]>
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> | |
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> |
fe4ba3c3 | 22 | #include <linux/tick.h> |
82607adc | 23 | #include <linux/workqueue.h> |
58687acb DZ |
24 | |
25 | #include <asm/irq_regs.h> | |
5d1c0f4a | 26 | #include <linux/kvm_para.h> |
58687acb | 27 | #include <linux/perf_event.h> |
81a4beef | 28 | #include <linux/kthread.h> |
58687acb | 29 | |
84d56e66 UO |
30 | /* |
31 | * The run state of the lockup detectors is controlled by the content of the | |
32 | * 'watchdog_enabled' variable. Each lockup detector has its dedicated bit - | |
33 | * bit 0 for the hard lockup detector and bit 1 for the soft lockup detector. | |
34 | * | |
35 | * 'watchdog_user_enabled', 'nmi_watchdog_enabled' and 'soft_watchdog_enabled' | |
36 | * are variables that are only used as an 'interface' between the parameters | |
37 | * in /proc/sys/kernel and the internal state bits in 'watchdog_enabled'. The | |
38 | * 'watchdog_thresh' variable is handled differently because its value is not | |
39 | * boolean, and the lockup detectors are 'suspended' while 'watchdog_thresh' | |
40 | * is equal zero. | |
41 | */ | |
42 | #define NMI_WATCHDOG_ENABLED_BIT 0 | |
43 | #define SOFT_WATCHDOG_ENABLED_BIT 1 | |
44 | #define NMI_WATCHDOG_ENABLED (1 << NMI_WATCHDOG_ENABLED_BIT) | |
45 | #define SOFT_WATCHDOG_ENABLED (1 << SOFT_WATCHDOG_ENABLED_BIT) | |
46 | ||
ab992dc3 PZ |
47 | static DEFINE_MUTEX(watchdog_proc_mutex); |
48 | ||
84d56e66 UO |
49 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
50 | static unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED|NMI_WATCHDOG_ENABLED; | |
51 | #else | |
52 | static unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED; | |
53 | #endif | |
54 | int __read_mostly nmi_watchdog_enabled; | |
55 | int __read_mostly soft_watchdog_enabled; | |
56 | int __read_mostly watchdog_user_enabled; | |
4eec42f3 | 57 | int __read_mostly watchdog_thresh = 10; |
84d56e66 | 58 | |
ed235875 AT |
59 | #ifdef CONFIG_SMP |
60 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; | |
55537871 | 61 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; |
ed235875 AT |
62 | #else |
63 | #define sysctl_softlockup_all_cpu_backtrace 0 | |
55537871 | 64 | #define sysctl_hardlockup_all_cpu_backtrace 0 |
ed235875 | 65 | #endif |
fe4ba3c3 CM |
66 | static struct cpumask watchdog_cpumask __read_mostly; |
67 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); | |
68 | ||
69 | /* Helper for online, unparked cpus. */ | |
70 | #define for_each_watchdog_cpu(cpu) \ | |
71 | for_each_cpu_and((cpu), cpu_online_mask, &watchdog_cpumask) | |
ed235875 | 72 | |
ec6a9066 UO |
73 | /* |
74 | * The 'watchdog_running' variable is set to 1 when the watchdog threads | |
75 | * are registered/started and is set to 0 when the watchdog threads are | |
76 | * unregistered/stopped, so it is an indicator whether the threads exist. | |
77 | */ | |
3c00ea82 | 78 | static int __read_mostly watchdog_running; |
ec6a9066 UO |
79 | /* |
80 | * If a subsystem has a need to deactivate the watchdog temporarily, it | |
81 | * can use the suspend/resume interface to achieve this. The content of | |
82 | * the 'watchdog_suspended' variable reflects this state. Existing threads | |
83 | * are parked/unparked by the lockup_detector_{suspend|resume} functions | |
84 | * (see comment blocks pertaining to those functions for further details). | |
85 | * | |
86 | * 'watchdog_suspended' also prevents threads from being registered/started | |
87 | * or unregistered/stopped via parameters in /proc/sys/kernel, so the state | |
88 | * of 'watchdog_running' cannot change while the watchdog is deactivated | |
89 | * temporarily (see related code in 'proc' handlers). | |
90 | */ | |
91 | static int __read_mostly watchdog_suspended; | |
92 | ||
0f34c400 | 93 | static u64 __read_mostly sample_period; |
58687acb DZ |
94 | |
95 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); | |
96 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); | |
97 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); | |
98 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); | |
58687acb | 99 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
bcd951cf TG |
100 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
101 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); | |
b1a8de1f | 102 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
23637d47 | 103 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
cafcd80d DZ |
104 | static DEFINE_PER_CPU(bool, hard_watchdog_warn); |
105 | static DEFINE_PER_CPU(bool, watchdog_nmi_touch); | |
58687acb DZ |
106 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
107 | static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); | |
108 | #endif | |
ed235875 | 109 | static unsigned long soft_lockup_nmi_warn; |
58687acb | 110 | |
58687acb DZ |
111 | /* boot commands */ |
112 | /* | |
113 | * Should we panic when a soft-lockup or hard-lockup occurs: | |
114 | */ | |
23637d47 | 115 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
ac1f5912 | 116 | unsigned int __read_mostly hardlockup_panic = |
fef2c9bc | 117 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; |
55537871 | 118 | static unsigned long hardlockup_allcpu_dumped; |
