Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
58687acb DZ |
2 | /* |
3 | * Detect hard and soft lockups on a system | |
4 | * | |
5 | * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. | |
6 | * | |
86f5e6a7 FLVC |
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 | |
58687acb DZ |
10 | * to those contributors as well. |
11 | */ | |
12 | ||
5f92a7b0 | 13 | #define pr_fmt(fmt) "watchdog: " fmt |
4501980a | 14 | |
58687acb DZ |
15 | #include <linux/mm.h> |
16 | #include <linux/cpu.h> | |
17 | #include <linux/nmi.h> | |
18 | #include <linux/init.h> | |
58687acb DZ |
19 | #include <linux/module.h> |
20 | #include <linux/sysctl.h> | |
fe4ba3c3 | 21 | #include <linux/tick.h> |
e6017571 | 22 | #include <linux/sched/clock.h> |
b17b0153 | 23 | #include <linux/sched/debug.h> |
78634061 | 24 | #include <linux/sched/isolation.h> |
9cf57731 | 25 | #include <linux/stop_machine.h> |
58687acb DZ |
26 | |
27 | #include <asm/irq_regs.h> | |
5d1c0f4a | 28 | #include <linux/kvm_para.h> |
58687acb | 29 | |
946d1977 | 30 | static DEFINE_MUTEX(watchdog_mutex); |
ab992dc3 | 31 | |
05a4a952 | 32 | #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG) |
df95d308 | 33 | # define WATCHDOG_HARDLOCKUP_DEFAULT 1 |
84d56e66 | 34 | #else |
df95d308 | 35 | # define WATCHDOG_HARDLOCKUP_DEFAULT 0 |
84d56e66 | 36 | #endif |
05a4a952 | 37 | |
09154985 TG |
38 | unsigned long __read_mostly watchdog_enabled; |
39 | int __read_mostly watchdog_user_enabled = 1; | |
df95d308 DA |
40 | static int __read_mostly watchdog_hardlockup_user_enabled = WATCHDOG_HARDLOCKUP_DEFAULT; |
41 | static int __read_mostly watchdog_softlockup_user_enabled = 1; | |
7feeb9cd | 42 | int __read_mostly watchdog_thresh = 10; |
df95d308 | 43 | static int __read_mostly watchdog_hardlockup_available; |
7feeb9cd | 44 | |
7feeb9cd TG |
45 | struct cpumask watchdog_cpumask __read_mostly; |
46 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); | |
47 | ||
05a4a952 | 48 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
f117955a GP |
49 | |
50 | # ifdef CONFIG_SMP | |
51 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; | |
52 | # endif /* CONFIG_SMP */ | |
53 | ||
05a4a952 NP |
54 | /* |
55 | * Should we panic when a soft-lockup or hard-lockup occurs: | |
56 | */ | |
57 | unsigned int __read_mostly hardlockup_panic = | |
67fca000 | 58 | IS_ENABLED(CONFIG_BOOTPARAM_HARDLOCKUP_PANIC); |
05a4a952 NP |
59 | /* |
60 | * We may not want to enable hard lockup detection by default in all cases, | |
61 | * for example when running the kernel as a guest on a hypervisor. In these | |
62 | * cases this function can be called to disable hard lockup detection. This | |
63 | * function should only be executed once by the boot processor before the | |
64 | * kernel command line parameters are parsed, because otherwise it is not | |
65 | * possible to override this in hardlockup_panic_setup(). | |
66 | */ | |
7a355820 | 67 | void __init hardlockup_detector_disable(void) |
05a4a952 | 68 | { |
df95d308 | 69 | watchdog_hardlockup_user_enabled = 0; |
05a4a952 NP |
70 | } |
71 | ||
72 | static int __init hardlockup_panic_setup(char *str) | |
73 | { | |
74 | if (!strncmp(str, "panic", 5)) | |
75 | hardlockup_panic = 1; | |
76 | else if (!strncmp(str, "nopanic", 7)) | |
77 | hardlockup_panic = 0; | |
78 | else if (!strncmp(str, "0", 1)) | |
df95d308 | 79 | watchdog_hardlockup_user_enabled = 0; |
05a4a952 | 80 | else if (!strncmp(str, "1", 1)) |
df95d308 | 81 | watchdog_hardlockup_user_enabled = 1; |
05a4a952 NP |
82 | return 1; |
83 | } | |
84 | __setup("nmi_watchdog=", hardlockup_panic_setup); | |
85 | ||
368a7e2c | 86 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
05a4a952 | 87 | |
1f423c90 | 88 | #if defined(CONFIG_HARDLOCKUP_DETECTOR_COUNTS_HRTIMER) |
81972551 | 89 | |
77c12fc9 DA |
90 | static DEFINE_PER_CPU(atomic_t, hrtimer_interrupts); |
91 | static DEFINE_PER_CPU(int, hrtimer_interrupts_saved); | |
1610611a | 92 | static DEFINE_PER_CPU(bool, watchdog_hardlockup_warned); |
ed92e1ef | 93 | static DEFINE_PER_CPU(bool, watchdog_hardlockup_touched); |
1610611a | 94 | static unsigned long watchdog_hardlockup_all_cpu_dumped; |
81972551 | 95 | |
ed92e1ef DA |
96 | notrace void arch_touch_nmi_watchdog(void) |
97 | { | |
98 | /* | |
99 | * Using __raw here because some code paths have | |
100 | * preemption enabled. If preemption is enabled | |
101 | * then interrupts should be enabled too, in which | |
102 | * case we shouldn't have to worry about the watchdog | |
103 | * going off. | |
104 | */ | |
105 | raw_cpu_write(watchdog_hardlockup_touched, true); | |
106 | } | |
107 | EXPORT_SYMBOL(arch_touch_nmi_watchdog); | |
108 | ||
1f423c90 DA |
109 | void watchdog_hardlockup_touch_cpu(unsigned int cpu) |
110 | { | |
111 | per_cpu(watchdog_hardlockup_touched, cpu) = true; | |
112 | ||
113 | /* Match with smp_rmb() in watchdog_hardlockup_check() */ | |
114 | smp_wmb(); | |
115 | } | |
116 | ||
77c12fc9 | 117 | static bool is_hardlockup(unsigned int cpu) |
81972551 | 118 | { |
77c12fc9 | 119 | int hrint = atomic_read(&per_cpu(hrtimer_interrupts, cpu)); |
81972551 | 120 | |
77c12fc9 | 121 | if (per_cpu(hrtimer_interrupts_saved, cpu) == hrint) |
