]>
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 | ||
5f92a7b0 | 12 | #define pr_fmt(fmt) "watchdog: " fmt |
4501980a | 13 | |
58687acb DZ |
14 | #include <linux/mm.h> |
15 | #include <linux/cpu.h> | |
16 | #include <linux/nmi.h> | |
17 | #include <linux/init.h> | |
58687acb DZ |
18 | #include <linux/module.h> |
19 | #include <linux/sysctl.h> | |
bcd951cf | 20 | #include <linux/smpboot.h> |
8bd75c77 | 21 | #include <linux/sched/rt.h> |
ae7e81c0 | 22 | #include <uapi/linux/sched/types.h> |
fe4ba3c3 | 23 | #include <linux/tick.h> |
82607adc | 24 | #include <linux/workqueue.h> |
e6017571 | 25 | #include <linux/sched/clock.h> |
b17b0153 | 26 | #include <linux/sched/debug.h> |
58687acb DZ |
27 | |
28 | #include <asm/irq_regs.h> | |
5d1c0f4a | 29 | #include <linux/kvm_para.h> |
81a4beef | 30 | #include <linux/kthread.h> |
58687acb | 31 | |
946d1977 | 32 | static DEFINE_MUTEX(watchdog_mutex); |
ab992dc3 | 33 | |
05a4a952 NP |
34 | int __read_mostly nmi_watchdog_enabled; |
35 | ||
36 | #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG) | |
37 | unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED | | |
38 | NMI_WATCHDOG_ENABLED; | |
84d56e66 | 39 | #else |
249e52e3 | 40 | unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED; |
84d56e66 | 41 | #endif |
05a4a952 NP |
42 | |
43 | #ifdef CONFIG_HARDLOCKUP_DETECTOR | |
44 | /* boot commands */ | |
45 | /* | |
46 | * Should we panic when a soft-lockup or hard-lockup occurs: | |
47 | */ | |
48 | unsigned int __read_mostly hardlockup_panic = | |
49 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; | |
50 | /* | |
51 | * We may not want to enable hard lockup detection by default in all cases, | |
52 | * for example when running the kernel as a guest on a hypervisor. In these | |
53 | * cases this function can be called to disable hard lockup detection. This | |
54 | * function should only be executed once by the boot processor before the | |
55 | * kernel command line parameters are parsed, because otherwise it is not | |
56 | * possible to override this in hardlockup_panic_setup(). | |
57 | */ | |
7a355820 | 58 | void __init hardlockup_detector_disable(void) |
05a4a952 NP |
59 | { |
60 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; | |
61 | } | |
62 | ||
63 | static int __init hardlockup_panic_setup(char *str) | |
64 | { | |
65 | if (!strncmp(str, "panic", 5)) | |
66 | hardlockup_panic = 1; | |
67 | else if (!strncmp(str, "nopanic", 7)) | |
68 | hardlockup_panic = 0; | |
69 | else if (!strncmp(str, "0", 1)) | |
70 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; | |
71 | else if (!strncmp(str, "1", 1)) | |
72 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; | |
73 | return 1; | |
74 | } | |
75 | __setup("nmi_watchdog=", hardlockup_panic_setup); | |
76 | ||
77 | #endif | |
78 | ||
79 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR | |
84d56e66 | 80 | int __read_mostly soft_watchdog_enabled; |
05a4a952 NP |
81 | #endif |
82 | ||
84d56e66 | 83 | int __read_mostly watchdog_user_enabled; |
4eec42f3 | 84 | int __read_mostly watchdog_thresh = 10; |
84d56e66 | 85 | |
ed235875 AT |
86 | #ifdef CONFIG_SMP |
87 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; | |
55537871 | 88 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; |
ed235875 | 89 | #endif |
05a4a952 | 90 | struct cpumask watchdog_cpumask __read_mostly; |
fe4ba3c3 CM |
91 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); |
92 | ||
ec6a9066 UO |
93 | /* |
94 | * The 'watchdog_running' variable is set to 1 when the watchdog threads | |
95 | * are registered/started and is set to 0 when the watchdog threads are | |
96 | * unregistered/stopped, so it is an indicator whether the threads exist. | |
97 | */ | |
3c00ea82 | 98 | static int __read_mostly watchdog_running; |
05a4a952 NP |
99 | |
100 | /* | |
101 | * These functions can be overridden if an architecture implements its | |
102 | * own hardlockup detector. | |
a10a842f NP |
103 | * |
104 | * watchdog_nmi_enable/disable can be implemented to start and stop when | |
105 | * softlockup watchdog threads start and stop. The arch must select the | |
106 | * SOFTLOCKUP_DETECTOR Kconfig. | |
05a4a952 NP |
107 | */ |
