summaryrefslogtreecommitdiffstats
path: root/drivers/perf/arm_pmu.c
blob: 1a0d340b65cf1d2b85d09a2d4e0857d7795ea112 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
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
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
#undef DEBUG

/*
 * ARM performance counter support.
 *
 * Copyright (C) 2009 picoChip Designs, Ltd., Jamie Iles
 * Copyright (C) 2010 ARM Ltd., Will Deacon <will.deacon@arm.com>
 *
 * This code is based on the sparc64 perf event code, which is in turn based
 * on the x86 code.
 */
#define pr_fmt(fmt) "hw perfevents: " fmt

#include <linux/bitmap.h>
#include <linux/cpumask.h>
#include <linux/cpu_pm.h>
#include <linux/export.h>
#include <linux/kernel.h>
#include <linux/perf/arm_pmu.h>
#include <linux/slab.h>
#include <linux/sched/clock.h>
#include <linux/spinlock.h>
#include <linux/irq.h>
#include <linux/irqdesc.h>

#include <asm/irq_regs.h>

static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu);
static DEFINE_PER_CPU(int, cpu_irq);

static int
armpmu_map_cache_event(const unsigned (*cache_map)
				      [PERF_COUNT_HW_CACHE_MAX]
				      [PERF_COUNT_HW_CACHE_OP_MAX]
				      [PERF_COUNT_HW_CACHE_RESULT_MAX],
		       u64 config)
{
	unsigned int cache_type, cache_op, cache_result, ret;

	cache_type = (config >>  0) & 0xff;
	if (cache_type >= PERF_COUNT_HW_CACHE_MAX)
		return -EINVAL;

	cache_op = (config >>  8) & 0xff;
	if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX)
		return -EINVAL;

	cache_result = (config >> 16) & 0xff;
	if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
		return -EINVAL;

	if (!cache_map)
		return -ENOENT;

	ret = (int)(*cache_map)[cache_type][cache_op][cache_result];

	if (ret == CACHE_OP_UNSUPPORTED)
		return -ENOENT;

	return ret;
}

static int
armpmu_map_hw_event(const unsigned (*event_map)[PERF_COUNT_HW_MAX], u64 config)
{
	int mapping;

	if (config >= PERF_COUNT_HW_MAX)
		return -EINVAL;

	if (!event_map)
		return -ENOENT;

	mapping = (*event_map)[config];
	return mapping == HW_OP_UNSUPPORTED ? -ENOENT : mapping;
}

static int
armpmu_map_raw_event(u32 raw_event_mask, u64 config)
{
	return (int)(config & raw_event_mask);
}

int
armpmu_map_event(struct perf_event *event,
		 const unsigned (*event_map)[PERF_COUNT_HW_MAX],
		 const unsigned (*cache_map)
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX],
		 u32 raw_event_mask)
{
	u64 config = event->attr.config;
	int type = event->attr.type;

	if (type == event->pmu->type)
		return armpmu_map_raw_event(raw_event_mask, config);

	switch (type) {
	case PERF_TYPE_HARDWARE:
		return armpmu_map_hw_event(event_map, config);
	case PERF_TYPE_HW_CACHE:
		return armpmu_map_cache_event(cache_map, config);
	case PERF_TYPE_RAW:
		return armpmu_map_raw_event(raw_event_mask, config);
	}

	return -ENOENT;
}

int armpmu_event_set_period(struct perf_event *event)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct hw_perf_event *hwc = &event->hw;
	s64 left = local64_read(&hwc->period_left);
	s64 period = hwc->sample_period;
	int ret = 0;

	if (unlikely(left <= -period)) {
		left = period;
		local64_set(&hwc->period_left, left);
		hwc->last_period = period;
		ret = 1;
	}

	if (unlikely(left <= 0)) {
		left += period;
		local64_set(&hwc->period_left, left);
		hwc->last_period = period;
		ret = 1;
	}

