watchdog.c 17 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_user_enabled = 1;
33
int __read_mostly watchdog_thresh = 10;
34 35 36 37 38 39
#ifdef CONFIG_SMP
int __read_mostly sysctl_softlockup_all_cpu_backtrace;
#else
#define sysctl_softlockup_all_cpu_backtrace 0
#endif

40
static int __read_mostly watchdog_running;
41
static u64 __read_mostly sample_period;
42 43 44 45 46 47

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);
48 49
static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
50
#ifdef CONFIG_HARDLOCKUP_DETECTOR
51 52
static DEFINE_PER_CPU(bool, hard_watchdog_warn);
static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
53 54 55
static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
#endif
56
static unsigned long soft_lockup_nmi_warn;
57 58 59 60 61

/* boot commands */
/*
 * Should we panic when a soft-lockup or hard-lockup occurs:
 */
62
#ifdef CONFIG_HARDLOCKUP_DETECTOR
63 64
static int hardlockup_panic =
			CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
65 66 67 68 69

static int __init hardlockup_panic_setup(char *str)
{
	if (!strncmp(str, "panic", 5))
		hardlockup_panic = 1;
70 71
	else if (!strncmp(str, "nopanic", 7))
		hardlockup_panic = 0;
72
	else if (!strncmp(str, "0", 1))
73
		watchdog_user_enabled = 0;
74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
	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)
{
92
	watchdog_user_enabled = 0;
93 94 95 96 97 98 99
	return 1;
}
__setup("nowatchdog", nowatchdog_setup);

/* deprecated */
static int __init nosoftlockup_setup(char *str)
{
100
	watchdog_user_enabled = 0;
101 102 103 104
	return 1;
}
__setup("nosoftlockup", nosoftlockup_setup);
/*  */
105 106 107 108 109 110 111 112 113
#ifdef CONFIG_SMP
static int __init softlockup_all_cpu_backtrace_setup(char *str)
{
	sysctl_softlockup_all_cpu_backtrace =
		!!simple_strtol(str, NULL, 0);
	return 1;
}
__setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
#endif
114

115 116 117 118 119 120 121
/*
 * 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.
 */
122
static int get_softlockup_thresh(void)
123 124 125
{
	return watchdog_thresh * 2;
}
126 127 128 129 130 131

/*
 * 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.
 */
132
static unsigned long get_timestamp(void)
133
{
134
	return local_clock() >> 30LL;  /* 2^30 ~= 10^9 */
135 136
}

137
static void set_sample_period(void)
138 139
{
	/*
140
	 * convert watchdog_thresh from seconds to ns
141 142 143 144
	 * 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
145
	 */
146
	sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
147 148 149 150 151
}

/* Commands for resetting the watchdog */
static void __touch_watchdog(void)
{
152
	__this_cpu_write(watchdog_touch_ts, get_timestamp());
153 154
}

155
void touch_softlockup_watchdog(void)
156
{
157 158 159 160 161
	/*
	 * Preemption can be enabled.  It doesn't matter which CPU's timestamp
	 * gets zeroed here, so use the raw_ operation.
	 */
	raw_cpu_write(watchdog_touch_ts, 0);
162
}
163
EXPORT_SYMBOL(touch_softlockup_watchdog);
164

165
void touch_all_softlockup_watchdogs(void)
166 167 168 169 170 171 172 173 174 175 176 177
{
	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;
}

178
#ifdef CONFIG_HARDLOCKUP_DETECTOR
179 180
void touch_nmi_watchdog(void)
{
181 182 183 184 185 186 187 188
	/*
	 * Using __raw here because some code paths have
	 * preemption enabled.  If preemption is enabled
	 * then interrupts should be enabled too, in which
	 * case we shouldn't have to worry about the watchdog
	 * going off.
	 */
	__raw_get_cpu_var(watchdog_nmi_touch) = true;
189
	touch_softlockup_watchdog();
190 191 192
}
EXPORT_SYMBOL(touch_nmi_watchdog);

