Loading...
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/interrupt.h>
13#include <linux/kmsg_dump.h>
14#include <linux/kallsyms.h>
15#include <linux/notifier.h>
16#include <linux/module.h>
17#include <linux/random.h>
18#include <linux/ftrace.h>
19#include <linux/reboot.h>
20#include <linux/delay.h>
21#include <linux/kexec.h>
22#include <linux/sched.h>
23#include <linux/sysrq.h>
24#include <linux/init.h>
25#include <linux/nmi.h>
26#include <linux/console.h>
27#include <linux/bug.h>
28
29#define PANIC_TIMER_STEP 100
30#define PANIC_BLINK_SPD 18
31
32int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
33static unsigned long tainted_mask;
34static int pause_on_oops;
35static int pause_on_oops_flag;
36static DEFINE_SPINLOCK(pause_on_oops_lock);
37bool crash_kexec_post_notifiers;
38int panic_on_warn __read_mostly;
39
40int panic_timeout = CONFIG_PANIC_TIMEOUT;
41EXPORT_SYMBOL_GPL(panic_timeout);
42
43ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
44
45EXPORT_SYMBOL(panic_notifier_list);
46
47static long no_blink(int state)
48{
49 return 0;
50}
51
52/* Returns how long it waited in ms */
53long (*panic_blink)(int state);
54EXPORT_SYMBOL(panic_blink);
55
56/*
57 * Stop ourself in panic -- architecture code may override this
58 */
59void __weak panic_smp_self_stop(void)
60{
61 while (1)
62 cpu_relax();
63}
64
65/*
66 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
67 * may override this to prepare for crash dumping, e.g. save regs info.
68 */
69void __weak nmi_panic_self_stop(struct pt_regs *regs)
70{
71 panic_smp_self_stop();
72}
73
74/*
75 * Stop other CPUs in panic. Architecture dependent code may override this
76 * with more suitable version. For example, if the architecture supports
77 * crash dump, it should save registers of each stopped CPU and disable
78 * per-CPU features such as virtualization extensions.
79 */
80void __weak crash_smp_send_stop(void)
81{
82 static int cpus_stopped;
83
84 /*
85 * This function can be called twice in panic path, but obviously
86 * we execute this only once.
87 */
88 if (cpus_stopped)
89 return;
90
91 /*
92 * Note smp_send_stop is the usual smp shutdown function, which
93 * unfortunately means it may not be hardened to work in a panic
94 * situation.
95 */
96 smp_send_stop();
97 cpus_stopped = 1;
98}
99
100atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
101
102/*
103 * A variant of panic() called from NMI context. We return if we've already
104 * panicked on this CPU. If another CPU already panicked, loop in
105 * nmi_panic_self_stop() which can provide architecture dependent code such
106 * as saving register state for crash dump.
107 */
108void nmi_panic(struct pt_regs *regs, const char *msg)
109{
110 int old_cpu, cpu;
111
112 cpu = raw_smp_processor_id();
113 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
114
115 if (old_cpu == PANIC_CPU_INVALID)
116 panic("%s", msg);
117 else if (old_cpu != cpu)
118 nmi_panic_self_stop(regs);
119}
120EXPORT_SYMBOL(nmi_panic);
121
122/**
123 * panic - halt the system
124 * @fmt: The text string to print
125 *
126 * Display a message, then perform cleanups.
127 *
128 * This function never returns.
129 */
130void panic(const char *fmt, ...)
131{
132 static char buf[1024];
133 va_list args;
134 long i, i_next = 0;
135 int state = 0;
136 int old_cpu, this_cpu;
137 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
138
139 /*
140 * Disable local interrupts. This will prevent panic_smp_self_stop
141 * from deadlocking the first cpu that invokes the panic, since
142 * there is nothing to prevent an interrupt handler (that runs
143 * after setting panic_cpu) from invoking panic() again.
144 */
145 local_irq_disable();
146
147 /*
148 * It's possible to come here directly from a panic-assertion and
149 * not have preempt disabled. Some functions called from here want
150 * preempt to be disabled. No point enabling it later though...
151 *
152 * Only one CPU is allowed to execute the panic code from here. For
153 * multiple parallel invocations of panic, all other CPUs either
154 * stop themself or will wait until they are stopped by the 1st CPU
155 * with smp_send_stop().
156 *
157 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
158 * comes here, so go ahead.
159 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
160 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
161 */
162 this_cpu = raw_smp_processor_id();
163 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
164
165 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
166 panic_smp_self_stop();
167
168 console_verbose();
169 bust_spinlocks(1);
170 va_start(args, fmt);
171 vsnprintf(buf, sizeof(buf), fmt, args);
172 va_end(args);
173 pr_emerg("Kernel panic - not syncing: %s\n", buf);
174#ifdef CONFIG_DEBUG_BUGVERBOSE
175 /*
176 * Avoid nested stack-dumping if a panic occurs during oops processing
177 */
178 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
179 dump_stack();
180#endif
181
182 /*
183 * If we have crashed and we have a crash kernel loaded let it handle
184 * everything else.
