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 | ||
4501980a AM |
12 | #define pr_fmt(fmt) "NMI watchdog: " fmt |
13 | ||
58687acb DZ |
14 | #include <linux/mm.h> |
15 | #include <linux/cpu.h> | |
16 | #include <linux/nmi.h> | |
17 | #include <linux/init.h> | |
18 | #include <linux/delay.h> | |
19 | #include <linux/freezer.h> | |
20 | #include <linux/kthread.h> | |
21 | #include <linux/lockdep.h> | |
22 | #include <linux/notifier.h> | |
23 | #include <linux/module.h> | |
24 | #include <linux/sysctl.h> | |
25 | ||
26 | #include <asm/irq_regs.h> | |
5d1c0f4a | 27 | #include <linux/kvm_para.h> |
58687acb DZ |
28 | #include <linux/perf_event.h> |
29 | ||
4135038a | 30 | int watchdog_enabled = 1; |
4eec42f3 | 31 | int __read_mostly watchdog_thresh = 10; |
58687acb DZ |
32 | |
33 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); | |
34 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); | |
35 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); | |
36 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); | |
58687acb | 37 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
23637d47 | 38 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
cafcd80d DZ |
39 | static DEFINE_PER_CPU(bool, hard_watchdog_warn); |
40 | static DEFINE_PER_CPU(bool, watchdog_nmi_touch); | |
58687acb DZ |
41 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
42 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); | |
43 | static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); | |
44 | #endif | |
45 | ||
58687acb DZ |
46 | /* boot commands */ |
47 | /* | |
48 | * Should we panic when a soft-lockup or hard-lockup occurs: | |
49 | */ | |
23637d47 | 50 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
fef2c9bc DZ |
51 | static int hardlockup_panic = |
52 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; | |
58687acb DZ |
53 | |
54 | static int __init hardlockup_panic_setup(char *str) | |
55 | { | |
56 | if (!strncmp(str, "panic", 5)) | |
57 | hardlockup_panic = 1; | |
fef2c9bc DZ |
58 | else if (!strncmp(str, "nopanic", 7)) |
59 | hardlockup_panic = 0; | |
5dc30558 | 60 | else if (!strncmp(str, "0", 1)) |
4135038a | 61 | watchdog_enabled = 0; |
58687acb DZ |
62 | return 1; |
63 | } | |
64 | __setup("nmi_watchdog=", hardlockup_panic_setup); | |
65 | #endif | |
66 | ||
67 | unsigned int __read_mostly softlockup_panic = | |
68 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; | |
69 | ||
70 | static int __init softlockup_panic_setup(char *str) | |
71 | { | |
72 | softlockup_panic = simple_strtoul(str, NULL, 0); | |
73 | ||
74 | return 1; | |
75 | } | |
76 | __setup("softlockup_panic=", softlockup_panic_setup); | |
77 | ||
78 | static int __init nowatchdog_setup(char *str) | |
79 | { | |
4135038a | 80 | watchdog_enabled = 0; |
58687acb DZ |
81 | return 1; |
82 | } | |
83 | __setup("nowatchdog", nowatchdog_setup); | |
84 | ||
85 | /* deprecated */ | |
86 | static int __init nosoftlockup_setup(char *str) | |
87 | { | |
4135038a | 88 | watchdog_enabled = 0; |
58687acb DZ |
89 | return 1; |
90 | } | |
91 | __setup("nosoftlockup", nosoftlockup_setup); | |
92 | /* */ | |
93 | ||
4eec42f3 MSB |
94 | /* |
95 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- | |
96 | * lockups can have false positives under extreme conditions. So we generally | |
97 | * want a higher threshold for soft lockups than for hard lockups. So we couple | |
98 | * the thresholds with a factor: we make the soft threshold twice the amount of | |
99 | * time the hard threshold is. | |
100 | */ | |
6e9101ae | 101 | static int get_softlockup_thresh(void) |
4eec42f3 MSB |
102 | { |
103 | return watchdog_thresh * 2; | |
104 | } | |
