Commit | Line | Data |
---|---|---|
58687acb DZ |
1 | /* |
2 | * Detect hard and soft lockups on a system | |
3 | * | |
4 | * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. | |
5 | * | |
86f5e6a7 FLVC |
6 | * Note: Most of this code is borrowed heavily from the original softlockup |
7 | * detector, so thanks to Ingo for the initial implementation. | |
8 | * Some chunks also taken from the old x86-specific nmi watchdog code, thanks | |
58687acb DZ |
9 | * to those contributors as well. |
10 | */ | |
11 | ||
12 | #include <linux/mm.h> | |
13 | #include <linux/cpu.h> | |
14 | #include <linux/nmi.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/delay.h> | |
17 | #include <linux/freezer.h> | |
18 | #include <linux/kthread.h> | |
19 | #include <linux/lockdep.h> | |
20 | #include <linux/notifier.h> | |
21 | #include <linux/module.h> | |
22 | #include <linux/sysctl.h> | |
23 | ||
24 | #include <asm/irq_regs.h> | |
25 | #include <linux/perf_event.h> | |
26 | ||
4135038a | 27 | int watchdog_enabled = 1; |
4eec42f3 | 28 | int __read_mostly watchdog_thresh = 10; |
58687acb DZ |
29 | |
30 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); | |
31 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); | |
32 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); | |
33 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); | |
58687acb | 34 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
23637d47 | 35 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
cafcd80d DZ |
36 | static DEFINE_PER_CPU(bool, hard_watchdog_warn); |
37 | static DEFINE_PER_CPU(bool, watchdog_nmi_touch); | |
58687acb DZ |
38 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
39 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); | |
40 | static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); | |
41 | #endif | |
42 | ||
58687acb DZ |
43 | /* boot commands */ |
44 | /* | |
45 | * Should we panic when a soft-lockup or hard-lockup occurs: | |
46 | */ | |
23637d47 | 47 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
fef2c9bc DZ |
48 | static int hardlockup_panic = |
49 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; | |
58687acb DZ |
50 | |
51 | static int __init hardlockup_panic_setup(char *str) | |
52 | { | |
53 | if (!strncmp(str, "panic", 5)) | |
54 | hardlockup_panic = 1; | |
fef2c9bc DZ |
55 | else if (!strncmp(str, "nopanic", 7)) |
56 | hardlockup_panic = 0; | |
5dc30558 | 57 | else if (!strncmp(str, "0", 1)) |
4135038a | 58 | watchdog_enabled = 0; |
58687acb DZ |
59 | return 1; |
60 | } | |
61 | __setup("nmi_watchdog=", hardlockup_panic_setup); | |
62 | #endif | |
63 | ||
64 | unsigned int __read_mostly softlockup_panic = | |
65 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; | |
66 | ||
67 | static int __init softlockup_panic_setup(char *str) | |
68 | { | |
69 | softlockup_panic = simple_strtoul(str, NULL, 0); | |
70 | ||
71 | return 1; | |
72 | } | |
73 | __setup("softlockup_panic=", softlockup_panic_setup); | |
74 | ||
75 | static int __init nowatchdog_setup(char *str) | |
76 | { | |
4135038a | 77 | watchdog_enabled = 0; |
58687acb DZ |
78 | return 1; |
79 | } | |
80 | __setup("nowatchdog", nowatchdog_setup); | |
81 | ||
82 | /* deprecated */ | |
83 | static int __init nosoftlockup_setup(char *str) | |
84 | { | |
4135038a | 85 | watchdog_enabled = 0; |
58687acb DZ |
86 | return 1; |
87 | } | |
88 | __setup("nosoftlockup", nosoftlockup_setup); | |
89 | /* */ | |
90 | ||
4eec42f3 MSB |
91 | /* |
92 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- | |
93 | * lockups can have false positives under extreme conditions. So we generally | |
