projects/linux·blob assoc-array-fixes · refs

1400 lines · 37.9 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Detect hard and soft lockups on a system
4 *
5 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
6 *
7 * Note: Most of this code is borrowed heavily from the original softlockup
8 * detector, so thanks to Ingo for the initial implementation.
9 * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
10 * to those contributors as well.
11 */
12
13#define pr_fmt(fmt) "watchdog: " fmt
14
15#include <linux/cpu.h>
16#include <linux/init.h>
17#include <linux/irq.h>
18#include <linux/irqdesc.h>
19#include <linux/kernel_stat.h>
20#include <linux/kvm_para.h>
21#include <linux/math64.h>
22#include <linux/mm.h>
23#include <linux/module.h>
24#include <linux/nmi.h>
25#include <linux/stop_machine.h>
26#include <linux/sysctl.h>
27#include <linux/tick.h>
28#include <linux/sys_info.h>
29
30#include <linux/sched/clock.h>
31#include <linux/sched/debug.h>
32#include <linux/sched/isolation.h>
33
34#include <asm/irq_regs.h>
35
36static DEFINE_MUTEX(watchdog_mutex);
37
38#if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HARDLOCKUP_DETECTOR_SPARC64)
39# define WATCHDOG_HARDLOCKUP_DEFAULT 1
40#else
41# define WATCHDOG_HARDLOCKUP_DEFAULT 0
42#endif
43
44#define NUM_SAMPLE_PERIODS 5
45
46unsigned long __read_mostly watchdog_enabled;
47int __read_mostly watchdog_user_enabled = 1;
48static int __read_mostly watchdog_hardlockup_user_enabled = WATCHDOG_HARDLOCKUP_DEFAULT;
49static int __read_mostly watchdog_softlockup_user_enabled = 1;
50int __read_mostly watchdog_thresh = 10;
51static int __read_mostly watchdog_thresh_next;
52static int __read_mostly watchdog_hardlockup_available;
53
54struct cpumask watchdog_cpumask __read_mostly;
55unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
56
57#ifdef CONFIG_HARDLOCKUP_DETECTOR
58
59# ifdef CONFIG_SMP
60int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
61# endif /* CONFIG_SMP */
62
63/*
64 * Number of consecutive missed interrupts before declaring a lockup.
65 * Default to 1 (immediate) for NMI/Perf. Buddy will overwrite this to 3.
66 */
67int __read_mostly watchdog_hardlockup_miss_thresh = 1;
68EXPORT_SYMBOL_GPL(watchdog_hardlockup_miss_thresh);
69
70/*
71 * Should we panic when a soft-lockup or hard-lockup occurs:
72 */
73unsigned int __read_mostly hardlockup_panic =
74 IS_ENABLED(CONFIG_BOOTPARAM_HARDLOCKUP_PANIC);
75
76/*
77 * bitmasks to control what kinds of system info to be printed when
78 * hard lockup is detected, it could be task, memory, lock etc.
79 * Refer include/linux/sys_info.h for detailed bit definition.
80 */
81unsigned long hardlockup_si_mask;
82
83#ifdef CONFIG_SYSFS
84
85static unsigned int hardlockup_count;
86
87static ssize_t hardlockup_count_show(struct kobject *kobj, struct kobj_attribute *attr,
88 char *page)
89{
90 return sysfs_emit(page, "%u\n", hardlockup_count);
91}
92
93static struct kobj_attribute hardlockup_count_attr = __ATTR_RO(hardlockup_count);
94
95static __init int kernel_hardlockup_sysfs_init(void)
96{
97 sysfs_add_file_to_group(kernel_kobj, &hardlockup_count_attr.attr, NULL);
98 return 0;
99}
100
101late_initcall(kernel_hardlockup_sysfs_init);
102
103#endif // CONFIG_SYSFS
104
105/*
106 * We may not want to enable hard lockup detection by default in all cases,
107 * for example when running the kernel as a guest on a hypervisor. In these
108 * cases this function can be called to disable hard lockup detection. This
109 * function should only be executed once by the boot processor before the
110 * kernel command line parameters are parsed, because otherwise it is not
111 * possible to override this in hardlockup_panic_setup().
112 */
113void __init hardlockup_detector_disable(void)
114{
115 watchdog_hardlockup_user_enabled = 0;
116}
117
118static int __init hardlockup_panic_setup(char *str)
119{
120next:
121 if (!strncmp(str, "panic", 5))
122 hardlockup_panic = 1;
123 else if (!strncmp(str, "nopanic", 7))
124 hardlockup_panic = 0;
125 else if (!strncmp(str, "0", 1))
126 watchdog_hardlockup_user_enabled = 0;
127 else if (!strncmp(str, "1", 1))
128 watchdog_hardlockup_user_enabled = 1;
129 else if (!strncmp(str, "r", 1))
130 hardlockup_config_perf_event(str + 1);
131 while (*(str++)) {
132 if (*str == ',') {
133 str++;
134 goto next;
135 }
136 }
137 return 1;
138}
139__setup("nmi_watchdog=", hardlockup_panic_setup);
140
141#endif /* CONFIG_HARDLOCKUP_DETECTOR */
142
143#if defined(CONFIG_HARDLOCKUP_DETECTOR_COUNTS_HRTIMER)
144
145static DEFINE_PER_CPU(atomic_t, hrtimer_interrupts);
146static DEFINE_PER_CPU(int, hrtimer_interrupts_saved);
147static DEFINE_PER_CPU(int, hrtimer_interrupts_missed);
148static DEFINE_PER_CPU(bool, watchdog_hardlockup_warned);
149static DEFINE_PER_CPU(bool, watchdog_hardlockup_touched);
150static unsigned long hard_lockup_nmi_warn;
151
152notrace void arch_touch_nmi_watchdog(void)
153{
154 /*
155 * Using __raw here because some code paths have
156 * preemption enabled. If preemption is enabled
157 * then interrupts should be enabled too, in which
158 * case we shouldn't have to worry about the watchdog
159 * going off.
