint_flag *t = &taint_flags[i];
388 *s++ = test_bit(i, &tainted_mask) ?
389 t->c_true : t->c_false;
390 }
391 *s = 0;
392 } else
393 snprintf(buf, sizeof(buf), "Not tainted");
394
395 return buf;
396}
397
398int test_taint(unsigned flag)
399{
400 return test_bit(flag, &tainted_mask);
401}
402EXPORT_SYMBOL(test_taint);
403
404unsigned long get_taint(void)
405{
406 return tainted_mask;
407}
408
409/**
410 * add_taint: add a taint flag if not already set.
411 * @flag: one of the TAINT_* constants.
412 * @lockdep_ok: whether lock debugging is still OK.
413 *
414 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
415 * some notewortht-but-not-corrupting cases, it can be set to true.
416 */
417void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
418{
419 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
420 pr_warn("Disabling lock debugging due to kernel taint\n");
421
422 set_bit(flag, &tainted_mask);
423}
424EXPORT_SYMBOL(add_taint);
425
426static void spin_msec(int msecs)
427{
428 int i;
429
430 for (i = 0; i < msecs; i++) {
431 touch_nmi_watchdog();
432 mdelay(1);
433 }
434}
435
436/*
437 * It just happens that oops_enter() and oops_exit() are identically
438 * implemented...
439 */
440static void do_oops_enter_exit(void)
441{
442 unsigned long flags;
443 static int spin_counter;
444
445 if (!pause_on_oops)
446 return;
447
448 spin_lock_irqsave(&pause_on_oops_lock, flags);
449 if (pause_on_oops_flag == 0) {
450 /* This CPU may now print the oops message */
451 pause_on_oops_flag = 1;
452 } else {
453 /* We need to stall this CPU */
454 if (!spin_counter) {
455 /* This CPU gets to do the counting */
456 spin_counter = pause_on_oops;
457 do {
458 spin_unlock(&pause_on_oops_lock);
459 spin_msec(MSEC_PER_SEC);
460 spin_lock(&pause_on_oops_lock);
461 } while (--spin_counter);
462 pause_on_oops_flag = 0;
463 } else {
464 /* This CPU waits for a different one */
465 while (spin_counter) {
466 spin_unlock(&pause_on_oops_lock);
467 spin_msec(1);
468 spin_lock(&pause_on_oops_lock);
469 }
470 }
471 }
472 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
473}
474
475/*
476 * Return true if the calling CPU is allowed to print oops-related info.
477 * This is a bit racy..
478 */
479int oops_may_print(void)
480{
481 return pause_on_oops_flag == 0;
482}
483
484/*
485 * Called when the architecture enters its oops handler, before it prints
486 * anything. If this is the first CPU to oops, and it's oopsing the first
487 * time then let it proceed.
488 *
489 * This is all enabled by the pause_on_oops kernel boot option. We do all
490 * this to ensure that oopses don't scroll off the screen. It has the
491 * side-effect of preventing later-oopsing CPUs from mucking up the display,
492 * too.
493 *
494 * It turns out that the CPU which is allowed to print ends up pausing for
495 * the right duration, whereas all the other CPUs pause for twice as long:
496 * once in oops_enter(), once in oops_exit().
497 */
498void oops_enter(void)
499{
500 tracing_off();
501 /* can't trust the integrity of the kernel anymore: */
502 debug_locks_off();
503 do_oops_enter_exit();
504}
505
506/*
507 * 64-bit random ID for oopses:
508 */
509static u64 oops_id;
510
511static int init_oops_id(void)
512{
513 if (!oops_id)
514 get_random_bytes(&oops_id, sizeof(oops_id));
515 else
516 oops_id++;
517
518 return 0;
519}
520late_initcall(init_oops_id);
521
522void print_oops_end_marker(void)
523{
524 init_oops_id();
525 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
526}
527
528/*
529 * Called when the architecture exits its oops handler, after printing
530 * everything.
531 */
532void oops_exit(void)
533{
534 do_oops_enter_exit();
535 print_oops_end_marker();
536 kmsg_dump(KMSG_DUMP_OOPS);
537}
538
539struct warn_args {
540 const char *fmt;
541 va_list args;
542};
543
544void __warn(const char *file, int line, void *caller, unsigned taint,
545 struct pt_regs *regs, struct warn_args *args)
546{
547 disable_trace_on_warning();
548
549 if (args)
550 pr_warn(CUT_HERE);
551
552 if (file)
553 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
554 raw_smp_processor_id(), current->pid, file, line,
555 caller);
556 else
557 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
558 raw_smp_processor_id(), current->pid, caller);
559
560 if (args)
561 vprintk(args->fmt, args->args);
562
563 if (panic_on_warn) {
564 /*
565 * This thread may hit another WARN() in the panic path.
566 * Resetting this prevents additional WARN() from panicking the
567 * system on this thread. Other threads are blocked by the
568 * panic_mutex in panic().
569 */
570 panic_on_warn = 0;
571 panic("panic_on_warn set ...\n");
572 }
573
574 print_modules();
575
576 if (regs)
577 show_regs(regs);
578 else
579 dump_stack();
580
581 print_irqtrace_events(current);
582
583 print_oops_end_marker();
584
585 /* Just a warning, don't kill lockdep. */
586 add_taint(taint, LOCKDEP_STILL_OK);
587}
588
589#ifdef WANT_WARN_ON_SLOWPATH
590void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
591{
592 struct warn_args args;
593
594 args.fmt = fmt;
595 va_start(args.args, fmt);
596 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
597 &args);
598 va_end(args.args);
599}
600EXPORT_SYMBOL(warn_slowpath_fmt);
601
602void warn_slowpath_fmt_taint(const char *file, int line,
603 unsigned taint, const char *fmt, ...)
604{
605 struct warn_args args;
606
607 args.fmt = fmt;
608 va_start(args.args, fmt);
609 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
610 va_end(args.args);
611}
612EXPORT_SYMBOL(warn_slowpath_fmt_taint);
613
614void warn_slowpath_null(const char *file, int line)
615{
616 pr_warn(CUT_HERE);
617 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
618}
619EXPORT_SYMBOL(warn_slowpath_null);
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_SIMPLE_ATTRIBUTE(clear_warn_once_fops,
646 NULL,
647 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("clear_warn_once", 0200, NULL,
654 NULL, &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 void __stack_chk_fail(void)
668{
669 panic("stack-protector: Kernel stack is corrupted in: %pB",
670 __builtin_return_address(0));
671}
672EXPORT_SYMBOL(__stack_chk_fail);
673
674#endif
675
676#ifdef CONFIG_ARCH_HAS_REFCOUNT
677void refcount_error_report(struct pt_regs *regs, const char *err)
678{
679 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
680 err, (void *)instruction_pointer(regs),
681 current->comm, task_pid_nr(current),
682 from_kuid_munged(&init_user_ns, current_uid()),
683 from_kuid_munged(&init_user_ns, current_euid()));
684}
685#endif
686
687core_param(panic, panic_timeout, int, 0644);
688core_param(panic_print, panic_print, ulong, 0644);
689core_param(pause_on_oops, pause_on_oops, int, 0644);
690core_param(panic_on_warn, panic_on_warn, int, 0644);
691core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
692
693static int __init oops_setup(char *s)
694{
695 if (!s)
696 return -EINVAL;
697 if (!strcmp(s, "panic"))
698 panic_on_oops = 1;
699 return 0;
700}
701early_param("oops", oops_setup);