]>
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) |
09154985 TG |
33 | # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED) |
34 | # define NMI_WATCHDOG_DEFAULT 1 | |
84d56e66 | 35 | #else |
09154985 TG |
36 | # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED) |
37 | # define NMI_WATCHDOG_DEFAULT 0 | |
84d56e66 | 38 | #endif |
05a4a952 | 39 | |
09154985 TG |
40 | unsigned long __read_mostly watchdog_enabled; |
41 | int __read_mostly watchdog_user_enabled = 1; | |
42 | int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT; | |
43 | int __read_mostly soft_watchdog_user_enabled = 1; | |
7feeb9cd | 44 | int __read_mostly watchdog_thresh = 10; |
48084abf | 45 | static int __read_mostly nmi_watchdog_available; |
7feeb9cd | 46 | |
7feeb9cd TG |
47 | struct cpumask watchdog_cpumask __read_mostly; |
48 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); | |
49 | ||
05a4a952 | 50 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
f117955a GP |
51 | |
52 | # ifdef CONFIG_SMP | |
53 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; | |
54 | # endif /* CONFIG_SMP */ | |
55 | ||
05a4a952 NP |
56 | /* |
57 | * Should we panic when a soft-lockup or hard-lockup occurs: | |
58 | */ | |
59 | unsigned int __read_mostly hardlockup_panic = | |
60 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; | |
61 | /* | |
62 | * We may not want to enable hard lockup detection by default in all cases, | |
63 | * for example when running the kernel as a guest on a hypervisor. In these | |
64 | * cases this function can be called to disable hard lockup detection. This | |
65 | * function should only be executed once by the boot processor before the | |
66 | * kernel command line parameters are parsed, because otherwise it is not | |
67 | * possible to override this in hardlockup_panic_setup(). | |
68 | */ | |
7a355820 | 69 | void __init hardlockup_detector_disable(void) |
05a4a952 | 70 | { |
09154985 | 71 | nmi_watchdog_user_enabled = 0; |
05a4a952 NP |
72 | } |
73 | ||
74 | static int __init hardlockup_panic_setup(char *str) | |
75 | { | |
76 | if (!strncmp(str, "panic", 5)) | |
77 | hardlockup_panic = 1; | |
78 | else if (!strncmp(str, "nopanic", 7)) | |
79 | hardlockup_panic = 0; | |
80 | else if (!strncmp(str, "0", 1)) | |
09154985 | 81 | nmi_watchdog_user_enabled = 0; |
05a4a952 | 82 | else if (!strncmp(str, "1", 1)) |
09154985 | 83 | nmi_watchdog_user_enabled = 1; |
05a4a952 NP |
84 | return 1; |
85 | } | |
86 | __setup("nmi_watchdog=", hardlockup_panic_setup); | |
87 | ||
368a7e2c | 88 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
05a4a952 | 89 | |
05a4a952 NP |
90 | /* |
91 | * These functions can be overridden if an architecture implements its | |
92 | * own hardlockup detector. | |
a10a842f NP |
93 | * |
94 | * watchdog_nmi_enable/disable can be implemented to start and stop when | |
95 | * softlockup watchdog threads start and stop. The arch must select the | |
96 | * SOFTLOCKUP_DETECTOR Kconfig. | |
05a4a952 NP |
97 | */ |
98 | int __weak watchdog_nmi_enable(unsigned int cpu) | |
99 | { | |
146c9d0e | 100 | hardlockup_detector_perf_enable(); |
05a4a952 NP |
101 | return 0; |
102 | } | |
941154bd | 103 | |
05a4a952 NP |
104 | void __weak watchdog_nmi_disable(unsigned int cpu) |
105 | { | |
941154bd | 106 | hardlockup_detector_perf_disable(); |
05a4a952 NP |
107 | } |
108 | ||
a994a314 TG |
109 | /* Return 0, if a NMI watchdog is available. Error code otherwise */ |
110 | int __weak __init watchdog_nmi_probe(void) | |
111 | { | |
112 | return hardlockup_detector_perf_init(); | |
113 | } | |
114 | ||
6592ad2f | 115 | /** |