160 */
161 raw_cpu_write(watchdog_hardlockup_touched, true);
162}
163EXPORT_SYMBOL(arch_touch_nmi_watchdog);
164
165void watchdog_hardlockup_touch_cpu(unsigned int cpu)
166{
167 per_cpu(watchdog_hardlockup_touched, cpu) = true;
168}
169
170static void watchdog_hardlockup_update_reset(unsigned int cpu)
171{
172 int hrint = atomic_read(&per_cpu(hrtimer_interrupts, cpu));
173
174 /*
175 * NOTE: we don't need any fancy atomic_t or READ_ONCE/WRITE_ONCE
176 * for hrtimer_interrupts_saved. hrtimer_interrupts_saved is
177 * written/read by a single CPU.
178 */
179 per_cpu(hrtimer_interrupts_saved, cpu) = hrint;
180 per_cpu(hrtimer_interrupts_missed, cpu) = 0;
181}
182
183static bool is_hardlockup(unsigned int cpu)
184{
185 int hrint = atomic_read(&per_cpu(hrtimer_interrupts, cpu));
186
187 if (per_cpu(hrtimer_interrupts_saved, cpu) != hrint) {
188 watchdog_hardlockup_update_reset(cpu);
189 return false;
190 }
191
192 per_cpu(hrtimer_interrupts_missed, cpu)++;
193 if (per_cpu(hrtimer_interrupts_missed, cpu) % watchdog_hardlockup_miss_thresh)
194 return false;
195
196 return true;
197}
198
199static void watchdog_hardlockup_kick(void)
200{
201 int new_interrupts;
202
203 new_interrupts = atomic_inc_return(this_cpu_ptr(&hrtimer_interrupts));
204 watchdog_buddy_check_hardlockup(new_interrupts);
205}
206
207void watchdog_hardlockup_check(unsigned int cpu, struct pt_regs *regs)
208{
209 int hardlockup_all_cpu_backtrace;
210 unsigned int this_cpu;
211 unsigned long flags;
212
213 if (per_cpu(watchdog_hardlockup_touched, cpu)) {
214 watchdog_hardlockup_update_reset(cpu);
215 per_cpu(watchdog_hardlockup_touched, cpu) = false;
216 return;
217 }
218
219 hardlockup_all_cpu_backtrace = (hardlockup_si_mask & SYS_INFO_ALL_BT) ?
220 1 : sysctl_hardlockup_all_cpu_backtrace;
221 /*
222 * Check for a hardlockup by making sure the CPU's timer
223 * interrupt is incrementing. The timer interrupt should have
224 * fired multiple times before we overflow'd. If it hasn't
225 * then this is a good indication the cpu is stuck
226 */
227 if (!is_hardlockup(cpu)) {
228 per_cpu(watchdog_hardlockup_warned, cpu) = false;
229 return;
230 }
231
232#ifdef CONFIG_SYSFS
233 ++hardlockup_count;
234#endif
235 /*
236 * A poorly behaving BPF scheduler can trigger hard lockup by
237 * e.g. putting numerous affinitized tasks in a single queue and
238 * directing all CPUs at it. The following call can return true
239 * only once when sched_ext is enabled and will immediately
240 * abort the BPF scheduler and print out a warning message.
241 */
242 if (scx_hardlockup(cpu))
243 return;
244
245 /* Only print hardlockups once. */
246 if (per_cpu(watchdog_hardlockup_warned, cpu))
247 return;
248
249 /*
250 * Prevent multiple hard-lockup reports if one cpu is already
251 * engaged in dumping all cpu back traces.
252 */
253 if (hardlockup_all_cpu_backtrace) {
254 if (test_and_set_bit_lock(0, &hard_lockup_nmi_warn))
255 return;
256 }
257
258 /*
259 * NOTE: we call printk_cpu_sync_get_irqsave() after printing
260 * the lockup message. While it would be nice to serialize
261 * that printout, we really want to make sure that if some
262 * other CPU somehow locked up while holding the lock associated
263 * with printk_cpu_sync_get_irqsave() that we can still at least
264 * get the message about the lockup out.
265 */
266 this_cpu = smp_processor_id();
267 pr_emerg("CPU%u: Watchdog detected hard LOCKUP on cpu %u\n", this_cpu, cpu);
268 printk_cpu_sync_get_irqsave(flags);
269
270 print_modules();
271 print_irqtrace_events(current);
272 if (cpu == this_cpu) {
273 if (regs)
274 show_regs(regs);
275 else
276 dump_stack();
277 printk_cpu_sync_put_irqrestore(flags);
278 } else {
279 printk_cpu_sync_put_irqrestore(flags);
280 trigger_single_cpu_backtrace(cpu);
281 }
282
283 if (hardlockup_all_cpu_backtrace) {
284 trigger_allbutcpu_cpu_backtrace(cpu);
285 if (!hardlockup_panic)
286 clear_bit_unlock(0, &hard_lockup_nmi_warn);
287 }
288
289 sys_info(hardlockup_si_mask & ~SYS_INFO_ALL_BT);
290 if (hardlockup_panic)
291 nmi_panic(regs, "Hard LOCKUP");
292
293 per_cpu(watchdog_hardlockup_warned, cpu) = true;
294}
295
296#else /* CONFIG_HARDLOCKUP_DETECTOR_COUNTS_HRTIMER */
297
298static inline void watchdog_hardlockup_kick(void) { }
299
300#endif /* !CONFIG_HARDLOCKUP_DETECTOR_COUNTS_HRTIMER */
301
302/*
303 * These functions can be overridden based on the configured hardlockdup detector.
304 *
305 * watchdog_hardlockup_enable/disable can be implemented to start and stop when
306 * softlockup watchdog start and stop. The detector must select the
307 * SOFTLOCKUP_DETECTOR Kconfig.
308 */
309void __weak watchdog_hardlockup_enable(unsigned int cpu) { }
310
311void __weak watchdog_hardlockup_disable(unsigned int cpu) { }
312
313/*
314 * Watchdog-detector specific API.
315 *
316 * Return 0 when hardlockup watchdog is available, negative value otherwise.
317 * Note that the negative value means that a delayed probe might
318 * succeed later.
319 */
320int __weak __init watchdog_hardlockup_probe(void)
321{
322 return -ENODEV;
323}
324
325/**
326 * watchdog_hardlockup_stop - Stop the watchdog for reconfiguration
327 *
328 * The reconfiguration steps are:
329 * watchdog_hardlockup_stop();
330 * update_variables();
331 * watchdog_hardlockup_start();
332 */
333void __weak watchdog_hardlockup_stop(void) { }
334
335/**
336 * watchdog_hardlockup_start - Start the watchdog after reconfiguration
337 *
338 * Counterpart to watchdog_hardlockup_stop().
