projects/linux·blob v4.20 · refs

669 lines · 16.7 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/module.h>
18#include <linux/random.h>
19#include <linux/ftrace.h>
20#include <linux/reboot.h>
21#include <linux/delay.h>
22#include <linux/kexec.h>
23#include <linux/sched.h>
24#include <linux/sysrq.h>
25#include <linux/init.h>
26#include <linux/nmi.h>
27#include <linux/console.h>
28#include <linux/bug.h>
29#include <linux/ratelimit.h>
30#include <linux/debugfs.h>
31#include <asm/sections.h>
32
33#define PANIC_TIMER_STEP 100
34#define PANIC_BLINK_SPD 18
35
36int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
37static unsigned long tainted_mask =
38 IS_ENABLED(CONFIG_GCC_PLUGIN_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
39static int pause_on_oops;
40static int pause_on_oops_flag;
41static DEFINE_SPINLOCK(pause_on_oops_lock);
42bool crash_kexec_post_notifiers;
43int panic_on_warn __read_mostly;
44
45int panic_timeout = CONFIG_PANIC_TIMEOUT;
46EXPORT_SYMBOL_GPL(panic_timeout);
47
48ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
49
50EXPORT_SYMBOL(panic_notifier_list);
51
52static long no_blink(int state)
53{
54 return 0;
55}
56
57/* Returns how long it waited in ms */
58long (*panic_blink)(int state);
59EXPORT_SYMBOL(panic_blink);
60
61/*
62 * Stop ourself in panic -- architecture code may override this
63 */
64void __weak panic_smp_self_stop(void)
65{
66 while (1)
67 cpu_relax();
68}
69
70/*
71 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
72 * may override this to prepare for crash dumping, e.g. save regs info.
73 */
74void __weak nmi_panic_self_stop(struct pt_regs *regs)
75{
76 panic_smp_self_stop();
77}
78
79/*
80 * Stop other CPUs in panic. Architecture dependent code may override this
81 * with more suitable version. For example, if the architecture supports
82 * crash dump, it should save registers of each stopped CPU and disable
83 * per-CPU features such as virtualization extensions.
84 */
85void __weak crash_smp_send_stop(void)
86{
87 static int cpus_stopped;
88
89 /*
90 * This function can be called twice in panic path, but obviously
91 * we execute this only once.
92 */
93 if (cpus_stopped)
94 return;
95
96 /*
97 * Note smp_send_stop is the usual smp shutdown function, which
98 * unfortunately means it may not be hardened to work in a panic
99 * situation.
100 */
101 smp_send_stop();
102 cpus_stopped = 1;
103}
104
105atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
106
107/*
108 * A variant of panic() called from NMI context. We return if we've already
109 * panicked on this CPU. If another CPU already panicked, loop in
110 * nmi_panic_self_stop() which can provide architecture dependent code such
111 * as saving register state for crash dump.
112 */
113void nmi_panic(struct pt_regs *regs, const char *msg)
114{
115 int old_cpu, cpu;
116
117 cpu = raw_smp_processor_id();
118 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
119
120 if (old_cpu == PANIC_CPU_INVALID)
121 panic("%s", msg);
122 else if (old_cpu != cpu)
123 nmi_panic_self_stop(regs);
124}
125EXPORT_SYMBOL(nmi_panic);
126
127/**
128 * panic - halt the system
129 * @fmt: The text string to print
130 *
131 * Display a message, then perform cleanups.
132 *
133 * This function never returns.
134 */
135void panic(const char *fmt, ...)
136{
137 static char buf[1024];
138 va_list args;
139 long i, i_next = 0, len;
140 int state = 0;
141 int old_cpu, this_cpu;
142 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
143
144 /*
145 * Disable local interrupts. This will prevent panic_smp_self_stop
146 * from deadlocking the first cpu that invokes the panic, since
147 * there is nothing to prevent an interrupt handler (that runs
148 * after setting panic_cpu) from invoking panic() again.
149 */
150 local_irq_disable();
151
152 /*
153 * It's possible to come here directly from a panic-assertion and
154 * not have preempt disabled. Some functions called from here want
155 * preempt to be disabled. No point enabling it later though...
156 *
157 * Only one CPU is allowed to execute the panic code from here. For
158 * multiple parallel invocations of panic, all other CPUs either
159 * stop themself or will wait until they are stopped by the 1st CPU
160 * with smp_send_stop().
161 *
162 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
163 * comes here, so go ahead.