108 | int __weak watchdog_nmi_enable(unsigned int cpu) | |
109 | { | |
110 | return 0; | |
111 | } | |
941154bd | 112 | |
05a4a952 NP |
113 | void __weak watchdog_nmi_disable(unsigned int cpu) |
114 | { | |
941154bd | 115 | hardlockup_detector_perf_disable(); |
05a4a952 NP |
116 | } |
117 | ||
a10a842f NP |
118 | /* |
119 | * watchdog_nmi_reconfigure can be implemented to be notified after any | |
120 | * watchdog configuration change. The arch hardlockup watchdog should | |
121 | * respond to the following variables: | |
122 | * - nmi_watchdog_enabled | |
123 | * - watchdog_thresh | |
124 | * - watchdog_cpumask | |
125 | * - sysctl_hardlockup_all_cpu_backtrace | |
126 | * - hardlockup_panic | |
a10a842f NP |
127 | */ |
128 | void __weak watchdog_nmi_reconfigure(void) | |
129 | { | |
130 | } | |
131 | ||
132 | ||
05a4a952 NP |
133 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR |
134 | ||
135 | /* Helper for online, unparked cpus. */ | |
136 | #define for_each_watchdog_cpu(cpu) \ | |
137 | for_each_cpu_and((cpu), cpu_online_mask, &watchdog_cpumask) | |
138 | ||
139 | atomic_t watchdog_park_in_progress = ATOMIC_INIT(0); | |
ec6a9066 | 140 | |
0f34c400 | 141 | static u64 __read_mostly sample_period; |
58687acb DZ |
142 | |
143 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); | |
144 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); | |
145 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); | |
146 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); | |
58687acb | 147 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
bcd951cf TG |
148 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
149 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); | |
b1a8de1f | 150 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
58687acb | 151 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
ed235875 | 152 | static unsigned long soft_lockup_nmi_warn; |
58687acb | 153 | |
58687acb DZ |
154 | unsigned int __read_mostly softlockup_panic = |
155 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; | |
156 | ||
157 | static int __init softlockup_panic_setup(char *str) | |
158 | { | |
159 | softlockup_panic = simple_strtoul(str, NULL, 0); | |
160 | ||
161 | return 1; | |
162 | } | |
163 | __setup("softlockup_panic=", softlockup_panic_setup); | |
164 | ||
165 | static int __init nowatchdog_setup(char *str) | |
166 | { | |
195daf66 | 167 | watchdog_enabled = 0; |
58687acb DZ |
168 | return 1; |
169 | } | |
170 | __setup("nowatchdog", nowatchdog_setup); | |
171 | ||
58687acb DZ |
172 | static int __init nosoftlockup_setup(char *str) |
173 | { | |
195daf66 | 174 | watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED; |
58687acb DZ |
175 | return 1; |
176 | } | |
177 | __setup("nosoftlockup", nosoftlockup_setup); | |
195daf66 | 178 | |
ed235875 AT |
179 | #ifdef CONFIG_SMP |
180 | static int __init softlockup_all_cpu_backtrace_setup(char *str) | |
181 | { | |
182 | sysctl_softlockup_all_cpu_backtrace = | |
183 | !!simple_strtol(str, NULL, 0); | |
184 | return 1; | |
185 | } | |
186 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); | |
05a4a952 | 187 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
55537871 JK |
188 | static int __init hardlockup_all_cpu_backtrace_setup(char *str) |
189 | { | |
190 | sysctl_hardlockup_all_cpu_backtrace = | |
191 | !!simple_strtol(str, NULL, 0); | |
192 | return 1; | |
193 | } | |
194 | __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup); | |
ed235875 | 195 | #endif |
05a4a952 | 196 | #endif |
58687acb | 197 | |
941154bd TG |
198 | static void __lockup_detector_cleanup(void); |
199 | ||
4eec42f3 MSB |
200 | /* |
201 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- | |
202 | * lockups can have false positives under extreme conditions. So we generally | |
203 | * want a higher threshold for soft lockups than for hard lockups. So we couple | |
204 | * the thresholds with a factor: we make the soft threshold twice the amount of | |
205 | * time the hard threshold is. | |
206 | */ | |
6e9101ae | 207 | static int get_softlockup_thresh(void) |
4eec42f3 MSB |
208 | { |
209 | return watchdog_thresh * 2; | |
210 | } | |
58687acb DZ |
211 | |