339 *
340 * The following variables have been updated in update_variables() and
341 * contain the currently valid configuration:
342 * - watchdog_enabled
343 * - watchdog_thresh
344 * - watchdog_cpumask
345 */
346void __weak watchdog_hardlockup_start(void) { }
347
348/**
349 * lockup_detector_update_enable - Update the sysctl enable bit
350 *
351 * Caller needs to make sure that the hard watchdogs are off, so this
352 * can't race with watchdog_hardlockup_disable().
353 */
354static void lockup_detector_update_enable(void)
355{
356 watchdog_enabled = 0;
357 if (!watchdog_user_enabled)
358 return;
359 if (watchdog_hardlockup_available && watchdog_hardlockup_user_enabled)
360 watchdog_enabled |= WATCHDOG_HARDLOCKUP_ENABLED;
361 if (watchdog_softlockup_user_enabled)
362 watchdog_enabled |= WATCHDOG_SOFTLOCKUP_ENABLED;
363}
364
365#ifdef CONFIG_SOFTLOCKUP_DETECTOR
366
367/*
368 * Delay the softlockup report when running a known slow code.
369 * It does _not_ affect the timestamp of the last successful reschedule.
370 */
371#define SOFTLOCKUP_DELAY_REPORT ULONG_MAX
372
373#ifdef CONFIG_SMP
374int __read_mostly sysctl_softlockup_all_cpu_backtrace;
375#endif
376
377/*
378 * bitmasks to control what kinds of system info to be printed when
379 * soft lockup is detected, it could be task, memory, lock etc.
380 * Refer include/linux/sys_info.h for detailed bit definition.
381 */
382static unsigned long softlockup_si_mask;
383
384static struct cpumask watchdog_allowed_mask __read_mostly;
385
386/* Global variables, exported for sysctl */
387unsigned int __read_mostly softlockup_panic =
388 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC;
389
390static bool softlockup_initialized __read_mostly;
391static u64 __read_mostly sample_period;
392
393#ifdef CONFIG_SYSFS
394
395static unsigned int softlockup_count;
396
397static ssize_t softlockup_count_show(struct kobject *kobj, struct kobj_attribute *attr,
398 char *page)
399{
400 return sysfs_emit(page, "%u\n", softlockup_count);
401}
402
403static struct kobj_attribute softlockup_count_attr = __ATTR_RO(softlockup_count);
404
405static __init int kernel_softlockup_sysfs_init(void)
406{
407 sysfs_add_file_to_group(kernel_kobj, &softlockup_count_attr.attr, NULL);
408 return 0;
409}
410
411late_initcall(kernel_softlockup_sysfs_init);
412
413#endif // CONFIG_SYSFS
414
415/* Timestamp taken after the last successful reschedule. */
416static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
417/* Timestamp of the last softlockup report. */
418static DEFINE_PER_CPU(unsigned long, watchdog_report_ts);
419static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
420static DEFINE_PER_CPU(bool, softlockup_touch_sync);
421static unsigned long soft_lockup_nmi_warn;
422
423static int __init softlockup_panic_setup(char *str)
424{
425 softlockup_panic = simple_strtoul(str, NULL, 0);
426 return 1;
427}
428__setup("softlockup_panic=", softlockup_panic_setup);
429
430static int __init nowatchdog_setup(char *str)
431{
432 watchdog_user_enabled = 0;
433 return 1;
434}
435__setup("nowatchdog", nowatchdog_setup);
436
437static int __init nosoftlockup_setup(char *str)
438{
439 watchdog_softlockup_user_enabled = 0;
440 return 1;
441}
442__setup("nosoftlockup", nosoftlockup_setup);
443
444static int __init watchdog_thresh_setup(char *str)
445{
446 get_option(&str, &watchdog_thresh);
447 return 1;
448}
449__setup("watchdog_thresh=", watchdog_thresh_setup);
450
451#ifdef CONFIG_SOFTLOCKUP_DETECTOR_INTR_STORM
452enum stats_per_group {
453 STATS_SYSTEM,
454 STATS_SOFTIRQ,
455 STATS_HARDIRQ,
456 STATS_IDLE,
457 NUM_STATS_PER_GROUP,
458};
459
460static const enum cpu_usage_stat tracked_stats[NUM_STATS_PER_GROUP] = {
461 CPUTIME_SYSTEM,
462 CPUTIME_SOFTIRQ,
463 CPUTIME_IRQ,
464 CPUTIME_IDLE,
465};
466
467static DEFINE_PER_CPU(u16, cpustat_old[NUM_STATS_PER_GROUP]);
468static DEFINE_PER_CPU(u8, cpustat_util[NUM_SAMPLE_PERIODS][NUM_STATS_PER_GROUP]);
469static DEFINE_PER_CPU(u8, cpustat_tail);
470
471/*
472 * We don't need nanosecond resolution. A granularity of 16ms is
473 * sufficient for our precision, allowing us to use u16 to store
474 * cpustats, which will roll over roughly every ~1000 seconds.
475 * 2^24 ~= 16 * 10^6
476 */
477static u16 get_16bit_precision(u64 data_ns)
478{
479 /*
480 * 2^24ns ~= 16.8ms
481 * Round to the nearest multiple of 16.8 milliseconds.
482 */
483 return (data_ns + (1 << 23)) >> 24LL;
484}
485
486static void update_cpustat(void)
487{
488 int i;
489 u8 util;
490 u16 old_stat, new_stat;
491 struct kernel_cpustat kcpustat;
492 u64 *cpustat = kcpustat.cpustat;
493 u8 tail = __this_cpu_read(cpustat_tail);
494 u16 sample_period_16 = get_16bit_precision(sample_period);
495
496 kcpustat_cpu_fetch(&kcpustat, smp_processor_id());
497
498 for (i = 0; i < NUM_STATS_PER_GROUP; i++) {
499 old_stat = __this_cpu_read(cpustat_old[i]);
500 new_stat = get_16bit_precision(cpustat[tracked_stats[i]]);
501 util = DIV_ROUND_UP(100 * (new_stat - old_stat), sample_period_16);
502 /*
503 * Since we use 16-bit precision, the raw data will undergo
504 * integer division, which may sometimes result in data loss,
505 * and then result might exceed 100%. To avoid confusion,
506 * we enforce a 100% display cap when calculations exceed this threshold.
