]>
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(); | |
55537871 | 347 | struct pt_regs *regs = get_irq_regs(); |
26e09c6e | 348 | |
58687acb | 349 | /* only print hardlockups once */ |
909ea964 | 350 | if (__this_cpu_read(hard_watchdog_warn) == true) |
58687acb DZ |
351 | return; |
352 | ||
55537871 JK |
353 | pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu); |
354 | print_modules(); | |
355 | print_irqtrace_events(current); | |
356 | if (regs) | |
357 | show_regs(regs); | |
58687acb | 358 | else |
55537871 JK |
359 | dump_stack(); |
360 | ||
361 | /* | |
362 | * Perform all-CPU dump only once to avoid multiple hardlockups | |
363 | * generating interleaving traces | |
364 | */ | |
365 | if (sysctl_hardlockup_all_cpu_backtrace && | |
366 | !test_and_set_bit(0, &hardlockup_allcpu_dumped)) | |
367 | trigger_allbutself_cpu_backtrace(); | |
368 | ||
369 | if (hardlockup_panic) | |
58c5661f | 370 | nmi_panic(regs, "Hard LOCKUP"); |
58687acb | 371 | |
909ea964 | 372 | __this_cpu_write(hard_watchdog_warn, true); |
58687acb DZ |
373 | return; |
374 | } | |
375 | ||
909ea964 | 376 | __this_cpu_write(hard_watchdog_warn, false); |
58687acb DZ |
377 | return; |
378 | } | |
bcd951cf TG |
379 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
380 | ||
58687acb DZ |
381 | static void watchdog_interrupt_count(void) |
382 | { | |
909ea964 | 383 | __this_cpu_inc(hrtimer_interrupts); |
58687acb | 384 | } |
bcd951cf TG |
385 | |
386 | static int watchdog_nmi_enable(unsigned int cpu); | |
387 | static void watchdog_nmi_disable(unsigned int cpu); | |
58687acb | 388 | |
58cf690a UO |
389 | static int watchdog_enable_all_cpus(void); |
390 | static void watchdog_disable_all_cpus(void); | |
391 | ||
58687acb DZ |
392 | /* watchdog kicker functions */ |
393 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) | |
394 | { | |
909ea964 | 395 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
58687acb DZ |
396 | struct pt_regs *regs = get_irq_regs(); |
397 | int duration; | |
ed235875 | 398 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
58687acb DZ |
399 | |
400 | /* kick the hardlockup detector */ | |
401 | watchdog_interrupt_count(); | |
402 | ||
403 | /* kick the softlockup detector */ | |
909ea964 | 404 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
58687acb DZ |
405 | |
406 | /* .. and repeat */ | |
0f34c400 | 407 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
58687acb DZ |
408 | |
409 | if (touch_ts == 0) { | |
909ea964 | 410 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
58687acb DZ |
411 | /* |
412 | * If the time stamp was touched atomically | |
413 | * make sure the scheduler tick is up to date. | |
414 | */ | |
909ea964 | 415 | __this_cpu_write(softlockup_touch_sync, false); |
58687acb DZ |
416 | sched_clock_tick(); |
417 | } | |
5d1c0f4a EM |
418 | |
419 | /* Clear the guest paused flag on watchdog reset */ | |
420 | kvm_check_and_clear_guest_paused(); | |
58687acb DZ |
421 | __touch_watchdog(); |
422 | return HRTIMER_RESTART; | |
423 | } | |
424 | ||
425 | /* check for a softlockup | |
426 | * This is done by making sure a high priority task is | |
427 | * being scheduled. The task touches the watchdog to | |
428 | * indicate it is getting cpu time. If it hasn't then | |
429 | * this is a good indication some task is hogging the cpu | |
430 | */ | |
26e09c6e | 431 | duration = is_softlockup(touch_ts); |
58687acb | 432 | if (unlikely(duration)) { |