94 | * want a higher threshold for soft lockups than for hard lockups. So we couple | |
95 | * the thresholds with a factor: we make the soft threshold twice the amount of | |
96 | * time the hard threshold is. | |
97 | */ | |
6e9101ae | 98 | static int get_softlockup_thresh(void) |
4eec42f3 MSB |
99 | { |
100 | return watchdog_thresh * 2; | |
101 | } | |
58687acb DZ |
102 | |
103 | /* | |
104 | * Returns seconds, approximately. We don't need nanosecond | |
105 | * resolution, and we don't need to waste time with a big divide when | |
106 | * 2^30ns == 1.074s. | |
107 | */ | |
108 | static unsigned long get_timestamp(int this_cpu) | |
109 | { | |
110 | return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ | |
111 | } | |
112 | ||
113 | static unsigned long get_sample_period(void) | |
114 | { | |
115 | /* | |
586692a5 | 116 | * convert watchdog_thresh from seconds to ns |
86f5e6a7 FLVC |
117 | * the divide by 5 is to give hrtimer several chances (two |
118 | * or three with the current relation between the soft | |
119 | * and hard thresholds) to increment before the | |
120 | * hardlockup detector generates a warning | |
58687acb | 121 | */ |
4eec42f3 | 122 | return get_softlockup_thresh() * (NSEC_PER_SEC / 5); |
58687acb DZ |
123 | } |
124 | ||
125 | /* Commands for resetting the watchdog */ | |
126 | static void __touch_watchdog(void) | |
127 | { | |
26e09c6e | 128 | int this_cpu = smp_processor_id(); |
58687acb | 129 | |
909ea964 | 130 | __this_cpu_write(watchdog_touch_ts, get_timestamp(this_cpu)); |
58687acb DZ |
131 | } |
132 | ||
332fbdbc | 133 | void touch_softlockup_watchdog(void) |
58687acb | 134 | { |
909ea964 | 135 | __this_cpu_write(watchdog_touch_ts, 0); |
58687acb | 136 | } |
0167c781 | 137 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
58687acb | 138 | |
332fbdbc | 139 | void touch_all_softlockup_watchdogs(void) |
58687acb DZ |
140 | { |
141 | int cpu; | |
142 | ||
143 | /* | |
144 | * this is done lockless | |
145 | * do we care if a 0 races with a timestamp? | |
146 | * all it means is the softlock check starts one cycle later | |
147 | */ | |
148 | for_each_online_cpu(cpu) | |
149 | per_cpu(watchdog_touch_ts, cpu) = 0; | |
150 | } | |
151 | ||
cafcd80d | 152 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb DZ |
153 | void touch_nmi_watchdog(void) |
154 | { | |
68d3f1d8 DZ |
155 | if (watchdog_enabled) { |
156 | unsigned cpu; | |
157 | ||
158 | for_each_present_cpu(cpu) { | |
159 | if (per_cpu(watchdog_nmi_touch, cpu) != true) | |
160 | per_cpu(watchdog_nmi_touch, cpu) = true; | |
161 | } | |
162 | } | |
332fbdbc | 163 | touch_softlockup_watchdog(); |
58687acb DZ |
164 | } |
165 | EXPORT_SYMBOL(touch_nmi_watchdog); | |
166 | ||
cafcd80d DZ |
167 | #endif |
168 | ||
58687acb DZ |
169 | void touch_softlockup_watchdog_sync(void) |
170 | { | |
171 | __raw_get_cpu_var(softlockup_touch_sync) = true; | |
172 | __raw_get_cpu_var(watchdog_touch_ts) = 0; | |
173 | } | |
174 | ||
23637d47 | 175 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb | 176 | /* watchdog detector functions */ |
26e09c6e | 177 | static int is_hardlockup(void) |
58687acb | 178 | { |
909ea964 | 179 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
58687acb | 180 | |
909ea964 | 181 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
58687acb DZ |
182 | return 1; |
183 | ||
909ea964 | 184 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
58687acb DZ |
185 | return 0; |
186 | } | |
187 | #endif | |
188 | ||
26e09c6e | 189 | static int is_softlockup(unsigned long touch_ts) |
58687acb | 190 | { |
26e09c6e | 191 | unsigned long now = get_timestamp(smp_processor_id()); |