6e7458a6 UO |
119 | /* |
120 | * We may not want to enable hard lockup detection by default in all cases, | |
121 | * for example when running the kernel as a guest on a hypervisor. In these | |
122 | * cases this function can be called to disable hard lockup detection. This | |
123 | * function should only be executed once by the boot processor before the | |
124 | * kernel command line parameters are parsed, because otherwise it is not | |
125 | * possible to override this in hardlockup_panic_setup(). | |
126 | */ | |
692297d8 | 127 | void hardlockup_detector_disable(void) |
6e7458a6 | 128 | { |
692297d8 | 129 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
6e7458a6 UO |
130 | } |
131 | ||
58687acb DZ |
132 | static int __init hardlockup_panic_setup(char *str) |
133 | { | |
134 | if (!strncmp(str, "panic", 5)) | |
135 | hardlockup_panic = 1; | |
fef2c9bc DZ |
136 | else if (!strncmp(str, "nopanic", 7)) |
137 | hardlockup_panic = 0; | |
5dc30558 | 138 | else if (!strncmp(str, "0", 1)) |
195daf66 UO |
139 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
140 | else if (!strncmp(str, "1", 1)) | |
141 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; | |
58687acb DZ |
142 | return 1; |
143 | } | |
144 | __setup("nmi_watchdog=", hardlockup_panic_setup); | |
145 | #endif | |
146 | ||
147 | unsigned int __read_mostly softlockup_panic = | |
148 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; | |
149 | ||
150 | static int __init softlockup_panic_setup(char *str) | |
151 | { | |
152 | softlockup_panic = simple_strtoul(str, NULL, 0); | |
153 | ||
154 | return 1; | |
155 | } | |
156 | __setup("softlockup_panic=", softlockup_panic_setup); | |
157 | ||
158 | static int __init nowatchdog_setup(char *str) | |
159 | { | |
195daf66 | 160 | watchdog_enabled = 0; |
58687acb DZ |
161 | return 1; |
162 | } | |
163 | __setup("nowatchdog", nowatchdog_setup); | |
164 | ||
58687acb DZ |
165 | static int __init nosoftlockup_setup(char *str) |
166 | { | |
195daf66 | 167 | watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED; |
58687acb DZ |
168 | return 1; |
169 | } | |
170 | __setup("nosoftlockup", nosoftlockup_setup); | |
195daf66 | 171 | |
ed235875 AT |
172 | #ifdef CONFIG_SMP |
173 | static int __init softlockup_all_cpu_backtrace_setup(char *str) | |
174 | { | |
175 | sysctl_softlockup_all_cpu_backtrace = | |
176 | !!simple_strtol(str, NULL, 0); | |
177 | return 1; | |
178 | } | |
179 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); | |
55537871 JK |
180 | static int __init hardlockup_all_cpu_backtrace_setup(char *str) |
181 | { | |
182 | sysctl_hardlockup_all_cpu_backtrace = | |
183 | !!simple_strtol(str, NULL, 0); | |
184 | return 1; | |
185 | } | |
186 | __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup); | |
ed235875 | 187 | #endif |
58687acb | 188 | |
4eec42f3 MSB |
189 | /* |
190 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- | |
191 | * lockups can have false positives under extreme conditions. So we generally | |
192 | * want a higher threshold for soft lockups than for hard lockups. So we couple | |
193 | * the thresholds with a factor: we make the soft threshold twice the amount of | |
194 | * time the hard threshold is. | |
195 | */ | |
6e9101ae | 196 | static int get_softlockup_thresh(void) |
4eec42f3 MSB |
197 | { |
198 | return watchdog_thresh * 2; | |
199 | } | |
58687acb DZ |
200 | |
201 | /* | |
202 | * Returns seconds, approximately. We don't need nanosecond | |
203 | * resolution, and we don't need to waste time with a big divide when | |
204 | * 2^30ns == 1.074s. | |
205 | */ | |
c06b4f19 | 206 | static unsigned long get_timestamp(void) |
58687acb | 207 | { |
545a2bf7 | 208 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
58687acb DZ |
209 | } |
210 | ||
0f34c400 | 211 | static void set_sample_period(void) |
58687acb DZ |
212 | { |
213 | /* | |
586692a5 | 214 | * convert watchdog_thresh from seconds to ns |
86f5e6a7 FLVC |
215 | * the divide by 5 is to give hrtimer several chances (two |
216 | * or three with the current relation between the soft | |
217 | * and hard thresholds) to increment before the | |
218 | * hardlockup detector generates a warning | |
58687acb | 219 | */ |
0f34c400 | 220 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
58687acb DZ |
221 | } |
222 | ||
223 | /* Commands for resetting the watchdog */ | |
224 | static void __touch_watchdog(void) | |
225 | { | |
c06b4f19 | 226 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
58687acb DZ |
227 | } |
228 | ||
03e0d461 TH |
229 | /** |
230 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls | |
231 | * | |
232 | * Call when the scheduler may have stalled for legitimate reasons | |
233 | * preventing the watchdog task from executing - e.g. the scheduler | |
234 | * entering idle state. This should only be used for scheduler events. | |
235 | * Use touch_softlockup_watchdog() for everything else. | |
236 | */ | |
237 | void touch_softlockup_watchdog_sched(void) | |
58687acb | 238 | { |
7861144b AM |
239 | /* |
240 | * Preemption can be enabled. It doesn't matter which CPU's timestamp | |
241 | * gets zeroed here, so use the raw_ operation. | |
242 | */ | |
243 | raw_cpu_write(watchdog_touch_ts, 0); | |