5d1c0f4a EM |
433 | /* |
434 | * If a virtual machine is stopped by the host it can look to | |
435 | * the watchdog like a soft lockup, check to see if the host | |
436 | * stopped the vm before we issue the warning | |
437 | */ | |
438 | if (kvm_check_and_clear_guest_paused()) | |
439 | return HRTIMER_RESTART; | |
440 | ||
58687acb | 441 | /* only warn once */ |
b1a8de1f | 442 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
443 | /* | |
444 | * When multiple processes are causing softlockups the | |
445 | * softlockup detector only warns on the first one | |
446 | * because the code relies on a full quiet cycle to | |
447 | * re-arm. The second process prevents the quiet cycle | |
448 | * and never gets reported. Use task pointers to detect | |
449 | * this. | |
450 | */ | |
451 | if (__this_cpu_read(softlockup_task_ptr_saved) != | |
452 | current) { | |
453 | __this_cpu_write(soft_watchdog_warn, false); | |
454 | __touch_watchdog(); | |
455 | } | |
58687acb | 456 | return HRTIMER_RESTART; |
b1a8de1f | 457 | } |
58687acb | 458 | |
ed235875 AT |
459 | if (softlockup_all_cpu_backtrace) { |
460 | /* Prevent multiple soft-lockup reports if one cpu is already | |
461 | * engaged in dumping cpu back traces | |
462 | */ | |
463 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { | |
464 | /* Someone else will report us. Let's give up */ | |
465 | __this_cpu_write(soft_watchdog_warn, true); | |
466 | return HRTIMER_RESTART; | |
467 | } | |
468 | } | |
469 | ||
656c3b79 | 470 | pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
26e09c6e | 471 | smp_processor_id(), duration, |
58687acb | 472 | current->comm, task_pid_nr(current)); |
b1a8de1f | 473 | __this_cpu_write(softlockup_task_ptr_saved, current); |
58687acb DZ |
474 | print_modules(); |
475 | print_irqtrace_events(current); | |
476 | if (regs) | |
477 | show_regs(regs); | |
478 | else | |
479 | dump_stack(); | |
480 | ||
ed235875 AT |
481 | if (softlockup_all_cpu_backtrace) { |
482 | /* Avoid generating two back traces for current | |
483 | * given that one is already made above | |
484 | */ | |
485 | trigger_allbutself_cpu_backtrace(); | |
486 | ||
487 | clear_bit(0, &soft_lockup_nmi_warn); | |
488 | /* Barrier to sync with other cpus */ | |
489 | smp_mb__after_atomic(); | |
490 | } | |
491 | ||
69361eef | 492 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
58687acb DZ |
493 | if (softlockup_panic) |
494 | panic("softlockup: hung tasks"); | |
909ea964 | 495 | __this_cpu_write(soft_watchdog_warn, true); |
58687acb | 496 | } else |
909ea964 | 497 | __this_cpu_write(soft_watchdog_warn, false); |
58687acb DZ |
498 | |
499 | return HRTIMER_RESTART; | |
500 | } | |
501 | ||
bcd951cf TG |
502 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
503 | { | |
504 | struct sched_param param = { .sched_priority = prio }; | |
58687acb | 505 | |
bcd951cf TG |
506 | sched_setscheduler(current, policy, ¶m); |
507 | } | |
508 | ||
509 | static void watchdog_enable(unsigned int cpu) | |
58687acb | 510 | { |
f7f66b05 | 511 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
58687acb | 512 | |
3935e895 BM |
513 | /* kick off the timer for the hardlockup detector */ |
514 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
515 | hrtimer->function = watchdog_timer_fn; | |
516 | ||
bcd951cf TG |
517 | /* Enable the perf event */ |
518 | watchdog_nmi_enable(cpu); | |
58687acb | 519 | |
58687acb | 520 | /* done here because hrtimer_start can only pin to smp_processor_id() */ |
