summaryrefslogtreecommitdiffstats
path: root/arch/s390/kernel/perf_cpum_cf.c
blob: e1c54d28713a8d1daaf4f416e4e7b16fc427059b (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
// SPDX-License-Identifier: GPL-2.0
/*
 * Performance event support for s390x - CPU-measurement Counter Facility
 *
 *  Copyright IBM Corp. 2012, 2017
 *  Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
 */
#define KMSG_COMPONENT	"cpum_cf"
#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 <asm/cpu_mcf.h>

static enum cpumf_ctr_set get_counter_set(u64 event)
{
	int set = CPUMF_CTR_SET_MAX;

	if (event < 32)
		set = CPUMF_CTR_SET_BASIC;
	else if (event < 64)
		set = CPUMF_CTR_SET_USER;
	else if (event < 128)
		set = CPUMF_CTR_SET_CRYPTO;
	else if (event < 256)
		set = CPUMF_CTR_SET_EXT;
	else if (event >= 448 && event < 496)
		set = CPUMF_CTR_SET_MT_DIAG;

	return set;
}

static int validate_ctr_version(const struct hw_perf_event *hwc)
{
	struct cpu_cf_events *cpuhw;
	int err = 0;
	u16 mtdiag_ctl;

	cpuhw = &get_cpu_var(cpu_cf_events);

	/* check required version for counter sets */
	switch (hwc->config_base) {
	case CPUMF_CTR_SET_BASIC:
	case CPUMF_CTR_SET_USER:
		if (cpuhw->info.cfvn < 1)
			err = -EOPNOTSUPP;
		break;
	case CPUMF_CTR_SET_CRYPTO:
	case CPUMF_CTR_SET_EXT:
		if (cpuhw->info.csvn < 1)
			err = -EOPNOTSUPP;
		if ((cpuhw->info.csvn == 1 && hwc->config > 159) ||
		    (cpuhw->info.csvn == 2 && hwc->config > 175) ||
		    (cpuhw->info.csvn  > 2 && hwc->config > 255))
			err = -EOPNOTSUPP;
		break;
	case CPUMF_CTR_SET_MT_DIAG:
		if (cpuhw->info.csvn <= 3)
			err = -EOPNOTSUPP;
		/*
		 * MT-diagnostic counters are read-only.  The counter set
		 * is automatically enabled and activated on all CPUs with
		 * multithreading (SMT).  Deactivation of multithreading
		 * also disables the counter set.  State changes are ignored
		 * by lcctl().	Because Linux controls SMT enablement through
		 * a kernel parameter only, the counter set is either disabled
		 * or enabled and active.
		 *
		 * Thus, the counters can only be used if SMT is on and the
		 * counter set is enabled and active.
		 */
		mtdiag_ctl = cpumf_ctr_ctl[CPUMF_CTR_SET_MT_DIAG];
		if (!((cpuhw->info.auth_ctl & mtdiag_ctl) &&
		      (cpuhw->info.enable_ctl & mtdiag_ctl) &&
		      (cpuhw->info.act_ctl & mtdiag_ctl)))
			err = -EOPNOTSUPP;
		break;
	}

	put_cpu_var(cpu_cf_events);
	return err;
}

static int validate_ctr_auth(const struct hw_perf_event *hwc)
{
	struct cpu_cf_events *cpuhw;
	u64 ctrs_state;
	int err = 0;

	cpuhw = &get_cpu_var(cpu_cf_events);

	/* Check authorization for cpu counter sets.
	 * If the particular CPU counter set is not authorized,
	 * return with -ENOENT in order to fall back to other
	 * PMUs that might suffice the event request.
	 */
	ctrs_state = cpumf_ctr_ctl[hwc->config_base];
	if (!(ctrs_state & cpuhw->info.auth_ctl))
		err = -ENOENT;

	put_cpu_var(cpu_cf_events);
	return err;
}

/*
 * Change the CPUMF state to active.
 * Enable and activate the CPU-counter sets according
 * to the per-cpu control state.
 */
static void cpumf_pmu_enable(struct pmu *pmu)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
	int err;

	if (cpuhw->flags & PMU_F_ENABLED)
		return;

	err = lcctl(cpuhw->state);
	if (err) {
		pr_err("Enabling the performance measuring unit "
		       "failed with rc=%x\n", err);
		return;
	}

	cpuhw->flags |= PMU_F_ENABLED;
}

/*
 * Change the CPUMF state to inactive.
 * Disable and enable (inactive) the CPU-counter sets according
 * to the per-cpu control state.
 */
static void cpumf_pmu_disable(struct pmu *pmu)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
	int err;
	u64 inactive;