81972551 DA |
122 | return true; |
123 | ||
77c12fc9 DA |
124 | /* |
125 | * NOTE: we don't need any fancy atomic_t or READ_ONCE/WRITE_ONCE | |
126 | * for hrtimer_interrupts_saved. hrtimer_interrupts_saved is | |
127 | * written/read by a single CPU. | |
128 | */ | |
129 | per_cpu(hrtimer_interrupts_saved, cpu) = hrint; | |
1610611a | 130 | |
81972551 DA |
131 | return false; |
132 | } | |
133 | ||
d3b62ace | 134 | static void watchdog_hardlockup_kick(void) |
81972551 | 135 | { |
d3b62ace DA |
136 | int new_interrupts; |
137 | ||
138 | new_interrupts = atomic_inc_return(this_cpu_ptr(&hrtimer_interrupts)); | |
139 | watchdog_buddy_check_hardlockup(new_interrupts); | |
81972551 DA |
140 | } |
141 | ||
77c12fc9 | 142 | void watchdog_hardlockup_check(unsigned int cpu, struct pt_regs *regs) |
81972551 | 143 | { |
1f423c90 DA |
144 | /* Match with smp_wmb() in watchdog_hardlockup_touch_cpu() */ |
145 | smp_rmb(); | |
146 | ||
ed92e1ef DA |
147 | if (per_cpu(watchdog_hardlockup_touched, cpu)) { |
148 | per_cpu(watchdog_hardlockup_touched, cpu) = false; | |
149 | return; | |
150 | } | |
151 | ||
1610611a DA |
152 | /* |
153 | * Check for a hardlockup by making sure the CPU's timer | |
154 | * interrupt is incrementing. The timer interrupt should have | |
155 | * fired multiple times before we overflow'd. If it hasn't | |
81972551 DA |
156 | * then this is a good indication the cpu is stuck |
157 | */ | |
77c12fc9 | 158 | if (is_hardlockup(cpu)) { |
1610611a | 159 | unsigned int this_cpu = smp_processor_id(); |
7a71d8e6 DA |
160 | struct cpumask backtrace_mask; |
161 | ||
162 | cpumask_copy(&backtrace_mask, cpu_online_mask); | |
81972551 | 163 | |
1610611a | 164 | /* Only print hardlockups once. */ |
77c12fc9 | 165 | if (per_cpu(watchdog_hardlockup_warned, cpu)) |
81972551 DA |
166 | return; |
167 | ||
77c12fc9 | 168 | pr_emerg("Watchdog detected hard LOCKUP on cpu %d\n", cpu); |
81972551 DA |
169 | print_modules(); |
170 | print_irqtrace_events(current); | |
77c12fc9 DA |
171 | if (cpu == this_cpu) { |
172 | if (regs) | |
173 | show_regs(regs); | |
174 | else | |
175 | dump_stack(); | |
176 | cpumask_clear_cpu(cpu, &backtrace_mask); | |
177 | } else { | |
178 | if (trigger_single_cpu_backtrace(cpu)) | |
179 | cpumask_clear_cpu(cpu, &backtrace_mask); | |
180 | } | |
81972551 DA |
181 | |
182 | /* | |
77c12fc9 DA |
183 | * Perform multi-CPU dump only once to avoid multiple |
184 | * hardlockups generating interleaving traces | |
81972551 DA |
185 | */ |
186 | if (sysctl_hardlockup_all_cpu_backtrace && | |
1610611a | 187 | !test_and_set_bit(0, &watchdog_hardlockup_all_cpu_dumped)) |
77c12fc9 | 188 | trigger_cpumask_backtrace(&backtrace_mask); |
81972551 DA |
189 | |
190 | if (hardlockup_panic) | |
191 | nmi_panic(regs, "Hard LOCKUP"); | |
192 | ||
77c12fc9 | 193 | per_cpu(watchdog_hardlockup_warned, cpu) = true; |
1610611a | 194 | } else { |
77c12fc9 | 195 | per_cpu(watchdog_hardlockup_warned, cpu) = false; |
81972551 | 196 | } |
81972551 DA |
197 | } |
198 | ||
1f423c90 | 199 | #else /* CONFIG_HARDLOCKUP_DETECTOR_COUNTS_HRTIMER */ |
81972551 | 200 | |
d3b62ace | 201 | static inline void watchdog_hardlockup_kick(void) { } |
81972551 | 202 | |
1f423c90 | 203 | #endif /* !CONFIG_HARDLOCKUP_DETECTOR_COUNTS_HRTIMER */ |
81972551 | 204 | |
05a4a952 | 205 | /* |
d9b3629a | 206 | * These functions can be overridden based on the configured hardlockdup detector. |
a10a842f | 207 | * |
df95d308 | 208 | * watchdog_hardlockup_enable/disable can be implemented to start and stop when |
d9b3629a | 209 | * softlockup watchdog start and stop. The detector must select the |
a10a842f | 210 | * SOFTLOCKUP_DETECTOR Kconfig. |
05a4a952 | 211 | */ |
d9b3629a | 212 | void __weak watchdog_hardlockup_enable(unsigned int cpu) { } |
941154bd | 213 | |
d9b3629a | 214 | void __weak watchdog_hardlockup_disable(unsigned int cpu) { } |
05a4a952 | 215 | |
930d8f8d LC |
216 | /* |
217 | * Watchdog-detector specific API. | |
218 | * | |
219 | * Return 0 when hardlockup watchdog is available, negative value otherwise. | |
220 | * Note that the negative value means that a delayed probe might | |
221 | * succeed later. | |
222 | */ | |
df95d308 | 223 | int __weak __init watchdog_hardlockup_probe(void) |
a994a314 | 224 | { |
d9b3629a | 225 | return -ENODEV; |
a994a314 TG |
226 | } |
227 | ||
6592ad2f | 228 | /** |
df95d308 | 229 | * watchdog_hardlockup_stop - Stop the watchdog for reconfiguration |
6592ad2f | 230 | * |
6b9dc480 | 231 | * The reconfiguration steps are: |
df95d308 | 232 | * watchdog_hardlockup_stop(); |
6592ad2f | 233 | * update_variables(); |
df95d308 | 234 | * watchdog_hardlockup_start(); |
6b9dc480 | 235 | */ |
df95d308 | 236 | void __weak watchdog_hardlockup_stop(void) { } |
6b9dc480 TG |
237 | |
238 | /** | |
df95d308 | 239 | * watchdog_hardlockup_start - Start the watchdog after reconfiguration |
6592ad2f | 240 | * |
df95d308 | 241 | * Counterpart to watchdog_hardlockup_stop(). |
6b9dc480 TG |
242 | * |
243 | * The following variables have been updated in update_variables() and | |
244 | * contain the currently valid configuration: | |
