aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/perf_counter/builtin-stat.c
blob: 9711e5524233bf429084d85bcda6c9b3d5255c10 (plain) (blame)
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
/*
 * builtin-stat.c
 *
 * Builtin stat command: Give a precise performance counters summary
 * overview about any workload, CPU or specific PID.
 *
 * Sample output:

   $ perf stat ~/hackbench 10
   Time: 0.104

    Performance counter stats for '/home/mingo/hackbench':

       1255.538611  task clock ticks     #      10.143 CPU utilization factor
             54011  context switches     #       0.043 M/sec
               385  CPU migrations       #       0.000 M/sec
             17755  pagefaults           #       0.014 M/sec
        3808323185  CPU cycles           #    3033.219 M/sec
        1575111190  instructions         #    1254.530 M/sec
          17367895  cache references     #      13.833 M/sec
           7674421  cache misses         #       6.112 M/sec

    Wall-clock time elapsed:   123.786620 msecs

 *
 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
 *
 * Improvements and fixes by:
 *
 *   Arjan van de Ven <arjan@linux.intel.com>
 *   Yanmin Zhang <yanmin.zhang@intel.com>
 *   Wu Fengguang <fengguang.wu@intel.com>
 *   Mike Galbraith <efault@gmx.de>
 *   Paul Mackerras <paulus@samba.org>
 *
 * Released under the GPL v2. (and only v2, not any later version)
 */

#include "perf.h"
#include "builtin.h"
#include "util/util.h"
#include "util/parse-options.h"
#include "util/parse-events.h"

#include <sys/prctl.h>

static struct perf_counter_attr default_attrs[MAX_COUNTERS] = {

  { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_TASK_CLOCK		},
  { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_CONTEXT_SWITCHES	},
  { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_CPU_MIGRATIONS	},
  { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_PAGE_FAULTS	},

  { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_CPU_CYCLES		},
  { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_INSTRUCTIONS	},
  { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_CACHE_REFERENCES	},
  { .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_CACHE_MISSES	},
};

static int			system_wide			=  0;
static int			inherit				=  1;

static int			fd[MAX_NR_CPUS][MAX_COUNTERS];

static int			target_pid			= -1;
static int			nr_cpus				=  0;
static unsigned int		page_size;

static int			scale				=  1;

static const unsigned int default_count[] = {
	1000000,
	1000000,
	  10000,
	  10000,
	1000000,
	  10000,
};

static __u64			event_res[MAX_COUNTERS][3];
static __u64			event_scaled[MAX_COUNTERS];

static __u64			runtime_nsecs;
static __u64			walltime_nsecs;

static void create_perfstat_counter(int counter)
{
	struct perf_counter_attr *attr = attrs + counter;

	if (scale)
		attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
				    PERF_FORMAT_TOTAL_TIME_RUNNING;

	if (system_wide) {
		int cpu;
		for (cpu = 0; cpu < nr_cpus; cpu ++) {
			fd[cpu][counter] = sys_perf_counter_open(attr, -1, cpu, -1, 0);
			if (fd[cpu][counter] < 0) {
				printf("perfstat error: syscall returned with %d (%s)\n",
						fd[cpu][counter], strerror(errno));
				exit(-1);
			}
		}
	} else {
		attr->inherit	= inherit;
		attr->disabled	= 1;

		fd[0][counter] = sys_perf_counter_open(attr, 0, -1, -1, 0);
		if (fd[0][counter] < 0) {
			printf("perfstat error: syscall returned with %d (%s)\n",
					fd[0][counter], strerror(errno));
			exit(-1);
		}
	}
}

/*
 * Does the counter have nsecs as a unit?
 */
static inline int nsec_counter(int counter)
{
	if (attrs[counter].type != PERF_TYPE_SOFTWARE)
		return 0;

	if (attrs[counter].config == PERF_COUNT_CPU_CLOCK)
		return 1;

	if (attrs[counter].config == PERF_COUNT_TASK_CLOCK)
		return 1;

	return 0;
}

/*
 * Read out the results of a single counter:
 */
static void read_counter(int counter)
{
	__u64 *count, single_count[3];
	ssize_t res;
	int cpu, nv;
	int scaled;

	count = event_res[counter];

	count[0] = count[1] = count[2] = 0;

	nv = scale ? 3 : 1;
	for (cpu = 0; cpu < nr_cpus; cpu ++) {
		res = read(fd[cpu][counter], single_count, nv * sizeof(__u64));
		assert(res == nv * sizeof(__u64));

		count[0] += single_count[0];
		if (scale) {
			count[1] += single_count[1];
			count[2] += single_count[2];
		}
	}

	scaled = 0;
	if (scale) {
		if (count[2] == 0) {
			event_scaled[counter] = -1;
			count[0] = 0;
			return;
		}

		if (count[2] < count[1]) {
			event_scaled[counter] = 1;
			count[0] = (unsigned long long)
				((double)count[0] * count[1] / count[2] + 0.5);
		}
	}
	/*
	 * Save the full runtime - to allow normalization during printout:
	 */
	if (attrs[counter].type == PERF_TYPE_SOFTWARE &&
		attrs[counter].config == PERF_COUNT_TASK_CLOCK)
		runtime_nsecs = count[0];
}

