Root/
Source at commit be977234bfb4a6dca8a39e7c52165e4cd536ad71 created 12 years 9 months ago. By Lars-Peter Clausen, jz4740: Fix compile error | |
---|---|
1 | /* |
2 | * Detect hard and soft lockups on a system |
3 | * |
4 | * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. |
5 | * |
6 | * this code detects hard lockups: incidents in where on a CPU |
7 | * the kernel does not respond to anything except NMI. |
8 | * |
9 | * Note: Most of this code is borrowed heavily from softlockup.c, |
10 | * so thanks to Ingo for the initial implementation. |
11 | * Some chunks also taken from arch/x86/kernel/apic/nmi.c, thanks |
12 | * to those contributors as well. |
13 | */ |
14 | |
15 | #include <linux/mm.h> |
16 | #include <linux/cpu.h> |
17 | #include <linux/nmi.h> |
18 | #include <linux/init.h> |
19 | #include <linux/delay.h> |
20 | #include <linux/freezer.h> |
21 | #include <linux/kthread.h> |
22 | #include <linux/lockdep.h> |
23 | #include <linux/notifier.h> |
24 | #include <linux/module.h> |
25 | #include <linux/sysctl.h> |
26 | |
27 | #include <asm/irq_regs.h> |
28 | #include <linux/perf_event.h> |
29 | |
30 | int watchdog_enabled = 1; |
31 | int __read_mostly softlockup_thresh = 60; |
32 | |
33 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
34 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); |
35 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
36 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); |
37 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
38 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
39 | static DEFINE_PER_CPU(bool, hard_watchdog_warn); |
40 | static DEFINE_PER_CPU(bool, watchdog_nmi_touch); |
41 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
42 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
43 | static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); |
44 | #endif |
45 | |
46 | /* boot commands */ |
47 | /* |
48 | * Should we panic when a soft-lockup or hard-lockup occurs: |
49 | */ |
50 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
51 | static int hardlockup_panic = |
52 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; |
53 | |
54 | static int __init hardlockup_panic_setup(char *str) |
55 | { |
56 | if (!strncmp(str, "panic", 5)) |
57 | hardlockup_panic = 1; |
58 | else if (!strncmp(str, "nopanic", 7)) |
59 | hardlockup_panic = 0; |
60 | else if (!strncmp(str, "0", 1)) |
61 | watchdog_enabled = 0; |
62 | return 1; |
63 | } |
64 | __setup("nmi_watchdog=", hardlockup_panic_setup); |
65 | #endif |
66 | |
67 | unsigned int __read_mostly softlockup_panic = |
68 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; |
69 | |
70 | static int __init softlockup_panic_setup(char *str) |
71 | { |
72 | softlockup_panic = simple_strtoul(str, NULL, 0); |
73 | |
74 | return 1; |
75 | } |
76 | __setup("softlockup_panic=", softlockup_panic_setup); |
77 | |
78 | static int __init nowatchdog_setup(char *str) |
79 | { |
80 | watchdog_enabled = 0; |
81 | return 1; |
82 | } |
83 | __setup("nowatchdog", nowatchdog_setup); |
84 | |
85 | /* deprecated */ |
86 | static int __init nosoftlockup_setup(char *str) |
87 | { |
88 | watchdog_enabled = 0; |
89 | return 1; |
90 | } |
91 | __setup("nosoftlockup", nosoftlockup_setup); |
92 | /* */ |
93 | |
94 | |
95 | /* |
96 | * Returns seconds, approximately. We don't need nanosecond |
97 | * resolution, and we don't need to waste time with a big divide when |
98 | * 2^30ns == 1.074s. |
99 | */ |
100 | static unsigned long get_timestamp(int this_cpu) |
101 | { |
102 | return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ |
103 | } |
104 | |
105 | static unsigned long get_sample_period(void) |
106 | { |
107 | /* |
