aboutsummaryrefslogtreecommitdiff
path: root/driver/gator_events_perf_pmu.c
blob: 34a6bc78756af0655e40dcd9256aec2b7d68b285 (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
/**
 * Copyright (C) ARM Limited 2010-2013. All rights reserved.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

#include <linux/slab.h>
#include <linux/perf_event.h>
#include "gator.h"

// gator_events_armvX.c is used for Linux 2.6.x
#if GATOR_PERF_PMU_SUPPORT

extern bool event_based_sampling;

#define CNTMAX 16
#define CCI_400 4
// + 1 for the cci-400 cycles counter
#define UCCNT (CCI_400 + 1)

struct gator_attr {
	char name[40];
	unsigned long enabled;
	unsigned long type;
	unsigned long event;
	unsigned long count;
	unsigned long key;
};

static struct gator_attr attrs[CNTMAX];
static int attr_count;
static struct gator_attr uc_attrs[UCCNT];
static int uc_attr_count;

struct gator_event {
	int curr;
	int prev;
	int prev_delta;
	bool zero;
	struct perf_event *pevent;
	struct perf_event_attr *pevent_attr;
};

static DEFINE_PER_CPU(struct gator_event[CNTMAX], events);
static struct gator_event uc_events[UCCNT];
static DEFINE_PER_CPU(int[(CNTMAX + UCCNT)*2], perf_cnt);

static void gator_events_perf_pmu_stop(void);

static int __create_files(struct super_block *sb, struct dentry *root, struct gator_attr *const attr)
{
	struct dentry *dir;

	if (attr->name[0] == '\0') {
		return 0;
	}
	dir = gatorfs_mkdir(sb, root, attr->name);
	if (!dir) {
		return -1;
	}
	gatorfs_create_ulong(sb, dir, "enabled", &attr->enabled);
	gatorfs_create_ulong(sb, dir, "count", &attr->count);
	gatorfs_create_ro_ulong(sb, dir, "key", &attr->key);
	gatorfs_create_ulong(sb, dir, "event", &attr->event);

	return 0;
}

static int gator_events_perf_pmu_create_files(struct super_block *sb, struct dentry *root)
{
	int cnt;

	for (cnt = 0; cnt < attr_count; cnt++) {
		if (__create_files(sb, root, &attrs[cnt]) != 0) {
			return -1;
		}
	}

	for (cnt = 0; cnt < uc_attr_count; cnt++) {
		if (__create_files(sb, root, &uc_attrs[cnt]) != 0) {
			return -1;
		}
	}

	return 0;
}

#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
static void ebs_overflow_handler(struct perf_event *event, int unused, struct perf_sample_data *data, struct pt_regs *regs)
#else
static void ebs_overflow_handler(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs)
#endif
{
	gator_backtrace_handler(regs);
}

#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
static void dummy_handler(struct perf_event *event, int unused, struct perf_sample_data *data, struct pt_regs *regs)
#else
static void dummy_handler(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs)
#endif
{
// Required as perf_event_create_kernel_counter() requires an overflow handler, even though all we do is poll
}

static int gator_events_perf_pmu_read(int **buffer);

static int gator_events_perf_pmu_online(int **buffer, bool migrate)
{
	return gator_events_perf_pmu_read(buffer);
}

static void __online_dispatch(int cpu, bool migrate, struct gator_attr *const attr, struct gator_event *const event)
{
	perf_overflow_handler_t handler;

	event->zero = true;

	if (event->pevent != NULL || event->pevent_attr == 0 || migrate) {
		return;
	}

	if (attr->count > 0) {
		handler = ebs_overflow_handler;
	} else {
		handler = dummy_handler;
	}

#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
	event->pevent = perf_event_create_kernel_counter(event->pevent_attr, cpu, 0, handler);
#else
	event->pevent = perf_event_create_kernel_counter(event->pevent_attr, cpu, 0, handler, 0);
#endif
	if (IS_ERR(event->pevent)) {
		pr_debug("gator: unable to online a counter on cpu %d\n", cpu);
		event->pevent = NULL;
		return;
	}

	if (event->pevent->state != PERF_EVENT_STATE_ACTIVE) {
		pr_debug("gator: inactive counter on cpu %d\n", cpu);
		perf_event_release_kernel(event->pevent);
		event->pevent = NULL;
		return;
	}
}

static void gator_events_perf_pmu_online_dispatch(int cpu, bool migrate)
{
	int cnt;

	cpu = pcpu_to_lcpu(cpu);

	for (cnt = 0; cnt < attr_count; cnt++) {
		__online_dispatch(cpu, migrate, &attrs[cnt], &per_cpu(events, cpu)[cnt]);
	}

	if (cpu == 0) {
		for (cnt = 0; cnt < uc_attr_count; cnt++) {
			__online_dispatch(cpu, migrate, &uc_attrs[cnt], &uc_events[cnt]);
		}
	}
}

static void __offline_dispatch(int cpu, struct gator_event *const event)
{
	struct perf_event *pe = NULL;