	if (!(cpuhw->flags & PMU_F_ENABLED))
		return;

	inactive = cpuhw->state & ~((1 << CPUMF_LCCTL_ENABLE_SHIFT) - 1);
	err = lcctl(inactive);
	if (err) {
		pr_err("Disabling the performance measuring unit "
		       "failed with rc=%x\n", err);
		return;
	}

	cpuhw->flags &= ~PMU_F_ENABLED;
}


/* Number of perf events counting hardware events */
static atomic_t num_events = ATOMIC_INIT(0);
/* Used to avoid races in calling reserve/release_cpumf_hardware */
static DEFINE_MUTEX(pmc_reserve_mutex);

/* Release the PMU if event is the last perf event */
static void hw_perf_event_destroy(struct perf_event *event)
{
	if (!atomic_add_unless(&num_events, -1, 1)) {
		mutex_lock(&pmc_reserve_mutex);
		if (atomic_dec_return(&num_events) == 0)
			__kernel_cpumcf_end();
		mutex_unlock(&pmc_reserve_mutex);
	}
}

/* CPUMF <-> perf event mappings for kernel+userspace (basic set) */
static const int cpumf_generic_events_basic[] = {
	[PERF_COUNT_HW_CPU_CYCLES]	    = 0,
	[PERF_COUNT_HW_INSTRUCTIONS]	    = 1,
	[PERF_COUNT_HW_CACHE_REFERENCES]    = -1,
	[PERF_COUNT_HW_CACHE_MISSES]	    = -1,
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = -1,
	[PERF_COUNT_HW_BRANCH_MISSES]	    = -1,
	[PERF_COUNT_HW_BUS_CYCLES]	    = -1,
};
/* CPUMF <-> perf event mappings for userspace (problem-state set) */
static const int cpumf_generic_events_user[] = {
	[PERF_COUNT_HW_CPU_CYCLES]	    = 32,
	[PERF_COUNT_HW_INSTRUCTIONS]	    = 33,
	[PERF_COUNT_HW_CACHE_REFERENCES]    = -1,
	[PERF_COUNT_HW_CACHE_MISSES]	    = -1,
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = -1,
	[PERF_COUNT_HW_BRANCH_MISSES]	    = -1,
	[PERF_COUNT_HW_BUS_CYCLES]	    = -1,
};

static int __hw_perf_event_init(struct perf_event *event)
{
	struct perf_event_attr *attr = &event->attr;
	struct hw_perf_event *hwc = &event->hw;
	enum cpumf_ctr_set set;
	int err = 0;
	u64 ev;

	switch (attr->type) {
	case PERF_TYPE_RAW:
		/* Raw events are used to access counters directly,
		 * hence do not permit excludes */
		if (attr->exclude_kernel || attr->exclude_user ||
		    attr->exclude_hv)
			return -EOPNOTSUPP;
		ev = attr->config;
		break;

	case PERF_TYPE_HARDWARE:
		if (is_sampling_event(event))	/* No sampling support */
			return -ENOENT;
		ev = attr->config;
		/* Count user space (problem-state) only */
		if (!attr->exclude_user && attr->exclude_kernel) {
			if (ev >= ARRAY_SIZE(cpumf_generic_events_user))
				return -EOPNOTSUPP;
			ev = cpumf_generic_events_user[ev];

		/* No support for kernel space counters only */
		} else if (!attr->exclude_kernel && attr->exclude_user) {
			return -EOPNOTSUPP;

		/* Count user and kernel space */
		} else {
			if (ev >= ARRAY_SIZE(cpumf_generic_events_basic))
				return -EOPNOTSUPP;
			ev = cpumf_generic_events_basic[ev];
		}
		break;

	default:
		return -ENOENT;
	}

	if (ev == -1)
		return -ENOENT;

	if (ev > PERF_CPUM_CF_MAX_CTR)
		return -ENOENT;

	/* Obtain the counter set to which the specified counter belongs */
	set = get_counter_set(ev);
	switch (set) {
	case CPUMF_CTR_SET_BASIC:
	case CPUMF_CTR_SET_USER:
	case CPUMF_CTR_SET_CRYPTO:
	case CPUMF_CTR_SET_EXT:
	case CPUMF_CTR_SET_MT_DIAG:
		/*
		 * Use the hardware perf event structure to store the
		 * counter number in the 'config' member and the counter
		 * set number in the 'config_base'.  The counter set number
		 * is then later used to enable/disable the counter(s).
		 */
		hwc->config = ev;
		hwc->config_base = set;
		break;
	case CPUMF_CTR_SET_MAX:
		/* The counter could not be associated to a counter set */
		return -EINVAL;
	};

