]>
Commit | Line | Data |
---|---|---|
8035e428 IM |
1 | /* |
2 | * builtin-annotate.c | |
3 | * | |
4 | * Builtin annotate command: Analyze the perf.data input file, | |
5 | * look up and read DSOs and symbol information and display | |
6 | * a histogram of results, along various sorting keys. | |
7 | */ | |
8 | #include "builtin.h" | |
9 | ||
10 | #include "util/util.h" | |
11 | ||
12 | #include "util/color.h" | |
13 | #include "util/list.h" | |
14 | #include "util/cache.h" | |
15 | #include "util/rbtree.h" | |
16 | #include "util/symbol.h" | |
17 | #include "util/string.h" | |
18 | ||
19 | #include "perf.h" | |
20 | ||
21 | #include "util/parse-options.h" | |
22 | #include "util/parse-events.h" | |
23 | ||
24 | #define SHOW_KERNEL 1 | |
25 | #define SHOW_USER 2 | |
26 | #define SHOW_HV 4 | |
27 | ||
28 | static char const *input_name = "perf.data"; | |
29 | static char *vmlinux = NULL; | |
30 | ||
31 | static char default_sort_order[] = "comm,dso"; | |
32 | static char *sort_order = default_sort_order; | |
33 | ||
34 | static int input; | |
35 | static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV; | |
36 | ||
37 | static int dump_trace = 0; | |
38 | #define dprintf(x...) do { if (dump_trace) printf(x); } while (0) | |
39 | ||
40 | static int verbose; | |
41 | static int full_paths; | |
42 | ||
43 | static unsigned long page_size; | |
44 | static unsigned long mmap_window = 32; | |
45 | ||
46 | struct ip_event { | |
47 | struct perf_event_header header; | |
48 | __u64 ip; | |
49 | __u32 pid, tid; | |
50 | }; | |
51 | ||
52 | struct mmap_event { | |
53 | struct perf_event_header header; | |
54 | __u32 pid, tid; | |
55 | __u64 start; | |
56 | __u64 len; | |
57 | __u64 pgoff; | |
58 | char filename[PATH_MAX]; | |
59 | }; | |
60 | ||
61 | struct comm_event { | |
62 | struct perf_event_header header; | |
63 | __u32 pid, tid; | |
64 | char comm[16]; | |
65 | }; | |
66 | ||
67 | struct fork_event { | |
68 | struct perf_event_header header; | |
69 | __u32 pid, ppid; | |
70 | }; | |
71 | ||
72 | struct period_event { | |
73 | struct perf_event_header header; | |
74 | __u64 time; | |
75 | __u64 id; | |
76 | __u64 sample_period; | |
77 | }; | |
78 | ||
79 | typedef union event_union { | |
80 | struct perf_event_header header; | |
81 | struct ip_event ip; | |
82 | struct mmap_event mmap; | |
83 | struct comm_event comm; | |
84 | struct fork_event fork; | |
85 | struct period_event period; | |
86 | } event_t; | |
87 | ||
88 | static LIST_HEAD(dsos); | |
89 | static struct dso *kernel_dso; | |
90 | static struct dso *vdso; | |
91 | ||
92 | static void dsos__add(struct dso *dso) | |
93 | { | |
94 | list_add_tail(&dso->node, &dsos); | |
95 | } | |
96 | ||
97 | static struct dso *dsos__find(const char *name) | |
98 | { | |
99 | struct dso *pos; | |
100 | ||
101 | list_for_each_entry(pos, &dsos, node) | |
102 | if (strcmp(pos->name, name) == 0) | |
103 | return pos; | |
104 | return NULL; | |
105 | } | |
106 | ||
107 | static struct dso *dsos__findnew(const char *name) | |
108 | { | |
109 | struct dso *dso = dsos__find(name); | |
110 | int nr; | |
111 | ||
112 | if (dso) | |
113 | return dso; | |
114 | ||
115 | dso = dso__new(name, 0); | |
116 | if (!dso) | |
117 | goto out_delete_dso; | |
118 | ||
119 | nr = dso__load(dso, NULL, verbose); | |
120 | if (nr < 0) { | |
121 | if (verbose) | |
122 | fprintf(stderr, "Failed to open: %s\n", name); | |
123 | goto out_delete_dso; | |
124 | } | |
125 | if (!nr && verbose) { | |
126 | fprintf(stderr, | |
127 | "No symbols found in: %s, maybe install a debug package?\n", | |
128 | name); | |
129 | } | |
130 | ||
131 | dsos__add(dso); | |
132 | ||
133 | return dso; | |
134 | ||
135 | out_delete_dso: | |
136 | dso__delete(dso); | |
137 | return NULL; | |
138 | } | |
139 | ||
140 | static void dsos__fprintf(FILE *fp) | |
141 | { | |
142 | struct dso *pos; | |
143 | ||
144 | list_for_each_entry(pos, &dsos, node) | |
145 | dso__fprintf(pos, fp); | |
146 | } | |
147 | ||
148 | static struct symbol *vdso__find_symbol(struct dso *dso, uint64_t ip) | |
149 | { | |
150 | return dso__find_symbol(kernel_dso, ip); | |
151 | } | |
152 | ||
153 | static int load_kernel(void) | |
154 | { | |
155 | int err; | |
156 | ||
157 | kernel_dso = dso__new("[kernel]", 0); | |
158 | if (!kernel_dso) | |
159 | return -1; | |
160 | ||
161 | err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose); | |
162 | if (err) { | |
163 | dso__delete(kernel_dso); | |
164 | kernel_dso = NULL; | |
165 | } else | |
166 | dsos__add(kernel_dso); | |
167 | ||
168 | vdso = dso__new("[vdso]", 0); | |