0f34c400 | 521 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
58687acb DZ |
522 | HRTIMER_MODE_REL_PINNED); |
523 | ||
bcd951cf TG |
524 | /* initialize timestamp */ |
525 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); | |
526 | __touch_watchdog(); | |
527 | } | |
58687acb | 528 | |
bcd951cf TG |
529 | static void watchdog_disable(unsigned int cpu) |
530 | { | |
f7f66b05 | 531 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
58687acb | 532 | |
bcd951cf TG |
533 | watchdog_set_prio(SCHED_NORMAL, 0); |
534 | hrtimer_cancel(hrtimer); | |
535 | /* disable the perf event */ | |
536 | watchdog_nmi_disable(cpu); | |
58687acb DZ |
537 | } |
538 | ||
b8900bc0 FW |
539 | static void watchdog_cleanup(unsigned int cpu, bool online) |
540 | { | |
541 | watchdog_disable(cpu); | |
542 | } | |
543 | ||
bcd951cf TG |
544 | static int watchdog_should_run(unsigned int cpu) |
545 | { | |
546 | return __this_cpu_read(hrtimer_interrupts) != | |
547 | __this_cpu_read(soft_lockup_hrtimer_cnt); | |
548 | } | |
549 | ||
550 | /* | |
551 | * The watchdog thread function - touches the timestamp. | |
552 | * | |
0f34c400 | 553 | * It only runs once every sample_period seconds (4 seconds by |
bcd951cf TG |
554 | * default) to reset the softlockup timestamp. If this gets delayed |
555 | * for more than 2*watchdog_thresh seconds then the debug-printout | |
556 | * triggers in watchdog_timer_fn(). | |
557 | */ | |
558 | static void watchdog(unsigned int cpu) | |
559 | { | |
560 | __this_cpu_write(soft_lockup_hrtimer_cnt, | |
561 | __this_cpu_read(hrtimer_interrupts)); | |
562 | __touch_watchdog(); | |
bcfba4f4 UO |
563 | |
564 | /* | |
565 | * watchdog_nmi_enable() clears the NMI_WATCHDOG_ENABLED bit in the | |
566 | * failure path. Check for failures that can occur asynchronously - | |
567 | * for example, when CPUs are on-lined - and shut down the hardware | |
568 | * perf event on each CPU accordingly. | |
569 | * | |
570 | * The only non-obvious place this bit can be cleared is through | |
571 | * watchdog_nmi_enable(), so a pr_info() is placed there. Placing a | |
572 | * pr_info here would be too noisy as it would result in a message | |
573 | * every few seconds if the hardlockup was disabled but the softlockup | |
574 | * enabled. | |
575 | */ | |
576 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) | |
577 | watchdog_nmi_disable(cpu); | |
bcd951cf | 578 | } |
58687acb | 579 | |
23637d47 | 580 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
a7027046 DZ |
581 | /* |
582 | * People like the simple clean cpu node info on boot. | |
583 | * Reduce the watchdog noise by only printing messages | |
584 | * that are different from what cpu0 displayed. | |
585 | */ | |
586 | static unsigned long cpu0_err; | |
587 | ||
bcd951cf | 588 | static int watchdog_nmi_enable(unsigned int cpu) |
58687acb DZ |
589 | { |
590 | struct perf_event_attr *wd_attr; | |
591 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
592 | ||
195daf66 UO |
593 | /* nothing to do if the hard lockup detector is disabled */ |
594 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) | |
595 | goto out; | |
6e7458a6 | 596 | |
58687acb DZ |
597 | /* is it already setup and enabled? */ |
598 | if (event && event->state > PERF_EVENT_STATE_OFF) | |
599 | goto out; | |
600 | ||
601 | /* it is setup but not enabled */ | |
602 | if (event != NULL) | |
603 | goto out_enable; | |