212 | /* | |
213 | * Returns seconds, approximately. We don't need nanosecond | |
214 | * resolution, and we don't need to waste time with a big divide when | |
215 | * 2^30ns == 1.074s. | |
216 | */ | |
c06b4f19 | 217 | static unsigned long get_timestamp(void) |
58687acb | 218 | { |
545a2bf7 | 219 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
58687acb DZ |
220 | } |
221 | ||
0f34c400 | 222 | static void set_sample_period(void) |
58687acb DZ |
223 | { |
224 | /* | |
586692a5 | 225 | * convert watchdog_thresh from seconds to ns |
86f5e6a7 FLVC |
226 | * the divide by 5 is to give hrtimer several chances (two |
227 | * or three with the current relation between the soft | |
228 | * and hard thresholds) to increment before the | |
229 | * hardlockup detector generates a warning | |
58687acb | 230 | */ |
0f34c400 | 231 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
7edaeb68 | 232 | watchdog_update_hrtimer_threshold(sample_period); |
58687acb DZ |
233 | } |
234 | ||
235 | /* Commands for resetting the watchdog */ | |
236 | static void __touch_watchdog(void) | |
237 | { | |
c06b4f19 | 238 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
58687acb DZ |
239 | } |
240 | ||
03e0d461 TH |
241 | /** |
242 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls | |
243 | * | |
244 | * Call when the scheduler may have stalled for legitimate reasons | |
245 | * preventing the watchdog task from executing - e.g. the scheduler | |
246 | * entering idle state. This should only be used for scheduler events. | |
247 | * Use touch_softlockup_watchdog() for everything else. | |
248 | */ | |
249 | void touch_softlockup_watchdog_sched(void) | |
58687acb | 250 | { |
7861144b AM |
251 | /* |
252 | * Preemption can be enabled. It doesn't matter which CPU's timestamp | |
253 | * gets zeroed here, so use the raw_ operation. | |
254 | */ | |
255 | raw_cpu_write(watchdog_touch_ts, 0); | |
58687acb | 256 | } |
03e0d461 TH |
257 | |
258 | void touch_softlockup_watchdog(void) | |
259 | { | |
260 | touch_softlockup_watchdog_sched(); | |
82607adc | 261 | wq_watchdog_touch(raw_smp_processor_id()); |
03e0d461 | 262 | } |
0167c781 | 263 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
58687acb | 264 | |
332fbdbc | 265 | void touch_all_softlockup_watchdogs(void) |
58687acb DZ |
266 | { |
267 | int cpu; | |
268 | ||
269 | /* | |
270 | * this is done lockless | |
271 | * do we care if a 0 races with a timestamp? | |
272 | * all it means is the softlock check starts one cycle later | |
273 | */ | |
fe4ba3c3 | 274 | for_each_watchdog_cpu(cpu) |
58687acb | 275 | per_cpu(watchdog_touch_ts, cpu) = 0; |
82607adc | 276 | wq_watchdog_touch(-1); |
58687acb DZ |
277 | } |
278 | ||
58687acb DZ |
279 | void touch_softlockup_watchdog_sync(void) |
280 | { | |
f7f66b05 CL |
281 | __this_cpu_write(softlockup_touch_sync, true); |
282 | __this_cpu_write(watchdog_touch_ts, 0); | |
58687acb DZ |
283 | } |
284 | ||
26e09c6e | 285 | static int is_softlockup(unsigned long touch_ts) |
58687acb | 286 | { |
c06b4f19 | 287 | unsigned long now = get_timestamp(); |
58687acb | 288 | |
39d2da21 | 289 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
195daf66 UO |
290 | /* Warn about unreasonable delays. */ |
291 | if (time_after(now, touch_ts + get_softlockup_thresh())) | |
292 | return now - touch_ts; | |
293 | } | |
58687acb DZ |
294 | return 0; |
295 | } | |
296 | ||
05a4a952 NP |
297 | /* watchdog detector functions */ |
298 | bool is_hardlockup(void) | |
58687acb | 299 | { |
05a4a952 | 300 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
bcd951cf | 301 | |
05a4a952 NP |
302 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
303 | return true; | |
304 | ||
305 | __this_cpu_write(hrtimer_interrupts_saved, hrint); | |
306 | return false; | |
73ce0511 | 307 | } |
05a4a952 NP |
308 | |
309 | static void watchdog_interrupt_count(void) | |
73ce0511 | 310 | { |
05a4a952 | 311 | __this_cpu_inc(hrtimer_interrupts); |
73ce0511 | 312 | } |
58687acb | 313 | |
58cf690a UO |
314 | static int watchdog_enable_all_cpus(void); |
315 | static void watchdog_disable_all_cpus(void); | |
316 | ||
58687acb DZ |
317 | /* watchdog kicker functions */ |
