watchdog.c 14 KB
Newer Older
1 2 3 4 5
/*
 * Detect hard and soft lockups on a system
 *
 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
 *
6 7 8
 * Note: Most of this code is borrowed heavily from the original softlockup
 * detector, so thanks to Ingo for the initial implementation.
 * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
9 10 11
 * to those contributors as well.
 */

12 13
#define pr_fmt(fmt) "NMI watchdog: " fmt

14 15 16 17 18 19 20 21 22 23 24
#include <linux/mm.h>
#include <linux/cpu.h>
#include <linux/nmi.h>
#include <linux/init.h>
#include <linux/delay.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
#include <linux/lockdep.h>
#include <linux/notifier.h>
#include <linux/module.h>
#include <linux/sysctl.h>
25
#include <linux/smpboot.h>
26
#include <linux/sched/rt.h>
27 28

#include <asm/irq_regs.h>
29
#include <linux/kvm_para.h>
30 31
#include <linux/perf_event.h>

32
int watchdog_enabled = 1;
33
int __read_mostly watchdog_thresh = 10;
34
static int __read_mostly watchdog_disabled;
35
static u64 __read_mostly sample_period;
36 37 38 39 40 41

static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
static DEFINE_PER_CPU(bool, softlockup_touch_sync);
static DEFINE_PER_CPU(bool, soft_watchdog_warn);
42 43
static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
44
#ifdef CONFIG_HARDLOCKUP_DETECTOR
45 46
static DEFINE_PER_CPU(bool, hard_watchdog_warn);
static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
47 48 49 50 51 52 53 54
static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
#endif

/* boot commands */
/*
 * Should we panic when a soft-lockup or hard-lockup occurs:
 */
55
#ifdef CONFIG_HARDLOCKUP_DETECTOR
56 57
static int hardlockup_panic =
			CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
58 59 60 61 62

static int __init hardlockup_panic_setup(char *str)
{
	if (!strncmp(str, "panic", 5))
		hardlockup_panic = 1;
63 64
	else if (!strncmp(str, "nopanic", 7))
		hardlockup_panic = 0;
65
	else if (!strncmp(str, "0", 1))
66
		watchdog_enabled = 0;
67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
	return 1;
}
__setup("nmi_watchdog=", hardlockup_panic_setup);
#endif

unsigned int __read_mostly softlockup_panic =
			CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;

static int __init softlockup_panic_setup(char *str)
{
	softlockup_panic = simple_strtoul(str, NULL, 0);

	return 1;
}
__setup("softlockup_panic=", softlockup_panic_setup);

static int __init nowatchdog_setup(char *str)
{
85
	watchdog_enabled = 0;
86 87 88 89 90 91 92
	return 1;
}
__setup("nowatchdog", nowatchdog_setup);

/* deprecated */
static int __init nosoftlockup_setup(char *str)
{
93
	watchdog_enabled = 0;
94 95 96 97 98
	return 1;
}
__setup("nosoftlockup", nosoftlockup_setup);
/*  */

99 100 101 102 103 104 105
/*
 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
 * lockups can have false positives under extreme conditions. So we generally
 * want a higher threshold for soft lockups than for hard lockups. So we couple
 * the thresholds with a factor: we make the soft threshold twice the amount of
 * time the hard threshold is.
 */
106
static int get_softlockup_thresh(void)
107 108 109
{
	return watchdog_thresh * 2;
}
110 111 112 113 114 115

/*
 * Returns seconds, approximately.  We don't need nanosecond
 * resolution, and we don't need to waste time with a big divide when
 * 2^30ns == 1.074s.
 */
116
static unsigned long get_timestamp(void)
117
{
118
	return local_clock() >> 30LL;  /* 2^30 ~= 10^9 */
119 120
}

121
static void set_sample_period(void)
122 123
{
	/*
124
	 * convert watchdog_thresh from seconds to ns
125 126 127 128
	 * the divide by 5 is to give hrtimer several chances (two
	 * or three with the current relation between the soft
	 * and hard thresholds) to increment before the
	 * hardlockup detector generates a warning
129
	 */
130
	sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
131 132 133 134 135
}

/* Commands for resetting the watchdog */
static void __touch_watchdog(void)
{
136
	__this_cpu_write(watchdog_touch_ts, get_timestamp());
137 138
}

139
void touch_softlockup_watchdog(void)
140
{
141
	__this_cpu_write(watchdog_touch_ts, 0);
142
}
143
EXPORT_SYMBOL(touch_softlockup_watchdog);
144