58687acb | 244 | } |
03e0d461 TH |
245 | |
246 | void touch_softlockup_watchdog(void) | |
247 | { | |
248 | touch_softlockup_watchdog_sched(); | |
82607adc | 249 | wq_watchdog_touch(raw_smp_processor_id()); |
03e0d461 | 250 | } |
0167c781 | 251 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
58687acb | 252 | |
332fbdbc | 253 | void touch_all_softlockup_watchdogs(void) |
58687acb DZ |
254 | { |
255 | int cpu; | |
256 | ||
257 | /* | |
258 | * this is done lockless | |
259 | * do we care if a 0 races with a timestamp? | |
260 | * all it means is the softlock check starts one cycle later | |
261 | */ | |
fe4ba3c3 | 262 | for_each_watchdog_cpu(cpu) |
58687acb | 263 | per_cpu(watchdog_touch_ts, cpu) = 0; |
82607adc | 264 | wq_watchdog_touch(-1); |
58687acb DZ |
265 | } |
266 | ||
cafcd80d | 267 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb DZ |
268 | void touch_nmi_watchdog(void) |
269 | { | |
62572e29 BZ |
270 | /* |
271 | * Using __raw here because some code paths have | |
272 | * preemption enabled. If preemption is enabled | |
273 | * then interrupts should be enabled too, in which | |
274 | * case we shouldn't have to worry about the watchdog | |
275 | * going off. | |
276 | */ | |
f7f66b05 | 277 | raw_cpu_write(watchdog_nmi_touch, true); |
332fbdbc | 278 | touch_softlockup_watchdog(); |
58687acb DZ |
279 | } |
280 | EXPORT_SYMBOL(touch_nmi_watchdog); | |
281 | ||
cafcd80d DZ |
282 | #endif |
283 | ||
58687acb DZ |
284 | void touch_softlockup_watchdog_sync(void) |
285 | { | |
f7f66b05 CL |
286 | __this_cpu_write(softlockup_touch_sync, true); |
287 | __this_cpu_write(watchdog_touch_ts, 0); | |
58687acb DZ |
288 | } |
289 | ||
23637d47 | 290 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
58687acb | 291 | /* watchdog detector functions */ |
451637e4 | 292 | static bool is_hardlockup(void) |
58687acb | 293 | { |
909ea964 | 294 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
58687acb | 295 | |
909ea964 | 296 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
451637e4 | 297 | return true; |
58687acb | 298 | |
909ea964 | 299 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
451637e4 | 300 | return false; |
58687acb DZ |
301 | } |
302 | #endif | |
303 | ||
26e09c6e | 304 | static int is_softlockup(unsigned long touch_ts) |
58687acb | 305 | { |
c06b4f19 | 306 | unsigned long now = get_timestamp(); |
58687acb | 307 | |
39d2da21 | 308 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
195daf66 UO |
309 | /* Warn about unreasonable delays. */ |
310 | if (time_after(now, touch_ts + get_softlockup_thresh())) | |
311 | return now - touch_ts; | |
312 | } | |
58687acb DZ |
313 | return 0; |
314 | } | |
315 | ||
23637d47 | 316 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
1880c4ae | 317 | |
58687acb DZ |
318 | static struct perf_event_attr wd_hw_attr = { |
319 | .type = PERF_TYPE_HARDWARE, | |
320 | .config = PERF_COUNT_HW_CPU_CYCLES, | |
321 | .size = sizeof(struct perf_event_attr), | |
322 | .pinned = 1, | |
323 | .disabled = 1, | |
324 | }; | |
325 | ||
326 | /* Callback function for perf event subsystem */ | |
a8b0ca17 | 327 | static void watchdog_overflow_callback(struct perf_event *event, |
58687acb DZ |
328 | struct perf_sample_data *data, |
329 | struct pt_regs *regs) | |
330 | { | |
c6db67cd PZ |
331 | /* Ensure the watchdog never gets throttled */ |
332 | event->hw.interrupts = 0; | |
333 | ||
909ea964 CL |
334 | if (__this_cpu_read(watchdog_nmi_touch) == true) { |
335 | __this_cpu_write(watchdog_nmi_touch, false); | |
58687acb DZ |
336 | return; |
337 | } | |
338 | ||
339 | /* check for a hardlockup | |
340 | * This is done by making sure our timer interrupt | |
341 | * is incrementing. The timer interrupt should have | |
342 | * fired multiple times before we overflow'd. If it hasn't | |
343 | * then this is a good indication the cpu is stuck | |
344 | */ | |
26e09c6e DZ |
345 | if (is_hardlockup()) { |
346 | int this_cpu = smp_processor_id(); | |
347 | ||
58687acb | 348 | /* only print hardlockups once */ |
909ea964 | 349 | if (__this_cpu_read(hard_watchdog_warn) == true) |
58687acb DZ |
350 | return; |
351 | ||
55537871 JK |
352 | pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu); |
353 | print_modules(); | |
354 | print_irqtrace_events(current); | |
355 | if (regs) | |
356 | show_regs(regs); | |
58687acb | 357 | else |
55537871 JK |
358 | dump_stack(); |
359 | ||
360 | /* | |
361 | * Perform all-CPU dump only once to avoid multiple hardlockups | |
362 | * generating interleaving traces | |
363 | */ | |
364 | if (sysctl_hardlockup_all_cpu_backtrace && | |
365 | !test_and_set_bit(0, &hardlockup_allcpu_dumped)) | |
366 | trigger_allbutself_cpu_backtrace(); | |
367 | ||
368 | if (hardlockup_panic) | |
58c5661f | 369 | nmi_panic(regs, "Hard LOCKUP"); |
58687acb | 370 | |
909ea964 | 371 | __this_cpu_write(hard_watchdog_warn, true); |
58687acb DZ |
372 | return; |
373 | } | |
374 | ||
909ea964 | 375 | __this_cpu_write(hard_watchdog_warn, false); |
58687acb DZ |
376 | return; |
377 | } | |
bcd951cf TG |
378 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
379 | ||
58687acb DZ |
380 | static void watchdog_interrupt_count(void) |
381 | { | |
909ea964 | 382 | __this_cpu_inc(hrtimer_interrupts); |
58687acb | 383 | } |