7feeb9cd | 245 | * - watchdog_enabled |
a10a842f NP |
246 | * - watchdog_thresh |
247 | * - watchdog_cpumask | |
a10a842f | 248 | */ |
df95d308 | 249 | void __weak watchdog_hardlockup_start(void) { } |
a10a842f | 250 | |
09154985 TG |
251 | /** |
252 | * lockup_detector_update_enable - Update the sysctl enable bit | |
253 | * | |
df95d308 DA |
254 | * Caller needs to make sure that the hard watchdogs are off, so this |
255 | * can't race with watchdog_hardlockup_disable(). | |
09154985 TG |
256 | */ |
257 | static void lockup_detector_update_enable(void) | |
258 | { | |
259 | watchdog_enabled = 0; | |
260 | if (!watchdog_user_enabled) | |
261 | return; | |
df95d308 DA |
262 | if (watchdog_hardlockup_available && watchdog_hardlockup_user_enabled) |
263 | watchdog_enabled |= WATCHDOG_HARDLOCKUP_ENABLED; | |
264 | if (watchdog_softlockup_user_enabled) | |
265 | watchdog_enabled |= WATCHDOG_SOFTOCKUP_ENABLED; | |
09154985 TG |
266 | } |
267 | ||
05a4a952 NP |
268 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR |
269 | ||
fef06efc PM |
270 | /* |
271 | * Delay the soflockup report when running a known slow code. | |
272 | * It does _not_ affect the timestamp of the last successdul reschedule. | |
273 | */ | |
274 | #define SOFTLOCKUP_DELAY_REPORT ULONG_MAX | |
11e31f60 | 275 | |
f117955a GP |
276 | #ifdef CONFIG_SMP |
277 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; | |
278 | #endif | |
279 | ||
e7e04615 SS |
280 | static struct cpumask watchdog_allowed_mask __read_mostly; |
281 | ||
2b9d7f23 TG |
282 | /* Global variables, exported for sysctl */ |
283 | unsigned int __read_mostly softlockup_panic = | |
67fca000 | 284 | IS_ENABLED(CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC); |
2eb2527f | 285 | |
9cf57731 | 286 | static bool softlockup_initialized __read_mostly; |
0f34c400 | 287 | static u64 __read_mostly sample_period; |
58687acb | 288 | |
fef06efc | 289 | /* Timestamp taken after the last successful reschedule. */ |
58687acb | 290 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
fef06efc PM |
291 | /* Timestamp of the last softlockup report. */ |
292 | static DEFINE_PER_CPU(unsigned long, watchdog_report_ts); | |
58687acb DZ |
293 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
294 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); | |
ed235875 | 295 | static unsigned long soft_lockup_nmi_warn; |
58687acb | 296 | |
58687acb DZ |
297 | static int __init nowatchdog_setup(char *str) |
298 | { | |
09154985 | 299 | watchdog_user_enabled = 0; |
58687acb DZ |
300 | return 1; |
301 | } | |
302 | __setup("nowatchdog", nowatchdog_setup); | |
303 | ||
58687acb DZ |
304 | static int __init nosoftlockup_setup(char *str) |
305 | { | |
df95d308 | 306 | watchdog_softlockup_user_enabled = 0; |
58687acb DZ |
307 | return 1; |
308 | } | |
309 | __setup("nosoftlockup", nosoftlockup_setup); | |
195daf66 | 310 | |
11295055 LO |
311 | static int __init watchdog_thresh_setup(char *str) |
312 | { | |
313 | get_option(&str, &watchdog_thresh); | |
314 | return 1; | |
315 | } | |
316 | __setup("watchdog_thresh=", watchdog_thresh_setup); | |
317 | ||
941154bd TG |
318 | static void __lockup_detector_cleanup(void); |
319 | ||
4eec42f3 MSB |
320 | /* |
321 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- | |
322 | * lockups can have false positives under extreme conditions. So we generally | |
323 | * want a higher threshold for soft lockups than for hard lockups. So we couple | |
324 | * the thresholds with a factor: we make the soft threshold twice the amount of | |
325 | * time the hard threshold is. | |
326 | */ | |
6e9101ae | 327 | static int get_softlockup_thresh(void) |
4eec42f3 MSB |
328 | { |
329 | return watchdog_thresh * 2; | |
330 | } | |
58687acb DZ |
331 | |
332 | /* | |
333 | * Returns seconds, approximately. We don't need nanosecond | |
334 | * resolution, and we don't need to waste time with a big divide when | |
335 | * 2^30ns == 1.074s. | |
336 | */ | |
c06b4f19 | 337 | static unsigned long get_timestamp(void) |
58687acb | 338 | { |
545a2bf7 | 339 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
58687acb DZ |
340 | } |
341 | ||
0f34c400 | 342 | static void set_sample_period(void) |
58687acb DZ |
343 | { |
344 | /* | |
586692a5 | 345 | * convert watchdog_thresh from seconds to ns |
86f5e6a7 FLVC |
346 | * the divide by 5 is to give hrtimer several chances (two |
347 | * or three with the current relation between the soft | |
348 | * and hard thresholds) to increment before the | |
349 | * hardlockup detector generates a warning | |
58687acb | 350 | */ |
0f34c400 | 351 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
7edaeb68 | 352 | watchdog_update_hrtimer_threshold(sample_period); |
58687acb DZ |
353 | } |
354 | ||
fef06efc PM |
355 | static void update_report_ts(void) |
356 | { | |
357 | __this_cpu_write(watchdog_report_ts, get_timestamp()); | |
358 | } | |
359 | ||
58687acb | 360 | /* Commands for resetting the watchdog */ |
7c0012f5 | 361 | static void update_touch_ts(void) |
58687acb | 362 | { |
c06b4f19 | 363 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
fef06efc | 364 | update_report_ts(); |