6b9dc480 | 116 | * watchdog_nmi_stop - Stop the watchdog for reconfiguration |
6592ad2f | 117 | * |
6b9dc480 TG |
118 | * The reconfiguration steps are: |
119 | * watchdog_nmi_stop(); | |
6592ad2f | 120 | * update_variables(); |
6b9dc480 TG |
121 | * watchdog_nmi_start(); |
122 | */ | |
123 | void __weak watchdog_nmi_stop(void) { } | |
124 | ||
125 | /** | |
126 | * watchdog_nmi_start - Start the watchdog after reconfiguration | |
6592ad2f | 127 | * |
6b9dc480 TG |
128 | * Counterpart to watchdog_nmi_stop(). |
129 | * | |
130 | * The following variables have been updated in update_variables() and | |
131 | * contain the currently valid configuration: | |
7feeb9cd | 132 | * - watchdog_enabled |
a10a842f NP |
133 | * - watchdog_thresh |
134 | * - watchdog_cpumask | |
a10a842f | 135 | */ |
6b9dc480 | 136 | void __weak watchdog_nmi_start(void) { } |
a10a842f | 137 | |
09154985 TG |
138 | /** |
139 | * lockup_detector_update_enable - Update the sysctl enable bit | |
140 | * | |
141 | * Caller needs to make sure that the NMI/perf watchdogs are off, so this | |
142 | * can't race with watchdog_nmi_disable(). | |
143 | */ | |
144 | static void lockup_detector_update_enable(void) | |
145 | { | |
146 | watchdog_enabled = 0; | |
147 | if (!watchdog_user_enabled) | |
148 | return; | |
a994a314 | 149 | if (nmi_watchdog_available && nmi_watchdog_user_enabled) |
09154985 TG |
150 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; |
151 | if (soft_watchdog_user_enabled) | |
152 | watchdog_enabled |= SOFT_WATCHDOG_ENABLED; | |
153 | } | |
154 | ||
05a4a952 NP |
155 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR |
156 | ||
11e31f60 TG |
157 | #define SOFTLOCKUP_RESET ULONG_MAX |
158 | ||
f117955a GP |
159 | #ifdef CONFIG_SMP |
160 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; | |
161 | #endif | |
162 | ||
e7e04615 SS |
163 | static struct cpumask watchdog_allowed_mask __read_mostly; |
164 | ||
2b9d7f23 TG |
165 | /* Global variables, exported for sysctl */ |
166 | unsigned int __read_mostly softlockup_panic = | |
167 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; | |
2eb2527f | 168 | |
9cf57731 | 169 | static bool softlockup_initialized __read_mostly; |
0f34c400 | 170 | static u64 __read_mostly sample_period; |
58687acb DZ |
171 | |
172 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); | |
58687acb DZ |
173 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
174 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); | |
58687acb | 175 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
bcd951cf | 176 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
58687acb | 177 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
ed235875 | 178 | static unsigned long soft_lockup_nmi_warn; |
58687acb | 179 | |
58687acb DZ |
180 | static int __init nowatchdog_setup(char *str) |
181 | { | |
09154985 | 182 | watchdog_user_enabled = 0; |
58687acb DZ |
183 | return 1; |
184 | } | |
185 | __setup("nowatchdog", nowatchdog_setup); | |
186 | ||
58687acb DZ |
187 | static int __init nosoftlockup_setup(char *str) |
188 | { | |
09154985 | 189 | soft_watchdog_user_enabled = 0; |
58687acb DZ |
190 | return 1; |
191 | } | |
192 | __setup("nosoftlockup", nosoftlockup_setup); | |
195daf66 | 193 | |
11295055 LO |
194 | static int __init watchdog_thresh_setup(char *str) |
195 | { | |
196 | get_option(&str, &watchdog_thresh); | |
197 | return 1; | |
198 | } | |
199 | __setup("watchdog_thresh=", watchdog_thresh_setup); | |
200 | ||
941154bd TG |
201 | static void __lockup_detector_cleanup(void); |
202 | ||
4eec42f3 MSB |
203 | /* |
204 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- | |
