xref: /openbmc/linux/tools/perf/tests/perf-record.c (revision 275876e2)
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 record_opts opts = {
38 		.target = {
39 			.uid = UINT_MAX,
40 			.uses_mmap = true,
41 		},
42 		.no_buffering = 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_default();
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, *mmap_filename;
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 	 * Create maps of threads and cpus to monitor. In this case
70 	 * we start with all threads and cpus (-1, -1) but then in
71 	 * perf_evlist__prepare_workload we'll fill in the only thread
72 	 * we're monitoring, the one forked there.
73 	 */
74 	err = perf_evlist__create_maps(evlist, &opts.target);
75 	if (err < 0) {
76 		pr_debug("Not enough memory to create thread/cpu maps\n");
77 		goto out_delete_evlist;
78 	}
79 
80 	/*
81 	 * Prepare the workload in argv[] to run, it'll fork it, and then wait
82 	 * for perf_evlist__start_workload() to exec it. This is done this way
83 	 * so that we have time to open the evlist (calling sys_perf_event_open
84 	 * on all the fds) and then mmap them.
85 	 */
86 	err = perf_evlist__prepare_workload(evlist, &opts.target, argv, false, NULL);
87 	if (err < 0) {
88 		pr_debug("Couldn't run the workload!\n");
89 		goto out_delete_evlist;
90 	}
91 
92 	/*
93 	 * Config the evsels, setting attr->comm on the first one, etc.
94 	 */
95 	evsel = perf_evlist__first(evlist);
96 	perf_evsel__set_sample_bit(evsel, CPU);
97 	perf_evsel__set_sample_bit(evsel, TID);
98 	perf_evsel__set_sample_bit(evsel, TIME);
99 	perf_evlist__config(evlist, &opts);
100 
101 	err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
102 	if (err < 0) {
103 		pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno));
104 		goto out_delete_evlist;
105 	}
106 
107 	cpu = err;
108 
109 	/*
110 	 * So that we can check perf_sample.cpu on all the samples.
111 	 */
112 	if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) {
113 		pr_debug("sched_setaffinity: %s\n", strerror(errno));
114 		goto out_delete_evlist;
115 	}
116 
117 	/*
118 	 * Call sys_perf_event_open on all the fds on all the evsels,
119 	 * grouping them if asked to.
120 	 */
121 	err = perf_evlist__open(evlist);
122 	if (err < 0) {
123 		pr_debug("perf_evlist__open: %s\n", strerror(errno));
124 		goto out_delete_evlist;
125 	}
126 
127 	/*
128 	 * mmap the first fd on a given CPU and ask for events for the other
129 	 * fds in the same CPU to be injected in the same mmap ring buffer
130 	 * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
131 	 */
132 	err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
133 	if (err < 0) {
134 		pr_debug("perf_evlist__mmap: %s\n", strerror(errno));
135 		goto out_delete_evlist;
136 	}
137 
138 	/*
139 	 * Now that all is properly set up, enable the events, they will
140 	 * count just on workload.pid, which will start...
141 	 */
142 	perf_evlist__enable(evlist);
143 
144 	/*
145 	 * Now!
