1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
4  *
5  * Parts came from builtin-{top,stat,record}.c, see those files for further
6  * copyright notes.
7  */
8 #include <api/fs/fs.h>
9 #include <errno.h>
10 #include <inttypes.h>
11 #include <poll.h>
12 #include "cpumap.h"
13 #include "util/mmap.h"
14 #include "thread_map.h"
15 #include "target.h"
16 #include "evlist.h"
17 #include "evsel.h"
18 #include "record.h"
19 #include "debug.h"
20 #include "units.h"
21 #include "bpf_counter.h"
22 #include <internal/lib.h> // page_size
23 #include "affinity.h"
24 #include "../perf.h"
25 #include "asm/bug.h"
26 #include "bpf-event.h"
27 #include "util/event.h"
28 #include "util/string2.h"
29 #include "util/perf_api_probe.h"
30 #include "util/evsel_fprintf.h"
31 #include "util/pmu.h"
32 #include "util/sample.h"
33 #include "util/bpf-filter.h"
34 #include "util/stat.h"
35 #include "util/util.h"
36 #include <signal.h>
37 #include <unistd.h>
38 #include <sched.h>
39 #include <stdlib.h>
40 
41 #include "parse-events.h"
42 #include <subcmd/parse-options.h>
43 
44 #include <fcntl.h>
45 #include <sys/ioctl.h>
46 #include <sys/mman.h>
47 #include <sys/prctl.h>
48 #include <sys/timerfd.h>
49 
50 #include <linux/bitops.h>
51 #include <linux/hash.h>
52 #include <linux/log2.h>
53 #include <linux/err.h>
54 #include <linux/string.h>
55 #include <linux/time64.h>
56 #include <linux/zalloc.h>
57 #include <perf/evlist.h>
58 #include <perf/evsel.h>
59 #include <perf/cpumap.h>
60 #include <perf/mmap.h>
61 
62 #include <internal/xyarray.h>
63 
64 #ifdef LACKS_SIGQUEUE_PROTOTYPE
65 int sigqueue(pid_t pid, int sig, const union sigval value);
66 #endif
67 
68 #define FD(e, x, y) (*(int *)xyarray__entry(e->core.fd, x, y))
69 #define SID(e, x, y) xyarray__entry(e->core.sample_id, x, y)
70 
evlist__init(struct evlist * evlist,struct perf_cpu_map * cpus,struct perf_thread_map * threads)71 void evlist__init(struct evlist *evlist, struct perf_cpu_map *cpus,
72 		  struct perf_thread_map *threads)
73 {
74 	perf_evlist__init(&evlist->core);
75 	perf_evlist__set_maps(&evlist->core, cpus, threads);
76 	evlist->workload.pid = -1;
77 	evlist->bkw_mmap_state = BKW_MMAP_NOTREADY;
78 	evlist->ctl_fd.fd = -1;
79 	evlist->ctl_fd.ack = -1;
80 	evlist->ctl_fd.pos = -1;
81 }
82 
evlist__new(void)83 struct evlist *evlist__new(void)
84 {
85 	struct evlist *evlist = zalloc(sizeof(*evlist));
86 
87 	if (evlist != NULL)
88 		evlist__init(evlist, NULL, NULL);
89 
90 	return evlist;
91 }
92 
evlist__new_default(void)93 struct evlist *evlist__new_default(void)
94 {
95 	struct evlist *evlist = evlist__new();
96 	bool can_profile_kernel;
97 	int err;
98 
99 	if (!evlist)
100 		return NULL;
101 
102 	can_profile_kernel = perf_event_paranoid_check(1);
103 	err = parse_event(evlist, can_profile_kernel ? "cycles:P" : "cycles:Pu");
104 	if (err) {
105 		evlist__delete(evlist);
106 		return NULL;
107 	}
108 
109 	if (evlist->core.nr_entries > 1) {
110 		struct evsel *evsel;
111 
112 		evlist__for_each_entry(evlist, evsel)
113 			evsel__set_sample_id(evsel, /*can_sample_identifier=*/false);
114 	}
115 
116 	return evlist;
117 }
118 
evlist__new_dummy(void)119 struct evlist *evlist__new_dummy(void)
120 {
121 	struct evlist *evlist = evlist__new();
122 
123 	if (evlist && evlist__add_dummy(evlist)) {
124 		evlist__delete(evlist);
125 		evlist = NULL;
126 	}
127 
128 	return evlist;
129 }
130 
131 /**
132  * evlist__set_id_pos - set the positions of event ids.
