summaryrefslogtreecommitdiff
path: root/arch/s390/kernel/perf_pai_ext.c
blob: 8ba0f1a3a39dc0b7b4a2a2e300604a1b7968208c (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
// SPDX-License-Identifier: GPL-2.0
/*
 * Performance event support - Processor Activity Instrumentation Extension
 * Facility
 *
 *  Copyright IBM Corp. 2022
 *  Author(s): Thomas Richter <tmricht@linux.ibm.com>
 */
#define KMSG_COMPONENT	"pai_ext"
#define pr_fmt(fmt)	KMSG_COMPONENT ": " fmt

#include <linux/kernel.h>
#include <linux/kernel_stat.h>
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/init.h>
#include <linux/export.h>
#include <linux/io.h>
#include <linux/perf_event.h>
#include <asm/ctlreg.h>
#include <asm/pai.h>
#include <asm/debug.h>

#define	PAIE1_CB_SZ		0x200	/* Size of PAIE1 control block */
#define	PAIE1_CTRBLOCK_SZ	0x400	/* Size of PAIE1 counter blocks */

static debug_info_t *paiext_dbg;
static unsigned int paiext_cnt;	/* Extracted with QPACI instruction */

struct pai_userdata {
	u16 num;
	u64 value;
} __packed;

/* Create the PAI extension 1 control block area.
 * The PAI extension control block 1 is pointed to by lowcore
 * address 0x1508 for each CPU. This control block is 512 bytes in size
 * and requires a 512 byte boundary alignment.
 */
struct paiext_cb {		/* PAI extension 1 control block */
	u64 header;		/* Not used */
	u64 reserved1;
	u64 acc;		/* Addr to analytics counter control block */
	u8 reserved2[488];
} __packed;

struct paiext_map {
	unsigned long *area;		/* Area for CPU to store counters */
	struct pai_userdata *save;	/* Area to store non-zero counters */
	enum paievt_mode mode;		/* Type of event */
	unsigned int active_events;	/* # of PAI Extension users */
	refcount_t refcnt;
	struct perf_event *event;	/* Perf event for sampling */
	struct paiext_cb *paiext_cb;	/* PAI extension control block area */
};

struct paiext_mapptr {
	struct paiext_map *mapptr;
};

static struct paiext_root {		/* Anchor to per CPU data */
	refcount_t refcnt;		/* Overall active events */
	struct paiext_mapptr __percpu *mapptr;
} paiext_root;

/* Free per CPU data when the last event is removed. */
static void paiext_root_free(void)
{
	if (refcount_dec_and_test(&paiext_root.refcnt)) {
		free_percpu(paiext_root.mapptr);
		paiext_root.mapptr = NULL;
	}
}

/* On initialization of first event also allocate per CPU data dynamically.
 * Start with an array of pointers, the array size is the maximum number of
 * CPUs possible, which might be larger than the number of CPUs currently
 * online.
 */
static int paiext_root_alloc(void)
{
	if (!refcount_inc_not_zero(&paiext_root.refcnt)) {
		/* The memory is already zeroed. */
		paiext_root.mapptr = alloc_percpu(struct paiext_mapptr);
		if (!paiext_root.mapptr) {
			/* Returning without refcnt adjustment is ok. The
			 * error code is handled by paiext_alloc() which
			 * decrements refcnt when an event can not be
			 * created.
			 */
			return -ENOMEM;
		}
		refcount_set(&paiext_root.refcnt, 1);
	}
	return 0;
}

/* Protects against concurrent increment of sampler and counter member
 * increments at the same time and prohibits concurrent execution of
 * counting and sampling events.
 * Ensures that analytics counter block is deallocated only when the
 * sampling and counting on that cpu is zero.
 * For details see paiext_alloc().
 */
static DEFINE_MUTEX(paiext_reserve_mutex);

