]>
Commit | Line | Data |
---|---|---|
1c6a800c ACM |
1 | /* |
2 | * builtin-test.c | |
3 | * | |
4 | * Builtin regression testing command: ever growing number of sanity tests | |
5 | */ | |
6 | #include "builtin.h" | |
7 | ||
8 | #include "util/cache.h" | |
9 | #include "util/debug.h" | |
ebf294bf | 10 | #include "util/debugfs.h" |
de5fa3a8 | 11 | #include "util/evlist.h" |
1c6a800c | 12 | #include "util/parse-options.h" |
de5fa3a8 | 13 | #include "util/parse-events.h" |
1c6a800c | 14 | #include "util/symbol.h" |
fd78260b | 15 | #include "util/thread_map.h" |
13b62567 | 16 | #include "../../include/linux/hw_breakpoint.h" |
1c6a800c | 17 | |
1c6a800c ACM |
18 | static int vmlinux_matches_kallsyms_filter(struct map *map __used, struct symbol *sym) |
19 | { | |
20 | bool *visited = symbol__priv(sym); | |
21 | *visited = true; | |
22 | return 0; | |
23 | } | |
24 | ||
25 | static int test__vmlinux_matches_kallsyms(void) | |
26 | { | |
27 | int err = -1; | |
28 | struct rb_node *nd; | |
29 | struct symbol *sym; | |
30 | struct map *kallsyms_map, *vmlinux_map; | |
31 | struct machine kallsyms, vmlinux; | |
32 | enum map_type type = MAP__FUNCTION; | |
e60770a0 | 33 | long page_size = sysconf(_SC_PAGE_SIZE); |
1c6a800c ACM |
34 | struct ref_reloc_sym ref_reloc_sym = { .name = "_stext", }; |
35 | ||
36 | /* | |
37 | * Step 1: | |
38 | * | |
39 | * Init the machines that will hold kernel, modules obtained from | |
40 | * both vmlinux + .ko files and from /proc/kallsyms split by modules. | |
41 | */ | |
42 | machine__init(&kallsyms, "", HOST_KERNEL_ID); | |
43 | machine__init(&vmlinux, "", HOST_KERNEL_ID); | |
44 | ||
45 | /* | |
46 | * Step 2: | |
47 | * | |
48 | * Create the kernel maps for kallsyms and the DSO where we will then | |
49 | * load /proc/kallsyms. Also create the modules maps from /proc/modules | |
50 | * and find the .ko files that match them in /lib/modules/`uname -r`/. | |
51 | */ | |
52 | if (machine__create_kernel_maps(&kallsyms) < 0) { | |
53 | pr_debug("machine__create_kernel_maps "); | |
54 | return -1; | |
55 | } | |
56 | ||
57 | /* | |
58 | * Step 3: | |
59 | * | |
60 | * Load and split /proc/kallsyms into multiple maps, one per module. | |
61 | */ | |
62 | if (machine__load_kallsyms(&kallsyms, "/proc/kallsyms", type, NULL) <= 0) { | |
63 | pr_debug("dso__load_kallsyms "); | |
64 | goto out; | |
65 | } | |
66 | ||
67 | /* | |
68 | * Step 4: | |
69 | * | |
70 | * kallsyms will be internally on demand sorted by name so that we can | |
71 | * find the reference relocation * symbol, i.e. the symbol we will use | |
72 | * to see if the running kernel was relocated by checking if it has the | |
73 | * same value in the vmlinux file we load. | |
74 | */ | |
75 | kallsyms_map = machine__kernel_map(&kallsyms, type); | |
76 | ||
77 | sym = map__find_symbol_by_name(kallsyms_map, ref_reloc_sym.name, NULL); | |
78 | if (sym == NULL) { | |
79 | pr_debug("dso__find_symbol_by_name "); | |
80 | goto out; | |
81 | } | |
82 | ||
83 | ref_reloc_sym.addr = sym->start; | |
84 | ||
85 | /* | |
86 | * Step 5: | |
87 | * | |
88 | * Now repeat step 2, this time for the vmlinux file we'll auto-locate. | |
89 | */ | |
90 | if (machine__create_kernel_maps(&vmlinux) < 0) { | |
91 | pr_debug("machine__create_kernel_maps "); | |
92 | goto out; | |
93 | } | |
94 | ||
95 | vmlinux_map = machine__kernel_map(&vmlinux, type); | |
96 | map__kmap(vmlinux_map)->ref_reloc_sym = &ref_reloc_sym; | |
97 | ||
98 | /* | |
99 | * Step 6: | |
100 | * | |
101 | * Locate a vmlinux file in the vmlinux path that has a buildid that | |
102 | * matches the one of the running kernel. | |
103 | * | |
104 | * While doing that look if we find the ref reloc symbol, if we find it | |
105 | * we'll have its ref_reloc_symbol.unrelocated_addr and then | |
106 | * maps__reloc_vmlinux will notice and set proper ->[un]map_ip routines | |
107 | * to fixup the symbols. | |
108 | */ | |
109 | if (machine__load_vmlinux_path(&vmlinux, type, | |
110 | vmlinux_matches_kallsyms_filter) <= 0) { | |
111 | pr_debug("machine__load_vmlinux_path "); | |
112 | goto out; | |
113 | } | |
114 | ||
115 | err = 0; | |
116 | /* | |
117 | * Step 7: | |
118 | * | |
119 | * Now look at the symbols in the vmlinux DSO and check if we find all of them | |
120 | * in the kallsyms dso. For the ones that are in both, check its names and | |
121 | * end addresses too. | |
122 | */ | |
123 | for (nd = rb_first(&vmlinux_map->dso->symbols[type]); nd; nd = rb_next(nd)) { | |
d3678758 ACM |
124 | struct symbol *pair, *first_pair; |
125 | bool backwards = true; | |
1c6a800c ACM |
126 | |
127 | sym = rb_entry(nd, struct symbol, rb_node); | |
d3678758 ACM |
128 | |
129 | if (sym->start == sym->end) | |
130 | continue; | |
131 | ||
132 | first_pair = machine__find_kernel_symbol(&kallsyms, type, sym->start, NULL, NULL); | |
133 | pair = first_pair; | |
1c6a800c ACM |
134 | |
135 | if (pair && pair->start == sym->start) { | |
136 | next_pair: | |
137 | if (strcmp(sym->name, pair->name) == 0) { | |
138 | /* | |
139 | * kallsyms don't have the symbol end, so we | |
140 | * set that by using the next symbol start - 1, | |
141 | * in some cases we get this up to a page | |
142 | * wrong, trace_kmalloc when I was developing | |
143 | * this code was one such example, 2106 bytes | |
144 | * off the real size. More than that and we | |
145 | * _really_ have a problem. | |
146 | */ | |
147 | s64 skew = sym->end - pair->end; | |
148 | if (llabs(skew) < page_size) | |
149 | continue; | |
150 | ||
9486aa38 | 151 | pr_debug("%#" PRIx64 ": diff end addr for %s v: %#" PRIx64 " k: %#" PRIx64 "\n", |
1c6a800c ACM |
152 | sym->start, sym->name, sym->end, pair->end); |
153 | } else { | |
d3678758 ACM |
154 | struct rb_node *nnd; |
155 | detour: | |
156 | nnd = backwards ? rb_prev(&pair->rb_node) : | |
157 | rb_next(&pair->rb_node); | |
1c6a800c ACM |
158 | if (nnd) { |
159 | struct symbol *next = rb_entry(nnd, struct symbol, rb_node); | |
160 | ||
161 | if (next->start == sym->start) { | |
162 | pair = next; | |
163 | goto next_pair; | |
164 | } | |
165 | } | |
d3678758 ACM |
166 | |
167 | if (backwards) { | |
168 | backwards = false; | |
169 | pair = first_pair; | |
170 | goto detour; | |
171 | } | |
172 | ||
9486aa38 | 173 | pr_debug("%#" PRIx64 ": diff name v: %s k: %s\n", |
1c6a800c ACM |