	if (event->pevent) {
		pe = event->pevent;
		event->pevent = NULL;
	}

	if (pe) {
		perf_event_release_kernel(pe);
	}
}

static void gator_events_perf_pmu_offline_dispatch(int cpu, bool migrate)
{
	int cnt;

	if (migrate) {
		return;
	}
	cpu = pcpu_to_lcpu(cpu);

	for (cnt = 0; cnt < attr_count; cnt++) {
		__offline_dispatch(cpu, &per_cpu(events, cpu)[cnt]);
	}

	if (cpu == 0) {
		for (cnt = 0; cnt < uc_attr_count; cnt++) {
			__offline_dispatch(cpu, &uc_events[cnt]);
		}
	}
}

static int __check_ebs(struct gator_attr *const attr)
{
	if (attr->count > 0) {
		if (!event_based_sampling) {
			event_based_sampling = true;
		} else {
			printk(KERN_WARNING "gator: Only one ebs counter is allowed\n");
			return -1;
		}
	}

	return 0;
}

static int __start(struct gator_attr *const attr, struct gator_event *const event)
{
	u32 size = sizeof(struct perf_event_attr);

	event->pevent = NULL;
	if (!attr->enabled) {	// Skip disabled counters
		return 0;
	}

	event->prev = 0;
	event->curr = 0;
	event->prev_delta = 0;
	event->pevent_attr = kmalloc(size, GFP_KERNEL);
	if (!event->pevent_attr) {
		gator_events_perf_pmu_stop();
		return -1;
	}

	memset(event->pevent_attr, 0, size);
	event->pevent_attr->type = attr->type;
	event->pevent_attr->size = size;
	event->pevent_attr->config = attr->event;
	event->pevent_attr->sample_period = attr->count;
	event->pevent_attr->pinned = 1;

	return 0;
}

static int gator_events_perf_pmu_start(void)
{
	int cnt, cpu;

	event_based_sampling = false;
	for (cnt = 0; cnt < attr_count; cnt++) {
		if (__check_ebs(&attrs[cnt]) != 0) {
			return -1;
		}
	}

	for (cnt = 0; cnt < uc_attr_count; cnt++) {
		if (__check_ebs(&uc_attrs[cnt]) != 0) {
			return -1;
		}
	}

	for_each_present_cpu(cpu) {
		for (cnt = 0; cnt < attr_count; cnt++) {
			if (__start(&attrs[cnt], &per_cpu(events, cpu)[cnt]) != 0) {
				return -1;
			}
		}
	}

	for (cnt = 0; cnt < uc_attr_count; cnt++) {
		if (__start(&uc_attrs[cnt], &uc_events[cnt]) != 0) {
			return -1;
		}
	}

	return 0;
}

static void __event_stop(struct gator_event *const event)
{
	if (event->pevent_attr) {
		kfree(event->pevent_attr);
		event->pevent_attr = NULL;
	}
}

static void __attr_stop(struct gator_attr *const attr)
{
	attr->enabled = 0;
	attr->event = 0;
	attr->count = 0;
}

static void gator_events_perf_pmu_stop(void)
{
	unsigned int cnt, cpu;

	for_each_present_cpu(cpu) {
		for (cnt = 0; cnt < attr_count; cnt++) {
			__event_stop(&per_cpu(events, cpu)[cnt]);
		}
	}

	for (cnt = 0; cnt < uc_attr_count; cnt++) {
		__event_stop(&uc_events[cnt]);
	}

	for (cnt = 0; cnt < attr_count; cnt++) {
		__attr_stop(&attrs[cnt]);
	}

	for (cnt = 0; cnt < uc_attr_count; cnt++) {
		__attr_stop(&uc_attrs[cnt]);
	}
}

static void __read(int *const len, int cpu, struct gator_attr *const attr, struct gator_event *const event)
{
	int delta;

	struct perf_event *const ev = event->pevent;
	if (ev != NULL && ev->state == PERF_EVENT_STATE_ACTIVE) {
		/* After creating the perf counter in __online_dispatch, there
		 * is a race condition between gator_events_perf_pmu_online and
		 * gator_events_perf_pmu_read. So have
		 * gator_events_perf_pmu_online call gator_events_perf_pmu_read
		 * and in __read check to see if it's the first call after
		 * __online_dispatch and if so, run the online code.
		 */
		if (event->zero) {
			ev->pmu->read(ev);
			event->prev = event->curr = local64_read(&ev->count);
			event->prev_delta = 0;
			per_cpu(perf_cnt, cpu)[(*len)++] = attr->key;
			per_cpu(perf_cnt, cpu)[(*len)++] = 0;
			event->zero = false;
		} else {
			ev->pmu->read(ev);
			event->curr = local64_read(&ev->count);
			delta = event->curr - event->prev;
			if (delta != 0 || delta != event->prev_delta) {
				event->prev_delta = delta;
				event->prev = event->curr;
				per_cpu(perf_cnt, cpu)[(*len)++] = attr->key;
				if (delta < 0) {
					delta *= -1;
				}
				per_cpu(perf_cnt, cpu)[(*len)++] = delta;
			}
		}
	}
}

