xref: /linux/tools/perf/tests/perf-record.c (revision d0b73b488c55df905ea8faaad079f8535629ed26)
1 #include <sched.h>
2 #include "evlist.h"
3 #include "evsel.h"
4 #include "perf.h"
5 #include "debug.h"
6 #include "tests.h"
7 
8 static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t *maskp)
9 {
10 	int i, cpu = -1, nrcpus = 1024;
11 realloc:
12 	CPU_ZERO(maskp);
13 
14 	if (sched_getaffinity(pid, sizeof(*maskp), maskp) == -1) {
15 		if (errno == EINVAL && nrcpus < (1024 << 8)) {
16 			nrcpus = nrcpus << 2;
17 			goto realloc;
18 		}
19 		perror("sched_getaffinity");
20 			return -1;
21 	}
22 
23 	for (i = 0; i < nrcpus; i++) {
24 		if (CPU_ISSET(i, maskp)) {
25 			if (cpu == -1)
26 				cpu = i;
27 			else
28 				CPU_CLR(i, maskp);
29 		}
30 	}
31 
32 	return cpu;
33 }
34 
35 int test__PERF_RECORD(void)
36 {
37 	struct perf_record_opts opts = {
38 		.target = {
39 			.uid = UINT_MAX,
40 			.uses_mmap = true,
41 		},
42 		.no_delay   = true,
43 		.freq	    = 10,
44 		.mmap_pages = 256,
45 	};
46 	cpu_set_t cpu_mask;
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, };
53 	char *bname;
54 	u64 prev_time = 0;
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;
60 	u32 cpu;
61 	int total_events = 0, nr_events[PERF_RECORD_MAX] = { 0, };
62 
63 	if (evlist == NULL || argv == NULL) {
64 		pr_debug("Not enough memory to create evlist\n");
65 		goto out;
66 	}
67 
68 	/*
69 	 * We need at least one evsel in the evlist, use the default
70 	 * one: "cycles".
71 	 */
72 	err = perf_evlist__add_default(evlist);
73 	if (err < 0) {
74 		pr_debug("Not enough memory to create evsel\n");
75 		goto out_delete_evlist;
76 	}
77 
78 	/*
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.
83 	 */
84 	err = perf_evlist__create_maps(evlist, &opts.target);
85 	if (err < 0) {
86 		pr_debug("Not enough memory to create thread/cpu maps\n");
87 		goto out_delete_evlist;
88 	}
89 
90 	/*
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.
95 	 */
96 	err = perf_evlist__prepare_workload(evlist, &opts, argv);
97 	if (err < 0) {
98 		pr_debug("Couldn't run the workload!\n");
99 		goto out_delete_evlist;
100 	}
101 
102 	/*
103 	 * Config the evsels, setting attr->comm on the first one, etc.
104 	 */
105 	evsel = perf_evlist__first(evlist);
106 	evsel->attr.sample_type |= PERF_SAMPLE_CPU;
107 	evsel->attr.sample_type |= PERF_SAMPLE_TID;
108 	evsel->attr.sample_type |= PERF_SAMPLE_TIME;
109 	perf_evlist__config_attrs(evlist, &opts);
110 
111 	err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
112 	if (err < 0) {
113 		pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno));
114 		goto out_delete_evlist;
115 	}
116 
117 	cpu = err;
118 
119 	/*
120 	 * So that we can check perf_sample.cpu on all the samples.
121 	 */
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_evlist;
125 	}
126 
127 	/*
128 	 * Call sys_perf_event_open on all the fds on all the evsels,
129 	 * grouping them if asked to.
130 	 */
131 	err = perf_evlist__open(evlist);
132 	if (err < 0) {
133 		pr_debug("perf_evlist__open: %s\n", strerror(errno));
134 		goto out_delete_evlist;
135 	}
136 
137 	/*
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)).
141 	 */
142 	err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
143 	if (err < 0) {
144 		pr_debug("perf_evlist__mmap: %s\n", strerror(errno));
145 		goto out_delete_evlist;
146 	}
147 
148 	/*
149 	 * Now that all is properly set up, enable the events, they will
150 	 * count just on workload.pid, which will start...