174 | sym->start, sym->name, pair->name); |
175 | } | |
176 | } else | |
9486aa38 | 177 | pr_debug("%#" PRIx64 ": %s not on kallsyms\n", sym->start, sym->name); |
1c6a800c ACM |
178 | |
179 | err = -1; | |
180 | } | |
181 | ||
182 | if (!verbose) | |
183 | goto out; | |
184 | ||
185 | pr_info("Maps only in vmlinux:\n"); | |
186 | ||
187 | for (nd = rb_first(&vmlinux.kmaps.maps[type]); nd; nd = rb_next(nd)) { | |
188 | struct map *pos = rb_entry(nd, struct map, rb_node), *pair; | |
189 | /* | |
190 | * If it is the kernel, kallsyms is always "[kernel.kallsyms]", while | |
191 | * the kernel will have the path for the vmlinux file being used, | |
192 | * so use the short name, less descriptive but the same ("[kernel]" in | |
193 | * both cases. | |
194 | */ | |
195 | pair = map_groups__find_by_name(&kallsyms.kmaps, type, | |
196 | (pos->dso->kernel ? | |
197 | pos->dso->short_name : | |
198 | pos->dso->name)); | |
199 | if (pair) | |
200 | pair->priv = 1; | |
201 | else | |
202 | map__fprintf(pos, stderr); | |
203 | } | |
204 | ||
205 | pr_info("Maps in vmlinux with a different name in kallsyms:\n"); | |
206 | ||
207 | for (nd = rb_first(&vmlinux.kmaps.maps[type]); nd; nd = rb_next(nd)) { | |
208 | struct map *pos = rb_entry(nd, struct map, rb_node), *pair; | |
209 | ||
210 | pair = map_groups__find(&kallsyms.kmaps, type, pos->start); | |
211 | if (pair == NULL || pair->priv) | |
212 | continue; | |
213 | ||
214 | if (pair->start == pos->start) { | |
215 | pair->priv = 1; | |
9486aa38 | 216 | pr_info(" %" PRIx64 "-%" PRIx64 " %" PRIx64 " %s in kallsyms as", |
1c6a800c ACM |
217 | pos->start, pos->end, pos->pgoff, pos->dso->name); |
218 | if (pos->pgoff != pair->pgoff || pos->end != pair->end) | |
9486aa38 | 219 | pr_info(": \n*%" PRIx64 "-%" PRIx64 " %" PRIx64 "", |
1c6a800c ACM |
220 | pair->start, pair->end, pair->pgoff); |
221 | pr_info(" %s\n", pair->dso->name); | |
222 | pair->priv = 1; | |
223 | } | |
224 | } | |
225 | ||
226 | pr_info("Maps only in kallsyms:\n"); | |
227 | ||
228 | for (nd = rb_first(&kallsyms.kmaps.maps[type]); | |
229 | nd; nd = rb_next(nd)) { | |
230 | struct map *pos = rb_entry(nd, struct map, rb_node); | |
231 | ||
232 | if (!pos->priv) | |
233 | map__fprintf(pos, stderr); | |
234 | } | |
235 | out: | |
236 | return err; | |
237 | } | |
238 | ||
0252208e | 239 | #include "util/cpumap.h" |
d854861c ACM |
240 | #include "util/evsel.h" |
241 | #include <sys/types.h> | |
242 | ||
de5fa3a8 | 243 | static int trace_event__id(const char *evname) |
d854861c ACM |
244 | { |
245 | char *filename; | |
246 | int err = -1, fd; | |
247 | ||
248 | if (asprintf(&filename, | |
baf040a0 | 249 | "%s/syscalls/%s/id", |
ebf294bf | 250 | tracing_events_path, evname) < 0) |
d854861c ACM |
251 | return -1; |
252 | ||
253 | fd = open(filename, O_RDONLY); | |
254 | if (fd >= 0) { | |
255 | char id[16]; | |
256 | if (read(fd, id, sizeof(id)) > 0) | |
257 | err = atoi(id); | |
258 | close(fd); | |
259 | } | |
260 | ||
261 | free(filename); | |
262 | return err; | |
263 | } | |
264 | ||
265 | static int test__open_syscall_event(void) | |
266 | { | |
267 | int err = -1, fd; | |
268 | struct thread_map *threads; | |
269 | struct perf_evsel *evsel; | |
23a2f3ab | 270 | struct perf_event_attr attr; |
d854861c ACM |
271 | unsigned int nr_open_calls = 111, i; |
272 | int id = trace_event__id("sys_enter_open"); | |
273 | ||
274 | if (id < 0) { | |
454a3bbe | 275 | pr_debug("is debugfs mounted on /sys/kernel/debug?\n"); |
d854861c ACM |
276 | return -1; |
277 | } | |
278 | ||
279 | threads = thread_map__new(-1, getpid()); | |
280 | if (threads == NULL) { | |
454a3bbe | 281 | pr_debug("thread_map__new\n"); |
d854861c ACM |
282 | return -1; |
283 | } | |
284 | ||
23a2f3ab LM |
285 | memset(&attr, 0, sizeof(attr)); |
286 | attr.type = PERF_TYPE_TRACEPOINT; | |
287 | attr.config = id; | |
288 | evsel = perf_evsel__new(&attr, 0); | |
d854861c | 289 | if (evsel == NULL) { |
454a3bbe | 290 | pr_debug("perf_evsel__new\n"); |
d854861c ACM |
291 | goto out_thread_map_delete; |
292 | } | |
293 | ||
727ab04e | 294 | if (perf_evsel__open_per_thread(evsel, threads, false, NULL) < 0) { |
454a3bbe ACM |
295 | pr_debug("failed to open counter: %s, " |
296 | "tweak /proc/sys/kernel/perf_event_paranoid?\n", | |
297 | strerror(errno)); | |
d854861c ACM |
298 | goto out_evsel_delete; |
299 | } | |
300 | ||
301 | for (i = 0; i < nr_open_calls; ++i) { | |
302 | fd = open("/etc/passwd", O_RDONLY); | |
303 | close(fd); | |
304 | } | |
305 | ||
306 | if (perf_evsel__read_on_cpu(evsel, 0, 0) < 0) { | |
5d2cd909 | 307 | pr_debug("perf_evsel__read_on_cpu\n"); |
d854861c ACM |
308 | goto out_close_fd; |
309 | } | |
310 | ||
454a3bbe | 311 | if (evsel->counts->cpu[0].val != nr_open_calls) { |
9486aa38 | 312 | pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls, got %" PRIu64 "\n", |
d854861c | 313 | nr_open_calls, evsel->counts->cpu[0].val); |
454a3bbe ACM |
314 | goto out_close_fd; |
315 | } | |
d854861c ACM |
316 | |
317 | err = 0; | |
318 | out_close_fd: | |
319 | perf_evsel__close_fd(evsel, 1, threads->nr); | |
320 | out_evsel_delete: | |
321 | perf_evsel__delete(evsel); | |
322 | out_thread_map_delete: | |
323 | thread_map__delete(threads); | |
324 | return err; | |
325 | } | |
326 | ||
0252208e ACM |
327 | #include <sched.h> |
328 | ||
329 | static int test__open_syscall_event_on_all_cpus(void) | |
330 | { | |
331 | int err = -1, fd, cpu; | |
332 | struct thread_map *threads; | |
333 | struct cpu_map *cpus; | |
334 | struct perf_evsel *evsel; | |
335 | struct perf_event_attr attr; | |
336 | unsigned int nr_open_calls = 111, i; | |
57b84e53 | 337 | cpu_set_t cpu_set; |
0252208e ACM |
338 | int id = trace_event__id("sys_enter_open"); |
339 | ||
340 | if (id < 0) { | |
341 | pr_debug("is debugfs mounted on /sys/kernel/debug?\n"); | |
342 | return -1; | |
343 | } | |
344 | ||
345 | threads = thread_map__new(-1, getpid()); | |
346 | if (threads == NULL) { | |
347 | pr_debug("thread_map__new\n"); | |
348 | return -1; | |
349 | } | |
350 | ||
351 | cpus = cpu_map__new(NULL); | |
98d77b78 HP |
352 | if (cpus == NULL) { |
353 | pr_debug("cpu_map__new\n"); | |
354 | goto out_thread_map_delete; | |
0252208e ACM |
355 | } |
356 | ||
0252208e | 357 | |
57b84e53 | 358 | CPU_ZERO(&cpu_set); |
0252208e ACM |
359 | |
360 | memset(&attr, 0, sizeof(attr)); | |
