File: gpu-perf.c

package info (click to toggle)
intel-gpu-tools 1.22-1
  • links: PTS, VCS
  • area: main
  • in suites: bullseye, sid
  • size: 8,408 kB
  • sloc: ansic: 145,957; yacc: 2,780; perl: 1,200; makefile: 878; sh: 525; lex: 487; xml: 404; python: 257
file content (509 lines) | stat: -rw-r--r-- 12,218 bytes parent folder | download
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
/*
 * Copyright © 2013 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 */

#include <stdint.h>
#include <stdbool.h>
#include <sys/types.h>
#include <sys/mman.h>
#include <sys/ioctl.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <fcntl.h>
#include <errno.h>

#include "igt_perf.h"

#include "gpu-perf.h"
#include "debugfs.h"

#if defined(__i386__)
#define rmb()           asm volatile("lock; addl $0,0(%%esp)" ::: "memory")
#define wmb()           asm volatile("lock; addl $0,0(%%esp)" ::: "memory")
#endif

#if defined(__x86_64__)
#define rmb()           asm volatile("lfence" ::: "memory")
#define wmb()           asm volatile("sfence" ::: "memory")
#endif

#define N_PAGES 32

struct sample_event {
	struct perf_event_header header;
	uint32_t pid, tid;
	uint64_t time;
	uint64_t id;
	uint32_t raw_size;
	uint8_t tracepoint_data[0];
};

enum {
	TP_GEM_REQUEST_ADD,
	TP_GEM_REQUEST_WAIT_BEGIN,
	TP_GEM_REQUEST_WAIT_END,
	TP_FLIP_COMPLETE,
	TP_GEM_RING_SYNC_TO,
	TP_GEM_RING_SWITCH_CONTEXT,

	TP_NB
};

struct tracepoint {
	const char *name;
	int event_id;

	struct {
		char name[128];
		int offset;
		int size;
		int is_signed;
	} fields[20];
	int n_fields;

	int device_field;
	int ctx_field;
	int ring_field;
	int seqno_field;
	int global_seqno_field;
	int plane_field;
} tracepoints[TP_NB] = {
	[TP_GEM_REQUEST_ADD]         = { .name = "i915/i915_gem_request_add", },
	[TP_GEM_REQUEST_WAIT_BEGIN]  = { .name = "i915/i915_gem_request_wait_begin", },
	[TP_GEM_REQUEST_WAIT_END]    = { .name = "i915/i915_gem_request_wait_end", },
	[TP_FLIP_COMPLETE]           = { .name = "i915/flip_complete", },
	[TP_GEM_RING_SYNC_TO]        = { .name = "i915/gem_ring_sync_to", },
	[TP_GEM_RING_SWITCH_CONTEXT] = { .name = "i915/gem_ring_switch_context", },
};

union parser_value {
    char *string;
    int integer;
};

struct parser_ctx {
	struct tracepoint *tp;
	FILE *fp;
};

#define YY_CTX_LOCAL
#define YY_CTX_MEMBERS struct parser_ctx ctx;
#define YYSTYPE union parser_value
#define YY_LOCAL(T) static __attribute__((unused)) T
#define YY_PARSE(T) static T
#define YY_INPUT(yy, buf, result, max)				\
	{							\
		int c = getc(yy->ctx.fp);			\
		result = (EOF == c) ? 0 : (*(buf)= c, 1);	\
		if (EOF != c) {					\
			yyprintf((stderr, "<%c>", c));		\
		}						\
	}

#include "tracepoint_format.h"

static int
tracepoint_id(int tp_id)
{
	struct tracepoint *tp = &tracepoints[tp_id];
	yycontext ctx;
	char buf[1024];

	/* Already parsed? */
	if (tp->event_id != 0)
		return tp->event_id;

	snprintf(buf, sizeof(buf), "%s/tracing/events/%s/format",
		 debugfs_path, tp->name);

	memset(&ctx, 0, sizeof(ctx));
	ctx.ctx.tp = tp;
	ctx.ctx.fp = fopen(buf, "r");

	if (ctx.ctx.fp == NULL)
		return 0;

	if (yyparse(&ctx)) {
		for (int f = 0; f < tp->n_fields; f++) {
			if (!strcmp(tp->fields[f].name, "device")) {
				tp->device_field = f;
			} else if (!strcmp(tp->fields[f].name, "ctx")) {
				tp->ctx_field = f;
			} else if (!strcmp(tp->fields[f].name, "ring")) {
				tp->ring_field = f;
			} else if (!strcmp(tp->fields[f].name, "seqno")) {
				tp->seqno_field = f;
			} else if (!strcmp(tp->fields[f].name, "global_seqno")) {
				tp->global_seqno_field = f;
			} else if (!strcmp(tp->fields[f].name, "plane")) {
				tp->plane_field = f;
			}
		}
	} else
		tp->event_id = tp->n_fields = 0;

