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