58687acb DZ |
365 | } |
366 | ||
03e0d461 TH |
367 | /** |
368 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls | |
369 | * | |
370 | * Call when the scheduler may have stalled for legitimate reasons | |
371 | * preventing the watchdog task from executing - e.g. the scheduler | |
372 | * entering idle state. This should only be used for scheduler events. | |
373 | * Use touch_softlockup_watchdog() for everything else. | |
374 | */ | |
cb9d7fd5 | 375 | notrace void touch_softlockup_watchdog_sched(void) |
58687acb | 376 | { |
7861144b | 377 | /* |
fef06efc PM |
378 | * Preemption can be enabled. It doesn't matter which CPU's watchdog |
379 | * report period gets restarted here, so use the raw_ operation. | |
7861144b | 380 | */ |
fef06efc | 381 | raw_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT); |
58687acb | 382 | } |
03e0d461 | 383 | |
cb9d7fd5 | 384 | notrace void touch_softlockup_watchdog(void) |
03e0d461 TH |
385 | { |
386 | touch_softlockup_watchdog_sched(); | |
82607adc | 387 | wq_watchdog_touch(raw_smp_processor_id()); |
03e0d461 | 388 | } |
0167c781 | 389 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
58687acb | 390 | |
332fbdbc | 391 | void touch_all_softlockup_watchdogs(void) |
58687acb DZ |
392 | { |
393 | int cpu; | |
394 | ||
395 | /* | |
d57108d4 TG |
396 | * watchdog_mutex cannpt be taken here, as this might be called |
397 | * from (soft)interrupt context, so the access to | |
398 | * watchdog_allowed_cpumask might race with a concurrent update. | |
399 | * | |
400 | * The watchdog time stamp can race against a concurrent real | |
401 | * update as well, the only side effect might be a cycle delay for | |
402 | * the softlockup check. | |
58687acb | 403 | */ |
89e28ce6 | 404 | for_each_cpu(cpu, &watchdog_allowed_mask) { |
fef06efc | 405 | per_cpu(watchdog_report_ts, cpu) = SOFTLOCKUP_DELAY_REPORT; |
89e28ce6 WQ |
406 | wq_watchdog_touch(cpu); |
407 | } | |
58687acb DZ |
408 | } |
409 | ||
58687acb DZ |
410 | void touch_softlockup_watchdog_sync(void) |
411 | { | |
f7f66b05 | 412 | __this_cpu_write(softlockup_touch_sync, true); |
fef06efc | 413 | __this_cpu_write(watchdog_report_ts, SOFTLOCKUP_DELAY_REPORT); |
58687acb DZ |
414 | } |
415 | ||
0f90b88d PM |
416 | static int is_softlockup(unsigned long touch_ts, |
417 | unsigned long period_ts, | |
418 | unsigned long now) | |
58687acb | 419 | { |
df95d308 | 420 | if ((watchdog_enabled & WATCHDOG_SOFTOCKUP_ENABLED) && watchdog_thresh) { |
195daf66 | 421 | /* Warn about unreasonable delays. */ |
fef06efc | 422 | if (time_after(now, period_ts + get_softlockup_thresh())) |
195daf66 UO |
423 | return now - touch_ts; |
424 | } | |
58687acb DZ |
425 | return 0; |
426 | } | |
427 | ||
05a4a952 | 428 | /* watchdog detector functions */ |
be45bf53 PZ |
429 | static DEFINE_PER_CPU(struct completion, softlockup_completion); |
430 | static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work); | |
431 | ||
9cf57731 | 432 | /* |
b124ac45 | 433 | * The watchdog feed function - touches the timestamp. |
9cf57731 PZ |
434 | * |
435 | * It only runs once every sample_period seconds (4 seconds by | |
436 | * default) to reset the softlockup timestamp. If this gets delayed | |
437 | * for more than 2*watchdog_thresh seconds then the debug-printout | |
438 | * triggers in watchdog_timer_fn(). | |
439 | */ | |
440 | static int softlockup_fn(void *data) | |
441 | { | |
7c0012f5 | 442 | update_touch_ts(); |
be45bf53 | 443 | complete(this_cpu_ptr(&softlockup_completion)); |
9cf57731 PZ |
444 | |
445 | return 0; | |
446 | } | |
447 | ||
58687acb DZ |
448 | /* watchdog kicker functions */ |
449 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
450 | { | |
0f90b88d | 451 | unsigned long touch_ts, period_ts, now; |
58687acb DZ |
452 | struct pt_regs *regs = get_irq_regs(); |
453 | int duration; | |
ed235875 | 454 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
58687acb | 455 | |
01f0a027 | 456 | if (!watchdog_enabled) |
b94f5118 DZ |
457 | return HRTIMER_NORESTART; |
458 | ||
d3b62ace | 459 | watchdog_hardlockup_kick(); |
58687acb DZ |
460 | |
461 | /* kick the softlockup detector */ | |
be45bf53 PZ |
462 | if (completion_done(this_cpu_ptr(&softlockup_completion))) { |
463 | reinit_completion(this_cpu_ptr(&softlockup_completion)); | |
464 | stop_one_cpu_nowait(smp_processor_id(), | |
465 | softlockup_fn, NULL, | |
466 | this_cpu_ptr(&softlockup_stop_work)); | |
467 | } | |
58687acb DZ |
468 | |
469 | /* .. and repeat */ | |
0f34c400 | 470 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
58687acb | 471 | |
0f90b88d PM |
472 | /* |
473 | * Read the current timestamp first. It might become invalid anytime | |
474 | * when a virtual machine is stopped by the host or when the watchog | |
475 | * is touched from NMI. | |
476 | */ | |
477 | now = get_timestamp(); | |
9bf3bc94 PM |
478 | /* |
479 | * If a virtual machine is stopped by the host it can look to | |
0f90b88d | 480 | * the watchdog like a soft lockup. This function touches the watchdog. |
9bf3bc94 PM |
481 | */ |
482 | kvm_check_and_clear_guest_paused(); | |
0f90b88d PM |
483 | /* |