604 | ||
58687acb | 605 | wd_attr = &wd_hw_attr; |
4eec42f3 | 606 | wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh); |
1880c4ae CG |
607 | |
608 | /* Try to register using hardware perf events */ | |
4dc0da86 | 609 | event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL); |
a7027046 DZ |
610 | |
611 | /* save cpu0 error for future comparision */ | |
612 | if (cpu == 0 && IS_ERR(event)) | |
613 | cpu0_err = PTR_ERR(event); | |
614 | ||
58687acb | 615 | if (!IS_ERR(event)) { |
a7027046 DZ |
616 | /* only print for cpu0 or different than cpu0 */ |
617 | if (cpu == 0 || cpu0_err) | |
618 | pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n"); | |
58687acb DZ |
619 | goto out_save; |
620 | } | |
621 | ||
bcfba4f4 UO |
622 | /* |
623 | * Disable the hard lockup detector if _any_ CPU fails to set up | |
624 | * set up the hardware perf event. The watchdog() function checks | |
625 | * the NMI_WATCHDOG_ENABLED bit periodically. | |
626 | * | |
627 | * The barriers are for syncing up watchdog_enabled across all the | |
628 | * cpus, as clear_bit() does not use barriers. | |
629 | */ | |
630 | smp_mb__before_atomic(); | |
631 | clear_bit(NMI_WATCHDOG_ENABLED_BIT, &watchdog_enabled); | |
632 | smp_mb__after_atomic(); | |
633 | ||
a7027046 DZ |
634 | /* skip displaying the same error again */ |
635 | if (cpu > 0 && (PTR_ERR(event) == cpu0_err)) | |
636 | return PTR_ERR(event); | |
5651f7f4 DZ |
637 | |
638 | /* vary the KERN level based on the returned errno */ | |
639 | if (PTR_ERR(event) == -EOPNOTSUPP) | |
4501980a | 640 | pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu); |
5651f7f4 | 641 | else if (PTR_ERR(event) == -ENOENT) |
656c3b79 | 642 | pr_warn("disabled (cpu%i): hardware events not enabled\n", |
4501980a | 643 | cpu); |
5651f7f4 | 644 | else |
4501980a AM |
645 | pr_err("disabled (cpu%i): unable to create perf event: %ld\n", |
646 | cpu, PTR_ERR(event)); | |
bcfba4f4 UO |
647 | |
648 | pr_info("Shutting down hard lockup detector on all cpus\n"); | |
649 | ||
eac24335 | 650 | return PTR_ERR(event); |
58687acb DZ |
651 | |
652 | /* success path */ | |
653 | out_save: | |
654 | per_cpu(watchdog_ev, cpu) = event; | |
655 | out_enable: | |
656 | perf_event_enable(per_cpu(watchdog_ev, cpu)); | |
657 | out: | |
658 | return 0; | |
659 | } | |
660 | ||
bcd951cf | 661 | static void watchdog_nmi_disable(unsigned int cpu) |
58687acb DZ |
662 | { |
663 | struct perf_event *event = per_cpu(watchdog_ev, cpu); | |
664 | ||
665 | if (event) { | |
666 | perf_event_disable(event); | |
667 | per_cpu(watchdog_ev, cpu) = NULL; | |
668 | ||
669 | /* should be in cleanup, but blocks oprofile */ | |
670 | perf_event_release_kernel(event); | |
671 | } | |
df577149 UO |
672 | if (cpu == 0) { |
673 | /* watchdog_nmi_enable() expects this to be zero initially. */ | |
674 | cpu0_err = 0; | |
675 | } | |
58687acb | 676 | } |
b3738d29 | 677 | |
58687acb | 678 | #else |
bcd951cf TG |
679 | static int watchdog_nmi_enable(unsigned int cpu) { return 0; } |
680 | static void watchdog_nmi_disable(unsigned int cpu) { return; } | |
23637d47 | 681 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
58687acb | 682 | |
b8900bc0 FW |
683 | static struct smp_hotplug_thread watchdog_threads = { |
684 | .store = &softlockup_watchdog, | |
685 | .thread_should_run = watchdog_should_run, | |
686 | .thread_fn = watchdog, | |
687 | .thread_comm = "watchdog/%u", | |
688 | .setup = watchdog_enable, | |