318 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
319 | { | |
909ea964 | 320 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
58687acb DZ |
321 | struct pt_regs *regs = get_irq_regs(); |
322 | int duration; | |
ed235875 | 323 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
58687acb | 324 | |
6554fd8c TG |
325 | if (!watchdog_enabled || |
326 | atomic_read(&watchdog_park_in_progress) != 0) | |
b94f5118 DZ |
327 | return HRTIMER_NORESTART; |
328 | ||
58687acb DZ |
329 | /* kick the hardlockup detector */ |
330 | watchdog_interrupt_count(); | |
331 | ||
332 | /* kick the softlockup detector */ | |
909ea964 | 333 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
58687acb DZ |
334 | |
335 | /* .. and repeat */ | |
0f34c400 | 336 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
58687acb DZ |
337 | |
338 | if (touch_ts == 0) { | |
909ea964 | 339 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
340 | /* |
341 | * If the time stamp was touched atomically | |
342 | * make sure the scheduler tick is up to date. | |
343 | */ | |
909ea964 | 344 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
345 | sched_clock_tick(); |
346 | } | |
5d1c0f4a EM |
347 | |
348 | /* Clear the guest paused flag on watchdog reset */ | |
349 | kvm_check_and_clear_guest_paused(); | |
58687acb DZ |
350 | __touch_watchdog(); |
351 | return HRTIMER_RESTART; | |
352 | } | |
353 | ||
354 | /* check for a softlockup | |
355 | * This is done by making sure a high priority task is | |
356 | * being scheduled. The task touches the watchdog to | |
357 | * indicate it is getting cpu time. If it hasn't then | |
358 | * this is a good indication some task is hogging the cpu | |
359 | */ | |
26e09c6e | 360 | duration = is_softlockup(touch_ts); |
58687acb | 361 | if (unlikely(duration)) { |
5d1c0f4a EM |
362 | /* |
363 | * If a virtual machine is stopped by the host it can look to | |
364 | * the watchdog like a soft lockup, check to see if the host | |
365 | * stopped the vm before we issue the warning | |
366 | */ | |
367 | if (kvm_check_and_clear_guest_paused()) | |
368 | return HRTIMER_RESTART; | |
369 | ||
58687acb | 370 | /* only warn once */ |
b1a8de1f | 371 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
372 | /* | |
373 | * When multiple processes are causing softlockups the | |
374 | * softlockup detector only warns on the first one | |
375 | * because the code relies on a full quiet cycle to | |
376 | * re-arm. The second process prevents the quiet cycle | |
377 | * and never gets reported. Use task pointers to detect | |
378 | * this. | |
379 | */ | |
380 | if (__this_cpu_read(softlockup_task_ptr_saved) != | |
381 | current) { | |
382 | __this_cpu_write(soft_watchdog_warn, false); | |
383 | __touch_watchdog(); | |
384 | } | |
58687acb | 385 | return HRTIMER_RESTART; |
b1a8de1f | 386 | } |
58687acb | 387 | |
ed235875 AT |
388 | if (softlockup_all_cpu_backtrace) { |
389 | /* Prevent multiple soft-lockup reports if one cpu is already | |
390 | * engaged in dumping cpu back traces | |
391 | */ | |
392 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { | |
393 | /* Someone else will report us. Let's give up */ | |
394 | __this_cpu_write(soft_watchdog_warn, true); | |
395 | return HRTIMER_RESTART; | |
396 | } | |
397 | } | |
398 | ||
656c3b79 | 399 | pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 400 | smp_processor_id(), duration, |
58687acb | 401 | current->comm, task_pid_nr(current)); |
b1a8de1f | 402 | __this_cpu_write(softlockup_task_ptr_saved, current); |
58687acb DZ |
403 | print_modules(); |
404 | print_irqtrace_events(current); | |
405 | if (regs) | |
406 | show_regs(regs); | |
407 | else | |
408 | dump_stack(); | |
409 | ||
ed235875 AT |
410 | if (softlockup_all_cpu_backtrace) { |
411 | /* Avoid generating two back traces for current | |
412 | * given that one is already made above | |
413 | */ | |
414 | trigger_allbutself_cpu_backtrace(); | |
415 | ||
416 | clear_bit(0, &soft_lockup_nmi_warn); | |
417 | /* Barrier to sync with other cpus */ | |
418 | smp_mb__after_atomic(); | |
419 | } | |
420 | ||
69361eef | 421 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
58687acb DZ |
422 | if (softlockup_panic) |