205 | * lockups can have false positives under extreme conditions. So we generally | |
206 | * want a higher threshold for soft lockups than for hard lockups. So we couple | |
207 | * the thresholds with a factor: we make the soft threshold twice the amount of | |
208 | * time the hard threshold is. | |
209 | */ | |
6e9101ae | 210 | static int get_softlockup_thresh(void) |
4eec42f3 MSB |
211 | { |
212 | return watchdog_thresh * 2; | |
213 | } | |
58687acb DZ |
214 | |
215 | /* | |
216 | * Returns seconds, approximately. We don't need nanosecond | |
217 | * resolution, and we don't need to waste time with a big divide when | |
218 | * 2^30ns == 1.074s. | |
219 | */ | |
c06b4f19 | 220 | static unsigned long get_timestamp(void) |
58687acb | 221 | { |
545a2bf7 | 222 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
58687acb DZ |
223 | } |
224 | ||
0f34c400 | 225 | static void set_sample_period(void) |
58687acb DZ |
226 | { |
227 | /* | |
586692a5 | 228 | * convert watchdog_thresh from seconds to ns |
86f5e6a7 FLVC |
229 | * the divide by 5 is to give hrtimer several chances (two |
230 | * or three with the current relation between the soft | |
231 | * and hard thresholds) to increment before the | |
232 | * hardlockup detector generates a warning | |
58687acb | 233 | */ |
0f34c400 | 234 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
7edaeb68 | 235 | watchdog_update_hrtimer_threshold(sample_period); |
58687acb DZ |
236 | } |
237 | ||
238 | /* Commands for resetting the watchdog */ | |
239 | static void __touch_watchdog(void) | |
240 | { | |
c06b4f19 | 241 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
58687acb DZ |
242 | } |
243 | ||
03e0d461 TH |
244 | /** |
245 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls | |
246 | * | |
247 | * Call when the scheduler may have stalled for legitimate reasons | |
248 | * preventing the watchdog task from executing - e.g. the scheduler | |
249 | * entering idle state. This should only be used for scheduler events. | |
250 | * Use touch_softlockup_watchdog() for everything else. | |
251 | */ | |
cb9d7fd5 | 252 | notrace void touch_softlockup_watchdog_sched(void) |
58687acb | 253 | { |
7861144b AM |
254 | /* |
255 | * Preemption can be enabled. It doesn't matter which CPU's timestamp | |
256 | * gets zeroed here, so use the raw_ operation. | |
257 | */ | |
11e31f60 | 258 | raw_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET); |
58687acb | 259 | } |
03e0d461 | 260 | |
cb9d7fd5 | 261 | notrace void touch_softlockup_watchdog(void) |
03e0d461 TH |
262 | { |
263 | touch_softlockup_watchdog_sched(); | |
82607adc | 264 | wq_watchdog_touch(raw_smp_processor_id()); |
03e0d461 | 265 | } |
0167c781 | 266 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
58687acb | 267 | |
332fbdbc | 268 | void touch_all_softlockup_watchdogs(void) |
58687acb DZ |
269 | { |
270 | int cpu; | |
271 | ||
272 | /* | |
d57108d4 TG |
273 | * watchdog_mutex cannpt be taken here, as this might be called |
274 | * from (soft)interrupt context, so the access to | |
275 | * watchdog_allowed_cpumask might race with a concurrent update. | |
276 | * | |
277 | * The watchdog time stamp can race against a concurrent real | |
278 | * update as well, the only side effect might be a cycle delay for | |
279 | * the softlockup check. | |
58687acb | 280 | */ |
89e28ce6 | 281 | for_each_cpu(cpu, &watchdog_allowed_mask) { |
11e31f60 | 282 | per_cpu(watchdog_touch_ts, cpu) = SOFTLOCKUP_RESET; |
89e28ce6 WQ |
283 | wq_watchdog_touch(cpu); |
284 | } | |
58687acb DZ |
285 | } |
286 | ||
58687acb DZ |
287 | void touch_softlockup_watchdog_sync(void) |
288 | { | |
f7f66b05 | 289 | __this_cpu_write(softlockup_touch_sync, true); |