58687acb DZ |
192 | |
193 | /* Warn about unreasonable delays: */ | |
4eec42f3 | 194 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
58687acb DZ |
195 | return now - touch_ts; |
196 | ||
197 | return 0; | |
198 | } | |
199 | ||
23637d47 | 200 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
1880c4ae | 201 | |
58687acb DZ |
202 | static struct perf_event_attr wd_hw_attr = { |
203 | .type = PERF_TYPE_HARDWARE, | |
204 | .config = PERF_COUNT_HW_CPU_CYCLES, | |
205 | .size = sizeof(struct perf_event_attr), | |
206 | .pinned = 1, | |
207 | .disabled = 1, | |
208 | }; | |
209 | ||
210 | /* Callback function for perf event subsystem */ | |
a8b0ca17 | 211 | static void watchdog_overflow_callback(struct perf_event *event, |
58687acb DZ |
212 | struct perf_sample_data *data, |
213 | struct pt_regs *regs) | |
214 | { | |
c6db67cd PZ |
215 | /* Ensure the watchdog never gets throttled */ |
216 | event->hw.interrupts = 0; | |
217 | ||
909ea964 CL |
218 | if (__this_cpu_read(watchdog_nmi_touch) == true) { |
219 | __this_cpu_write(watchdog_nmi_touch, false); | |
58687acb DZ |
220 | return; |
221 | } | |
222 | ||
223 | /* check for a hardlockup | |
224 | * This is done by making sure our timer interrupt | |
225 | * is incrementing. The timer interrupt should have | |
226 | * fired multiple times before we overflow'd. If it hasn't | |
227 | * then this is a good indication the cpu is stuck | |
228 | */ | |
26e09c6e DZ |
229 | if (is_hardlockup()) { |
230 | int this_cpu = smp_processor_id(); | |
231 | ||
58687acb | 232 | /* only print hardlockups once */ |
909ea964 | 233 | if (__this_cpu_read(hard_watchdog_warn) == true) |
58687acb DZ |
234 | return; |
235 | ||
236 | if (hardlockup_panic) | |
237 | panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu); | |
238 | else | |
239 | WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu); | |
240 | ||
909ea964 | 241 | __this_cpu_write(hard_watchdog_warn, true); |
58687acb DZ |
242 | return; |
243 | } | |
244 | ||
909ea964 | 245 | __this_cpu_write(hard_watchdog_warn, false); |
58687acb DZ |
246 | return; |
247 | } | |
248 | static void watchdog_interrupt_count(void) | |
249 | { | |
909ea964 | 250 | __this_cpu_inc(hrtimer_interrupts); |
58687acb DZ |
251 | } |
252 | #else | |
253 | static inline void watchdog_interrupt_count(void) { return; } | |
23637d47 | 254 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
58687acb DZ |
255 | |
256 | /* watchdog kicker functions */ | |
257 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
258 | { | |
909ea964 | 259 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
58687acb DZ |
260 | struct pt_regs *regs = get_irq_regs(); |
261 | int duration; | |
262 | ||
263 | /* kick the hardlockup detector */ | |
264 | watchdog_interrupt_count(); | |
265 | ||
266 | /* kick the softlockup detector */ | |
909ea964 | 267 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
58687acb DZ |
268 | |
269 | /* .. and repeat */ | |
270 | hrtimer_forward_now(hrtimer, ns_to_ktime(get_sample_period())); | |
271 | ||
272 | if (touch_ts == 0) { | |
909ea964 | 273 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
274 | /* |
275 | * If the time stamp was touched atomically | |
276 | * make sure the scheduler tick is up to date. | |
277 | */ | |
909ea964 | 278 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
279 | sched_clock_tick(); |
280 | } | |
281 | __touch_watchdog(); | |
282 | return HRTIMER_RESTART; | |
283 | } | |
284 | ||
285 | /* check for a softlockup | |
286 | * This is done by making sure a high priority task is | |