108 | * convert softlockup_thresh from seconds to ns |
109 | * the divide by 5 is to give hrtimer 5 chances to |
110 | * increment before the hardlockup detector generates |
111 | * a warning |
112 | */ |
113 | return softlockup_thresh / 5 * NSEC_PER_SEC; |
114 | } |
115 | |
116 | /* Commands for resetting the watchdog */ |
117 | static void __touch_watchdog(void) |
118 | { |
119 | int this_cpu = smp_processor_id(); |
120 | |
121 | __this_cpu_write(watchdog_touch_ts, get_timestamp(this_cpu)); |
122 | } |
123 | |
124 | void touch_softlockup_watchdog(void) |
125 | { |
126 | __this_cpu_write(watchdog_touch_ts, 0); |
127 | } |
128 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
129 | |
130 | void touch_all_softlockup_watchdogs(void) |
131 | { |
132 | int cpu; |
133 | |
134 | /* |
135 | * this is done lockless |
136 | * do we care if a 0 races with a timestamp? |
137 | * all it means is the softlock check starts one cycle later |
138 | */ |
139 | for_each_online_cpu(cpu) |
140 | per_cpu(watchdog_touch_ts, cpu) = 0; |
141 | } |
142 | |
143 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
144 | void touch_nmi_watchdog(void) |
145 | { |
146 | if (watchdog_enabled) { |
147 | unsigned cpu; |
148 | |
149 | for_each_present_cpu(cpu) { |
150 | if (per_cpu(watchdog_nmi_touch, cpu) != true) |
151 | per_cpu(watchdog_nmi_touch, cpu) = true; |
152 | } |
153 | } |
154 | touch_softlockup_watchdog(); |
155 | } |
156 | EXPORT_SYMBOL(touch_nmi_watchdog); |
157 | |
158 | #endif |
159 | |
160 | void touch_softlockup_watchdog_sync(void) |
161 | { |
162 | __raw_get_cpu_var(softlockup_touch_sync) = true; |
163 | __raw_get_cpu_var(watchdog_touch_ts) = 0; |
164 | } |
165 | |
166 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
167 | /* watchdog detector functions */ |
168 | static int is_hardlockup(void) |
169 | { |
170 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
171 | |
172 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
173 | return 1; |
174 | |
175 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
176 | return 0; |
177 | } |
178 | #endif |
179 | |
180 | static int is_softlockup(unsigned long touch_ts) |
181 | { |
182 | unsigned long now = get_timestamp(smp_processor_id()); |
183 | |
184 | /* Warn about unreasonable delays: */ |
185 | if (time_after(now, touch_ts + softlockup_thresh)) |
186 | return now - touch_ts; |
187 | |
188 | return 0; |
189 | } |
190 | |
191 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
192 | static struct perf_event_attr wd_hw_attr = { |
193 | .type = PERF_TYPE_HARDWARE, |
194 | .config = PERF_COUNT_HW_CPU_CYCLES, |
195 | .size = sizeof(struct perf_event_attr), |
196 | .pinned = 1, |
197 | .disabled = 1, |
198 | }; |
199 | |
200 | /* Callback function for perf event subsystem */ |
201 | static void watchdog_overflow_callback(struct perf_event *event, int nmi, |
202 | struct perf_sample_data *data, |
203 | struct pt_regs *regs) |
204 | { |
205 | /* Ensure the watchdog never gets throttled */ |
206 | event->hw.interrupts = 0; |
207 | |
208 | if (__this_cpu_read(watchdog_nmi_touch) == true) { |
209 | __this_cpu_write(watchdog_nmi_touch, false); |
210 | return; |
211 | } |
212 | |
213 | /* check for a hardlockup |
214 | * This is done by making sure our timer interrupt |
215 | * is incrementing. The timer interrupt should have |
216 | * fired multiple times before we overflow'd. If it hasn't |
217 | * then this is a good indication the cpu is stuck |
218 | */ |
219 | if (is_hardlockup()) { |
220 | int this_cpu = smp_processor_id(); |
221 | |
222 | /* only print hardlockups once */ |
223 | if (__this_cpu_read(hard_watchdog_warn) == true) |
224 | return; |
225 | |
226 | if (hardlockup_panic) |