361 | attr.type = PERF_TYPE_TRACEPOINT; | |
362 | attr.config = id; | |
363 | evsel = perf_evsel__new(&attr, 0); | |
364 | if (evsel == NULL) { | |
365 | pr_debug("perf_evsel__new\n"); | |
57b84e53 | 366 | goto out_thread_map_delete; |
0252208e ACM |
367 | } |
368 | ||
727ab04e | 369 | if (perf_evsel__open(evsel, cpus, threads, false, NULL) < 0) { |
0252208e ACM |
370 | pr_debug("failed to open counter: %s, " |
371 | "tweak /proc/sys/kernel/perf_event_paranoid?\n", | |
372 | strerror(errno)); | |
373 | goto out_evsel_delete; | |
374 | } | |
375 | ||
376 | for (cpu = 0; cpu < cpus->nr; ++cpu) { | |
377 | unsigned int ncalls = nr_open_calls + cpu; | |
57b84e53 ACM |
378 | /* |
379 | * XXX eventually lift this restriction in a way that | |
380 | * keeps perf building on older glibc installations | |
381 | * without CPU_ALLOC. 1024 cpus in 2010 still seems | |
382 | * a reasonable upper limit tho :-) | |
383 | */ | |
384 | if (cpus->map[cpu] >= CPU_SETSIZE) { | |
385 | pr_debug("Ignoring CPU %d\n", cpus->map[cpu]); | |
386 | continue; | |
387 | } | |
0252208e | 388 | |
57b84e53 ACM |
389 | CPU_SET(cpus->map[cpu], &cpu_set); |
390 | if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) { | |
ffb5e0fb HP |
391 | pr_debug("sched_setaffinity() failed on CPU %d: %s ", |
392 | cpus->map[cpu], | |
393 | strerror(errno)); | |
394 | goto out_close_fd; | |
395 | } | |
0252208e ACM |
396 | for (i = 0; i < ncalls; ++i) { |
397 | fd = open("/etc/passwd", O_RDONLY); | |
398 | close(fd); | |
399 | } | |
57b84e53 | 400 | CPU_CLR(cpus->map[cpu], &cpu_set); |
0252208e ACM |
401 | } |
402 | ||
403 | /* | |
404 | * Here we need to explicitely preallocate the counts, as if | |
405 | * we use the auto allocation it will allocate just for 1 cpu, | |
406 | * as we start by cpu 0. | |
407 | */ | |
408 | if (perf_evsel__alloc_counts(evsel, cpus->nr) < 0) { | |
409 | pr_debug("perf_evsel__alloc_counts(ncpus=%d)\n", cpus->nr); | |
410 | goto out_close_fd; | |
411 | } | |
412 | ||
d2af9687 ACM |
413 | err = 0; |
414 | ||
0252208e ACM |
415 | for (cpu = 0; cpu < cpus->nr; ++cpu) { |
416 | unsigned int expected; | |
417 | ||
57b84e53 ACM |
418 | if (cpus->map[cpu] >= CPU_SETSIZE) |
419 | continue; | |
420 | ||
0252208e | 421 | if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) { |
5d2cd909 | 422 | pr_debug("perf_evsel__read_on_cpu\n"); |
d2af9687 ACM |
423 | err = -1; |
424 | break; | |
0252208e ACM |
425 | } |
426 | ||
427 | expected = nr_open_calls + cpu; | |
428 | if (evsel->counts->cpu[cpu].val != expected) { | |
9486aa38 | 429 | pr_debug("perf_evsel__read_on_cpu: expected to intercept %d calls on cpu %d, got %" PRIu64 "\n", |
ffb5e0fb | 430 | expected, cpus->map[cpu], evsel->counts->cpu[cpu].val); |
d2af9687 | 431 | err = -1; |
0252208e ACM |
432 | } |
433 | } | |
434 | ||
0252208e ACM |
435 | out_close_fd: |
436 | perf_evsel__close_fd(evsel, 1, threads->nr); | |
437 | out_evsel_delete: | |
438 | perf_evsel__delete(evsel); | |
0252208e ACM |
439 | out_thread_map_delete: |
440 | thread_map__delete(threads); | |
441 | return err; | |
442 | } | |
443 | ||
de5fa3a8 ACM |
444 | /* |
445 | * This test will generate random numbers of calls to some getpid syscalls, | |
446 | * then establish an mmap for a group of events that are created to monitor | |
447 | * the syscalls. | |
448 | * | |
449 | * It will receive the events, using mmap, use its PERF_SAMPLE_ID generated | |
450 | * sample.id field to map back to its respective perf_evsel instance. | |
451 | * | |
452 | * Then it checks if the number of syscalls reported as perf events by | |
453 | * the kernel corresponds to the number of syscalls made. | |
454 | */ | |
455 | static int test__basic_mmap(void) | |
456 | { | |
457 | int err = -1; | |
8115d60c | 458 | union perf_event *event; |
de5fa3a8 | 459 | struct thread_map *threads; |
de5fa3a8 ACM |
460 | struct cpu_map *cpus; |
461 | struct perf_evlist *evlist; | |
462 | struct perf_event_attr attr = { | |
463 | .type = PERF_TYPE_TRACEPOINT, | |
464 | .read_format = PERF_FORMAT_ID, | |
465 | .sample_type = PERF_SAMPLE_ID, | |
466 | .watermark = 0, | |
467 | }; | |
468 | cpu_set_t cpu_set; | |
469 | const char *syscall_names[] = { "getsid", "getppid", "getpgrp", | |
470 | "getpgid", }; | |
471 | pid_t (*syscalls[])(void) = { (void *)getsid, getppid, getpgrp, | |
472 | (void*)getpgid }; | |
473 | #define nsyscalls ARRAY_SIZE(syscall_names) | |
474 | int ids[nsyscalls]; | |
475 | unsigned int nr_events[nsyscalls], | |
476 | expected_nr_events[nsyscalls], i, j; | |
477 | struct perf_evsel *evsels[nsyscalls], *evsel; | |
c2a70653 | 478 | int sample_size = __perf_evsel__sample_size(attr.sample_type); |
de5fa3a8 ACM |
479 | |
480 | for (i = 0; i < nsyscalls; ++i) { | |
481 | char name[64]; | |
482 | ||
483 | snprintf(name, sizeof(name), "sys_enter_%s", syscall_names[i]); | |
484 | ids[i] = trace_event__id(name); | |
485 | if (ids[i] < 0) { | |
486 | pr_debug("Is debugfs mounted on /sys/kernel/debug?\n"); | |
487 | return -1; | |
488 | } | |
489 | nr_events[i] = 0; | |
490 | expected_nr_events[i] = random() % 257; | |
491 | } | |
492 | ||
493 | threads = thread_map__new(-1, getpid()); | |
494 | if (threads == NULL) { | |
495 | pr_debug("thread_map__new\n"); | |
496 | return -1; | |
497 | } | |
498 | ||
499 | cpus = cpu_map__new(NULL); | |
54489c18 HP |
500 | if (cpus == NULL) { |
501 | pr_debug("cpu_map__new\n"); | |
de5fa3a8 ACM |
502 | goto out_free_threads; |
503 | } | |
504 | ||
505 | CPU_ZERO(&cpu_set); | |
506 | CPU_SET(cpus->map[0], &cpu_set); | |
507 | sched_setaffinity(0, sizeof(cpu_set), &cpu_set); | |
508 | if (sched_setaffinity(0, sizeof(cpu_set), &cpu_set) < 0) { | |
509 | pr_debug("sched_setaffinity() failed on CPU %d: %s ", | |
510 | cpus->map[0], strerror(errno)); | |
511 | goto out_free_cpus; | |
512 | } | |
513 | ||
7e2ed097 | 514 | evlist = perf_evlist__new(cpus, threads); |
54489c18 | 515 | if (evlist == NULL) { |
de5fa3a8 ACM |
516 | pr_debug("perf_evlist__new\n"); |
517 | goto out_free_cpus; | |
518 | } | |
519 | ||
520 | /* anonymous union fields, can't be initialized above */ | |
521 | attr.wakeup_events = 1; | |
522 | attr.sample_period = 1; | |
523 | ||
de5fa3a8 ACM |
524 | for (i = 0; i < nsyscalls; ++i) { |
525 | attr.config = ids[i]; | |
526 | evsels[i] = perf_evsel__new(&attr, i); | |
527 | if (evsels[i] == NULL) { | |