	/* Initialize for using the CPU-measurement counter facility */
	if (!atomic_inc_not_zero(&num_events)) {
		mutex_lock(&pmc_reserve_mutex);
		if (atomic_read(&num_events) == 0 && __kernel_cpumcf_begin())
			err = -EBUSY;
		else
			atomic_inc(&num_events);
		mutex_unlock(&pmc_reserve_mutex);
	}
	if (err)
		return err;
	event->destroy = hw_perf_event_destroy;

	/* Finally, validate version and authorization of the counter set */
	err = validate_ctr_auth(hwc);
	if (!err)
		err = validate_ctr_version(hwc);

	return err;
}

static int cpumf_pmu_event_init(struct perf_event *event)
{
	int err;

	switch (event->attr.type) {
	case PERF_TYPE_HARDWARE:
	case PERF_TYPE_HW_CACHE:
	case PERF_TYPE_RAW:
		err = __hw_perf_event_init(event);
		break;
	default:
		return -ENOENT;
	}

	if (unlikely(err) && event->destroy)
		event->destroy(event);

	return err;
}

static int hw_perf_event_reset(struct perf_event *event)
{
	u64 prev, new;
	int err;

	do {
		prev = local64_read(&event->hw.prev_count);
		err = ecctr(event->hw.config, &new);
		if (err) {
			if (err != 3)
				break;
			/* The counter is not (yet) available. This
			 * might happen if the counter set to which
			 * this counter belongs is in the disabled
			 * state.
			 */
			new = 0;
		}
	} while (local64_cmpxchg(&event->hw.prev_count, prev, new) != prev);

	return err;
}

static void hw_perf_event_update(struct perf_event *event)
{
	u64 prev, new, delta;
	int err;

	do {
		prev = local64_read(&event->hw.prev_count);
		err = ecctr(event->hw.config, &new);
		if (err)
			return;
	} while (local64_cmpxchg(&event->hw.prev_count, prev, new) != prev);

	delta = (prev <= new) ? new - prev
			      : (-1ULL - prev) + new + 1;	 /* overflow */
	local64_add(delta, &event->count);
}

static void cpumf_pmu_read(struct perf_event *event)
{
	if (event->hw.state & PERF_HES_STOPPED)
		return;

	hw_perf_event_update(event);
}

static void cpumf_pmu_start(struct perf_event *event, int flags)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
	struct hw_perf_event *hwc = &event->hw;

	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
		return;

	if (WARN_ON_ONCE(hwc->config == -1))
		return;

	if (flags & PERF_EF_RELOAD)
		WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));

	hwc->state = 0;

	/* (Re-)enable and activate the counter set */
	ctr_set_enable(&cpuhw->state, hwc->config_base);
	ctr_set_start(&cpuhw->state, hwc->config_base);

	/* The counter set to which this counter belongs can be already active.
	 * Because all counters in a set are active, the event->hw.prev_count
	 * needs to be synchronized.  At this point, the counter set can be in
	 * the inactive or disabled state.
	 */
	hw_perf_event_reset(event);

	/* increment refcount for this counter set */
	atomic_inc(&cpuhw->ctr_set[hwc->config_base]);
}

static void cpumf_pmu_stop(struct perf_event *event, int flags)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
	struct hw_perf_event *hwc = &event->hw;

	if (!(hwc->state & PERF_HES_STOPPED)) {
		/* Decrement reference count for this counter set and if this
		 * is the last used counter in the set, clear activation
		 * control and set the counter set state to inactive.
		 */
		if (!atomic_dec_return(&cpuhw->ctr_set[hwc->config_base]))
			ctr_set_stop(&cpuhw->state, hwc->config_base);
		event->hw.state |= PERF_HES_STOPPED;
	}

	if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
		hw_perf_event_update(event);
		event->hw.state |= PERF_HES_UPTODATE;
	}
}

static int cpumf_pmu_add(struct perf_event *event, int flags)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	/* Check authorization for the counter set to which this
	 * counter belongs.
	 * For group events transaction, the authorization check is
	 * done in cpumf_pmu_commit_txn().
	 */
	if (!(cpuhw->txn_flags & PERF_PMU_TXN_ADD))
		if (validate_ctr_auth(&event->hw))
			return -ENOENT;

	ctr_set_enable(&cpuhw->state, event->hw.config_base);
	event->hw.state = PERF_HES_UPTODATE | PERF_HES_STOPPED;

	if (flags & PERF_EF_START)
		cpumf_pmu_start(event, PERF_EF_RELOAD);

	perf_event_update_userpage(event);

	return 0;
}

static void cpumf_pmu_del(struct perf_event *event, int flags)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	cpumf_pmu_stop(event, PERF_EF_UPDATE);