227 | panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu); |
228 | else |
229 | WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu); |
230 | |
231 | __this_cpu_write(hard_watchdog_warn, true); |
232 | return; |
233 | } |
234 | |
235 | __this_cpu_write(hard_watchdog_warn, false); |
236 | return; |
237 | } |
238 | static void watchdog_interrupt_count(void) |
239 | { |
240 | __this_cpu_inc(hrtimer_interrupts); |
241 | } |
242 | #else |
243 | static inline void watchdog_interrupt_count(void) { return; } |
244 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
245 | |
246 | /* watchdog kicker functions */ |
247 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) |
248 | { |
249 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
250 | struct pt_regs *regs = get_irq_regs(); |
251 | int duration; |
252 | |
253 | /* kick the hardlockup detector */ |
254 | watchdog_interrupt_count(); |
255 | |
256 | /* kick the softlockup detector */ |
257 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
258 | |
259 | /* .. and repeat */ |
260 | hrtimer_forward_now(hrtimer, ns_to_ktime(get_sample_period())); |
261 | |
262 | if (touch_ts == 0) { |
263 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
264 | /* |
265 | * If the time stamp was touched atomically |
266 | * make sure the scheduler tick is up to date. |
267 | */ |
268 | __this_cpu_write(softlockup_touch_sync, false); |
269 | sched_clock_tick(); |
270 | } |
271 | __touch_watchdog(); |
272 | return HRTIMER_RESTART; |
273 | } |
274 | |
275 | /* check for a softlockup |
276 | * This is done by making sure a high priority task is |
277 | * being scheduled. The task touches the watchdog to |
278 | * indicate it is getting cpu time. If it hasn't then |
279 | * this is a good indication some task is hogging the cpu |
280 | */ |
281 | duration = is_softlockup(touch_ts); |
282 | if (unlikely(duration)) { |
283 | /* only warn once */ |
284 | if (__this_cpu_read(soft_watchdog_warn) == true) |
285 | return HRTIMER_RESTART; |
286 | |
287 | printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
288 | smp_processor_id(), duration, |
289 | current->comm, task_pid_nr(current)); |
290 | print_modules(); |
291 | print_irqtrace_events(current); |
292 | if (regs) |
293 | show_regs(regs); |
294 | else |
295 | dump_stack(); |
296 | |
297 | if (softlockup_panic) |
298 | panic("softlockup: hung tasks"); |
299 | __this_cpu_write(soft_watchdog_warn, true); |
300 | } else |
301 | __this_cpu_write(soft_watchdog_warn, false); |
302 | |
303 | return HRTIMER_RESTART; |
304 | } |
305 | |
306 | |
307 | /* |
308 | * The watchdog thread - touches the timestamp. |
309 | */ |
310 | static int watchdog(void *unused) |
311 | { |
312 | static struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
313 | struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer); |
314 | |
315 | sched_setscheduler(current, SCHED_FIFO, ¶m); |
316 | |
317 | /* initialize timestamp */ |
318 | __touch_watchdog(); |
319 | |
320 | /* kick off the timer for the hardlockup detector */ |
321 | /* done here because hrtimer_start can only pin to smp_processor_id() */ |
322 | hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()), |
323 | HRTIMER_MODE_REL_PINNED); |
324 | |
325 | set_current_state(TASK_INTERRUPTIBLE); |
326 | /* |
327 | * Run briefly once per second to reset the softlockup timestamp. |
328 | * If this gets delayed for more than 60 seconds then the |
329 | * debug-printout triggers in watchdog_timer_fn(). |
330 | */ |
331 | while (!kthread_should_stop()) { |
332 | __touch_watchdog(); |
333 | schedule(); |
334 | |
335 | if (kthread_should_stop()) |
336 | break; |
337 | |
338 | set_current_state(TASK_INTERRUPTIBLE); |
339 | } |