507 */
508 if (util > 100)
509 util = 100;
510 __this_cpu_write(cpustat_util[tail][i], util);
511 __this_cpu_write(cpustat_old[i], new_stat);
512 }
513
514 __this_cpu_write(cpustat_tail, (tail + 1) % NUM_SAMPLE_PERIODS);
515}
516
517static void print_cpustat(void)
518{
519 int i, group;
520 u8 tail = __this_cpu_read(cpustat_tail);
521 u64 sample_period_msecond = sample_period;
522
523 do_div(sample_period_msecond, NSEC_PER_MSEC);
524
525 /*
526 * Outputting the "watchdog" prefix on every line is redundant and not
527 * concise, and the original alarm information is sufficient for
528 * positioning in logs, hence here printk() is used instead of pr_crit().
529 */
530 printk(KERN_CRIT "CPU#%d Utilization every %llums during lockup:\n",
531 smp_processor_id(), sample_period_msecond);
532
533 for (i = 0; i < NUM_SAMPLE_PERIODS; i++) {
534 group = (tail + i) % NUM_SAMPLE_PERIODS;
535 printk(KERN_CRIT "\t#%d: %3u%% system,\t%3u%% softirq,\t"
536 "%3u%% hardirq,\t%3u%% idle\n", i + 1,
537 __this_cpu_read(cpustat_util[group][STATS_SYSTEM]),
538 __this_cpu_read(cpustat_util[group][STATS_SOFTIRQ]),
539 __this_cpu_read(cpustat_util[group][STATS_HARDIRQ]),
540 __this_cpu_read(cpustat_util[group][STATS_IDLE]));
541 }
542}
543
544#define HARDIRQ_PERCENT_THRESH 50
545#define NUM_HARDIRQ_REPORT 5
546struct irq_counts {
547 int irq;
548 u32 counts;
549};
550
551static DEFINE_PER_CPU(bool, snapshot_taken);
552
553/* Tabulate the most frequent interrupts. */
554static void tabulate_irq_count(struct irq_counts *irq_counts, int irq, u32 counts, int rank)
555{
556 int i;
557 struct irq_counts new_count = {irq, counts};
558
559 for (i = 0; i < rank; i++) {
560 if (counts > irq_counts[i].counts)
561 swap(new_count, irq_counts[i]);
562 }
563}
564
565/*
566 * If the hardirq time exceeds HARDIRQ_PERCENT_THRESH% of the sample_period,
567 * then the cause of softlockup might be interrupt storm. In this case, it
568 * would be useful to start interrupt counting.
569 */
570static bool need_counting_irqs(void)
571{
572 u8 util;
573 int tail = __this_cpu_read(cpustat_tail);
574
575 tail = (tail + NUM_SAMPLE_PERIODS - 1) % NUM_SAMPLE_PERIODS;
576 util = __this_cpu_read(cpustat_util[tail][STATS_HARDIRQ]);
577 return util > HARDIRQ_PERCENT_THRESH;
578}
579
580static void start_counting_irqs(void)
581{
582 if (!__this_cpu_read(snapshot_taken)) {
583 kstat_snapshot_irqs();
584 __this_cpu_write(snapshot_taken, true);
585 }
586}
587
588static void stop_counting_irqs(void)
589{
590 __this_cpu_write(snapshot_taken, false);
591}
592
593static void print_irq_counts(void)
594{
595 unsigned int i, count;
596 struct irq_counts irq_counts_sorted[NUM_HARDIRQ_REPORT] = {
597 {-1, 0}, {-1, 0}, {-1, 0}, {-1, 0}, {-1, 0}
598 };
599
600 if (__this_cpu_read(snapshot_taken)) {
601 for_each_active_irq(i) {
602 count = kstat_get_irq_since_snapshot(i);
603 tabulate_irq_count(irq_counts_sorted, i, count, NUM_HARDIRQ_REPORT);
604 }
605
606 /*
607 * Outputting the "watchdog" prefix on every line is redundant and not
608 * concise, and the original alarm information is sufficient for
609 * positioning in logs, hence here printk() is used instead of pr_crit().
610 */
611 printk(KERN_CRIT "CPU#%d Detect HardIRQ Time exceeds %d%%. Most frequent HardIRQs:\n",
612 smp_processor_id(), HARDIRQ_PERCENT_THRESH);
613
614 for (i = 0; i < NUM_HARDIRQ_REPORT; i++) {
615 if (irq_counts_sorted[i].irq == -1)
616 break;
617
618 printk(KERN_CRIT "\t#%u: %-10u\tirq#%d\n",
619 i + 1, irq_counts_sorted[i].counts,
620 irq_counts_sorted[i].irq);
621 }
622
623 /*
624 * If the hardirq time is less than HARDIRQ_PERCENT_THRESH% in the last
625 * sample_period, then we suspect the interrupt storm might be subsiding.
626 */
627 if (!need_counting_irqs())
628 stop_counting_irqs();
629 }
630}
631
632static void report_cpu_status(void)
633{
634 print_cpustat();
635 print_irq_counts();
636}
637#else
638static inline void update_cpustat(void) { }
639static inline void report_cpu_status(void) { }
640static inline bool need_counting_irqs(void) { return false; }
641static inline void start_counting_irqs(void) { }
642static inline void stop_counting_irqs(void) { }
643#endif
644
645/*
646 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
647 * lockups can have false positives under extreme conditions. So we generally
648 * want a higher threshold for soft lockups than for hard lockups. So we couple
649 * the thresholds with a factor: we make the soft threshold twice the amount of
650 * time the hard threshold is.
651 */
652static int get_softlockup_thresh(void)
653{
654 return watchdog_thresh * 2;
655}
656
657/*
658 * Returns seconds, approximately. We don't need nanosecond
659 * resolution, and we don't need to waste time with a big divide when
660 * 2^30ns == 1.074s.