	/*
	 * Limit the maximum period to prevent the counter value
	 * from overtaking the one we are about to program. In
	 * effect we are reducing max_period to account for
	 * interrupt latency (and we are being very conservative).
	 */
	if (left > (armpmu->max_period >> 1))
		left = armpmu->max_period >> 1;

	local64_set(&hwc->prev_count, (u64)-left);

	armpmu->write_counter(event, (u64)(-left) & 0xffffffff);

	perf_event_update_userpage(event);

	return ret;
}

u64 armpmu_event_update(struct perf_event *event)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct hw_perf_event *hwc = &event->hw;
	u64 delta, prev_raw_count, new_raw_count;

again:
	prev_raw_count = local64_read(&hwc->prev_count);
	new_raw_count = armpmu->read_counter(event);

	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
			     new_raw_count) != prev_raw_count)
		goto again;

	delta = (new_raw_count - prev_raw_count) & armpmu->max_period;

	local64_add(delta, &event->count);
	local64_sub(delta, &hwc->period_left);

	return new_raw_count;
}

static void
armpmu_read(struct perf_event *event)
{
	armpmu_event_update(event);
}

static void
armpmu_stop(struct perf_event *event, int flags)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct hw_perf_event *hwc = &event->hw;

	/*
	 * ARM pmu always has to update the counter, so ignore
	 * PERF_EF_UPDATE, see comments in armpmu_start().
	 */
	if (!(hwc->state & PERF_HES_STOPPED)) {
		armpmu->disable(event);
		armpmu_event_update(event);
		hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
	}
}

static void armpmu_start(struct perf_event *event, int flags)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct hw_perf_event *hwc = &event->hw;

	/*
	 * ARM pmu always has to reprogram the period, so ignore
	 * PERF_EF_RELOAD, see the comment below.
	 */
	if (flags & PERF_EF_RELOAD)
		WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));

	hwc->state = 0;
	/*
	 * Set the period again. Some counters can't be stopped, so when we
	 * were stopped we simply disabled the IRQ source and the counter
	 * may have been left counting. If we don't do this step then we may
	 * get an interrupt too soon or *way* too late if the overflow has
	 * happened since disabling.
	 */
	armpmu_event_set_period(event);
	armpmu->enable(event);
}

static void
armpmu_del(struct perf_event *event, int flags)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
	struct hw_perf_event *hwc = &event->hw;
	int idx = hwc->idx;

	armpmu_stop(event, PERF_EF_UPDATE);
	hw_events->events[idx] = NULL;
	clear_bit(idx, hw_events->used_mask);
	if (armpmu->clear_event_idx)
		armpmu->clear_event_idx(hw_events, event);

	perf_event_update_userpage(event);
}

static int
armpmu_add(struct perf_event *event, int flags)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
	struct hw_perf_event *hwc = &event->hw;
	int idx;

	/* An event following a process won't be stopped earlier */
	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
		return -ENOENT;

	/* If we don't have a space for the counter then finish early. */
	idx = armpmu->get_event_idx(hw_events, event);
	if (idx < 0)
		return idx;

	/*
	 * If there is an event in the counter we are going to use then make
	 * sure it is disabled.
	 */
	event->hw.idx = idx;
	armpmu->disable(event);
	hw_events->events[idx] = event;

	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
	if (flags & PERF_EF_START)
		armpmu_start(event, PERF_EF_RELOAD);

	/* Propagate our changes to the userspace mapping. */
	perf_event_update_userpage(event);

	return 0;
}

static int
validate_event(struct pmu *pmu, struct pmu_hw_events *hw_events,
			       struct perf_event *event)
{
	struct arm_pmu *armpmu;

	if (is_software_event(event))
		return 1;

	/*
	 * Reject groups spanning multiple HW PMUs (e.g. CPU + CCI). The
	 * core perf code won't check that the pmu->ctx == leader->ctx
	 * until after pmu->event_init(event).
	 */
	if (event->pmu != pmu)
		return 0;

	if (event->state < PERF_EVENT_STATE_OFF)
		return 1;

	if (event->state == PERF_EVENT_STATE_OFF && !event->attr.enable_on_exec)
		return 1;

	armpmu = to_arm_pmu(event->pmu);
	return armpmu->get_event_idx(hw_events, event) >= 0;
}

static int
validate_group(struct perf_event *event)
{
	struct perf_event *sibling, *leader = event->group_leader;
	struct pmu_hw_events fake_pmu;