11e31f60 | 290 | __this_cpu_write(watchdog_touch_ts, SOFTLOCKUP_RESET); |
58687acb DZ |
291 | } |
292 | ||
26e09c6e | 293 | static int is_softlockup(unsigned long touch_ts) |
58687acb | 294 | { |
c06b4f19 | 295 | unsigned long now = get_timestamp(); |
58687acb | 296 | |
39d2da21 | 297 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
195daf66 UO |
298 | /* Warn about unreasonable delays. */ |
299 | if (time_after(now, touch_ts + get_softlockup_thresh())) | |
300 | return now - touch_ts; | |
301 | } | |
58687acb DZ |
302 | return 0; |
303 | } | |
304 | ||
05a4a952 NP |
305 | /* watchdog detector functions */ |
306 | bool is_hardlockup(void) | |
58687acb | 307 | { |
05a4a952 | 308 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
bcd951cf | 309 | |
05a4a952 NP |
310 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
311 | return true; | |
312 | ||
313 | __this_cpu_write(hrtimer_interrupts_saved, hrint); | |
314 | return false; | |
73ce0511 | 315 | } |
05a4a952 NP |
316 | |
317 | static void watchdog_interrupt_count(void) | |
73ce0511 | 318 | { |
05a4a952 | 319 | __this_cpu_inc(hrtimer_interrupts); |
73ce0511 | 320 | } |
58687acb | 321 | |
be45bf53 PZ |
322 | static DEFINE_PER_CPU(struct completion, softlockup_completion); |
323 | static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work); | |
324 | ||
9cf57731 PZ |
325 | /* |
326 | * The watchdog thread function - touches the timestamp. | |
327 | * | |
328 | * It only runs once every sample_period seconds (4 seconds by | |
329 | * default) to reset the softlockup timestamp. If this gets delayed | |
330 | * for more than 2*watchdog_thresh seconds then the debug-printout | |
331 | * triggers in watchdog_timer_fn(). | |
332 | */ | |
333 | static int softlockup_fn(void *data) | |
334 | { | |
9cf57731 | 335 | __touch_watchdog(); |
be45bf53 | 336 | complete(this_cpu_ptr(&softlockup_completion)); |
9cf57731 PZ |
337 | |
338 | return 0; | |
339 | } | |
340 | ||
58687acb DZ |
341 | /* watchdog kicker functions */ |
342 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
343 | { | |
909ea964 | 344 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
58687acb DZ |
345 | struct pt_regs *regs = get_irq_regs(); |
346 | int duration; | |
ed235875 | 347 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
58687acb | 348 | |
01f0a027 | 349 | if (!watchdog_enabled) |
b94f5118 DZ |
350 | return HRTIMER_NORESTART; |
351 | ||
58687acb DZ |
352 | /* kick the hardlockup detector */ |
353 | watchdog_interrupt_count(); | |
354 | ||
355 | /* kick the softlockup detector */ | |
be45bf53 PZ |
356 | if (completion_done(this_cpu_ptr(&softlockup_completion))) { |
357 | reinit_completion(this_cpu_ptr(&softlockup_completion)); | |
358 | stop_one_cpu_nowait(smp_processor_id(), | |
359 | softlockup_fn, NULL, | |
360 | this_cpu_ptr(&softlockup_stop_work)); | |
361 | } | |
58687acb DZ |
362 | |
363 | /* .. and repeat */ | |
0f34c400 | 364 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
58687acb | 365 | |
11e31f60 | 366 | if (touch_ts == SOFTLOCKUP_RESET) { |
909ea964 | 367 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
368 | /* |
369 | * If the time stamp was touched atomically | |
370 | * make sure the scheduler tick is up to date. | |
371 | */ | |
909ea964 | 372 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
373 | sched_clock_tick(); |
374 | } | |
5d1c0f4a EM |
375 | |
376 | /* Clear the guest paused flag on watchdog reset */ | |
377 | kvm_check_and_clear_guest_paused(); | |
58687acb DZ |
378 | __touch_watchdog(); |
379 | return HRTIMER_RESTART; | |
380 | } | |
381 | ||
382 | /* check for a softlockup | |
