8 static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t *maskp)
10 int i, cpu = -1, nrcpus = 1024;
14 if (sched_getaffinity(pid, sizeof(*maskp), maskp) == -1) {
15 if (errno == EINVAL && nrcpus < (1024 << 8)) {
19 perror("sched_getaffinity");
23 for (i = 0; i < nrcpus; i++) {
24 if (CPU_ISSET(i, maskp)) {
35 int test__PERF_RECORD(void)
37 struct perf_record_opts opts = {
47 size_t cpu_mask_size = sizeof(cpu_mask);
48 struct perf_evlist *evlist = perf_evlist__new(NULL, NULL);
49 struct perf_evsel *evsel;
50 struct perf_sample sample;
51 const char *cmd = "sleep";
52 const char *argv[] = { cmd, "1", NULL, };
55 bool found_cmd_mmap = false,
56 found_libc_mmap = false,
57 found_vdso_mmap = false,
58 found_ld_mmap = false;
59 int err = -1, errs = 0, i, wakeups = 0;
61 int total_events = 0, nr_events[PERF_RECORD_MAX] = { 0, };
63 if (evlist == NULL || argv == NULL) {
64 pr_debug("Not enough memory to create evlist\n");
69 * We need at least one evsel in the evlist, use the default
72 err = perf_evlist__add_default(evlist);
74 pr_debug("Not enough memory to create evsel\n");
75 goto out_delete_evlist;
79 * Create maps of threads and cpus to monitor. In this case
80 * we start with all threads and cpus (-1, -1) but then in
81 * perf_evlist__prepare_workload we'll fill in the only thread
82 * we're monitoring, the one forked there.
84 err = perf_evlist__create_maps(evlist, &opts.target);
86 pr_debug("Not enough memory to create thread/cpu maps\n");
87 goto out_delete_evlist;
91 * Prepare the workload in argv[] to run, it'll fork it, and then wait
92 * for perf_evlist__start_workload() to exec it. This is done this way
93 * so that we have time to open the evlist (calling sys_perf_event_open
94 * on all the fds) and then mmap them.
96 err = perf_evlist__prepare_workload(evlist, &opts, argv);
98 pr_debug("Couldn't run the workload!\n");
103 * Config the evsels, setting attr->comm on the first one, etc.
105 evsel = perf_evlist__first(evlist);
106 perf_evsel__set_sample_bit(evsel, CPU);
107 perf_evsel__set_sample_bit(evsel, TID);
108 perf_evsel__set_sample_bit(evsel, TIME);
109 perf_evlist__config(evlist, &opts);
111 err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
113 pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno));
114 goto out_delete_maps;
120 * So that we can check perf_sample.cpu on all the samples.
122 if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) {
123 pr_debug("sched_setaffinity: %s\n", strerror(errno));
124 goto out_delete_maps;
128 * Call sys_perf_event_open on all the fds on all the evsels,
129 * grouping them if asked to.
131 err = perf_evlist__open(evlist);
133 pr_debug("perf_evlist__open: %s\n", strerror(errno));
134 goto out_delete_maps;
138 * mmap the first fd on a given CPU and ask for events for the other
139 * fds in the same CPU to be injected in the same mmap ring buffer
140 * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
142 err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
144 pr_debug("perf_evlist__mmap: %s\n", strerror(errno));
145 goto out_delete_maps;
149 * Now that all is properly set up, enable the events, they will
150 * count just on workload.pid, which will start...
152 perf_evlist__enable(evlist);
157 perf_evlist__start_workload(evlist);
160 int before = total_events;
162 for (i = 0; i < evlist->nr_mmaps; i++) {
163 union perf_event *event;
165 while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
166 const u32 type = event->header.type;
167 const char *name = perf_event__name(type);
170 if (type < PERF_RECORD_MAX)
173 err = perf_evlist__parse_sample(evlist, event, &sample);
176 perf_event__fprintf(event, stderr);
177 pr_debug("Couldn't parse sample\n");
182 pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
183 perf_event__fprintf(event, stderr);
186 if (prev_time > sample.time) {
187 pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n",
188 name, prev_time, sample.time);
192 prev_time = sample.time;
194 if (sample.cpu != cpu) {
195 pr_debug("%s with unexpected cpu, expected %d, got %d\n",
196 name, cpu, sample.cpu);
200 if ((pid_t)sample.pid != evlist->workload.pid) {
201 pr_debug("%s with unexpected pid, expected %d, got %d\n",
202 name, evlist->workload.pid, sample.pid);
206 if ((pid_t)sample.tid != evlist->workload.pid) {
207 pr_debug("%s with unexpected tid, expected %d, got %d\n",
208 name, evlist->workload.pid, sample.tid);
212 if ((type == PERF_RECORD_COMM ||
213 type == PERF_RECORD_MMAP ||
214 type == PERF_RECORD_FORK ||
215 type == PERF_RECORD_EXIT) &&
216 (pid_t)event->comm.pid != evlist->workload.pid) {
217 pr_debug("%s with unexpected pid/tid\n", name);
221 if ((type == PERF_RECORD_COMM ||
222 type == PERF_RECORD_MMAP) &&
223 event->comm.pid != event->comm.tid) {
224 pr_debug("%s with different pid/tid!\n", name);
229 case PERF_RECORD_COMM:
230 if (strcmp(event->comm.comm, cmd)) {
231 pr_debug("%s with unexpected comm!\n", name);
235 case PERF_RECORD_EXIT:
237 case PERF_RECORD_MMAP:
238 bname = strrchr(event->mmap.filename, '/');
241 found_cmd_mmap = !strcmp(bname + 1, cmd);
242 if (!found_libc_mmap)
243 found_libc_mmap = !strncmp(bname + 1, "libc", 4);
245 found_ld_mmap = !strncmp(bname + 1, "ld", 2);
246 } else if (!found_vdso_mmap)
247 found_vdso_mmap = !strcmp(event->mmap.filename, "[vdso]");
250 case PERF_RECORD_SAMPLE:
251 /* Just ignore samples for now */
254 pr_debug("Unexpected perf_event->header.type %d!\n",
262 * We don't use poll here because at least at 3.1 times the
263 * PERF_RECORD_{!SAMPLE} events don't honour
264 * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
266 if (total_events == before && false)
267 poll(evlist->pollfd, evlist->nr_fds, -1);
271 pr_debug("No PERF_RECORD_EXIT event!\n");
277 if (nr_events[PERF_RECORD_COMM] > 1) {
278 pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
282 if (nr_events[PERF_RECORD_COMM] == 0) {
283 pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
287 if (!found_cmd_mmap) {
288 pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
292 if (!found_libc_mmap) {
293 pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
297 if (!found_ld_mmap) {
298 pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
302 if (!found_vdso_mmap) {
303 pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
307 perf_evlist__munmap(evlist);
309 perf_evlist__delete_maps(evlist);
311 perf_evlist__delete(evlist);
313 return (err < 0 || errs > 0) ? -1 : 0;