/* Free all memory allocated for event counting/sampling setup */
static void paiext_free(struct paiext_mapptr *mp)
{
	kfree(mp->mapptr->area);
	kfree(mp->mapptr->paiext_cb);
	kvfree(mp->mapptr->save);
	kfree(mp->mapptr);
	mp->mapptr = NULL;
}

/* Release the PMU if event is the last perf event */
static void paiext_event_destroy(struct perf_event *event)
{
	struct paiext_mapptr *mp = per_cpu_ptr(paiext_root.mapptr, event->cpu);
	struct paiext_map *cpump = mp->mapptr;

	mutex_lock(&paiext_reserve_mutex);
	cpump->event = NULL;
	if (refcount_dec_and_test(&cpump->refcnt))	/* Last reference gone */
		paiext_free(mp);
	paiext_root_free();
	mutex_unlock(&paiext_reserve_mutex);
	debug_sprintf_event(paiext_dbg, 4, "%s cpu %d mapptr %p\n", __func__,
			    event->cpu, mp->mapptr);

}

/* Used to avoid races in checking concurrent access of counting and
 * sampling for pai_extension events.
 *
 * Only one instance of event pai_ext/NNPA_ALL/ for sampling is
 * allowed and when this event is running, no counting event is allowed.
 * Several counting events are allowed in parallel, but no sampling event
 * is allowed while one (or more) counting events are running.
 *
 * This function is called in process context and it is safe to block.
 * When the event initialization functions fails, no other call back will
 * be invoked.
 *
 * Allocate the memory for the event.
 */
static int paiext_alloc(struct perf_event_attr *a, struct perf_event *event)
{
	struct paiext_mapptr *mp;
	struct paiext_map *cpump;
	int rc;

	mutex_lock(&paiext_reserve_mutex);

	rc = paiext_root_alloc();
	if (rc)
		goto unlock;

	mp = per_cpu_ptr(paiext_root.mapptr, event->cpu);
	cpump = mp->mapptr;
	if (!cpump) {			/* Paiext_map allocated? */
		rc = -ENOMEM;
		cpump = kzalloc(sizeof(*cpump), GFP_KERNEL);
		if (!cpump)
			goto undo;

		/* Allocate memory for counter area and counter extraction.
		 * These are
		 * - a 512 byte block and requires 512 byte boundary alignment.
		 * - a 1KB byte block and requires 1KB boundary alignment.
		 * Only the first counting event has to allocate the area.
		 *
		 * Note: This works with commit 59bb47985c1d by default.
		 * Backporting this to kernels without this commit might
		 * need adjustment.
		 */
		mp->mapptr = cpump;
		cpump->area = kzalloc(PAIE1_CTRBLOCK_SZ, GFP_KERNEL);
		cpump->paiext_cb = kzalloc(PAIE1_CB_SZ, GFP_KERNEL);
		cpump->save = kvmalloc_array(paiext_cnt + 1,
					     sizeof(struct pai_userdata),
					     GFP_KERNEL);
		if (!cpump->save || !cpump->area || !cpump->paiext_cb) {
			paiext_free(mp);
			goto undo;
		}
		refcount_set(&cpump->refcnt, 1);
		cpump->mode = a->sample_period ? PAI_MODE_SAMPLING
					       : PAI_MODE_COUNTING;
	} else {
		/* Multiple invocation, check what is active.
		 * Supported are multiple counter events or only one sampling
		 * event concurrently at any one time.
		 */
		if (cpump->mode == PAI_MODE_SAMPLING ||
		    (cpump->mode == PAI_MODE_COUNTING && a->sample_period)) {
			rc = -EBUSY;
			goto undo;
		}
		refcount_inc(&cpump->refcnt);
	}

	rc = 0;
	cpump->event = event;

undo:
	if (rc) {
		/* Error in allocation of event, decrement anchor. Since
		 * the event in not created, its destroy() function is never
		 * invoked. Adjust the reference counter for the anchor.
		 */
		paiext_root_free();
	}
unlock:
	mutex_unlock(&paiext_reserve_mutex);
	/* If rc is non-zero, no increment of counter/sampler was done. */
	return rc;
}