169 | if (!vdso) | |
170 | return -1; | |
171 | ||
172 | vdso->find_symbol = vdso__find_symbol; | |
173 | ||
174 | dsos__add(vdso); | |
175 | ||
176 | return err; | |
177 | } | |
178 | ||
179 | static char __cwd[PATH_MAX]; | |
180 | static char *cwd = __cwd; | |
181 | static int cwdlen; | |
182 | ||
183 | static int strcommon(const char *pathname) | |
184 | { | |
185 | int n = 0; | |
186 | ||
187 | while (pathname[n] == cwd[n] && n < cwdlen) | |
188 | ++n; | |
189 | ||
190 | return n; | |
191 | } | |
192 | ||
193 | struct map { | |
194 | struct list_head node; | |
195 | uint64_t start; | |
196 | uint64_t end; | |
197 | uint64_t pgoff; | |
198 | uint64_t (*map_ip)(struct map *, uint64_t); | |
199 | struct dso *dso; | |
200 | }; | |
201 | ||
202 | static uint64_t map__map_ip(struct map *map, uint64_t ip) | |
203 | { | |
204 | return ip - map->start + map->pgoff; | |
205 | } | |
206 | ||
207 | static uint64_t vdso__map_ip(struct map *map, uint64_t ip) | |
208 | { | |
209 | return ip; | |
210 | } | |
211 | ||
212 | static struct map *map__new(struct mmap_event *event) | |
213 | { | |
214 | struct map *self = malloc(sizeof(*self)); | |
215 | ||
216 | if (self != NULL) { | |
217 | const char *filename = event->filename; | |
218 | char newfilename[PATH_MAX]; | |
219 | ||
220 | if (cwd) { | |
221 | int n = strcommon(filename); | |
222 | ||
223 | if (n == cwdlen) { | |
224 | snprintf(newfilename, sizeof(newfilename), | |
225 | ".%s", filename + n); | |
226 | filename = newfilename; | |
227 | } | |
228 | } | |
229 | ||
230 | self->start = event->start; | |
231 | self->end = event->start + event->len; | |
232 | self->pgoff = event->pgoff; | |
233 | ||
234 | self->dso = dsos__findnew(filename); | |
235 | if (self->dso == NULL) | |
236 | goto out_delete; | |
237 | ||
238 | if (self->dso == vdso) | |
239 | self->map_ip = vdso__map_ip; | |
240 | else | |
241 | self->map_ip = map__map_ip; | |
242 | } | |
243 | return self; | |
244 | out_delete: | |
245 | free(self); | |
246 | return NULL; | |
247 | } | |
248 | ||
249 | static struct map *map__clone(struct map *self) | |
250 | { | |
251 | struct map *map = malloc(sizeof(*self)); | |
252 | ||
253 | if (!map) | |
254 | return NULL; | |
255 | ||
256 | memcpy(map, self, sizeof(*self)); | |
257 | ||
258 | return map; | |
259 | } | |
260 | ||
261 | static int map__overlap(struct map *l, struct map *r) | |
262 | { | |
263 | if (l->start > r->start) { | |
264 | struct map *t = l; | |
265 | l = r; | |
266 | r = t; | |
267 | } | |
268 | ||
269 | if (l->end > r->start) | |
270 | return 1; | |
271 | ||
272 | return 0; | |
273 | } | |
274 | ||
275 | static size_t map__fprintf(struct map *self, FILE *fp) | |
276 | { | |
277 | return fprintf(fp, " %"PRIx64"-%"PRIx64" %"PRIx64" %s\n", | |
278 | self->start, self->end, self->pgoff, self->dso->name); | |
279 | } | |
280 | ||
281 | ||
282 | struct thread { | |
283 | struct rb_node rb_node; | |
284 | struct list_head maps; | |
285 | pid_t pid; | |
286 | char *comm; | |
287 | }; | |
288 | ||
289 | static struct thread *thread__new(pid_t pid) | |
290 | { | |
291 | struct thread *self = malloc(sizeof(*self)); | |
292 | ||
293 | if (self != NULL) { | |
294 | self->pid = pid; | |
295 | self->comm = malloc(32); | |
296 | if (self->comm) | |
297 | snprintf(self->comm, 32, ":%d", self->pid); | |
298 | INIT_LIST_HEAD(&self->maps); | |
299 | } | |
300 | ||
301 | return self; | |
302 | } | |
303 | ||
304 | static int thread__set_comm(struct thread *self, const char *comm) | |
305 | { | |
306 | if (self->comm) | |
307 | free(self->comm); | |
308 | self->comm = strdup(comm); | |
309 | return self->comm ? 0 : -ENOMEM; | |
310 | } | |
311 | ||
312 | static size_t thread__fprintf(struct thread *self, FILE *fp) | |
313 | { | |
314 | struct map *pos; | |
315 | size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm); | |
316 | ||
317 | list_for_each_entry(pos, &self->maps, node) | |
318 | ret += map__fprintf(pos, fp); | |
319 | ||
320 | return ret; | |
321 | } | |
322 | ||
323 | ||
324 | static struct rb_root threads; | |
325 | static struct thread *last_match; | |
326 | ||
327 | static struct thread *threads__findnew(pid_t pid) | |
328 | { | |
329 | struct rb_node **p = &threads.rb_node; | |
330 | struct rb_node *parent = NULL; | |
331 | struct thread *th; | |
332 | ||
333 | /* | |
334 | * Font-end cache - PID lookups come in blocks, | |
335 | * so most of the time we dont have to look up | |
336 | * the full rbtree: | |
337 | */ | |
338 | if (last_match && last_match->pid == pid) | |
339 | return last_match; | |
340 | ||
341 | while (*p != NULL) { | |
