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 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	char sbuf[STRERR_BUFSIZE];
63
64	if (evlist == NULL || argv == NULL) {
65		pr_debug("Not enough memory to create evlist\n");
66		goto out;
67	}
68
69	/*
70	 * Create maps of threads and cpus to monitor. In this case
71	 * we start with all threads and cpus (-1, -1) but then in
72	 * perf_evlist__prepare_workload we'll fill in the only thread
73	 * we're monitoring, the one forked there.
74	 */
75	err = perf_evlist__create_maps(evlist, &opts.target);
76	if (err < 0) {
77		pr_debug("Not enough memory to create thread/cpu maps\n");
78		goto out_delete_evlist;
79	}
80
81	/*
82	 * Prepare the workload in argv[] to run, it'll fork it, and then wait
83	 * for perf_evlist__start_workload() to exec it. This is done this way
84	 * so that we have time to open the evlist (calling sys_perf_event_open
85	 * on all the fds) and then mmap them.
86	 */
87	err = perf_evlist__prepare_workload(evlist, &opts.target, argv, false, NULL);
88	if (err < 0) {
89		pr_debug("Couldn't run the workload!\n");
90		goto out_delete_evlist;
91	}
92
93	/*
94	 * Config the evsels, setting attr->comm on the first one, etc.
95	 */
96	evsel = perf_evlist__first(evlist);
97	perf_evsel__set_sample_bit(evsel, CPU);
98	perf_evsel__set_sample_bit(evsel, TID);
99	perf_evsel__set_sample_bit(evsel, TIME);
100	perf_evlist__config(evlist, &opts);
101
102	err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
103	if (err < 0) {
104		pr_debug("sched__get_first_possible_cpu: %s\n",
105			 strerror_r(errno, sbuf, sizeof(sbuf)));
106		goto out_delete_evlist;
107	}
108
109	cpu = err;
110
111	/*
112	 * So that we can check perf_sample.cpu on all the samples.
113	 */
114	if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) {
115		pr_debug("sched_setaffinity: %s\n",
116			 strerror_r(errno, sbuf, sizeof(sbuf)));
117		goto out_delete_evlist;
118	}
119
120	/*
121	 * Call sys_perf_event_open on all the fds on all the evsels,
122	 * grouping them if asked to.
123	 */
124	err = perf_evlist__open(evlist);
125	if (err < 0) {
126		pr_debug("perf_evlist__open: %s\n",
127			 strerror_r(errno, sbuf, sizeof(sbuf)));
128		goto out_delete_evlist;
129	}
130
131	/*
132	 * mmap the first fd on a given CPU and ask for events for the other
133	 * fds in the same CPU to be injected in the same mmap ring buffer
134	 * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)).
135	 */
136	err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
137	if (err < 0) {
138		pr_debug("perf_evlist__mmap: %s\n",
139			 strerror_r(errno, sbuf, sizeof(sbuf)));
140		goto out_delete_evlist;
141	}
142
143	/*
144	 * Now that all is properly set up, enable the events, they will
145	 * count just on workload.pid, which will start...
146	 */
147	perf_evlist__enable(evlist);
148
149	/*
150	 * Now!