bcd951cf TG |
384 | |
385 | static int watchdog_nmi_enable(unsigned int cpu); | |
386 | static void watchdog_nmi_disable(unsigned int cpu); | |
58687acb | 387 | |
58cf690a UO |
388 | static int watchdog_enable_all_cpus(void); |
389 | static void watchdog_disable_all_cpus(void); | |
390 | ||
58687acb DZ |
391 | /* watchdog kicker functions */ |
392 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
393 | { | |
909ea964 | 394 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
58687acb DZ |
395 | struct pt_regs *regs = get_irq_regs(); |
396 | int duration; | |
ed235875 | 397 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
58687acb DZ |
398 | |
399 | /* kick the hardlockup detector */ | |
400 | watchdog_interrupt_count(); | |
401 | ||
402 | /* kick the softlockup detector */ | |
909ea964 | 403 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
58687acb DZ |
404 | |
405 | /* .. and repeat */ | |
0f34c400 | 406 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
58687acb DZ |
407 | |
408 | if (touch_ts == 0) { | |
909ea964 | 409 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
410 | /* |
411 | * If the time stamp was touched atomically | |
412 | * make sure the scheduler tick is up to date. | |
413 | */ | |
909ea964 | 414 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
415 | sched_clock_tick(); |
416 | } | |
5d1c0f4a EM |
417 | |
418 | /* Clear the guest paused flag on watchdog reset */ | |
419 | kvm_check_and_clear_guest_paused(); | |
58687acb DZ |
420 | __touch_watchdog(); |
421 | return HRTIMER_RESTART; | |
422 | } | |
423 | ||
424 | /* check for a softlockup | |
425 | * This is done by making sure a high priority task is | |
426 | * being scheduled. The task touches the watchdog to | |
427 | * indicate it is getting cpu time. If it hasn't then | |
428 | * this is a good indication some task is hogging the cpu | |
429 | */ | |
26e09c6e | 430 | duration = is_softlockup(touch_ts); |
58687acb | 431 | if (unlikely(duration)) { |
5d1c0f4a EM |
432 | /* |
433 | * If a virtual machine is stopped by the host it can look to | |
434 | * the watchdog like a soft lockup, check to see if the host | |
435 | * stopped the vm before we issue the warning | |
436 | */ | |
437 | if (kvm_check_and_clear_guest_paused()) | |
438 | return HRTIMER_RESTART; | |
439 | ||
58687acb | 440 | /* only warn once */ |
b1a8de1f | 441 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
442 | /* | |
443 | * When multiple processes are causing softlockups the | |
444 | * softlockup detector only warns on the first one | |
445 | * because the code relies on a full quiet cycle to | |
446 | * re-arm. The second process prevents the quiet cycle | |
447 | * and never gets reported. Use task pointers to detect | |
448 | * this. | |
449 | */ | |
450 | if (__this_cpu_read(softlockup_task_ptr_saved) != | |
451 | current) { | |
452 | __this_cpu_write(soft_watchdog_warn, false); | |
453 | __touch_watchdog(); | |
454 | } | |
58687acb | 455 | return HRTIMER_RESTART; |
b1a8de1f | 456 | } |
58687acb | 457 | |
ed235875 AT |
458 | if (softlockup_all_cpu_backtrace) { |
459 | /* Prevent multiple soft-lockup reports if one cpu is already | |
460 | * engaged in dumping cpu back traces | |
461 | */ | |
462 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { | |
463 | /* Someone else will report us. Let's give up */ | |
464 | __this_cpu_write(soft_watchdog_warn, true); | |
465 | return HRTIMER_RESTART; | |
466 | } | |
467 | } | |
468 | ||
656c3b79 | 469 | pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 470 | smp_processor_id(), duration, |
58687acb | 471 | current->comm, task_pid_nr(current)); |
b1a8de1f | 472 | __this_cpu_write(softlockup_task_ptr_saved, current); |
58687acb DZ |
473 | print_modules(); |
474 | print_irqtrace_events(current); | |
475 | if (regs) | |
476 | show_regs(regs); | |
477 | else | |
478 | dump_stack(); | |
479 | ||
ed235875 AT |
480 | if (softlockup_all_cpu_backtrace) { |
481 | /* Avoid generating two back traces for current | |
482 | * given that one is already made above | |
483 | */ | |
484 | trigger_allbutself_cpu_backtrace(); | |
485 | ||
486 | clear_bit(0, &soft_lockup_nmi_warn); | |
487 | /* Barrier to sync with other cpus */ | |
488 | smp_mb__after_atomic(); | |
489 | } | |
490 | ||
69361eef | 491 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
58687acb DZ |
492 | if (softlockup_panic) |
493 | panic("softlockup: hung tasks"); | |
909ea964 | 494 | __this_cpu_write(soft_watchdog_warn, true); |
58687acb | 495 | } else |
909ea964 | 496 | __this_cpu_write(soft_watchdog_warn, false); |
58687acb DZ |
497 | |
498 | return HRTIMER_RESTART; | |
499 | } | |
500 | ||
bcd951cf TG |
501 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
502 | { | |
503 | struct sched_param param = { .sched_priority = prio }; | |
58687acb | 504 | |
bcd951cf TG |
505 | sched_setscheduler(current, policy, ¶m); |
506 | } | |
507 | ||
508 | static void watchdog_enable(unsigned int cpu) | |
58687acb | 509 | { |
f7f66b05 | 510 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
58687acb | 511 | |
3935e895 BM |
512 | /* kick off the timer for the hardlockup detector */ |
513 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
514 | hrtimer->function = watchdog_timer_fn; | |