342 | parent = *p; | |
343 | th = rb_entry(parent, struct thread, rb_node); | |
344 | ||
345 | if (th->pid == pid) { | |
346 | last_match = th; | |
347 | return th; | |
348 | } | |
349 | ||
350 | if (pid < th->pid) | |
351 | p = &(*p)->rb_left; | |
352 | else | |
353 | p = &(*p)->rb_right; | |
354 | } | |
355 | ||
356 | th = thread__new(pid); | |
357 | if (th != NULL) { | |
358 | rb_link_node(&th->rb_node, parent, p); | |
359 | rb_insert_color(&th->rb_node, &threads); | |
360 | last_match = th; | |
361 | } | |
362 | ||
363 | return th; | |
364 | } | |
365 | ||
366 | static void thread__insert_map(struct thread *self, struct map *map) | |
367 | { | |
368 | struct map *pos, *tmp; | |
369 | ||
370 | list_for_each_entry_safe(pos, tmp, &self->maps, node) { | |
371 | if (map__overlap(pos, map)) { | |
372 | list_del_init(&pos->node); | |
373 | /* XXX leaks dsos */ | |
374 | free(pos); | |
375 | } | |
376 | } | |
377 | ||
378 | list_add_tail(&map->node, &self->maps); | |
379 | } | |
380 | ||
381 | static int thread__fork(struct thread *self, struct thread *parent) | |
382 | { | |
383 | struct map *map; | |
384 | ||
385 | if (self->comm) | |
386 | free(self->comm); | |
387 | self->comm = strdup(parent->comm); | |
388 | if (!self->comm) | |
389 | return -ENOMEM; | |
390 | ||
391 | list_for_each_entry(map, &parent->maps, node) { | |
392 | struct map *new = map__clone(map); | |
393 | if (!new) | |
394 | return -ENOMEM; | |
395 | thread__insert_map(self, new); | |
396 | } | |
397 | ||
398 | return 0; | |
399 | } | |
400 | ||
401 | static struct map *thread__find_map(struct thread *self, uint64_t ip) | |
402 | { | |
403 | struct map *pos; | |
404 | ||
405 | if (self == NULL) | |
406 | return NULL; | |
407 | ||
408 | list_for_each_entry(pos, &self->maps, node) | |
409 | if (ip >= pos->start && ip <= pos->end) | |
410 | return pos; | |
411 | ||
412 | return NULL; | |
413 | } | |
414 | ||
415 | static size_t threads__fprintf(FILE *fp) | |
416 | { | |
417 | size_t ret = 0; | |
418 | struct rb_node *nd; | |
419 | ||
420 | for (nd = rb_first(&threads); nd; nd = rb_next(nd)) { | |
421 | struct thread *pos = rb_entry(nd, struct thread, rb_node); | |
422 | ||
423 | ret += thread__fprintf(pos, fp); | |
424 | } | |
425 | ||
426 | return ret; | |
427 | } | |
428 | ||
429 | /* | |
430 | * histogram, sorted on item, collects counts | |
431 | */ | |
432 | ||
433 | static struct rb_root hist; | |
434 | ||
435 | struct hist_entry { | |
436 | struct rb_node rb_node; | |
437 | ||
438 | struct thread *thread; | |
439 | struct map *map; | |
440 | struct dso *dso; | |
441 | struct symbol *sym; | |
442 | uint64_t ip; | |
443 | char level; | |
444 | ||
445 | uint32_t count; | |
446 | }; | |
447 | ||
448 | /* | |
449 | * configurable sorting bits | |
450 | */ | |
451 | ||
452 | struct sort_entry { | |
453 | struct list_head list; | |
454 | ||
455 | char *header; | |
456 | ||
457 | int64_t (*cmp)(struct hist_entry *, struct hist_entry *); | |
458 | int64_t (*collapse)(struct hist_entry *, struct hist_entry *); | |
459 | size_t (*print)(FILE *fp, struct hist_entry *); | |
460 | }; | |
461 | ||
462 | /* --sort pid */ | |
463 | ||
464 | static int64_t | |
465 | sort__thread_cmp(struct hist_entry *left, struct hist_entry *right) | |
466 | { | |
467 | return right->thread->pid - left->thread->pid; | |
468 | } | |
469 | ||
470 | static size_t | |
471 | sort__thread_print(FILE *fp, struct hist_entry *self) | |
472 | { | |
473 | return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid); | |
474 | } | |
475 | ||
476 | static struct sort_entry sort_thread = { | |
477 | .header = " Command: Pid", | |
478 | .cmp = sort__thread_cmp, | |
479 | .print = sort__thread_print, | |
480 | }; | |
481 | ||
482 | /* --sort comm */ | |
483 | ||
484 | static int64_t | |
485 | sort__comm_cmp(struct hist_entry *left, struct hist_entry *right) | |
486 | { | |
487 | return right->thread->pid - left->thread->pid; | |
488 | } | |
489 | ||
490 | static int64_t | |
491 | sort__comm_collapse(struct hist_entry *left, struct hist_entry *right) | |
492 | { | |
493 | char *comm_l = left->thread->comm; | |
494 | char *comm_r = right->thread->comm; | |
495 | ||
496 | if (!comm_l || !comm_r) { | |
497 | if (!comm_l && !comm_r) | |
498 | return 0; | |
499 | else if (!comm_l) | |
500 | return -1; | |
501 | else | |
502 | return 1; | |
503 | } | |
504 | ||
505 | return strcmp(comm_l, comm_r); | |
506 | } | |
507 | ||
508 | static size_t | |
509 | sort__comm_print(FILE *fp, struct hist_entry *self) | |
510 | { | |
511 | return fprintf(fp, "%16s", self->thread->comm); | |
512 | } | |