58687acb DZ |
105 | |
106 | /* | |
107 | * Returns seconds, approximately. We don't need nanosecond | |
108 | * resolution, and we don't need to waste time with a big divide when | |
109 | * 2^30ns == 1.074s. | |
110 | */ | |
111 | static unsigned long get_timestamp(int this_cpu) | |
112 | { | |
113 | return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ | |
114 | } | |
115 | ||
116 | static unsigned long get_sample_period(void) | |
117 | { | |
118 | /* | |
586692a5 | 119 | * convert watchdog_thresh from seconds to ns |
86f5e6a7 FLVC |
120 | * the divide by 5 is to give hrtimer several chances (two |
121 | * or three with the current relation between the soft | |
122 | * and hard thresholds) to increment before the | |
123 | * hardlockup detector generates a warning | |
58687acb | 124 | */ |
4eec42f3 | 125 | return get_softlockup_thresh() * (NSEC_PER_SEC / 5); |
58687acb DZ |
126 | } |
127 | ||
128 | /* Commands for resetting the watchdog */ | |
129 | static void __touch_watchdog(void) | |
130 | { | |
26e09c6e | 131 | int this_cpu = smp_processor_id(); |
58687acb | 132 | |
909ea964 | 133 | __this_cpu_write(watchdog_touch_ts, get_timestamp(this_cpu)); |
58687acb DZ |
134 | } |
135 | ||
332fbdbc | 136 | void touch_softlockup_watchdog(void) |
58687acb | 137 | { |
909ea964 | 138 | __this_cpu_write(watchdog_touch_ts, 0); |
58687acb | 139 | } |
0167c781 | 140 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
58687acb | 141 | |
332fbdbc | 142 | void touch_all_softlockup_watchdogs(void) |
58687acb DZ |
143 | { |
144 | int cpu; | |
145 | ||
146 | /* | |
147 | * this is done lockless | |
148 | * do we care if a 0 races with a timestamp? | |
149 | * all it means is the softlock check starts one cycle later | |
150 | */ | |
151 | for_each_online_cpu(cpu) | |
152 | per_cpu(watchdog_touch_ts, cpu) = 0; | |
153 | } | |
154 | ||
cafcd80d | 155 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb DZ |
156 | void touch_nmi_watchdog(void) |
157 | { | |
68d3f1d8 DZ |
158 | if (watchdog_enabled) { |
159 | unsigned cpu; | |
160 | ||
161 | for_each_present_cpu(cpu) { | |
162 | if (per_cpu(watchdog_nmi_touch, cpu) != true) | |
163 | per_cpu(watchdog_nmi_touch, cpu) = true; | |
164 | } | |
165 | } | |
332fbdbc | 166 | touch_softlockup_watchdog(); |
58687acb DZ |
167 | } |
168 | EXPORT_SYMBOL(touch_nmi_watchdog); | |
169 | ||
cafcd80d DZ |
170 | #endif |
171 | ||
58687acb DZ |
172 | void touch_softlockup_watchdog_sync(void) |
173 | { | |
174 | __raw_get_cpu_var(softlockup_touch_sync) = true; | |
175 | __raw_get_cpu_var(watchdog_touch_ts) = 0; | |
176 | } | |
177 | ||
23637d47 | 178 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb | 179 | /* watchdog detector functions */ |
26e09c6e | 180 | static int is_hardlockup(void) |
58687acb | 181 | { |
909ea964 | 182 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
58687acb | 183 | |
909ea964 | 184 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
58687acb DZ |
185 | return 1; |
186 | ||
909ea964 | 187 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
58687acb DZ |
188 | return 0; |
189 | } | |
190 | #endif | |
191 | ||
26e09c6e | 192 | static int is_softlockup(unsigned long touch_ts) |
58687acb | 193 | { |
26e09c6e | 194 | unsigned long now = get_timestamp(smp_processor_id()); |
58687acb DZ |
195 | |
196 | /* Warn about unreasonable delays: */ | |
4eec42f3 | 197 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
58687acb DZ |
198 | return now - touch_ts; |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
23637d47 | 203 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
1880c4ae | 204 | |
58687acb DZ |
205 | static struct perf_event_attr wd_hw_attr = { |