383 | * This is done by making sure a high priority task is | |
384 | * being scheduled. The task touches the watchdog to | |
385 | * indicate it is getting cpu time. If it hasn't then | |
386 | * this is a good indication some task is hogging the cpu | |
387 | */ | |
26e09c6e | 388 | duration = is_softlockup(touch_ts); |
58687acb | 389 | if (unlikely(duration)) { |
5d1c0f4a EM |
390 | /* |
391 | * If a virtual machine is stopped by the host it can look to | |
392 | * the watchdog like a soft lockup, check to see if the host | |
393 | * stopped the vm before we issue the warning | |
394 | */ | |
395 | if (kvm_check_and_clear_guest_paused()) | |
396 | return HRTIMER_RESTART; | |
397 | ||
58687acb | 398 | /* only warn once */ |
3a51449b | 399 | if (__this_cpu_read(soft_watchdog_warn) == true) |
58687acb DZ |
400 | return HRTIMER_RESTART; |
401 | ||
ed235875 AT |
402 | if (softlockup_all_cpu_backtrace) { |
403 | /* Prevent multiple soft-lockup reports if one cpu is already | |
404 | * engaged in dumping cpu back traces | |
405 | */ | |
406 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { | |
407 | /* Someone else will report us. Let's give up */ | |
408 | __this_cpu_write(soft_watchdog_warn, true); | |
409 | return HRTIMER_RESTART; | |
410 | } | |
411 | } | |
412 | ||
656c3b79 | 413 | pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 414 | smp_processor_id(), duration, |
58687acb DZ |
415 | current->comm, task_pid_nr(current)); |
416 | print_modules(); | |
417 | print_irqtrace_events(current); | |
418 | if (regs) | |
419 | show_regs(regs); | |
420 | else | |
421 | dump_stack(); | |
422 | ||
ed235875 AT |
423 | if (softlockup_all_cpu_backtrace) { |
424 | /* Avoid generating two back traces for current | |
425 | * given that one is already made above | |
426 | */ | |
427 | trigger_allbutself_cpu_backtrace(); | |
428 | ||
429 | clear_bit(0, &soft_lockup_nmi_warn); | |
430 | /* Barrier to sync with other cpus */ | |
431 | smp_mb__after_atomic(); | |
432 | } | |
433 | ||
69361eef | 434 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
58687acb DZ |
435 | if (softlockup_panic) |
436 | panic("softlockup: hung tasks"); | |
909ea964 | 437 | __this_cpu_write(soft_watchdog_warn, true); |
58687acb | 438 | } else |
909ea964 | 439 | __this_cpu_write(soft_watchdog_warn, false); |
58687acb DZ |
440 | |
441 | return HRTIMER_RESTART; | |
442 | } | |
443 | ||
bcd951cf | 444 | static void watchdog_enable(unsigned int cpu) |
58687acb | 445 | { |
01f0a027 | 446 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
be45bf53 | 447 | struct completion *done = this_cpu_ptr(&softlockup_completion); |
58687acb | 448 | |
9cf57731 PZ |
449 | WARN_ON_ONCE(cpu != smp_processor_id()); |
450 | ||
be45bf53 PZ |
451 | init_completion(done); |
452 | complete(done); | |
453 | ||
01f0a027 TG |
454 | /* |
455 | * Start the timer first to prevent the NMI watchdog triggering | |
456 | * before the timer has a chance to fire. | |
457 | */ | |
d2ab4cf4 | 458 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
3935e895 | 459 | hrtimer->function = watchdog_timer_fn; |
01f0a027 | 460 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
d2ab4cf4 | 461 | HRTIMER_MODE_REL_PINNED_HARD); |
3935e895 | 462 | |
01f0a027 TG |
463 | /* Initialize timestamp */ |
464 | __touch_watchdog(); | |
bcd951cf | 465 | /* Enable the perf event */ |
146c9d0e TG |
466 | if (watchdog_enabled & NMI_WATCHDOG_ENABLED) |
467 | watchdog_nmi_enable(cpu); | |
bcd951cf | 468 | } |
58687acb | 469 | |
bcd951cf TG |
470 | static void watchdog_disable(unsigned int cpu) |
471 | { | |
01f0a027 | 472 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