661 */
662static unsigned long get_timestamp(void)
663{
664 return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
665}
666
667static void set_sample_period(void)
668{
669 /*
670 * convert watchdog_thresh from seconds to ns
671 * the divide by 5 is to give hrtimer several chances (two
672 * or three with the current relation between the soft
673 * and hard thresholds) to increment before the
674 * hardlockup detector generates a warning
675 */
676 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / NUM_SAMPLE_PERIODS);
677 watchdog_update_hrtimer_threshold(sample_period);
678}
679
680static void update_report_ts(void)
681{
682 __this_cpu_write(watchdog_report_ts, get_timestamp());
683}
684
685/* Commands for resetting the watchdog */
686static void update_touch_ts(void)
687{
688 __this_cpu_write(watchdog_touch_ts, get_timestamp());
689 update_report_ts();
690}
691
692/**
693 * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
694 *
695 * Call when the scheduler may have stalled for legitimate reasons
696 * preventing the watchdog task from executing - e.g. the scheduler
697 * entering idle state. This should only be used for scheduler events.
698 * Use touch_softlockup_watchdog() for everything else.
699 */
700notrace void touch_softlockup_watchdog_sched(void)
701{
702 /*
703 * Preemption can be enabled. It doesn't matter which CPU's watchdog
704 * report period gets restarted here, so use the raw_ operation.
705 */
706 raw_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
707}
708
709notrace void touch_softlockup_watchdog(void)
710{
711 touch_softlockup_watchdog_sched();
712 wq_watchdog_touch(raw_smp_processor_id());
713}
714EXPORT_SYMBOL(touch_softlockup_watchdog);
715
716void touch_all_softlockup_watchdogs(void)
717{
718 int cpu;
719
720 /*
721 * watchdog_mutex cannpt be taken here, as this might be called
722 * from (soft)interrupt context, so the access to
723 * watchdog_allowed_cpumask might race with a concurrent update.
724 *
725 * The watchdog time stamp can race against a concurrent real
726 * update as well, the only side effect might be a cycle delay for
727 * the softlockup check.
728 */
729 for_each_cpu(cpu, &watchdog_allowed_mask) {
730 per_cpu(watchdog_report_ts, cpu) = SOFTLOCKUP_DELAY_REPORT;
731 wq_watchdog_touch(cpu);
732 }
733}
734
735void touch_softlockup_watchdog_sync(void)
736{
737 __this_cpu_write(softlockup_touch_sync, true);
738 __this_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT);
739}
740
741static int is_softlockup(unsigned long touch_ts,
742 unsigned long period_ts,
743 unsigned long now)
744{
745 if ((watchdog_enabled & WATCHDOG_SOFTLOCKUP_ENABLED) && watchdog_thresh) {
746 /*
747 * If period_ts has not been updated during a sample_period, then
748 * in the subsequent few sample_periods, period_ts might also not
749 * be updated, which could indicate a potential softlockup. In
750 * this case, if we suspect the cause of the potential softlockup
751 * might be interrupt storm, then we need to count the interrupts
752 * to find which interrupt is storming.
753 */
754 if (time_after_eq(now, period_ts + get_softlockup_thresh() / NUM_SAMPLE_PERIODS) &&
755 need_counting_irqs())
756 start_counting_irqs();
757
758 /*
759 * A poorly behaving BPF scheduler can live-lock the system into
760 * soft lockups. Tell sched_ext to try ejecting the BPF
761 * scheduler when close to a soft lockup.
762 */
763 if (time_after_eq(now, period_ts + get_softlockup_thresh() * 3 / 4))
764 scx_softlockup(now - touch_ts);
765
766 /* Warn about unreasonable delays. */
767 if (time_after(now, period_ts + get_softlockup_thresh()))
768 return now - touch_ts;
769 }
770 return 0;
771}
772
773/* watchdog detector functions */
774static DEFINE_PER_CPU(struct completion, softlockup_completion);
775static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
776
777/*
778 * The watchdog feed function - touches the timestamp.
779 *
780 * It only runs once every sample_period seconds (4 seconds by
781 * default) to reset the softlockup timestamp. If this gets delayed
782 * for more than 2*watchdog_thresh seconds then the debug-printout
783 * triggers in watchdog_timer_fn().
784 */
785static int softlockup_fn(void *data)
786{
787 update_touch_ts();
788 stop_counting_irqs();
789 complete(this_cpu_ptr(&softlockup_completion));
790
791 return 0;
792}
793
794/* watchdog kicker functions */
795static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
796{
797 unsigned long touch_ts, period_ts, now;
798 struct pt_regs *regs = get_irq_regs();
799 int softlockup_all_cpu_backtrace;
800 int duration, thresh_count;
801 unsigned long flags;
802
803 if (!watchdog_enabled)
804 return HRTIMER_NORESTART;
805
806 /*
807 * pass the buddy check if a panic is in process
808 */
809 if (panic_in_progress())
810 return HRTIMER_NORESTART;
811
812 softlockup_all_cpu_backtrace = (softlockup_si_mask & SYS_INFO_ALL_BT) ?
813 1 : sysctl_softlockup_all_cpu_backtrace;
814
815 watchdog_hardlockup_kick();
816
817 /* kick the softlockup detector */
818 if (completion_done(this_cpu_ptr(&softlockup_completion))) {
819 reinit_completion(this_cpu_ptr(&softlockup_completion));
820 stop_one_cpu_nowait(smp_processor_id(),
821 softlockup_fn, NULL,
822 this_cpu_ptr(&softlockup_stop_work));
823 }
824
825 /* .. and repeat */
826 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
827
828 /*
829 * Read the current timestamp first. It might become invalid anytime
830 * when a virtual machine is stopped by the host or when the watchog
831 * is touched from NMI.
832 */
833 now = get_timestamp();
834 /*
835 * If a virtual machine is stopped by the host it can look to
836 * the watchdog like a soft lockup. This function touches the watchdog.
837 */
838 kvm_check_and_clear_guest_paused();
839 /*
840 * The stored timestamp is comparable with @now only when not touched.
841 * It might get touched anytime from NMI. Make sure that is_softlockup()
842 * uses the same (valid) value.
843 */
844 period_ts = READ_ONCE(*this_cpu_ptr(&watchdog_report_ts));
845
846 update_cpustat();
847
848 /* Reset the interval when touched by known problematic code. */
849 if (period_ts == SOFTLOCKUP_DELAY_REPORT) {
850 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
851 /*
852 * If the time stamp was touched atomically
853 * make sure the scheduler tick is up to date.