133  * @evlist: selected event list
134  *
135  * Events with compatible sample types all have the same id_pos
136  * and is_pos.  For convenience, put a copy on evlist.
137  */
evlist__set_id_pos(struct evlist * evlist)138 void evlist__set_id_pos(struct evlist *evlist)
139 {
140 	struct evsel *first = evlist__first(evlist);
141 
142 	evlist->id_pos = first->id_pos;
143 	evlist->is_pos = first->is_pos;
144 }
145 
evlist__update_id_pos(struct evlist * evlist)146 static void evlist__update_id_pos(struct evlist *evlist)
147 {
148 	struct evsel *evsel;
149 
150 	evlist__for_each_entry(evlist, evsel)
151 		evsel__calc_id_pos(evsel);
152 
153 	evlist__set_id_pos(evlist);
154 }
155 
evlist__purge(struct evlist * evlist)156 static void evlist__purge(struct evlist *evlist)
157 {
158 	struct evsel *pos, *n;
159 
160 	evlist__for_each_entry_safe(evlist, n, pos) {
161 		list_del_init(&pos->core.node);
162 		pos->evlist = NULL;
163 		evsel__delete(pos);
164 	}
165 
166 	evlist->core.nr_entries = 0;
167 }
168 
evlist__exit(struct evlist * evlist)169 void evlist__exit(struct evlist *evlist)
170 {
171 	event_enable_timer__exit(&evlist->eet);
172 	zfree(&evlist->mmap);
173 	zfree(&evlist->overwrite_mmap);
174 	perf_evlist__exit(&evlist->core);
175 }
176 
evlist__delete(struct evlist * evlist)177 void evlist__delete(struct evlist *evlist)
178 {
179 	if (evlist == NULL)
180 		return;
181 
182 	evlist__free_stats(evlist);
183 	evlist__munmap(evlist);
184 	evlist__close(evlist);
185 	evlist__purge(evlist);
186 	evlist__exit(evlist);
187 	free(evlist);
188 }
189 
evlist__add(struct evlist * evlist,struct evsel * entry)190 void evlist__add(struct evlist *evlist, struct evsel *entry)
191 {
192 	perf_evlist__add(&evlist->core, &entry->core);
193 	entry->evlist = evlist;
194 	entry->tracking = !entry->core.idx;
195 
196 	if (evlist->core.nr_entries == 1)
197 		evlist__set_id_pos(evlist);
198 }
199 
evlist__remove(struct evlist * evlist,struct evsel * evsel)200 void evlist__remove(struct evlist *evlist, struct evsel *evsel)
201 {
202 	evsel->evlist = NULL;
203 	perf_evlist__remove(&evlist->core, &evsel->core);
204 }
205 
evlist__splice_list_tail(struct evlist * evlist,struct list_head * list)206 void evlist__splice_list_tail(struct evlist *evlist, struct list_head *list)
207 {
208 	while (!list_empty(list)) {
209 		struct evsel *evsel, *temp, *leader = NULL;
210 
211 		__evlist__for_each_entry_safe(list, temp, evsel) {
212 			list_del_init(&evsel->core.node);
213 			evlist__add(evlist, evsel);
214 			leader = evsel;
215 			break;
216 		}
217 
218 		__evlist__for_each_entry_safe(list, temp, evsel) {
219 			if (evsel__has_leader(evsel, leader)) {
220 				list_del_init(&evsel->core.node);
221 				evlist__add(evlist, evsel);
222 			}
223 		}
224 	}
225 }
226 
__evlist__set_tracepoints_handlers(struct evlist * evlist,const struct evsel_str_handler * assocs,size_t nr_assocs)227 int __evlist__set_tracepoints_handlers(struct evlist *evlist,
228 				       const struct evsel_str_handler *assocs, size_t nr_assocs)
229 {
230 	size_t i;
231 	int err;
232 
233 	for (i = 0; i < nr_assocs; i++) {
234 		// Adding a handler for an event not in this evlist, just ignore it.