206 | .type = PERF_TYPE_HARDWARE, | |
207 | .config = PERF_COUNT_HW_CPU_CYCLES, | |
208 | .size = sizeof(struct perf_event_attr), | |
209 | .pinned = 1, | |
210 | .disabled = 1, | |
211 | }; | |
212 | ||
213 | /* Callback function for perf event subsystem */ | |
a8b0ca17 | 214 | static void watchdog_overflow_callback(struct perf_event *event, |
58687acb DZ |
215 | struct perf_sample_data *data, |
216 | struct pt_regs *regs) | |
217 | { | |
c6db67cd PZ |
218 | /* Ensure the watchdog never gets throttled */ |
219 | event->hw.interrupts = 0; | |
220 | ||
909ea964 CL |
221 | if (__this_cpu_read(watchdog_nmi_touch) == true) { |
222 | __this_cpu_write(watchdog_nmi_touch, false); | |
58687acb DZ |
223 | return; |
224 | } | |
225 | ||
226 | /* check for a hardlockup | |
227 | * This is done by making sure our timer interrupt | |
228 | * is incrementing. The timer interrupt should have | |
229 | * fired multiple times before we overflow'd. If it hasn't | |
230 | * then this is a good indication the cpu is stuck | |
231 | */ | |
26e09c6e DZ |
232 | if (is_hardlockup()) { |
233 | int this_cpu = smp_processor_id(); | |
234 | ||
58687acb | 235 | /* only print hardlockups once */ |
909ea964 | 236 | if (__this_cpu_read(hard_watchdog_warn) == true) |
58687acb DZ |
237 | return; |
238 | ||
239 | if (hardlockup_panic) | |
240 | panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu); | |
241 | else | |
242 | WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu); | |
243 | ||
909ea964 | 244 | __this_cpu_write(hard_watchdog_warn, true); |
58687acb DZ |
245 | return; |
246 | } | |
247 | ||
909ea964 | 248 | __this_cpu_write(hard_watchdog_warn, false); |
58687acb DZ |
249 | return; |
250 | } | |
251 | static void watchdog_interrupt_count(void) | |
252 | { | |
909ea964 | 253 | __this_cpu_inc(hrtimer_interrupts); |
58687acb DZ |
254 | } |
255 | #else | |
256 | static inline void watchdog_interrupt_count(void) { return; } | |
23637d47 | 257 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
58687acb DZ |
258 | |
259 | /* watchdog kicker functions */ | |
260 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
261 | { | |
909ea964 | 262 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
58687acb DZ |
263 | struct pt_regs *regs = get_irq_regs(); |
264 | int duration; | |
265 | ||
266 | /* kick the hardlockup detector */ | |
267 | watchdog_interrupt_count(); | |
268 | ||
269 | /* kick the softlockup detector */ | |
909ea964 | 270 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
58687acb DZ |
271 | |
272 | /* .. and repeat */ | |
273 | hrtimer_forward_now(hrtimer, ns_to_ktime(get_sample_period())); | |
274 | ||
275 | if (touch_ts == 0) { | |
909ea964 | 276 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
277 | /* |
278 | * If the time stamp was touched atomically | |
279 | * make sure the scheduler tick is up to date. | |
280 | */ | |
909ea964 | 281 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
282 | sched_clock_tick(); |
283 | } | |
5d1c0f4a EM |
284 | |
285 | /* Clear the guest paused flag on watchdog reset */ | |
286 | kvm_check_and_clear_guest_paused(); | |
58687acb DZ |
287 | __touch_watchdog(); |
288 | return HRTIMER_RESTART; | |
289 | } | |
290 | ||
291 | /* check for a softlockup | |
292 | * This is done by making sure a high priority task is | |
293 | * being scheduled. The task touches the watchdog to | |
294 | * indicate it is getting cpu time. If it hasn't then | |
295 | * this is a good indication some task is hogging the cpu | |
296 | */ | |
26e09c6e | 297 | duration = is_softlockup(touch_ts); |
58687acb | 298 | if (unlikely(duration)) { |
5d1c0f4a EM |
299 | /* |
300 | * If a virtual machine is stopped by the host it can look to | |