854 */
855 __this_cpu_write(softlockup_touch_sync, false);
856 sched_clock_tick();
857 }
858
859 update_report_ts();
860 return HRTIMER_RESTART;
861 }
862
863 /* Check for a softlockup. */
864 touch_ts = __this_cpu_read(watchdog_touch_ts);
865 duration = is_softlockup(touch_ts, period_ts, now);
866 if (unlikely(duration)) {
867#ifdef CONFIG_SYSFS
868 ++softlockup_count;
869#endif
870
871 /*
872 * Prevent multiple soft-lockup reports if one cpu is already
873 * engaged in dumping all cpu back traces.
874 */
875 if (softlockup_all_cpu_backtrace) {
876 if (test_and_set_bit_lock(0, &soft_lockup_nmi_warn))
877 return HRTIMER_RESTART;
878 }
879
880 /* Start period for the next softlockup warning. */
881 update_report_ts();
882
883 printk_cpu_sync_get_irqsave(flags);
884 pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
885 smp_processor_id(), duration,
886 current->comm, task_pid_nr(current));
887 report_cpu_status();
888 print_modules();
889 print_irqtrace_events(current);
890 if (regs)
891 show_regs(regs);
892 else
893 dump_stack();
894 printk_cpu_sync_put_irqrestore(flags);
895
896 if (softlockup_all_cpu_backtrace) {
897 trigger_allbutcpu_cpu_backtrace(smp_processor_id());
898 if (!softlockup_panic)
899 clear_bit_unlock(0, &soft_lockup_nmi_warn);
900 }
901
902 add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
903 sys_info(softlockup_si_mask & ~SYS_INFO_ALL_BT);
904 thresh_count = duration / get_softlockup_thresh();
905
906 if (softlockup_panic && thresh_count >= softlockup_panic)
907 panic("softlockup: hung tasks");
908 }
909
910 return HRTIMER_RESTART;
911}
912
913static void watchdog_enable(unsigned int cpu)
914{
915 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
916 struct completion *done = this_cpu_ptr(&softlockup_completion);
917
918 WARN_ON_ONCE(cpu != smp_processor_id());
919
920 init_completion(done);
921 complete(done);
922
923 /*
924 * Start the timer first to prevent the hardlockup watchdog triggering
925 * before the timer has a chance to fire.
926 */
927 hrtimer_setup(hrtimer, watchdog_timer_fn, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
928 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
929 HRTIMER_MODE_REL_PINNED_HARD);
930
931 /* Initialize timestamp */
932 update_touch_ts();
933 /* Enable the hardlockup detector */
934 if (watchdog_enabled & WATCHDOG_HARDLOCKUP_ENABLED)
935 watchdog_hardlockup_enable(cpu);
936}
937
938static void watchdog_disable(unsigned int cpu)
939{
940 struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
941
942 WARN_ON_ONCE(cpu != smp_processor_id());
943
944 /*
945 * Disable the hardlockup detector first. That prevents that a large
946 * delay between disabling the timer and disabling the hardlockup
947 * detector causes a false positive.
948 */
949 watchdog_hardlockup_disable(cpu);
950 hrtimer_cancel(hrtimer);
951 wait_for_completion(this_cpu_ptr(&softlockup_completion));
952}
953
954static int softlockup_stop_fn(void *data)
955{
956 watchdog_disable(smp_processor_id());
957 return 0;
958}
959
960static void softlockup_stop_all(void)
961{
962 int cpu;
963
964 if (!softlockup_initialized)
965 return;
966
967 for_each_cpu(cpu, &watchdog_allowed_mask)
968 smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
969
970 cpumask_clear(&watchdog_allowed_mask);
971}
972
973static int softlockup_start_fn(void *data)
974{
975 watchdog_enable(smp_processor_id());
976 return 0;
977}
978
979static void softlockup_start_all(void)
980{
981 int cpu;
982
983 cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
984 for_each_cpu(cpu, &watchdog_allowed_mask)
985 smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
986}
987
988int lockup_detector_online_cpu(unsigned int cpu)
989{
990 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
991 watchdog_enable(cpu);
992 return 0;
993}
994
995int lockup_detector_offline_cpu(unsigned int cpu)
996{
997 if (cpumask_test_cpu(cpu, &watchdog_allowed_mask))
998 watchdog_disable(cpu);
999 return 0;
1000}
1001
1002static void __lockup_detector_reconfigure(bool thresh_changed)
1003{
1004 cpus_read_lock();
1005 watchdog_hardlockup_stop();
1006
1007 softlockup_stop_all();
1008 /*
1009 * To prevent watchdog_timer_fn from using the old interval and
1010 * the new watchdog_thresh at the same time, which could lead to
1011 * false softlockup reports, it is necessary to update the
1012 * watchdog_thresh after the softlockup is completed.
1013 */
1014 if (thresh_changed)
1015 watchdog_thresh = READ_ONCE(watchdog_thresh_next);
1016 set_sample_period();
1017 lockup_detector_update_enable();
1018 if (watchdog_enabled && watchdog_thresh)
1019 softlockup_start_all();
1020
1021 watchdog_hardlockup_start();
1022 cpus_read_unlock();
1023}
1024
1025void lockup_detector_reconfigure(void)
1026{
1027 mutex_lock(&watchdog_mutex);
1028 __lockup_detector_reconfigure(false);
1029 mutex_unlock(&watchdog_mutex);
1030}
1031
1032/*
1033 * Create the watchdog infrastructure and configure the detector(s).
1034 */
1035static __init void lockup_detector_setup(void)
1036{
1037 /*
1038 * If sysctl is off and watchdog got disabled on the command line,
1039 * nothing to do here.