193 194
#endif

195 196 197 198 199 200
void touch_softlockup_watchdog_sync(void)
{
	__raw_get_cpu_var(softlockup_touch_sync) = true;
	__raw_get_cpu_var(watchdog_touch_ts) = 0;
}

201
#ifdef CONFIG_HARDLOCKUP_DETECTOR
202
/* watchdog detector functions */
203
static int is_hardlockup(void)
204
{
205
	unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
206

207
	if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
208 209
		return 1;

210
	__this_cpu_write(hrtimer_interrupts_saved, hrint);
211 212 213 214
	return 0;
}
#endif

215
static int is_softlockup(unsigned long touch_ts)
216
{
217
	unsigned long now = get_timestamp();
218 219

	/* Warn about unreasonable delays: */
220
	if (time_after(now, touch_ts + get_softlockup_thresh()))
221 222 223 224 225
		return now - touch_ts;

	return 0;
}

226
#ifdef CONFIG_HARDLOCKUP_DETECTOR
227

228 229 230 231 232 233 234 235 236
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 */
237
static void watchdog_overflow_callback(struct perf_event *event,
238 239 240
		 struct perf_sample_data *data,
		 struct pt_regs *regs)
{
241 242 243
	/* Ensure the watchdog never gets throttled */
	event->hw.interrupts = 0;

244 245
	if (__this_cpu_read(watchdog_nmi_touch) == true) {
		__this_cpu_write(watchdog_nmi_touch, false);
246 247 248 249 250 251 252 253 254
		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
	 */
255 256 257
	if (is_hardlockup()) {
		int this_cpu = smp_processor_id();

258
		/* only print hardlockups once */
259
		if (__this_cpu_read(hard_watchdog_warn) == true)
260 261 262
			return;

		if (hardlockup_panic)
263 264
			panic("Watchdog detected hard LOCKUP on cpu %d",
			      this_cpu);
265
		else
266 267
			WARN(1, "Watchdog detected hard LOCKUP on cpu %d",
			     this_cpu);
268

269
		__this_cpu_write(hard_watchdog_warn, true);
270 271 272
		return;
	}

273
	__this_cpu_write(hard_watchdog_warn, false);
274 275
	return;
}
276 277
#endif /* CONFIG_HARDLOCKUP_DETECTOR */

278 279
static void watchdog_interrupt_count(void)
{
280
	__this_cpu_inc(hrtimer_interrupts);
281
}
282 283 284

static int watchdog_nmi_enable(unsigned int cpu);
static void watchdog_nmi_disable(unsigned int cpu);
285 286 287 288

/* watchdog kicker functions */
static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
{
289
	unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
290 291
	struct pt_regs *regs = get_irq_regs();
	int duration;
292
	int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
293 294 295 296 297

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

	/* kick the softlockup detector */
298
	wake_up_process(__this_cpu_read(softlockup_watchdog));
299 300

	/* .. and repeat */
301
	hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
302 303

	if (touch_ts == 0) {
304
		if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
305 306 307 308
			/*
			 * If the time stamp was touched atomically
			 * make sure the scheduler tick is up to date.
			 */
309
			__this_cpu_write(softlockup_touch_sync, false);
310 311
			sched_clock_tick();
		}
312 313 314

		/* Clear the guest paused flag on watchdog reset */
		kvm_check_and_clear_guest_paused();
315 316 317 318 319 320 321 322 323 324
		__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
	 */
325
	duration = is_softlockup(touch_ts);
326
	if (unlikely(duration)) {
327 328 329 330 331 332 333 334
		/*
		 * 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;

335
		/* only warn once */
336
		if (__this_cpu_read(soft_watchdog_warn) == true)
337 338
			return HRTIMER_RESTART;

339 340 341 342 343 344 345 346 347 348 349
		if (softlockup_all_cpu_backtrace) {
			/* Prevent multiple soft-lockup reports if one cpu is already
			 * engaged in dumping cpu back traces
			 */
			if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
				/* Someone else will report us. Let's give up */
				__this_cpu_write(soft_watchdog_warn, true);
				return HRTIMER_RESTART;
			}
		}