145
void touch_all_softlockup_watchdogs(void)
146 147 148 149 150 151 152 153 154 155 156 157
{
	int cpu;

	/*
	 * this is done lockless
	 * do we care if a 0 races with a timestamp?
	 * all it means is the softlock check starts one cycle later
	 */
	for_each_online_cpu(cpu)
		per_cpu(watchdog_touch_ts, cpu) = 0;
}

158
#ifdef CONFIG_HARDLOCKUP_DETECTOR
159 160
void touch_nmi_watchdog(void)
{
161 162 163 164 165 166 167 168
	if (watchdog_enabled) {
		unsigned cpu;

		for_each_present_cpu(cpu) {
			if (per_cpu(watchdog_nmi_touch, cpu) != true)
				per_cpu(watchdog_nmi_touch, cpu) = true;
		}
	}
169
	touch_softlockup_watchdog();
170 171 172
}
EXPORT_SYMBOL(touch_nmi_watchdog);

173 174
#endif

175 176 177 178 179 180
void touch_softlockup_watchdog_sync(void)
{
	__raw_get_cpu_var(softlockup_touch_sync) = true;
	__raw_get_cpu_var(watchdog_touch_ts) = 0;
}

181
#ifdef CONFIG_HARDLOCKUP_DETECTOR
182
/* watchdog detector functions */
183
static int is_hardlockup(void)
184
{
185
	unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
186

187
	if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
188 189
		return 1;

190
	__this_cpu_write(hrtimer_interrupts_saved, hrint);
191 192 193 194
	return 0;
}
#endif

195
static int is_softlockup(unsigned long touch_ts)
196
{
197
	unsigned long now = get_timestamp();
198 199

	/* Warn about unreasonable delays: */
200
	if (time_after(now, touch_ts + get_softlockup_thresh()))
201 202 203 204 205
		return now - touch_ts;

	return 0;
}

206
#ifdef CONFIG_HARDLOCKUP_DETECTOR
207

208 209 210 211 212 213 214 215 216
static struct perf_event_attr wd_hw_attr = {
	.type		= PERF_TYPE_HARDWARE,
	.config		= PERF_COUNT_HW_CPU_CYCLES,
	.size		= sizeof(struct perf_event_attr),
	.pinned		= 1,
	.disabled	= 1,
};

/* Callback function for perf event subsystem */
217
static void watchdog_overflow_callback(struct perf_event *event,
218 219 220
		 struct perf_sample_data *data,
		 struct pt_regs *regs)
{
221 222 223
	/* Ensure the watchdog never gets throttled */
	event->hw.interrupts = 0;

224 225
	if (__this_cpu_read(watchdog_nmi_touch) == true) {
		__this_cpu_write(watchdog_nmi_touch, false);
226 227 228 229 230 231 232 233 234
		return;
	}

	/* check for a hardlockup
	 * This is done by making sure our timer interrupt
	 * is incrementing.  The timer interrupt should have
	 * fired multiple times before we overflow'd.  If it hasn't
	 * then this is a good indication the cpu is stuck
	 */
235 236 237
	if (is_hardlockup()) {
		int this_cpu = smp_processor_id();

238
		/* only print hardlockups once */
239
		if (__this_cpu_read(hard_watchdog_warn) == true)
240 241 242 243 244 245 246
			return;

		if (hardlockup_panic)
			panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
		else
			WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu);

247
		__this_cpu_write(hard_watchdog_warn, true);
248 249 250
		return;
	}

251
	__this_cpu_write(hard_watchdog_warn, false);
252 253
	return;
}
254 255
#endif /* CONFIG_HARDLOCKUP_DETECTOR */

256 257
static void watchdog_interrupt_count(void)
{
258
	__this_cpu_inc(hrtimer_interrupts);
259
}
260 261 262

static int watchdog_nmi_enable(unsigned int cpu);
static void watchdog_nmi_disable(unsigned int cpu);
263 264 265 266

/* watchdog kicker functions */
static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
{
267
	unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
268 269 270 271 272 273 274
	struct pt_regs *regs = get_irq_regs();
	int duration;

	/* kick the hardlockup detector */
	watchdog_interrupt_count();

	/* kick the softlockup detector */
275
	wake_up_process(__this_cpu_read(softlockup_watchdog));
276 277

	/* .. and repeat */
278
	hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
279 280

	if (touch_ts == 0) {
281
		if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
282 283 284 285
			/*
			 * If the time stamp was touched atomically
			 * make sure the scheduler tick is up to date.
			 */
286
			__this_cpu_write(softlockup_touch_sync, false);
287 288
			sched_clock_tick();
		}
289 290 291