423 | panic("softlockup: hung tasks"); | |
909ea964 | 424 | __this_cpu_write(soft_watchdog_warn, true); |
58687acb | 425 | } else |
909ea964 | 426 | __this_cpu_write(soft_watchdog_warn, false); |
58687acb DZ |
427 | |
428 | return HRTIMER_RESTART; | |
429 | } | |
430 | ||
bcd951cf TG |
431 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
432 | { | |
433 | struct sched_param param = { .sched_priority = prio }; | |
58687acb | 434 | |
bcd951cf TG |
435 | sched_setscheduler(current, policy, ¶m); |
436 | } | |
437 | ||
438 | static void watchdog_enable(unsigned int cpu) | |
58687acb | 439 | { |
f7f66b05 | 440 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
58687acb | 441 | |
3935e895 BM |
442 | /* kick off the timer for the hardlockup detector */ |
443 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
444 | hrtimer->function = watchdog_timer_fn; | |
445 | ||
bcd951cf TG |
446 | /* Enable the perf event */ |
447 | watchdog_nmi_enable(cpu); | |
58687acb | 448 | |
58687acb | 449 | /* done here because hrtimer_start can only pin to smp_processor_id() */ |
0f34c400 | 450 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
58687acb DZ |
451 | HRTIMER_MODE_REL_PINNED); |
452 | ||
bcd951cf TG |
453 | /* initialize timestamp */ |
454 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); | |
455 | __touch_watchdog(); | |
456 | } | |
58687acb | 457 | |
bcd951cf TG |
458 | static void watchdog_disable(unsigned int cpu) |
459 | { | |
f7f66b05 | 460 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
58687acb | 461 | |
bcd951cf TG |
462 | watchdog_set_prio(SCHED_NORMAL, 0); |
463 | hrtimer_cancel(hrtimer); | |
464 | /* disable the perf event */ | |
465 | watchdog_nmi_disable(cpu); | |
58687acb DZ |
466 | } |
467 | ||
b8900bc0 FW |
468 | static void watchdog_cleanup(unsigned int cpu, bool online) |
469 | { | |
470 | watchdog_disable(cpu); | |
471 | } | |
472 | ||
bcd951cf TG |
473 | static int watchdog_should_run(unsigned int cpu) |
474 | { | |
475 | return __this_cpu_read(hrtimer_interrupts) != | |
476 | __this_cpu_read(soft_lockup_hrtimer_cnt); | |
477 | } | |
478 | ||
479 | /* | |
480 | * The watchdog thread function - touches the timestamp. | |
481 | * | |
0f34c400 | 482 | * It only runs once every sample_period seconds (4 seconds by |
bcd951cf TG |
483 | * default) to reset the softlockup timestamp. If this gets delayed |
484 | * for more than 2*watchdog_thresh seconds then the debug-printout | |
485 | * triggers in watchdog_timer_fn(). | |
486 | */ | |
487 | static void watchdog(unsigned int cpu) | |
488 | { | |
489 | __this_cpu_write(soft_lockup_hrtimer_cnt, | |
490 | __this_cpu_read(hrtimer_interrupts)); | |
491 | __touch_watchdog(); | |
492 | } | |
58687acb | 493 | |
b8900bc0 FW |
494 | static struct smp_hotplug_thread watchdog_threads = { |
495 | .store = &softlockup_watchdog, | |
496 | .thread_should_run = watchdog_should_run, | |
497 | .thread_fn = watchdog, | |
498 | .thread_comm = "watchdog/%u", | |
499 | .setup = watchdog_enable, | |
500 | .cleanup = watchdog_cleanup, | |
501 | .park = watchdog_disable, | |
502 | .unpark = watchdog_enable, | |
503 | }; | |
504 | ||
81a4beef UO |
505 | /* |
506 | * park all watchdog threads that are specified in 'watchdog_cpumask' | |
ee7fed54 UO |
507 | * |
508 | * This function returns an error if kthread_park() of a watchdog thread | |
509 | * fails. In this situation, the watchdog threads of some CPUs can already | |
510 | * be parked and the watchdog threads of other CPUs can still be runnable. | |
511 | * Callers are expected to handle this special condition as appropriate in | |
512 | * their context. | |
a2a45b85 UO |
513 | * |
514 | * This function may only be called in a context that is protected against | |
515 | * races with CPU hotplug - for example, via get_online_cpus(). | |
81a4beef UO |
516 | */ |
517 | static int watchdog_park_threads(void) | |
518 | { | |
519 | int cpu, ret = 0; | |
520 | ||
b94f5118 DZ |
521 | atomic_set(&watchdog_park_in_progress, 1); |
522 | ||
81a4beef UO |
523 | for_each_watchdog_cpu(cpu) { |
524 | ret = kthread_park(per_cpu(softlockup_watchdog, cpu)); | |
525 | if (ret) | |
526 | break; | |
527 | } | |
81a4beef | 528 | |
b94f5118 DZ |