	/*
	 * Initialise the fake PMU. We only need to populate the
	 * used_mask for the purposes of validation.
	 */
	memset(&fake_pmu.used_mask, 0, sizeof(fake_pmu.used_mask));

	if (!validate_event(event->pmu, &fake_pmu, leader))
		return -EINVAL;

	for_each_sibling_event(sibling, leader) {
		if (!validate_event(event->pmu, &fake_pmu, sibling))
			return -EINVAL;
	}

	if (!validate_event(event->pmu, &fake_pmu, event))
		return -EINVAL;

	return 0;
}

static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
{
	struct arm_pmu *armpmu;
	int ret;
	u64 start_clock, finish_clock;

	/*
	 * we request the IRQ with a (possibly percpu) struct arm_pmu**, but
	 * the handlers expect a struct arm_pmu*. The percpu_irq framework will
	 * do any necessary shifting, we just need to perform the first
	 * dereference.
	 */
	armpmu = *(void **)dev;
	if (WARN_ON_ONCE(!armpmu))
		return IRQ_NONE;

	start_clock = sched_clock();
	ret = armpmu->handle_irq(irq, armpmu);
	finish_clock = sched_clock();

	perf_sample_event_took(finish_clock - start_clock);
	return ret;
}

static int
event_requires_mode_exclusion(struct perf_event_attr *attr)
{
	return attr->exclude_idle || attr->exclude_user ||
	       attr->exclude_kernel || attr->exclude_hv;
}

static int
__hw_perf_event_init(struct perf_event *event)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	struct hw_perf_event *hwc = &event->hw;
	int mapping;

	mapping = armpmu->map_event(event);

	if (mapping < 0) {
		pr_debug("event %x:%llx not supported\n", event->attr.type,
			 event->attr.config);
		return mapping;
	}

	/*
	 * We don't assign an index until we actually place the event onto
	 * hardware. Use -1 to signify that we haven't decided where to put it
	 * yet. For SMP systems, each core has it's own PMU so we can't do any
	 * clever allocation or constraints checking at this point.
	 */
	hwc->idx		= -1;
	hwc->config_base	= 0;
	hwc->config		= 0;
	hwc->event_base		= 0;

	/*
	 * Check whether we need to exclude the counter from certain modes.
	 */
	if ((!armpmu->set_event_filter ||
	     armpmu->set_event_filter(hwc, &event->attr)) &&
	     event_requires_mode_exclusion(&event->attr)) {
		pr_debug("ARM performance counters do not support "
			 "mode exclusion\n");
		return -EOPNOTSUPP;
	}

	/*
	 * Store the event encoding into the config_base field.
	 */
	hwc->config_base	    |= (unsigned long)mapping;

	if (!is_sampling_event(event)) {
		/*
		 * For non-sampling runs, limit the sample_period to half
		 * of the counter width. That way, the new counter value
		 * is far less likely to overtake the previous one unless
		 * you have some serious IRQ latency issues.
		 */
		hwc->sample_period  = armpmu->max_period >> 1;
		hwc->last_period    = hwc->sample_period;
		local64_set(&hwc->period_left, hwc->sample_period);
	}

	if (event->group_leader != event) {
		if (validate_group(event) != 0)
			return -EINVAL;
	}

	return 0;
}

static int armpmu_event_init(struct perf_event *event)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);

	/*
	 * Reject CPU-affine events for CPUs that are of a different class to
	 * that which this PMU handles. Process-following events (where
	 * event->cpu == -1) can be migrated between CPUs, and thus we have to
	 * reject them later (in armpmu_add) if they're scheduled on a
	 * different class of CPU.
	 */
	if (event->cpu != -1 &&
		!cpumask_test_cpu(event->cpu, &armpmu->supported_cpus))
		return -ENOENT;

	/* does not support taken branch sampling */
	if (has_branch_stack(event))
		return -EOPNOTSUPP;

	if (armpmu->map_event(event) == -ENOENT)
		return -ENOENT;

	return __hw_perf_event_init(event);
}

static void armpmu_enable(struct pmu *pmu)
{
	struct arm_pmu *armpmu = to_arm_pmu(pmu);
	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
	int enabled = bitmap_weight(hw_events->used_mask, armpmu->num_events);