340 | __set_current_state(TASK_RUNNING); |
341 | |
342 | return 0; |
343 | } |
344 | |
345 | |
346 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
347 | static int watchdog_nmi_enable(int cpu) |
348 | { |
349 | struct perf_event_attr *wd_attr; |
350 | struct perf_event *event = per_cpu(watchdog_ev, cpu); |
351 | |
352 | /* is it already setup and enabled? */ |
353 | if (event && event->state > PERF_EVENT_STATE_OFF) |
354 | goto out; |
355 | |
356 | /* it is setup but not enabled */ |
357 | if (event != NULL) |
358 | goto out_enable; |
359 | |
360 | /* Try to register using hardware perf events */ |
361 | wd_attr = &wd_hw_attr; |
362 | wd_attr->sample_period = hw_nmi_get_sample_period(); |
363 | event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback); |
364 | if (!IS_ERR(event)) { |
365 | printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n"); |
366 | goto out_save; |
367 | } |
368 | |
369 | |
370 | /* vary the KERN level based on the returned errno */ |
371 | if (PTR_ERR(event) == -EOPNOTSUPP) |
372 | printk(KERN_INFO "NMI watchdog disabled (cpu%i): not supported (no LAPIC?)\n", cpu); |
373 | else if (PTR_ERR(event) == -ENOENT) |
374 | printk(KERN_WARNING "NMI watchdog disabled (cpu%i): hardware events not enabled\n", cpu); |
375 | else |
376 | printk(KERN_ERR "NMI watchdog disabled (cpu%i): unable to create perf event: %ld\n", cpu, PTR_ERR(event)); |
377 | return PTR_ERR(event); |
378 | |
379 | /* success path */ |
380 | out_save: |
381 | per_cpu(watchdog_ev, cpu) = event; |
382 | out_enable: |
383 | perf_event_enable(per_cpu(watchdog_ev, cpu)); |
384 | out: |
385 | return 0; |
386 | } |
387 | |
388 | static void watchdog_nmi_disable(int cpu) |
389 | { |
390 | struct perf_event *event = per_cpu(watchdog_ev, cpu); |
391 | |
392 | if (event) { |
393 | perf_event_disable(event); |
394 | per_cpu(watchdog_ev, cpu) = NULL; |
395 | |
396 | /* should be in cleanup, but blocks oprofile */ |
397 | perf_event_release_kernel(event); |
398 | } |
399 | return; |
400 | } |
401 | #else |
402 | static int watchdog_nmi_enable(int cpu) { return 0; } |
403 | static void watchdog_nmi_disable(int cpu) { return; } |
404 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
405 | |
406 | /* prepare/enable/disable routines */ |
407 | static int watchdog_prepare_cpu(int cpu) |
408 | { |
409 | struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); |
410 | |
411 | WARN_ON(per_cpu(softlockup_watchdog, cpu)); |
412 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
413 | hrtimer->function = watchdog_timer_fn; |
414 | |
415 | return 0; |
416 | } |
417 | |
418 | static int watchdog_enable(int cpu) |
419 | { |
420 | struct task_struct *p = per_cpu(softlockup_watchdog, cpu); |
421 | int err = 0; |
422 | |
423 | /* enable the perf event */ |
424 | err = watchdog_nmi_enable(cpu); |
425 | |
426 | /* Regardless of err above, fall through and start softlockup */ |
427 | |
428 | /* create the watchdog thread */ |
429 | if (!p) { |
430 | p = kthread_create(watchdog, (void *)(unsigned long)cpu, "watchdog/%d", cpu); |
431 | if (IS_ERR(p)) { |
432 | printk(KERN_ERR "softlockup watchdog for %i failed\n", cpu); |
433 | if (!err) { |
434 | /* if hardlockup hasn't already set this */ |
435 | err = PTR_ERR(p); |
436 | /* and disable the perf event */ |
437 | watchdog_nmi_disable(cpu); |
438 | } |
439 | goto out; |
440 | } |
441 | kthread_bind(p, cpu); |
442 | per_cpu(watchdog_touch_ts, cpu) = 0; |
443 | per_cpu(softlockup_watchdog, cpu) = p; |
444 | wake_up_process(p); |
445 | } |
446 | |
447 | out: |
448 | return err; |
449 | } |
450 | |
451 | static void watchdog_disable(int cpu) |
452 | { |
453 | struct task_struct *p = per_cpu(softlockup_watchdog, cpu); |