529 | atomic_set(&watchdog_park_in_progress, 0); |
530 | ||
81a4beef UO |
531 | return ret; |
532 | } | |
533 | ||
534 | /* | |
535 | * unpark all watchdog threads that are specified in 'watchdog_cpumask' | |
a2a45b85 UO |
536 | * |
537 | * This function may only be called in a context that is protected against | |
538 | * races with CPU hotplug - for example, via get_online_cpus(). | |
81a4beef UO |
539 | */ |
540 | static void watchdog_unpark_threads(void) | |
541 | { | |
542 | int cpu; | |
543 | ||
81a4beef UO |
544 | for_each_watchdog_cpu(cpu) |
545 | kthread_unpark(per_cpu(softlockup_watchdog, cpu)); | |
81a4beef UO |
546 | } |
547 | ||
b43cb43c | 548 | static int update_watchdog_all_cpus(void) |
9809b18f | 549 | { |
b43cb43c UO |
550 | int ret; |
551 | ||
552 | ret = watchdog_park_threads(); | |
553 | if (ret) | |
554 | return ret; | |
555 | ||
d4bdd0b2 | 556 | watchdog_unpark_threads(); |
b43cb43c UO |
557 | |
558 | return 0; | |
9809b18f MH |
559 | } |
560 | ||
b2f57c3a | 561 | static int watchdog_enable_all_cpus(void) |
58687acb | 562 | { |
b8900bc0 | 563 | int err = 0; |
58687acb | 564 | |
3c00ea82 | 565 | if (!watchdog_running) { |
230ec939 FW |
566 | err = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
567 | &watchdog_cpumask); | |
b8900bc0 FW |
568 | if (err) |
569 | pr_err("Failed to create watchdog threads, disabled\n"); | |
230ec939 | 570 | else |
3c00ea82 | 571 | watchdog_running = 1; |
b2f57c3a UO |
572 | } else { |
573 | /* | |
574 | * Enable/disable the lockup detectors or | |
575 | * change the sample period 'on the fly'. | |
576 | */ | |
b43cb43c UO |
577 | err = update_watchdog_all_cpus(); |
578 | ||
579 | if (err) { | |
580 | watchdog_disable_all_cpus(); | |
581 | pr_err("Failed to update lockup detectors, disabled\n"); | |
582 | } | |
bcd951cf | 583 | } |
b8900bc0 | 584 | |
b43cb43c UO |
585 | if (err) |
586 | watchdog_enabled = 0; | |
587 | ||
b8900bc0 | 588 | return err; |
58687acb DZ |
589 | } |
590 | ||
591 | static void watchdog_disable_all_cpus(void) | |
592 | { | |
3c00ea82 FW |
593 | if (watchdog_running) { |
594 | watchdog_running = 0; | |
b8900bc0 | 595 | smpboot_unregister_percpu_thread(&watchdog_threads); |
bcd951cf | 596 | } |
58687acb DZ |
597 | } |
598 | ||
a10a842f NP |
599 | #ifdef CONFIG_SYSCTL |
600 | static int watchdog_update_cpus(void) | |
601 | { | |
602 | return smpboot_update_cpumask_percpu_thread( | |
603 | &watchdog_threads, &watchdog_cpumask); | |
604 | } | |
605 | #endif | |
606 | ||
05a4a952 NP |
607 | #else /* SOFTLOCKUP */ |
608 | static int watchdog_park_threads(void) | |
609 | { | |
610 | return 0; | |
611 | } | |
612 | ||
613 | static void watchdog_unpark_threads(void) | |
614 | { | |
615 | } | |
616 | ||
617 | static int watchdog_enable_all_cpus(void) | |
618 | { | |
619 | return 0; | |
620 | } | |
621 | ||
622 | static void watchdog_disable_all_cpus(void) | |
623 | { | |
624 | } | |
625 | ||
a10a842f NP |
626 | #ifdef CONFIG_SYSCTL |
627 | static int watchdog_update_cpus(void) | |
628 | { | |
629 | return 0; | |
630 | } | |
631 | #endif | |
632 | ||
05a4a952 NP |
633 | static void set_sample_period(void) |
634 | { | |
635 | } | |
636 | #endif /* SOFTLOCKUP */ | |
637 | ||
941154bd TG |
638 | static void __lockup_detector_cleanup(void) |
639 | { | |
640 | lockdep_assert_held(&watchdog_mutex); | |
641 | hardlockup_detector_perf_cleanup(); | |
642 | } | |
643 | ||
644 | /** | |
645 | * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes | |
646 | * | |
647 | * Caller must not hold the cpu hotplug rwsem. | |
648 | */ | |
649 | void lockup_detector_cleanup(void) | |
650 | { | |
651 | mutex_lock(&watchdog_mutex); | |
652 | __lockup_detector_cleanup(); | |
653 | mutex_unlock(&watchdog_mutex); | |
654 | } | |
655 | ||
6554fd8c TG |
656 | /** |
657 | * lockup_detector_soft_poweroff - Interface to stop lockup detector(s) | |
658 | * | |
659 | * Special interface for parisc. It prevents lockup detector warnings from | |
660 | * the default pm_poweroff() function which busy loops forever. | |
661 | */ | |
662 | void lockup_detector_soft_poweroff(void) | |
663 | { | |
664 | watchdog_enabled = 0; | |
665 | } | |
666 | ||
58cf690a UO |
667 | #ifdef CONFIG_SYSCTL |