287 | * being scheduled. The task touches the watchdog to | |
288 | * indicate it is getting cpu time. If it hasn't then | |
289 | * this is a good indication some task is hogging the cpu | |
290 | */ | |
26e09c6e | 291 | duration = is_softlockup(touch_ts); |
58687acb DZ |
292 | if (unlikely(duration)) { |
293 | /* only warn once */ | |
909ea964 | 294 | if (__this_cpu_read(soft_watchdog_warn) == true) |
58687acb DZ |
295 | return HRTIMER_RESTART; |
296 | ||
b0f4c4b3 | 297 | printk(KERN_EMERG "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 298 | smp_processor_id(), duration, |
58687acb DZ |
299 | current->comm, task_pid_nr(current)); |
300 | print_modules(); | |
301 | print_irqtrace_events(current); | |
302 | if (regs) | |
303 | show_regs(regs); | |
304 | else | |
305 | dump_stack(); | |
306 | ||
307 | if (softlockup_panic) | |
308 | panic("softlockup: hung tasks"); | |
909ea964 | 309 | __this_cpu_write(soft_watchdog_warn, true); |
58687acb | 310 | } else |
909ea964 | 311 | __this_cpu_write(soft_watchdog_warn, false); |
58687acb DZ |
312 | |
313 | return HRTIMER_RESTART; | |
314 | } | |
315 | ||
316 | ||
317 | /* | |
318 | * The watchdog thread - touches the timestamp. | |
319 | */ | |
26e09c6e | 320 | static int watchdog(void *unused) |
58687acb | 321 | { |
cba9bd22 | 322 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
26e09c6e | 323 | struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer); |
58687acb DZ |
324 | |
325 | sched_setscheduler(current, SCHED_FIFO, ¶m); | |
326 | ||
327 | /* initialize timestamp */ | |
328 | __touch_watchdog(); | |
329 | ||
330 | /* kick off the timer for the hardlockup detector */ | |
331 | /* done here because hrtimer_start can only pin to smp_processor_id() */ | |
332 | hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()), | |
333 | HRTIMER_MODE_REL_PINNED); | |
334 | ||
335 | set_current_state(TASK_INTERRUPTIBLE); | |
336 | /* | |
86f5e6a7 FLVC |
337 | * Run briefly (kicked by the hrtimer callback function) once every |
338 | * get_sample_period() seconds (4 seconds by default) to reset the | |
339 | * softlockup timestamp. If this gets delayed for more than | |
340 | * 2*watchdog_thresh seconds then the debug-printout triggers in | |
341 | * watchdog_timer_fn(). | |
58687acb DZ |
342 | */ |
343 | while (!kthread_should_stop()) { | |
344 | __touch_watchdog(); | |
345 | schedule(); | |
346 | ||
347 | if (kthread_should_stop()) | |
348 | break; | |
349 | ||
350 | set_current_state(TASK_INTERRUPTIBLE); | |
351 | } | |
352 | __set_current_state(TASK_RUNNING); | |
cba9bd22 TG |
353 | param.sched_priority = 0; |
354 | sched_setscheduler(current, SCHED_NORMAL, ¶m); | |
58687acb DZ |
355 | return 0; |
356 | } | |
357 | ||
358 | ||
23637d47 | 359 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb DZ |
360 | static int watchdog_nmi_enable(int cpu) |
361 | { | |
362 | struct perf_event_attr *wd_attr; | |
363 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
364 | ||
365 | /* is it already setup and enabled? */ | |
366 | if (event && event->state > PERF_EVENT_STATE_OFF) | |
367 | goto out; | |
368 | ||
369 | /* it is setup but not enabled */ | |
370 | if (event != NULL) | |
371 | goto out_enable; | |
372 | ||
58687acb | 373 | wd_attr = &wd_hw_attr; |
4eec42f3 | 374 | wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh); |
1880c4ae CG |
375 | |
376 | /* Try to register using hardware perf events */ | |
4dc0da86 | 377 | event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL); |
58687acb DZ |
378 | if (!IS_ERR(event)) { |
379 | printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n"); | |
380 | goto out_save; | |
