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
74atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
75
76/*
77 * A variant of panic() called from NMI context. We return if we've already
78 * panicked on this CPU. If another CPU already panicked, loop in
79 * nmi_panic_self_stop() which can provide architecture dependent code such
80 * as saving register state for crash dump.
81 */
82void nmi_panic(struct pt_regs *regs, const char *msg)
83{
84 int old_cpu, cpu;
85
86 cpu = raw_smp_processor_id();
87 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
88
89 if (old_cpu == PANIC_CPU_INVALID)
90 panic("%s", msg);
91 else if (old_cpu != cpu)
92 nmi_panic_self_stop(regs);
93}
94EXPORT_SYMBOL(nmi_panic);
95
96/**
97 * panic - halt the system
98 * @fmt: The text string to print
99 *
100 * Display a message, then perform cleanups.
101 *
102 * This function never returns.
103 */
104void panic(const char *fmt, ...)
105{
106 static char buf[1024];
107 va_list args;
108 long i, i_next = 0;
109 int state = 0;
110 int old_cpu, this_cpu;
111
112 /*
113 * Disable local interrupts. This will prevent panic_smp_self_stop
114 * from deadlocking the first cpu that invokes the panic, since
115 * there is nothing to prevent an interrupt handler (that runs
116 * after setting panic_cpu) from invoking panic() again.
117 */
118 local_irq_disable();
119
120 /*
121 * It's possible to come here directly from a panic-assertion and
122 * not have preempt disabled. Some functions called from here want
123 * preempt to be disabled. No point enabling it later though...
124 *
125 * Only one CPU is allowed to execute the panic code from here. For
126 * multiple parallel invocations of panic, all other CPUs either
127 * stop themself or will wait until they are stopped by the 1st CPU
128 * with smp_send_stop().
129 *
130 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
131 * comes here, so go ahead.
132 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
133 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
134 */
135 this_cpu = raw_smp_processor_id();
136 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
137
138 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
139 panic_smp_self_stop();
140
141 console_verbose();
142 bust_spinlocks(1);
143 va_start(args, fmt);
144 vsnprintf(buf, sizeof(buf), fmt, args);
145 va_end(args);
146 pr_emerg("Kernel panic - not syncing: %s\n", buf);
147#ifdef CONFIG_DEBUG_BUGVERBOSE
148 /*
149 * Avoid nested stack-dumping if a panic occurs during oops processing
150 */
151 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
152 dump_stack();
153#endif
154
155 /*
156 * If we have crashed and we have a crash kernel loaded let it handle
157 * everything else.
158 * If we want to run this after calling panic_notifiers, pass
159 * the "crash_kexec_post_notifiers" option to the kernel.
160 *
161 * Bypass the panic_cpu check and call __crash_kexec directly.
162 */
163 if (!crash_kexec_post_notifiers)
164 __crash_kexec(NULL);
165
166 /*
167 * Note smp_send_stop is the usual smp shutdown function, which
168 * unfortunately means it may not be hardened to work in a panic
169 * situation.
170 */
171 smp_send_stop();
172
173 /*
174 * Run any panic handlers, including those that might need to
175 * add information to the kmsg dump output.
176 */
177 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
178
179 kmsg_dump(KMSG_DUMP_PANIC);
180
181 /*
182 * If you doubt kdump always works fine in any situation,
183 * "crash_kexec_post_notifiers" offers you a chance to run
184 * panic_notifiers and dumping kmsg before kdump.
185 * Note: since some panic_notifiers can make crashed kernel
186 * more unstable, it can increase risks of the kdump failure too.
187 *
188 * Bypass the panic_cpu check and call __crash_kexec directly.
189 */
190 if (crash_kexec_post_notifiers)
191 __crash_kexec(NULL);
192
193 bust_spinlocks(0);
194
195 /*
196 * We may have ended up stopping the CPU holding the lock (in
197 * smp_send_stop()) while still having some valuable data in the console
198 * buffer. Try to acquire the lock then release it regardless of the
199 * result. The release will also print the buffers out. Locks debug
200 * should be disabled to avoid reporting bad unlock balance when
201 * panic() is not being callled from OOPS.
202 */
203 debug_locks_off();
204 console_flush_on_panic();
205
206 if (!panic_blink)
207 panic_blink = no_blink;
208
209 if (panic_timeout > 0) {
210 /*
211 * Delay timeout seconds before rebooting the machine.
212 * We can't use the "normal" timers since we just panicked.
213 */
214 pr_emerg("Rebooting in %d seconds..", panic_timeout);
215
216 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
217 touch_nmi_watchdog();
218 if (i >= i_next) {
219 i += panic_blink(state ^= 1);
220 i_next = i + 3600 / PANIC_BLINK_SPD;
221 }
222 mdelay(PANIC_TIMER_STEP);
223 }
224 }
225 if (panic_timeout != 0) {
226 /*
227 * This will not be a clean reboot, with everything
228 * shutting down. But if there is a chance of
229 * rebooting the system it will be rebooted.