/* The PAI extension 1 control block supports up to 128 entries. Return
 * the index within PAIE1_CB given the event number. Also validate event
 * number.
 */
static int paiext_event_valid(struct perf_event *event)
{
	u64 cfg = event->attr.config;

	if (cfg >= PAI_NNPA_BASE && cfg <= PAI_NNPA_BASE + paiext_cnt) {
		/* Offset NNPA in paiext_cb */
		event->hw.config_base = offsetof(struct paiext_cb, acc);
		return 0;
	}
	return -EINVAL;
}

/* Might be called on different CPU than the one the event is intended for. */
static int paiext_event_init(struct perf_event *event)
{
	struct perf_event_attr *a = &event->attr;
	int rc;

	/* PMU pai_ext registered as PERF_TYPE_RAW, check event type */
	if (a->type != PERF_TYPE_RAW && event->pmu->type != a->type)
		return -ENOENT;
	/* PAI extension event must be valid and in supported range */
	rc = paiext_event_valid(event);
	if (rc)
		return rc;
	/* Allow only CPU wide operation, no process context for now. */
	if ((event->attach_state & PERF_ATTACH_TASK) || event->cpu == -1)
		return -ENOENT;
	/* Allow only event NNPA_ALL for sampling. */
	if (a->sample_period && a->config != PAI_NNPA_BASE)
		return -EINVAL;
	/* Prohibit exclude_user event selection */
	if (a->exclude_user)
		return -EINVAL;

	rc = paiext_alloc(a, event);
	if (rc)
		return rc;
	event->hw.last_tag = 0;
	event->destroy = paiext_event_destroy;

	if (a->sample_period) {
		a->sample_period = 1;
		a->freq = 0;
		/* Register for paicrypt_sched_task() to be called */
		event->attach_state |= PERF_ATTACH_SCHED_CB;
		/* Add raw data which are the memory mapped counters */
		a->sample_type |= PERF_SAMPLE_RAW;
		/* Turn off inheritance */
		a->inherit = 0;
	}

	return 0;
}

static u64 paiext_getctr(struct paiext_map *cpump, int nr)
{
	return cpump->area[nr];
}

/* Read the counter values. Return value from location in buffer. For event
 * NNPA_ALL sum up all events.
 */
static u64 paiext_getdata(struct perf_event *event)
{
	struct paiext_mapptr *mp = this_cpu_ptr(paiext_root.mapptr);
	struct paiext_map *cpump = mp->mapptr;
	u64 sum = 0;
	int i;

	if (event->attr.config != PAI_NNPA_BASE)
		return paiext_getctr(cpump, event->attr.config - PAI_NNPA_BASE);

	for (i = 1; i <= paiext_cnt; i++)
		sum += paiext_getctr(cpump, i);

	return sum;
}

static u64 paiext_getall(struct perf_event *event)
{
	return paiext_getdata(event);
}

static void paiext_read(struct perf_event *event)
{
	u64 prev, new, delta;

	prev = local64_read(&event->hw.prev_count);
	new = paiext_getall(event);
	local64_set(&event->hw.prev_count, new);
	delta = new - prev;
	local64_add(delta, &event->count);
}

static void paiext_start(struct perf_event *event, int flags)
{
	u64 sum;

	if (event->hw.last_tag)
		return;
	event->hw.last_tag = 1;
	sum = paiext_getall(event);		/* Get current value */
	local64_set(&event->hw.prev_count, sum);
}

static int paiext_add(struct perf_event *event, int flags)
{
	struct paiext_mapptr *mp = this_cpu_ptr(paiext_root.mapptr);
	struct paiext_map *cpump = mp->mapptr;
	struct paiext_cb *pcb = cpump->paiext_cb;

