perf-record.c 7.8 KB
Newer Older
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
#include <sched.h>
#include "evlist.h"
#include "evsel.h"
#include "perf.h"
#include "debug.h"
#include "tests.h"

static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t *maskp)
{
	int i, cpu = -1, nrcpus = 1024;
realloc:
	CPU_ZERO(maskp);

	if (sched_getaffinity(pid, sizeof(*maskp), maskp) == -1) {
		if (errno == EINVAL && nrcpus < (1024 << 8)) {
			nrcpus = nrcpus << 2;
			goto realloc;
		}
		perror("sched_getaffinity");
			return -1;
	}

	for (i = 0; i < nrcpus; i++) {
		if (CPU_ISSET(i, maskp)) {
			if (cpu == -1)
				cpu = i;
			else
				CPU_CLR(i, maskp);
		}
	}

	return cpu;
}

int test__PERF_RECORD(void)
{
37
	struct record_opts opts = {
38 39 40 41
		.target = {
			.uid = UINT_MAX,
			.uses_mmap = true,
		},
42 43 44
		.no_buffering = true,
		.freq	      = 10,
		.mmap_pages   = 256,
45 46 47
	};
	cpu_set_t cpu_mask;
	size_t cpu_mask_size = sizeof(cpu_mask);
48
	struct perf_evlist *evlist = perf_evlist__new_default();
49 50 51 52
	struct perf_evsel *evsel;
	struct perf_sample sample;
	const char *cmd = "sleep";
	const char *argv[] = { cmd, "1", NULL, };
53
	char *bname, *mmap_filename;
54 55 56 57 58 59 60 61
	u64 prev_time = 0;
	bool found_cmd_mmap = false,
	     found_libc_mmap = false,
	     found_vdso_mmap = false,
	     found_ld_mmap = false;
	int err = -1, errs = 0, i, wakeups = 0;
	u32 cpu;
	int total_events = 0, nr_events[PERF_RECORD_MAX] = { 0, };
62
	char sbuf[STRERR_BUFSIZE];
63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86

	if (evlist == NULL || argv == NULL) {
		pr_debug("Not enough memory to create evlist\n");
		goto out;
	}

	/*
	 * Create maps of threads and cpus to monitor. In this case
	 * we start with all threads and cpus (-1, -1) but then in
	 * perf_evlist__prepare_workload we'll fill in the only thread
	 * we're monitoring, the one forked there.
	 */
	err = perf_evlist__create_maps(evlist, &opts.target);
	if (err < 0) {
		pr_debug("Not enough memory to create thread/cpu maps\n");
		goto out_delete_evlist;
	}

	/*
	 * Prepare the workload in argv[] to run, it'll fork it, and then wait
	 * for perf_evlist__start_workload() to exec it. This is done this way
	 * so that we have time to open the evlist (calling sys_perf_event_open
	 * on all the fds) and then mmap them.
	 */
87
	err = perf_evlist__prepare_workload(evlist, &opts.target, argv, false, NULL);
88 89
	if (err < 0) {
		pr_debug("Couldn't run the workload!\n");
90
		goto out_delete_evlist;
91 92 93 94 95 96
	}

	/*
	 * Config the evsels, setting attr->comm on the first one, etc.
	 */
	evsel = perf_evlist__first(evlist);
97 98 99
	perf_evsel__set_sample_bit(evsel, CPU);
	perf_evsel__set_sample_bit(evsel, TID);
	perf_evsel__set_sample_bit(evsel, TIME);
100
	perf_evlist__config(evlist, &opts);
101 102 103

	err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
	if (err < 0) {
104 105
		pr_debug("sched__get_first_possible_cpu: %s\n",
			 strerror_r(errno, sbuf, sizeof(sbuf)));
106
		goto out_delete_evlist;
107 108 109 110 111 112 113 114
	}

	cpu = err;

	/*
	 * So that we can check perf_sample.cpu on all the samples.
	 */
	if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) {
115 116
		pr_debug("sched_setaffinity: %s\n",
			 strerror_r(errno, sbuf, sizeof(sbuf)));
117
		goto out_delete_evlist;
118 119 120 121 122 123 124 125
	}

	/*
	 * Call sys_perf_event_open on all the fds on all the evsels,
	 * grouping them if asked to.
	 */
	err = perf_evlist__open(evlist);
	if (err < 0) {
126 127
		pr_debug("perf_evlist__open: %s\n",
			 strerror_r(errno, sbuf, sizeof(sbuf)));
128
		goto out_delete_evlist;
129 130 131 132 133 134 135 136 137
	}

	/*
	 * mmap the first fd on a given CPU and ask for events for the other
	 * fds in the same CPU to be injected in the same mmap ring buffer
	 * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
	 */
	err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
	if (err < 0) {
138 139
		pr_debug("perf_evlist__mmap: %s\n",
			 strerror_r(errno, sbuf, sizeof(sbuf)));
140
		goto out_delete_evlist;
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
	}