230 */
231 emergency_restart();
232 }
233#ifdef __sparc__
234 {
235 extern int stop_a_enabled;
236 /* Make sure the user can actually press Stop-A (L1-A) */
237 stop_a_enabled = 1;
238 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
239 }
240#endif
241#if defined(CONFIG_S390)
242 {
243 unsigned long caller;
244
245 caller = (unsigned long)__builtin_return_address(0);
246 disabled_wait(caller);
247 }
248#endif
249 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
250 local_irq_enable();
251 for (i = 0; ; i += PANIC_TIMER_STEP) {
252 touch_softlockup_watchdog();
253 if (i >= i_next) {
254 i += panic_blink(state ^= 1);
255 i_next = i + 3600 / PANIC_BLINK_SPD;
256 }
257 mdelay(PANIC_TIMER_STEP);
258 }
259}
260
261EXPORT_SYMBOL(panic);
262
263
264struct tnt {
265 u8 bit;
266 char true;
267 char false;
268};
269
270static const struct tnt tnts[] = {
271 { TAINT_PROPRIETARY_MODULE, 'P', 'G' },
272 { TAINT_FORCED_MODULE, 'F', ' ' },
273 { TAINT_CPU_OUT_OF_SPEC, 'S', ' ' },
274 { TAINT_FORCED_RMMOD, 'R', ' ' },
275 { TAINT_MACHINE_CHECK, 'M', ' ' },
276 { TAINT_BAD_PAGE, 'B', ' ' },
277 { TAINT_USER, 'U', ' ' },
278 { TAINT_DIE, 'D', ' ' },
279 { TAINT_OVERRIDDEN_ACPI_TABLE, 'A', ' ' },
280 { TAINT_WARN, 'W', ' ' },
281 { TAINT_CRAP, 'C', ' ' },
282 { TAINT_FIRMWARE_WORKAROUND, 'I', ' ' },
283 { TAINT_OOT_MODULE, 'O', ' ' },
284 { TAINT_UNSIGNED_MODULE, 'E', ' ' },
285 { TAINT_SOFTLOCKUP, 'L', ' ' },
286 { TAINT_LIVEPATCH, 'K', ' ' },
287};
288
289/**
290 * print_tainted - return a string to represent the kernel taint state.
291 *
292 * 'P' - Proprietary module has been loaded.
293 * 'F' - Module has been forcibly loaded.
294 * 'S' - SMP with CPUs not designed for SMP.
295 * 'R' - User forced a module unload.
296 * 'M' - System experienced a machine check exception.
297 * 'B' - System has hit bad_page.
298 * 'U' - Userspace-defined naughtiness.
299 * 'D' - Kernel has oopsed before
300 * 'A' - ACPI table overridden.
301 * 'W' - Taint on warning.
302 * 'C' - modules from drivers/staging are loaded.
303 * 'I' - Working around severe firmware bug.
304 * 'O' - Out-of-tree module has been loaded.
305 * 'E' - Unsigned module has been loaded.
306 * 'L' - A soft lockup has previously occurred.
307 * 'K' - Kernel has been live patched.
308 *
309 * The string is overwritten by the next call to print_tainted().
310 */
311const char *print_tainted(void)
312{
313 static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
314
315 if (tainted_mask) {
316 char *s;
317 int i;
318
319 s = buf + sprintf(buf, "Tainted: ");
320 for (i = 0; i < ARRAY_SIZE(tnts); i++) {
321 const struct tnt *t = &tnts[i];
322 *s++ = test_bit(t->bit, &tainted_mask) ?
323 t->true : t->false;
324 }
325 *s = 0;
326 } else
327 snprintf(buf, sizeof(buf), "Not tainted");
328
329 return buf;
330}
331
332int test_taint(unsigned flag)
333{
334 return test_bit(flag, &tainted_mask);
335}
336EXPORT_SYMBOL(test_taint);
337
338unsigned long get_taint(void)
339{
340 return tainted_mask;
341}
342
343/**
344 * add_taint: add a taint flag if not already set.
345 * @flag: one of the TAINT_* constants.
346 * @lockdep_ok: whether lock debugging is still OK.
347 *
348 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
349 * some notewortht-but-not-corrupting cases, it can be set to true.
350 */
351void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
352{
353 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
354 pr_warn("Disabling lock debugging due to kernel taint\n");
355
356 set_bit(flag, &tainted_mask);
357}
358EXPORT_SYMBOL(add_taint);
359
360static void spin_msec(int msecs)
361{
362 int i;
363
364 for (i = 0; i < msecs; i++) {
365 touch_nmi_watchdog();
366 mdelay(1);
367 }
368}
369
370/*
371 * It just happens that oops_enter() and oops_exit() are identically
372 * implemented...
373 */
374static void do_oops_enter_exit(void)
375{
376 unsigned long flags;
377 static int spin_counter;
378
379 if (!pause_on_oops)
380 return;
381
382 spin_lock_irqsave(&pause_on_oops_lock, flags);
383 if (pause_on_oops_flag == 0) {
384 /* This CPU may now print the oops message */
385 pause_on_oops_flag = 1;
386 } else {
387 /* We need to stall this CPU */
388 if (!spin_counter) {
389 /* This CPU gets to do the counting */
390 spin_counter = pause_on_oops;
391 do {
392 spin_unlock(&pause_on_oops_lock);
393 spin_msec(MSEC_PER_SEC);
394 spin_lock(&pause_on_oops_lock);
395 } while (--spin_counter);
396 pause_on_oops_flag = 0;
397 } else {
398 /* This CPU waits for a different one */
399 while (spin_counter) {
400 spin_unlock(&pause_on_oops_lock);
401 spin_msec(1);
402 spin_lock(&pause_on_oops_lock);
403 }
404 }
405 }
406 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
407}
408
409/*
410 * Return true if the calling CPU is allowed to print oops-related info.