185 * If we want to run this after calling panic_notifiers, pass
186 * the "crash_kexec_post_notifiers" option to the kernel.
187 *
188 * Bypass the panic_cpu check and call __crash_kexec directly.
189 */
190 if (!_crash_kexec_post_notifiers) {
191 printk_nmi_flush_on_panic();
192 __crash_kexec(NULL);
193
194 /*
195 * Note smp_send_stop is the usual smp shutdown function, which
196 * unfortunately means it may not be hardened to work in a
197 * panic situation.
198 */
199 smp_send_stop();
200 } else {
201 /*
202 * If we want to do crash dump after notifier calls and
203 * kmsg_dump, we will need architecture dependent extra
204 * works in addition to stopping other CPUs.
205 */
206 crash_smp_send_stop();
207 }
208
209 /*
210 * Run any panic handlers, including those that might need to
211 * add information to the kmsg dump output.
212 */
213 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
214
215 /* Call flush even twice. It tries harder with a single online CPU */
216 printk_nmi_flush_on_panic();
217 kmsg_dump(KMSG_DUMP_PANIC);
218
219 /*
220 * If you doubt kdump always works fine in any situation,
221 * "crash_kexec_post_notifiers" offers you a chance to run
222 * panic_notifiers and dumping kmsg before kdump.
223 * Note: since some panic_notifiers can make crashed kernel
224 * more unstable, it can increase risks of the kdump failure too.
225 *
226 * Bypass the panic_cpu check and call __crash_kexec directly.
227 */
228 if (_crash_kexec_post_notifiers)
229 __crash_kexec(NULL);
230
231 bust_spinlocks(0);
232
233 /*
234 * We may have ended up stopping the CPU holding the lock (in
235 * smp_send_stop()) while still having some valuable data in the console
236 * buffer. Try to acquire the lock then release it regardless of the
237 * result. The release will also print the buffers out. Locks debug
238 * should be disabled to avoid reporting bad unlock balance when
239 * panic() is not being callled from OOPS.
240 */
241 debug_locks_off();
242 console_flush_on_panic();
243
244 if (!panic_blink)
245 panic_blink = no_blink;
246
247 if (panic_timeout > 0) {
248 /*
249 * Delay timeout seconds before rebooting the machine.
250 * We can't use the "normal" timers since we just panicked.
251 */
252 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
253
254 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
255 touch_nmi_watchdog();
256 if (i >= i_next) {
257 i += panic_blink(state ^= 1);
258 i_next = i + 3600 / PANIC_BLINK_SPD;
259 }
260 mdelay(PANIC_TIMER_STEP);
261 }
262 }
263 if (panic_timeout != 0) {
264 /*
265 * This will not be a clean reboot, with everything
266 * shutting down. But if there is a chance of
267 * rebooting the system it will be rebooted.
268 */
269 emergency_restart();
270 }
271#ifdef __sparc__
272 {
273 extern int stop_a_enabled;
274 /* Make sure the user can actually press Stop-A (L1-A) */
275 stop_a_enabled = 1;
276 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
277 }
278#endif
279#if defined(CONFIG_S390)
280 {
281 unsigned long caller;
282
283 caller = (unsigned long)__builtin_return_address(0);
284 disabled_wait(caller);
285 }
286#endif
287 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
288 local_irq_enable();
289 for (i = 0; ; i += PANIC_TIMER_STEP) {
290 touch_softlockup_watchdog();
291 if (i >= i_next) {
292 i += panic_blink(state ^= 1);
293 i_next = i + 3600 / PANIC_BLINK_SPD;
294 }
295 mdelay(PANIC_TIMER_STEP);
296 }
297}
298
299EXPORT_SYMBOL(panic);
300
301/*
302 * TAINT_FORCED_RMMOD could be a per-module flag but the module
303 * is being removed anyway.