		/* Clear the guest paused flag on watchdog reset */
		kvm_check_and_clear_guest_paused();
292 293 294 295 296 297 298 299 300 301
		__touch_watchdog();
		return HRTIMER_RESTART;
	}

	/* check for a softlockup
	 * This is done by making sure a high priority task is
	 * being scheduled.  The task touches the watchdog to
	 * indicate it is getting cpu time.  If it hasn't then
	 * this is a good indication some task is hogging the cpu
	 */
302
	duration = is_softlockup(touch_ts);
303
	if (unlikely(duration)) {
304 305 306 307 308 309 310 311
		/*
		 * If a virtual machine is stopped by the host it can look to
		 * the watchdog like a soft lockup, check to see if the host
		 * stopped the vm before we issue the warning
		 */
		if (kvm_check_and_clear_guest_paused())
			return HRTIMER_RESTART;

312
		/* only warn once */
313
		if (__this_cpu_read(soft_watchdog_warn) == true)
314 315
			return HRTIMER_RESTART;

316
		printk(KERN_EMERG "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
317
			smp_processor_id(), duration,
318 319 320 321 322 323 324 325 326 327
			current->comm, task_pid_nr(current));
		print_modules();
		print_irqtrace_events(current);
		if (regs)
			show_regs(regs);
		else
			dump_stack();

		if (softlockup_panic)
			panic("softlockup: hung tasks");
328
		__this_cpu_write(soft_watchdog_warn, true);
329
	} else
330
		__this_cpu_write(soft_watchdog_warn, false);
331 332 333 334

	return HRTIMER_RESTART;
}

335 336 337
static void watchdog_set_prio(unsigned int policy, unsigned int prio)
{
	struct sched_param param = { .sched_priority = prio };
338

339 340 341 342
	sched_setscheduler(current, policy, &param);
}

static void watchdog_enable(unsigned int cpu)
343
{
344
	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
345

346 347 348 349
	/* kick off the timer for the hardlockup detector */
	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hrtimer->function = watchdog_timer_fn;

350 351 352 353 354 355 356
	if (!watchdog_enabled) {
		kthread_park(current);
		return;
	}

	/* Enable the perf event */
	watchdog_nmi_enable(cpu);
357 358

	/* done here because hrtimer_start can only pin to smp_processor_id() */
359
	hrtimer_start(hrtimer, ns_to_ktime(sample_period),
360 361
		      HRTIMER_MODE_REL_PINNED);

362 363 364 365
	/* initialize timestamp */
	watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
	__touch_watchdog();
}
366

367 368 369
static void watchdog_disable(unsigned int cpu)
{
	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
370

371 372 373 374
	watchdog_set_prio(SCHED_NORMAL, 0);
	hrtimer_cancel(hrtimer);
	/* disable the perf event */
	watchdog_nmi_disable(cpu);
375 376
}

377 378 379 380 381 382 383 384 385
static int watchdog_should_run(unsigned int cpu)
{
	return __this_cpu_read(hrtimer_interrupts) !=
		__this_cpu_read(soft_lockup_hrtimer_cnt);
}

/*
 * The watchdog thread function - touches the timestamp.
 *
386
 * It only runs once every sample_period seconds (4 seconds by
387 388 389 390 391 392 393 394 395 396
 * default) to reset the softlockup timestamp. If this gets delayed
 * for more than 2*watchdog_thresh seconds then the debug-printout
 * triggers in watchdog_timer_fn().
 */
static void watchdog(unsigned int cpu)
{
	__this_cpu_write(soft_lockup_hrtimer_cnt,
			 __this_cpu_read(hrtimer_interrupts));
	__touch_watchdog();
}
397

398
#ifdef CONFIG_HARDLOCKUP_DETECTOR
399 400 401 402 403 404 405
/*
 * People like the simple clean cpu node info on boot.
 * Reduce the watchdog noise by only printing messages
 * that are different from what cpu0 displayed.
 */
static unsigned long cpu0_err;

406
static int watchdog_nmi_enable(unsigned int cpu)
407 408 409 410 411 412 413 414 415 416 417 418 419
{
	struct perf_event_attr *wd_attr;
	struct perf_event *event = per_cpu(watchdog_ev, cpu);

	/* is it already setup and enabled? */
	if (event && event->state > PERF_EVENT_STATE_OFF)
		goto out;