151	 */
152	perf_evlist__start_workload(evlist);
153
154	while (1) {
155		int before = total_events;
156
157		for (i = 0; i < evlist->nr_mmaps; i++) {
158			union perf_event *event;
159
160			while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
161				const u32 type = event->header.type;
162				const char *name = perf_event__name(type);
163
164				++total_events;
165				if (type < PERF_RECORD_MAX)
166					nr_events[type]++;
167
168				err = perf_evlist__parse_sample(evlist, event, &sample);
169				if (err < 0) {
170					if (verbose)
171						perf_event__fprintf(event, stderr);
172					pr_debug("Couldn't parse sample\n");
173					goto out_delete_evlist;
174				}
175
176				if (verbose) {
177					pr_info("%" PRIu64" %d ", sample.time, sample.cpu);
178					perf_event__fprintf(event, stderr);
179				}
180
181				if (prev_time > sample.time) {
182					pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n",
183						 name, prev_time, sample.time);
184					++errs;
185				}
186
187				prev_time = sample.time;
188
189				if (sample.cpu != cpu) {
190					pr_debug("%s with unexpected cpu, expected %d, got %d\n",
191						 name, cpu, sample.cpu);
192					++errs;
193				}
194
195				if ((pid_t)sample.pid != evlist->workload.pid) {
196					pr_debug("%s with unexpected pid, expected %d, got %d\n",
197						 name, evlist->workload.pid, sample.pid);
198					++errs;
199				}
200
201				if ((pid_t)sample.tid != evlist->workload.pid) {
202					pr_debug("%s with unexpected tid, expected %d, got %d\n",
203						 name, evlist->workload.pid, sample.tid);
204					++errs;
205				}
206
207				if ((type == PERF_RECORD_COMM ||
208				     type == PERF_RECORD_MMAP ||
209				     type == PERF_RECORD_MMAP2 ||
210				     type == PERF_RECORD_FORK ||
211				     type == PERF_RECORD_EXIT) &&
212				     (pid_t)event->comm.pid != evlist->workload.pid) {
213					pr_debug("%s with unexpected pid/tid\n", name);
214					++errs;
215				}
216
217				if ((type == PERF_RECORD_COMM ||
218				     type == PERF_RECORD_MMAP ||
219				     type == PERF_RECORD_MMAP2) &&
220				     event->comm.pid != event->comm.tid) {
221					pr_debug("%s with different pid/tid!\n", name);
222					++errs;
223				}
224
225				switch (type) {
226				case PERF_RECORD_COMM:
227					if (strcmp(event->comm.comm, cmd)) {
228						pr_debug("%s with unexpected comm!\n", name);
229						++errs;
230					}
231					break;
232				case PERF_RECORD_EXIT:
233					goto found_exit;
234				case PERF_RECORD_MMAP:
235					mmap_filename = event->mmap.filename;
236					goto check_bname;
237				case PERF_RECORD_MMAP2:
238					mmap_filename = event->mmap2.filename;
239				check_bname:
240					bname = strrchr(mmap_filename, '/');
241					if (bname != NULL) {
242						if (!found_cmd_mmap)
243							found_cmd_mmap = !strcmp(bname + 1, cmd);
244						if (!found_libc_mmap)
245							found_libc_mmap = !strncmp(bname + 1, "libc", 4);
246						if (!found_ld_mmap)
247							found_ld_mmap = !strncmp(bname + 1, "ld", 2);
248					} else if (!found_vdso_mmap)
249						found_vdso_mmap = !strcmp(mmap_filename, "[vdso]");
250					break;
251
252				case PERF_RECORD_SAMPLE:
253					/* Just ignore samples for now */
254					break;
255				default:
256					pr_debug("Unexpected perf_event->header.type %d!\n",
257						 type);
258					++errs;
259				}
260
261				perf_evlist__mmap_consume(evlist, i);
262			}
263		}
264
265		/*
266		 * We don't use poll here because at least at 3.1 times the
267		 * PERF_RECORD_{!SAMPLE} events don't honour
268		 * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does.
269		 */
270		if (total_events == before && false)
271			perf_evlist__poll(evlist, -1);
272
273		sleep(1);
274		if (++wakeups > 5) {
275			pr_debug("No PERF_RECORD_EXIT event!\n");
276			break;
277		}
278	}
279
280found_exit:
281	if (nr_events[PERF_RECORD_COMM] > 1) {
282		pr_debug("Excessive number of PERF_RECORD_COMM events!\n");
283		++errs;
284	}
285
286	if (nr_events[PERF_RECORD_COMM] == 0) {
287		pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd);
288		++errs;
289	}
290
291	if (!found_cmd_mmap) {
292		pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd);
293		++errs;
294	}
295
296	if (!found_libc_mmap) {
297		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc");
298		++errs;
299	}
300
301	if (!found_ld_mmap) {
302		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld");
303		++errs;
304	}
305
306	if (!found_vdso_mmap) {
307		pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]");
308		++errs;
309	}
310out_delete_evlist:
311	perf_evlist__delete(evlist);
312out:
313	return (err < 0 || errs > 0) ? -1 : 0;
314}
315