411 * This is a bit racy..
412 */
413int oops_may_print(void)
414{
415 return pause_on_oops_flag == 0;
416}
417
418/*
419 * Called when the architecture enters its oops handler, before it prints
420 * anything. If this is the first CPU to oops, and it's oopsing the first
421 * time then let it proceed.
422 *
423 * This is all enabled by the pause_on_oops kernel boot option. We do all
424 * this to ensure that oopses don't scroll off the screen. It has the
425 * side-effect of preventing later-oopsing CPUs from mucking up the display,
426 * too.
427 *
428 * It turns out that the CPU which is allowed to print ends up pausing for
429 * the right duration, whereas all the other CPUs pause for twice as long:
430 * once in oops_enter(), once in oops_exit().
431 */
432void oops_enter(void)
433{
434 tracing_off();
435 /* can't trust the integrity of the kernel anymore: */
436 debug_locks_off();
437 do_oops_enter_exit();
438}
439
440/*
441 * 64-bit random ID for oopses:
442 */
443static u64 oops_id;
444
445static int init_oops_id(void)
446{
447 if (!oops_id)
448 get_random_bytes(&oops_id, sizeof(oops_id));
449 else
450 oops_id++;
451
452 return 0;
453}
454late_initcall(init_oops_id);
455
456void print_oops_end_marker(void)
457{
458 init_oops_id();
459 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
460}
461
462/*
463 * Called when the architecture exits its oops handler, after printing
464 * everything.
465 */
466void oops_exit(void)
467{
468 do_oops_enter_exit();
469 print_oops_end_marker();
470 kmsg_dump(KMSG_DUMP_OOPS);
471}
472
473struct warn_args {
474 const char *fmt;
475 va_list args;
476};
477
478void __warn(const char *file, int line, void *caller, unsigned taint,
479 struct pt_regs *regs, struct warn_args *args)
480{
481 disable_trace_on_warning();
482
483 pr_warn("------------[ cut here ]------------\n");
484
485 if (file)
486 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
487 raw_smp_processor_id(), current->pid, file, line,
488 caller);
489 else
490 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
491 raw_smp_processor_id(), current->pid, caller);
492
493 if (args)
494 vprintk(args->fmt, args->args);
495
496 if (panic_on_warn) {
497 /*
498 * This thread may hit another WARN() in the panic path.
499 * Resetting this prevents additional WARN() from panicking the
500 * system on this thread. Other threads are blocked by the
501 * panic_mutex in panic().
502 */
503 panic_on_warn = 0;
504 panic("panic_on_warn set ...\n");
505 }
506
507 print_modules();
508
509 if (regs)
510 show_regs(regs);
511 else
512 dump_stack();
513
514 print_oops_end_marker();
515
516 /* Just a warning, don't kill lockdep. */
517 add_taint(taint, LOCKDEP_STILL_OK);
518}
519
520#ifdef WANT_WARN_ON_SLOWPATH
521void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
522{
523 struct warn_args args;
524
525 args.fmt = fmt;
526 va_start(args.args, fmt);
527 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
528 &args);
529 va_end(args.args);
530}
531EXPORT_SYMBOL(warn_slowpath_fmt);
532
533void warn_slowpath_fmt_taint(const char *file, int line,
534 unsigned taint, const char *fmt, ...)
535{
536 struct warn_args args;
537
538 args.fmt = fmt;
539 va_start(args.args, fmt);
540 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
541 va_end(args.args);
542}
543EXPORT_SYMBOL(warn_slowpath_fmt_taint);
544
545void warn_slowpath_null(const char *file, int line)
546{
547 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
548}
549EXPORT_SYMBOL(warn_slowpath_null);
550#endif
551
552#ifdef CONFIG_CC_STACKPROTECTOR
553
554/*
555 * Called when gcc's -fstack-protector feature is used, and
556 * gcc detects corruption of the on-stack canary value
557 */
558__visible void __stack_chk_fail(void)
559{
560 panic("stack-protector: Kernel stack is corrupted in: %p\n",
561 __builtin_return_address(0));
562}
563EXPORT_SYMBOL(__stack_chk_fail);
564
565#endif
566
567core_param(panic, panic_timeout, int, 0644);
568core_param(pause_on_oops, pause_on_oops, int, 0644);
569core_param(panic_on_warn, panic_on_warn, int, 0644);
570
571static int __init setup_crash_kexec_post_notifiers(char *s)
572{
573 crash_kexec_post_notifiers = true;
574 return 0;
575}
576early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
577
578static int __init oops_setup(char *s)
579{
580 if (!s)
581 return -EINVAL;
582 if (!strcmp(s, "panic"))
583 panic_on_oops = 1;
584 return 0;
585}
586early_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 <linux/seq_buf.h>
39#include <trace/events/error_report.h>
40#include <asm/sections.h>
41
42#define PANIC_TIMER_STEP 100
43#define PANIC_BLINK_SPD 18
44
45#ifdef CONFIG_SMP
46/*
47 * Should we dump all CPUs backtraces in an oops event?