484 | * The stored timestamp is comparable with @now only when not touched. | |
485 | * It might get touched anytime from NMI. Make sure that is_softlockup() | |
486 | * uses the same (valid) value. | |
487 | */ | |
488 | period_ts = READ_ONCE(*this_cpu_ptr(&watchdog_report_ts)); | |
9bf3bc94 PM |
489 | |
490 | /* Reset the interval when touched by known problematic code. */ | |
fef06efc | 491 | if (period_ts == SOFTLOCKUP_DELAY_REPORT) { |
909ea964 | 492 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
493 | /* |
494 | * If the time stamp was touched atomically | |
495 | * make sure the scheduler tick is up to date. | |
496 | */ | |
909ea964 | 497 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
498 | sched_clock_tick(); |
499 | } | |
5d1c0f4a | 500 | |
fef06efc | 501 | update_report_ts(); |
58687acb DZ |
502 | return HRTIMER_RESTART; |
503 | } | |
504 | ||
0f90b88d PM |
505 | /* Check for a softlockup. */ |
506 | touch_ts = __this_cpu_read(watchdog_touch_ts); | |
507 | duration = is_softlockup(touch_ts, period_ts, now); | |
58687acb | 508 | if (unlikely(duration)) { |
9f113bf7 PM |
509 | /* |
510 | * Prevent multiple soft-lockup reports if one cpu is already | |
511 | * engaged in dumping all cpu back traces. | |
512 | */ | |
ed235875 | 513 | if (softlockup_all_cpu_backtrace) { |
9f113bf7 | 514 | if (test_and_set_bit_lock(0, &soft_lockup_nmi_warn)) |
ed235875 | 515 | return HRTIMER_RESTART; |
ed235875 AT |
516 | } |
517 | ||
c9ad17c9 | 518 | /* Start period for the next softlockup warning. */ |
fef06efc | 519 | update_report_ts(); |
c9ad17c9 | 520 | |
656c3b79 | 521 | pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 522 | smp_processor_id(), duration, |
58687acb DZ |
523 | current->comm, task_pid_nr(current)); |
524 | print_modules(); | |
525 | print_irqtrace_events(current); | |
526 | if (regs) | |
527 | show_regs(regs); | |
528 | else | |
529 | dump_stack(); | |
530 | ||
ed235875 | 531 | if (softlockup_all_cpu_backtrace) { |
ed235875 | 532 | trigger_allbutself_cpu_backtrace(); |
9f113bf7 | 533 | clear_bit_unlock(0, &soft_lockup_nmi_warn); |
ed235875 AT |
534 | } |
535 | ||
69361eef | 536 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
58687acb DZ |
537 | if (softlockup_panic) |
538 | panic("softlockup: hung tasks"); | |
1bc503cb | 539 | } |
58687acb DZ |
540 | |
541 | return HRTIMER_RESTART; | |
542 | } | |
543 | ||
bcd951cf | 544 | static void watchdog_enable(unsigned int cpu) |
58687acb | 545 | { |
01f0a027 | 546 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
be45bf53 | 547 | struct completion *done = this_cpu_ptr(&softlockup_completion); |
58687acb | 548 | |
9cf57731 PZ |
549 | WARN_ON_ONCE(cpu != smp_processor_id()); |
550 | ||
be45bf53 PZ |
551 | init_completion(done); |
552 | complete(done); | |
553 | ||
01f0a027 | 554 | /* |
df95d308 | 555 | * Start the timer first to prevent the hardlockup watchdog triggering |
01f0a027 TG |
556 | * before the timer has a chance to fire. |
557 | */ | |
d2ab4cf4 | 558 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
3935e895 | 559 | hrtimer->function = watchdog_timer_fn; |
01f0a027 | 560 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
d2ab4cf4 | 561 | HRTIMER_MODE_REL_PINNED_HARD); |
3935e895 | 562 | |
01f0a027 | 563 | /* Initialize timestamp */ |
7c0012f5 | 564 | update_touch_ts(); |
df95d308 DA |
565 | /* Enable the hardlockup detector */ |
566 | if (watchdog_enabled & WATCHDOG_HARDLOCKUP_ENABLED) | |
567 | watchdog_hardlockup_enable(cpu); | |
bcd951cf | 568 | } |
58687acb | 569 | |
bcd951cf TG |
570 | static void watchdog_disable(unsigned int cpu) |
571 | { | |
01f0a027 | 572 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
58687acb | 573 | |
9cf57731 PZ |
574 | WARN_ON_ONCE(cpu != smp_processor_id()); |
575 | ||
01f0a027 | 576 | /* |
df95d308 DA |
577 | * Disable the hardlockup detector first. That prevents that a large |
578 | * delay between disabling the timer and disabling the hardlockup | |
579 | * detector causes a false positive. | |
01f0a027 | 580 | */ |
df95d308 | 581 | watchdog_hardlockup_disable(cpu); |
01f0a027 | 582 | hrtimer_cancel(hrtimer); |
be45bf53 | 583 | wait_for_completion(this_cpu_ptr(&softlockup_completion)); |
58687acb DZ |
584 | } |
585 | ||
9cf57731 | 586 | static int softlockup_stop_fn(void *data) |
b8900bc0 | 587 | { |
9cf57731 PZ |
588 | watchdog_disable(smp_processor_id()); |
589 | return 0; | |
b8900bc0 FW |
590 | } |
591 | ||
9cf57731 | 592 | static void softlockup_stop_all(void) |
bcd951cf | 593 | { |
9cf57731 PZ |
594 | int cpu; |
595 | ||
596 | if (!softlockup_initialized) | |
597 | return; | |
598 | ||
599 | for_each_cpu(cpu, &watchdog_allowed_mask) | |
600 | smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false); | |
601 | ||
602 | cpumask_clear(&watchdog_allowed_mask); | |
bcd951cf TG |
603 | } |
604 | ||
9cf57731 | 605 | static int softlockup_start_fn(void *data) |
bcd951cf | 606 | { |
9cf57731 PZ |
607 | watchdog_enable(smp_processor_id()); |
608 | return 0; | |
bcd951cf | 609 | } |
58687acb | 610 | |
9cf57731 | 611 | static void softlockup_start_all(void) |
2eb2527f | 612 | { |
9cf57731 | 613 | int cpu; |
2eb2527f | 614 | |
9cf57731 PZ |
