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
8static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t *maskp)
9{
10	int i, cpu = -1, nrcpus = 1024;
11realloc:
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
35int 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();
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	 * 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.target, argv,
97					    false, false);
98	if (err < 0) {
99		pr_debug("Couldn't run the workload!\n");
100		goto out_delete_maps;
101	}
102
103	/*
104	 * Config the evsels, setting attr->comm on the first one, etc.
105	 */
106	evsel = perf_evlist__first(evlist);
107	perf_evsel__set_sample_bit(evsel, CPU);
108	perf_evsel__set_sample_bit(evsel, TID);
109	perf_evsel__set_sample_bit(evsel, TIME);
110	perf_evlist__config(evlist, &opts);
111
112	err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
113	if (err < 0) {
114		pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno));
115		goto out_delete_maps;
116	}
117
118	cpu = err;
119
120	/*
121	 * So that we can check perf_sample.cpu on all the samples.
122	 */
123	if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) {
124		pr_debug("sched_setaffinity: %s\n", strerror(errno));
125		goto out_delete_maps;
126	}
127
128	/*
129	 * Call sys_perf_event_open on all the fds on all the evsels,
130	 * grouping them if asked to.
131	 */
132	err = perf_evlist__open(evlist);
133	if (err < 0) {
134		pr_debug("perf_evlist__open: %s\n", strerror(errno));
135		goto out_delete_maps;
136	}
137
138	/*
139	 * mmap the first fd on a given CPU and ask for events for the other
140	 * fds in the same CPU to be injected in the same mmap ring buffer
141	 * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
142	 */
143	err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
144	if (err < 0) {
145		pr_debug("perf_evlist__mmap: %s\n", strerror(errno));
146		goto out_close_evlist;
147	}
148
149	/*
150	 * Now that all is properly set up, enable the events, they will
151	 * count just on workload.pid, which will start...
152	 */
153	perf_evlist__enable(evlist);
154
155	/*
156	 * Now!
157	 */
158	perf_evlist__start_workload(evlist);
159
160	while (1) {
161		int before = total_events;
162
163		for (i = 0; i < evlist->nr_mmaps; i++) {
164			union perf_event *event;
165
166			while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
167				const u32 type = event->header.type;
168				const char *name = perf_event__name(type);
169
170				++total_events;
171				if (type < PERF_RECORD_MAX)
172					nr_events[type]++;
173
174				err = perf_evlist__parse_sample(evlist, event, &sample);
175				if (err < 0) {
176					if (verbose)
177						perf_event__fprintf(event, stderr);
178					pr_debug("Couldn't parse sample\n");
179					goto out_err;
180				}
181
182				if (verbose) {
183					pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
184					perf_event__fprintf(event, stderr);
185				}
186
187				if (prev_time > sample.time) {
188					pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n",
189						 name, prev_time, sample.time);
190					++errs;
191				}
192
193				prev_time = sample.time;
194
195				if (sample.cpu != cpu) {
196					pr_debug("%s with unexpected cpu, expected %d, got %d\n",
197						 name, cpu, sample.cpu);
198					++errs;
199				}
200
201				if ((pid_t)sample.pid != evlist->workload.pid) {
202					pr_debug("%s with unexpected pid, expected %d, got %d\n",
203						 name, evlist->workload.pid, sample.pid);
204					++errs;
205				}
206
207				if ((pid_t)sample.tid != evlist->workload.pid) {
208					pr_debug("%s with unexpected tid, expected %d, got %d\n",
209						 name, evlist->workload.pid, sample.tid);
210					++errs;
211				}
212
213				if ((type == PERF_RECORD_COMM ||
214				     type == PERF_RECORD_MMAP ||
215				     type == PERF_RECORD_MMAP2 ||
216				     type == PERF_RECORD_FORK ||
217				     type == PERF_RECORD_EXIT) &&
218				     (pid_t)event->comm.pid != evlist->workload.pid) {
219					pr_debug("%s with unexpected pid/tid\n", name);
220					++errs;
221				}
222
223				if ((type == PERF_RECORD_COMM ||
224				     type == PERF_RECORD_MMAP ||
225				     type == PERF_RECORD_MMAP2) &&
226				     event->comm.pid != event->comm.tid) {
227					pr_debug("%s with different pid/tid!\n", name);
228					++errs;
229				}
230
231				switch (type) {
232				case PERF_RECORD_COMM:
233					if (strcmp(event->comm.comm, cmd)) {
234						pr_debug("%s with unexpected comm!\n", name);
235						++errs;
236					}
237					break;
238				case PERF_RECORD_EXIT:
239					goto found_exit;
240				case PERF_RECORD_MMAP:
241					mmap_filename = event->mmap.filename;
242					goto check_bname;
243				case PERF_RECORD_MMAP2:
244					mmap_filename = event->mmap2.filename;
245				check_bname:
246					bname = strrchr(mmap_filename, '/');
247					if (bname != NULL) {
248						if (!found_cmd_mmap)
249							found_cmd_mmap = !strcmp(bname + 1, cmd);
250						if (!found_libc_mmap)
251							found_libc_mmap = !strncmp(bname + 1, "libc", 4);
252						if (!found_ld_mmap)
253							found_ld_mmap = !strncmp(bname + 1, "ld", 2);
254					} else if (!found_vdso_mmap)
255						found_vdso_mmap = !strcmp(mmap_filename, "[vdso]");
256					break;
257
258				case PERF_RECORD_SAMPLE:
259					/* Just ignore samples for now */
260					break;
261				default:
262					pr_debug("Unexpected perf_event->header.type %d!\n",
263						 type);
264					++errs;
265				}
266
267				perf_evlist__mmap_consume(evlist, i);
268			}
269		}
270
271		/*
272		 * We don't use poll here because at least at 3.1 times the
273		 * PERF_RECORD_{!SAMPLE} events don't honour
274		 * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
275		 */
276		if (total_events == before && false)
277			poll(evlist->pollfd, evlist->nr_fds, -1);
278
279		sleep(1);
280		if (++wakeups > 5) {
281			pr_debug("No PERF_RECORD_EXIT event!\n");
282			break;
283		}
284	}
285
286found_exit:
287	if (nr_events[PERF_RECORD_COMM] > 1) {
288		pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
289		++errs;
290	}
291
292	if (nr_events[PERF_RECORD_COMM] == 0) {
293		pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
294		++errs;
295	}
296
297	if (!found_cmd_mmap) {
298		pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
299		++errs;
300	}
301
302	if (!found_libc_mmap) {
303		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
304		++errs;
305	}
306
307	if (!found_ld_mmap) {
308		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
309		++errs;
310	}
311
312	if (!found_vdso_mmap) {
313		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
314		++errs;
315	}
316out_err:
317	perf_evlist__munmap(evlist);
318out_close_evlist:
319	perf_evlist__close(evlist);
320out_delete_maps:
321	perf_evlist__delete_maps(evlist);
322out_delete_evlist:
323	perf_evlist__delete(evlist);
324out:
325	return (err < 0 || errs > 0) ? -1 : 0;
326}
327