164 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
165 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
166 */
167 this_cpu = raw_smp_processor_id();
168 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
169
170 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
171 panic_smp_self_stop();
172
173 console_verbose();
174 bust_spinlocks(1);
175 va_start(args, fmt);
176 len = vscnprintf(buf, sizeof(buf), fmt, args);
177 va_end(args);
178
179 if (len && buf[len - 1] == '\n')
180 buf[len - 1] = '\0';
181
182 pr_emerg("Kernel panic - not syncing: %s\n", buf);
183#ifdef CONFIG_DEBUG_BUGVERBOSE
184 /*
185 * Avoid nested stack-dumping if a panic occurs during oops processing
186 */
187 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
188 dump_stack();
189#endif
190
191 /*
192 * If we have crashed and we have a crash kernel loaded let it handle
193 * everything else.
194 * If we want to run this after calling panic_notifiers, pass
195 * the "crash_kexec_post_notifiers" option to the kernel.
196 *
197 * Bypass the panic_cpu check and call __crash_kexec directly.
198 */
199 if (!_crash_kexec_post_notifiers) {
200 printk_safe_flush_on_panic();
201 __crash_kexec(NULL);
202
203 /*
204 * Note smp_send_stop is the usual smp shutdown function, which
205 * unfortunately means it may not be hardened to work in a
206 * panic situation.
207 */
208 smp_send_stop();
209 } else {
210 /*
211 * If we want to do crash dump after notifier calls and
212 * kmsg_dump, we will need architecture dependent extra
213 * works in addition to stopping other CPUs.
214 */
215 crash_smp_send_stop();
216 }
217
218 /*
219 * Run any panic handlers, including those that might need to
220 * add information to the kmsg dump output.
221 */
222 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
223
224 /* Call flush even twice. It tries harder with a single online CPU */
225 printk_safe_flush_on_panic();
226 kmsg_dump(KMSG_DUMP_PANIC);
227
228 /*
229 * If you doubt kdump always works fine in any situation,
230 * "crash_kexec_post_notifiers" offers you a chance to run
231 * panic_notifiers and dumping kmsg before kdump.
232 * Note: since some panic_notifiers can make crashed kernel
233 * more unstable, it can increase risks of the kdump failure too.
234 *
235 * Bypass the panic_cpu check and call __crash_kexec directly.
236 */
237 if (_crash_kexec_post_notifiers)
238 __crash_kexec(NULL);
239
240 bust_spinlocks(0);
241
242 /*
243 * We may have ended up stopping the CPU holding the lock (in
244 * smp_send_stop()) while still having some valuable data in the console
245 * buffer. Try to acquire the lock then release it regardless of the
246 * result. The release will also print the buffers out. Locks debug
247 * should be disabled to avoid reporting bad unlock balance when
248 * panic() is not being callled from OOPS.
249 */
250 debug_locks_off();
251 console_flush_on_panic();
252
253 if (!panic_blink)
254 panic_blink = no_blink;
255
256 if (panic_timeout > 0) {
257 /*
258 * Delay timeout seconds before rebooting the machine.
259 * We can't use the "normal" timers since we just panicked.
260 */
261 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
262
263 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
264 touch_nmi_watchdog();
265 if (i >= i_next) {
266 i += panic_blink(state ^= 1);
267 i_next = i + 3600 / PANIC_BLINK_SPD;
268 }
269 mdelay(PANIC_TIMER_STEP);
270 }
271 }
272 if (panic_timeout != 0) {
273 /*
274 * This will not be a clean reboot, with everything
275 * shutting down. But if there is a chance of
276 * rebooting the system it will be rebooted.
277 */
278 emergency_restart();
279 }
280#ifdef __sparc__
281 {
282 extern int stop_a_enabled;
283 /* Make sure the user can actually press Stop-A (L1-A) */
284 stop_a_enabled = 1;
285 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
286 "twice on console to return to the boot prom\n");
287 }
288#endif
289#if defined(CONFIG_S390)
290 {
291 unsigned long caller;
292
293 caller = (unsigned long)__builtin_return_address(0);
294 disabled_wait(caller);
295 }
296#endif
297 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
298 local_irq_enable();
299 for (i = 0; ; i += PANIC_TIMER_STEP) {
300 touch_softlockup_watchdog();
301 if (i >= i_next) {
302 i += panic_blink(state ^= 1);
303 i_next = i + 3600 / PANIC_BLINK_SPD;
304 }
305 mdelay(PANIC_TIMER_STEP);
306 }
307}
308
309EXPORT_SYMBOL(panic);
310
311/*
312 * TAINT_FORCED_RMMOD could be a per-module flag but the module
313 * is being removed anyway.
314 */
315const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
316 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
317 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
318 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
319 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
320 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
321 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
322 [ TAINT_USER ] = { 'U', ' ', false },
323 [ TAINT_DIE ] = { 'D', ' ', false },
324 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
325 [ TAINT_WARN ] = { 'W', ' ', false },
326 [ TAINT_CRAP ] = { 'C', ' ', true },
327 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
328 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
329 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
330 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
331 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
332 [ TAINT_AUX ] = { 'X', ' ', true },
333 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
334};
335
336/**
337 * print_tainted - return a string to represent the kernel taint state.