615 | cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask); |
616 | for_each_cpu(cpu, &watchdog_allowed_mask) | |
617 | smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false); | |
2eb2527f TG |
618 | } |
619 | ||
9cf57731 | 620 | int lockup_detector_online_cpu(unsigned int cpu) |
2eb2527f | 621 | { |
7dd47617 TG |
622 | if (cpumask_test_cpu(cpu, &watchdog_allowed_mask)) |
623 | watchdog_enable(cpu); | |
9cf57731 | 624 | return 0; |
2eb2527f TG |
625 | } |
626 | ||
9cf57731 | 627 | int lockup_detector_offline_cpu(unsigned int cpu) |
2eb2527f | 628 | { |
7dd47617 TG |
629 | if (cpumask_test_cpu(cpu, &watchdog_allowed_mask)) |
630 | watchdog_disable(cpu); | |
9cf57731 | 631 | return 0; |
2eb2527f TG |
632 | } |
633 | ||
7c56a873 | 634 | static void __lockup_detector_reconfigure(void) |
2eb2527f | 635 | { |
e31d6883 | 636 | cpus_read_lock(); |
df95d308 | 637 | watchdog_hardlockup_stop(); |
9cf57731 PZ |
638 | |
639 | softlockup_stop_all(); | |
2eb2527f | 640 | set_sample_period(); |
09154985 TG |
641 | lockup_detector_update_enable(); |
642 | if (watchdog_enabled && watchdog_thresh) | |
9cf57731 PZ |
643 | softlockup_start_all(); |
644 | ||
df95d308 | 645 | watchdog_hardlockup_start(); |
e31d6883 TG |
646 | cpus_read_unlock(); |
647 | /* | |
648 | * Must be called outside the cpus locked section to prevent | |
649 | * recursive locking in the perf code. | |
650 | */ | |
651 | __lockup_detector_cleanup(); | |
2eb2527f TG |
652 | } |
653 | ||
7c56a873 LD |
654 | void lockup_detector_reconfigure(void) |
655 | { | |
656 | mutex_lock(&watchdog_mutex); | |
657 | __lockup_detector_reconfigure(); | |
658 | mutex_unlock(&watchdog_mutex); | |
659 | } | |
660 | ||
2eb2527f | 661 | /* |
b124ac45 | 662 | * Create the watchdog infrastructure and configure the detector(s). |
2eb2527f | 663 | */ |
5587185d | 664 | static __init void lockup_detector_setup(void) |
2eb2527f | 665 | { |
2eb2527f TG |
666 | /* |
667 | * If sysctl is off and watchdog got disabled on the command line, | |
668 | * nothing to do here. | |
669 | */ | |
09154985 TG |
670 | lockup_detector_update_enable(); |
671 | ||
2eb2527f TG |
672 | if (!IS_ENABLED(CONFIG_SYSCTL) && |
673 | !(watchdog_enabled && watchdog_thresh)) | |
674 | return; | |
675 | ||
2eb2527f | 676 | mutex_lock(&watchdog_mutex); |
7c56a873 | 677 | __lockup_detector_reconfigure(); |
9cf57731 | 678 | softlockup_initialized = true; |
2eb2527f TG |
679 | mutex_unlock(&watchdog_mutex); |
680 | } | |
681 | ||
2b9d7f23 | 682 | #else /* CONFIG_SOFTLOCKUP_DETECTOR */ |
7c56a873 | 683 | static void __lockup_detector_reconfigure(void) |
6592ad2f | 684 | { |
e31d6883 | 685 | cpus_read_lock(); |
df95d308 | 686 | watchdog_hardlockup_stop(); |
09154985 | 687 | lockup_detector_update_enable(); |
df95d308 | 688 | watchdog_hardlockup_start(); |
e31d6883 | 689 | cpus_read_unlock(); |
6592ad2f | 690 | } |
7c56a873 LD |
691 | void lockup_detector_reconfigure(void) |
692 | { | |
693 | __lockup_detector_reconfigure(); | |
694 | } | |
5587185d | 695 | static inline void lockup_detector_setup(void) |
34ddaa3e | 696 | { |
7c56a873 | 697 | __lockup_detector_reconfigure(); |
34ddaa3e | 698 | } |
2b9d7f23 | 699 | #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */ |
05a4a952 | 700 | |
941154bd TG |
701 | static void __lockup_detector_cleanup(void) |
702 | { | |
703 | lockdep_assert_held(&watchdog_mutex); | |
704 | hardlockup_detector_perf_cleanup(); | |
705 | } | |
706 | ||
707 | /** | |
708 | * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes | |
709 | * | |
710 | * Caller must not hold the cpu hotplug rwsem. | |
711 | */ | |
712 | void lockup_detector_cleanup(void) | |
713 | { | |
714 | mutex_lock(&watchdog_mutex); | |
715 | __lockup_detector_cleanup(); | |
716 | mutex_unlock(&watchdog_mutex); | |
717 | } | |
718 | ||
6554fd8c TG |
719 | /** |
720 | * lockup_detector_soft_poweroff - Interface to stop lockup detector(s) | |
721 | * | |
722 | * Special interface for parisc. It prevents lockup detector warnings from | |
723 | * the default pm_poweroff() function which busy loops forever. | |
724 | */ | |
725 | void lockup_detector_soft_poweroff(void) | |
726 | { | |
727 | watchdog_enabled = 0; | |
728 | } | |
729 | ||
58cf690a UO |
730 | #ifdef CONFIG_SYSCTL |
731 | ||
b124ac45 | 732 | /* Propagate any changes to the watchdog infrastructure */ |
d57108d4 | 733 | static void proc_watchdog_update(void) |
a0c9cbb9 | 734 | { |
e8b62b2d TG |
735 | /* Remove impossible cpus to keep sysctl output clean. */ |
736 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask); | |
7c56a873 | 737 | __lockup_detector_reconfigure(); |
a0c9cbb9 UO |
738 | } |
739 | ||
ef246a21 UO |
740 | /* |
741 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter | |
742 | * | |
df95d308 DA |
743 | * caller | table->data points to | 'which' |
744 | * -------------------|----------------------------------|------------------------------- | |
745 | * proc_watchdog | watchdog_user_enabled | WATCHDOG_HARDLOCKUP_ENABLED | | |
746 | * | | WATCHDOG_SOFTOCKUP_ENABLED | |
747 | * -------------------|----------------------------------|------------------------------- | |