48 * Defaults to 0, can be changed via sysctl.
49 */
50static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace;
51#else
52#define sysctl_oops_all_cpu_backtrace 0
53#endif /* CONFIG_SMP */
54
55int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
56static unsigned long tainted_mask =
57 IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0;
58static int pause_on_oops;
59static int pause_on_oops_flag;
60static DEFINE_SPINLOCK(pause_on_oops_lock);
61bool crash_kexec_post_notifiers;
62int panic_on_warn __read_mostly;
63unsigned long panic_on_taint;
64bool panic_on_taint_nousertaint = false;
65static unsigned int warn_limit __read_mostly;
66
67bool panic_triggering_all_cpu_backtrace;
68
69int panic_timeout = CONFIG_PANIC_TIMEOUT;
70EXPORT_SYMBOL_GPL(panic_timeout);
71
72#define PANIC_PRINT_TASK_INFO 0x00000001
73#define PANIC_PRINT_MEM_INFO 0x00000002
74#define PANIC_PRINT_TIMER_INFO 0x00000004
75#define PANIC_PRINT_LOCK_INFO 0x00000008
76#define PANIC_PRINT_FTRACE_INFO 0x00000010
77#define PANIC_PRINT_ALL_PRINTK_MSG 0x00000020
78#define PANIC_PRINT_ALL_CPU_BT 0x00000040
79#define PANIC_PRINT_BLOCKED_TASKS 0x00000080
80unsigned long panic_print;
81
82ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
83
84EXPORT_SYMBOL(panic_notifier_list);
85
86#ifdef CONFIG_SYSCTL
87static struct ctl_table kern_panic_table[] = {
88#ifdef CONFIG_SMP
89 {
90 .procname = "oops_all_cpu_backtrace",
91 .data = &sysctl_oops_all_cpu_backtrace,
92 .maxlen = sizeof(int),
93 .mode = 0644,
94 .proc_handler = proc_dointvec_minmax,
95 .extra1 = SYSCTL_ZERO,
96 .extra2 = SYSCTL_ONE,
97 },
98#endif
99 {
100 .procname = "warn_limit",
101 .data = &warn_limit,
102 .maxlen = sizeof(warn_limit),
103 .mode = 0644,
104 .proc_handler = proc_douintvec,
105 },
106};
107
108static __init int kernel_panic_sysctls_init(void)
109{
110 register_sysctl_init("kernel", kern_panic_table);
111 return 0;
112}
113late_initcall(kernel_panic_sysctls_init);
114#endif
115
116static atomic_t warn_count = ATOMIC_INIT(0);
117
118#ifdef CONFIG_SYSFS
119static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr,
120 char *page)
121{
122 return sysfs_emit(page, "%d\n", atomic_read(&warn_count));
123}
124
125static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count);
126
127static __init int kernel_panic_sysfs_init(void)
128{
129 sysfs_add_file_to_group(kernel_kobj, &warn_count_attr.attr, NULL);
130 return 0;
131}
132late_initcall(kernel_panic_sysfs_init);
133#endif
134
135static long no_blink(int state)
136{
137 return 0;
138}
139
140/* Returns how long it waited in ms */
141long (*panic_blink)(int state);
142EXPORT_SYMBOL(panic_blink);
143
144/*
145 * Stop ourself in panic -- architecture code may override this
146 */
147void __weak __noreturn panic_smp_self_stop(void)
148{
149 while (1)
150 cpu_relax();
151}
152
153/*
154 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
155 * may override this to prepare for crash dumping, e.g. save regs info.
156 */
157void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs)
158{
159 panic_smp_self_stop();
160}
161
162/*
163 * Stop other CPUs in panic. Architecture dependent code may override this
164 * with more suitable version. For example, if the architecture supports
165 * crash dump, it should save registers of each stopped CPU and disable
166 * per-CPU features such as virtualization extensions.
167 */
168void __weak crash_smp_send_stop(void)
169{
170 static int cpus_stopped;
171
172 /*
173 * This function can be called twice in panic path, but obviously
174 * we execute this only once.
175 */
176 if (cpus_stopped)
177 return;
178
179 /*
180 * Note smp_send_stop is the usual smp shutdown function, which
181 * unfortunately means it may not be hardened to work in a panic
182 * situation.
183 */
184 smp_send_stop();
185 cpus_stopped = 1;
186}
187
188atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
189
190/*
191 * A variant of panic() called from NMI context. We return if we've already
192 * panicked on this CPU. If another CPU already panicked, loop in
193 * nmi_panic_self_stop() which can provide architecture dependent code such
194 * as saving register state for crash dump.