301 | * the watchdog like a soft lockup, check to see if the host | |
302 | * stopped the vm before we issue the warning | |
303 | */ | |
304 | if (kvm_check_and_clear_guest_paused()) | |
305 | return HRTIMER_RESTART; | |
306 | ||
58687acb | 307 | /* only warn once */ |
909ea964 | 308 | if (__this_cpu_read(soft_watchdog_warn) == true) |
58687acb DZ |
309 | return HRTIMER_RESTART; |
310 | ||
b0f4c4b3 | 311 | printk(KERN_EMERG "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 312 | smp_processor_id(), duration, |
58687acb DZ |
313 | current->comm, task_pid_nr(current)); |
314 | print_modules(); | |
315 | print_irqtrace_events(current); | |
316 | if (regs) | |
317 | show_regs(regs); | |
318 | else | |
319 | dump_stack(); | |
320 | ||
321 | if (softlockup_panic) | |
322 | panic("softlockup: hung tasks"); | |
909ea964 | 323 | __this_cpu_write(soft_watchdog_warn, true); |
58687acb | 324 | } else |
909ea964 | 325 | __this_cpu_write(soft_watchdog_warn, false); |
58687acb DZ |
326 | |
327 | return HRTIMER_RESTART; | |
328 | } | |
329 | ||
330 | ||
331 | /* | |
332 | * The watchdog thread - touches the timestamp. | |
333 | */ | |
26e09c6e | 334 | static int watchdog(void *unused) |
58687acb | 335 | { |
7a05c0f7 | 336 | struct sched_param param = { .sched_priority = 0 }; |
26e09c6e | 337 | struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer); |
58687acb | 338 | |
58687acb DZ |
339 | /* initialize timestamp */ |
340 | __touch_watchdog(); | |
341 | ||
342 | /* kick off the timer for the hardlockup detector */ | |
343 | /* done here because hrtimer_start can only pin to smp_processor_id() */ | |
344 | hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()), | |
345 | HRTIMER_MODE_REL_PINNED); | |
346 | ||
347 | set_current_state(TASK_INTERRUPTIBLE); | |
348 | /* | |
86f5e6a7 FLVC |
349 | * Run briefly (kicked by the hrtimer callback function) once every |
350 | * get_sample_period() seconds (4 seconds by default) to reset the | |
351 | * softlockup timestamp. If this gets delayed for more than | |
352 | * 2*watchdog_thresh seconds then the debug-printout triggers in | |
353 | * watchdog_timer_fn(). | |
58687acb DZ |
354 | */ |
355 | while (!kthread_should_stop()) { | |
356 | __touch_watchdog(); | |
357 | schedule(); | |
358 | ||
359 | if (kthread_should_stop()) | |
360 | break; | |
361 | ||
362 | set_current_state(TASK_INTERRUPTIBLE); | |
363 | } | |
b60f796c AM |
364 | /* |
365 | * Drop the policy/priority elevation during thread exit to avoid a | |
366 | * scheduling latency spike. | |
367 | */ | |
58687acb | 368 | __set_current_state(TASK_RUNNING); |
cba9bd22 | 369 | sched_setscheduler(current, SCHED_NORMAL, ¶m); |
58687acb DZ |
370 | return 0; |
371 | } | |
372 | ||
373 | ||
23637d47 | 374 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb DZ |
375 | static int watchdog_nmi_enable(int cpu) |
376 | { | |
377 | struct perf_event_attr *wd_attr; | |
378 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
379 | ||
380 | /* is it already setup and enabled? */ | |
381 | if (event && event->state > PERF_EVENT_STATE_OFF) | |
382 | goto out; | |
383 | ||
384 | /* it is setup but not enabled */ | |
385 | if (event != NULL) | |
386 | goto out_enable; | |
387 | ||
58687acb | 388 | wd_attr = &wd_hw_attr; |
4eec42f3 | 389 | wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh); |
1880c4ae CG |
390 | |
391 | /* Try to register using hardware perf events */ | |
4dc0da86 | 392 | event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL); |
58687acb | 393 | if (!IS_ERR(event)) { |
4501980a | 394 | pr_info("enabled, takes one hw-pmu counter.\n"); |
58687acb DZ |
395 | goto out_save; |