	if (++cpump->active_events == 1) {
		S390_lowcore.aicd = virt_to_phys(cpump->paiext_cb);
		pcb->acc = virt_to_phys(cpump->area) | 0x1;
		/* Enable CPU instruction lookup for PAIE1 control block */
		local_ctl_set_bit(0, CR0_PAI_EXTENSION_BIT);
		debug_sprintf_event(paiext_dbg, 4, "%s 1508 %llx acc %llx\n",
				    __func__, S390_lowcore.aicd, pcb->acc);
	}
	if (flags & PERF_EF_START && !event->attr.sample_period) {
		/* Only counting needs initial counter value */
		paiext_start(event, PERF_EF_RELOAD);
	}
	event->hw.state = 0;
	if (event->attr.sample_period) {
		cpump->event = event;
		perf_sched_cb_inc(event->pmu);
	}
	return 0;
}

static void paiext_stop(struct perf_event *event, int flags)
{
	paiext_read(event);
	event->hw.state = PERF_HES_STOPPED;
}

static void paiext_del(struct perf_event *event, int flags)
{
	struct paiext_mapptr *mp = this_cpu_ptr(paiext_root.mapptr);
	struct paiext_map *cpump = mp->mapptr;
	struct paiext_cb *pcb = cpump->paiext_cb;

	if (event->attr.sample_period)
		perf_sched_cb_dec(event->pmu);
	if (!event->attr.sample_period) {
		/* Only counting needs to read counter */
		paiext_stop(event, PERF_EF_UPDATE);
	}
	if (--cpump->active_events == 0) {
		/* Disable CPU instruction lookup for PAIE1 control block */
		local_ctl_clear_bit(0, CR0_PAI_EXTENSION_BIT);
		pcb->acc = 0;
		S390_lowcore.aicd = 0;
		debug_sprintf_event(paiext_dbg, 4, "%s 1508 %llx acc %llx\n",
				    __func__, S390_lowcore.aicd, pcb->acc);
	}
}

/* Create raw data and save it in buffer. Returns number of bytes copied.
 * Saves only positive counter entries of the form
 * 2 bytes: Number of counter
 * 8 bytes: Value of counter
 */
static size_t paiext_copy(struct paiext_map *cpump)
{
	struct pai_userdata *userdata = cpump->save;
	int i, outidx = 0;

	for (i = 1; i <= paiext_cnt; i++) {
		u64 val = paiext_getctr(cpump, i);

		if (val) {
			userdata[outidx].num = i;
			userdata[outidx].value = val;
			outidx++;
		}
	}
	return outidx * sizeof(*userdata);
}

/* Write sample when one or more counters values are nonzero.
 *
 * Note: The function paiext_sched_task() and paiext_push_sample() are not
 * invoked after function paiext_del() has been called because of function
 * perf_sched_cb_dec().
 * The function paiext_sched_task() and paiext_push_sample() are only
 * called when sampling is active. Function perf_sched_cb_inc()
 * has been invoked to install function paiext_sched_task() as call back
 * to run at context switch time (see paiext_add()).
 *
 * This causes function perf_event_context_sched_out() and
 * perf_event_context_sched_in() to check whether the PMU has installed an
 * sched_task() callback. That callback is not active after paiext_del()
 * returns and has deleted the event on that CPU.
 */
static int paiext_push_sample(void)
{
	struct paiext_mapptr *mp = this_cpu_ptr(paiext_root.mapptr);
	struct paiext_map *cpump = mp->mapptr;
	struct perf_event *event = cpump->event;
	struct perf_sample_data data;
	struct perf_raw_record raw;
	struct pt_regs regs;
	size_t rawsize;
	int overflow;

	rawsize = paiext_copy(cpump);
	if (!rawsize)			/* No incremented counters */
		return 0;