689 | .cleanup = watchdog_cleanup, | |
690 | .park = watchdog_disable, | |
691 | .unpark = watchdog_enable, | |
692 | }; | |
693 | ||
81a4beef UO |
694 | /* |
695 | * park all watchdog threads that are specified in 'watchdog_cpumask' | |
ee7fed54 UO |
696 | * |
697 | * This function returns an error if kthread_park() of a watchdog thread | |
698 | * fails. In this situation, the watchdog threads of some CPUs can already | |
699 | * be parked and the watchdog threads of other CPUs can still be runnable. | |
700 | * Callers are expected to handle this special condition as appropriate in | |
701 | * their context. | |
a2a45b85 UO |
702 | * |
703 | * This function may only be called in a context that is protected against | |
704 | * races with CPU hotplug - for example, via get_online_cpus(). | |
81a4beef UO |
705 | */ |
706 | static int watchdog_park_threads(void) | |
707 | { | |
708 | int cpu, ret = 0; | |
709 | ||
81a4beef UO |
710 | for_each_watchdog_cpu(cpu) { |
711 | ret = kthread_park(per_cpu(softlockup_watchdog, cpu)); | |
712 | if (ret) | |
713 | break; | |
714 | } | |
81a4beef UO |
715 | |
716 | return ret; | |
717 | } | |
718 | ||
719 | /* | |
720 | * unpark all watchdog threads that are specified in 'watchdog_cpumask' | |
a2a45b85 UO |
721 | * |
722 | * This function may only be called in a context that is protected against | |
723 | * races with CPU hotplug - for example, via get_online_cpus(). | |
81a4beef UO |
724 | */ |
725 | static void watchdog_unpark_threads(void) | |
726 | { | |
727 | int cpu; | |
728 | ||
81a4beef UO |
729 | for_each_watchdog_cpu(cpu) |
730 | kthread_unpark(per_cpu(softlockup_watchdog, cpu)); | |
81a4beef UO |
731 | } |
732 | ||
8c073d27 UO |
733 | /* |
734 | * Suspend the hard and soft lockup detector by parking the watchdog threads. | |
735 | */ | |
ec6a9066 | 736 | int lockup_detector_suspend(void) |
8c073d27 UO |
737 | { |
738 | int ret = 0; | |
739 | ||
ee89e71e | 740 | get_online_cpus(); |
8c073d27 UO |
741 | mutex_lock(&watchdog_proc_mutex); |
742 | /* | |
743 | * Multiple suspend requests can be active in parallel (counted by | |
744 | * the 'watchdog_suspended' variable). If the watchdog threads are | |
745 | * running, the first caller takes care that they will be parked. | |
746 | * The state of 'watchdog_running' cannot change while a suspend | |
ec6a9066 | 747 | * request is active (see related code in 'proc' handlers). |
8c073d27 UO |
748 | */ |
749 | if (watchdog_running && !watchdog_suspended) | |
750 | ret = watchdog_park_threads(); | |
751 | ||
752 | if (ret == 0) | |
753 | watchdog_suspended++; | |
c993590c UO |
754 | else { |
755 | watchdog_disable_all_cpus(); | |
756 | pr_err("Failed to suspend lockup detectors, disabled\n"); | |
757 | watchdog_enabled = 0; | |
758 | } | |
8c073d27 UO |
759 | |
760 | mutex_unlock(&watchdog_proc_mutex); | |
761 | ||
762 | return ret; | |
763 | } | |
764 | ||
765 | /* | |
766 | * Resume the hard and soft lockup detector by unparking the watchdog threads. | |
767 | */ | |
ec6a9066 | 768 | void lockup_detector_resume(void) |
8c073d27 UO |
769 | { |
770 | mutex_lock(&watchdog_proc_mutex); | |
771 | ||
772 | watchdog_suspended--; | |
773 | /* | |
774 | * The watchdog threads are unparked if they were previously running | |
775 | * and if there is no more active suspend request. | |
776 | */ | |
777 | if (watchdog_running && !watchdog_suspended) | |
778 | watchdog_unpark_threads(); | |
779 | ||