748 | * proc_nmi_watchdog | watchdog_hardlockup_user_enabled | WATCHDOG_HARDLOCKUP_ENABLED | |
749 | * -------------------|----------------------------------|------------------------------- | |
750 | * proc_soft_watchdog | watchdog_softlockup_user_enabled | WATCHDOG_SOFTOCKUP_ENABLED | |
ef246a21 UO |
751 | */ |
752 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, | |
32927393 | 753 | void *buffer, size_t *lenp, loff_t *ppos) |
ef246a21 | 754 | { |
09154985 | 755 | int err, old, *param = table->data; |
ef246a21 | 756 | |
946d1977 | 757 | mutex_lock(&watchdog_mutex); |
ef246a21 | 758 | |
ef246a21 | 759 | if (!write) { |
09154985 TG |
760 | /* |
761 | * On read synchronize the userspace interface. This is a | |
762 | * racy snapshot. | |
763 | */ | |
764 | *param = (watchdog_enabled & which) != 0; | |
ef246a21 UO |
765 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
766 | } else { | |
09154985 | 767 | old = READ_ONCE(*param); |
ef246a21 | 768 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
09154985 | 769 | if (!err && old != READ_ONCE(*param)) |
d57108d4 | 770 | proc_watchdog_update(); |
ef246a21 | 771 | } |
946d1977 | 772 | mutex_unlock(&watchdog_mutex); |
ef246a21 UO |
773 | return err; |
774 | } | |
775 | ||
83a80a39 UO |
776 | /* |
777 | * /proc/sys/kernel/watchdog | |
778 | */ | |
779 | int proc_watchdog(struct ctl_table *table, int write, | |
32927393 | 780 | void *buffer, size_t *lenp, loff_t *ppos) |
83a80a39 | 781 | { |
df95d308 DA |
782 | return proc_watchdog_common(WATCHDOG_HARDLOCKUP_ENABLED | |
783 | WATCHDOG_SOFTOCKUP_ENABLED, | |
83a80a39 UO |
784 | table, write, buffer, lenp, ppos); |
785 | } | |
786 | ||
787 | /* | |
788 | * /proc/sys/kernel/nmi_watchdog | |
58687acb | 789 | */ |
83a80a39 | 790 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
32927393 | 791 | void *buffer, size_t *lenp, loff_t *ppos) |
83a80a39 | 792 | { |
df95d308 | 793 | if (!watchdog_hardlockup_available && write) |
a994a314 | 794 | return -ENOTSUPP; |
df95d308 | 795 | return proc_watchdog_common(WATCHDOG_HARDLOCKUP_ENABLED, |
83a80a39 UO |
796 | table, write, buffer, lenp, ppos); |
797 | } | |
798 | ||
799 | /* | |
800 | * /proc/sys/kernel/soft_watchdog | |
801 | */ | |
802 | int proc_soft_watchdog(struct ctl_table *table, int write, | |
32927393 | 803 | void *buffer, size_t *lenp, loff_t *ppos) |
83a80a39 | 804 | { |
df95d308 | 805 | return proc_watchdog_common(WATCHDOG_SOFTOCKUP_ENABLED, |
83a80a39 UO |
806 | table, write, buffer, lenp, ppos); |
807 | } | |
58687acb | 808 | |
83a80a39 UO |
809 | /* |
810 | * /proc/sys/kernel/watchdog_thresh | |
811 | */ | |
812 | int proc_watchdog_thresh(struct ctl_table *table, int write, | |
32927393 | 813 | void *buffer, size_t *lenp, loff_t *ppos) |
58687acb | 814 | { |
d57108d4 | 815 | int err, old; |
58687acb | 816 | |
946d1977 | 817 | mutex_lock(&watchdog_mutex); |
bcd951cf | 818 | |
d57108d4 | 819 | old = READ_ONCE(watchdog_thresh); |
b8900bc0 | 820 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
83a80a39 | 821 | |
d57108d4 TG |
822 | if (!err && write && old != READ_ONCE(watchdog_thresh)) |
823 | proc_watchdog_update(); | |
e04ab2bc | 824 | |
946d1977 | 825 | mutex_unlock(&watchdog_mutex); |
b8900bc0 | 826 | return err; |
58687acb | 827 | } |
fe4ba3c3 CM |
828 | |
829 | /* | |
830 | * The cpumask is the mask of possible cpus that the watchdog can run | |
831 | * on, not the mask of cpus it is actually running on. This allows the | |
832 | * user to specify a mask that will include cpus that have not yet | |
833 | * been brought online, if desired. | |
834 | */ | |
835 | int proc_watchdog_cpumask(struct ctl_table *table, int write, | |
32927393 | 836 | void *buffer, size_t *lenp, loff_t *ppos) |
fe4ba3c3 CM |
837 | { |
838 | int err; | |
839 | ||
946d1977 | 840 | mutex_lock(&watchdog_mutex); |
8c073d27 | 841 | |
fe4ba3c3 | 842 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
05ba3de7 | 843 | if (!err && write) |
e8b62b2d | 844 | proc_watchdog_update(); |
5490125d | 845 | |
946d1977 | 846 | mutex_unlock(&watchdog_mutex); |
fe4ba3c3 CM |
847 | return err; |
848 | } | |
dd0693fd XN |
849 | |
850 | static const int sixty = 60; | |
851 | ||
852 | static struct ctl_table watchdog_sysctls[] = { | |
853 | { | |
854 | .procname = "watchdog", | |
855 | .data = &watchdog_user_enabled, | |
856 | .maxlen = sizeof(int), | |
857 | .mode = 0644, | |
858 | .proc_handler = proc_watchdog, | |
859 | .extra1 = SYSCTL_ZERO, | |
860 | .extra2 = SYSCTL_ONE, | |
861 | }, | |
862 | { | |
863 | .procname = "watchdog_thresh", | |
864 | .data = &watchdog_thresh, | |
865 | .maxlen = sizeof(int), | |
866 | .mode = 0644, | |
867 | .proc_handler = proc_watchdog_thresh, | |
868 | .extra1 = SYSCTL_ZERO, | |
869 | .extra2 = (void *)&sixty, | |
870 | }, | |
dd0693fd XN |
871 | { |
872 | .procname = "watchdog_cpumask", | |
873 | .data = &watchdog_cpumask_bits, | |
874 | .maxlen = NR_CPUS, | |
875 | .mode = 0644, | |
876 | .proc_handler = proc_watchdog_cpumask, | |
877 | }, | |
878 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR | |
879 | { | |
880 | .procname = "soft_watchdog", | |