528 | pr_debug("perf_evsel__new\n"); | |
529 | goto out_free_evlist; | |
530 | } | |
531 | ||
532 | perf_evlist__add(evlist, evsels[i]); | |
533 | ||
727ab04e | 534 | if (perf_evsel__open(evsels[i], cpus, threads, false, NULL) < 0) { |
de5fa3a8 ACM |
535 | pr_debug("failed to open counter: %s, " |
536 | "tweak /proc/sys/kernel/perf_event_paranoid?\n", | |
537 | strerror(errno)); | |
538 | goto out_close_fd; | |
539 | } | |
540 | } | |
541 | ||
7e2ed097 | 542 | if (perf_evlist__mmap(evlist, 128, true) < 0) { |
de5fa3a8 ACM |
543 | pr_debug("failed to mmap events: %d (%s)\n", errno, |
544 | strerror(errno)); | |
545 | goto out_close_fd; | |
546 | } | |
547 | ||
548 | for (i = 0; i < nsyscalls; ++i) | |
549 | for (j = 0; j < expected_nr_events[i]; ++j) { | |
550 | int foo = syscalls[i](); | |
551 | ++foo; | |
552 | } | |
553 | ||
aece948f | 554 | while ((event = perf_evlist__mmap_read(evlist, 0)) != NULL) { |
8d50e5b4 | 555 | struct perf_sample sample; |
de5fa3a8 ACM |
556 | |
557 | if (event->header.type != PERF_RECORD_SAMPLE) { | |
558 | pr_debug("unexpected %s event\n", | |
8115d60c | 559 | perf_event__name(event->header.type)); |
de5fa3a8 ACM |
560 | goto out_munmap; |
561 | } | |
562 | ||
5538beca | 563 | err = perf_event__parse_sample(event, attr.sample_type, sample_size, |
936be503 | 564 | false, &sample, false); |
5538beca FW |
565 | if (err) { |
566 | pr_err("Can't parse sample, err = %d\n", err); | |
567 | goto out_munmap; | |
568 | } | |
569 | ||
de5fa3a8 ACM |
570 | evsel = perf_evlist__id2evsel(evlist, sample.id); |
571 | if (evsel == NULL) { | |
572 | pr_debug("event with id %" PRIu64 | |
573 | " doesn't map to an evsel\n", sample.id); | |
574 | goto out_munmap; | |
575 | } | |
576 | nr_events[evsel->idx]++; | |
577 | } | |
578 | ||
579 | list_for_each_entry(evsel, &evlist->entries, node) { | |
580 | if (nr_events[evsel->idx] != expected_nr_events[evsel->idx]) { | |
581 | pr_debug("expected %d %s events, got %d\n", | |
582 | expected_nr_events[evsel->idx], | |
583 | event_name(evsel), nr_events[evsel->idx]); | |
584 | goto out_munmap; | |
585 | } | |
586 | } | |
587 | ||
588 | err = 0; | |
589 | out_munmap: | |
7e2ed097 | 590 | perf_evlist__munmap(evlist); |
de5fa3a8 ACM |
591 | out_close_fd: |
592 | for (i = 0; i < nsyscalls; ++i) | |
593 | perf_evsel__close_fd(evsels[i], 1, threads->nr); | |
594 | out_free_evlist: | |
595 | perf_evlist__delete(evlist); | |
596 | out_free_cpus: | |
597 | cpu_map__delete(cpus); | |
598 | out_free_threads: | |
599 | thread_map__delete(threads); | |
600 | return err; | |
601 | #undef nsyscalls | |
602 | } | |
603 | ||
13b62567 JO |
604 | #define TEST_ASSERT_VAL(text, cond) \ |
605 | do { \ | |
65c1e045 | 606 | if (!(cond)) { \ |
13b62567 JO |
607 | pr_debug("FAILED %s:%d %s\n", __FILE__, __LINE__, text); \ |
608 | return -1; \ | |
609 | } \ | |
610 | } while (0) | |
611 | ||
612 | static int test__checkevent_tracepoint(struct perf_evlist *evlist) | |
613 | { | |
614 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
615 | struct perf_evsel, node); | |
616 | ||
617 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
618 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_TRACEPOINT == evsel->attr.type); | |
619 | TEST_ASSERT_VAL("wrong sample_type", | |
620 | (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | PERF_SAMPLE_CPU) == | |
621 | evsel->attr.sample_type); | |
622 | TEST_ASSERT_VAL("wrong sample_period", 1 == evsel->attr.sample_period); | |
623 | return 0; | |
624 | } | |
625 | ||
626 | static int test__checkevent_tracepoint_multi(struct perf_evlist *evlist) | |
627 | { | |
628 | struct perf_evsel *evsel; | |
629 | ||
630 | TEST_ASSERT_VAL("wrong number of entries", evlist->nr_entries > 1); | |
631 | ||
632 | list_for_each_entry(evsel, &evlist->entries, node) { | |
633 | TEST_ASSERT_VAL("wrong type", | |
634 | PERF_TYPE_TRACEPOINT == evsel->attr.type); | |
635 | TEST_ASSERT_VAL("wrong sample_type", | |
636 | (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | PERF_SAMPLE_CPU) | |
637 | == evsel->attr.sample_type); | |
638 | TEST_ASSERT_VAL("wrong sample_period", | |
639 | 1 == evsel->attr.sample_period); | |
640 | } | |
641 | return 0; | |
642 | } | |
643 | ||
644 | static int test__checkevent_raw(struct perf_evlist *evlist) | |
645 | { | |
646 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
647 | struct perf_evsel, node); | |
648 | ||
649 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
650 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->attr.type); | |
651 | TEST_ASSERT_VAL("wrong config", 1 == evsel->attr.config); | |
652 | return 0; | |
653 | } | |
654 | ||
655 | static int test__checkevent_numeric(struct perf_evlist *evlist) | |
656 | { | |
657 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
658 | struct perf_evsel, node); | |
659 | ||
660 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
661 | TEST_ASSERT_VAL("wrong type", 1 == evsel->attr.type); | |
662 | TEST_ASSERT_VAL("wrong config", 1 == evsel->attr.config); | |
663 | return 0; | |
664 | } | |
665 | ||
666 | static int test__checkevent_symbolic_name(struct perf_evlist *evlist) | |
667 | { | |
668 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
669 | struct perf_evsel, node); | |
670 | ||
671 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
672 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE == evsel->attr.type); | |
673 | TEST_ASSERT_VAL("wrong config", | |
674 | PERF_COUNT_HW_INSTRUCTIONS == evsel->attr.config); | |
675 | return 0; | |
676 | } | |
677 | ||
678 | static int test__checkevent_symbolic_alias(struct perf_evlist *evlist) | |
679 | { | |
680 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
681 | struct perf_evsel, node); | |
682 | ||
683 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
684 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_SOFTWARE == evsel->attr.type); | |
685 | TEST_ASSERT_VAL("wrong config", | |
686 | PERF_COUNT_SW_PAGE_FAULTS == evsel->attr.config); | |
687 | return 0; | |
688 | } | |
689 | ||
690 | static int test__checkevent_genhw(struct perf_evlist *evlist) | |
691 | { | |
692 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
693 | struct perf_evsel, node); | |
694 | ||
695 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