515 | ||
bcd951cf TG |
516 | /* Enable the perf event */ |
517 | watchdog_nmi_enable(cpu); | |
58687acb | 518 | |
58687acb | 519 | /* done here because hrtimer_start can only pin to smp_processor_id() */ |
0f34c400 | 520 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
58687acb DZ |
521 | HRTIMER_MODE_REL_PINNED); |
522 | ||
bcd951cf TG |
523 | /* initialize timestamp */ |
524 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); | |
525 | __touch_watchdog(); | |
526 | } | |
58687acb | 527 | |
bcd951cf TG |
528 | static void watchdog_disable(unsigned int cpu) |
529 | { | |
f7f66b05 | 530 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
58687acb | 531 | |
bcd951cf TG |
532 | watchdog_set_prio(SCHED_NORMAL, 0); |
533 | hrtimer_cancel(hrtimer); | |
534 | /* disable the perf event */ | |
535 | watchdog_nmi_disable(cpu); | |
58687acb DZ |
536 | } |
537 | ||
b8900bc0 FW |
538 | static void watchdog_cleanup(unsigned int cpu, bool online) |
539 | { | |
540 | watchdog_disable(cpu); | |
541 | } | |
542 | ||
bcd951cf TG |
543 | static int watchdog_should_run(unsigned int cpu) |
544 | { | |
545 | return __this_cpu_read(hrtimer_interrupts) != | |
546 | __this_cpu_read(soft_lockup_hrtimer_cnt); | |
547 | } | |
548 | ||
549 | /* | |
550 | * The watchdog thread function - touches the timestamp. | |
551 | * | |
0f34c400 | 552 | * It only runs once every sample_period seconds (4 seconds by |
bcd951cf TG |
553 | * default) to reset the softlockup timestamp. If this gets delayed |
554 | * for more than 2*watchdog_thresh seconds then the debug-printout | |
555 | * triggers in watchdog_timer_fn(). | |
556 | */ | |
557 | static void watchdog(unsigned int cpu) | |
558 | { | |
559 | __this_cpu_write(soft_lockup_hrtimer_cnt, | |
560 | __this_cpu_read(hrtimer_interrupts)); | |
561 | __touch_watchdog(); | |
bcfba4f4 UO |
562 | |
563 | /* | |
564 | * watchdog_nmi_enable() clears the NMI_WATCHDOG_ENABLED bit in the | |
565 | * failure path. Check for failures that can occur asynchronously - | |
566 | * for example, when CPUs are on-lined - and shut down the hardware | |
567 | * perf event on each CPU accordingly. | |
568 | * | |
569 | * The only non-obvious place this bit can be cleared is through | |
570 | * watchdog_nmi_enable(), so a pr_info() is placed there. Placing a | |
571 | * pr_info here would be too noisy as it would result in a message | |
572 | * every few seconds if the hardlockup was disabled but the softlockup | |
573 | * enabled. | |
574 | */ | |
575 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) | |
576 | watchdog_nmi_disable(cpu); | |
bcd951cf | 577 | } |
58687acb | 578 | |
23637d47 | 579 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
a7027046 DZ |
580 | /* |
581 | * People like the simple clean cpu node info on boot. | |
582 | * Reduce the watchdog noise by only printing messages | |
583 | * that are different from what cpu0 displayed. | |
584 | */ | |
585 | static unsigned long cpu0_err; | |
586 | ||
bcd951cf | 587 | static int watchdog_nmi_enable(unsigned int cpu) |
58687acb DZ |
588 | { |
589 | struct perf_event_attr *wd_attr; | |
590 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
591 | ||
195daf66 UO |
592 | /* nothing to do if the hard lockup detector is disabled */ |
593 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) | |
594 | goto out; | |
6e7458a6 | 595 | |
58687acb DZ |
596 | /* is it already setup and enabled? */ |
597 | if (event && event->state > PERF_EVENT_STATE_OFF) | |
598 | goto out; | |
599 | ||
600 | /* it is setup but not enabled */ | |
601 | if (event != NULL) | |
602 | goto out_enable; | |
603 | ||
58687acb | 604 | wd_attr = &wd_hw_attr; |
4eec42f3 | 605 | wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh); |
1880c4ae CG |
606 | |
607 | /* Try to register using hardware perf events */ | |
4dc0da86 | 608 | event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL); |
a7027046 DZ |
609 | |
610 | /* save cpu0 error for future comparision */ | |
611 | if (cpu == 0 && IS_ERR(event)) | |
612 | cpu0_err = PTR_ERR(event); | |
613 | ||
58687acb | 614 | if (!IS_ERR(event)) { |
a7027046 DZ |
615 | /* only print for cpu0 or different than cpu0 */ |
616 | if (cpu == 0 || cpu0_err) | |
617 | pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n"); | |
58687acb DZ |
618 | goto out_save; |
619 | } | |
620 | ||
bcfba4f4 UO |
621 | /* |
622 | * Disable the hard lockup detector if _any_ CPU fails to set up | |
623 | * set up the hardware perf event. The watchdog() function checks | |
624 | * the NMI_WATCHDOG_ENABLED bit periodically. | |
625 | * | |
626 | * The barriers are for syncing up watchdog_enabled across all the | |
627 | * cpus, as clear_bit() does not use barriers. | |
628 | */ | |
629 | smp_mb__before_atomic(); | |
630 | clear_bit(NMI_WATCHDOG_ENABLED_BIT, &watchdog_enabled); | |
631 | smp_mb__after_atomic(); | |
632 | ||
a7027046 DZ |
633 | /* skip displaying the same error again */ |
634 | if (cpu > 0 && (PTR_ERR(event) == cpu0_err)) | |
635 | return PTR_ERR(event); | |
5651f7f4 DZ |
636 | |
637 | /* vary the KERN level based on the returned errno */ | |
638 | if (PTR_ERR(event) == -EOPNOTSUPP) | |
4501980a | 639 | pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu); |
5651f7f4 | 640 | else if (PTR_ERR(event) == -ENOENT) |