338 *
339 * For individual taint flag meanings, see Documentation/sysctl/kernel.txt
340 *
341 * The string is overwritten by the next call to print_tainted(),
342 * but is always NULL terminated.
343 */
344const char *print_tainted(void)
345{
346 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
347
348 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
349
350 if (tainted_mask) {
351 char *s;
352 int i;
353
354 s = buf + sprintf(buf, "Tainted: ");
355 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
356 const struct taint_flag *t = &taint_flags[i];
357 *s++ = test_bit(i, &tainted_mask) ?
358 t->c_true : t->c_false;
359 }
360 *s = 0;
361 } else
362 snprintf(buf, sizeof(buf), "Not tainted");
363
364 return buf;
365}
366
367int test_taint(unsigned flag)
368{
369 return test_bit(flag, &tainted_mask);
370}
371EXPORT_SYMBOL(test_taint);
372
373unsigned long get_taint(void)
374{
375 return tainted_mask;
376}
377
378/**
379 * add_taint: add a taint flag if not already set.
380 * @flag: one of the TAINT_* constants.
381 * @lockdep_ok: whether lock debugging is still OK.
382 *
383 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
384 * some notewortht-but-not-corrupting cases, it can be set to true.
385 */
386void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
387{
388 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
389 pr_warn("Disabling lock debugging due to kernel taint\n");
390
391 set_bit(flag, &tainted_mask);
392}
393EXPORT_SYMBOL(add_taint);
394
395static void spin_msec(int msecs)
396{
397 int i;
398
399 for (i = 0; i < msecs; i++) {
400 touch_nmi_watchdog();
401 mdelay(1);
402 }
403}
404
405/*
406 * It just happens that oops_enter() and oops_exit() are identically
407 * implemented...
408 */
409static void do_oops_enter_exit(void)
410{
411 unsigned long flags;
412 static int spin_counter;
413
414 if (!pause_on_oops)
415 return;
416
417 spin_lock_irqsave(&pause_on_oops_lock, flags);
418 if (pause_on_oops_flag == 0) {
419 /* This CPU may now print the oops message */
420 pause_on_oops_flag = 1;
421 } else {
422 /* We need to stall this CPU */
423 if (!spin_counter) {
424 /* This CPU gets to do the counting */
425 spin_counter = pause_on_oops;
426 do {
427 spin_unlock(&pause_on_oops_lock);
428 spin_msec(MSEC_PER_SEC);
429 spin_lock(&pause_on_oops_lock);
430 } while (--spin_counter);
431 pause_on_oops_flag = 0;
432 } else {
433 /* This CPU waits for a different one */
434 while (spin_counter) {
435 spin_unlock(&pause_on_oops_lock);
436 spin_msec(1);
437 spin_lock(&pause_on_oops_lock);
438 }
439 }
440 }
441 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
442}
443
444/*
445 * Return true if the calling CPU is allowed to print oops-related info.
446 * This is a bit racy..
447 */
448int oops_may_print(void)
449{
450 return pause_on_oops_flag == 0;
451}
452
453/*
454 * Called when the architecture enters its oops handler, before it prints
455 * anything. If this is the first CPU to oops, and it's oopsing the first
456 * time then let it proceed.
457 *
458 * This is all enabled by the pause_on_oops kernel boot option. We do all
459 * this to ensure that oopses don't scroll off the screen. It has the
460 * side-effect of preventing later-oopsing CPUs from mucking up the display,
461 * too.
462 *
463 * It turns out that the CPU which is allowed to print ends up pausing for
464 * the right duration, whereas all the other CPUs pause for twice as long:
465 * once in oops_enter(), once in oops_exit().
466 */
467void oops_enter(void)
468{
469 tracing_off();
470 /* can't trust the integrity of the kernel anymore: */
471 debug_locks_off();
472 do_oops_enter_exit();
473}
474
475/*
476 * 64-bit random ID for oopses:
477 */
478static u64 oops_id;
479
480static int init_oops_id(void)
481{
482 if (!oops_id)
483 get_random_bytes(&oops_id, sizeof(oops_id));
484 else
485 oops_id++;
486
487 return 0;
488}
489late_initcall(init_oops_id);
490
491void print_oops_end_marker(void)
492{
493 init_oops_id();
494 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
495}
496
497/*
498 * Called when the architecture exits its oops handler, after printing
499 * everything.