58687acb | 473 | |
9cf57731 PZ |
474 | WARN_ON_ONCE(cpu != smp_processor_id()); |
475 | ||
01f0a027 TG |
476 | /* |
477 | * Disable the perf event first. That prevents that a large delay | |
478 | * between disabling the timer and disabling the perf event causes | |
479 | * the perf NMI to detect a false positive. | |
480 | */ | |
bcd951cf | 481 | watchdog_nmi_disable(cpu); |
01f0a027 | 482 | hrtimer_cancel(hrtimer); |
be45bf53 | 483 | wait_for_completion(this_cpu_ptr(&softlockup_completion)); |
58687acb DZ |
484 | } |
485 | ||
9cf57731 | 486 | static int softlockup_stop_fn(void *data) |
b8900bc0 | 487 | { |
9cf57731 PZ |
488 | watchdog_disable(smp_processor_id()); |
489 | return 0; | |
b8900bc0 FW |
490 | } |
491 | ||
9cf57731 | 492 | static void softlockup_stop_all(void) |
bcd951cf | 493 | { |
9cf57731 PZ |
494 | int cpu; |
495 | ||
496 | if (!softlockup_initialized) | |
497 | return; | |
498 | ||
499 | for_each_cpu(cpu, &watchdog_allowed_mask) | |
500 | smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false); | |
501 | ||
502 | cpumask_clear(&watchdog_allowed_mask); | |
bcd951cf TG |
503 | } |
504 | ||
9cf57731 | 505 | static int softlockup_start_fn(void *data) |
bcd951cf | 506 | { |
9cf57731 PZ |
507 | watchdog_enable(smp_processor_id()); |
508 | return 0; | |
bcd951cf | 509 | } |
58687acb | 510 | |
9cf57731 | 511 | static void softlockup_start_all(void) |
2eb2527f | 512 | { |
9cf57731 | 513 | int cpu; |
2eb2527f | 514 | |
9cf57731 PZ |
515 | cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask); |
516 | for_each_cpu(cpu, &watchdog_allowed_mask) | |
517 | smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false); | |
2eb2527f TG |
518 | } |
519 | ||
9cf57731 | 520 | int lockup_detector_online_cpu(unsigned int cpu) |
2eb2527f | 521 | { |
7dd47617 TG |
522 | if (cpumask_test_cpu(cpu, &watchdog_allowed_mask)) |
523 | watchdog_enable(cpu); | |
9cf57731 | 524 | return 0; |
2eb2527f TG |
525 | } |
526 | ||
9cf57731 | 527 | int lockup_detector_offline_cpu(unsigned int cpu) |
2eb2527f | 528 | { |
7dd47617 TG |
529 | if (cpumask_test_cpu(cpu, &watchdog_allowed_mask)) |
530 | watchdog_disable(cpu); | |
9cf57731 | 531 | return 0; |
2eb2527f TG |
532 | } |
533 | ||
5587185d | 534 | static void lockup_detector_reconfigure(void) |
2eb2527f | 535 | { |
e31d6883 | 536 | cpus_read_lock(); |
6b9dc480 | 537 | watchdog_nmi_stop(); |
9cf57731 PZ |
538 | |
539 | softlockup_stop_all(); | |
2eb2527f | 540 | set_sample_period(); |
09154985 TG |
541 | lockup_detector_update_enable(); |
542 | if (watchdog_enabled && watchdog_thresh) | |
9cf57731 PZ |
543 | softlockup_start_all(); |
544 | ||
6b9dc480 | 545 | watchdog_nmi_start(); |
e31d6883 TG |
546 | cpus_read_unlock(); |
547 | /* | |
548 | * Must be called outside the cpus locked section to prevent | |
549 | * recursive locking in the perf code. | |
550 | */ | |
551 | __lockup_detector_cleanup(); | |
2eb2527f TG |
552 | } |
553 | ||
554 | /* | |
5587185d | 555 | * Create the watchdog thread infrastructure and configure the detector(s). |
2eb2527f TG |
556 | * |
557 | * The threads are not unparked as watchdog_allowed_mask is empty. When | |
76e15524 | 558 | * the threads are successfully initialized, take the proper locks and |
2eb2527f TG |
559 | * unpark the threads in the watchdog_cpumask if the watchdog is enabled. |
560 | */ | |
5587185d | 561 | static __init void lockup_detector_setup(void) |
2eb2527f | 562 | { |
2eb2527f TG |
563 | /* |
564 | * If sysctl is off and watchdog got disabled on the command line, | |
565 | * nothing to do here. | |
566 | */ | |
09154985 TG |
567 | lockup_detector_update_enable(); |
568 | ||