513 | ||
514 | static struct sort_entry sort_comm = { | |
515 | .header = " Command", | |
516 | .cmp = sort__comm_cmp, | |
517 | .collapse = sort__comm_collapse, | |
518 | .print = sort__comm_print, | |
519 | }; | |
520 | ||
521 | /* --sort dso */ | |
522 | ||
523 | static int64_t | |
524 | sort__dso_cmp(struct hist_entry *left, struct hist_entry *right) | |
525 | { | |
526 | struct dso *dso_l = left->dso; | |
527 | struct dso *dso_r = right->dso; | |
528 | ||
529 | if (!dso_l || !dso_r) { | |
530 | if (!dso_l && !dso_r) | |
531 | return 0; | |
532 | else if (!dso_l) | |
533 | return -1; | |
534 | else | |
535 | return 1; | |
536 | } | |
537 | ||
538 | return strcmp(dso_l->name, dso_r->name); | |
539 | } | |
540 | ||
541 | static size_t | |
542 | sort__dso_print(FILE *fp, struct hist_entry *self) | |
543 | { | |
544 | if (self->dso) | |
545 | return fprintf(fp, "%-25s", self->dso->name); | |
546 | ||
547 | return fprintf(fp, "%016llx ", (__u64)self->ip); | |
548 | } | |
549 | ||
550 | static struct sort_entry sort_dso = { | |
551 | .header = "Shared Object ", | |
552 | .cmp = sort__dso_cmp, | |
553 | .print = sort__dso_print, | |
554 | }; | |
555 | ||
556 | /* --sort symbol */ | |
557 | ||
558 | static int64_t | |
559 | sort__sym_cmp(struct hist_entry *left, struct hist_entry *right) | |
560 | { | |
561 | uint64_t ip_l, ip_r; | |
562 | ||
563 | if (left->sym == right->sym) | |
564 | return 0; | |
565 | ||
566 | ip_l = left->sym ? left->sym->start : left->ip; | |
567 | ip_r = right->sym ? right->sym->start : right->ip; | |
568 | ||
569 | return (int64_t)(ip_r - ip_l); | |
570 | } | |
571 | ||
572 | static size_t | |
573 | sort__sym_print(FILE *fp, struct hist_entry *self) | |
574 | { | |
575 | size_t ret = 0; | |
576 | ||
577 | if (verbose) | |
578 | ret += fprintf(fp, "%#018llx ", (__u64)self->ip); | |
579 | ||
580 | if (self->sym) { | |
581 | ret += fprintf(fp, "[%c] %s", | |
582 | self->dso == kernel_dso ? 'k' : '.', self->sym->name); | |
583 | } else { | |
584 | ret += fprintf(fp, "%#016llx", (__u64)self->ip); | |
585 | } | |
586 | ||
587 | return ret; | |
588 | } | |
589 | ||
590 | static struct sort_entry sort_sym = { | |
591 | .header = "Symbol", | |
592 | .cmp = sort__sym_cmp, | |
593 | .print = sort__sym_print, | |
594 | }; | |
595 | ||
596 | static int sort__need_collapse = 0; | |
597 | ||
598 | struct sort_dimension { | |
599 | char *name; | |
600 | struct sort_entry *entry; | |
601 | int taken; | |
602 | }; | |
603 | ||
604 | static struct sort_dimension sort_dimensions[] = { | |
605 | { .name = "pid", .entry = &sort_thread, }, | |
606 | { .name = "comm", .entry = &sort_comm, }, | |
607 | { .name = "dso", .entry = &sort_dso, }, | |
608 | { .name = "symbol", .entry = &sort_sym, }, | |
609 | }; | |
610 | ||
611 | static LIST_HEAD(hist_entry__sort_list); | |
612 | ||
613 | static int sort_dimension__add(char *tok) | |
614 | { | |
615 | int i; | |
616 | ||
617 | for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) { | |
618 | struct sort_dimension *sd = &sort_dimensions[i]; | |
619 | ||
620 | if (sd->taken) | |
621 | continue; | |
622 | ||
623 | if (strncasecmp(tok, sd->name, strlen(tok))) | |
624 | continue; | |
625 | ||
626 | if (sd->entry->collapse) | |
627 | sort__need_collapse = 1; | |
628 | ||
629 | list_add_tail(&sd->entry->list, &hist_entry__sort_list); | |
630 | sd->taken = 1; | |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
635 | return -ESRCH; | |
636 | } | |
637 | ||
638 | static int64_t | |
639 | hist_entry__cmp(struct hist_entry *left, struct hist_entry *right) | |
640 | { | |
641 | struct sort_entry *se; | |
642 | int64_t cmp = 0; | |
643 | ||
644 | list_for_each_entry(se, &hist_entry__sort_list, list) { | |
645 | cmp = se->cmp(left, right); | |
646 | if (cmp) | |
647 | break; | |
648 | } | |
649 | ||
650 | return cmp; | |
651 | } | |
652 | ||
653 | static int64_t | |
654 | hist_entry__collapse(struct hist_entry *left, struct hist_entry *right) | |
655 | { | |
656 | struct sort_entry *se; | |
657 | int64_t cmp = 0; | |
658 | ||
659 | list_for_each_entry(se, &hist_entry__sort_list, list) { | |
660 | int64_t (*f)(struct hist_entry *, struct hist_entry *); | |
661 | ||
662 | f = se->collapse ?: se->cmp; | |
663 | ||
664 | cmp = f(left, right); | |
665 | if (cmp) | |
666 | break; | |
667 | } | |
668 | ||
669 | return cmp; | |
670 | } | |
671 | ||
672 | static size_t | |
673 | hist_entry__fprintf(FILE *fp, struct hist_entry *self, uint64_t total_samples) | |
674 | { | |
675 | struct sort_entry *se; | |
676 | size_t ret; | |
677 | ||
678 | if (total_samples) { | |