668 | ||
58687acb | 669 | /* |
a0c9cbb9 UO |
670 | * Update the run state of the lockup detectors. |
671 | */ | |
672 | static int proc_watchdog_update(void) | |
673 | { | |
674 | int err = 0; | |
675 | ||
676 | /* | |
677 | * Watchdog threads won't be started if they are already active. | |
678 | * The 'watchdog_running' variable in watchdog_*_all_cpus() takes | |
679 | * care of this. If those threads are already active, the sample | |
680 | * period will be updated and the lockup detectors will be enabled | |
681 | * or disabled 'on the fly'. | |
682 | */ | |
683 | if (watchdog_enabled && watchdog_thresh) | |
b2f57c3a | 684 | err = watchdog_enable_all_cpus(); |
a0c9cbb9 UO |
685 | else |
686 | watchdog_disable_all_cpus(); | |
687 | ||
a10a842f NP |
688 | watchdog_nmi_reconfigure(); |
689 | ||
941154bd TG |
690 | __lockup_detector_cleanup(); |
691 | ||
a0c9cbb9 UO |
692 | return err; |
693 | ||
694 | } | |
695 | ||
ef246a21 UO |
696 | /* |
697 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter | |
698 | * | |
699 | * caller | table->data points to | 'which' contains the flag(s) | |
700 | * -------------------|-----------------------|----------------------------- | |
701 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed | |
702 | * | | with SOFT_WATCHDOG_ENABLED | |
703 | * -------------------|-----------------------|----------------------------- | |
704 | * proc_nmi_watchdog | nmi_watchdog_enabled | NMI_WATCHDOG_ENABLED | |
705 | * -------------------|-----------------------|----------------------------- | |
706 | * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED | |
707 | */ | |
708 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, | |
709 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
710 | { | |
711 | int err, old, new; | |
712 | int *watchdog_param = (int *)table->data; | |
713 | ||
b7a34981 | 714 | cpu_hotplug_disable(); |
946d1977 | 715 | mutex_lock(&watchdog_mutex); |
ef246a21 UO |
716 | |
717 | /* | |
718 | * If the parameter is being read return the state of the corresponding | |
719 | * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the | |
720 | * run state of the lockup detectors. | |
721 | */ | |
722 | if (!write) { | |
723 | *watchdog_param = (watchdog_enabled & which) != 0; | |
724 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
725 | } else { | |
726 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
727 | if (err) | |
728 | goto out; | |
729 | ||
730 | /* | |
731 | * There is a race window between fetching the current value | |
732 | * from 'watchdog_enabled' and storing the new value. During | |
733 | * this race window, watchdog_nmi_enable() can sneak in and | |
734 | * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'. | |
735 | * The 'cmpxchg' detects this race and the loop retries. | |
736 | */ | |
737 | do { | |
738 | old = watchdog_enabled; | |
739 | /* | |
740 | * If the parameter value is not zero set the | |
741 | * corresponding bit(s), else clear it(them). | |
742 | */ | |
743 | if (*watchdog_param) | |
744 | new = old | which; | |
745 | else | |
746 | new = old & ~which; | |
747 | } while (cmpxchg(&watchdog_enabled, old, new) != old); | |
748 | ||
749 | /* | |
b43cb43c UO |
750 | * Update the run state of the lockup detectors. There is _no_ |
751 | * need to check the value returned by proc_watchdog_update() | |
752 | * and to restore the previous value of 'watchdog_enabled' as | |
753 | * both lockup detectors are disabled if proc_watchdog_update() | |
754 | * returns an error. | |
ef246a21 | 755 | */ |
a1ee1932 JH |
756 | if (old == new) |
757 | goto out; | |
758 | ||
ef246a21 | 759 | err = proc_watchdog_update(); |
ef246a21 UO |
760 | } |
761 | out: | |
946d1977 | 762 | mutex_unlock(&watchdog_mutex); |
b7a34981 | 763 | cpu_hotplug_enable(); |
ef246a21 UO |
764 | return err; |
765 | } | |
766 | ||
83a80a39 UO |
767 | /* |
768 | * /proc/sys/kernel/watchdog | |
769 | */ | |
770 | int proc_watchdog(struct ctl_table *table, int write, | |
771 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
772 | { | |
773 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, | |
774 | table, write, buffer, lenp, ppos); | |
775 | } | |
776 | ||