	/*
	 * Now that all is properly set up, enable the events, they will
	 * count just on workload.pid, which will start...
	 */
	perf_evlist__enable(evlist);

	/*
	 * Now!
	 */
	perf_evlist__start_workload(evlist);

	while (1) {
		int before = total_events;

		for (i = 0; i < evlist->nr_mmaps; i++) {
			union perf_event *event;

			while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
				const u32 type = event->header.type;
				const char *name = perf_event__name(type);

				++total_events;
				if (type < PERF_RECORD_MAX)
					nr_events[type]++;

				err = perf_evlist__parse_sample(evlist, event, &sample);
				if (err < 0) {
					if (verbose)
						perf_event__fprintf(event, stderr);
					pr_debug("Couldn't parse sample\n");
173
					goto out_delete_evlist;
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
				}

				if (verbose) {
					pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
					perf_event__fprintf(event, stderr);
				}

				if (prev_time > sample.time) {
					pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n",
						 name, prev_time, sample.time);
					++errs;
				}

				prev_time = sample.time;

				if (sample.cpu != cpu) {
					pr_debug("%s with unexpected cpu, expected %d, got %d\n",
						 name, cpu, sample.cpu);
					++errs;
				}

				if ((pid_t)sample.pid != evlist->workload.pid) {
					pr_debug("%s with unexpected pid, expected %d, got %d\n",
						 name, evlist->workload.pid, sample.pid);
					++errs;
				}

				if ((pid_t)sample.tid != evlist->workload.pid) {
					pr_debug("%s with unexpected tid, expected %d, got %d\n",
						 name, evlist->workload.pid, sample.tid);
					++errs;
				}

				if ((type == PERF_RECORD_COMM ||
				     type == PERF_RECORD_MMAP ||
209
				     type == PERF_RECORD_MMAP2 ||
210 211 212 213 214 215 216 217
				     type == PERF_RECORD_FORK ||
				     type == PERF_RECORD_EXIT) &&
				     (pid_t)event->comm.pid != evlist->workload.pid) {
					pr_debug("%s with unexpected pid/tid\n", name);
					++errs;
				}

				if ((type == PERF_RECORD_COMM ||
218 219
				     type == PERF_RECORD_MMAP ||
				     type == PERF_RECORD_MMAP2) &&
220 221 222 223 224 225 226 227 228 229 230 231 232 233 234
				     event->comm.pid != event->comm.tid) {
					pr_debug("%s with different pid/tid!\n", name);
					++errs;
				}

				switch (type) {
				case PERF_RECORD_COMM:
					if (strcmp(event->comm.comm, cmd)) {
						pr_debug("%s with unexpected comm!\n", name);
						++errs;
					}
					break;
				case PERF_RECORD_EXIT:
					goto found_exit;
				case PERF_RECORD_MMAP:
235 236 237 238 239 240
					mmap_filename = event->mmap.filename;
					goto check_bname;
				case PERF_RECORD_MMAP2:
					mmap_filename = event->mmap2.filename;
				check_bname:
					bname = strrchr(mmap_filename, '/');
241 242 243 244 245 246 247 248
					if (bname != NULL) {
						if (!found_cmd_mmap)
							found_cmd_mmap = !strcmp(bname + 1, cmd);
						if (!found_libc_mmap)
							found_libc_mmap = !strncmp(bname + 1, "libc", 4);
						if (!found_ld_mmap)
							found_ld_mmap = !strncmp(bname + 1, "ld", 2);
					} else if (!found_vdso_mmap)
249
						found_vdso_mmap = !strcmp(mmap_filename, "[vdso]");
250 251 252 253 254 255 256 257 258 259
					break;

				case PERF_RECORD_SAMPLE:
					/* Just ignore samples for now */
					break;
				default:
					pr_debug("Unexpected perf_event->header.type %d!\n",
						 type);
					++errs;
				}
260 261

				perf_evlist__mmap_consume(evlist, i);
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
			}
		}

		/*
		 * We don't use poll here because at least at 3.1 times the
		 * PERF_RECORD_{!SAMPLE} events don't honour
		 * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
		 */
		if (total_events == before && false)
			poll(evlist->pollfd, evlist->nr_fds, -1);

		sleep(1);
		if (++wakeups > 5) {
			pr_debug("No PERF_RECORD_EXIT event!\n");
			break;
		}
	}

found_exit:
	if (nr_events[PERF_RECORD_COMM] > 1) {
		pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
		++errs;
	}

	if (nr_events[PERF_RECORD_COMM] == 0) {
		pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
		++errs;
	}

	if (!found_cmd_mmap) {
		pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
		++errs;
	}

	if (!found_libc_mmap) {
		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
		++errs;
	}

	if (!found_ld_mmap) {
		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
		++errs;
	}

	if (!found_vdso_mmap) {
		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
		++errs;
	}
out_delete_evlist:
	perf_evlist__delete(evlist);
out:
	return (err < 0 || errs > 0) ? -1 : 0;
}