	/* Check if any counter in the counter set is still used.  If not used,
	 * change the counter set to the disabled state.  This also clears the
	 * content of all counters in the set.
	 *
	 * When a new perf event has been added but not yet started, this can
	 * clear enable control and resets all counters in a set.  Therefore,
	 * cpumf_pmu_start() always has to reenable a counter set.
	 */
	if (!atomic_read(&cpuhw->ctr_set[event->hw.config_base]))
		ctr_set_disable(&cpuhw->state, event->hw.config_base);

	perf_event_update_userpage(event);
}

/*
 * Start group events scheduling transaction.
 * Set flags to perform a single test at commit time.
 *
 * We only support PERF_PMU_TXN_ADD transactions. Save the
 * transaction flags but otherwise ignore non-PERF_PMU_TXN_ADD
 * transactions.
 */
static void cpumf_pmu_start_txn(struct pmu *pmu, unsigned int txn_flags)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	WARN_ON_ONCE(cpuhw->txn_flags);		/* txn already in flight */

	cpuhw->txn_flags = txn_flags;
	if (txn_flags & ~PERF_PMU_TXN_ADD)
		return;

	perf_pmu_disable(pmu);
	cpuhw->tx_state = cpuhw->state;
}

/*
 * Stop and cancel a group events scheduling tranctions.
 * Assumes cpumf_pmu_del() is called for each successful added
 * cpumf_pmu_add() during the transaction.
 */
static void cpumf_pmu_cancel_txn(struct pmu *pmu)
{
	unsigned int txn_flags;
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	WARN_ON_ONCE(!cpuhw->txn_flags);	/* no txn in flight */

	txn_flags = cpuhw->txn_flags;
	cpuhw->txn_flags = 0;
	if (txn_flags & ~PERF_PMU_TXN_ADD)
		return;

	WARN_ON(cpuhw->tx_state != cpuhw->state);

	perf_pmu_enable(pmu);
}

/*
 * Commit the group events scheduling transaction.  On success, the
 * transaction is closed.   On error, the transaction is kept open
 * until cpumf_pmu_cancel_txn() is called.
 */
static int cpumf_pmu_commit_txn(struct pmu *pmu)
{
	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);
	u64 state;

	WARN_ON_ONCE(!cpuhw->txn_flags);	/* no txn in flight */

	if (cpuhw->txn_flags & ~PERF_PMU_TXN_ADD) {
		cpuhw->txn_flags = 0;
		return 0;
	}

	/* check if the updated state can be scheduled */
	state = cpuhw->state & ~((1 << CPUMF_LCCTL_ENABLE_SHIFT) - 1);
	state >>= CPUMF_LCCTL_ENABLE_SHIFT;
	if ((state & cpuhw->info.auth_ctl) != state)
		return -ENOENT;

	cpuhw->txn_flags = 0;
	perf_pmu_enable(pmu);
	return 0;
}

/* Performance monitoring unit for s390x */
static struct pmu cpumf_pmu = {
	.task_ctx_nr  = perf_sw_context,
	.capabilities = PERF_PMU_CAP_NO_INTERRUPT,
	.pmu_enable   = cpumf_pmu_enable,
	.pmu_disable  = cpumf_pmu_disable,
	.event_init   = cpumf_pmu_event_init,
	.add	      = cpumf_pmu_add,
	.del	      = cpumf_pmu_del,
	.start	      = cpumf_pmu_start,
	.stop	      = cpumf_pmu_stop,
	.read	      = cpumf_pmu_read,
	.start_txn    = cpumf_pmu_start_txn,
	.commit_txn   = cpumf_pmu_commit_txn,
	.cancel_txn   = cpumf_pmu_cancel_txn,
};

static int __init cpumf_pmu_init(void)
{
	int rc;

	if (!kernel_cpumcf_avail())
		return -ENODEV;

	cpumf_pmu.attr_groups = cpumf_cf_event_group();
	rc = perf_pmu_register(&cpumf_pmu, "cpum_cf", PERF_TYPE_RAW);
	if (rc)
		pr_err("Registering the cpum_cf PMU failed with rc=%i\n", rc);
	return rc;
}
subsys_initcall(cpumf_pmu_init);