780 | mutex_unlock(&watchdog_proc_mutex); | |
ee89e71e | 781 | put_online_cpus(); |
8c073d27 UO |
782 | } |
783 | ||
b43cb43c | 784 | static int update_watchdog_all_cpus(void) |
9809b18f | 785 | { |
b43cb43c UO |
786 | int ret; |
787 | ||
788 | ret = watchdog_park_threads(); | |
789 | if (ret) | |
790 | return ret; | |
791 | ||
d4bdd0b2 | 792 | watchdog_unpark_threads(); |
b43cb43c UO |
793 | |
794 | return 0; | |
9809b18f MH |
795 | } |
796 | ||
b2f57c3a | 797 | static int watchdog_enable_all_cpus(void) |
58687acb | 798 | { |
b8900bc0 | 799 | int err = 0; |
58687acb | 800 | |
3c00ea82 | 801 | if (!watchdog_running) { |
230ec939 FW |
802 | err = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
803 | &watchdog_cpumask); | |
b8900bc0 FW |
804 | if (err) |
805 | pr_err("Failed to create watchdog threads, disabled\n"); | |
230ec939 | 806 | else |
3c00ea82 | 807 | watchdog_running = 1; |
b2f57c3a UO |
808 | } else { |
809 | /* | |
810 | * Enable/disable the lockup detectors or | |
811 | * change the sample period 'on the fly'. | |
812 | */ | |
b43cb43c UO |
813 | err = update_watchdog_all_cpus(); |
814 | ||
815 | if (err) { | |
816 | watchdog_disable_all_cpus(); | |
817 | pr_err("Failed to update lockup detectors, disabled\n"); | |
818 | } | |
bcd951cf | 819 | } |
b8900bc0 | 820 | |
b43cb43c UO |
821 | if (err) |
822 | watchdog_enabled = 0; | |
823 | ||
b8900bc0 | 824 | return err; |
58687acb DZ |
825 | } |
826 | ||
827 | static void watchdog_disable_all_cpus(void) | |
828 | { | |
3c00ea82 FW |
829 | if (watchdog_running) { |
830 | watchdog_running = 0; | |
b8900bc0 | 831 | smpboot_unregister_percpu_thread(&watchdog_threads); |
bcd951cf | 832 | } |
58687acb DZ |
833 | } |
834 | ||
58cf690a UO |
835 | #ifdef CONFIG_SYSCTL |
836 | ||
58687acb | 837 | /* |
a0c9cbb9 UO |
838 | * Update the run state of the lockup detectors. |
839 | */ | |
840 | static int proc_watchdog_update(void) | |
841 | { | |
842 | int err = 0; | |
843 | ||
844 | /* | |
845 | * Watchdog threads won't be started if they are already active. | |
846 | * The 'watchdog_running' variable in watchdog_*_all_cpus() takes | |
847 | * care of this. If those threads are already active, the sample | |
848 | * period will be updated and the lockup detectors will be enabled | |
849 | * or disabled 'on the fly'. | |
850 | */ | |
851 | if (watchdog_enabled && watchdog_thresh) | |
b2f57c3a | 852 | err = watchdog_enable_all_cpus(); |
a0c9cbb9 UO |
853 | else |
854 | watchdog_disable_all_cpus(); | |
855 | ||
856 | return err; | |
857 | ||
858 | } | |
859 | ||
ef246a21 UO |
860 | /* |
861 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter | |
862 | * | |
863 | * caller | table->data points to | 'which' contains the flag(s) | |
864 | * -------------------|-----------------------|----------------------------- | |
865 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed | |
866 | * | | with SOFT_WATCHDOG_ENABLED | |
867 | * -------------------|-----------------------|----------------------------- | |
868 | * proc_nmi_watchdog | nmi_watchdog_enabled | NMI_WATCHDOG_ENABLED | |
869 | * -------------------|-----------------------|----------------------------- | |
870 | * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED | |
871 | */ | |
872 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, | |
873 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
874 | { | |
875 | int err, old, new; | |
876 | int *watchdog_param = (int *)table->data; | |