679 | double percent = self->count * 100.0 / total_samples; | |
680 | char *color = PERF_COLOR_NORMAL; | |
681 | ||
682 | /* | |
683 | * We color high-overhead entries in red, low-overhead | |
684 | * entries in green - and keep the middle ground normal: | |
685 | */ | |
686 | if (percent >= 5.0) | |
687 | color = PERF_COLOR_RED; | |
688 | if (percent < 0.5) | |
689 | color = PERF_COLOR_GREEN; | |
690 | ||
691 | ret = color_fprintf(fp, color, " %6.2f%%", | |
692 | (self->count * 100.0) / total_samples); | |
693 | } else | |
694 | ret = fprintf(fp, "%12d ", self->count); | |
695 | ||
696 | list_for_each_entry(se, &hist_entry__sort_list, list) { | |
697 | fprintf(fp, " "); | |
698 | ret += se->print(fp, self); | |
699 | } | |
700 | ||
701 | ret += fprintf(fp, "\n"); | |
702 | ||
703 | return ret; | |
704 | } | |
705 | ||
706 | /* | |
707 | * collect histogram counts | |
708 | */ | |
709 | ||
710 | static int | |
711 | hist_entry__add(struct thread *thread, struct map *map, struct dso *dso, | |
712 | struct symbol *sym, uint64_t ip, char level) | |
713 | { | |
714 | struct rb_node **p = &hist.rb_node; | |
715 | struct rb_node *parent = NULL; | |
716 | struct hist_entry *he; | |
717 | struct hist_entry entry = { | |
718 | .thread = thread, | |
719 | .map = map, | |
720 | .dso = dso, | |
721 | .sym = sym, | |
722 | .ip = ip, | |
723 | .level = level, | |
724 | .count = 1, | |
725 | }; | |
726 | int cmp; | |
727 | ||
728 | while (*p != NULL) { | |
729 | parent = *p; | |
730 | he = rb_entry(parent, struct hist_entry, rb_node); | |
731 | ||
732 | cmp = hist_entry__cmp(&entry, he); | |
733 | ||
734 | if (!cmp) { | |
735 | he->count++; | |
736 | return 0; | |
737 | } | |
738 | ||
739 | if (cmp < 0) | |
740 | p = &(*p)->rb_left; | |
741 | else | |
742 | p = &(*p)->rb_right; | |
743 | } | |
744 | ||
745 | he = malloc(sizeof(*he)); | |
746 | if (!he) | |
747 | return -ENOMEM; | |
748 | *he = entry; | |
749 | rb_link_node(&he->rb_node, parent, p); | |
750 | rb_insert_color(&he->rb_node, &hist); | |
751 | ||
752 | return 0; | |
753 | } | |
754 | ||
755 | static void hist_entry__free(struct hist_entry *he) | |
756 | { | |
757 | free(he); | |
758 | } | |
759 | ||
760 | /* | |
761 | * collapse the histogram | |
762 | */ | |
763 | ||
764 | static struct rb_root collapse_hists; | |
765 | ||
766 | static void collapse__insert_entry(struct hist_entry *he) | |
767 | { | |
768 | struct rb_node **p = &collapse_hists.rb_node; | |
769 | struct rb_node *parent = NULL; | |
770 | struct hist_entry *iter; | |
771 | int64_t cmp; | |
772 | ||
773 | while (*p != NULL) { | |
774 | parent = *p; | |
775 | iter = rb_entry(parent, struct hist_entry, rb_node); | |
776 | ||
777 | cmp = hist_entry__collapse(iter, he); | |
778 | ||
779 | if (!cmp) { | |
780 | iter->count += he->count; | |
781 | hist_entry__free(he); | |
782 | return; | |
783 | } | |
784 | ||
785 | if (cmp < 0) | |
786 | p = &(*p)->rb_left; | |
787 | else | |
788 | p = &(*p)->rb_right; | |
789 | } | |
790 | ||
791 | rb_link_node(&he->rb_node, parent, p); | |
792 | rb_insert_color(&he->rb_node, &collapse_hists); | |
793 | } | |
794 | ||
795 | static void collapse__resort(void) | |
796 | { | |
797 | struct rb_node *next; | |
798 | struct hist_entry *n; | |
799 | ||
800 | if (!sort__need_collapse) | |
801 | return; | |
802 | ||
803 | next = rb_first(&hist); | |
804 | while (next) { | |
805 | n = rb_entry(next, struct hist_entry, rb_node); | |
806 | next = rb_next(&n->rb_node); | |
807 | ||
808 | rb_erase(&n->rb_node, &hist); | |
809 | collapse__insert_entry(n); | |
810 | } | |
811 | } | |
812 | ||
813 | /* | |
814 | * reverse the map, sort on count. | |
815 | */ | |
816 | ||
817 | static struct rb_root output_hists; | |
818 | ||
819 | static void output__insert_entry(struct hist_entry *he) | |
820 | { | |
821 | struct rb_node **p = &output_hists.rb_node; | |
822 | struct rb_node *parent = NULL; | |
823 | struct hist_entry *iter; | |
824 | ||
825 | while (*p != NULL) { | |
826 | parent = *p; | |
827 | iter = rb_entry(parent, struct hist_entry, rb_node); | |
828 | ||
829 | if (he->count > iter->count) | |
830 | p = &(*p)->rb_left; | |
831 | else | |
832 | p = &(*p)->rb_right; | |
833 | } | |
834 | ||
835 | rb_link_node(&he->rb_node, parent, p); | |
836 | rb_insert_color(&he->rb_node, &output_hists); | |
837 | } | |
838 | ||
839 | static void output__resort(void) | |
840 | { | |
841 | struct rb_node *next; | |
842 | struct hist_entry *n; | |
843 | struct rb_root *tree = &hist; | |
844 | ||
845 | if (sort__need_collapse) | |
846 | tree = &collapse_hists; | |
847 | ||
848 | next = rb_first(tree); | |