195 */
196void nmi_panic(struct pt_regs *regs, const char *msg)
197{
198 int old_cpu, this_cpu;
199
200 old_cpu = PANIC_CPU_INVALID;
201 this_cpu = raw_smp_processor_id();
202
203 /* atomic_try_cmpxchg updates old_cpu on failure */
204 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu))
205 panic("%s", msg);
206 else if (old_cpu != this_cpu)
207 nmi_panic_self_stop(regs);
208}
209EXPORT_SYMBOL(nmi_panic);
210
211static void panic_print_sys_info(bool console_flush)
212{
213 if (console_flush) {
214 if (panic_print & PANIC_PRINT_ALL_PRINTK_MSG)
215 console_flush_on_panic(CONSOLE_REPLAY_ALL);
216 return;
217 }
218
219 if (panic_print & PANIC_PRINT_TASK_INFO)
220 show_state();
221
222 if (panic_print & PANIC_PRINT_MEM_INFO)
223 show_mem();
224
225 if (panic_print & PANIC_PRINT_TIMER_INFO)
226 sysrq_timer_list_show();
227
228 if (panic_print & PANIC_PRINT_LOCK_INFO)
229 debug_show_all_locks();
230
231 if (panic_print & PANIC_PRINT_FTRACE_INFO)
232 ftrace_dump(DUMP_ALL);
233
234 if (panic_print & PANIC_PRINT_BLOCKED_TASKS)
235 show_state_filter(TASK_UNINTERRUPTIBLE);
236}
237
238void check_panic_on_warn(const char *origin)
239{
240 unsigned int limit;
241
242 if (panic_on_warn)
243 panic("%s: panic_on_warn set ...\n", origin);
244
245 limit = READ_ONCE(warn_limit);
246 if (atomic_inc_return(&warn_count) >= limit && limit)
247 panic("%s: system warned too often (kernel.warn_limit is %d)",
248 origin, limit);
249}
250
251/*
252 * Helper that triggers the NMI backtrace (if set in panic_print)
253 * and then performs the secondary CPUs shutdown - we cannot have
254 * the NMI backtrace after the CPUs are off!
255 */
256static void panic_other_cpus_shutdown(bool crash_kexec)
257{
258 if (panic_print & PANIC_PRINT_ALL_CPU_BT) {
259 /* Temporary allow non-panic CPUs to write their backtraces. */
260 panic_triggering_all_cpu_backtrace = true;
261 trigger_all_cpu_backtrace();
262 panic_triggering_all_cpu_backtrace = false;
263 }
264
265 /*
266 * Note that smp_send_stop() is the usual SMP shutdown function,
267 * which unfortunately may not be hardened to work in a panic
268 * situation. If we want to do crash dump after notifier calls
269 * and kmsg_dump, we will need architecture dependent extra
270 * bits in addition to stopping other CPUs, hence we rely on
271 * crash_smp_send_stop() for that.
272 */
273 if (!crash_kexec)
274 smp_send_stop();
275 else
276 crash_smp_send_stop();
277}
278
279/**
280 * panic - halt the system
281 * @fmt: The text string to print
282 *
283 * Display a message, then perform cleanups.
284 *
285 * This function never returns.
286 */
287void panic(const char *fmt, ...)
288{
289 static char buf[1024];
290 va_list args;
291 long i, i_next = 0, len;
292 int state = 0;
293 int old_cpu, this_cpu;
294 bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers;
295
296 if (panic_on_warn) {
297 /*
298 * This thread may hit another WARN() in the panic path.
299 * Resetting this prevents additional WARN() from panicking the
300 * system on this thread. Other threads are blocked by the
301 * panic_mutex in panic().
302 */
303 panic_on_warn = 0;
304 }
305
306 /*
307 * Disable local interrupts. This will prevent panic_smp_self_stop
308 * from deadlocking the first cpu that invokes the panic, since
309 * there is nothing to prevent an interrupt handler (that runs
310 * after setting panic_cpu) from invoking panic() again.
311 */
312 local_irq_disable();
313 preempt_disable_notrace();
314
315 /*
316 * It's possible to come here directly from a panic-assertion and
317 * not have preempt disabled. Some functions called from here want
318 * preempt to be disabled. No point enabling it later though...
319 *
320 * Only one CPU is allowed to execute the panic code from here. For
321 * multiple parallel invocations of panic, all other CPUs either
322 * stop themself or will wait until they are stopped by the 1st CPU
323 * with smp_send_stop().
324 *
325 * cmpxchg success means this is the 1st CPU which comes here,
326 * so go ahead.
327 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
328 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
329 */
330 old_cpu = PANIC_CPU_INVALID;
331 this_cpu = raw_smp_processor_id();
332
333 /* atomic_try_cmpxchg updates old_cpu on failure */
334 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
335 /* go ahead */
336 } else if (old_cpu != this_cpu)
337 panic_smp_self_stop();
338
339 console_verbose();
340 bust_spinlocks(1);
341 va_start(args, fmt);
342 len = vscnprintf(buf, sizeof(buf), fmt, args);
343 va_end(args);
344
345 if (len && buf[len - 1] == '\n')
346 buf[len - 1] = '\0';
347
348 pr_emerg("Kernel panic - not syncing: %s\n", buf);
349#ifdef CONFIG_DEBUG_BUGVERBOSE
350 /*
351 * Avoid nested stack-dumping if a panic occurs during oops processing
352 */
353 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
354 dump_stack();
355#endif
356
357 /*
358 * If kgdb is enabled, give it a chance to run before we stop all
359 * the other CPUs or else we won't be able to debug processes left
360 * running on them.
361 */
362 kgdb_panic(buf);
363
364 /*
365 * If we have crashed and we have a crash kernel loaded let it handle
366 * everything else.
367 * If we want to run this after calling panic_notifiers, pass
368 * the "crash_kexec_post_notifiers" option to the kernel.