304 */
305const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
306 { 'P', 'G', true }, /* TAINT_PROPRIETARY_MODULE */
307 { 'F', ' ', true }, /* TAINT_FORCED_MODULE */
308 { 'S', ' ', false }, /* TAINT_CPU_OUT_OF_SPEC */
309 { 'R', ' ', false }, /* TAINT_FORCED_RMMOD */
310 { 'M', ' ', false }, /* TAINT_MACHINE_CHECK */
311 { 'B', ' ', false }, /* TAINT_BAD_PAGE */
312 { 'U', ' ', false }, /* TAINT_USER */
313 { 'D', ' ', false }, /* TAINT_DIE */
314 { 'A', ' ', false }, /* TAINT_OVERRIDDEN_ACPI_TABLE */
315 { 'W', ' ', false }, /* TAINT_WARN */
316 { 'C', ' ', true }, /* TAINT_CRAP */
317 { 'I', ' ', false }, /* TAINT_FIRMWARE_WORKAROUND */
318 { 'O', ' ', true }, /* TAINT_OOT_MODULE */
319 { 'E', ' ', true }, /* TAINT_UNSIGNED_MODULE */
320 { 'L', ' ', false }, /* TAINT_SOFTLOCKUP */
321 { 'K', ' ', true }, /* TAINT_LIVEPATCH */
322};
323
324/**
325 * print_tainted - return a string to represent the kernel taint state.
326 *
327 * 'P' - Proprietary module has been loaded.
328 * 'F' - Module has been forcibly loaded.
329 * 'S' - SMP with CPUs not designed for SMP.
330 * 'R' - User forced a module unload.
331 * 'M' - System experienced a machine check exception.
332 * 'B' - System has hit bad_page.
333 * 'U' - Userspace-defined naughtiness.
334 * 'D' - Kernel has oopsed before
335 * 'A' - ACPI table overridden.
336 * 'W' - Taint on warning.
337 * 'C' - modules from drivers/staging are loaded.
338 * 'I' - Working around severe firmware bug.
339 * 'O' - Out-of-tree module has been loaded.
340 * 'E' - Unsigned module has been loaded.
341 * 'L' - A soft lockup has previously occurred.
342 * 'K' - Kernel has been live patched.
343 *
344 * The string is overwritten by the next call to print_tainted().
345 */
346const char *print_tainted(void)
347{
348 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
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 pr_warn("------------[ cut here ]------------\n");
519
520 if (file)
521 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
522 raw_smp_processor_id(), current->pid, file, line,
523 caller);
524 else
525 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
526 raw_smp_processor_id(), current->pid, caller);
527
528 if (args)
529 vprintk(args->fmt, args->args);
530
531 if (panic_on_warn) {
532 /*
533 * This thread may hit another WARN() in the panic path.
534 * Resetting this prevents additional WARN() from panicking the
535 * system on this thread. Other threads are blocked by the
536 * panic_mutex in panic().
537 */
538 panic_on_warn = 0;
539 panic("panic_on_warn set ...\n");
540 }
541
542 print_modules();
543
544 if (regs)
545 show_regs(regs);
546 else
547 dump_stack();
548
549 print_oops_end_marker();
550
551 /* Just a warning, don't kill lockdep. */
552 add_taint(taint, LOCKDEP_STILL_OK);
553}
554
555#ifdef WANT_WARN_ON_SLOWPATH
556void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
557{
558 struct warn_args args;
559
560 args.fmt = fmt;
561 va_start(args.args, fmt);
562 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
563 &args);
564 va_end(args.args);
565}
566EXPORT_SYMBOL(warn_slowpath_fmt);
567
568void warn_slowpath_fmt_taint(const char *file, int line,
569 unsigned taint, const char *fmt, ...)
570{
571 struct warn_args args;
572
573 args.fmt = fmt;
574 va_start(args.args, fmt);
575 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
576 va_end(args.args);
577}
578EXPORT_SYMBOL(warn_slowpath_fmt_taint);
579
580void warn_slowpath_null(const char *file, int line)
581{
582 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
583}
584EXPORT_SYMBOL(warn_slowpath_null);
585#endif
586
587#ifdef CONFIG_CC_STACKPROTECTOR
588
589/*
590 * Called when gcc's -fstack-protector feature is used, and
591 * gcc detects corruption of the on-stack canary value
592 */
593__visible void __stack_chk_fail(void)
594{
595 panic("stack-protector: Kernel stack is corrupted in: %p\n",
596 __builtin_return_address(0));
597}
598EXPORT_SYMBOL(__stack_chk_fail);
599
600#endif
601
602core_param(panic, panic_timeout, int, 0644);
603core_param(pause_on_oops, pause_on_oops, int, 0644);
604core_param(panic_on_warn, panic_on_warn, int, 0644);
605core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
606
607static int __init oops_setup(char *s)
608{
609 if (!s)
610 return -EINVAL;
611 if (!strcmp(s, "panic"))
612 panic_on_oops = 1;
613 return 0;
614}
615early_param("oops", oops_setup);
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/string_helpers.h>
29#include <linux/sysrq.h>
30#include <linux/init.h>
31#include <linux/nmi.h>
32#include <linux/console.h>
33#include <linux/bug.h>
34#include <linux/ratelimit.h>
35#include <linux/debugfs.h>
36#include <linux/sysfs.h>
37#include <linux/context_tracking.h>
38#include <trace/events/error_report.h>
39#include <asm/sections.h>
40
41#define PANIC_TIMER_STEP 100
42#define PANIC_BLINK_SPD 18
43
44#ifdef CONFIG_SMP
45/*
46 * Should we dump all CPUs backtraces in an oops event?