849 | ||
850 | while (next) { | |
851 | n = rb_entry(next, struct hist_entry, rb_node); | |
852 | next = rb_next(&n->rb_node); | |
853 | ||
854 | rb_erase(&n->rb_node, tree); | |
855 | output__insert_entry(n); | |
856 | } | |
857 | } | |
858 | ||
859 | static size_t output__fprintf(FILE *fp, uint64_t total_samples) | |
860 | { | |
861 | struct hist_entry *pos; | |
862 | struct sort_entry *se; | |
863 | struct rb_node *nd; | |
864 | size_t ret = 0; | |
865 | ||
866 | fprintf(fp, "\n"); | |
867 | fprintf(fp, "#\n"); | |
868 | fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples); | |
869 | fprintf(fp, "#\n"); | |
870 | ||
871 | fprintf(fp, "# Overhead"); | |
872 | list_for_each_entry(se, &hist_entry__sort_list, list) | |
873 | fprintf(fp, " %s", se->header); | |
874 | fprintf(fp, "\n"); | |
875 | ||
876 | fprintf(fp, "# ........"); | |
877 | list_for_each_entry(se, &hist_entry__sort_list, list) { | |
878 | int i; | |
879 | ||
880 | fprintf(fp, " "); | |
881 | for (i = 0; i < strlen(se->header); i++) | |
882 | fprintf(fp, "."); | |
883 | } | |
884 | fprintf(fp, "\n"); | |
885 | ||
886 | fprintf(fp, "#\n"); | |
887 | ||
888 | for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) { | |
889 | pos = rb_entry(nd, struct hist_entry, rb_node); | |
890 | ret += hist_entry__fprintf(fp, pos, total_samples); | |
891 | } | |
892 | ||
893 | if (!strcmp(sort_order, default_sort_order)) { | |
894 | fprintf(fp, "#\n"); | |
895 | fprintf(fp, "# (For more details, try: perf annotate --sort comm,dso,symbol)\n"); | |
896 | fprintf(fp, "#\n"); | |
897 | } | |
898 | fprintf(fp, "\n"); | |
899 | ||
900 | return ret; | |
901 | } | |
902 | ||
903 | static void register_idle_thread(void) | |
904 | { | |
905 | struct thread *thread = threads__findnew(0); | |
906 | ||
907 | if (thread == NULL || | |
908 | thread__set_comm(thread, "[idle]")) { | |
909 | fprintf(stderr, "problem inserting idle task.\n"); | |
910 | exit(-1); | |
911 | } | |
912 | } | |
913 | ||
914 | static unsigned long total = 0, | |
915 | total_mmap = 0, | |
916 | total_comm = 0, | |
917 | total_fork = 0, | |
918 | total_unknown = 0; | |
919 | ||
920 | static int | |
921 | process_overflow_event(event_t *event, unsigned long offset, unsigned long head) | |
922 | { | |
923 | char level; | |
924 | int show = 0; | |
925 | struct dso *dso = NULL; | |
926 | struct thread *thread = threads__findnew(event->ip.pid); | |
927 | uint64_t ip = event->ip.ip; | |
928 | struct map *map = NULL; | |
929 | ||
930 | dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p\n", | |
931 | (void *)(offset + head), | |
932 | (void *)(long)(event->header.size), | |
933 | event->header.misc, | |
934 | event->ip.pid, | |
935 | (void *)(long)ip); | |
936 | ||
937 | dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid); | |
938 | ||
939 | if (thread == NULL) { | |
940 | fprintf(stderr, "problem processing %d event, skipping it.\n", | |
941 | event->header.type); | |
942 | return -1; | |
943 | } | |
944 | ||
945 | if (event->header.misc & PERF_EVENT_MISC_KERNEL) { | |
946 | show = SHOW_KERNEL; | |
947 | level = 'k'; | |
948 | ||
949 | dso = kernel_dso; | |
950 | ||
951 | dprintf(" ...... dso: %s\n", dso->name); | |
952 | ||
953 | } else if (event->header.misc & PERF_EVENT_MISC_USER) { | |
954 | ||
955 | show = SHOW_USER; | |
956 | level = '.'; | |
957 | ||
958 | map = thread__find_map(thread, ip); | |
959 | if (map != NULL) { | |
960 | ip = map->map_ip(map, ip); | |
961 | dso = map->dso; | |
962 | } else { | |
963 | /* | |
964 | * If this is outside of all known maps, | |
965 | * and is a negative address, try to look it | |
966 | * up in the kernel dso, as it might be a | |
967 | * vsyscall (which executes in user-mode): | |
968 | */ | |
969 | if ((long long)ip < 0) | |
970 | dso = kernel_dso; | |
971 | } | |
972 | dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>"); | |
973 | ||
974 | } else { | |
975 | show = SHOW_HV; | |
976 | level = 'H'; | |
977 | dprintf(" ...... dso: [hypervisor]\n"); | |
978 | } | |
979 | ||
980 | if (show & show_mask) { | |
981 | struct symbol *sym = NULL; | |
982 | ||
983 | if (dso) | |
984 | sym = dso->find_symbol(dso, ip); | |
985 | ||
986 | if (hist_entry__add(thread, map, dso, sym, ip, level)) { | |
987 | fprintf(stderr, | |
988 | "problem incrementing symbol count, skipping event\n"); | |
989 | return -1; | |
990 | } | |
991 | } | |
992 | total++; | |
993 | ||
994 | return 0; | |
995 | } | |
996 | ||
997 | static int | |
998 | process_mmap_event(event_t *event, unsigned long offset, unsigned long head) | |
999 | { | |