369 *
370 * Bypass the panic_cpu check and call __crash_kexec directly.
371 */
372 if (!_crash_kexec_post_notifiers)
373 __crash_kexec(NULL);
374
375 panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
376
377 printk_legacy_allow_panic_sync();
378
379 /*
380 * Run any panic handlers, including those that might need to
381 * add information to the kmsg dump output.
382 */
383 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
384
385 panic_print_sys_info(false);
386
387 kmsg_dump_desc(KMSG_DUMP_PANIC, buf);
388
389 /*
390 * If you doubt kdump always works fine in any situation,
391 * "crash_kexec_post_notifiers" offers you a chance to run
392 * panic_notifiers and dumping kmsg before kdump.
393 * Note: since some panic_notifiers can make crashed kernel
394 * more unstable, it can increase risks of the kdump failure too.
395 *
396 * Bypass the panic_cpu check and call __crash_kexec directly.
397 */
398 if (_crash_kexec_post_notifiers)
399 __crash_kexec(NULL);
400
401 console_unblank();
402
403 /*
404 * We may have ended up stopping the CPU holding the lock (in
405 * smp_send_stop()) while still having some valuable data in the console
406 * buffer. Try to acquire the lock then release it regardless of the
407 * result. The release will also print the buffers out. Locks debug
408 * should be disabled to avoid reporting bad unlock balance when
409 * panic() is not being callled from OOPS.
410 */
411 debug_locks_off();
412 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
413
414 panic_print_sys_info(true);
415
416 if (!panic_blink)
417 panic_blink = no_blink;
418
419 if (panic_timeout > 0) {
420 /*
421 * Delay timeout seconds before rebooting the machine.
422 * We can't use the "normal" timers since we just panicked.
423 */
424 pr_emerg("Rebooting in %d seconds..\n", panic_timeout);
425
426 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
427 touch_nmi_watchdog();
428 if (i >= i_next) {
429 i += panic_blink(state ^= 1);
430 i_next = i + 3600 / PANIC_BLINK_SPD;
431 }
432 mdelay(PANIC_TIMER_STEP);
433 }
434 }
435 if (panic_timeout != 0) {
436 /*
437 * This will not be a clean reboot, with everything
438 * shutting down. But if there is a chance of
439 * rebooting the system it will be rebooted.
440 */
441 if (panic_reboot_mode != REBOOT_UNDEFINED)
442 reboot_mode = panic_reboot_mode;
443 emergency_restart();
444 }
445#ifdef __sparc__
446 {
447 extern int stop_a_enabled;
448 /* Make sure the user can actually press Stop-A (L1-A) */
449 stop_a_enabled = 1;
450 pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n"
451 "twice on console to return to the boot prom\n");
452 }
453#endif
454#if defined(CONFIG_S390)
455 disabled_wait();
456#endif
457 pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n", buf);
458
459 /* Do not scroll important messages printed above */
460 suppress_printk = 1;
461
462 /*
463 * The final messages may not have been printed if in a context that
464 * defers printing (such as NMI) and irq_work is not available.
465 * Explicitly flush the kernel log buffer one last time.
466 */
467 console_flush_on_panic(CONSOLE_FLUSH_PENDING);
468 nbcon_atomic_flush_unsafe();
469
470 local_irq_enable();
471 for (i = 0; ; i += PANIC_TIMER_STEP) {
472 touch_softlockup_watchdog();
473 if (i >= i_next) {
474 i += panic_blink(state ^= 1);
475 i_next = i + 3600 / PANIC_BLINK_SPD;
476 }
477 mdelay(PANIC_TIMER_STEP);
478 }
479}
480
481EXPORT_SYMBOL(panic);
482
483#define TAINT_FLAG(taint, _c_true, _c_false, _module) \
484 [ TAINT_##taint ] = { \
485 .c_true = _c_true, .c_false = _c_false, \
486 .module = _module, \
487 .desc = #taint, \
488 }
489
490/*
491 * TAINT_FORCED_RMMOD could be a per-module flag but the module
492 * is being removed anyway.