146 	 */
147 	perf_evlist__start_workload(evlist);
148 
149 	while (1) {
150 		int before = total_events;
151 
152 		for (i = 0; i < evlist->nr_mmaps; i++) {
153 			union perf_event *event;
154 
155 			while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
156 				const u32 type = event->header.type;
157 				const char *name = perf_event__name(type);
158 
159 				++total_events;
160 				if (type < PERF_RECORD_MAX)
161 					nr_events[type]++;
162 
163 				err = perf_evlist__parse_sample(evlist, event, &sample);
164 				if (err < 0) {
165 					if (verbose)
166 						perf_event__fprintf(event, stderr);
167 					pr_debug("Couldn't parse sample\n");
168 					goto out_delete_evlist;
169 				}
170 
171 				if (verbose) {
172 					pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
173 					perf_event__fprintf(event, stderr);
174 				}
175 
176 				if (prev_time > sample.time) {
177 					pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n",
178 						 name, prev_time, sample.time);
179 					++errs;
180 				}
181 
182 				prev_time = sample.time;
183 
184 				if (sample.cpu != cpu) {
185 					pr_debug("%s with unexpected cpu, expected %d, got %d\n",
186 						 name, cpu, sample.cpu);
187 					++errs;
188 				}
189 
190 				if ((pid_t)sample.pid != evlist->workload.pid) {
191 					pr_debug("%s with unexpected pid, expected %d, got %d\n",
192 						 name, evlist->workload.pid, sample.pid);
193 					++errs;
194 				}
195 
196 				if ((pid_t)sample.tid != evlist->workload.pid) {
197 					pr_debug("%s with unexpected tid, expected %d, got %d\n",
198 						 name, evlist->workload.pid, sample.tid);
199 					++errs;
200 				}
201 
202 				if ((type == PERF_RECORD_COMM ||
203 				     type == PERF_RECORD_MMAP ||
204 				     type == PERF_RECORD_MMAP2 ||
205 				     type == PERF_RECORD_FORK ||
206 				     type == PERF_RECORD_EXIT) &&
207 				     (pid_t)event->comm.pid != evlist->workload.pid) {
208 					pr_debug("%s with unexpected pid/tid\n", name);
209 					++errs;
210 				}
211 
212 				if ((type == PERF_RECORD_COMM ||
213 				     type == PERF_RECORD_MMAP ||
214 				     type == PERF_RECORD_MMAP2) &&
215 				     event->comm.pid != event->comm.tid) {
216 					pr_debug("%s with different pid/tid!\n", name);
217 					++errs;
218 				}
219 
220 				switch (type) {
221 				case PERF_RECORD_COMM:
222 					if (strcmp(event->comm.comm, cmd)) {
223 						pr_debug("%s with unexpected comm!\n", name);
224 						++errs;
225 					}
226 					break;
227 				case PERF_RECORD_EXIT:
228 					goto found_exit;
229 				case PERF_RECORD_MMAP:
230 					mmap_filename = event->mmap.filename;
231 					goto check_bname;
232 				case PERF_RECORD_MMAP2:
233 					mmap_filename = event->mmap2.filename;
234 				check_bname:
235 					bname = strrchr(mmap_filename, '/');
236 					if (bname != NULL) {
237 						if (!found_cmd_mmap)
238 							found_cmd_mmap = !strcmp(bname + 1, cmd);
239 						if (!found_libc_mmap)
240 							found_libc_mmap = !strncmp(bname + 1, "libc", 4);
241 						if (!found_ld_mmap)
242 							found_ld_mmap = !strncmp(bname + 1, "ld", 2);
243 					} else if (!found_vdso_mmap)
244 						found_vdso_mmap = !strcmp(mmap_filename, "[vdso]");
245 					break;
246 
247 				case PERF_RECORD_SAMPLE:
248 					/* Just ignore samples for now */
249 					break;
250 				default:
251 					pr_debug("Unexpected perf_event->header.type %d!\n",
252 						 type);
253 					++errs;
254 				}
255 
256 				perf_evlist__mmap_consume(evlist, i);
257 			}
258 		}
259 
260 		/*
261 		 * We don't use poll here because at least at 3.1 times the
262 		 * PERF_RECORD_{!SAMPLE} events don't honour
263 		 * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
264 		 */
265 		if (total_events == before && false)
266 			poll(evlist->pollfd, evlist->nr_fds, -1);
267 
268 		sleep(1);
269 		if (++wakeups > 5) {
270 			pr_debug("No PERF_RECORD_EXIT event!\n");
271 			break;
272 		}
273 	}
274 
275 found_exit:
276 	if (nr_events[PERF_RECORD_COMM] > 1) {
277 		pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
278 		++errs;
279 	}
280 
281 	if (nr_events[PERF_RECORD_COMM] == 0) {
282 		pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
283 		++errs;
284 	}
285 
286 	if (!found_cmd_mmap) {
287 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
288 		++errs;
289 	}
290 
291 	if (!found_libc_mmap) {
292 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
293 		++errs;
294 	}
295 
296 	if (!found_ld_mmap) {
297 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
298 		++errs;
299 	}
300 
301 	if (!found_vdso_mmap) {
302 		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
303 		++errs;
304 	}
305 out_delete_evlist:
306 	perf_evlist__delete(evlist);
307 out:
308 	return (err < 0 || errs > 0) ? -1 : 0;
309 }
310