47 * Defaults to 0, can be changed via sysctl.
48 */
49static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
50#else
51#define sysctl_oops_all_cpu_backtrace 0
52#endif /* CONFIG_SMP */
53
54int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
55static unsigned long tainted_mask =
56 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
57static int pause_on_oops;
58static int pause_on_oops_flag;
59static DEFINE_SPINLOCK(pause_on_oops_lock);
60bool crash_kexec_post_notifiers;
61int panic_on_warn __read_mostly;
62unsigned long panic_on_taint;
63bool panic_on_taint_nousertaint = false;
64static unsigned int warn_limit __read_mostly;
65
66int panic_timeout = CONFIG_PANIC_TIMEOUT;
67EXPORT_SYMBOL_GPL(panic_timeout);
68
69#define PANIC_PRINT_TASK_INFO 0x00000001
70#define PANIC_PRINT_MEM_INFO 0x00000002
71#define PANIC_PRINT_TIMER_INFO 0x00000004
72#define PANIC_PRINT_LOCK_INFO 0x00000008
73#define PANIC_PRINT_FTRACE_INFO 0x00000010
74#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
75#define PANIC_PRINT_ALL_CPU_BT 0x00000040
76unsigned long panic_print;
77
78ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
79
80EXPORT_SYMBOL(panic_notifier_list);
81
82#ifdef CONFIG_SYSCTL
83static struct ctl_table kern_panic_table[] = {
84#ifdef CONFIG_SMP
85 {
86 .procname = "oops_all_cpu_backtrace",
87 .data = &sysctl_oops_all_cpu_backtrace,
88 .maxlen = sizeof(int),
89 .mode = 0644,
90 .proc_handler = proc_dointvec_minmax,
91 .extra1 = SYSCTL_ZERO,
92 .extra2 = SYSCTL_ONE,
93 },
94#endif
95 {
96 .procname = "warn_limit",
97 .data = &warn_limit,
98 .maxlen = sizeof(warn_limit),
99 .mode = 0644,
100 .proc_handler = proc_douintvec,
101 },
102 { }
103};
104
105static __init int kernel_panic_sysctls_init(void)
106{
107 register_sysctl_init("kernel", kern_panic_table);
108 return 0;
109}
110late_initcall(kernel_panic_sysctls_init);
111#endif
112
113static atomic_t warn_count = ATOMIC_INIT(0);
114
115#ifdef CONFIG_SYSFS
116static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
117 char *page)
118{
119 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
120}
121
122static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
123
124static __init int kernel_panic_sysfs_init(void)
125{
126 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
127 return 0;
128}
129late_initcall(kernel_panic_sysfs_init);
130#endif
131
132static long no_blink(int state)
133{
134 return 0;
135}
136
137/* Returns how long it waited in ms */
138long (*panic_blink)(int state);
139EXPORT_SYMBOL(panic_blink);
140
141/*
142 * Stop ourself in panic -- architecture code may override this
143 */
144void __weak __noreturn panic_smp_self_stop(void)
145{
146 while (1)
147 cpu_relax();
148}
149
150/*
151 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
152 * may override this to prepare for crash dumping, e.g. save regs info.
153 */
154void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
155{
156 panic_smp_self_stop();
157}
158
159/*
160 * Stop other CPUs in panic. Architecture dependent code may override this
161 * with more suitable version. For example, if the architecture supports
162 * crash dump, it should save registers of each stopped CPU and disable
163 * per-CPU features such as virtualization extensions.
164 */
165void __weak crash_smp_send_stop(void)
166{
167 static int cpus_stopped;
168
169 /*
170 * This function can be called twice in panic path, but obviously
171 * we execute this only once.
172 */
173 if (cpus_stopped)
174 return;
175
176 /*
177 * Note smp_send_stop is the usual smp shutdown function, which
178 * unfortunately means it may not be hardened to work in a panic
179 * situation.
180 */
181 smp_send_stop();
182 cpus_stopped = 1;
183}
184
185atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
186
187/*
188 * A variant of panic() called from NMI context. We return if we've already
189 * panicked on this CPU. If another CPU already panicked, loop in
190 * nmi_panic_self_stop() which can provide architecture dependent code such
191 * as saving register state for crash dump.