2eb2527f TG |
569 | if (!IS_ENABLED(CONFIG_SYSCTL) && |
570 | !(watchdog_enabled && watchdog_thresh)) | |
571 | return; | |
572 | ||
2eb2527f | 573 | mutex_lock(&watchdog_mutex); |
5587185d | 574 | lockup_detector_reconfigure(); |
9cf57731 | 575 | softlockup_initialized = true; |
2eb2527f TG |
576 | mutex_unlock(&watchdog_mutex); |
577 | } | |
578 | ||
2b9d7f23 | 579 | #else /* CONFIG_SOFTLOCKUP_DETECTOR */ |
5587185d | 580 | static void lockup_detector_reconfigure(void) |
6592ad2f | 581 | { |
e31d6883 | 582 | cpus_read_lock(); |
6b9dc480 | 583 | watchdog_nmi_stop(); |
09154985 | 584 | lockup_detector_update_enable(); |
6b9dc480 | 585 | watchdog_nmi_start(); |
e31d6883 | 586 | cpus_read_unlock(); |
6592ad2f | 587 | } |
5587185d | 588 | static inline void lockup_detector_setup(void) |
34ddaa3e | 589 | { |
5587185d | 590 | lockup_detector_reconfigure(); |
34ddaa3e | 591 | } |
2b9d7f23 | 592 | #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */ |
05a4a952 | 593 | |
941154bd TG |
594 | static void __lockup_detector_cleanup(void) |
595 | { | |
596 | lockdep_assert_held(&watchdog_mutex); | |
597 | hardlockup_detector_perf_cleanup(); | |
598 | } | |
599 | ||
600 | /** | |
601 | * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes | |
602 | * | |
603 | * Caller must not hold the cpu hotplug rwsem. | |
604 | */ | |
605 | void lockup_detector_cleanup(void) | |
606 | { | |
607 | mutex_lock(&watchdog_mutex); | |
608 | __lockup_detector_cleanup(); | |
609 | mutex_unlock(&watchdog_mutex); | |
610 | } | |
611 | ||
6554fd8c TG |
612 | /** |
613 | * lockup_detector_soft_poweroff - Interface to stop lockup detector(s) | |
614 | * | |
615 | * Special interface for parisc. It prevents lockup detector warnings from | |
616 | * the default pm_poweroff() function which busy loops forever. | |
617 | */ | |
618 | void lockup_detector_soft_poweroff(void) | |
619 | { | |
620 | watchdog_enabled = 0; | |
621 | } | |
622 | ||
58cf690a UO |
623 | #ifdef CONFIG_SYSCTL |
624 | ||
e8b62b2d | 625 | /* Propagate any changes to the watchdog threads */ |
d57108d4 | 626 | static void proc_watchdog_update(void) |
a0c9cbb9 | 627 | { |
e8b62b2d TG |
628 | /* Remove impossible cpus to keep sysctl output clean. */ |
629 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask); | |
5587185d | 630 | lockup_detector_reconfigure(); |
a0c9cbb9 UO |
631 | } |
632 | ||
ef246a21 UO |
633 | /* |
634 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter | |
635 | * | |
7feeb9cd TG |
636 | * caller | table->data points to | 'which' |
637 | * -------------------|----------------------------|-------------------------- | |
638 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED | | |
639 | * | | SOFT_WATCHDOG_ENABLED | |
640 | * -------------------|----------------------------|-------------------------- | |
641 | * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED | |
642 | * -------------------|----------------------------|-------------------------- | |
643 | * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED | |
ef246a21 UO |
644 | */ |
645 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, | |
32927393 | 646 | void *buffer, size_t *lenp, loff_t *ppos) |
ef246a21 | 647 | { |
09154985 | 648 | int err, old, *param = table->data; |
ef246a21 | 649 | |
946d1977 | 650 | mutex_lock(&watchdog_mutex); |
ef246a21 | 651 | |
ef246a21 | 652 | if (!write) { |
09154985 TG |
653 | /* |
654 | * On read synchronize the userspace interface. This is a | |
655 | * racy snapshot. | |
656 | */ | |
657 | *param = (watchdog_enabled & which) != 0; | |
ef246a21 UO |
658 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