/*
 * Print out the results of a single counter:
 */
static void print_counter(int counter)
{
	__u64 *count;
	int scaled;

	count = event_res[counter];
	scaled = event_scaled[counter];

	if (scaled == -1) {
		fprintf(stderr, " %14s  %-20s\n",
			"<not counted>", event_name(counter));
		return;
	}

	if (nsec_counter(counter)) {
		double msecs = (double)count[0] / 1000000;

		fprintf(stderr, " %14.6f  %-20s",
			msecs, event_name(counter));
		if (attrs[counter].type == PERF_TYPE_SOFTWARE &&
			attrs[counter].config == PERF_COUNT_TASK_CLOCK) {

			fprintf(stderr, " # %11.3f CPU utilization factor",
				(double)count[0] / (double)walltime_nsecs);
		}
	} else {
		fprintf(stderr, " %14Ld  %-20s",
			count[0], event_name(counter));
		if (runtime_nsecs)
			fprintf(stderr, " # %11.3f M/sec",
				(double)count[0]/runtime_nsecs*1000.0);
	}
	if (scaled)
		fprintf(stderr, "  (scaled from %.2f%%)",
			(double) count[2] / count[1] * 100);
	fprintf(stderr, "\n");
}

static int do_perfstat(int argc, const char **argv)
{
	unsigned long long t0, t1;
	int counter;
	int status;
	int pid;
	int i;

	if (!system_wide)
		nr_cpus = 1;

	for (counter = 0; counter < nr_counters; counter++)
		create_perfstat_counter(counter);

	/*
	 * Enable counters and exec the command:
	 */
	t0 = rdclock();
	prctl(PR_TASK_PERF_COUNTERS_ENABLE);

	if ((pid = fork()) < 0)
		perror("failed to fork");

	if (!pid) {
		if (execvp(argv[0], (char **)argv)) {
			perror(argv[0]);
			exit(-1);
		}
	}

	while (wait(&status) >= 0)
		;

	prctl(PR_TASK_PERF_COUNTERS_DISABLE);
	t1 = rdclock();

	walltime_nsecs = t1 - t0;

	fflush(stdout);

	fprintf(stderr, "\n");
	fprintf(stderr, " Performance counter stats for \'%s", argv[0]);

	for (i = 1; i < argc; i++)
		fprintf(stderr, " %s", argv[i]);

	fprintf(stderr, "\':\n");
	fprintf(stderr, "\n");

	for (counter = 0; counter < nr_counters; counter++)
		read_counter(counter);

	for (counter = 0; counter < nr_counters; counter++)
		print_counter(counter);


	fprintf(stderr, "\n");
	fprintf(stderr, " Wall-clock time elapsed: %12.6f msecs\n",
			(double)(t1-t0)/1e6);
	fprintf(stderr, "\n");

	return 0;
}

static void skip_signal(int signo)
{
}

static const char * const stat_usage[] = {
	"perf stat [<options>] <command>",
	NULL
};

static char events_help_msg[EVENTS_HELP_MAX];

static const struct option options[] = {
	OPT_CALLBACK('e', "event", NULL, "event",
		     events_help_msg, parse_events),
	OPT_BOOLEAN('i', "inherit", &inherit,
		    "child tasks inherit counters"),
	OPT_INTEGER('p', "pid", &target_pid,
		    "stat events on existing pid"),
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
	OPT_BOOLEAN('l', "scale", &scale,
			    "scale/normalize counters"),
	OPT_END()
};

int cmd_stat(int argc, const char **argv, const char *prefix)
{
	page_size = sysconf(_SC_PAGE_SIZE);

	create_events_help(events_help_msg);

	memcpy(attrs, default_attrs, sizeof(attrs));

	argc = parse_options(argc, argv, options, stat_usage, 0);
	if (!argc)
		usage_with_options(stat_usage, options);

	if (!nr_counters)
		nr_counters = 8;

	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	assert(nr_cpus <= MAX_NR_CPUS);
	assert(nr_cpus >= 0);

	/*
	 * We dont want to block the signals - that would cause
	 * child tasks to inherit that and Ctrl-C would not work.
	 * What we want is for Ctrl-C to work in the exec()-ed
	 * task, but being ignored by perf stat itself:
	 */
	signal(SIGINT,  skip_signal);
	signal(SIGALRM, skip_signal);
	signal(SIGABRT, skip_signal);

	return do_perfstat(argc, argv);
}