1000 | struct thread *thread = threads__findnew(event->mmap.pid); | |
1001 | struct map *map = map__new(&event->mmap); | |
1002 | ||
1003 | dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n", | |
1004 | (void *)(offset + head), | |
1005 | (void *)(long)(event->header.size), | |
1006 | event->mmap.pid, | |
1007 | (void *)(long)event->mmap.start, | |
1008 | (void *)(long)event->mmap.len, | |
1009 | (void *)(long)event->mmap.pgoff, | |
1010 | event->mmap.filename); | |
1011 | ||
1012 | if (thread == NULL || map == NULL) { | |
1013 | dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n"); | |
1014 | return 0; | |
1015 | } | |
1016 | ||
1017 | thread__insert_map(thread, map); | |
1018 | total_mmap++; | |
1019 | ||
1020 | return 0; | |
1021 | } | |
1022 | ||
1023 | static int | |
1024 | process_comm_event(event_t *event, unsigned long offset, unsigned long head) | |
1025 | { | |
1026 | struct thread *thread = threads__findnew(event->comm.pid); | |
1027 | ||
1028 | dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n", | |
1029 | (void *)(offset + head), | |
1030 | (void *)(long)(event->header.size), | |
1031 | event->comm.comm, event->comm.pid); | |
1032 | ||
1033 | if (thread == NULL || | |
1034 | thread__set_comm(thread, event->comm.comm)) { | |
1035 | dprintf("problem processing PERF_EVENT_COMM, skipping event.\n"); | |
1036 | return -1; | |
1037 | } | |
1038 | total_comm++; | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
1043 | static int | |
1044 | process_fork_event(event_t *event, unsigned long offset, unsigned long head) | |
1045 | { | |
1046 | struct thread *thread = threads__findnew(event->fork.pid); | |
1047 | struct thread *parent = threads__findnew(event->fork.ppid); | |
1048 | ||
1049 | dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n", | |
1050 | (void *)(offset + head), | |
1051 | (void *)(long)(event->header.size), | |
1052 | event->fork.pid, event->fork.ppid); | |
1053 | ||
1054 | if (!thread || !parent || thread__fork(thread, parent)) { | |
1055 | dprintf("problem processing PERF_EVENT_FORK, skipping event.\n"); | |
1056 | return -1; | |
1057 | } | |
1058 | total_fork++; | |
1059 | ||
1060 | return 0; | |
1061 | } | |
1062 | ||
1063 | static int | |
1064 | process_period_event(event_t *event, unsigned long offset, unsigned long head) | |
1065 | { | |
1066 | dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n", | |
1067 | (void *)(offset + head), | |
1068 | (void *)(long)(event->header.size), | |
1069 | event->period.time, | |
1070 | event->period.id, | |
1071 | event->period.sample_period); | |
1072 | ||
1073 | return 0; | |
1074 | } | |
1075 | ||
1076 | static int | |
1077 | process_event(event_t *event, unsigned long offset, unsigned long head) | |
1078 | { | |
1079 | if (event->header.misc & PERF_EVENT_MISC_OVERFLOW) | |
1080 | return process_overflow_event(event, offset, head); | |
1081 | ||
1082 | switch (event->header.type) { | |
1083 | case PERF_EVENT_MMAP: | |
1084 | return process_mmap_event(event, offset, head); | |
1085 | ||
1086 | case PERF_EVENT_COMM: | |
1087 | return process_comm_event(event, offset, head); | |
1088 | ||
1089 | case PERF_EVENT_FORK: | |
1090 | return process_fork_event(event, offset, head); | |
1091 | ||
1092 | case PERF_EVENT_PERIOD: | |
1093 | return process_period_event(event, offset, head); | |
1094 | /* | |
1095 | * We dont process them right now but they are fine: | |
1096 | */ | |
1097 | ||
1098 | case PERF_EVENT_THROTTLE: | |
1099 | case PERF_EVENT_UNTHROTTLE: | |
1100 | return 0; | |
1101 | ||
1102 | default: | |
1103 | return -1; | |
1104 | } | |
1105 | ||
1106 | return 0; | |
1107 | } | |
1108 | ||
1109 | static int __cmd_annotate(void) | |
1110 | { | |
1111 | int ret, rc = EXIT_FAILURE; | |
1112 | unsigned long offset = 0; | |
1113 | unsigned long head = 0; | |
1114 | struct stat stat; | |
1115 | event_t *event; | |
1116 | uint32_t size; | |
1117 | char *buf; | |
1118 | ||
1119 | register_idle_thread(); | |
1120 | ||
1121 | input = open(input_name, O_RDONLY); | |
1122 | if (input < 0) { | |
1123 | perror("failed to open file"); | |
1124 | exit(-1); | |
1125 | } | |
1126 | ||
1127 | ret = fstat(input, &stat); | |
1128 | if (ret < 0) { | |
1129 | perror("failed to stat file"); | |
1130 | exit(-1); | |
1131 | } | |
1132 | ||
1133 | if (!stat.st_size) { | |
1134 | fprintf(stderr, "zero-sized file, nothing to do!\n"); | |
1135 | exit(0); | |
1136 | } | |
1137 | ||
1138 | if (load_kernel() < 0) { | |
1139 | perror("failed to load kernel symbols"); | |
1140 | return EXIT_FAILURE; | |
1141 | } | |
1142 | ||
1143 | if (!full_paths) { | |