235 		struct evsel *evsel = evlist__find_tracepoint_by_name(evlist, assocs[i].name);
236 		if (evsel == NULL)
237 			continue;
238 
239 		err = -EEXIST;
240 		if (evsel->handler != NULL)
241 			goto out;
242 		evsel->handler = assocs[i].handler;
243 	}
244 
245 	err = 0;
246 out:
247 	return err;
248 }
249 
evlist__set_leader(struct evlist * evlist)250 static void evlist__set_leader(struct evlist *evlist)
251 {
252 	perf_evlist__set_leader(&evlist->core);
253 }
254 
evlist__dummy_event(struct evlist * evlist)255 static struct evsel *evlist__dummy_event(struct evlist *evlist)
256 {
257 	struct perf_event_attr attr = {
258 		.type	= PERF_TYPE_SOFTWARE,
259 		.config = PERF_COUNT_SW_DUMMY,
260 		.size	= sizeof(attr), /* to capture ABI version */
261 		/* Avoid frequency mode for dummy events to avoid associated timers. */
262 		.freq = 0,
263 		.sample_period = 1,
264 	};
265 
266 	return evsel__new_idx(&attr, evlist->core.nr_entries);
267 }
268 
evlist__add_dummy(struct evlist * evlist)269 int evlist__add_dummy(struct evlist *evlist)
270 {
271 	struct evsel *evsel = evlist__dummy_event(evlist);
272 
273 	if (evsel == NULL)
274 		return -ENOMEM;
275 
276 	evlist__add(evlist, evsel);
277 	return 0;
278 }
279 
evlist__add_aux_dummy(struct evlist * evlist,bool system_wide)280 struct evsel *evlist__add_aux_dummy(struct evlist *evlist, bool system_wide)
281 {
282 	struct evsel *evsel = evlist__dummy_event(evlist);
283 
284 	if (!evsel)
285 		return NULL;
286 
287 	evsel->core.attr.exclude_kernel = 1;
288 	evsel->core.attr.exclude_guest = 1;
289 	evsel->core.attr.exclude_hv = 1;
290 	evsel->core.system_wide = system_wide;
291 	evsel->no_aux_samples = true;
292 	evsel->name = strdup("dummy:u");
293 
294 	evlist__add(evlist, evsel);
295 	return evsel;
296 }
297 
298 #ifdef HAVE_LIBTRACEEVENT
evlist__add_sched_switch(struct evlist * evlist,bool system_wide)299 struct evsel *evlist__add_sched_switch(struct evlist *evlist, bool system_wide)
300 {
301 	struct evsel *evsel = evsel__newtp_idx("sched", "sched_switch", 0);
302 
303 	if (IS_ERR(evsel))
304 		return evsel;
305 
306 	evsel__set_sample_bit(evsel, CPU);
307 	evsel__set_sample_bit(evsel, TIME);
308 
309 	evsel->core.system_wide = system_wide;
310 	evsel->no_aux_samples = true;
311 
312 	evlist__add(evlist, evsel);
313 	return evsel;
314 }
315 #endif
316 
evlist__add_attrs(struct evlist * evlist,struct perf_event_attr * attrs,size_t nr_attrs)317 int evlist__add_attrs(struct evlist *evlist, struct perf_event_attr *attrs, size_t nr_attrs)
318 {
319 	struct evsel *evsel, *n;
320 	LIST_HEAD(head);
321 	size_t i;
322 
323 	for (i = 0; i < nr_attrs; i++) {
324 		evsel = evsel__new_idx(attrs + i, evlist->core.nr_entries + i);
325 		if (evsel == NULL)
326 			goto out_delete_partial_list;
327 		list_add_tail(&evsel->core.node, &head);
328 	}
329 
330 	evlist__splice_list_tail(evlist, &head);
331 
332 	return 0;
333 
334 out_delete_partial_list:
335 	__evlist__for_each_entry_safe(&head, n, evsel)
336 		evsel__delete(evsel);
337 	return -1;
338 }
339 
__evlist__add_default_attrs(struct evlist * evlist,struct perf_event_attr * attrs,size_t nr_attrs)340 int __evlist__add_default_attrs(struct evlist *evlist, struct perf_event_attr *attrs, size_t nr_attrs)
341 {
342 	size_t i;
343 
344 	for (i = 0; i < nr_attrs; i++)
345 		event_attr_init(attrs + i);
346 
347 	return evlist__add_attrs(evlist, attrs, nr_attrs);
348 }
349 
arch_evlist__add_default_attrs(struct evlist * evlist,struct perf_event_attr * attrs,size_t nr_attrs)350 __weak int arch_evlist__add_default_attrs(struct evlist *evlist,
351 					  struct perf_event_attr *attrs,
352 					  size_t nr_attrs)