1040 */
1041 lockup_detector_update_enable();
1042
1043 if (!IS_ENABLED(CONFIG_SYSCTL) &&
1044 !(watchdog_enabled && watchdog_thresh))
1045 return;
1046
1047 mutex_lock(&watchdog_mutex);
1048 __lockup_detector_reconfigure(false);
1049 softlockup_initialized = true;
1050 mutex_unlock(&watchdog_mutex);
1051}
1052
1053#else /* CONFIG_SOFTLOCKUP_DETECTOR */
1054static void __lockup_detector_reconfigure(bool thresh_changed)
1055{
1056 cpus_read_lock();
1057 watchdog_hardlockup_stop();
1058 if (thresh_changed)
1059 watchdog_thresh = READ_ONCE(watchdog_thresh_next);
1060 lockup_detector_update_enable();
1061 watchdog_hardlockup_start();
1062 cpus_read_unlock();
1063}
1064void lockup_detector_reconfigure(void)
1065{
1066 __lockup_detector_reconfigure(false);
1067}
1068static inline void lockup_detector_setup(void)
1069{
1070 __lockup_detector_reconfigure(false);
1071}
1072#endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
1073
1074/**
1075 * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
1076 *
1077 * Special interface for parisc. It prevents lockup detector warnings from
1078 * the default pm_poweroff() function which busy loops forever.
1079 */
1080void lockup_detector_soft_poweroff(void)
1081{
1082 watchdog_enabled = 0;
1083}
1084
1085#ifdef CONFIG_SYSCTL
1086
1087/* Propagate any changes to the watchdog infrastructure */
1088static void proc_watchdog_update(bool thresh_changed)
1089{
1090 /* Remove impossible cpus to keep sysctl output clean. */
1091 cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
1092 __lockup_detector_reconfigure(thresh_changed);
1093}
1094
1095/*
1096 * common function for watchdog, nmi_watchdog and soft_watchdog parameter
1097 *
1098 * caller | table->data points to | 'which'
1099 * -------------------|----------------------------------|-------------------------------
1100 * proc_watchdog | watchdog_user_enabled | WATCHDOG_HARDLOCKUP_ENABLED |
1101 * | | WATCHDOG_SOFTLOCKUP_ENABLED
1102 * -------------------|----------------------------------|-------------------------------
1103 * proc_nmi_watchdog | watchdog_hardlockup_user_enabled | WATCHDOG_HARDLOCKUP_ENABLED
1104 * -------------------|----------------------------------|-------------------------------
1105 * proc_soft_watchdog | watchdog_softlockup_user_enabled | WATCHDOG_SOFTLOCKUP_ENABLED
1106 */
1107static int proc_watchdog_common(int which, const struct ctl_table *table, int write,
1108 void *buffer, size_t *lenp, loff_t *ppos)
1109{
1110 int err, old, *param = table->data;
1111
1112 mutex_lock(&watchdog_mutex);
1113
1114 old = *param;
1115 if (!write) {
1116 /*
1117 * On read synchronize the userspace interface. This is a
1118 * racy snapshot.
1119 */
1120 *param = (watchdog_enabled & which) != 0;
1121 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
1122 *param = old;
1123 } else {
1124 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
1125 if (!err && old != READ_ONCE(*param))
1126 proc_watchdog_update(false);
1127 }
1128 mutex_unlock(&watchdog_mutex);
1129 return err;
1130}
1131
1132/*
1133 * /proc/sys/kernel/watchdog
1134 */
1135static int proc_watchdog(const struct ctl_table *table, int write,
1136 void *buffer, size_t *lenp, loff_t *ppos)
1137{
1138 return proc_watchdog_common(WATCHDOG_HARDLOCKUP_ENABLED |
1139 WATCHDOG_SOFTLOCKUP_ENABLED,
1140 table, write, buffer, lenp, ppos);
1141}
1142
1143/*
1144 * /proc/sys/kernel/nmi_watchdog
1145 */
1146static int proc_nmi_watchdog(const struct ctl_table *table, int write,
1147 void *buffer, size_t *lenp, loff_t *ppos)
1148{
1149 if (!watchdog_hardlockup_available && write)
1150 return -ENOTSUPP;
1151 return proc_watchdog_common(WATCHDOG_HARDLOCKUP_ENABLED,
1152 table, write, buffer, lenp, ppos);
1153}
1154
1155#ifdef CONFIG_SOFTLOCKUP_DETECTOR
1156/*
1157 * /proc/sys/kernel/soft_watchdog
1158 */
1159static int proc_soft_watchdog(const struct ctl_table *table, int write,
1160 void *buffer, size_t *lenp, loff_t *ppos)
1161{
1162 return proc_watchdog_common(WATCHDOG_SOFTLOCKUP_ENABLED,
1163 table, write, buffer, lenp, ppos);
1164}
1165#endif
1166
1167/*
1168 * /proc/sys/kernel/watchdog_thresh
1169 */
1170static int proc_watchdog_thresh(const struct ctl_table *table, int write,
1171 void *buffer, size_t *lenp, loff_t *ppos)
1172{
1173 int err, old;
1174
1175 mutex_lock(&watchdog_mutex);
1176
1177 watchdog_thresh_next = READ_ONCE(watchdog_thresh);
1178
1179 old = watchdog_thresh_next;
1180 err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
1181
1182 if (!err && write && old != READ_ONCE(watchdog_thresh_next))
1183 proc_watchdog_update(true);
1184
1185 mutex_unlock(&watchdog_mutex);
1186 return err;
1187}
1188
1189/*
1190 * The cpumask is the mask of possible cpus that the watchdog can run
1191 * on, not the mask of cpus it is actually running on. This allows the
1192 * user to specify a mask that will include cpus that have not yet
1193 * been brought online, if desired.