454 | struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); |
455 | |
456 | /* |
457 | * cancel the timer first to stop incrementing the stats |
458 | * and waking up the kthread |
459 | */ |
460 | hrtimer_cancel(hrtimer); |
461 | |
462 | /* disable the perf event */ |
463 | watchdog_nmi_disable(cpu); |
464 | |
465 | /* stop the watchdog thread */ |
466 | if (p) { |
467 | per_cpu(softlockup_watchdog, cpu) = NULL; |
468 | kthread_stop(p); |
469 | } |
470 | } |
471 | |
472 | static void watchdog_enable_all_cpus(void) |
473 | { |
474 | int cpu; |
475 | |
476 | watchdog_enabled = 0; |
477 | |
478 | for_each_online_cpu(cpu) |
479 | if (!watchdog_enable(cpu)) |
480 | /* if any cpu succeeds, watchdog is considered |
481 | enabled for the system */ |
482 | watchdog_enabled = 1; |
483 | |
484 | if (!watchdog_enabled) |
485 | printk(KERN_ERR "watchdog: failed to be enabled on some cpus\n"); |
486 | |
487 | } |
488 | |
489 | static void watchdog_disable_all_cpus(void) |
490 | { |
491 | int cpu; |
492 | |
493 | for_each_online_cpu(cpu) |
494 | watchdog_disable(cpu); |
495 | |
496 | /* if all watchdogs are disabled, then they are disabled for the system */ |
497 | watchdog_enabled = 0; |
498 | } |
499 | |
500 | |
501 | /* sysctl functions */ |
502 | #ifdef CONFIG_SYSCTL |
503 | /* |
504 | * proc handler for /proc/sys/kernel/nmi_watchdog |
505 | */ |
506 | |
507 | int proc_dowatchdog_enabled(struct ctl_table *table, int write, |
508 | void __user *buffer, size_t *length, loff_t *ppos) |
509 | { |
510 | proc_dointvec(table, write, buffer, length, ppos); |
511 | |
512 | if (write) { |
513 | if (watchdog_enabled) |
514 | watchdog_enable_all_cpus(); |
515 | else |
516 | watchdog_disable_all_cpus(); |
517 | } |
518 | return 0; |
519 | } |
520 | |
521 | int proc_dowatchdog_thresh(struct ctl_table *table, int write, |
522 | void __user *buffer, |
523 | size_t *lenp, loff_t *ppos) |
524 | { |
525 | return proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
526 | } |
527 | #endif /* CONFIG_SYSCTL */ |
528 | |
529 | |
530 | /* |
531 | * Create/destroy watchdog threads as CPUs come and go: |
532 | */ |
533 | static int __cpuinit |
534 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) |
535 | { |
536 | int hotcpu = (unsigned long)hcpu; |
537 | int err = 0; |
538 | |
539 | switch (action) { |
540 | case CPU_UP_PREPARE: |
541 | case CPU_UP_PREPARE_FROZEN: |
542 | err = watchdog_prepare_cpu(hotcpu); |
543 | break; |
544 | case CPU_ONLINE: |
545 | case CPU_ONLINE_FROZEN: |
546 | if (watchdog_enabled) |
547 | err = watchdog_enable(hotcpu); |
548 | break; |
549 | #ifdef CONFIG_HOTPLUG_CPU |
550 | case CPU_UP_CANCELED: |
551 | case CPU_UP_CANCELED_FROZEN: |
552 | watchdog_disable(hotcpu); |
553 | break; |
554 | case CPU_DEAD: |
555 | case CPU_DEAD_FROZEN: |
556 | watchdog_disable(hotcpu); |
557 | break; |
558 | #endif /* CONFIG_HOTPLUG_CPU */ |
559 | } |
560 | |
561 | /* |
562 | * hardlockup and softlockup are not important enough |
563 | * to block cpu bring up. Just always succeed and |
564 | * rely on printk output to flag problems. |
565 | */ |
566 | return NOTIFY_OK; |
567 | } |
568 | |
569 | static struct notifier_block __cpuinitdata cpu_nfb = { |
570 | .notifier_call = cpu_callback |
571 | }; |
572 | |
573 | void __init lockup_detector_init(void) |
574 | { |
575 | void *cpu = (void *)(long)smp_processor_id(); |
576 | int err; |
577 | |
578 | err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
579 | WARN_ON(notifier_to_errno(err)); |
580 | |
581 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
582 | register_cpu_notifier(&cpu_nfb); |
583 | |
584 | return; |
585 | } |
586 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9