192 */
193void nmi_panic(struct pt_regs *regs, const char *msg)
194{
195 int old_cpu, this_cpu;
196
197 old_cpu = PANIC_CPU_INVALID;
198 this_cpu = raw_smp_processor_id();
199
200 /* atomic_try_cmpxchg updates old_cpu on failure */
201 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu))
202 panic("%s", msg);
203 else if (old_cpu != this_cpu)
204 nmi_panic_self_stop(regs);
205}
206EXPORT_SYMBOL(nmi_panic);
207
208static void panic_print_sys_info(bool console_flush)
209{
210 if (console_flush) {
211 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
212 console_flush_on_panic(CONSOLE_REPLAY_ALL);
213 return;
214 }
215
216 if (panic_print & PANIC_PRINT_TASK_INFO)
217 show_state();
218
219 if (panic_print & PANIC_PRINT_MEM_INFO)
220 show_mem();
221
222 if (panic_print & PANIC_PRINT_TIMER_INFO)
223 sysrq_timer_list_show();
224
225 if (panic_print & PANIC_PRINT_LOCK_INFO)
226 debug_show_all_locks();
227
228 if (panic_print & PANIC_PRINT_FTRACE_INFO)
229 ftrace_dump(DUMP_ALL);
230}
231
232void check_panic_on_warn(const char *origin)
233{
234 unsigned int limit;
235
236 if (panic_on_warn)
237 panic("%s: panic_on_warn set ...\n", origin);
238
239 limit = READ_ONCE(warn_limit);
240 if (atomic_inc_return(&warn_count) >= limit && limit)
241 panic("%s: system warned too often (kernel.warn_limit is %d)",
242 origin, limit);
243}
244
245/*
246 * Helper that triggers the NMI backtrace (if set in panic_print)
247 * and then performs the secondary CPUs shutdown - we cannot have
248 * the NMI backtrace after the CPUs are off!
249 */
250static void panic_other_cpus_shutdown(bool crash_kexec)
251{
252 if (panic_print & PANIC_PRINT_ALL_CPU_BT)
253 trigger_all_cpu_backtrace();
254
255 /*
256 * Note that smp_send_stop() is the usual SMP shutdown function,
257 * which unfortunately may not be hardened to work in a panic
258 * situation. If we want to do crash dump after notifier calls
259 * and kmsg_dump, we will need architecture dependent extra
260 * bits in addition to stopping other CPUs, hence we rely on
261 * crash_smp_send_stop() for that.
262 */
263 if (!crash_kexec)
264 smp_send_stop();
265 else
266 crash_smp_send_stop();
267}
268
269/**
270 * panic - halt the system
271 * @fmt: The text string to print
272 *
273 * Display a message, then perform cleanups.
274 *
275 * This function never returns.
276 */
277void panic(const char *fmt, ...)
278{
279 static char buf[1024];
280 va_list args;
281 long i, i_next = 0, len;
282 int state = 0;
283 int old_cpu, this_cpu;
284 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
285
286 if (panic_on_warn) {
287 /*
288 * This thread may hit another WARN() in the panic path.
289 * Resetting this prevents additional WARN() from panicking the
290 * system on this thread. Other threads are blocked by the
291 * panic_mutex in panic().
292 */
293 panic_on_warn = 0;
294 }
295
296 /*
297 * Disable local interrupts. This will prevent panic_smp_self_stop
298 * from deadlocking the first cpu that invokes the panic, since
299 * there is nothing to prevent an interrupt handler (that runs
300 * after setting panic_cpu) from invoking panic() again.
301 */
302 local_irq_disable();
303 preempt_disable_notrace();
304
305 /*
306 * It's possible to come here directly from a panic-assertion and
307 * not have preempt disabled. Some functions called from here want
308 * preempt to be disabled. No point enabling it later though...
309 *
310 * Only one CPU is allowed to execute the panic code from here. For
311 * multiple parallel invocations of panic, all other CPUs either
312 * stop themself or will wait until they are stopped by the 1st CPU
313 * with smp_send_stop().
314 *
315 * cmpxchg success means this is the 1st CPU which comes here,
316 * so go ahead.
317 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
318 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
319 */
320 old_cpu = PANIC_CPU_INVALID;
321 this_cpu = raw_smp_processor_id();
322
323 /* atomic_try_cmpxchg updates old_cpu on failure */
324 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
325 /* go ahead */
326 } else if (old_cpu != this_cpu)
327 panic_smp_self_stop();
328
329 console_verbose();
330 bust_spinlocks(1);
331 va_start(args, fmt);
332 len = vscnprintf(buf, sizeof(buf), fmt, args);
333 va_end(args);
334
335 if (len && buf[len - 1] == '\n')
336 buf[len - 1] = '\0';
337
338 pr_emerg("Kernel panic - not syncing: %s\n", buf);
339#ifdef CONFIG_DEBUG_BUGVERBOSE
340 /*
341 * Avoid nested stack-dumping if a panic occurs during oops processing
342 */
343 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
344 dump_stack();
345#endif
346
347 /*
348 * If kgdb is enabled, give it a chance to run before we stop all
349 * the other CPUs or else we won't be able to debug processes left
350 * running on them.
