summaryrefslogtreecommitdiff
path: root/benchmarks/gem_syslatency.c
blob: 7671dc43efe93a3abac96300f77935c07f366c34 (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
/*
 * Copyright © 2016 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 "igt.h"
#include <unistd.h>
#include <stdlib.h>
#include <stdint.h>
#include <stdio.h>
#include <string.h>
#include <fcntl.h>
#include <ftw.h>
#include <inttypes.h>
#include <pthread.h>
#include <sched.h>
#include <signal.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/ioctl.h>
#include <sys/time.h>
#include <time.h>
#include <limits.h>
#include "drm.h"

#include <linux/unistd.h>

#define sigev_notify_thread_id _sigev_un._tid

static volatile int done;

struct gem_busyspin {
	pthread_t thread;
	unsigned long sz;
	unsigned long count;
	bool leak;
	bool interrupts;
};

struct sys_wait {
	pthread_t thread;
	struct igt_mean mean;
};

static void force_low_latency(void)
{
	int32_t target = 0;
	int fd = open("/dev/cpu_dma_latency", O_RDWR);
	if (fd < 0 || write(fd, &target, sizeof(target)) < 0)
		fprintf(stderr,
			"Unable to prevent CPU sleeps and force low latency using /dev/cpu_dma_latency: %s\n",
			strerror(errno));
}

#define LOCAL_I915_EXEC_NO_RELOC (1<<11)
#define LOCAL_I915_EXEC_HANDLE_LUT (1<<12)

#define LOCAL_I915_EXEC_BSD_SHIFT      (13)
#define LOCAL_I915_EXEC_BSD_MASK       (3 << LOCAL_I915_EXEC_BSD_SHIFT)

#define ENGINE_FLAGS  (I915_EXEC_RING_MASK | LOCAL_I915_EXEC_BSD_MASK)

static bool ignore_engine(int fd, unsigned engine)
{
	if (engine == 0)
		return true;

	if (gem_has_bsd2(fd) && engine == I915_EXEC_BSD)
		return true;

	return false;
}

static void *gem_busyspin(void *arg)
{
	const uint32_t bbe = MI_BATCH_BUFFER_END;
	struct gem_busyspin *bs = arg;
	struct drm_i915_gem_execbuffer2 execbuf;
	struct drm_i915_gem_exec_object2 obj[2];
	const unsigned sz =
		bs->sz ? bs->sz + sizeof(bbe) : bs->leak ? 16 << 20 : 4 << 10;
	unsigned engines[16];
	unsigned nengine;
	unsigned engine;
	int fd;

	fd = drm_open_driver(DRIVER_INTEL);

	nengine = 0;
	for_each_engine(fd, engine)
		if (!ignore_engine(fd, engine)) engines[nengine++] = engine;

	memset(obj, 0, sizeof(obj));
	obj[0].handle = gem_create(fd, 4096);
	obj[0].flags = EXEC_OBJECT_WRITE;
	obj[1].handle = gem_create(fd, sz);
	gem_write(fd, obj[1].handle, bs->sz, &bbe, sizeof(bbe));

	memset(&execbuf, 0, sizeof(execbuf));
	if (bs->interrupts) {
		execbuf.buffers_ptr = (uintptr_t)&obj[0];
		execbuf.buffer_count = 2;
	} else {
		execbuf.buffers_ptr = (uintptr_t)&obj[1];
		execbuf.buffer_count = 1;
	}
	execbuf.flags |= LOCAL_I915_EXEC_HANDLE_LUT;
	execbuf.flags |= LOCAL_I915_EXEC_NO_RELOC;
	if (__gem_execbuf(fd, &execbuf)) {
		execbuf.flags = 0;
		gem_execbuf(fd, &execbuf);
	}

	while (!done) {
		for (int n = 0; n < nengine; n++) {
			const int m = rand() % nengine;
			unsigned int tmp = engines[n];
			engines[n] = engines[m];
			engines[m] = tmp;
		}
		for (int n = 0; n < nengine; n++) {
			execbuf.flags &= ~ENGINE_FLAGS;
			execbuf.flags |= engines[n];
			gem_execbuf(fd, &execbuf);
		}
		bs->count += nengine;
		if (bs->leak) {
			gem_madvise(fd, obj[1].handle, I915_MADV_DONTNEED);
			obj[1].handle = gem_create(fd, sz);
			gem_write(fd, obj[1].handle, bs->sz, &bbe, sizeof(bbe));
		}
	}

	close(fd);
	return NULL;
}

static double elapsed(const struct timespec *a, const struct timespec *b)
{
	return 1e9*(b->tv_sec - a->tv_sec) + (b->tv_nsec - a ->tv_nsec);
}

static void *sys_wait(void *arg)
{
	struct sys_wait *w = arg;
	struct sigevent sev;
	timer_t timer;
	sigset_t mask;
	struct timespec now;
#define SIG SIGRTMIN