696 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_HW_CACHE == evsel->attr.type); | |
697 | TEST_ASSERT_VAL("wrong config", (1 << 16) == evsel->attr.config); | |
698 | return 0; | |
699 | } | |
700 | ||
701 | static int test__checkevent_breakpoint(struct perf_evlist *evlist) | |
702 | { | |
703 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
704 | struct perf_evsel, node); | |
705 | ||
706 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
707 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type); | |
708 | TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config); | |
709 | TEST_ASSERT_VAL("wrong bp_type", (HW_BREAKPOINT_R | HW_BREAKPOINT_W) == | |
710 | evsel->attr.bp_type); | |
711 | TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_4 == | |
712 | evsel->attr.bp_len); | |
713 | return 0; | |
714 | } | |
715 | ||
716 | static int test__checkevent_breakpoint_x(struct perf_evlist *evlist) | |
717 | { | |
718 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
719 | struct perf_evsel, node); | |
720 | ||
721 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
722 | TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type); | |
723 | TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config); | |
724 | TEST_ASSERT_VAL("wrong bp_type", | |
725 | HW_BREAKPOINT_X == evsel->attr.bp_type); | |
726 | TEST_ASSERT_VAL("wrong bp_len", sizeof(long) == evsel->attr.bp_len); | |
727 | return 0; | |
728 | } | |
729 | ||
730 | static int test__checkevent_breakpoint_r(struct perf_evlist *evlist) | |
731 | { | |
732 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
733 | struct perf_evsel, node); | |
734 | ||
735 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
736 | TEST_ASSERT_VAL("wrong type", | |
737 | PERF_TYPE_BREAKPOINT == evsel->attr.type); | |
738 | TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config); | |
739 | TEST_ASSERT_VAL("wrong bp_type", | |
740 | HW_BREAKPOINT_R == evsel->attr.bp_type); | |
741 | TEST_ASSERT_VAL("wrong bp_len", | |
742 | HW_BREAKPOINT_LEN_4 == evsel->attr.bp_len); | |
743 | return 0; | |
744 | } | |
745 | ||
746 | static int test__checkevent_breakpoint_w(struct perf_evlist *evlist) | |
747 | { | |
748 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
749 | struct perf_evsel, node); | |
750 | ||
751 | TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries); | |
752 | TEST_ASSERT_VAL("wrong type", | |
753 | PERF_TYPE_BREAKPOINT == evsel->attr.type); | |
754 | TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config); | |
755 | TEST_ASSERT_VAL("wrong bp_type", | |
756 | HW_BREAKPOINT_W == evsel->attr.bp_type); | |
757 | TEST_ASSERT_VAL("wrong bp_len", | |
758 | HW_BREAKPOINT_LEN_4 == evsel->attr.bp_len); | |
759 | return 0; | |
760 | } | |
761 | ||
65c1e045 JO |
762 | static int test__checkevent_tracepoint_modifier(struct perf_evlist *evlist) |
763 | { | |
764 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
765 | struct perf_evsel, node); | |
766 | ||
767 | TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user); | |
768 | TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel); | |
769 | TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv); | |
770 | TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip); | |
771 | ||
772 | return test__checkevent_tracepoint(evlist); | |
773 | } | |
774 | ||
775 | static int | |
776 | test__checkevent_tracepoint_multi_modifier(struct perf_evlist *evlist) | |
777 | { | |
778 | struct perf_evsel *evsel; | |
779 | ||
780 | TEST_ASSERT_VAL("wrong number of entries", evlist->nr_entries > 1); | |
781 | ||
782 | list_for_each_entry(evsel, &evlist->entries, node) { | |
783 | TEST_ASSERT_VAL("wrong exclude_user", | |
784 | !evsel->attr.exclude_user); | |
785 | TEST_ASSERT_VAL("wrong exclude_kernel", | |
786 | evsel->attr.exclude_kernel); | |
787 | TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv); | |
788 | TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip); | |
789 | } | |
790 | ||
791 | return test__checkevent_tracepoint_multi(evlist); | |
792 | } | |
793 | ||
794 | static int test__checkevent_raw_modifier(struct perf_evlist *evlist) | |
795 | { | |
796 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
797 | struct perf_evsel, node); | |
798 | ||
799 | TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user); | |
800 | TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel); | |
801 | TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv); | |
802 | TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip); | |
803 | ||
804 | return test__checkevent_raw(evlist); | |
805 | } | |
806 | ||
807 | static int test__checkevent_numeric_modifier(struct perf_evlist *evlist) | |
808 | { | |
809 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
810 | struct perf_evsel, node); | |
811 | ||
812 | TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user); | |
813 | TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel); | |
814 | TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv); | |
815 | TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip); | |
816 | ||
817 | return test__checkevent_numeric(evlist); | |
818 | } | |
819 | ||
820 | static int test__checkevent_symbolic_name_modifier(struct perf_evlist *evlist) | |
821 | { | |
822 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
823 | struct perf_evsel, node); | |
824 | ||
825 | TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user); | |
826 | TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel); | |
827 | TEST_ASSERT_VAL("wrong exclude_hv", !evsel->attr.exclude_hv); | |
828 | TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip); | |
829 | ||
830 | return test__checkevent_symbolic_name(evlist); | |
831 | } | |
832 | ||
833 | static int test__checkevent_symbolic_alias_modifier(struct perf_evlist *evlist) | |
834 | { | |
835 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
836 | struct perf_evsel, node); | |
837 | ||
838 | TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user); | |
839 | TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel); | |
840 | TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv); | |
841 | TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip); | |
842 | ||
843 | return test__checkevent_symbolic_alias(evlist); | |
844 | } | |
845 | ||