1194 */
1195static int proc_watchdog_cpumask(const struct ctl_table *table, int write,
1196 void *buffer, size_t *lenp, loff_t *ppos)
1197{
1198 int err;
1199
1200 mutex_lock(&watchdog_mutex);
1201
1202 err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
1203 if (!err && write)
1204 proc_watchdog_update(false);
1205
1206 mutex_unlock(&watchdog_mutex);
1207 return err;
1208}
1209
1210static const int sixty = 60;
1211
1212static const struct ctl_table watchdog_sysctls[] = {
1213 {
1214 .procname = "watchdog",
1215 .data = &watchdog_user_enabled,
1216 .maxlen = sizeof(int),
1217 .mode = 0644,
1218 .proc_handler = proc_watchdog,
1219 .extra1 = SYSCTL_ZERO,
1220 .extra2 = SYSCTL_ONE,
1221 },
1222 {
1223 .procname = "watchdog_thresh",
1224 .data = &watchdog_thresh_next,
1225 .maxlen = sizeof(int),
1226 .mode = 0644,
1227 .proc_handler = proc_watchdog_thresh,
1228 .extra1 = SYSCTL_ZERO,
1229 .extra2 = (void *)&sixty,
1230 },
1231 {
1232 .procname = "watchdog_cpumask",
1233 .data = &watchdog_cpumask_bits,
1234 .maxlen = NR_CPUS,
1235 .mode = 0644,
1236 .proc_handler = proc_watchdog_cpumask,
1237 },
1238#ifdef CONFIG_SOFTLOCKUP_DETECTOR
1239 {
1240 .procname = "soft_watchdog",
1241 .data = &watchdog_softlockup_user_enabled,
1242 .maxlen = sizeof(int),
1243 .mode = 0644,
1244 .proc_handler = proc_soft_watchdog,
1245 .extra1 = SYSCTL_ZERO,
1246 .extra2 = SYSCTL_ONE,
1247 },
1248 {
1249 .procname = "softlockup_panic",
1250 .data = &softlockup_panic,
1251 .maxlen = sizeof(int),
1252 .mode = 0644,
1253 .proc_handler = proc_dointvec_minmax,
1254 .extra1 = SYSCTL_ZERO,
1255 .extra2 = SYSCTL_INT_MAX,
1256 },
1257 {
1258 .procname = "softlockup_sys_info",
1259 .data = &softlockup_si_mask,
1260 .maxlen = sizeof(softlockup_si_mask),
1261 .mode = 0644,
1262 .proc_handler = sysctl_sys_info_handler,
1263 },
1264#ifdef CONFIG_SMP
1265 {
1266 .procname = "softlockup_all_cpu_backtrace",
1267 .data = &sysctl_softlockup_all_cpu_backtrace,
1268 .maxlen = sizeof(int),
1269 .mode = 0644,
1270 .proc_handler = proc_dointvec_minmax,
1271 .extra1 = SYSCTL_ZERO,
1272 .extra2 = SYSCTL_ONE,
1273 },
1274#endif /* CONFIG_SMP */
1275#endif
1276#ifdef CONFIG_HARDLOCKUP_DETECTOR
1277 {
1278 .procname = "hardlockup_panic",
1279 .data = &hardlockup_panic,
1280 .maxlen = sizeof(int),
1281 .mode = 0644,
1282 .proc_handler = proc_dointvec_minmax,
1283 .extra1 = SYSCTL_ZERO,
1284 .extra2 = SYSCTL_ONE,
1285 },
1286 {
1287 .procname = "hardlockup_sys_info",
1288 .data = &hardlockup_si_mask,
1289 .maxlen = sizeof(hardlockup_si_mask),
1290 .mode = 0644,
1291 .proc_handler = sysctl_sys_info_handler,
1292 },
1293#ifdef CONFIG_SMP
1294 {
1295 .procname = "hardlockup_all_cpu_backtrace",
1296 .data = &sysctl_hardlockup_all_cpu_backtrace,
1297 .maxlen = sizeof(int),
1298 .mode = 0644,
1299 .proc_handler = proc_dointvec_minmax,
1300 .extra1 = SYSCTL_ZERO,
1301 .extra2 = SYSCTL_ONE,
1302 },
1303#endif /* CONFIG_SMP */
1304#endif
1305 {
1306 .procname = "nmi_watchdog",
1307 .data = &watchdog_hardlockup_user_enabled,
1308 .maxlen = sizeof(int),
1309 .mode = 0644,
1310 .proc_handler = proc_nmi_watchdog,
1311 .extra1 = SYSCTL_ZERO,
1312 .extra2 = SYSCTL_ONE,
1313 },
1314};
1315
1316static void __init watchdog_sysctl_init(void)
1317{
1318 register_sysctl_init("kernel", watchdog_sysctls);
1319}
1320
1321#else
1322#define watchdog_sysctl_init() do { } while (0)
1323#endif /* CONFIG_SYSCTL */
1324
1325static void __init lockup_detector_delay_init(struct work_struct *work);
1326static bool allow_lockup_detector_init_retry __initdata;
1327
1328static struct work_struct detector_work __initdata =
1329 __WORK_INITIALIZER(detector_work, lockup_detector_delay_init);
1330
1331static void __init lockup_detector_delay_init(struct work_struct *work)
1332{
1333 int ret;
1334
1335 ret = watchdog_hardlockup_probe();
1336 if (ret) {
1337 if (ret == -ENODEV)
1338 pr_info("NMI not fully supported\n");
1339 else
1340 pr_info("Delayed init of the lockup detector failed: %d\n", ret);
1341 pr_info("Hard watchdog permanently disabled\n");
1342 return;
1343 }
1344
1345 allow_lockup_detector_init_retry = false;
1346
1347 watchdog_hardlockup_available = true;
1348 lockup_detector_setup();
1349}
1350
1351/*
1352 * lockup_detector_retry_init - retry init lockup detector if possible.
1353 *
1354 * Retry hardlockup detector init. It is useful when it requires some
1355 * functionality that has to be initialized later on a particular
1356 * platform.
1357 */
1358void __init lockup_detector_retry_init(void)
1359{
1360 /* Must be called before late init calls */
1361 if (!allow_lockup_detector_init_retry)
1362 return;
1363
1364 schedule_work(&detector_work);
1365}
1366
1367/*
1368 * Ensure that optional delayed hardlockup init is proceed before
1369 * the init code and memory is freed.
1370 */
1371static int __init lockup_detector_check(void)
1372{
1373 /* Prevent any later retry. */
1374 allow_lockup_detector_init_retry = false;
1375
1376 /* Make sure no work is pending. */
1377 flush_work(&detector_work);
1378
1379 watchdog_sysctl_init();
1380
1381 return 0;
1382
1383}
1384late_initcall_sync(lockup_detector_check);
1385
1386void __init lockup_detector_init(void)
1387{
1388 if (tick_nohz_full_enabled())
1389 pr_info("Disabling watchdog on nohz_full cores by default\n");
1390
1391 cpumask_copy(&watchdog_cpumask,
1392 housekeeping_cpumask(HK_TYPE_TIMER));
1393
1394 if (!watchdog_hardlockup_probe())
1395 watchdog_hardlockup_available = true;
1396 else
1397 allow_lockup_detector_init_retry = true;
1398
1399 lockup_detector_setup();
1400}