656c3b79 | 641 | pr_warn("disabled (cpu%i): hardware events not enabled\n", |
4501980a | 642 | cpu); |
5651f7f4 | 643 | else |
4501980a AM |
644 | pr_err("disabled (cpu%i): unable to create perf event: %ld\n", |
645 | cpu, PTR_ERR(event)); | |
bcfba4f4 UO |
646 | |
647 | pr_info("Shutting down hard lockup detector on all cpus\n"); | |
648 | ||
eac24335 | 649 | return PTR_ERR(event); |
58687acb DZ |
650 | |
651 | /* success path */ | |
652 | out_save: | |
653 | per_cpu(watchdog_ev, cpu) = event; | |
654 | out_enable: | |
655 | perf_event_enable(per_cpu(watchdog_ev, cpu)); | |
656 | out: | |
657 | return 0; | |
658 | } | |
659 | ||
bcd951cf | 660 | static void watchdog_nmi_disable(unsigned int cpu) |
58687acb DZ |
661 | { |
662 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
663 | ||
664 | if (event) { | |
665 | perf_event_disable(event); | |
666 | per_cpu(watchdog_ev, cpu) = NULL; | |
667 | ||
668 | /* should be in cleanup, but blocks oprofile */ | |
669 | perf_event_release_kernel(event); | |
670 | } | |
df577149 UO |
671 | if (cpu == 0) { |
672 | /* watchdog_nmi_enable() expects this to be zero initially. */ | |
673 | cpu0_err = 0; | |
674 | } | |
58687acb | 675 | } |
b3738d29 | 676 | |
58687acb | 677 | #else |
bcd951cf TG |
678 | static int watchdog_nmi_enable(unsigned int cpu) { return 0; } |
679 | static void watchdog_nmi_disable(unsigned int cpu) { return; } | |
23637d47 | 680 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
58687acb | 681 | |
b8900bc0 FW |
682 | static struct smp_hotplug_thread watchdog_threads = { |
683 | .store = &softlockup_watchdog, | |
684 | .thread_should_run = watchdog_should_run, | |
685 | .thread_fn = watchdog, | |
686 | .thread_comm = "watchdog/%u", | |
687 | .setup = watchdog_enable, | |
688 | .cleanup = watchdog_cleanup, | |
689 | .park = watchdog_disable, | |
690 | .unpark = watchdog_enable, | |
691 | }; | |
692 | ||
81a4beef UO |
693 | /* |
694 | * park all watchdog threads that are specified in 'watchdog_cpumask' | |
ee7fed54 UO |
695 | * |
696 | * This function returns an error if kthread_park() of a watchdog thread | |
697 | * fails. In this situation, the watchdog threads of some CPUs can already | |
698 | * be parked and the watchdog threads of other CPUs can still be runnable. | |
699 | * Callers are expected to handle this special condition as appropriate in | |
700 | * their context. | |
a2a45b85 UO |
701 | * |
702 | * This function may only be called in a context that is protected against | |
703 | * races with CPU hotplug - for example, via get_online_cpus(). | |
81a4beef UO |
704 | */ |
705 | static int watchdog_park_threads(void) | |
706 | { | |
707 | int cpu, ret = 0; | |
708 | ||
81a4beef UO |
709 | for_each_watchdog_cpu(cpu) { |
710 | ret = kthread_park(per_cpu(softlockup_watchdog, cpu)); | |
711 | if (ret) | |
712 | break; | |
713 | } | |
81a4beef UO |
714 | |
715 | return ret; | |
716 | } | |
717 | ||
718 | /* | |
719 | * unpark all watchdog threads that are specified in 'watchdog_cpumask' | |
a2a45b85 UO |
720 | * |
721 | * This function may only be called in a context that is protected against | |
722 | * races with CPU hotplug - for example, via get_online_cpus(). | |
81a4beef UO |
723 | */ |
724 | static void watchdog_unpark_threads(void) | |
725 | { | |
726 | int cpu; | |
727 | ||
81a4beef UO |
728 | for_each_watchdog_cpu(cpu) |
729 | kthread_unpark(per_cpu(softlockup_watchdog, cpu)); | |
81a4beef UO |
730 | } |
731 | ||
8c073d27 UO |
732 | /* |
733 | * Suspend the hard and soft lockup detector by parking the watchdog threads. | |
734 | */ | |
ec6a9066 | 735 | int lockup_detector_suspend(void) |
8c073d27 UO |
736 | { |
737 | int ret = 0; | |
738 | ||
ee89e71e | 739 | get_online_cpus(); |
8c073d27 UO |
740 | mutex_lock(&watchdog_proc_mutex); |
741 | /* | |
742 | * Multiple suspend requests can be active in parallel (counted by | |
743 | * the 'watchdog_suspended' variable). If the watchdog threads are | |
744 | * running, the first caller takes care that they will be parked. | |
745 | * The state of 'watchdog_running' cannot change while a suspend | |
ec6a9066 | 746 | * request is active (see related code in 'proc' handlers). |
8c073d27 UO |
747 | */ |
748 | if (watchdog_running && !watchdog_suspended) | |
749 | ret = watchdog_park_threads(); | |
750 | ||
751 | if (ret == 0) | |
752 | watchdog_suspended++; | |
c993590c UO |
753 | else { |
754 | watchdog_disable_all_cpus(); | |
755 | pr_err("Failed to suspend lockup detectors, disabled\n"); | |
756 | watchdog_enabled = 0; | |
757 | } | |
8c073d27 UO |
758 | |
759 | mutex_unlock(&watchdog_proc_mutex); | |
760 | ||
761 | return ret; | |
762 | } | |
763 | ||
764 | /* | |
765 | * Resume the hard and soft lockup detector by unparking the watchdog threads. | |
766 | */ | |
ec6a9066 | 767 | void lockup_detector_resume(void) |
8c073d27 UO |
768 | { |
769 | mutex_lock(&watchdog_proc_mutex); | |
770 | ||
771 | watchdog_suspended--; | |
772 | /* | |
773 | * The watchdog threads are unparked if they were previously running | |
774 | * and if there is no more active suspend request. | |
775 | */ | |
776 | if (watchdog_running && !watchdog_suspended) | |
777 | watchdog_unpark_threads(); | |
778 | ||
779 | mutex_unlock(&watchdog_proc_mutex); | |