static int gator_events_perf_pmu_read(int **buffer)
{
	int cnt, len = 0;
	const int cpu = get_logical_cpu();

	for (cnt = 0; cnt < attr_count; cnt++) {
		__read(&len, cpu, &attrs[cnt], &per_cpu(events, cpu)[cnt]);
	}

	if (cpu == 0) {
		for (cnt = 0; cnt < uc_attr_count; cnt++) {
			__read(&len, cpu, &uc_attrs[cnt], &uc_events[cnt]);
		}
	}

	if (buffer) {
		*buffer = per_cpu(perf_cnt, cpu);
	}

	return len;
}

static struct gator_interface gator_events_perf_pmu_interface = {
	.create_files = gator_events_perf_pmu_create_files,
	.start = gator_events_perf_pmu_start,
	.stop = gator_events_perf_pmu_stop,
	.online = gator_events_perf_pmu_online,
	.online_dispatch = gator_events_perf_pmu_online_dispatch,
	.offline_dispatch = gator_events_perf_pmu_offline_dispatch,
	.read = gator_events_perf_pmu_read,
};

static void __attr_init(struct gator_attr *const attr)
{
	attr->name[0] = '\0';
	attr->enabled = 0;
	attr->type = 0;
	attr->event = 0;
	attr->count = 0;
	attr->key = gator_events_get_key();
}

static void gator_events_perf_pmu_cci_init(const int type)
{
	int cnt;

	strncpy(uc_attrs[uc_attr_count].name, "cci-400_ccnt", sizeof(uc_attrs[uc_attr_count].name));
	uc_attrs[uc_attr_count].type = type;
	++uc_attr_count;

	for (cnt = 0; cnt < CCI_400; ++cnt, ++uc_attr_count) {
		struct gator_attr *const attr = &uc_attrs[uc_attr_count];
		snprintf(attr->name, sizeof(attr->name), "cci-400_cnt%d", cnt);
		attr->type = type;
	}
}

static void gator_events_perf_pmu_cpu_init(const struct gator_cpu *const gator_cpu, const int type)
{
	int cnt;

	snprintf(attrs[attr_count].name, sizeof(attrs[attr_count].name), "%s_ccnt", gator_cpu->pmnc_name);
	attrs[attr_count].type = type;
	++attr_count;

	for (cnt = 0; cnt < gator_cpu->pmnc_counters; ++cnt, ++attr_count) {
		struct gator_attr *const attr = &attrs[attr_count];
		snprintf(attr->name, sizeof(attr->name), "%s_cnt%d", gator_cpu->pmnc_name, cnt);
		attr->type = type;
	}
}

int gator_events_perf_pmu_init(void)
{
	struct perf_event_attr pea;
	struct perf_event *pe;
	const struct gator_cpu *gator_cpu;
	int type;
	int cpu;
	int cnt;
	bool found_cpu = false;

	for (cnt = 0; cnt < CNTMAX; cnt++) {
		__attr_init(&attrs[cnt]);
	}
	for (cnt = 0; cnt < UCCNT; cnt++) {
		__attr_init(&uc_attrs[cnt]);
	}

	memset(&pea, 0, sizeof(pea));
	pea.size = sizeof(pea);
	pea.config = 0xFF;
	attr_count = 0;
	uc_attr_count = 0;
	for (type = PERF_TYPE_MAX; type < 0x20; ++type) {
		pea.type = type;

		// A particular PMU may work on some but not all cores, so try on each core
		pe = NULL;
		for_each_present_cpu(cpu) {
#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
			pe = perf_event_create_kernel_counter(&pea, cpu, 0, dummy_handler);
#else
			pe = perf_event_create_kernel_counter(&pea, cpu, 0, dummy_handler, 0);
#endif
			if (!IS_ERR(pe)) {
				break;
			}
		}
		// Assume that valid PMUs are contigious
		if (IS_ERR(pe)) {
			break;
		}

		if (pe->pmu != NULL && type == pe->pmu->type) {
			if (strcmp("CCI", pe->pmu->name) == 0) {
				gator_events_perf_pmu_cci_init(type);
			} else if ((gator_cpu = gator_find_cpu_by_pmu_name(pe->pmu->name)) != NULL) {
				found_cpu = true;
				gator_events_perf_pmu_cpu_init(gator_cpu, type);
			}
		}

		perf_event_release_kernel(pe);
	}

	if (!found_cpu) {
		const struct gator_cpu *const gator_cpu = gator_find_cpu_by_cpuid(gator_cpuid());
		if (gator_cpu == NULL) {
			return -1;
		}
		gator_events_perf_pmu_cpu_init(gator_cpu, PERF_TYPE_RAW);
	}

	if (attr_count > CNTMAX) {
		printk(KERN_ERR "gator: Too many perf counters\n");
		return -1;
	}

	if (uc_attr_count > UCCNT) {
		printk(KERN_ERR "gator: Too many perf uncore counters\n");
		return -1;
	}

	return gator_events_install(&gator_events_perf_pmu_interface);
}

gator_events_init(gator_events_perf_pmu_init);
#endif