1144 | if (getcwd(__cwd, sizeof(__cwd)) == NULL) { | |
1145 | perror("failed to get the current directory"); | |
1146 | return EXIT_FAILURE; | |
1147 | } | |
1148 | cwdlen = strlen(cwd); | |
1149 | } else { | |
1150 | cwd = NULL; | |
1151 | cwdlen = 0; | |
1152 | } | |
1153 | remap: | |
1154 | buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ, | |
1155 | MAP_SHARED, input, offset); | |
1156 | if (buf == MAP_FAILED) { | |
1157 | perror("failed to mmap file"); | |
1158 | exit(-1); | |
1159 | } | |
1160 | ||
1161 | more: | |
1162 | event = (event_t *)(buf + head); | |
1163 | ||
1164 | size = event->header.size; | |
1165 | if (!size) | |
1166 | size = 8; | |
1167 | ||
1168 | if (head + event->header.size >= page_size * mmap_window) { | |
1169 | unsigned long shift = page_size * (head / page_size); | |
1170 | int ret; | |
1171 | ||
1172 | ret = munmap(buf, page_size * mmap_window); | |
1173 | assert(ret == 0); | |
1174 | ||
1175 | offset += shift; | |
1176 | head -= shift; | |
1177 | goto remap; | |
1178 | } | |
1179 | ||
1180 | size = event->header.size; | |
1181 | ||
1182 | dprintf("%p [%p]: event: %d\n", | |
1183 | (void *)(offset + head), | |
1184 | (void *)(long)event->header.size, | |
1185 | event->header.type); | |
1186 | ||
1187 | if (!size || process_event(event, offset, head) < 0) { | |
1188 | ||
1189 | dprintf("%p [%p]: skipping unknown header type: %d\n", | |
1190 | (void *)(offset + head), | |
1191 | (void *)(long)(event->header.size), | |
1192 | event->header.type); | |
1193 | ||
1194 | total_unknown++; | |
1195 | ||
1196 | /* | |
1197 | * assume we lost track of the stream, check alignment, and | |
1198 | * increment a single u64 in the hope to catch on again 'soon'. | |
1199 | */ | |
1200 | ||
1201 | if (unlikely(head & 7)) | |
1202 | head &= ~7ULL; | |
1203 | ||
1204 | size = 8; | |
1205 | } | |
1206 | ||
1207 | head += size; | |
1208 | ||
1209 | if (offset + head < stat.st_size) | |
1210 | goto more; | |
1211 | ||
1212 | rc = EXIT_SUCCESS; | |
1213 | close(input); | |
1214 | ||
1215 | dprintf(" IP events: %10ld\n", total); | |
1216 | dprintf(" mmap events: %10ld\n", total_mmap); | |
1217 | dprintf(" comm events: %10ld\n", total_comm); | |
1218 | dprintf(" fork events: %10ld\n", total_fork); | |
1219 | dprintf(" unknown events: %10ld\n", total_unknown); | |
1220 | ||
1221 | if (dump_trace) | |
1222 | return 0; | |
1223 | ||
1224 | if (verbose >= 3) | |
1225 | threads__fprintf(stdout); | |
1226 | ||
1227 | if (verbose >= 2) | |
1228 | dsos__fprintf(stdout); | |
1229 | ||
1230 | collapse__resort(); | |
1231 | output__resort(); | |
1232 | output__fprintf(stdout, total); | |
1233 | ||
1234 | return rc; | |
1235 | } | |
1236 | ||
1237 | static const char * const annotate_usage[] = { | |
1238 | "perf annotate [<options>] <command>", | |
1239 | NULL | |
1240 | }; | |
1241 | ||
1242 | static const struct option options[] = { | |
1243 | OPT_STRING('i', "input", &input_name, "file", | |
1244 | "input file name"), | |
1245 | OPT_BOOLEAN('v', "verbose", &verbose, | |
1246 | "be more verbose (show symbol address, etc)"), | |
1247 | OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, | |
1248 | "dump raw trace in ASCII"), | |
1249 | OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"), | |
1250 | OPT_STRING('s', "sort", &sort_order, "key[,key2...]", | |
1251 | "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"), | |
1252 | OPT_BOOLEAN('P', "full-paths", &full_paths, | |
1253 | "Don't shorten the pathnames taking into account the cwd"), | |
1254 | OPT_END() | |
1255 | }; | |
1256 | ||
1257 | static void setup_sorting(void) | |
1258 | { | |
1259 | char *tmp, *tok, *str = strdup(sort_order); | |
1260 | ||
1261 | for (tok = strtok_r(str, ", ", &tmp); | |
1262 | tok; tok = strtok_r(NULL, ", ", &tmp)) { | |
1263 | if (sort_dimension__add(tok) < 0) { | |
1264 | error("Unknown --sort key: `%s'", tok); | |
1265 | usage_with_options(annotate_usage, options); | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | free(str); | |
1270 | } | |
1271 | ||
1272 | int cmd_annotate(int argc, const char **argv, const char *prefix) | |
1273 | { | |
1274 | symbol__init(); | |
1275 | ||
1276 | page_size = getpagesize(); | |
1277 | ||
1278 | argc = parse_options(argc, argv, options, annotate_usage, 0); | |
1279 | ||
1280 | setup_sorting(); | |
1281 | ||
1282 | /* | |
1283 | * Any (unrecognized) arguments left? | |
1284 | */ | |
1285 | if (argc) | |
1286 | usage_with_options(annotate_usage, options); | |
1287 | ||
1288 | setup_pager(); | |
1289 | ||
1290 | return __cmd_annotate(); | |
1291 | } |