	yyrelease(&ctx);
	fclose(ctx.ctx.fp);

	return tp->event_id;
}

#define READ_TP_FIELD_U32(sample, tp_id, field_name)			\
	(*(const uint32_t *)((sample)->tracepoint_data +		\
			     tracepoints[tp_id].fields[			\
				     tracepoints[tp_id].field_name##_field].offset))

static int perf_tracepoint_open(struct gpu_perf *gp, int tp_id,
				int (*func)(struct gpu_perf *, const void *))
{
	struct perf_event_attr attr;
	struct gpu_perf_sample *sample;
	int n, *fd;

	memset(&attr, 0, sizeof (attr));

	attr.type = PERF_TYPE_TRACEPOINT;
	attr.config = tracepoint_id(tp_id);
	if (attr.config == 0)
		return ENOENT;

	attr.sample_period = 1;
	attr.sample_type = (PERF_SAMPLE_TIME | PERF_SAMPLE_STREAM_ID | PERF_SAMPLE_TID | PERF_SAMPLE_RAW);
	attr.read_format = PERF_FORMAT_ID;

	attr.exclude_guest = 1;

	n = gp->nr_cpus * (gp->nr_events+1);
	fd = realloc(gp->fd, n*sizeof(int));
	sample = realloc(gp->sample, n*sizeof(*gp->sample));
	if (fd == NULL || sample == NULL)
		return ENOMEM;
	gp->fd = fd;
	gp->sample = sample;

	fd += gp->nr_events * gp->nr_cpus;
	sample += gp->nr_events * gp->nr_cpus;
	for (n = 0; n < gp->nr_cpus; n++) {
		uint64_t track[2];

		fd[n] = perf_event_open(&attr, -1, n, -1, 0);
		if (fd[n] == -1)
			return errno;

		/* read back the event to establish id->tracepoint */
		if (read(fd[n], track, sizeof(track)) < 0)
			return errno;
		sample[n].id = track[1];
		sample[n].func = func;
	}

	gp->nr_events++;
	return 0;
}

static int perf_mmap(struct gpu_perf *gp)
{
	int size = (1 + N_PAGES) * gp->page_size;
	int *fd, i, j;

	gp->map = malloc(sizeof(void *)*gp->nr_cpus);
	if (gp->map == NULL)
		return ENOMEM;

	fd = gp->fd;
	for (j = 0; j < gp->nr_cpus; j++) {
		gp->map[j] = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, *fd++, 0);
		if (gp->map[j] == (void *)-1)
			goto err;
	}

	for (i = 1; i < gp->nr_events; i++) {
		for (j = 0; j < gp->nr_cpus; j++)
			ioctl(*fd++, PERF_EVENT_IOC_SET_OUTPUT, gp->fd[j]);
	}

	return 0;

err:
	while (--j > 0)
		munmap(gp->map[j], size);
	free(gp->map);
	gp->map = NULL;
	return EINVAL;
}

static int get_comm(pid_t pid, char *comm, int len)
{
	char filename[1024];
	int fd;

	*comm = '\0';
	snprintf(filename, sizeof(filename), "/proc/%d/comm", pid);

	fd = open(filename, 0);
	if (fd >= 0) {
		len = read(fd, comm, len-1);
		if (len >= 0)
			comm[len-1] = '\0';
		close(fd);
	} else
		len = -1;

	return len;
}

static struct gpu_perf_comm *
lookup_comm(struct gpu_perf *gp, pid_t pid)
{
	struct gpu_perf_comm *comm;

	if (pid == 0)
		return NULL;

	for (comm = gp->comm; comm != NULL; comm = comm->next) {
		if (comm->pid == pid)
			break;
	}
	if (comm == NULL) {
		comm = calloc(1, sizeof(*comm));
		if (comm == NULL)
			return NULL;

		if (get_comm(pid, comm->name, sizeof(comm->name)) < 0) {
			free(comm);
			return NULL;
		}

		comm->pid = pid;
		comm->next = gp->comm;
		gp->comm = comm;
	}

	return comm;
}

static int request_add(struct gpu_perf *gp, const void *event)
{
	const struct sample_event *sample = event;
	struct gpu_perf_comm *comm;

	comm = lookup_comm(gp, sample->pid);
	if (comm == NULL)
		return 0;

	comm->nr_requests[READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_ADD, ring)]++;
	return 1;
}

static int flip_complete(struct gpu_perf *gp, const void *event)
{
	const struct sample_event *sample = event;

	gp->flip_complete[READ_TP_FIELD_U32(sample, TP_FLIP_COMPLETE, plane)]++;
	return 1;
}

static int ctx_switch(struct gpu_perf *gp, const void *event)
{
	const struct sample_event *sample = event;

	gp->ctx_switch[READ_TP_FIELD_U32(sample, TP_GEM_RING_SWITCH_CONTEXT, ring)]++;
	return 1;
}

static int ring_sync(struct gpu_perf *gp, const void *event)
{
	const struct sample_event *sample = event;
	struct gpu_perf_comm *comm;

	comm = lookup_comm(gp, sample->pid);
	if (comm == NULL)
		return 0;

	comm->nr_sema++;
	return 1;
}

static int wait_begin(struct gpu_perf *gp, const void *event)
{
	const struct sample_event *sample = event;
	struct gpu_perf_comm *comm;
	struct gpu_perf_time *wait;

	comm = lookup_comm(gp, sample->pid);
	if (comm == NULL)
		return 0;

	wait = malloc(sizeof(*wait));
	if (wait == NULL)
		return 0;