381 | } | |
382 | ||
5651f7f4 DZ |
383 | |
384 | /* vary the KERN level based on the returned errno */ | |
385 | if (PTR_ERR(event) == -EOPNOTSUPP) | |
386 | printk(KERN_INFO "NMI watchdog disabled (cpu%i): not supported (no LAPIC?)\n", cpu); | |
387 | else if (PTR_ERR(event) == -ENOENT) | |
388 | printk(KERN_WARNING "NMI watchdog disabled (cpu%i): hardware events not enabled\n", cpu); | |
389 | else | |
390 | printk(KERN_ERR "NMI watchdog disabled (cpu%i): unable to create perf event: %ld\n", cpu, PTR_ERR(event)); | |
eac24335 | 391 | return PTR_ERR(event); |
58687acb DZ |
392 | |
393 | /* success path */ | |
394 | out_save: | |
395 | per_cpu(watchdog_ev, cpu) = event; | |
396 | out_enable: | |
397 | perf_event_enable(per_cpu(watchdog_ev, cpu)); | |
398 | out: | |
399 | return 0; | |
400 | } | |
401 | ||
402 | static void watchdog_nmi_disable(int cpu) | |
403 | { | |
404 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
405 | ||
406 | if (event) { | |
407 | perf_event_disable(event); | |
408 | per_cpu(watchdog_ev, cpu) = NULL; | |
409 | ||
410 | /* should be in cleanup, but blocks oprofile */ | |
411 | perf_event_release_kernel(event); | |
412 | } | |
413 | return; | |
414 | } | |
415 | #else | |
416 | static int watchdog_nmi_enable(int cpu) { return 0; } | |
417 | static void watchdog_nmi_disable(int cpu) { return; } | |
23637d47 | 418 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
58687acb DZ |
419 | |
420 | /* prepare/enable/disable routines */ | |
6e9101ae | 421 | static void watchdog_prepare_cpu(int cpu) |
58687acb DZ |
422 | { |
423 | struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); | |
424 | ||
425 | WARN_ON(per_cpu(softlockup_watchdog, cpu)); | |
426 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
427 | hrtimer->function = watchdog_timer_fn; | |
58687acb DZ |
428 | } |
429 | ||
430 | static int watchdog_enable(int cpu) | |
431 | { | |
432 | struct task_struct *p = per_cpu(softlockup_watchdog, cpu); | |
f99a9933 | 433 | int err = 0; |
58687acb DZ |
434 | |
435 | /* enable the perf event */ | |
eac24335 | 436 | err = watchdog_nmi_enable(cpu); |
f99a9933 DZ |
437 | |
438 | /* Regardless of err above, fall through and start softlockup */ | |
58687acb DZ |
439 | |
440 | /* create the watchdog thread */ | |
441 | if (!p) { | |
18e5a45d | 442 | p = kthread_create_on_node(watchdog, NULL, cpu_to_node(cpu), "watchdog/%d", cpu); |
58687acb DZ |
443 | if (IS_ERR(p)) { |
444 | printk(KERN_ERR "softlockup watchdog for %i failed\n", cpu); | |
1409f141 | 445 | if (!err) { |
f99a9933 DZ |
446 | /* if hardlockup hasn't already set this */ |
447 | err = PTR_ERR(p); | |
1409f141 HD |
448 | /* and disable the perf event */ |
449 | watchdog_nmi_disable(cpu); | |
450 | } | |
f99a9933 | 451 | goto out; |
58687acb DZ |
452 | } |
453 | kthread_bind(p, cpu); | |
454 | per_cpu(watchdog_touch_ts, cpu) = 0; | |
455 | per_cpu(softlockup_watchdog, cpu) = p; | |
456 | wake_up_process(p); | |
457 | } | |
458 | ||
f99a9933 DZ |
459 | out: |
460 | return err; | |
58687acb DZ |
461 | } |
462 | ||
463 | static void watchdog_disable(int cpu) | |
464 | { | |
465 | struct task_struct *p = per_cpu(softlockup_watchdog, cpu); | |
466 | struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); | |
467 | ||
468 | /* | |
469 | * cancel the timer first to stop incrementing the stats | |
470 | * and waking up the kthread | |
471 | */ | |
472 | hrtimer_cancel(hrtimer); | |
473 | ||
474 | /* disable the perf event */ | |
475 | watchdog_nmi_disable(cpu); | |
476 | ||
477 | /* stop the watchdog thread */ | |
478 | if (p) { | |
479 | per_cpu(softlockup_watchdog, cpu) = NULL; | |