	sigemptyset(&mask);
	sigaddset(&mask, SIG);
	sigprocmask(SIG_SETMASK, &mask, NULL);

	sev.sigev_notify = SIGEV_SIGNAL | SIGEV_THREAD_ID;
	sev.sigev_notify_thread_id = gettid();
	sev.sigev_signo = SIG;
	timer_create(CLOCK_MONOTONIC, &sev, &timer);

	clock_gettime(CLOCK_MONOTONIC, &now);
	while (!done) {
		struct itimerspec its;
		int sigs;

		its.it_value = now;
		its.it_value.tv_nsec += 100 * 1000;
		its.it_value.tv_nsec += rand() % (NSEC_PER_SEC / 1000);
		if (its.it_value.tv_nsec >= NSEC_PER_SEC) {
			its.it_value.tv_nsec -= NSEC_PER_SEC;
			its.it_value.tv_sec += 1;
		}
		its.it_interval.tv_sec = its.it_interval.tv_nsec = 0;
		timer_settime(timer, TIMER_ABSTIME, &its, NULL);

		sigwait(&mask, &sigs);
		clock_gettime(CLOCK_MONOTONIC, &now);
		igt_mean_add(&w->mean, elapsed(&its.it_value, &now));
	}

	sigprocmask(SIG_UNBLOCK, &mask, NULL);
	timer_delete(timer);

	return NULL;
}

#define PAGE_SIZE 4096
static void *sys_thp_alloc(void *arg)
{
	struct sys_wait *w = arg;
	struct timespec now;

	clock_gettime(CLOCK_MONOTONIC, &now);
	while (!done) {
		const size_t sz = 2 << 20;
		const struct timespec start = now;
		void *ptr;

		ptr = mmap(NULL, sz,
			   PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS,
			   -1, 0);
		assert(ptr != MAP_FAILED);
		madvise(ptr, sz, MADV_HUGEPAGE);
		for (size_t page = 0; page < sz; page += PAGE_SIZE)
			*(volatile uint32_t *)((unsigned char *)ptr + page) = 0;
		munmap(ptr, sz);

		clock_gettime(CLOCK_MONOTONIC, &now);
		igt_mean_add(&w->mean, elapsed(&start, &now));
	}

	return NULL;
}

static void bind_cpu(pthread_attr_t *attr, int cpu)
{
#ifdef __USE_GNU
	cpu_set_t mask;

	if (cpu == -1)
		return;

	CPU_ZERO(&mask);
	CPU_SET(cpu, &mask);

	pthread_attr_setaffinity_np(attr, sizeof(mask), &mask);
#endif
}

static void rtprio(pthread_attr_t *attr, int prio)
{
#ifdef PTHREAD_EXPLICIT_SCHED
	struct sched_param param = { .sched_priority = 99 };
	pthread_attr_setinheritsched(attr, PTHREAD_EXPLICIT_SCHED);
	pthread_attr_setschedpolicy(attr, SCHED_FIFO);
	pthread_attr_setschedparam(attr, &param);
#endif
}

static double l_estimate(igt_stats_t *stats)
{
	if (stats->n_values > 9)
		return igt_stats_get_trimean(stats);
	else if (stats->n_values > 5)
		return igt_stats_get_median(stats);
	else
		return igt_stats_get_mean(stats);
}

static double min_measurement_error(void)
{
	struct timespec start, end;
	int n;

	clock_gettime(CLOCK_MONOTONIC, &start);
	for (n = 0; n < 1024; n++)
		clock_gettime(CLOCK_MONOTONIC, &end);

	return elapsed(&start, &end) / n;
}

static int print_entry(const char *filepath, const struct stat *info,
		       const int typeflag, struct FTW *pathinfo)
{
	int fd;

	fd = open(filepath, O_RDONLY);
	if (fd != -1)  {
		void *ptr;

		ptr = mmap(NULL, info->st_size,
			   PROT_READ, MAP_SHARED | MAP_POPULATE,
			   fd, 0);
		if (ptr != MAP_FAILED)
			munmap(ptr, info->st_size);

		close(fd);
	}

	return 0;
}

static void *background_fs(void *path)
{
	while (1)
		nftw(path, print_entry, 20, FTW_PHYS | FTW_MOUNT);
	return NULL;
}

static unsigned long calibrate_nop(unsigned int target_us,
				   unsigned int tolerance_pct)
{
	const uint32_t bbe = MI_BATCH_BUFFER_END;
	const unsigned int loops = 100;
	struct drm_i915_gem_exec_object2 obj = {};
	struct drm_i915_gem_execbuffer2 eb =
		{ .buffer_count = 1, .buffers_ptr = (uintptr_t)&obj};
	struct timespec t_0, t_end;
	long sz, prev;
	int fd;

	fd = drm_open_driver(DRIVER_INTEL);

	clock_gettime(CLOCK_MONOTONIC, &t_0);