353 {
354 	if (!nr_attrs)
355 		return 0;
356 
357 	return __evlist__add_default_attrs(evlist, attrs, nr_attrs);
358 }
359 
evlist__find_tracepoint_by_id(struct evlist * evlist,int id)360 struct evsel *evlist__find_tracepoint_by_id(struct evlist *evlist, int id)
361 {
362 	struct evsel *evsel;
363 
364 	evlist__for_each_entry(evlist, evsel) {
365 		if (evsel->core.attr.type   == PERF_TYPE_TRACEPOINT &&
366 		    (int)evsel->core.attr.config == id)
367 			return evsel;
368 	}
369 
370 	return NULL;
371 }
372 
evlist__find_tracepoint_by_name(struct evlist * evlist,const char * name)373 struct evsel *evlist__find_tracepoint_by_name(struct evlist *evlist, const char *name)
374 {
375 	struct evsel *evsel;
376 
377 	evlist__for_each_entry(evlist, evsel) {
378 		if ((evsel->core.attr.type == PERF_TYPE_TRACEPOINT) &&
379 		    (strcmp(evsel->name, name) == 0))
380 			return evsel;
381 	}
382 
383 	return NULL;
384 }
385 
386 #ifdef HAVE_LIBTRACEEVENT
evlist__add_newtp(struct evlist * evlist,const char * sys,const char * name,void * handler)387 int evlist__add_newtp(struct evlist *evlist, const char *sys, const char *name, void *handler)
388 {
389 	struct evsel *evsel = evsel__newtp(sys, name);
390 
391 	if (IS_ERR(evsel))
392 		return -1;
393 
394 	evsel->handler = handler;
395 	evlist__add(evlist, evsel);
396 	return 0;
397 }
398 #endif
399 
evlist__cpu_begin(struct evlist * evlist,struct affinity * affinity)400 struct evlist_cpu_iterator evlist__cpu_begin(struct evlist *evlist, struct affinity *affinity)
401 {
402 	struct evlist_cpu_iterator itr = {
403 		.container = evlist,
404 		.evsel = NULL,
405 		.cpu_map_idx = 0,
406 		.evlist_cpu_map_idx = 0,
407 		.evlist_cpu_map_nr = perf_cpu_map__nr(evlist->core.all_cpus),
408 		.cpu = (struct perf_cpu){ .cpu = -1},
409 		.affinity = affinity,
410 	};
411 
412 	if (evlist__empty(evlist)) {
413 		/* Ensure the empty list doesn't iterate. */
414 		itr.evlist_cpu_map_idx = itr.evlist_cpu_map_nr;
415 	} else {
416 		itr.evsel = evlist__first(evlist);
417 		if (itr.affinity) {
418 			itr.cpu = perf_cpu_map__cpu(evlist->core.all_cpus, 0);
419 			affinity__set(itr.affinity, itr.cpu.cpu);
420 			itr.cpu_map_idx = perf_cpu_map__idx(itr.evsel->core.cpus, itr.cpu);
421 			/*
422 			 * If this CPU isn't in the evsel's cpu map then advance
423 			 * through the list.
424 			 */
425 			if (itr.cpu_map_idx == -1)
426 				evlist_cpu_iterator__next(&itr);
427 		}
428 	}
429 	return itr;
430 }
431 
evlist_cpu_iterator__next(struct evlist_cpu_iterator * evlist_cpu_itr)432 void evlist_cpu_iterator__next(struct evlist_cpu_iterator *evlist_cpu_itr)
433 {
434 	while (evlist_cpu_itr->evsel != evlist__last(evlist_cpu_itr->container)) {
435 		evlist_cpu_itr->evsel = evsel__next(evlist_cpu_itr->evsel);
436 		evlist_cpu_itr->cpu_map_idx =
437 			perf_cpu_map__idx(evlist_cpu_itr->evsel->core.cpus,
438 					  evlist_cpu_itr->cpu);
439 		if (evlist_cpu_itr->cpu_map_idx != -1)
440 			return;
441 	}
442 	evlist_cpu_itr->evlist_cpu_map_idx++;
443 	if (evlist_cpu_itr->evlist_cpu_map_idx < evlist_cpu_itr->evlist_cpu_map_nr) {
444 		evlist_cpu_itr->evsel = evlist__first(evlist_cpu_itr->container);
445 		evlist_cpu_itr->cpu =
446 			perf_cpu_map__cpu(evlist_cpu_itr->container->core.all_cpus,
447 					  evlist_cpu_itr->evlist_cpu_map_idx);
448 		if (evlist_cpu_itr->affinity)
449 			affinity__set(evlist_cpu_itr->affinity, evlist_cpu_itr->cpu.cpu);
450 		evlist_cpu_itr->cpu_map_idx =
451 			perf_cpu_map__idx(evlist_cpu_itr->evsel->core.cpus,
452 					  evlist_cpu_itr->cpu);
453 		/*
454 		 * If this CPU isn't in the evsel's cpu map then advance through
455 		 * the list.