351 */
352 kgdb_panic(buf);
353
354 /*
355 * If we have crashed and we have a crash kernel loaded let it handle
356 * everything else.
357 * If we want to run this after calling panic_notifiers, pass
358 * the "crash_kexec_post_notifiers" option to the kernel.
359 *
360 * Bypass the panic_cpu check and call __crash_kexec directly.
361 */
362 if (!_crash_kexec_post_notifiers)
363 __crash_kexec(NULL);
364
365 panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
366
367 /*
368 * Run any panic handlers, including those that might need to
369 * add information to the kmsg dump output.
370 */
371 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
372
373 panic_print_sys_info(false);
374
375 kmsg_dump(KMSG_DUMP_PANIC);
376
377 /*
378 * If you doubt kdump always works fine in any situation,
379 * "crash_kexec_post_notifiers" offers you a chance to run
380 * panic_notifiers and dumping kmsg before kdump.
381 * Note: since some panic_notifiers can make crashed kernel
382 * more unstable, it can increase risks of the kdump failure too.
383 *
384 * Bypass the panic_cpu check and call __crash_kexec directly.
385 */
386 if (_crash_kexec_post_notifiers)
387 __crash_kexec(NULL);
388
389 console_unblank();
390
391 /*
392 * We may have ended up stopping the CPU holding the lock (in
393 * smp_send_stop()) while still having some valuable data in the console
394 * buffer. Try to acquire the lock then release it regardless of the
395 * result. The release will also print the buffers out. Locks debug
396 * should be disabled to avoid reporting bad unlock balance when
397 * panic() is not being callled from OOPS.
398 */
399 debug_locks_off();
400 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
401
402 panic_print_sys_info(true);
403
404 if (!panic_blink)
405 panic_blink = no_blink;
406
407 if (panic_timeout > 0) {
408 /*
409 * Delay timeout seconds before rebooting the machine.
410 * We can't use the "normal" timers since we just panicked.
411 */
412 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
413
414 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
415 touch_nmi_watchdog();
416 if (i >= i_next) {
417 i += panic_blink(state ^= 1);
418 i_next = i + 3600 / PANIC_BLINK_SPD;
419 }
420 mdelay(PANIC_TIMER_STEP);
421 }
422 }
423 if (panic_timeout != 0) {
424 /*
425 * This will not be a clean reboot, with everything
426 * shutting down. But if there is a chance of
427 * rebooting the system it will be rebooted.
428 */
429 if (panic_reboot_mode != REBOOT_UNDEFINED)
430 reboot_mode = panic_reboot_mode;
431 emergency_restart();
432 }
433#ifdef __sparc__
434 {
435 extern int stop_a_enabled;
436 /* Make sure the user can actually press Stop-A (L1-A) */
437 stop_a_enabled = 1;
438 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
439 "twice on console to return to the boot prom\n");
440 }
441#endif
442#if defined(CONFIG_S390)
443 disabled_wait();
444#endif
445 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
446
447 /* Do not scroll important messages printed above */
448 suppress_printk = 1;
449 local_irq_enable();
450 for (i = 0; ; i += PANIC_TIMER_STEP) {
451 touch_softlockup_watchdog();
452 if (i >= i_next) {
453 i += panic_blink(state ^= 1);
454 i_next = i + 3600 / PANIC_BLINK_SPD;
455 }
456 mdelay(PANIC_TIMER_STEP);
457 }
458}
459
460EXPORT_SYMBOL(panic);
461
462/*
463 * TAINT_FORCED_RMMOD could be a per-module flag but the module
464 * is being removed anyway.
465 */
466const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
467 [ TAINT_PROPRIETARY_MODULE ] = { 'P', 'G', true },
468 [ TAINT_FORCED_MODULE ] = { 'F', ' ', true },
469 [ TAINT_CPU_OUT_OF_SPEC ] = { 'S', ' ', false },
470 [ TAINT_FORCED_RMMOD ] = { 'R', ' ', false },
471 [ TAINT_MACHINE_CHECK ] = { 'M', ' ', false },
472 [ TAINT_BAD_PAGE ] = { 'B', ' ', false },
473 [ TAINT_USER ] = { 'U', ' ', false },
474 [ TAINT_DIE ] = { 'D', ' ', false },
475 [ TAINT_OVERRIDDEN_ACPI_TABLE ] = { 'A', ' ', false },
476 [ TAINT_WARN ] = { 'W', ' ', false },
477 [ TAINT_CRAP ] = { 'C', ' ', true },
478 [ TAINT_FIRMWARE_WORKAROUND ] = { 'I', ' ', false },
479 [ TAINT_OOT_MODULE ] = { 'O', ' ', true },
480 [ TAINT_UNSIGNED_MODULE ] = { 'E', ' ', true },
481 [ TAINT_SOFTLOCKUP ] = { 'L', ' ', false },
482 [ TAINT_LIVEPATCH ] = { 'K', ' ', true },
483 [ TAINT_AUX ] = { 'X', ' ', true },
484 [ TAINT_RANDSTRUCT ] = { 'T', ' ', true },
485 [ TAINT_TEST ] = { 'N', ' ', true },
486};
487
488/**
489 * print_tainted - return a string to represent the kernel taint state.