	/* For task-bound events we may be called on other CPUs */
	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
		return;

	if (enabled)
		armpmu->start(armpmu);
}

static void armpmu_disable(struct pmu *pmu)
{
	struct arm_pmu *armpmu = to_arm_pmu(pmu);

	/* For task-bound events we may be called on other CPUs */
	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
		return;

	armpmu->stop(armpmu);
}

/*
 * In heterogeneous systems, events are specific to a particular
 * microarchitecture, and aren't suitable for another. Thus, only match CPUs of
 * the same microarchitecture.
 */
static int armpmu_filter_match(struct perf_event *event)
{
	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
	unsigned int cpu = smp_processor_id();
	return cpumask_test_cpu(cpu, &armpmu->supported_cpus);
}

static ssize_t armpmu_cpumask_show(struct device *dev,
				   struct device_attribute *attr, char *buf)
{
	struct arm_pmu *armpmu = to_arm_pmu(dev_get_drvdata(dev));
	return cpumap_print_to_pagebuf(true, buf, &armpmu->supported_cpus);
}

static DEVICE_ATTR(cpus, S_IRUGO, armpmu_cpumask_show, NULL);

static struct attribute *armpmu_common_attrs[] = {
	&dev_attr_cpus.attr,
	NULL,
};

static struct attribute_group armpmu_common_attr_group = {
	.attrs = armpmu_common_attrs,
};

/* Set at runtime when we know what CPU type we are. */
static struct arm_pmu *__oprofile_cpu_pmu;

/*
 * Despite the names, these two functions are CPU-specific and are used
 * by the OProfile/perf code.
 */
const char *perf_pmu_name(void)
{
	if (!__oprofile_cpu_pmu)
		return NULL;

	return __oprofile_cpu_pmu->name;
}
EXPORT_SYMBOL_GPL(perf_pmu_name);

int perf_num_counters(void)
{
	int max_events = 0;

	if (__oprofile_cpu_pmu != NULL)
		max_events = __oprofile_cpu_pmu->num_events;

	return max_events;
}
EXPORT_SYMBOL_GPL(perf_num_counters);

static int armpmu_count_irq_users(const int irq)
{
	int cpu, count = 0;

	for_each_possible_cpu(cpu) {
		if (per_cpu(cpu_irq, cpu) == irq)
			count++;
	}

	return count;
}

void armpmu_free_irq(int irq, int cpu)
{
	if (per_cpu(cpu_irq, cpu) == 0)
		return;
	if (WARN_ON(irq != per_cpu(cpu_irq, cpu)))
		return;

	if (!irq_is_percpu_devid(irq))
		free_irq(irq, per_cpu_ptr(&cpu_armpmu, cpu));
	else if (armpmu_count_irq_users(irq) == 1)
		free_percpu_irq(irq, &cpu_armpmu);

	per_cpu(cpu_irq, cpu) = 0;
}

int armpmu_request_irq(int irq, int cpu)
{
	int err = 0;
	const irq_handler_t handler = armpmu_dispatch_irq;
	if (!irq)
		return 0;

	if (!irq_is_percpu_devid(irq)) {
		unsigned long irq_flags;

		err = irq_force_affinity(irq, cpumask_of(cpu));

		if (err && num_possible_cpus() > 1) {
			pr_warn("unable to set irq affinity (irq=%d, cpu=%u)\n",
				irq, cpu);
			goto err_out;
		}

		irq_flags = IRQF_PERCPU |
			    IRQF_NOBALANCING |
			    IRQF_NO_THREAD;

		irq_set_status_flags(irq, IRQ_NOAUTOEN);
		err = request_irq(irq, handler, irq_flags, "arm-pmu",
				  per_cpu_ptr(&cpu_armpmu, cpu));
	} else if (armpmu_count_irq_users(irq) == 0) {
		err = request_percpu_irq(irq, handler, "arm-pmu",
					 &cpu_armpmu);
	}

	if (err)
		goto err_out;

	per_cpu(cpu_irq, cpu) = irq;
	return 0;

err_out:
	pr_err("unable to request IRQ%d for ARM PMU counters\n", irq);
	return err;
}

static int armpmu_get_cpu_irq(struct arm_pmu *pmu, int cpu)
{
	struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
	return per_cpu(hw_events->irq, cpu);
}