456 		 */
457 		if (evlist_cpu_itr->cpu_map_idx == -1)
458 			evlist_cpu_iterator__next(evlist_cpu_itr);
459 	}
460 }
461 
evlist_cpu_iterator__end(const struct evlist_cpu_iterator * evlist_cpu_itr)462 bool evlist_cpu_iterator__end(const struct evlist_cpu_iterator *evlist_cpu_itr)
463 {
464 	return evlist_cpu_itr->evlist_cpu_map_idx >= evlist_cpu_itr->evlist_cpu_map_nr;
465 }
466 
evsel__strcmp(struct evsel * pos,char * evsel_name)467 static int evsel__strcmp(struct evsel *pos, char *evsel_name)
468 {
469 	if (!evsel_name)
470 		return 0;
471 	if (evsel__is_dummy_event(pos))
472 		return 1;
473 	return !evsel__name_is(pos, evsel_name);
474 }
475 
evlist__is_enabled(struct evlist * evlist)476 static int evlist__is_enabled(struct evlist *evlist)
477 {
478 	struct evsel *pos;
479 
480 	evlist__for_each_entry(evlist, pos) {
481 		if (!evsel__is_group_leader(pos) || !pos->core.fd)
482 			continue;
483 		/* If at least one event is enabled, evlist is enabled. */
484 		if (!pos->disabled)
485 			return true;
486 	}
487 	return false;
488 }
489 
__evlist__disable(struct evlist * evlist,char * evsel_name,bool excl_dummy)490 static void __evlist__disable(struct evlist *evlist, char *evsel_name, bool excl_dummy)
491 {
492 	struct evsel *pos;
493 	struct evlist_cpu_iterator evlist_cpu_itr;
494 	struct affinity saved_affinity, *affinity = NULL;
495 	bool has_imm = false;
496 
497 	// See explanation in evlist__close()
498 	if (!cpu_map__is_dummy(evlist->core.user_requested_cpus)) {
499 		if (affinity__setup(&saved_affinity) < 0)
500 			return;
501 		affinity = &saved_affinity;
502 	}
503 
504 	/* Disable 'immediate' events last */
505 	for (int imm = 0; imm <= 1; imm++) {
506 		evlist__for_each_cpu(evlist_cpu_itr, evlist, affinity) {
507 			pos = evlist_cpu_itr.evsel;
508 			if (evsel__strcmp(pos, evsel_name))
509 				continue;
510 			if (pos->disabled || !evsel__is_group_leader(pos) || !pos->core.fd)
511 				continue;
512 			if (excl_dummy && evsel__is_dummy_event(pos))
513 				continue;
514 			if (pos->immediate)
515 				has_imm = true;
516 			if (pos->immediate != imm)
517 				continue;
518 			evsel__disable_cpu(pos, evlist_cpu_itr.cpu_map_idx);
519 		}
520 		if (!has_imm)
521 			break;
522 	}
523 
524 	affinity__cleanup(affinity);
525 	evlist__for_each_entry(evlist, pos) {
526 		if (evsel__strcmp(pos, evsel_name))
527 			continue;
528 		if (!evsel__is_group_leader(pos) || !pos->core.fd)
529 			continue;
530 		if (excl_dummy && evsel__is_dummy_event(pos))
531 			continue;
532 		pos->disabled = true;
533 	}
534 
535 	/*
536 	 * If we disabled only single event, we need to check
537 	 * the enabled state of the evlist manually.