350
		pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
351
			smp_processor_id(), duration,
352 353 354 355 356 357 358 359
			current->comm, task_pid_nr(current));
		print_modules();
		print_irqtrace_events(current);
		if (regs)
			show_regs(regs);
		else
			dump_stack();

360 361 362 363 364 365 366 367 368 369 370
		if (softlockup_all_cpu_backtrace) {
			/* Avoid generating two back traces for current
			 * given that one is already made above
			 */
			trigger_allbutself_cpu_backtrace();

			clear_bit(0, &soft_lockup_nmi_warn);
			/* Barrier to sync with other cpus */
			smp_mb__after_atomic();
		}

Josh Hunt's avatar
Josh Hunt committed
371
		add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
372 373
		if (softlockup_panic)
			panic("softlockup: hung tasks");
374
		__this_cpu_write(soft_watchdog_warn, true);
375
	} else
376
		__this_cpu_write(soft_watchdog_warn, false);
377 378 379 380

	return HRTIMER_RESTART;
}

381 382 383
static void watchdog_set_prio(unsigned int policy, unsigned int prio)
{
	struct sched_param param = { .sched_priority = prio };
384

385 386 387 388
	sched_setscheduler(current, policy, &param);
}

static void watchdog_enable(unsigned int cpu)
389
{
390
	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
391

392 393 394 395
	/* kick off the timer for the hardlockup detector */
	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hrtimer->function = watchdog_timer_fn;

396 397
	/* Enable the perf event */
	watchdog_nmi_enable(cpu);
398 399

	/* done here because hrtimer_start can only pin to smp_processor_id() */
400
	hrtimer_start(hrtimer, ns_to_ktime(sample_period),
401 402
		      HRTIMER_MODE_REL_PINNED);

403 404 405 406
	/* initialize timestamp */
	watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
	__touch_watchdog();
}
407

408 409 410
static void watchdog_disable(unsigned int cpu)
{
	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
411

412 413 414 415
	watchdog_set_prio(SCHED_NORMAL, 0);
	hrtimer_cancel(hrtimer);
	/* disable the perf event */
	watchdog_nmi_disable(cpu);
416 417
}

418 419 420 421 422
static void watchdog_cleanup(unsigned int cpu, bool online)
{
	watchdog_disable(cpu);
}

423 424 425 426 427 428 429 430 431
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.
 *
432
 * It only runs once every sample_period seconds (4 seconds by
433 434 435 436 437 438 439 440 441 442
 * 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();
}
443

444
#ifdef CONFIG_HARDLOCKUP_DETECTOR
445 446 447 448 449 450 451
/*
 * 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;

452
static int watchdog_nmi_enable(unsigned int cpu)
453 454 455 456 457 458 459 460 461 462 463 464 465
{
	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;
466
	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
467 468

	/* Try to register using hardware perf events */
469
	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
470 471 472 473 474

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

475
	if (!IS_ERR(event)) {
476 477 478
		/* 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");
479 480 481
		goto out_save;
	}

482 483 484
	/* skip displaying the same error again */
	if (cpu > 0 && (PTR_ERR(event) == cpu0_err))
		return PTR_ERR(event);
485 486 487

	/* vary the KERN level based on the returned errno */
	if (PTR_ERR(event) == -EOPNOTSUPP)
488
		pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
489
	else if (PTR_ERR(event) == -ENOENT)
490
		pr_warn("disabled (cpu%i): hardware events not enabled\n",
491
			 cpu);
492
	else
493 494
		pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
			cpu, PTR_ERR(event));
495
	return PTR_ERR(event);
496 497 498 499 500 501 502 503 504 505

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

506
static void watchdog_nmi_disable(unsigned int cpu)
507 508 509 510 511 512 513 514 515 516 517 518 519
{
	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
520 521
static int watchdog_nmi_enable(unsigned int cpu) { return 0; }
static void watchdog_nmi_disable(unsigned int cpu) { return; }
522
#endif /* CONFIG_HARDLOCKUP_DETECTOR */
523