/*
 * PMU hardware loses all context when a CPU goes offline.
 * When a CPU is hotplugged back in, since some hardware registers are
 * UNKNOWN at reset, the PMU must be explicitly reset to avoid reading
 * junk values out of them.
 */
static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node)
{
	struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node);
	int irq;

	if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
		return 0;
	if (pmu->reset)
		pmu->reset(pmu);

	per_cpu(cpu_armpmu, cpu) = pmu;

	irq = armpmu_get_cpu_irq(pmu, cpu);
	if (irq) {
		if (irq_is_percpu_devid(irq))
			enable_percpu_irq(irq, IRQ_TYPE_NONE);
		else
			enable_irq(irq);
	}

	return 0;
}

static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)
{
	struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node);
	int irq;

	if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
		return 0;

	irq = armpmu_get_cpu_irq(pmu, cpu);
	if (irq) {
		if (irq_is_percpu_devid(irq))
			disable_percpu_irq(irq);
		else
			disable_irq_nosync(irq);
	}

	per_cpu(cpu_armpmu, cpu) = NULL;

	return 0;
}

#ifdef CONFIG_CPU_PM
static void cpu_pm_pmu_setup(struct arm_pmu *armpmu, unsigned long cmd)
{
	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
	struct perf_event *event;
	int idx;

	for (idx = 0; idx < armpmu->num_events; idx++) {
		/*
		 * If the counter is not used skip it, there is no
		 * need of stopping/restarting it.
		 */
		if (!test_bit(idx, hw_events->used_mask))
			continue;

		event = hw_events->events[idx];

		switch (cmd) {
		case CPU_PM_ENTER:
			/*
			 * Stop and update the counter
			 */
			armpmu_stop(event, PERF_EF_UPDATE);
			break;
		case CPU_PM_EXIT:
		case CPU_PM_ENTER_FAILED:
			 /*
			  * Restore and enable the counter.
			  * armpmu_start() indirectly calls
			  *
			  * perf_event_update_userpage()
			  *
			  * that requires RCU read locking to be functional,
			  * wrap the call within RCU_NONIDLE to make the
			  * RCU subsystem aware this cpu is not idle from
			  * an RCU perspective for the armpmu_start() call
			  * duration.
			  */
			RCU_NONIDLE(armpmu_start(event, PERF_EF_RELOAD));
			break;
		default:
			break;
		}
	}
}

static int cpu_pm_pmu_notify(struct notifier_block *b, unsigned long cmd,
			     void *v)
{
	struct arm_pmu *armpmu = container_of(b, struct arm_pmu, cpu_pm_nb);
	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
	int enabled = bitmap_weight(hw_events->used_mask, armpmu->num_events);

	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
		return NOTIFY_DONE;

	/*
	 * Always reset the PMU registers on power-up even if
	 * there are no events running.
	 */
	if (cmd == CPU_PM_EXIT && armpmu->reset)
		armpmu->reset(armpmu);

	if (!enabled)
		return NOTIFY_OK;

	switch (cmd) {
	case CPU_PM_ENTER:
		armpmu->stop(armpmu);
		cpu_pm_pmu_setup(armpmu, cmd);
		break;
	case CPU_PM_EXIT:
		cpu_pm_pmu_setup(armpmu, cmd);
	case CPU_PM_ENTER_FAILED:
		armpmu->start(armpmu);
		break;
	default:
		return NOTIFY_DONE;
	}

	return NOTIFY_OK;
}

static int cpu_pm_pmu_register(struct arm_pmu *cpu_pmu)
{
	cpu_pmu->cpu_pm_nb.notifier_call = cpu_pm_pmu_notify;
	return cpu_pm_register_notifier(&cpu_pmu->cpu_pm_nb);
}