	/* XXX argument order CTX == ENGINE! */

	wait->comm = comm;
	wait->comm->active = true;
	wait->context = READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_BEGIN, ctx);
	wait->seqno = READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_BEGIN, seqno);
	wait->time = sample->time;
	wait->next = gp->wait[READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_BEGIN, ring)];
	gp->wait[READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_BEGIN, ring)] = wait;

	return 0;
}

static int wait_end(struct gpu_perf *gp, const void *event)
{
	const struct sample_event *sample = event;
	struct gpu_perf_time *wait, **prev;
	uint32_t engine = READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_END, ring);
	uint32_t context = READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_END, ctx);
	uint32_t seqno = READ_TP_FIELD_U32(sample, TP_GEM_REQUEST_WAIT_END, seqno);

	for (prev = &gp->wait[engine]; (wait = *prev) != NULL; prev = &wait->next) {
		if (wait->context != context || wait->seqno != seqno)
			continue;

		wait->comm->wait_time += sample->time - wait->time;
		wait->comm->active = false;

		*prev = wait->next;
		free(wait);
		return 1;
	}

	return 0;
}

void gpu_perf_init(struct gpu_perf *gp, unsigned flags)
{
	memset(gp, 0, sizeof(*gp));
	gp->nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	gp->page_size = getpagesize();

	perf_tracepoint_open(gp, TP_GEM_REQUEST_ADD, request_add);
	if (perf_tracepoint_open(gp, TP_GEM_REQUEST_WAIT_BEGIN, wait_begin) == 0)
		perf_tracepoint_open(gp, TP_GEM_REQUEST_WAIT_END, wait_end);
	perf_tracepoint_open(gp, TP_FLIP_COMPLETE, flip_complete);
	perf_tracepoint_open(gp, TP_GEM_RING_SYNC_TO, ring_sync);
	perf_tracepoint_open(gp, TP_GEM_RING_SWITCH_CONTEXT, ctx_switch);

	if (gp->nr_events == 0) {
		gp->error = "i915.ko tracepoints not available";
		return;
	}

	if (perf_mmap(gp))
		return;
}

static int process_sample(struct gpu_perf *gp, int cpu,
			  const struct perf_event_header *header)
{
	const struct sample_event *sample = (const struct sample_event *)header;
	int n, update = 0;

	/* hash me! */
	for (n = 0; n < gp->nr_events; n++) {
		int m = n * gp->nr_cpus + cpu;
		if (gp->sample[m].id != sample->id)
			continue;

		update = gp->sample[m].func(gp, sample);
		break;
	}

	return update;
}

int gpu_perf_update(struct gpu_perf *gp)
{
	const int size = N_PAGES * gp->page_size;
	const int mask = size - 1;
	uint8_t *buffer = NULL;
	int buffer_size = 0;
	int n, update = 0;

	if (gp->map == NULL)
		return 0;

	for (n = 0; n < gp->nr_cpus; n++) {
		struct perf_event_mmap_page *mmap = gp->map[n];
		const uint8_t *data;
		uint64_t head, tail;
		int wrap = 0;

		tail = mmap->data_tail;
		head = mmap->data_head;
		rmb();

		if (head < tail) {
			wrap = 1;
			tail &= mask;
			head &= mask;
			head += size;
		}

		data = (uint8_t *)mmap + gp->page_size;
		while (head - tail >= sizeof (struct perf_event_header)) {
			const struct perf_event_header *header;

			header = (const struct perf_event_header *)(data + (tail & mask));
			if (header->size > head - tail)
				break;

			if ((const uint8_t *)header + header->size > data + size) {
				int before;

				if (header->size > buffer_size) {
					uint8_t *b = realloc(buffer, header->size);
					if (b == NULL)
						break;

					buffer = b;
					buffer_size = header->size;
				}

				before = data + size - (const uint8_t *)header;

				memcpy(buffer, header, before);
				memcpy(buffer + before, data, header->size - before);

				header = (struct perf_event_header *)buffer;
			}

			if (header->type == PERF_RECORD_SAMPLE)
				update += process_sample(gp, n, header);
			tail += header->size;
		}

		if (wrap)
			tail &= mask;
		mmap->data_tail = tail;
		wmb();
	}

	free(buffer);
	return update;
}