524 525 526 527 528 529 530 531 532 533 534
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,
	.cleanup		= watchdog_cleanup,
	.park			= watchdog_disable,
	.unpark			= watchdog_enable,
};

535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562
static void restart_watchdog_hrtimer(void *info)
{
	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
	int ret;

	/*
	 * No need to cancel and restart hrtimer if it is currently executing
	 * because it will reprogram itself with the new period now.
	 * We should never see it unqueued here because we are running per-cpu
	 * with interrupts disabled.
	 */
	ret = hrtimer_try_to_cancel(hrtimer);
	if (ret == 1)
		hrtimer_start(hrtimer, ns_to_ktime(sample_period),
				HRTIMER_MODE_REL_PINNED);
}

static void update_timers(int cpu)
{
	/*
	 * Make sure that perf event counter will adopt to a new
	 * sampling period. Updating the sampling period directly would
	 * be much nicer but we do not have an API for that now so
	 * let's use a big hammer.
	 * Hrtimer will adopt the new period on the next tick but this
	 * might be late already so we have to restart the timer as well.
	 */
	watchdog_nmi_disable(cpu);
563
	smp_call_function_single(cpu, restart_watchdog_hrtimer, NULL, 1);
564 565 566 567 568 569 570 571 572 573 574 575 576 577
	watchdog_nmi_enable(cpu);
}

static void update_timers_all_cpus(void)
{
	int cpu;

	get_online_cpus();
	for_each_online_cpu(cpu)
		update_timers(cpu);
	put_online_cpus();
}

static int watchdog_enable_all_cpus(bool sample_period_changed)
578
{
579
	int err = 0;
580

581
	if (!watchdog_running) {
582 583 584 585
		err = smpboot_register_percpu_thread(&watchdog_threads);
		if (err)
			pr_err("Failed to create watchdog threads, disabled\n");
		else
586
			watchdog_running = 1;
587 588
	} else if (sample_period_changed) {
		update_timers_all_cpus();
589
	}
590 591

	return err;
592 593
}

594 595 596
/* prepare/enable/disable routines */
/* sysctl functions */
#ifdef CONFIG_SYSCTL
597 598
static void watchdog_disable_all_cpus(void)
{
599 600
	if (watchdog_running) {
		watchdog_running = 0;
601
		smpboot_unregister_percpu_thread(&watchdog_threads);
602
	}
603 604 605
}

/*
606
 * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh
607 608
 */

609 610
int proc_dowatchdog(struct ctl_table *table, int write,
		    void __user *buffer, size_t *lenp, loff_t *ppos)
611
{
612
	int err, old_thresh, old_enabled;
613
	static DEFINE_MUTEX(watchdog_proc_mutex);
614

615
	mutex_lock(&watchdog_proc_mutex);
616
	old_thresh = ACCESS_ONCE(watchdog_thresh);
617
	old_enabled = ACCESS_ONCE(watchdog_user_enabled);
618

619 620
	err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
	if (err || !write)
621
		goto out;
622

623
	set_sample_period();
624 625
	/*
	 * Watchdog threads shouldn't be enabled if they are
626
	 * disabled. The 'watchdog_running' variable check in
627 628
	 * watchdog_*_all_cpus() function takes care of this.
	 */
629
	if (watchdog_user_enabled && watchdog_thresh)
630
		err = watchdog_enable_all_cpus(old_thresh != watchdog_thresh);
631 632 633
	else
		watchdog_disable_all_cpus();

634 635 636
	/* Restore old values on failure */
	if (err) {
		watchdog_thresh = old_thresh;
637
		watchdog_user_enabled = old_enabled;
638
	}
639 640
out:
	mutex_unlock(&watchdog_proc_mutex);
641
	return err;
642 643 644
}
#endif /* CONFIG_SYSCTL */

645
void __init lockup_detector_init(void)
646
{
647
	set_sample_period();
648

649
	if (watchdog_user_enabled)
650
		watchdog_enable_all_cpus(false);
651
}