151 	 */
152 	perf_evlist__enable(evlist);
153 
154 	/*
155 	 * Now!
156 	 */
157 	perf_evlist__start_workload(evlist);
158 
159 	while (1) {
160 		int before = total_events;
161 
162 		for (i = 0; i < evlist->nr_mmaps; i++) {
163 			union perf_event *event;
164 
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);
168 
169 				++total_events;
170 				if (type < PERF_RECORD_MAX)
171 					nr_events[type]++;
172 
173 				err = perf_evlist__parse_sample(evlist, event, &sample);
174 				if (err < 0) {
175 					if (verbose)
176 						perf_event__fprintf(event, stderr);
177 					pr_debug("Couldn't parse sample\n");
178 					goto out_err;
179 				}
180 
181 				if (verbose) {
182 					pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
183 					perf_event__fprintf(event, stderr);
184 				}
185 
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);
189 					++errs;
190 				}
191 
192 				prev_time = sample.time;
193 
194 				if (sample.cpu != cpu) {
195 					pr_debug("%s with unexpected cpu, expected %d, got %d\n",
196 						 name, cpu, sample.cpu);
197 					++errs;
198 				}
199 
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);
203 					++errs;
204 				}
205 
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);
209 					++errs;
210 				}
211 
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);
218 					++errs;
219 				}
220 
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);
225 					++errs;
226 				}
227 
228 				switch (type) {
229 				case PERF_RECORD_COMM:
230 					if (strcmp(event->comm.comm, cmd)) {
231 						pr_debug("%s with unexpected comm!\n", name);
232 						++errs;
233 					}
234 					break;
235 				case PERF_RECORD_EXIT:
236 					goto found_exit;
237 				case PERF_RECORD_MMAP:
238 					bname = strrchr(event->mmap.filename, '/');
239 					if (bname != NULL) {
240 						if (!found_cmd_mmap)
241 							found_cmd_mmap = !strcmp(bname + 1, cmd);
242 						if (!found_libc_mmap)
243 							found_libc_mmap = !strncmp(bname + 1, "libc", 4);
244 						if (!found_ld_mmap)
245 							found_ld_mmap = !strncmp(bname + 1, "ld", 2);
246 					} else if (!found_vdso_mmap)
247 						found_vdso_mmap = !strcmp(event->mmap.filename, "[vdso]");
248 					break;
249 
250 				case PERF_RECORD_SAMPLE:
251 					/* Just ignore samples for now */
252 					break;
253 				default:
254 					pr_debug("Unexpected perf_event->header.type %d!\n",
255 						 type);
256 					++errs;
257 				}
258 			}
259 		}
260 
261 		/*
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.
265 		 */
266 		if (total_events == before && false)
267 			poll(evlist->pollfd, evlist->nr_fds, -1);
268 
269 		sleep(1);
270 		if (++wakeups > 5) {
271 			pr_debug("No PERF_RECORD_EXIT event!\n");
272 			break;
273 		}
274 	}
275 
276 found_exit:
277 	if (nr_events[PERF_RECORD_COMM] > 1) {
278 		pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
279 		++errs;
280 	}
281 
282 	if (nr_events[PERF_RECORD_COMM] == 0) {
283 		pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
284 		++errs;
285 	}
286 
287 	if (!found_cmd_mmap) {
288 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
289 		++errs;
290 	}
291 
292 	if (!found_libc_mmap) {
293 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
294 		++errs;
295 	}
296 
297 	if (!found_ld_mmap) {
298 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
299 		++errs;
300 	}
301 
302 	if (!found_vdso_mmap) {
303 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
304 		++errs;
305 	}
306 out_err:
307 	perf_evlist__munmap(evlist);
308 out_delete_evlist:
309 	perf_evlist__delete(evlist);
310 out:
311 	return (err < 0 || errs > 0) ? -1 : 0;
312 }
313