480 | kthread_stop(p); | |
481 | } | |
58687acb DZ |
482 | } |
483 | ||
4ff81951 VA |
484 | /* sysctl functions */ |
485 | #ifdef CONFIG_SYSCTL | |
58687acb DZ |
486 | static void watchdog_enable_all_cpus(void) |
487 | { | |
488 | int cpu; | |
39735766 MS |
489 | |
490 | watchdog_enabled = 0; | |
58687acb DZ |
491 | |
492 | for_each_online_cpu(cpu) | |
39735766 MS |
493 | if (!watchdog_enable(cpu)) |
494 | /* if any cpu succeeds, watchdog is considered | |
495 | enabled for the system */ | |
496 | watchdog_enabled = 1; | |
58687acb | 497 | |
39735766 | 498 | if (!watchdog_enabled) |
58687acb DZ |
499 | printk(KERN_ERR "watchdog: failed to be enabled on some cpus\n"); |
500 | ||
501 | } | |
502 | ||
503 | static void watchdog_disable_all_cpus(void) | |
504 | { | |
505 | int cpu; | |
506 | ||
507 | for_each_online_cpu(cpu) | |
508 | watchdog_disable(cpu); | |
509 | ||
510 | /* if all watchdogs are disabled, then they are disabled for the system */ | |
511 | watchdog_enabled = 0; | |
512 | } | |
513 | ||
514 | ||
58687acb | 515 | /* |
586692a5 | 516 | * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh |
58687acb DZ |
517 | */ |
518 | ||
586692a5 MSB |
519 | int proc_dowatchdog(struct ctl_table *table, int write, |
520 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
58687acb | 521 | { |
e04ab2bc | 522 | int ret; |
58687acb | 523 | |
586692a5 | 524 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
e04ab2bc MSB |
525 | if (ret || !write) |
526 | goto out; | |
527 | ||
586692a5 | 528 | if (watchdog_enabled && watchdog_thresh) |
e04ab2bc MSB |
529 | watchdog_enable_all_cpus(); |
530 | else | |
531 | watchdog_disable_all_cpus(); | |
532 | ||
533 | out: | |
534 | return ret; | |
58687acb | 535 | } |
58687acb DZ |
536 | #endif /* CONFIG_SYSCTL */ |
537 | ||
538 | ||
539 | /* | |
540 | * Create/destroy watchdog threads as CPUs come and go: | |
541 | */ | |
542 | static int __cpuinit | |
543 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
544 | { | |
545 | int hotcpu = (unsigned long)hcpu; | |
546 | ||
547 | switch (action) { | |
548 | case CPU_UP_PREPARE: | |
549 | case CPU_UP_PREPARE_FROZEN: | |
6e9101ae | 550 | watchdog_prepare_cpu(hotcpu); |
58687acb DZ |
551 | break; |
552 | case CPU_ONLINE: | |
553 | case CPU_ONLINE_FROZEN: | |
4135038a | 554 | if (watchdog_enabled) |
6e9101ae | 555 | watchdog_enable(hotcpu); |
58687acb DZ |
556 | break; |
557 | #ifdef CONFIG_HOTPLUG_CPU | |
558 | case CPU_UP_CANCELED: | |
559 | case CPU_UP_CANCELED_FROZEN: | |
560 | watchdog_disable(hotcpu); | |
561 | break; | |
562 | case CPU_DEAD: | |
563 | case CPU_DEAD_FROZEN: | |
564 | watchdog_disable(hotcpu); | |
565 | break; | |
566 | #endif /* CONFIG_HOTPLUG_CPU */ | |
567 | } | |
f99a9933 DZ |
568 | |
569 | /* | |
570 | * hardlockup and softlockup are not important enough | |
571 | * to block cpu bring up. Just always succeed and | |
572 | * rely on printk output to flag problems. | |
573 | */ | |
574 | return NOTIFY_OK; | |
58687acb DZ |
575 | } |
576 | ||
577 | static struct notifier_block __cpuinitdata cpu_nfb = { | |
578 | .notifier_call = cpu_callback | |
579 | }; | |
580 | ||
004417a6 | 581 | void __init lockup_detector_init(void) |
58687acb DZ |
582 | { |
583 | void *cpu = (void *)(long)smp_processor_id(); | |
584 | int err; | |
585 | ||
58687acb | 586 | err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
eac24335 | 587 | WARN_ON(notifier_to_errno(err)); |
58687acb DZ |
588 | |
589 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); | |
590 | register_cpu_notifier(&cpu_nfb); | |
591 | ||
004417a6 | 592 | return; |
58687acb | 593 | } |