500 */
501void oops_exit(void)
502{
503 do_oops_enter_exit();
504 print_oops_end_marker();
505 kmsg_dump(KMSG_DUMP_OOPS);
506}
507
508struct warn_args {
509 const char *fmt;
510 va_list args;
511};
512
513void __warn(const char *file, int line, void *caller, unsigned taint,
514 struct pt_regs *regs, struct warn_args *args)
515{
516 disable_trace_on_warning();
517
518 if (args)
519 pr_warn(CUT_HERE);
520
521 if (file)
522 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
523 raw_smp_processor_id(), current->pid, file, line,
524 caller);
525 else
526 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
527 raw_smp_processor_id(), current->pid, caller);
528
529 if (args)
530 vprintk(args->fmt, args->args);
531
532 if (panic_on_warn) {
533 /*
534 * This thread may hit another WARN() in the panic path.
535 * Resetting this prevents additional WARN() from panicking the
536 * system on this thread. Other threads are blocked by the
537 * panic_mutex in panic().
538 */
539 panic_on_warn = 0;
540 panic("panic_on_warn set ...\n");
541 }
542
543 print_modules();
544
545 if (regs)
546 show_regs(regs);
547 else
548 dump_stack();
549
550 print_irqtrace_events(current);
551
552 print_oops_end_marker();
553
554 /* Just a warning, don't kill lockdep. */
555 add_taint(taint, LOCKDEP_STILL_OK);
556}
557
558#ifdef WANT_WARN_ON_SLOWPATH
559void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
560{
561 struct warn_args args;
562
563 args.fmt = fmt;
564 va_start(args.args, fmt);
565 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
566 &args);
567 va_end(args.args);
568}
569EXPORT_SYMBOL(warn_slowpath_fmt);
570
571void warn_slowpath_fmt_taint(const char *file, int line,
572 unsigned taint, const char *fmt, ...)
573{
574 struct warn_args args;
575
576 args.fmt = fmt;
577 va_start(args.args, fmt);
578 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
579 va_end(args.args);
580}
581EXPORT_SYMBOL(warn_slowpath_fmt_taint);
582
583void warn_slowpath_null(const char *file, int line)
584{
585 pr_warn(CUT_HERE);
586 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
587}
588EXPORT_SYMBOL(warn_slowpath_null);
589#else
590void __warn_printk(const char *fmt, ...)
591{
592 va_list args;
593
594 pr_warn(CUT_HERE);
595
596 va_start(args, fmt);
597 vprintk(fmt, args);
598 va_end(args);
599}
600EXPORT_SYMBOL(__warn_printk);
601#endif
602
603#ifdef CONFIG_BUG
604
605/* Support resetting WARN*_ONCE state */
606
607static int clear_warn_once_set(void *data, u64 val)
608{
609 generic_bug_clear_once();
610 memset(__start_once, 0, __end_once - __start_once);
611 return 0;
612}
613
614DEFINE_SIMPLE_ATTRIBUTE(clear_warn_once_fops,
615 NULL,
616 clear_warn_once_set,
617 "%lld\n");
618
619static __init int register_warn_debugfs(void)
620{
621 /* Don't care about failure */
622 debugfs_create_file("clear_warn_once", 0200, NULL,
623 NULL, &clear_warn_once_fops);
624 return 0;
625}
626
627device_initcall(register_warn_debugfs);
628#endif
629
630#ifdef CONFIG_STACKPROTECTOR
631
632/*
633 * Called when gcc's -fstack-protector feature is used, and
634 * gcc detects corruption of the on-stack canary value
635 */
636__visible void __stack_chk_fail(void)
637{
638 panic("stack-protector: Kernel stack is corrupted in: %pB",
639 __builtin_return_address(0));
640}
641EXPORT_SYMBOL(__stack_chk_fail);
642
643#endif
644
645#ifdef CONFIG_ARCH_HAS_REFCOUNT
646void refcount_error_report(struct pt_regs *regs, const char *err)
647{
648 WARN_RATELIMIT(1, "refcount_t %s at %pB in %s[%d], uid/euid: %u/%u\n",
649 err, (void *)instruction_pointer(regs),
650 current->comm, task_pid_nr(current),
651 from_kuid_munged(&init_user_ns, current_uid()),
652 from_kuid_munged(&init_user_ns, current_euid()));
653}
654#endif
655
656core_param(panic, panic_timeout, int, 0644);
657core_param(pause_on_oops, pause_on_oops, int, 0644);
658core_param(panic_on_warn, panic_on_warn, int, 0644);
659core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
660
661static int __init oops_setup(char *s)
662{
663 if (!s)
664 return -EINVAL;
665 if (!strcmp(s, "panic"))
666 panic_on_oops = 1;
667 return 0;
668}
669early_param("oops", oops_setup);