	/* it is setup but not enabled */
	if (event != NULL)
		goto out_enable;

	wd_attr = &wd_hw_attr;
420
	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
421 422

	/* Try to register using hardware perf events */
423
	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
424 425 426 427 428

	/* save cpu0 error for future comparision */
	if (cpu == 0 && IS_ERR(event))
		cpu0_err = PTR_ERR(event);

429
	if (!IS_ERR(event)) {
430 431 432
		/* only print for cpu0 or different than cpu0 */
		if (cpu == 0 || cpu0_err)
			pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n");
433 434 435
		goto out_save;
	}

436 437 438
	/* skip displaying the same error again */
	if (cpu > 0 && (PTR_ERR(event) == cpu0_err))
		return PTR_ERR(event);
439 440 441

	/* vary the KERN level based on the returned errno */
	if (PTR_ERR(event) == -EOPNOTSUPP)
442
		pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
443
	else if (PTR_ERR(event) == -ENOENT)
444 445
		pr_warning("disabled (cpu%i): hardware events not enabled\n",
			 cpu);
446
	else
447 448
		pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
			cpu, PTR_ERR(event));
449
	return PTR_ERR(event);
450 451 452 453 454 455 456 457 458 459

	/* success path */
out_save:
	per_cpu(watchdog_ev, cpu) = event;
out_enable:
	perf_event_enable(per_cpu(watchdog_ev, cpu));
out:
	return 0;
}

460
static void watchdog_nmi_disable(unsigned int cpu)
461 462 463 464 465 466 467 468 469 470 471 472 473
{
	struct perf_event *event = per_cpu(watchdog_ev, cpu);

	if (event) {
		perf_event_disable(event);
		per_cpu(watchdog_ev, cpu) = NULL;

		/* should be in cleanup, but blocks oprofile */
		perf_event_release_kernel(event);
	}
	return;
}
#else
474 475
static int watchdog_nmi_enable(unsigned int cpu) { return 0; }
static void watchdog_nmi_disable(unsigned int cpu) { return; }
476
#endif /* CONFIG_HARDLOCKUP_DETECTOR */
477 478

/* prepare/enable/disable routines */
479 480
/* sysctl functions */
#ifdef CONFIG_SYSCTL
481 482
static void watchdog_enable_all_cpus(void)
{
483
	unsigned int cpu;
484

485 486 487 488 489
	if (watchdog_disabled) {
		watchdog_disabled = 0;
		for_each_online_cpu(cpu)
			kthread_unpark(per_cpu(softlockup_watchdog, cpu));
	}
490 491 492 493
}

static void watchdog_disable_all_cpus(void)
{
494
	unsigned int cpu;
495

496 497 498 499 500
	if (!watchdog_disabled) {
		watchdog_disabled = 1;
		for_each_online_cpu(cpu)
			kthread_park(per_cpu(softlockup_watchdog, cpu));
	}
501 502 503
}

/*
504
 * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh
505 506
 */

507 508
int proc_dowatchdog(struct ctl_table *table, int write,
		    void __user *buffer, size_t *lenp, loff_t *ppos)
509
{
510
	int ret;
511

512 513 514
	if (watchdog_disabled < 0)
		return -ENODEV;

515
	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
516
	if (ret || !write)
517
		return ret;
518

519
	set_sample_period();
520 521 522 523 524
	/*
	 * Watchdog threads shouldn't be enabled if they are
	 * disabled. The 'watchdog_disabled' variable check in
	 * watchdog_*_all_cpus() function takes care of this.
	 */
525
	if (watchdog_enabled && watchdog_thresh)
526 527 528 529 530
		watchdog_enable_all_cpus();
	else
		watchdog_disable_all_cpus();

	return ret;
531 532 533
}
#endif /* CONFIG_SYSCTL */

534 535 536 537 538 539 540 541
static struct smp_hotplug_thread watchdog_threads = {
	.store			= &softlockup_watchdog,
	.thread_should_run	= watchdog_should_run,
	.thread_fn		= watchdog,
	.thread_comm		= "watchdog/%u",
	.setup			= watchdog_enable,
	.park			= watchdog_disable,
	.unpark			= watchdog_enable,
542 543
};

544
void __init lockup_detector_init(void)
545
{
546
	set_sample_period();
547 548 549 550
	if (smpboot_register_percpu_thread(&watchdog_threads)) {
		pr_err("Failed to create watchdog threads, disabled\n");
		watchdog_disabled = -ENODEV;
	}
551
}