490 *
491 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
492 *
493 * The string is overwritten by the next call to print_tainted(),
494 * but is always NULL terminated.
495 */
496const char *print_tainted(void)
497{
498 static char buf[TAINT_FLAGS_COUNT + sizeof("Tainted: ")];
499
500 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
501
502 if (tainted_mask) {
503 char *s;
504 int i;
505
506 s = buf + sprintf(buf, "Tainted: ");
507 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
508 const struct taint_flag *t = &taint_flags[i];
509 *s++ = test_bit(i, &tainted_mask) ?
510 t->c_true : t->c_false;
511 }
512 *s = 0;
513 } else
514 snprintf(buf, sizeof(buf), "Not tainted");
515
516 return buf;
517}
518
519int test_taint(unsigned flag)
520{
521 return test_bit(flag, &tainted_mask);
522}
523EXPORT_SYMBOL(test_taint);
524
525unsigned long get_taint(void)
526{
527 return tainted_mask;
528}
529
530/**
531 * add_taint: add a taint flag if not already set.
532 * @flag: one of the TAINT_* constants.
533 * @lockdep_ok: whether lock debugging is still OK.
534 *
535 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
536 * some notewortht-but-not-corrupting cases, it can be set to true.
537 */
538void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
539{
540 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
541 pr_warn("Disabling lock debugging due to kernel taint\n");
542
543 set_bit(flag, &tainted_mask);
544
545 if (tainted_mask & panic_on_taint) {
546 panic_on_taint = 0;
547 panic("panic_on_taint set ...");
548 }
549}
550EXPORT_SYMBOL(add_taint);
551
552static void spin_msec(int msecs)
553{
554 int i;
555
556 for (i = 0; i < msecs; i++) {
557 touch_nmi_watchdog();
558 mdelay(1);
559 }
560}
561
562/*
563 * It just happens that oops_enter() and oops_exit() are identically
564 * implemented...
565 */
566static void do_oops_enter_exit(void)
567{
568 unsigned long flags;
569 static int spin_counter;
570
571 if (!pause_on_oops)
572 return;
573
574 spin_lock_irqsave(&pause_on_oops_lock, flags);
575 if (pause_on_oops_flag == 0) {
576 /* This CPU may now print the oops message */
577 pause_on_oops_flag = 1;
578 } else {
579 /* We need to stall this CPU */
580 if (!spin_counter) {
581 /* This CPU gets to do the counting */
582 spin_counter = pause_on_oops;
583 do {
584 spin_unlock(&pause_on_oops_lock);
585 spin_msec(MSEC_PER_SEC);
586 spin_lock(&pause_on_oops_lock);
587 } while (--spin_counter);
588 pause_on_oops_flag = 0;
589 } else {
590 /* This CPU waits for a different one */
591 while (spin_counter) {
592 spin_unlock(&pause_on_oops_lock);
593 spin_msec(1);
594 spin_lock(&pause_on_oops_lock);
595 }
596 }
597 }
598 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
599}
600
601/*
602 * Return true if the calling CPU is allowed to print oops-related info.
603 * This is a bit racy..
604 */
605bool oops_may_print(void)
606{
607 return pause_on_oops_flag == 0;
608}
609
610/*
611 * Called when the architecture enters its oops handler, before it prints
612 * anything. If this is the first CPU to oops, and it's oopsing the first
613 * time then let it proceed.
614 *
615 * This is all enabled by the pause_on_oops kernel boot option. We do all
616 * this to ensure that oopses don't scroll off the screen. It has the
617 * side-effect of preventing later-oopsing CPUs from mucking up the display,
618 * too.
619 *
620 * It turns out that the CPU which is allowed to print ends up pausing for
621 * the right duration, whereas all the other CPUs pause for twice as long:
622 * once in oops_enter(), once in oops_exit().