df95d308 | 881 | .data = &watchdog_softlockup_user_enabled, |
dd0693fd XN |
882 | .maxlen = sizeof(int), |
883 | .mode = 0644, | |
884 | .proc_handler = proc_soft_watchdog, | |
885 | .extra1 = SYSCTL_ZERO, | |
886 | .extra2 = SYSCTL_ONE, | |
887 | }, | |
888 | { | |
889 | .procname = "softlockup_panic", | |
890 | .data = &softlockup_panic, | |
891 | .maxlen = sizeof(int), | |
892 | .mode = 0644, | |
893 | .proc_handler = proc_dointvec_minmax, | |
894 | .extra1 = SYSCTL_ZERO, | |
895 | .extra2 = SYSCTL_ONE, | |
896 | }, | |
897 | #ifdef CONFIG_SMP | |
898 | { | |
899 | .procname = "softlockup_all_cpu_backtrace", | |
900 | .data = &sysctl_softlockup_all_cpu_backtrace, | |
901 | .maxlen = sizeof(int), | |
902 | .mode = 0644, | |
903 | .proc_handler = proc_dointvec_minmax, | |
904 | .extra1 = SYSCTL_ZERO, | |
905 | .extra2 = SYSCTL_ONE, | |
906 | }, | |
907 | #endif /* CONFIG_SMP */ | |
908 | #endif | |
909 | #ifdef CONFIG_HARDLOCKUP_DETECTOR | |
910 | { | |
911 | .procname = "hardlockup_panic", | |
912 | .data = &hardlockup_panic, | |
913 | .maxlen = sizeof(int), | |
914 | .mode = 0644, | |
915 | .proc_handler = proc_dointvec_minmax, | |
916 | .extra1 = SYSCTL_ZERO, | |
917 | .extra2 = SYSCTL_ONE, | |
918 | }, | |
919 | #ifdef CONFIG_SMP | |
920 | { | |
921 | .procname = "hardlockup_all_cpu_backtrace", | |
922 | .data = &sysctl_hardlockup_all_cpu_backtrace, | |
923 | .maxlen = sizeof(int), | |
924 | .mode = 0644, | |
925 | .proc_handler = proc_dointvec_minmax, | |
926 | .extra1 = SYSCTL_ZERO, | |
927 | .extra2 = SYSCTL_ONE, | |
928 | }, | |
929 | #endif /* CONFIG_SMP */ | |
930 | #endif | |
931 | {} | |
932 | }; | |
933 | ||
9ec272c5 DA |
934 | static struct ctl_table watchdog_hardlockup_sysctl[] = { |
935 | { | |
936 | .procname = "nmi_watchdog", | |
937 | .data = &watchdog_hardlockup_user_enabled, | |
938 | .maxlen = sizeof(int), | |
939 | .mode = 0444, | |
940 | .proc_handler = proc_nmi_watchdog, | |
941 | .extra1 = SYSCTL_ZERO, | |
942 | .extra2 = SYSCTL_ONE, | |
943 | }, | |
944 | {} | |
945 | }; | |
946 | ||
dd0693fd XN |
947 | static void __init watchdog_sysctl_init(void) |
948 | { | |
949 | register_sysctl_init("kernel", watchdog_sysctls); | |
9ec272c5 DA |
950 | |
951 | if (watchdog_hardlockup_available) | |
952 | watchdog_hardlockup_sysctl[0].mode = 0644; | |
953 | register_sysctl_init("kernel", watchdog_hardlockup_sysctl); | |
dd0693fd | 954 | } |
9ec272c5 | 955 | |
dd0693fd XN |
956 | #else |
957 | #define watchdog_sysctl_init() do { } while (0) | |
58687acb DZ |
958 | #endif /* CONFIG_SYSCTL */ |
959 | ||
930d8f8d LC |
960 | static void __init lockup_detector_delay_init(struct work_struct *work); |
961 | static bool allow_lockup_detector_init_retry __initdata; | |
962 | ||
963 | static struct work_struct detector_work __initdata = | |
964 | __WORK_INITIALIZER(detector_work, lockup_detector_delay_init); | |
965 | ||
966 | static void __init lockup_detector_delay_init(struct work_struct *work) | |
967 | { | |
968 | int ret; | |
969 | ||
970 | ret = watchdog_hardlockup_probe(); | |
971 | if (ret) { | |
972 | pr_info("Delayed init of the lockup detector failed: %d\n", ret); | |
973 | pr_info("Hard watchdog permanently disabled\n"); | |
974 | return; | |
975 | } | |
976 | ||
977 | allow_lockup_detector_init_retry = false; | |
978 | ||
979 | watchdog_hardlockup_available = true; | |
980 | lockup_detector_setup(); | |
981 | } | |
982 | ||
983 | /* | |
984 | * lockup_detector_retry_init - retry init lockup detector if possible. | |
985 | * | |
986 | * Retry hardlockup detector init. It is useful when it requires some | |
987 | * functionality that has to be initialized later on a particular | |
988 | * platform. | |
989 | */ | |
990 | void __init lockup_detector_retry_init(void) | |
991 | { | |
992 | /* Must be called before late init calls */ | |
993 | if (!allow_lockup_detector_init_retry) | |
994 | return; | |
995 | ||
996 | schedule_work(&detector_work); | |
997 | } | |
998 | ||
999 | /* | |
1000 | * Ensure that optional delayed hardlockup init is proceed before | |
1001 | * the init code and memory is freed. | |
1002 | */ | |
1003 | static int __init lockup_detector_check(void) | |
1004 | { | |
1005 | /* Prevent any later retry. */ | |
1006 | allow_lockup_detector_init_retry = false; | |
1007 | ||
1008 | /* Make sure no work is pending. */ | |
1009 | flush_work(&detector_work); | |
1010 | ||
9ec272c5 DA |
1011 | watchdog_sysctl_init(); |
1012 | ||
930d8f8d LC |
1013 | return 0; |
1014 | ||
1015 | } | |
1016 | late_initcall_sync(lockup_detector_check); | |
1017 | ||
004417a6 | 1018 | void __init lockup_detector_init(void) |
58687acb | 1019 | { |
13316b31 | 1020 | if (tick_nohz_full_enabled()) |
314b08ff | 1021 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
13316b31 | 1022 | |
de201559 | 1023 | cpumask_copy(&watchdog_cpumask, |
04d4e665 | 1024 | housekeeping_cpumask(HK_TYPE_TIMER)); |
fe4ba3c3 | 1025 | |
df95d308 DA |
1026 | if (!watchdog_hardlockup_probe()) |
1027 | watchdog_hardlockup_available = true; | |
930d8f8d LC |
1028 | else |
1029 | allow_lockup_detector_init_retry = true; | |
1030 | ||
5587185d | 1031 | lockup_detector_setup(); |
58687acb | 1032 | } |