396 | } | |
397 | ||
5651f7f4 DZ |
398 | |
399 | /* vary the KERN level based on the returned errno */ | |
400 | if (PTR_ERR(event) == -EOPNOTSUPP) | |
4501980a | 401 | pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu); |
5651f7f4 | 402 | else if (PTR_ERR(event) == -ENOENT) |
4501980a AM |
403 | pr_warning("disabled (cpu%i): hardware events not enabled\n", |
404 | cpu); | |
5651f7f4 | 405 | else |
4501980a AM |
406 | pr_err("disabled (cpu%i): unable to create perf event: %ld\n", |
407 | cpu, PTR_ERR(event)); | |
eac24335 | 408 | return PTR_ERR(event); |
58687acb DZ |
409 | |
410 | /* success path */ | |
411 | out_save: | |
412 | per_cpu(watchdog_ev, cpu) = event; | |
413 | out_enable: | |
414 | perf_event_enable(per_cpu(watchdog_ev, cpu)); | |
415 | out: | |
416 | return 0; | |
417 | } | |
418 | ||
419 | static void watchdog_nmi_disable(int cpu) | |
420 | { | |
421 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
422 | ||
423 | if (event) { | |
424 | perf_event_disable(event); | |
425 | per_cpu(watchdog_ev, cpu) = NULL; | |
426 | ||
427 | /* should be in cleanup, but blocks oprofile */ | |
428 | perf_event_release_kernel(event); | |
429 | } | |
430 | return; | |
431 | } | |
432 | #else | |
433 | static int watchdog_nmi_enable(int cpu) { return 0; } | |
434 | static void watchdog_nmi_disable(int cpu) { return; } | |
23637d47 | 435 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
58687acb DZ |
436 | |
437 | /* prepare/enable/disable routines */ | |
6e9101ae | 438 | static void watchdog_prepare_cpu(int cpu) |
58687acb DZ |
439 | { |
440 | struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); | |
441 | ||
442 | WARN_ON(per_cpu(softlockup_watchdog, cpu)); | |
443 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
444 | hrtimer->function = watchdog_timer_fn; | |
58687acb DZ |
445 | } |
446 | ||
447 | static int watchdog_enable(int cpu) | |
448 | { | |
449 | struct task_struct *p = per_cpu(softlockup_watchdog, cpu); | |
f99a9933 | 450 | int err = 0; |
58687acb DZ |
451 | |
452 | /* enable the perf event */ | |
eac24335 | 453 | err = watchdog_nmi_enable(cpu); |
f99a9933 DZ |
454 | |
455 | /* Regardless of err above, fall through and start softlockup */ | |
58687acb DZ |
456 | |
457 | /* create the watchdog thread */ | |
458 | if (!p) { | |
7a05c0f7 | 459 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
18e5a45d | 460 | p = kthread_create_on_node(watchdog, NULL, cpu_to_node(cpu), "watchdog/%d", cpu); |
58687acb | 461 | if (IS_ERR(p)) { |
4501980a | 462 | pr_err("softlockup watchdog for %i failed\n", cpu); |
1409f141 | 463 | if (!err) { |
f99a9933 DZ |
464 | /* if hardlockup hasn't already set this */ |
465 | err = PTR_ERR(p); | |
1409f141 HD |
466 | /* and disable the perf event */ |
467 | watchdog_nmi_disable(cpu); | |
468 | } | |
f99a9933 | 469 | goto out; |
58687acb | 470 | } |
7a05c0f7 | 471 | sched_setscheduler(p, SCHED_FIFO, ¶m); |
58687acb DZ |
472 | kthread_bind(p, cpu); |
473 | per_cpu(watchdog_touch_ts, cpu) = 0; | |
474 | per_cpu(softlockup_watchdog, cpu) = p; | |
475 | wake_up_process(p); | |
476 | } | |
477 | ||
f99a9933 DZ |
478 | out: |
479 | return err; | |
58687acb DZ |
480 | } |
481 | ||
482 | static void watchdog_disable(int cpu) | |
483 | { | |
484 | struct task_struct *p = per_cpu(softlockup_watchdog, cpu); | |
485 | struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); | |
486 | ||
487 | /* | |
488 | * cancel the timer first to stop incrementing the stats | |
489 | * and waking up the kthread | |
490 | */ | |
491 | hrtimer_cancel(hrtimer); | |
492 | ||
493 | /* disable the perf event */ | |
494 | watchdog_nmi_disable(cpu); | |
495 | ||
496 | /* stop the watchdog thread */ | |