846 | static int test__checkevent_genhw_modifier(struct perf_evlist *evlist) | |
847 | { | |
848 | struct perf_evsel *evsel = list_entry(evlist->entries.next, | |
849 | struct perf_evsel, node); | |
850 | ||
851 | TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user); | |
852 | TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel); | |
853 | TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv); | |
854 | TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip); | |
855 | ||
856 | return test__checkevent_genhw(evlist); | |
857 | } | |
858 | ||
13b62567 JO |
859 | static struct test__event_st { |
860 | const char *name; | |
861 | __u32 type; | |
862 | int (*check)(struct perf_evlist *evlist); | |
863 | } test__events[] = { | |
864 | { | |
865 | .name = "syscalls:sys_enter_open", | |
866 | .check = test__checkevent_tracepoint, | |
867 | }, | |
868 | { | |
869 | .name = "syscalls:*", | |
870 | .check = test__checkevent_tracepoint_multi, | |
871 | }, | |
872 | { | |
873 | .name = "r1", | |
874 | .check = test__checkevent_raw, | |
875 | }, | |
876 | { | |
877 | .name = "1:1", | |
878 | .check = test__checkevent_numeric, | |
879 | }, | |
880 | { | |
881 | .name = "instructions", | |
882 | .check = test__checkevent_symbolic_name, | |
883 | }, | |
884 | { | |
885 | .name = "faults", | |
886 | .check = test__checkevent_symbolic_alias, | |
887 | }, | |
888 | { | |
889 | .name = "L1-dcache-load-miss", | |
890 | .check = test__checkevent_genhw, | |
891 | }, | |
892 | { | |
893 | .name = "mem:0", | |
894 | .check = test__checkevent_breakpoint, | |
895 | }, | |
896 | { | |
897 | .name = "mem:0:x", | |
898 | .check = test__checkevent_breakpoint_x, | |
899 | }, | |
900 | { | |
901 | .name = "mem:0:r", | |
902 | .check = test__checkevent_breakpoint_r, | |
903 | }, | |
904 | { | |
905 | .name = "mem:0:w", | |
906 | .check = test__checkevent_breakpoint_w, | |
907 | }, | |
65c1e045 JO |
908 | { |
909 | .name = "syscalls:sys_enter_open:k", | |
910 | .check = test__checkevent_tracepoint_modifier, | |
911 | }, | |
912 | { | |
913 | .name = "syscalls:*:u", | |
914 | .check = test__checkevent_tracepoint_multi_modifier, | |
915 | }, | |
916 | { | |
917 | .name = "r1:kp", | |
918 | .check = test__checkevent_raw_modifier, | |
919 | }, | |
920 | { | |
921 | .name = "1:1:hp", | |
922 | .check = test__checkevent_numeric_modifier, | |
923 | }, | |
924 | { | |
925 | .name = "instructions:h", | |
926 | .check = test__checkevent_symbolic_name_modifier, | |
927 | }, | |
928 | { | |
929 | .name = "faults:u", | |
930 | .check = test__checkevent_symbolic_alias_modifier, | |
931 | }, | |
932 | { | |
933 | .name = "L1-dcache-load-miss:kp", | |
934 | .check = test__checkevent_genhw_modifier, | |
935 | }, | |
13b62567 JO |
936 | }; |
937 | ||
938 | #define TEST__EVENTS_CNT (sizeof(test__events) / sizeof(struct test__event_st)) | |
939 | ||
940 | static int test__parse_events(void) | |
941 | { | |
942 | struct perf_evlist *evlist; | |
943 | u_int i; | |
944 | int ret = 0; | |
945 | ||
946 | for (i = 0; i < TEST__EVENTS_CNT; i++) { | |
947 | struct test__event_st *e = &test__events[i]; | |
948 | ||
949 | evlist = perf_evlist__new(NULL, NULL); | |
950 | if (evlist == NULL) | |
951 | break; | |
952 | ||
953 | ret = parse_events(evlist, e->name, 0); | |
954 | if (ret) { | |
955 | pr_debug("failed to parse event '%s', err %d\n", | |
956 | e->name, ret); | |
957 | break; | |
958 | } | |
959 | ||
960 | ret = e->check(evlist); | |
961 | if (ret) | |
962 | break; | |
963 | ||
964 | perf_evlist__delete(evlist); | |
965 | } | |
966 | ||
967 | return ret; | |
968 | } | |
3e7c439a ACM |
969 | |
970 | static int sched__get_first_possible_cpu(pid_t pid, cpu_set_t **maskp, | |
971 | size_t *sizep) | |
972 | { | |
973 | cpu_set_t *mask; | |
974 | size_t size; | |
975 | int i, cpu = -1, nrcpus = 1024; | |
976 | realloc: | |
977 | mask = CPU_ALLOC(nrcpus); | |
978 | size = CPU_ALLOC_SIZE(nrcpus); | |
979 | CPU_ZERO_S(size, mask); | |
980 | ||
981 | if (sched_getaffinity(pid, size, mask) == -1) { | |
982 | CPU_FREE(mask); | |
983 | if (errno == EINVAL && nrcpus < (1024 << 8)) { | |
984 | nrcpus = nrcpus << 2; | |
985 | goto realloc; | |
986 | } | |
987 | perror("sched_getaffinity"); | |
988 | return -1; | |
989 | } | |
990 | ||
991 | for (i = 0; i < nrcpus; i++) { | |
992 | if (CPU_ISSET_S(i, size, mask)) { | |
993 | if (cpu == -1) { | |
994 | cpu = i; | |
995 | *maskp = mask; | |
996 | *sizep = size; | |
997 | } else | |
998 | CPU_CLR_S(i, size, mask); | |
999 | } | |
1000 | } | |
1001 | ||
1002 | if (cpu == -1) | |
1003 | CPU_FREE(mask); | |
1004 | ||
1005 | return cpu; | |
1006 | } | |
1007 | ||
1008 | static int test__PERF_RECORD(void) | |
1009 | { | |
1010 | struct perf_record_opts opts = { | |
1011 | .target_pid = -1, | |
1012 | .target_tid = -1, | |
1013 | .no_delay = true, | |
1014 | .freq = 10, | |
1015 | .mmap_pages = 256, | |
1016 | .sample_id_all_avail = true, | |
1017 | }; | |
1018 | cpu_set_t *cpu_mask = NULL; | |
1019 | size_t cpu_mask_size = 0; | |
1020 | struct perf_evlist *evlist = perf_evlist__new(NULL, NULL); | |
1021 | struct perf_evsel *evsel; | |
1022 | struct perf_sample sample; | |
1023 | const char *cmd = "sleep"; | |
1024 | const char *argv[] = { cmd, "1", NULL, }; | |
1025 | char *bname; | |
1026 | u64 sample_type, prev_time = 0; | |
1027 | bool found_cmd_mmap = false, | |
1028 | found_libc_mmap = false, | |
1029 | found_vdso_mmap = false, | |
1030 | found_ld_mmap = false; | |
f71c49e5 | 1031 | int err = -1, errs = 0, i, wakeups = 0, sample_size; |
3e7c439a ACM |
1032 | u32 cpu; |
1033 | int total_events = 0, nr_events[PERF_RECORD_MAX] = { 0, }; | |
1034 | ||
1035 | if (evlist == NULL || argv == NULL) { | |
1036 | pr_debug("Not enough memory to create evlist\n"); | |
1037 | goto out; | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * We need at least one evsel in the evlist, use the default | |
1042 | * one: "cycles". | |
1043 | */ | |
1044 | err = perf_evlist__add_default(evlist); | |
1045 | if (err < 0) { | |
1046 | pr_debug("Not enough memory to create evsel\n"); | |
1047 | goto out_delete_evlist; | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * Create maps of threads and cpus to monitor. In this case | |
1052 | * we start with all threads and cpus (-1, -1) but then in | |
1053 | * perf_evlist__prepare_workload we'll fill in the only thread | |