ee89e71e | 780 | put_online_cpus(); |
8c073d27 UO |
781 | } |
782 | ||
b43cb43c | 783 | static int update_watchdog_all_cpus(void) |
9809b18f | 784 | { |
b43cb43c UO |
785 | int ret; |
786 | ||
787 | ret = watchdog_park_threads(); | |
788 | if (ret) | |
789 | return ret; | |
790 | ||
d4bdd0b2 | 791 | watchdog_unpark_threads(); |
b43cb43c UO |
792 | |
793 | return 0; | |
9809b18f MH |
794 | } |
795 | ||
b2f57c3a | 796 | static int watchdog_enable_all_cpus(void) |
58687acb | 797 | { |
b8900bc0 | 798 | int err = 0; |
58687acb | 799 | |
3c00ea82 | 800 | if (!watchdog_running) { |
230ec939 FW |
801 | err = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
802 | &watchdog_cpumask); | |
b8900bc0 FW |
803 | if (err) |
804 | pr_err("Failed to create watchdog threads, disabled\n"); | |
230ec939 | 805 | else |
3c00ea82 | 806 | watchdog_running = 1; |
b2f57c3a UO |
807 | } else { |
808 | /* | |
809 | * Enable/disable the lockup detectors or | |
810 | * change the sample period 'on the fly'. | |
811 | */ | |
b43cb43c UO |
812 | err = update_watchdog_all_cpus(); |
813 | ||
814 | if (err) { | |
815 | watchdog_disable_all_cpus(); | |
816 | pr_err("Failed to update lockup detectors, disabled\n"); | |
817 | } | |
bcd951cf | 818 | } |
b8900bc0 | 819 | |
b43cb43c UO |
820 | if (err) |
821 | watchdog_enabled = 0; | |
822 | ||
b8900bc0 | 823 | return err; |
58687acb DZ |
824 | } |
825 | ||
826 | static void watchdog_disable_all_cpus(void) | |
827 | { | |
3c00ea82 FW |
828 | if (watchdog_running) { |
829 | watchdog_running = 0; | |
b8900bc0 | 830 | smpboot_unregister_percpu_thread(&watchdog_threads); |
bcd951cf | 831 | } |
58687acb DZ |
832 | } |
833 | ||
58cf690a UO |
834 | #ifdef CONFIG_SYSCTL |
835 | ||
58687acb | 836 | /* |
a0c9cbb9 UO |
837 | * Update the run state of the lockup detectors. |
838 | */ | |
839 | static int proc_watchdog_update(void) | |
840 | { | |
841 | int err = 0; | |
842 | ||
843 | /* | |
844 | * Watchdog threads won't be started if they are already active. | |
845 | * The 'watchdog_running' variable in watchdog_*_all_cpus() takes | |
846 | * care of this. If those threads are already active, the sample | |
847 | * period will be updated and the lockup detectors will be enabled | |
848 | * or disabled 'on the fly'. | |
849 | */ | |
850 | if (watchdog_enabled && watchdog_thresh) | |
b2f57c3a | 851 | err = watchdog_enable_all_cpus(); |
a0c9cbb9 UO |
852 | else |
853 | watchdog_disable_all_cpus(); | |
854 | ||
855 | return err; | |
856 | ||
857 | } | |
858 | ||
ef246a21 UO |
859 | /* |
860 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter | |
861 | * | |
862 | * caller | table->data points to | 'which' contains the flag(s) | |
863 | * -------------------|-----------------------|----------------------------- | |
864 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed | |
865 | * | | with SOFT_WATCHDOG_ENABLED | |
866 | * -------------------|-----------------------|----------------------------- | |
867 | * proc_nmi_watchdog | nmi_watchdog_enabled | NMI_WATCHDOG_ENABLED | |
868 | * -------------------|-----------------------|----------------------------- | |
869 | * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED | |
870 | */ | |
871 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, | |
872 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
873 | { | |
874 | int err, old, new; | |
875 | int *watchdog_param = (int *)table->data; | |
876 | ||
8614ddef | 877 | get_online_cpus(); |
ef246a21 UO |
878 | mutex_lock(&watchdog_proc_mutex); |
879 | ||
8c073d27 UO |
880 | if (watchdog_suspended) { |
881 | /* no parameter changes allowed while watchdog is suspended */ | |
882 | err = -EAGAIN; | |
883 | goto out; | |
884 | } | |
885 | ||
ef246a21 UO |
886 | /* |
887 | * If the parameter is being read return the state of the corresponding | |
888 | * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the | |
889 | * run state of the lockup detectors. | |
890 | */ | |
891 | if (!write) { | |
892 | *watchdog_param = (watchdog_enabled & which) != 0; | |
893 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
894 | } else { | |
895 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
896 | if (err) | |
897 | goto out; | |
898 | ||
899 | /* | |
900 | * There is a race window between fetching the current value | |
901 | * from 'watchdog_enabled' and storing the new value. During | |
902 | * this race window, watchdog_nmi_enable() can sneak in and | |
903 | * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'. | |
904 | * The 'cmpxchg' detects this race and the loop retries. | |
905 | */ | |
906 | do { | |
907 | old = watchdog_enabled; | |
908 | /* | |
909 | * If the parameter value is not zero set the | |
910 | * corresponding bit(s), else clear it(them). | |
911 | */ | |
912 | if (*watchdog_param) | |
913 | new = old | which; | |
914 | else | |
915 | new = old & ~which; | |
916 | } while (cmpxchg(&watchdog_enabled, old, new) != old); | |
917 | ||
918 | /* | |
b43cb43c UO |
919 | * Update the run state of the lockup detectors. There is _no_ |
920 | * need to check the value returned by proc_watchdog_update() | |
921 | * and to restore the previous value of 'watchdog_enabled' as | |
922 | * both lockup detectors are disabled if proc_watchdog_update() | |