static void cpu_pm_pmu_unregister(struct arm_pmu *cpu_pmu)
{
	cpu_pm_unregister_notifier(&cpu_pmu->cpu_pm_nb);
}
#else
static inline int cpu_pm_pmu_register(struct arm_pmu *cpu_pmu) { return 0; }
static inline void cpu_pm_pmu_unregister(struct arm_pmu *cpu_pmu) { }
#endif

static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
{
	int err;

	err = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_STARTING,
				       &cpu_pmu->node);
	if (err)
		goto out;

	err = cpu_pm_pmu_register(cpu_pmu);
	if (err)
		goto out_unregister;

	return 0;

out_unregister:
	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING,
					    &cpu_pmu->node);
out:
	return err;
}

static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu)
{
	cpu_pm_pmu_unregister(cpu_pmu);
	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING,
					    &cpu_pmu->node);
}

static struct arm_pmu *__armpmu_alloc(gfp_t flags)
{
	struct arm_pmu *pmu;
	int cpu;

	pmu = kzalloc(sizeof(*pmu), flags);
	if (!pmu) {
		pr_info("failed to allocate PMU device!\n");
		goto out;
	}

	pmu->hw_events = alloc_percpu_gfp(struct pmu_hw_events, flags);
	if (!pmu->hw_events) {
		pr_info("failed to allocate per-cpu PMU data.\n");
		goto out_free_pmu;
	}

	pmu->pmu = (struct pmu) {
		.pmu_enable	= armpmu_enable,
		.pmu_disable	= armpmu_disable,
		.event_init	= armpmu_event_init,
		.add		= armpmu_add,
		.del		= armpmu_del,
		.start		= armpmu_start,
		.stop		= armpmu_stop,
		.read		= armpmu_read,
		.filter_match	= armpmu_filter_match,
		.attr_groups	= pmu->attr_groups,
		/*
		 * This is a CPU PMU potentially in a heterogeneous
		 * configuration (e.g. big.LITTLE). This is not an uncore PMU,
		 * and we have taken ctx sharing into account (e.g. with our
		 * pmu::filter_match callback and pmu::event_init group
		 * validation).
		 */
		.capabilities	= PERF_PMU_CAP_HETEROGENEOUS_CPUS,
	};

	pmu->attr_groups[ARMPMU_ATTR_GROUP_COMMON] =
		&armpmu_common_attr_group;

	for_each_possible_cpu(cpu) {
		struct pmu_hw_events *events;

		events = per_cpu_ptr(pmu->hw_events, cpu);
		raw_spin_lock_init(&events->pmu_lock);
		events->percpu_pmu = pmu;
	}

	return pmu;

out_free_pmu:
	kfree(pmu);
out:
	return NULL;
}

struct arm_pmu *armpmu_alloc(void)
{
	return __armpmu_alloc(GFP_KERNEL);
}

struct arm_pmu *armpmu_alloc_atomic(void)
{
	return __armpmu_alloc(GFP_ATOMIC);
}


void armpmu_free(struct arm_pmu *pmu)
{
	free_percpu(pmu->hw_events);
	kfree(pmu);
}

int armpmu_register(struct arm_pmu *pmu)
{
	int ret;

	ret = cpu_pmu_init(pmu);
	if (ret)
		return ret;

	ret = perf_pmu_register(&pmu->pmu, pmu->name, -1);
	if (ret)
		goto out_destroy;

	if (!__oprofile_cpu_pmu)
		__oprofile_cpu_pmu = pmu;

	pr_info("enabled with %s PMU driver, %d counters available\n",
		pmu->name, pmu->num_events);

	return 0;

out_destroy:
	cpu_pmu_destroy(pmu);
	return ret;
}

static int arm_pmu_hp_init(void)
{
	int ret;

	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_STARTING,
				      "perf/arm/pmu:starting",
				      arm_perf_starting_cpu,
				      arm_perf_teardown_cpu);
	if (ret)
		pr_err("CPU hotplug notifier for ARM PMU could not be registered: %d\n",
		       ret);
	return ret;
}
subsys_initcall(arm_pmu_hp_init);