1054 | * we're monitoring, the one forked there. | |
1055 | */ | |
1056 | err = perf_evlist__create_maps(evlist, opts.target_pid, | |
1057 | opts.target_tid, opts.cpu_list); | |
1058 | if (err < 0) { | |
1059 | pr_debug("Not enough memory to create thread/cpu maps\n"); | |
1060 | goto out_delete_evlist; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * Prepare the workload in argv[] to run, it'll fork it, and then wait | |
1065 | * for perf_evlist__start_workload() to exec it. This is done this way | |
1066 | * so that we have time to open the evlist (calling sys_perf_event_open | |
1067 | * on all the fds) and then mmap them. | |
1068 | */ | |
1069 | err = perf_evlist__prepare_workload(evlist, &opts, argv); | |
1070 | if (err < 0) { | |
1071 | pr_debug("Couldn't run the workload!\n"); | |
1072 | goto out_delete_evlist; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Config the evsels, setting attr->comm on the first one, etc. | |
1077 | */ | |
1078 | evsel = list_entry(evlist->entries.next, struct perf_evsel, node); | |
1079 | evsel->attr.sample_type |= PERF_SAMPLE_CPU; | |
1080 | evsel->attr.sample_type |= PERF_SAMPLE_TID; | |
1081 | evsel->attr.sample_type |= PERF_SAMPLE_TIME; | |
1082 | perf_evlist__config_attrs(evlist, &opts); | |
1083 | ||
1084 | err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask, | |
1085 | &cpu_mask_size); | |
1086 | if (err < 0) { | |
1087 | pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno)); | |
1088 | goto out_delete_evlist; | |
1089 | } | |
1090 | ||
1091 | cpu = err; | |
1092 | ||
1093 | /* | |
1094 | * So that we can check perf_sample.cpu on all the samples. | |
1095 | */ | |
1096 | if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, cpu_mask) < 0) { | |
1097 | pr_debug("sched_setaffinity: %s\n", strerror(errno)); | |
1098 | goto out_free_cpu_mask; | |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * Call sys_perf_event_open on all the fds on all the evsels, | |
1103 | * grouping them if asked to. | |
1104 | */ | |
1105 | err = perf_evlist__open(evlist, opts.group); | |
1106 | if (err < 0) { | |
1107 | pr_debug("perf_evlist__open: %s\n", strerror(errno)); | |
1108 | goto out_delete_evlist; | |
1109 | } | |
1110 | ||
1111 | /* | |
1112 | * mmap the first fd on a given CPU and ask for events for the other | |
1113 | * fds in the same CPU to be injected in the same mmap ring buffer | |
1114 | * (using ioctl(PERF_EVENT_IOC_SET_OUTPUT)). | |
1115 | */ | |
1116 | err = perf_evlist__mmap(evlist, opts.mmap_pages, false); | |
1117 | if (err < 0) { | |
1118 | pr_debug("perf_evlist__mmap: %s\n", strerror(errno)); | |
1119 | goto out_delete_evlist; | |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * We'll need these two to parse the PERF_SAMPLE_* fields in each | |
1124 | * event. | |
1125 | */ | |
1126 | sample_type = perf_evlist__sample_type(evlist); | |
1127 | sample_size = __perf_evsel__sample_size(sample_type); | |
1128 | ||
1129 | /* | |
1130 | * Now that all is properly set up, enable the events, they will | |
1131 | * count just on workload.pid, which will start... | |
1132 | */ | |
1133 | perf_evlist__enable(evlist); | |
1134 | ||
1135 | /* | |
1136 | * Now! | |
1137 | */ | |
1138 | perf_evlist__start_workload(evlist); | |
1139 | ||
3e7c439a ACM |
1140 | while (1) { |
1141 | int before = total_events; | |
1142 | ||
1143 | for (i = 0; i < evlist->nr_mmaps; i++) { | |
1144 | union perf_event *event; | |
1145 | ||
1146 | while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) { | |
1147 | const u32 type = event->header.type; | |
1148 | const char *name = perf_event__name(type); | |
1149 | ||
1150 | ++total_events; | |
1151 | if (type < PERF_RECORD_MAX) | |
1152 | nr_events[type]++; | |
1153 | ||
f71c49e5 ACM |
1154 | err = perf_event__parse_sample(event, sample_type, |
1155 | sample_size, true, | |
1156 | &sample, false); | |
1157 | if (err < 0) { | |
3e7c439a ACM |
1158 | if (verbose) |
1159 | perf_event__fprintf(event, stderr); | |
1160 | pr_debug("Couldn't parse sample\n"); | |
1161 | goto out_err; | |
1162 | } | |
1163 | ||
1164 | if (verbose) { | |
1165 | pr_info("%" PRIu64" %d ", sample.time, sample.cpu); | |
1166 | perf_event__fprintf(event, stderr); | |
1167 | } | |
1168 | ||
1169 | if (prev_time > sample.time) { | |
1170 | pr_debug("%s going backwards in time, prev=%" PRIu64 ", curr=%" PRIu64 "\n", | |
1171 | name, prev_time, sample.time); | |
f71c49e5 | 1172 | ++errs; |
3e7c439a ACM |
1173 | } |
1174 | ||
1175 | prev_time = sample.time; | |
1176 | ||
1177 | if (sample.cpu != cpu) { | |
1178 | pr_debug("%s with unexpected cpu, expected %d, got %d\n", | |
1179 | name, cpu, sample.cpu); | |
f71c49e5 | 1180 | ++errs; |
3e7c439a ACM |
1181 | } |
1182 | ||
1183 | if ((pid_t)sample.pid != evlist->workload.pid) { | |
1184 | pr_debug("%s with unexpected pid, expected %d, got %d\n", | |
1185 | name, evlist->workload.pid, sample.pid); | |
f71c49e5 | 1186 | ++errs; |
3e7c439a ACM |
1187 | } |
1188 | ||
1189 | if ((pid_t)sample.tid != evlist->workload.pid) { | |
1190 | pr_debug("%s with unexpected tid, expected %d, got %d\n", | |
1191 | name, evlist->workload.pid, sample.tid); | |
f71c49e5 | 1192 | ++errs; |
3e7c439a ACM |
1193 | } |
1194 | ||
1195 | if ((type == PERF_RECORD_COMM || | |
1196 | type == PERF_RECORD_MMAP || | |
1197 | type == PERF_RECORD_FORK || | |
1198 | type == PERF_RECORD_EXIT) && | |
1199 | (pid_t)event->comm.pid != evlist->workload.pid) { | |
1200 | pr_debug("%s with unexpected pid/tid\n", name); | |
f71c49e5 | 1201 | ++errs; |
3e7c439a ACM |
1202 | } |
1203 | ||
1204 | if ((type == PERF_RECORD_COMM || | |
1205 | type == PERF_RECORD_MMAP) && | |
1206 | event->comm.pid != event->comm.tid) { | |
1207 | pr_debug("%s with different pid/tid!\n", name); | |
f71c49e5 | 1208 | ++errs; |
3e7c439a ACM |
1209 | } |
1210 | ||
1211 | switch (type) { | |
1212 | case PERF_RECORD_COMM: | |
1213 | if (strcmp(event->comm.comm, cmd)) { | |
1214 | pr_debug("%s with unexpected comm!\n", name); | |
f71c49e5 | 1215 | ++errs; |
3e7c439a ACM |
1216 | } |
1217 | break; | |
1218 | case PERF_RECORD_EXIT: | |
1219 | goto found_exit; | |
1220 | case PERF_RECORD_MMAP: | |
1221 | bname = strrchr(event->mmap.filename, '/'); | |
1222 | if (bname != NULL) { | |
1223 | if (!found_cmd_mmap) | |
1224 | found_cmd_mmap = !strcmp(bname + 1, cmd); | |