659 | } else { | |
09154985 | 660 | old = READ_ONCE(*param); |
ef246a21 | 661 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
09154985 | 662 | if (!err && old != READ_ONCE(*param)) |
d57108d4 | 663 | proc_watchdog_update(); |
ef246a21 | 664 | } |
946d1977 | 665 | mutex_unlock(&watchdog_mutex); |
ef246a21 UO |
666 | return err; |
667 | } | |
668 | ||
83a80a39 UO |
669 | /* |
670 | * /proc/sys/kernel/watchdog | |
671 | */ | |
672 | int proc_watchdog(struct ctl_table *table, int write, | |
32927393 | 673 | void *buffer, size_t *lenp, loff_t *ppos) |
83a80a39 UO |
674 | { |
675 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, | |
676 | table, write, buffer, lenp, ppos); | |
677 | } | |
678 | ||
679 | /* | |
680 | * /proc/sys/kernel/nmi_watchdog | |
58687acb | 681 | */ |
83a80a39 | 682 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
32927393 | 683 | void *buffer, size_t *lenp, loff_t *ppos) |
83a80a39 | 684 | { |
a994a314 TG |
685 | if (!nmi_watchdog_available && write) |
686 | return -ENOTSUPP; | |
83a80a39 UO |
687 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, |
688 | table, write, buffer, lenp, ppos); | |
689 | } | |
690 | ||
691 | /* | |
692 | * /proc/sys/kernel/soft_watchdog | |
693 | */ | |
694 | int proc_soft_watchdog(struct ctl_table *table, int write, | |
32927393 | 695 | void *buffer, size_t *lenp, loff_t *ppos) |
83a80a39 UO |
696 | { |
697 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, | |
698 | table, write, buffer, lenp, ppos); | |
699 | } | |
58687acb | 700 | |
83a80a39 UO |
701 | /* |
702 | * /proc/sys/kernel/watchdog_thresh | |
703 | */ | |
704 | int proc_watchdog_thresh(struct ctl_table *table, int write, | |
32927393 | 705 | void *buffer, size_t *lenp, loff_t *ppos) |
58687acb | 706 | { |
d57108d4 | 707 | int err, old; |
58687acb | 708 | |
946d1977 | 709 | mutex_lock(&watchdog_mutex); |
bcd951cf | 710 | |
d57108d4 | 711 | old = READ_ONCE(watchdog_thresh); |
b8900bc0 | 712 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
83a80a39 | 713 | |
d57108d4 TG |
714 | if (!err && write && old != READ_ONCE(watchdog_thresh)) |
715 | proc_watchdog_update(); | |
e04ab2bc | 716 | |
946d1977 | 717 | mutex_unlock(&watchdog_mutex); |
b8900bc0 | 718 | return err; |
58687acb | 719 | } |
fe4ba3c3 CM |
720 | |
721 | /* | |
722 | * The cpumask is the mask of possible cpus that the watchdog can run | |
723 | * on, not the mask of cpus it is actually running on. This allows the | |
724 | * user to specify a mask that will include cpus that have not yet | |
725 | * been brought online, if desired. | |
726 | */ | |
727 | int proc_watchdog_cpumask(struct ctl_table *table, int write, | |
32927393 | 728 | void *buffer, size_t *lenp, loff_t *ppos) |
fe4ba3c3 CM |
729 | { |
730 | int err; | |
731 | ||
946d1977 | 732 | mutex_lock(&watchdog_mutex); |
8c073d27 | 733 | |
fe4ba3c3 | 734 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
05ba3de7 | 735 | if (!err && write) |
e8b62b2d | 736 | proc_watchdog_update(); |
5490125d | 737 | |
946d1977 | 738 | mutex_unlock(&watchdog_mutex); |
fe4ba3c3 CM |
739 | return err; |
740 | } | |
58687acb DZ |
741 | #endif /* CONFIG_SYSCTL */ |
742 | ||
004417a6 | 743 | void __init lockup_detector_init(void) |
58687acb | 744 | { |
13316b31 | 745 | if (tick_nohz_full_enabled()) |
314b08ff | 746 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
13316b31 | 747 | |
de201559 FW |
748 | cpumask_copy(&watchdog_cpumask, |
749 | housekeeping_cpumask(HK_FLAG_TIMER)); | |
fe4ba3c3 | 750 | |
a994a314 TG |
751 | if (!watchdog_nmi_probe()) |
752 | nmi_watchdog_available = true; | |
5587185d | 753 | lockup_detector_setup(); |
58687acb | 754 | } |