493 */
494const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = {
495 TAINT_FLAG(PROPRIETARY_MODULE, 'P', 'G', true),
496 TAINT_FLAG(FORCED_MODULE, 'F', ' ', true),
497 TAINT_FLAG(CPU_OUT_OF_SPEC, 'S', ' ', false),
498 TAINT_FLAG(FORCED_RMMOD, 'R', ' ', false),
499 TAINT_FLAG(MACHINE_CHECK, 'M', ' ', false),
500 TAINT_FLAG(BAD_PAGE, 'B', ' ', false),
501 TAINT_FLAG(USER, 'U', ' ', false),
502 TAINT_FLAG(DIE, 'D', ' ', false),
503 TAINT_FLAG(OVERRIDDEN_ACPI_TABLE, 'A', ' ', false),
504 TAINT_FLAG(WARN, 'W', ' ', false),
505 TAINT_FLAG(CRAP, 'C', ' ', true),
506 TAINT_FLAG(FIRMWARE_WORKAROUND, 'I', ' ', false),
507 TAINT_FLAG(OOT_MODULE, 'O', ' ', true),
508 TAINT_FLAG(UNSIGNED_MODULE, 'E', ' ', true),
509 TAINT_FLAG(SOFTLOCKUP, 'L', ' ', false),
510 TAINT_FLAG(LIVEPATCH, 'K', ' ', true),
511 TAINT_FLAG(AUX, 'X', ' ', true),
512 TAINT_FLAG(RANDSTRUCT, 'T', ' ', true),
513 TAINT_FLAG(TEST, 'N', ' ', true),
514};
515
516#undef TAINT_FLAG
517
518static void print_tainted_seq(struct seq_buf *s, bool verbose)
519{
520 const char *sep = "";
521 int i;
522
523 if (!tainted_mask) {
524 seq_buf_puts(s, "Not tainted");
525 return;
526 }
527
528 seq_buf_printf(s, "Tainted: ");
529 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
530 const struct taint_flag *t = &taint_flags[i];
531 bool is_set = test_bit(i, &tainted_mask);
532 char c = is_set ? t->c_true : t->c_false;
533
534 if (verbose) {
535 if (is_set) {
536 seq_buf_printf(s, "%s[%c]=%s", sep, c, t->desc);
537 sep = ", ";
538 }
539 } else {
540 seq_buf_putc(s, c);
541 }
542 }
543}
544
545static const char *_print_tainted(bool verbose)
546{
547 /* FIXME: what should the size be? */
548 static char buf[sizeof(taint_flags)];
549 struct seq_buf s;
550
551 BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT);
552
553 seq_buf_init(&s, buf, sizeof(buf));
554
555 print_tainted_seq(&s, verbose);
556
557 return seq_buf_str(&s);
558}
559
560/**
561 * print_tainted - return a string to represent the kernel taint state.
562 *
563 * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst
564 *
565 * The string is overwritten by the next call to print_tainted(),
566 * but is always NULL terminated.
567 */
568const char *print_tainted(void)
569{
570 return _print_tainted(false);
571}
572
573/**
574 * print_tainted_verbose - A more verbose version of print_tainted()
575 */
576const char *print_tainted_verbose(void)
577{
578 return _print_tainted(true);
579}
580
581int test_taint(unsigned flag)
582{
583 return test_bit(flag, &tainted_mask);
584}
585EXPORT_SYMBOL(test_taint);
586
587unsigned long get_taint(void)
588{
589 return tainted_mask;
590}
591
592/**
593 * add_taint: add a taint flag if not already set.
594 * @flag: one of the TAINT_* constants.
595 * @lockdep_ok: whether lock debugging is still OK.
596 *
597 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
598 * some notewortht-but-not-corrupting cases, it can be set to true.
599 */
600void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
601{
602 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
603 pr_warn("Disabling lock debugging due to kernel taint\n");
604
605 set_bit(flag, &tainted_mask);
606
607 if (tainted_mask & panic_on_taint) {
608 panic_on_taint = 0;
609 panic("panic_on_taint set ...");
610 }
611}
612EXPORT_SYMBOL(add_taint);
613
614static void spin_msec(int msecs)
615{
616 int i;
617
618 for (i = 0; i < msecs; i++) {
619 touch_nmi_watchdog();
620 mdelay(1);
621 }
622}
623
624/*
625 * It just happens that oops_enter() and oops_exit() are identically
626 * implemented...
627 */
628static void do_oops_enter_exit(void)
629{
630 unsigned long flags;
631 static int spin_counter;
632
633 if (!pause_on_oops)
634 return;
635
636 spin_lock_irqsave(&pause_on_oops_lock, flags);
637 if (pause_on_oops_flag == 0) {
638 /* This CPU may now print the oops message */
639 pause_on_oops_flag = 1;
640 } else {
641 /* We need to stall this CPU */
642 if (!spin_counter) {
643 /* This CPU gets to do the counting */
644 spin_counter = pause_on_oops;
645 do {
646 spin_unlock(&pause_on_oops_lock);
647 spin_msec(MSEC_PER_SEC);
648 spin_lock(&pause_on_oops_lock);
649 } while (--spin_counter);
650 pause_on_oops_flag = 0;
651 } else {
652 /* This CPU waits for a different one */
653 while (spin_counter) {
654 spin_unlock(&pause_on_oops_lock);
655 spin_msec(1);
656 spin_lock(&pause_on_oops_lock);
657 }
658 }
659 }
660 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
661}
662
663/*
664 * Return true if the calling CPU is allowed to print oops-related info.
665 * This is a bit racy..
666 */
667bool oops_may_print(void)
668{
669 return pause_on_oops_flag == 0;
670}
671
672/*
673 * Called when the architecture enters its oops handler, before it prints
674 * anything. If this is the first CPU to oops, and it's oopsing the first
675 * time then let it proceed.
676 *
677 * This is all enabled by the pause_on_oops kernel boot option. We do all
678 * this to ensure that oopses don't scroll off the screen. It has the
679 * side-effect of preventing later-oopsing CPUs from mucking up the display,
680 * too.
681 *
682 * It turns out that the CPU which is allowed to print ends up pausing for
683 * the right duration, whereas all the other CPUs pause for twice as long:
684 * once in oops_enter(), once in oops_exit().