1225 | if (!found_libc_mmap) | |
1226 | found_libc_mmap = !strncmp(bname + 1, "libc", 4); | |
1227 | if (!found_ld_mmap) | |
1228 | found_ld_mmap = !strncmp(bname + 1, "ld", 2); | |
1229 | } else if (!found_vdso_mmap) | |
1230 | found_vdso_mmap = !strcmp(event->mmap.filename, "[vdso]"); | |
1231 | break; | |
1232 | ||
1233 | case PERF_RECORD_SAMPLE: | |
1234 | /* Just ignore samples for now */ | |
1235 | break; | |
1236 | default: | |
1237 | pr_debug("Unexpected perf_event->header.type %d!\n", | |
1238 | type); | |
f71c49e5 | 1239 | ++errs; |
3e7c439a ACM |
1240 | } |
1241 | } | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * We don't use poll here because at least at 3.1 times the | |
1246 | * PERF_RECORD_{!SAMPLE} events don't honour | |
1247 | * perf_event_attr.wakeup_events, just PERF_EVENT_SAMPLE does. | |
1248 | */ | |
1249 | if (total_events == before && false) | |
1250 | poll(evlist->pollfd, evlist->nr_fds, -1); | |
1251 | ||
1252 | sleep(1); | |
1253 | if (++wakeups > 5) { | |
1254 | pr_debug("No PERF_RECORD_EXIT event!\n"); | |
f71c49e5 | 1255 | break; |
3e7c439a ACM |
1256 | } |
1257 | } | |
1258 | ||
1259 | found_exit: | |
1260 | if (nr_events[PERF_RECORD_COMM] > 1) { | |
1261 | pr_debug("Excessive number of PERF_RECORD_COMM events!\n"); | |
f71c49e5 | 1262 | ++errs; |
3e7c439a ACM |
1263 | } |
1264 | ||
1265 | if (nr_events[PERF_RECORD_COMM] == 0) { | |
1266 | pr_debug("Missing PERF_RECORD_COMM for %s!\n", cmd); | |
f71c49e5 | 1267 | ++errs; |
3e7c439a ACM |
1268 | } |
1269 | ||
1270 | if (!found_cmd_mmap) { | |
1271 | pr_debug("PERF_RECORD_MMAP for %s missing!\n", cmd); | |
f71c49e5 | 1272 | ++errs; |
3e7c439a ACM |
1273 | } |
1274 | ||
1275 | if (!found_libc_mmap) { | |
1276 | pr_debug("PERF_RECORD_MMAP for %s missing!\n", "libc"); | |
f71c49e5 | 1277 | ++errs; |
3e7c439a ACM |
1278 | } |
1279 | ||
1280 | if (!found_ld_mmap) { | |
1281 | pr_debug("PERF_RECORD_MMAP for %s missing!\n", "ld"); | |
f71c49e5 | 1282 | ++errs; |
3e7c439a ACM |
1283 | } |
1284 | ||
1285 | if (!found_vdso_mmap) { | |
1286 | pr_debug("PERF_RECORD_MMAP for %s missing!\n", "[vdso]"); | |
f71c49e5 | 1287 | ++errs; |
3e7c439a | 1288 | } |
3e7c439a ACM |
1289 | out_err: |
1290 | perf_evlist__munmap(evlist); | |
1291 | out_free_cpu_mask: | |
1292 | CPU_FREE(cpu_mask); | |
1293 | out_delete_evlist: | |
1294 | perf_evlist__delete(evlist); | |
1295 | out: | |
f71c49e5 | 1296 | return (err < 0 || errs > 0) ? -1 : 0; |
3e7c439a ACM |
1297 | } |
1298 | ||
1c6a800c ACM |
1299 | static struct test { |
1300 | const char *desc; | |
1301 | int (*func)(void); | |
1302 | } tests[] = { | |
1303 | { | |
1304 | .desc = "vmlinux symtab matches kallsyms", | |
1305 | .func = test__vmlinux_matches_kallsyms, | |
1306 | }, | |
d854861c ACM |
1307 | { |
1308 | .desc = "detect open syscall event", | |
1309 | .func = test__open_syscall_event, | |
1310 | }, | |
0252208e ACM |
1311 | { |
1312 | .desc = "detect open syscall event on all cpus", | |
1313 | .func = test__open_syscall_event_on_all_cpus, | |
1314 | }, | |
de5fa3a8 ACM |
1315 | { |
1316 | .desc = "read samples using the mmap interface", | |
1317 | .func = test__basic_mmap, | |
1318 | }, | |
13b62567 JO |
1319 | { |
1320 | .desc = "parse events tests", | |
1321 | .func = test__parse_events, | |
1322 | }, | |
3e7c439a ACM |
1323 | { |
1324 | .desc = "Validate PERF_RECORD_* events & perf_sample fields", | |
1325 | .func = test__PERF_RECORD, | |
1326 | }, | |
1c6a800c ACM |
1327 | { |
1328 | .func = NULL, | |
1329 | }, | |
1330 | }; | |
1331 | ||
e60770a0 | 1332 | static bool perf_test__matches(int curr, int argc, const char *argv[]) |
1c6a800c | 1333 | { |
e60770a0 ACM |
1334 | int i; |
1335 | ||
1336 | if (argc == 0) | |
1337 | return true; | |
1338 | ||
1339 | for (i = 0; i < argc; ++i) { | |
1340 | char *end; | |
1341 | long nr = strtoul(argv[i], &end, 10); | |
1342 | ||
1343 | if (*end == '\0') { | |
1344 | if (nr == curr + 1) | |
1345 | return true; | |
1346 | continue; | |
1347 | } | |
1c6a800c | 1348 | |
e60770a0 ACM |
1349 | if (strstr(tests[curr].desc, argv[i])) |
1350 | return true; | |
1351 | } | |
1352 | ||
1353 | return false; | |
1354 | } | |
1355 | ||
1356 | static int __cmd_test(int argc, const char *argv[]) | |
1357 | { | |
1358 | int i = 0; | |
1c6a800c ACM |
1359 | |
1360 | while (tests[i].func) { | |
e60770a0 ACM |
1361 | int curr = i++, err; |
1362 | ||
1363 | if (!perf_test__matches(curr, argc, argv)) | |
1364 | continue; | |
1365 | ||
1366 | pr_info("%2d: %s:", i, tests[curr].desc); | |
1c6a800c | 1367 | pr_debug("\n--- start ---\n"); |
e60770a0 ACM |
1368 | err = tests[curr].func(); |
1369 | pr_debug("---- end ----\n%s:", tests[curr].desc); | |
1c6a800c | 1370 | pr_info(" %s\n", err ? "FAILED!\n" : "Ok"); |
1c6a800c ACM |
1371 | } |
1372 | ||
1373 | return 0; | |
1374 | } | |
1375 | ||
e60770a0 ACM |
1376 | static int perf_test__list(int argc, const char **argv) |
1377 | { | |
1378 | int i = 0; | |
1379 | ||
1380 | while (tests[i].func) { | |
1381 | int curr = i++; | |
1382 | ||
1383 | if (argc > 1 && !strstr(tests[curr].desc, argv[1])) | |
1384 | continue; | |
1385 | ||
1386 | pr_info("%2d: %s\n", i, tests[curr].desc); | |
1387 | } | |
1388 | ||
1389 | return 0; | |
1390 | } | |
1c6a800c | 1391 | |
e60770a0 ACM |
1392 | int cmd_test(int argc, const char **argv, const char *prefix __used) |
1393 | { | |
1394 | const char * const test_usage[] = { | |
1395 | "perf test [<options>] [{list <test-name-fragment>|[<test-name-fragments>|<test-numbers>]}]", | |
1396 | NULL, | |
1397 | }; | |
1398 | const struct option test_options[] = { | |
c30ab8aa | 1399 | OPT_INCR('v', "verbose", &verbose, |
1c6a800c ACM |
1400 | "be more verbose (show symbol address, etc)"), |
1401 | OPT_END() | |
e60770a0 | 1402 | }; |
1c6a800c | 1403 | |
1c6a800c | 1404 | argc = parse_options(argc, argv, test_options, test_usage, 0); |
e60770a0 ACM |
1405 | if (argc >= 1 && !strcmp(argv[0], "list")) |
1406 | return perf_test__list(argc, argv); | |
1c6a800c ACM |
1407 | |
1408 | symbol_conf.priv_size = sizeof(int); | |
1409 | symbol_conf.sort_by_name = true; | |
1410 | symbol_conf.try_vmlinux_path = true; | |
1411 | ||
1412 | if (symbol__init() < 0) | |
1413 | return -1; | |
1414 | ||
1415 | setup_pager(); | |
1416 | ||
e60770a0 | 1417 | return __cmd_test(argc, argv); |
1c6a800c | 1418 | } |