497 | if (p) { | |
498 | per_cpu(softlockup_watchdog, cpu) = NULL; | |
499 | kthread_stop(p); | |
500 | } | |
58687acb DZ |
501 | } |
502 | ||
4ff81951 VA |
503 | /* sysctl functions */ |
504 | #ifdef CONFIG_SYSCTL | |
58687acb DZ |
505 | static void watchdog_enable_all_cpus(void) |
506 | { | |
507 | int cpu; | |
39735766 MS |
508 | |
509 | watchdog_enabled = 0; | |
58687acb DZ |
510 | |
511 | for_each_online_cpu(cpu) | |
39735766 MS |
512 | if (!watchdog_enable(cpu)) |
513 | /* if any cpu succeeds, watchdog is considered | |
514 | enabled for the system */ | |
515 | watchdog_enabled = 1; | |
58687acb | 516 | |
39735766 | 517 | if (!watchdog_enabled) |
4501980a | 518 | pr_err("failed to be enabled on some cpus\n"); |
58687acb DZ |
519 | |
520 | } | |
521 | ||
522 | static void watchdog_disable_all_cpus(void) | |
523 | { | |
524 | int cpu; | |
525 | ||
526 | for_each_online_cpu(cpu) | |
527 | watchdog_disable(cpu); | |
528 | ||
529 | /* if all watchdogs are disabled, then they are disabled for the system */ | |
530 | watchdog_enabled = 0; | |
531 | } | |
532 | ||
533 | ||
58687acb | 534 | /* |
586692a5 | 535 | * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh |
58687acb DZ |
536 | */ |
537 | ||
586692a5 MSB |
538 | int proc_dowatchdog(struct ctl_table *table, int write, |
539 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
58687acb | 540 | { |
e04ab2bc | 541 | int ret; |
58687acb | 542 | |
586692a5 | 543 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
e04ab2bc MSB |
544 | if (ret || !write) |
545 | goto out; | |
546 | ||
586692a5 | 547 | if (watchdog_enabled && watchdog_thresh) |
e04ab2bc MSB |
548 | watchdog_enable_all_cpus(); |
549 | else | |
550 | watchdog_disable_all_cpus(); | |
551 | ||
552 | out: | |
553 | return ret; | |
58687acb | 554 | } |
58687acb DZ |
555 | #endif /* CONFIG_SYSCTL */ |
556 | ||
557 | ||
558 | /* | |
559 | * Create/destroy watchdog threads as CPUs come and go: | |
560 | */ | |
561 | static int __cpuinit | |
562 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
563 | { | |
564 | int hotcpu = (unsigned long)hcpu; | |
565 | ||
566 | switch (action) { | |
567 | case CPU_UP_PREPARE: | |
568 | case CPU_UP_PREPARE_FROZEN: | |
6e9101ae | 569 | watchdog_prepare_cpu(hotcpu); |
58687acb DZ |
570 | break; |
571 | case CPU_ONLINE: | |
572 | case CPU_ONLINE_FROZEN: | |
4135038a | 573 | if (watchdog_enabled) |
6e9101ae | 574 | watchdog_enable(hotcpu); |
58687acb DZ |
575 | break; |
576 | #ifdef CONFIG_HOTPLUG_CPU | |
577 | case CPU_UP_CANCELED: | |
578 | case CPU_UP_CANCELED_FROZEN: | |
579 | watchdog_disable(hotcpu); | |
580 | break; | |
581 | case CPU_DEAD: | |
582 | case CPU_DEAD_FROZEN: | |
583 | watchdog_disable(hotcpu); | |
584 | break; | |
585 | #endif /* CONFIG_HOTPLUG_CPU */ | |
586 | } | |
f99a9933 DZ |
587 | |
588 | /* | |
589 | * hardlockup and softlockup are not important enough | |
590 | * to block cpu bring up. Just always succeed and | |
591 | * rely on printk output to flag problems. | |
592 | */ | |
593 | return NOTIFY_OK; | |
58687acb DZ |
594 | } |
595 | ||
596 | static struct notifier_block __cpuinitdata cpu_nfb = { | |
597 | .notifier_call = cpu_callback | |
598 | }; | |
599 | ||
004417a6 | 600 | void __init lockup_detector_init(void) |
58687acb DZ |
601 | { |
602 | void *cpu = (void *)(long)smp_processor_id(); | |
603 | int err; | |
604 | ||
58687acb | 605 | err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
eac24335 | 606 | WARN_ON(notifier_to_errno(err)); |
58687acb DZ |
607 | |
608 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); | |
609 | register_cpu_notifier(&cpu_nfb); | |
610 | ||
004417a6 | 611 | return; |
58687acb | 612 | } |