623 */
624void oops_enter(void)
625{
626 tracing_off();
627 /* can't trust the integrity of the kernel anymore: */
628 debug_locks_off();
629 do_oops_enter_exit();
630
631 if (sysctl_oops_all_cpu_backtrace)
632 trigger_all_cpu_backtrace();
633}
634
635static void print_oops_end_marker(void)
636{
637 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
638}
639
640/*
641 * Called when the architecture exits its oops handler, after printing
642 * everything.
643 */
644void oops_exit(void)
645{
646 do_oops_enter_exit();
647 print_oops_end_marker();
648 kmsg_dump(KMSG_DUMP_OOPS);
649}
650
651struct warn_args {
652 const char *fmt;
653 va_list args;
654};
655
656void __warn(const char *file, int line, void *caller, unsigned taint,
657 struct pt_regs *regs, struct warn_args *args)
658{
659 disable_trace_on_warning();
660
661 if (file)
662 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
663 raw_smp_processor_id(), current->pid, file, line,
664 caller);
665 else
666 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
667 raw_smp_processor_id(), current->pid, caller);
668
669 if (args)
670 vprintk(args->fmt, args->args);
671
672 print_modules();
673
674 if (regs)
675 show_regs(regs);
676
677 check_panic_on_warn("kernel");
678
679 if (!regs)
680 dump_stack();
681
682 print_irqtrace_events(current);
683
684 print_oops_end_marker();
685 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
686
687 /* Just a warning, don't kill lockdep. */
688 add_taint(taint, LOCKDEP_STILL_OK);
689}
690
691#ifdef CONFIG_BUG
692#ifndef __WARN_FLAGS
693void warn_slowpath_fmt(const char *file, int line, unsigned taint,
694 const char *fmt, ...)
695{
696 bool rcu = warn_rcu_enter();
697 struct warn_args args;
698
699 pr_warn(CUT_HERE);
700
701 if (!fmt) {
702 __warn(file, line, __builtin_return_address(0), taint,
703 NULL, NULL);
704 warn_rcu_exit(rcu);
705 return;
706 }
707
708 args.fmt = fmt;
709 va_start(args.args, fmt);
710 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
711 va_end(args.args);
712 warn_rcu_exit(rcu);
713}
714EXPORT_SYMBOL(warn_slowpath_fmt);
715#else
716void __warn_printk(const char *fmt, ...)
717{
718 bool rcu = warn_rcu_enter();
719 va_list args;
720
721 pr_warn(CUT_HERE);
722
723 va_start(args, fmt);
724 vprintk(fmt, args);
725 va_end(args);
726 warn_rcu_exit(rcu);
727}
728EXPORT_SYMBOL(__warn_printk);
729#endif
730
731/* Support resetting WARN*_ONCE state */
732
733static int clear_warn_once_set(void *data, u64 val)
734{
735 generic_bug_clear_once();
736 memset(__start_once, 0, __end_once - __start_once);
737 return 0;
738}
739
740DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
741 "%lld\n");
742
743static __init int register_warn_debugfs(void)
744{
745 /* Don't care about failure */
746 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
747 &clear_warn_once_fops);
748 return 0;
749}
750
751device_initcall(register_warn_debugfs);
752#endif
753
754#ifdef CONFIG_STACKPROTECTOR
755
756/*
757 * Called when gcc's -fstack-protector feature is used, and
758 * gcc detects corruption of the on-stack canary value
759 */
760__visible noinstr void __stack_chk_fail(void)
761{
762 instrumentation_begin();
763 panic("stack-protector: Kernel stack is corrupted in: %pB",
764 __builtin_return_address(0));
765 instrumentation_end();
766}
767EXPORT_SYMBOL(__stack_chk_fail);
768
769#endif
770
771core_param(panic, panic_timeout, int, 0644);
772core_param(panic_print, panic_print, ulong, 0644);
773core_param(pause_on_oops, pause_on_oops, int, 0644);
774core_param(panic_on_warn, panic_on_warn, int, 0644);
775core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
776
777static int __init oops_setup(char *s)
778{
779 if (!s)
780 return -EINVAL;
781 if (!strcmp(s, "panic"))
782 panic_on_oops = 1;
783 return 0;
784}
785early_param("oops", oops_setup);
786
787static int __init panic_on_taint_setup(char *s)
788{
789 char *taint_str;
790
791 if (!s)
792 return -EINVAL;
793
794 taint_str = strsep(&s, ",");
795 if (kstrtoul(taint_str, 16, &panic_on_taint))
796 return -EINVAL;
797
798 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
799 panic_on_taint &= TAINT_FLAGS_MAX;
800
801 if (!panic_on_taint)
802 return -EINVAL;
803
804 if (s && !strcmp(s, "nousertaint"))
805 panic_on_taint_nousertaint = true;
806
807 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
808 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
809
810 return 0;
811}
812early_param("panic_on_taint", panic_on_taint_setup);