777 | /* | |
778 | * /proc/sys/kernel/nmi_watchdog | |
58687acb | 779 | */ |
83a80a39 UO |
780 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
781 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
782 | { | |
783 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, | |
784 | table, write, buffer, lenp, ppos); | |
785 | } | |
786 | ||
787 | /* | |
788 | * /proc/sys/kernel/soft_watchdog | |
789 | */ | |
790 | int proc_soft_watchdog(struct ctl_table *table, int write, | |
791 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
792 | { | |
793 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, | |
794 | table, write, buffer, lenp, ppos); | |
795 | } | |
58687acb | 796 | |
83a80a39 UO |
797 | /* |
798 | * /proc/sys/kernel/watchdog_thresh | |
799 | */ | |
800 | int proc_watchdog_thresh(struct ctl_table *table, int write, | |
801 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
58687acb | 802 | { |
a1ee1932 | 803 | int err, old, new; |
58687acb | 804 | |
b7a34981 | 805 | cpu_hotplug_disable(); |
946d1977 | 806 | mutex_lock(&watchdog_mutex); |
bcd951cf | 807 | |
83a80a39 | 808 | old = ACCESS_ONCE(watchdog_thresh); |
b8900bc0 | 809 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
83a80a39 | 810 | |
b8900bc0 | 811 | if (err || !write) |
359e6fab | 812 | goto out; |
e04ab2bc | 813 | |
b66a2356 | 814 | /* |
d283c640 | 815 | * Update the sample period. Restore on failure. |
b66a2356 | 816 | */ |
a1ee1932 JH |
817 | new = ACCESS_ONCE(watchdog_thresh); |
818 | if (old == new) | |
819 | goto out; | |
820 | ||
83a80a39 UO |
821 | set_sample_period(); |
822 | err = proc_watchdog_update(); | |
d283c640 | 823 | if (err) { |
83a80a39 | 824 | watchdog_thresh = old; |
d283c640 UO |
825 | set_sample_period(); |
826 | } | |
359e6fab | 827 | out: |
946d1977 | 828 | mutex_unlock(&watchdog_mutex); |
b7a34981 | 829 | cpu_hotplug_enable(); |
b8900bc0 | 830 | return err; |
58687acb | 831 | } |
fe4ba3c3 CM |
832 | |
833 | /* | |
834 | * The cpumask is the mask of possible cpus that the watchdog can run | |
835 | * on, not the mask of cpus it is actually running on. This allows the | |
836 | * user to specify a mask that will include cpus that have not yet | |
837 | * been brought online, if desired. | |
838 | */ | |
839 | int proc_watchdog_cpumask(struct ctl_table *table, int write, | |
840 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
841 | { | |
842 | int err; | |
843 | ||
b7a34981 | 844 | cpu_hotplug_disable(); |
946d1977 | 845 | mutex_lock(&watchdog_mutex); |
8c073d27 | 846 | |
fe4ba3c3 CM |
847 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
848 | if (!err && write) { | |
849 | /* Remove impossible cpus to keep sysctl output cleaner. */ | |
850 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, | |
851 | cpu_possible_mask); | |
852 | ||
853 | if (watchdog_running) { | |
854 | /* | |
855 | * Failure would be due to being unable to allocate | |
856 | * a temporary cpumask, so we are likely not in a | |
857 | * position to do much else to make things better. | |
858 | */ | |
a10a842f | 859 | if (watchdog_update_cpus() != 0) |
fe4ba3c3 CM |
860 | pr_err("cpumask update failed\n"); |
861 | } | |
a10a842f NP |
862 | |
863 | watchdog_nmi_reconfigure(); | |
941154bd | 864 | __lockup_detector_cleanup(); |
fe4ba3c3 | 865 | } |
5490125d | 866 | |
946d1977 | 867 | mutex_unlock(&watchdog_mutex); |
b7a34981 | 868 | cpu_hotplug_enable(); |
fe4ba3c3 CM |
869 | return err; |
870 | } | |
871 | ||
58687acb DZ |
872 | #endif /* CONFIG_SYSCTL */ |
873 | ||
004417a6 | 874 | void __init lockup_detector_init(void) |
58687acb | 875 | { |
0f34c400 | 876 | set_sample_period(); |
b8900bc0 | 877 | |
fe4ba3c3 CM |
878 | #ifdef CONFIG_NO_HZ_FULL |
879 | if (tick_nohz_full_enabled()) { | |
314b08ff FW |
880 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
881 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); | |
fe4ba3c3 CM |
882 | } else |
883 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); | |
884 | #else | |
885 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); | |
886 | #endif | |
887 | ||
195daf66 | 888 | if (watchdog_enabled) |
b2f57c3a | 889 | watchdog_enable_all_cpus(); |
58687acb | 890 | } |