685 */
686void oops_enter(void)
687{
688 nbcon_cpu_emergency_enter();
689 tracing_off();
690 /* can't trust the integrity of the kernel anymore: */
691 debug_locks_off();
692 do_oops_enter_exit();
693
694 if (sysctl_oops_all_cpu_backtrace)
695 trigger_all_cpu_backtrace();
696}
697
698static void print_oops_end_marker(void)
699{
700 pr_warn("---[ end trace %016llx ]---\n", 0ULL);
701}
702
703/*
704 * Called when the architecture exits its oops handler, after printing
705 * everything.
706 */
707void oops_exit(void)
708{
709 do_oops_enter_exit();
710 print_oops_end_marker();
711 nbcon_cpu_emergency_exit();
712 kmsg_dump(KMSG_DUMP_OOPS);
713}
714
715struct warn_args {
716 const char *fmt;
717 va_list args;
718};
719
720void __warn(const char *file, int line, void *caller, unsigned taint,
721 struct pt_regs *regs, struct warn_args *args)
722{
723 nbcon_cpu_emergency_enter();
724
725 disable_trace_on_warning();
726
727 if (file)
728 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
729 raw_smp_processor_id(), current->pid, file, line,
730 caller);
731 else
732 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
733 raw_smp_processor_id(), current->pid, caller);
734
735#pragma GCC diagnostic push
736#ifndef __clang__
737#pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
738#endif
739 if (args)
740 vprintk(args->fmt, args->args);
741#pragma GCC diagnostic pop
742
743 print_modules();
744
745 if (regs)
746 show_regs(regs);
747
748 check_panic_on_warn("kernel");
749
750 if (!regs)
751 dump_stack();
752
753 print_irqtrace_events(current);
754
755 print_oops_end_marker();
756 trace_error_report_end(ERROR_DETECTOR_WARN, (unsigned long)caller);
757
758 /* Just a warning, don't kill lockdep. */
759 add_taint(taint, LOCKDEP_STILL_OK);
760
761 nbcon_cpu_emergency_exit();
762}
763
764#ifdef CONFIG_BUG
765#ifndef __WARN_FLAGS
766void warn_slowpath_fmt(const char *file, int line, unsigned taint,
767 const char *fmt, ...)
768{
769 bool rcu = warn_rcu_enter();
770 struct warn_args args;
771
772 pr_warn(CUT_HERE);
773
774 if (!fmt) {
775 __warn(file, line, __builtin_return_address(0), taint,
776 NULL, NULL);
777 warn_rcu_exit(rcu);
778 return;
779 }
780
781 args.fmt = fmt;
782 va_start(args.args, fmt);
783 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
784 va_end(args.args);
785 warn_rcu_exit(rcu);
786}
787EXPORT_SYMBOL(warn_slowpath_fmt);
788#else
789void __warn_printk(const char *fmt, ...)
790{
791 bool rcu = warn_rcu_enter();
792 va_list args;
793
794 pr_warn(CUT_HERE);
795
796 va_start(args, fmt);
797 vprintk(fmt, args);
798 va_end(args);
799 warn_rcu_exit(rcu);
800}
801EXPORT_SYMBOL(__warn_printk);
802#endif
803
804/* Support resetting WARN*_ONCE state */
805
806static int clear_warn_once_set(void *data, u64 val)
807{
808 generic_bug_clear_once();
809 memset(__start_once, 0, __end_once - __start_once);
810 return 0;
811}
812
813DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set,
814 "%lld\n");
815
816static __init int register_warn_debugfs(void)
817{
818 /* Don't care about failure */
819 debugfs_create_file_unsafe("clear_warn_once", 0200, NULL, NULL,
820 &clear_warn_once_fops);
821 return 0;
822}
823
824device_initcall(register_warn_debugfs);
825#endif
826
827#ifdef CONFIG_STACKPROTECTOR
828
829/*
830 * Called when gcc's -fstack-protector feature is used, and
831 * gcc detects corruption of the on-stack canary value
832 */
833__visible noinstr void __stack_chk_fail(void)
834{
835 instrumentation_begin();
836 panic("stack-protector: Kernel stack is corrupted in: %pB",
837 __builtin_return_address(0));
838 instrumentation_end();
839}
840EXPORT_SYMBOL(__stack_chk_fail);
841
842#endif
843
844core_param(panic, panic_timeout, int, 0644);
845core_param(panic_print, panic_print, ulong, 0644);
846core_param(pause_on_oops, pause_on_oops, int, 0644);
847core_param(panic_on_warn, panic_on_warn, int, 0644);
848core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644);
849
850static int __init oops_setup(char *s)
851{
852 if (!s)
853 return -EINVAL;
854 if (!strcmp(s, "panic"))
855 panic_on_oops = 1;
856 return 0;
857}
858early_param("oops", oops_setup);
859
860static int __init panic_on_taint_setup(char *s)
861{
862 char *taint_str;
863
864 if (!s)
865 return -EINVAL;
866
867 taint_str = strsep(&s, ",");
868 if (kstrtoul(taint_str, 16, &panic_on_taint))
869 return -EINVAL;
870
871 /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */
872 panic_on_taint &= TAINT_FLAGS_MAX;
873
874 if (!panic_on_taint)
875 return -EINVAL;
876
877 if (s && !strcmp(s, "nousertaint"))
878 panic_on_taint_nousertaint = true;
879
880 pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n",
881 panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint));
882
883 return 0;
884}
885early_param("panic_on_taint", panic_on_taint_setup);