923 | * returns an error. | |
ef246a21 | 924 | */ |
a1ee1932 JH |
925 | if (old == new) |
926 | goto out; | |
927 | ||
ef246a21 | 928 | err = proc_watchdog_update(); |
ef246a21 UO |
929 | } |
930 | out: | |
931 | mutex_unlock(&watchdog_proc_mutex); | |
8614ddef | 932 | put_online_cpus(); |
ef246a21 UO |
933 | return err; |
934 | } | |
935 | ||
83a80a39 UO |
936 | /* |
937 | * /proc/sys/kernel/watchdog | |
938 | */ | |
939 | int proc_watchdog(struct ctl_table *table, int write, | |
940 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
941 | { | |
942 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, | |
943 | table, write, buffer, lenp, ppos); | |
944 | } | |
945 | ||
946 | /* | |
947 | * /proc/sys/kernel/nmi_watchdog | |
58687acb | 948 | */ |
83a80a39 UO |
949 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
950 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
951 | { | |
952 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, | |
953 | table, write, buffer, lenp, ppos); | |
954 | } | |
955 | ||
956 | /* | |
957 | * /proc/sys/kernel/soft_watchdog | |
958 | */ | |
959 | int proc_soft_watchdog(struct ctl_table *table, int write, | |
960 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
961 | { | |
962 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, | |
963 | table, write, buffer, lenp, ppos); | |
964 | } | |
58687acb | 965 | |
83a80a39 UO |
966 | /* |
967 | * /proc/sys/kernel/watchdog_thresh | |
968 | */ | |
969 | int proc_watchdog_thresh(struct ctl_table *table, int write, | |
970 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
58687acb | 971 | { |
a1ee1932 | 972 | int err, old, new; |
58687acb | 973 | |
8614ddef | 974 | get_online_cpus(); |
359e6fab | 975 | mutex_lock(&watchdog_proc_mutex); |
bcd951cf | 976 | |
8c073d27 UO |
977 | if (watchdog_suspended) { |
978 | /* no parameter changes allowed while watchdog is suspended */ | |
979 | err = -EAGAIN; | |
980 | goto out; | |
981 | } | |
982 | ||
83a80a39 | 983 | old = ACCESS_ONCE(watchdog_thresh); |
b8900bc0 | 984 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
83a80a39 | 985 | |
b8900bc0 | 986 | if (err || !write) |
359e6fab | 987 | goto out; |
e04ab2bc | 988 | |
b66a2356 | 989 | /* |
d283c640 | 990 | * Update the sample period. Restore on failure. |
b66a2356 | 991 | */ |
a1ee1932 JH |
992 | new = ACCESS_ONCE(watchdog_thresh); |
993 | if (old == new) | |
994 | goto out; | |
995 | ||
83a80a39 UO |
996 | set_sample_period(); |
997 | err = proc_watchdog_update(); | |
d283c640 | 998 | if (err) { |
83a80a39 | 999 | watchdog_thresh = old; |
d283c640 UO |
1000 | set_sample_period(); |
1001 | } | |
359e6fab MH |
1002 | out: |
1003 | mutex_unlock(&watchdog_proc_mutex); | |
8614ddef | 1004 | put_online_cpus(); |
b8900bc0 | 1005 | return err; |
58687acb | 1006 | } |
fe4ba3c3 CM |
1007 | |
1008 | /* | |
1009 | * The cpumask is the mask of possible cpus that the watchdog can run | |
1010 | * on, not the mask of cpus it is actually running on. This allows the | |
1011 | * user to specify a mask that will include cpus that have not yet | |
1012 | * been brought online, if desired. | |
1013 | */ | |
1014 | int proc_watchdog_cpumask(struct ctl_table *table, int write, | |
1015 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1016 | { | |
1017 | int err; | |
1018 | ||
8614ddef | 1019 | get_online_cpus(); |
fe4ba3c3 | 1020 | mutex_lock(&watchdog_proc_mutex); |
8c073d27 UO |
1021 | |
1022 | if (watchdog_suspended) { | |
1023 | /* no parameter changes allowed while watchdog is suspended */ | |
1024 | err = -EAGAIN; | |
1025 | goto out; | |
1026 | } | |
1027 | ||
fe4ba3c3 CM |
1028 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
1029 | if (!err && write) { | |
1030 | /* Remove impossible cpus to keep sysctl output cleaner. */ | |
1031 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, | |
1032 | cpu_possible_mask); | |
1033 | ||
1034 | if (watchdog_running) { | |
1035 | /* | |
1036 | * Failure would be due to being unable to allocate | |
1037 | * a temporary cpumask, so we are likely not in a | |
1038 | * position to do much else to make things better. | |
1039 | */ | |
1040 | if (smpboot_update_cpumask_percpu_thread( | |
1041 | &watchdog_threads, &watchdog_cpumask) != 0) | |
1042 | pr_err("cpumask update failed\n"); | |
1043 | } | |
1044 | } | |
8c073d27 | 1045 | out: |
fe4ba3c3 | 1046 | mutex_unlock(&watchdog_proc_mutex); |
8614ddef | 1047 | put_online_cpus(); |
fe4ba3c3 CM |
1048 | return err; |
1049 | } | |
1050 | ||
58687acb DZ |
1051 | #endif /* CONFIG_SYSCTL */ |
1052 | ||
004417a6 | 1053 | void __init lockup_detector_init(void) |
58687acb | 1054 | { |
0f34c400 | 1055 | set_sample_period(); |
b8900bc0 | 1056 | |
fe4ba3c3 CM |
1057 | #ifdef CONFIG_NO_HZ_FULL |
1058 | if (tick_nohz_full_enabled()) { | |
314b08ff FW |
1059 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
1060 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); | |
fe4ba3c3 CM |
1061 | } else |
1062 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); | |
1063 | #else | |
1064 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); | |
1065 | #endif | |
1066 | ||
195daf66 | 1067 | if (watchdog_enabled) |
b2f57c3a | 1068 | watchdog_enable_all_cpus(); |
58687acb | 1069 | } |