	sz = 256 * 1024;
	do {
		struct timespec t_start;

		obj.handle = gem_create(fd, sz + sizeof(bbe));
		gem_write(fd, obj.handle, sz, &bbe, sizeof(bbe));
		gem_execbuf(fd, &eb);
		gem_sync(fd, obj.handle);

		clock_gettime(CLOCK_MONOTONIC, &t_start);
		for (int loop = 0; loop < loops; loop++)
			gem_execbuf(fd, &eb);
		gem_sync(fd, obj.handle);
		clock_gettime(CLOCK_MONOTONIC, &t_end);

		gem_close(fd, obj.handle);

		prev = sz;
		sz = loops * sz / elapsed(&t_start, &t_end) * 1e3 * target_us;
		sz = ALIGN(sz, sizeof(uint32_t));
	} while (elapsed(&t_0, &t_end) < 5 ||
		 abs(sz - prev) > (sz * tolerance_pct / 100));

	close(fd);

	return sz;
}

int main(int argc, char **argv)
{
	struct gem_busyspin *busy;
	struct sys_wait *wait;
	void *sys_fn = sys_wait;
	pthread_attr_t attr;
	pthread_t bg_fs = 0;
	int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
	igt_stats_t cycles, mean, max;
	double min;
	int time = 10;
	int field = -1;
	int enable_gem_sysbusy = 1;
	bool leak = false;
	bool interrupts = false;
	long batch = 0;
	int n, c;

	while ((c = getopt(argc, argv, "r:t:f:bmni1")) != -1) {
		switch (c) {
		case '1':
			ncpus = 1;
			break;
		case 'n': /* dry run, measure baseline system latency */
			enable_gem_sysbusy = 0;
			break;
		case 'i': /* interrupts ahoy! */
			interrupts = true;
			break;
		case 't':
			/* How long to run the benchmark for (seconds) */
			time = atoi(optarg);
			if (time < 0)
				time = INT_MAX;
			break;
		case 'r':
			/* Duration of each batch (microseconds) */
			batch = atoi(optarg);
			break;
		case 'f':
			/* Select an output field */
			field = atoi(optarg);
			break;
		case 'b':
			pthread_create(&bg_fs, NULL,
				       background_fs, (void *)"/");
			sleep(5);
			break;
		case 'm':
			sys_fn = sys_thp_alloc;
			leak = true;
			break;
		default:
			break;
		}
	}

	/* Prevent CPU sleeps so that busy and idle loads are consistent. */
	force_low_latency();
	min = min_measurement_error();

	if (batch > 0)
		batch = calibrate_nop(batch, 2);
	else
		batch = -batch;

	busy = calloc(ncpus, sizeof(*busy));
	pthread_attr_init(&attr);
	if (enable_gem_sysbusy) {
		for (n = 0; n < ncpus; n++) {
			bind_cpu(&attr, n);
			busy[n].sz = batch;
			busy[n].leak = leak;
			busy[n].interrupts = interrupts;
			pthread_create(&busy[n].thread, &attr,
				       gem_busyspin, &busy[n]);
		}
	}

	wait = calloc(ncpus, sizeof(*wait));
	pthread_attr_init(&attr);
	rtprio(&attr, 99);
	for (n = 0; n < ncpus; n++) {
		igt_mean_init(&wait[n].mean);
		bind_cpu(&attr, n);
		pthread_create(&wait[n].thread, &attr, sys_fn, &wait[n]);
	}

	sleep(time);
	done = 1;

	igt_stats_init_with_size(&cycles, ncpus);
	if (enable_gem_sysbusy) {
		for (n = 0; n < ncpus; n++) {
			pthread_join(busy[n].thread, NULL);
			igt_stats_push(&cycles, busy[n].count);
		}
	}

	igt_stats_init_with_size(&mean, ncpus);
	igt_stats_init_with_size(&max, ncpus);
	for (n = 0; n < ncpus; n++) {
		pthread_join(wait[n].thread, NULL);
		igt_stats_push_float(&mean, wait[n].mean.mean);
		igt_stats_push_float(&max, wait[n].mean.max);
	}
	if (bg_fs) {
		pthread_cancel(bg_fs);
		pthread_join(bg_fs, NULL);
	}

	switch (field) {
	default:
		printf("gem_syslatency: cycles=%.0f, latency mean=%.3fus max=%.0fus\n",
		       igt_stats_get_mean(&cycles),
		       (igt_stats_get_mean(&mean) - min)/ 1000,
		       (l_estimate(&max) - min) / 1000);
		break;
	case 0:
		printf("%.0f\n", igt_stats_get_mean(&cycles));
		break;
	case 1:
		printf("%.3f\n", (igt_stats_get_mean(&mean) - min) / 1000);
		break;
	case 2:
		printf("%.0f\n", (l_estimate(&max) - min) / 1000);
		break;
	}

	return 0;

}