	/* Setup perf sample */
	memset(&regs, 0, sizeof(regs));
	memset(&raw, 0, sizeof(raw));
	memset(&data, 0, sizeof(data));
	perf_sample_data_init(&data, 0, event->hw.last_period);
	if (event->attr.sample_type & PERF_SAMPLE_TID) {
		data.tid_entry.pid = task_tgid_nr(current);
		data.tid_entry.tid = task_pid_nr(current);
	}
	if (event->attr.sample_type & PERF_SAMPLE_TIME)
		data.time = event->clock();
	if (event->attr.sample_type & (PERF_SAMPLE_ID | PERF_SAMPLE_IDENTIFIER))
		data.id = event->id;
	if (event->attr.sample_type & PERF_SAMPLE_CPU)
		data.cpu_entry.cpu = smp_processor_id();
	if (event->attr.sample_type & PERF_SAMPLE_RAW) {
		raw.frag.size = rawsize;
		raw.frag.data = cpump->save;
		perf_sample_save_raw_data(&data, &raw);
	}

	overflow = perf_event_overflow(event, &data, &regs);
	perf_event_update_userpage(event);
	/* Clear lowcore area after read */
	memset(cpump->area, 0, PAIE1_CTRBLOCK_SZ);
	return overflow;
}

/* Called on schedule-in and schedule-out. No access to event structure,
 * but for sampling only event NNPA_ALL is allowed.
 */
static void paiext_sched_task(struct perf_event_pmu_context *pmu_ctx, bool sched_in)
{
	/* We started with a clean page on event installation. So read out
	 * results on schedule_out and if page was dirty, clear values.
	 */
	if (!sched_in)
		paiext_push_sample();
}

/* Attribute definitions for pai extension1 interface. As with other CPU
 * Measurement Facilities, there is one attribute per mapped counter.
 * The number of mapped counters may vary per machine generation. Use
 * the QUERY PROCESSOR ACTIVITY COUNTER INFORMATION (QPACI) instruction
 * to determine the number of mapped counters. The instructions returns
 * a positive number, which is the highest number of supported counters.
 * All counters less than this number are also supported, there are no
 * holes. A returned number of zero means no support for mapped counters.
 *
 * The identification of the counter is a unique number. The chosen range
 * is 0x1800 + offset in mapped kernel page.
 * All CPU Measurement Facility counters identifiers must be unique and
 * the numbers from 0 to 496 are already used for the CPU Measurement
 * Counter facility. Number 0x1000 to 0x103e are used for PAI cryptography
 * counters.
 * Numbers 0xb0000, 0xbc000 and 0xbd000 are already
 * used for the CPU Measurement Sampling facility.
 */
PMU_FORMAT_ATTR(event, "config:0-63");

static struct attribute *paiext_format_attr[] = {
	&format_attr_event.attr,
	NULL,
};

static struct attribute_group paiext_events_group = {
	.name = "events",
	.attrs = NULL,			/* Filled in attr_event_init() */
};

static struct attribute_group paiext_format_group = {
	.name = "format",
	.attrs = paiext_format_attr,
};

static const struct attribute_group *paiext_attr_groups[] = {
	&paiext_events_group,
	&paiext_format_group,
	NULL,
};

/* Performance monitoring unit for mapped counters */
static struct pmu paiext = {
	.task_ctx_nr  = perf_invalid_context,
	.event_init   = paiext_event_init,
	.add	      = paiext_add,
	.del	      = paiext_del,
	.start	      = paiext_start,
	.stop	      = paiext_stop,
	.read	      = paiext_read,
	.sched_task   = paiext_sched_task,
	.attr_groups  = paiext_attr_groups,
};

/* List of symbolic PAI extension 1 NNPA counter names. */
static const char * const paiext_ctrnames[] = {
	[0] = "NNPA_ALL",
	[1] = "NNPA_ADD",
	[2] = "NNPA_SUB",
	[3] = "NNPA_MUL",
	[4] = "NNPA_DIV",
	[5] = "NNPA_MIN",
	[6] = "NNPA_MAX",
	[7] = "NNPA_LOG",
	[8] = "NNPA_EXP",
	[9] = "NNPA_IBM_RESERVED_9",
	[10] = "NNPA_RELU",
	[11] = "NNPA_TANH",
	[12] = "NNPA_SIGMOID",
	[13] = "NNPA_SOFTMAX",
	[14] = "NNPA_BATCHNORM",
	[15] = "NNPA_MAXPOOL2D",
	[16] = "NNPA_AVGPOOL2D",
	[17] = "NNPA_LSTMACT",
	[18] = "NNPA_GRUACT",
	[19] = "NNPA_CONVOLUTION",
	[20] = "NNPA_MATMUL_OP",
	[21] = "NNPA_MATMUL_OP_BCAST23",
	[22] = "NNPA_SMALLBATCH",
	[23] = "NNPA_LARGEDIM",
	[24] = "NNPA_SMALLTENSOR",
	[25] = "NNPA_1MFRAME",
	[26] = "NNPA_2GFRAME",
	[27] = "NNPA_ACCESSEXCEPT",
};