877 | ||
8614ddef | 878 | get_online_cpus(); |
ef246a21 UO |
879 | mutex_lock(&watchdog_proc_mutex); |
880 | ||
8c073d27 UO |
881 | if (watchdog_suspended) { |
882 | /* no parameter changes allowed while watchdog is suspended */ | |
883 | err = -EAGAIN; | |
884 | goto out; | |
885 | } | |
886 | ||
ef246a21 UO |
887 | /* |
888 | * If the parameter is being read return the state of the corresponding | |
889 | * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the | |
890 | * run state of the lockup detectors. | |
891 | */ | |
892 | if (!write) { | |
893 | *watchdog_param = (watchdog_enabled & which) != 0; | |
894 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
895 | } else { | |
896 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
897 | if (err) | |
898 | goto out; | |
899 | ||
900 | /* | |
901 | * There is a race window between fetching the current value | |
902 | * from 'watchdog_enabled' and storing the new value. During | |
903 | * this race window, watchdog_nmi_enable() can sneak in and | |
904 | * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'. | |
905 | * The 'cmpxchg' detects this race and the loop retries. | |
906 | */ | |
907 | do { | |
908 | old = watchdog_enabled; | |
909 | /* | |
910 | * If the parameter value is not zero set the | |
911 | * corresponding bit(s), else clear it(them). | |
912 | */ | |
913 | if (*watchdog_param) | |
914 | new = old | which; | |
915 | else | |
916 | new = old & ~which; | |
917 | } while (cmpxchg(&watchdog_enabled, old, new) != old); | |
918 | ||
919 | /* | |
b43cb43c UO |
920 | * Update the run state of the lockup detectors. There is _no_ |
921 | * need to check the value returned by proc_watchdog_update() | |
922 | * and to restore the previous value of 'watchdog_enabled' as | |
923 | * both lockup detectors are disabled if proc_watchdog_update() | |
924 | * returns an error. | |
ef246a21 UO |
925 | */ |
926 | err = proc_watchdog_update(); | |
ef246a21 UO |
927 | } |
928 | out: | |
929 | mutex_unlock(&watchdog_proc_mutex); | |
8614ddef | 930 | put_online_cpus(); |
ef246a21 UO |
931 | return err; |
932 | } | |
933 | ||
83a80a39 UO |
934 | /* |
935 | * /proc/sys/kernel/watchdog | |
936 | */ | |
937 | int proc_watchdog(struct ctl_table *table, int write, | |
938 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
939 | { | |
940 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, | |
941 | table, write, buffer, lenp, ppos); | |
942 | } | |
943 | ||
944 | /* | |
945 | * /proc/sys/kernel/nmi_watchdog | |
58687acb | 946 | */ |
83a80a39 UO |
947 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
948 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
949 | { | |
950 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, | |
951 | table, write, buffer, lenp, ppos); | |
952 | } | |
953 | ||
954 | /* | |
955 | * /proc/sys/kernel/soft_watchdog | |
956 | */ | |
957 | int proc_soft_watchdog(struct ctl_table *table, int write, | |
958 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
959 | { | |
960 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, | |
961 | table, write, buffer, lenp, ppos); | |
962 | } | |
58687acb | 963 | |
83a80a39 UO |
964 | /* |
965 | * /proc/sys/kernel/watchdog_thresh | |
966 | */ | |
967 | int proc_watchdog_thresh(struct ctl_table *table, int write, | |
968 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
58687acb | 969 | { |
83a80a39 | 970 | int err, old; |
58687acb | 971 | |
8614ddef | 972 | get_online_cpus(); |
359e6fab | 973 | mutex_lock(&watchdog_proc_mutex); |
bcd951cf | 974 | |
8c073d27 UO |
975 | if (watchdog_suspended) { |
976 | /* no parameter changes allowed while watchdog is suspended */ | |
977 | err = -EAGAIN; | |
978 | goto out; | |
979 | } | |
980 | ||
83a80a39 | 981 | old = ACCESS_ONCE(watchdog_thresh); |
b8900bc0 | 982 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
83a80a39 | 983 | |
b8900bc0 | 984 | if (err || !write) |
359e6fab | 985 | goto out; |
e04ab2bc | 986 | |
b66a2356 | 987 | /* |
d283c640 | 988 | * Update the sample period. Restore on failure. |
b66a2356 | 989 | */ |
83a80a39 UO |
990 | set_sample_period(); |
991 | err = proc_watchdog_update(); | |
d283c640 | 992 | if (err) { |
83a80a39 | 993 | watchdog_thresh = old; |
d283c640 UO |
994 | set_sample_period(); |
995 | } | |
359e6fab MH |
996 | out: |
997 | mutex_unlock(&watchdog_proc_mutex); | |
8614ddef | 998 | put_online_cpus(); |
b8900bc0 | 999 | return err; |
58687acb | 1000 | } |
fe4ba3c3 CM |
1001 | |
1002 | /* | |
1003 | * The cpumask is the mask of possible cpus that the watchdog can run | |
1004 | * on, not the mask of cpus it is actually running on. This allows the | |
1005 | * user to specify a mask that will include cpus that have not yet | |
1006 | * been brought online, if desired. | |
1007 | */ | |
1008 | int proc_watchdog_cpumask(struct ctl_table *table, int write, | |
1009 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1010 | { | |
1011 | int err; | |
1012 | ||
8614ddef | 1013 | get_online_cpus(); |
fe4ba3c3 | 1014 | mutex_lock(&watchdog_proc_mutex); |
8c073d27 UO |
1015 | |
1016 | if (watchdog_suspended) { | |
1017 | /* no parameter changes allowed while watchdog is suspended */ | |
1018 | err = -EAGAIN; | |
1019 | goto out; | |
1020 | } | |
1021 | ||
fe4ba3c3 CM |
1022 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
1023 | if (!err && write) { | |
1024 | /* Remove impossible cpus to keep sysctl output cleaner. */ | |
1025 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, | |
1026 | cpu_possible_mask); | |
1027 | ||
1028 | if (watchdog_running) { | |
1029 | /* | |
1030 | * Failure would be due to being unable to allocate | |
1031 | * a temporary cpumask, so we are likely not in a | |
1032 | * position to do much else to make things better. | |
1033 | */ | |
1034 | if (smpboot_update_cpumask_percpu_thread( | |
1035 | &watchdog_threads, &watchdog_cpumask) != 0) | |
1036 | pr_err("cpumask update failed\n"); | |
1037 | } | |
1038 | } | |
8c073d27 | 1039 | out: |
fe4ba3c3 | 1040 | mutex_unlock(&watchdog_proc_mutex); |
8614ddef | 1041 | put_online_cpus(); |
fe4ba3c3 CM |
1042 | return err; |
1043 | } | |
1044 | ||
58687acb DZ |
1045 | #endif /* CONFIG_SYSCTL */ |
1046 | ||
004417a6 | 1047 | void __init lockup_detector_init(void) |
58687acb | 1048 | { |
0f34c400 | 1049 | set_sample_period(); |
b8900bc0 | 1050 | |
fe4ba3c3 CM |
1051 | #ifdef CONFIG_NO_HZ_FULL |
1052 | if (tick_nohz_full_enabled()) { | |
314b08ff FW |
1053 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
1054 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); | |
fe4ba3c3 CM |
1055 | } else |
1056 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); | |
1057 | #else | |
1058 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); | |
1059 | #endif | |
1060 | ||
195daf66 | 1061 | if (watchdog_enabled) |
b2f57c3a | 1062 | watchdog_enable_all_cpus(); |
58687acb | 1063 | } |