static void __init attr_event_free(struct attribute **attrs, int num)
{
	struct perf_pmu_events_attr *pa;
	struct device_attribute *dap;
	int i;

	for (i = 0; i < num; i++) {
		dap = container_of(attrs[i], struct device_attribute, attr);
		pa = container_of(dap, struct perf_pmu_events_attr, attr);
		kfree(pa);
	}
	kfree(attrs);
}

static int __init attr_event_init_one(struct attribute **attrs, int num)
{
	struct perf_pmu_events_attr *pa;

	pa = kzalloc(sizeof(*pa), GFP_KERNEL);
	if (!pa)
		return -ENOMEM;

	sysfs_attr_init(&pa->attr.attr);
	pa->id = PAI_NNPA_BASE + num;
	pa->attr.attr.name = paiext_ctrnames[num];
	pa->attr.attr.mode = 0444;
	pa->attr.show = cpumf_events_sysfs_show;
	pa->attr.store = NULL;
	attrs[num] = &pa->attr.attr;
	return 0;
}

/* Create PMU sysfs event attributes on the fly. */
static int __init attr_event_init(void)
{
	struct attribute **attrs;
	int ret, i;

	attrs = kmalloc_array(ARRAY_SIZE(paiext_ctrnames) + 1, sizeof(*attrs),
			      GFP_KERNEL);
	if (!attrs)
		return -ENOMEM;
	for (i = 0; i < ARRAY_SIZE(paiext_ctrnames); i++) {
		ret = attr_event_init_one(attrs, i);
		if (ret) {
			attr_event_free(attrs, i - 1);
			return ret;
		}
	}
	attrs[i] = NULL;
	paiext_events_group.attrs = attrs;
	return 0;
}

static int __init paiext_init(void)
{
	struct qpaci_info_block ib;
	int rc = -ENOMEM;

	if (!test_facility(197))
		return 0;

	qpaci(&ib);
	paiext_cnt = ib.num_nnpa;
	if (paiext_cnt >= PAI_NNPA_MAXCTR)
		paiext_cnt = PAI_NNPA_MAXCTR;
	if (!paiext_cnt)
		return 0;

	rc = attr_event_init();
	if (rc) {
		pr_err("Creation of PMU " KMSG_COMPONENT " /sysfs failed\n");
		return rc;
	}

	/* Setup s390dbf facility */
	paiext_dbg = debug_register(KMSG_COMPONENT, 2, 256, 128);
	if (!paiext_dbg) {
		pr_err("Registration of s390dbf " KMSG_COMPONENT " failed\n");
		rc = -ENOMEM;
		goto out_init;
	}
	debug_register_view(paiext_dbg, &debug_sprintf_view);

	rc = perf_pmu_register(&paiext, KMSG_COMPONENT, -1);
	if (rc) {
		pr_err("Registration of " KMSG_COMPONENT " PMU failed with "
		       "rc=%i\n", rc);
		goto out_pmu;
	}

	return 0;

out_pmu:
	debug_unregister_view(paiext_dbg, &debug_sprintf_view);
	debug_unregister(paiext_dbg);
out_init:
	attr_event_free(paiext_events_group.attrs,
			ARRAY_SIZE(paiext_ctrnames) + 1);
	return rc;
}

device_initcall(paiext_init);