538 	 */
539 	if (evsel_name)
540 		evlist->enabled = evlist__is_enabled(evlist);
541 	else
542 		evlist->enabled = false;
543 }
544 
evlist__disable(struct evlist * evlist)545 void evlist__disable(struct evlist *evlist)
546 {
547 	__evlist__disable(evlist, NULL, false);
548 }
549 
evlist__disable_non_dummy(struct evlist * evlist)550 void evlist__disable_non_dummy(struct evlist *evlist)
551 {
552 	__evlist__disable(evlist, NULL, true);
553 }
554 
evlist__disable_evsel(struct evlist * evlist,char * evsel_name)555 void evlist__disable_evsel(struct evlist *evlist, char *evsel_name)
556 {
557 	__evlist__disable(evlist, evsel_name, false);
558 }
559 
__evlist__enable(struct evlist * evlist,char * evsel_name,bool excl_dummy)560 static void __evlist__enable(struct evlist *evlist, char *evsel_name, bool excl_dummy)
561 {
562 	struct evsel *pos;
563 	struct evlist_cpu_iterator evlist_cpu_itr;
564 	struct affinity saved_affinity, *affinity = NULL;
565 
566 	// See explanation in evlist__close()
567 	if (!cpu_map__is_dummy(evlist->core.user_requested_cpus)) {
568 		if (affinity__setup(&saved_affinity) < 0)
569 			return;
570 		affinity = &saved_affinity;
571 	}
572 
573 	evlist__for_each_cpu(evlist_cpu_itr, evlist, affinity) {
574 		pos = evlist_cpu_itr.evsel;
575 		if (evsel__strcmp(pos, evsel_name))
576 			continue;
577 		if (!evsel__is_group_leader(pos) || !pos->core.fd)
578 			continue;
579 		if (excl_dummy && evsel__is_dummy_event(pos))
580 			continue;
581 		evsel__enable_cpu(pos, evlist_cpu_itr.cpu_map_idx);
582 	}
583 	affinity__cleanup(affinity);
584 	evlist__for_each_entry(evlist, pos) {
585 		if (evsel__strcmp(pos, evsel_name))
586 			continue;
587 		if (!evsel__is_group_leader(pos) || !pos->core.fd)
588 			continue;
589 		if (excl_dummy && evsel__is_dummy_event(pos))
590 			continue;
591 		pos->disabled = false;
592 	}
593 
594 	/*
595 	 * Even single event sets the 'enabled' for evlist,
596 	 * so the toggle can work properly and toggle to
597 	 * 'disabled' state.
598 	 */
599 	evlist->enabled = true;
600 }
601 
evlist__enable(struct evlist * evlist)602 void evlist__enable(struct evlist *evlist)
603 {
604 	__evlist__enable(evlist, NULL, false);
605 }
606 
evlist__enable_non_dummy(struct evlist * evlist)607 void evlist__enable_non_dummy(struct evlist *evlist)
608 {
609 	__evlist__enable(evlist, NULL, true);
610 }
611 
evlist__enable_evsel(struct evlist * evlist,char * evsel_name)612 void evlist__enable_evsel(struct evlist *evlist, char *evsel_name)
613 {
614 	__evlist__enable(evlist, evsel_name, false);
615 }
616 
evlist__toggle_enable(struct evlist * evlist)617 void evlist__toggle_enable(struct evlist *evlist)
618 {
619 	(evlist->enabled ? evlist__disable : evlist__enable)(evlist);
620 }
621 
evlist__add_pollfd(struct evlist * evlist,int fd)622 int evlist__add_pollfd(struct evlist *evlist, int fd)
623 {
624 	return perf_evlist__add_pollfd(&evlist->core, fd, NULL, POLLIN, fdarray_flag__default);
625 }
626 
evlist__filter_pollfd(struct evlist * evlist,short revents_and_mask)627 int evlist__filter_pollfd(struct evlist *evlist, short revents_and_mask)
628 {
629 	return perf_evlist__filter_pollfd(&evlist->core, revents_and_mask);
630 }
631 
632 #ifdef HAVE_EVENTFD_SUPPORT
evlist__add_wakeup_eventfd(struct evlist * evlist,int fd)633 int evlist__add_wakeup_eventfd(struct evlist *evlist, int fd)
634 {
635 	return perf_evlist__add_pollfd(&evlist->core, fd, NULL, POLLIN,
636 				       fdarray_flag__nonfilterable |
637 				       fdarray_flag__non_perf_event);
638 }
639 #endif
640 
evlist__poll(struct evlist * evlist,int timeout)641 int evlist__poll(struct evlist *evlist, int timeout)
642 {
643 	return perf_evlist__poll(&evlist->core, timeout);
644 }
645 
evlist__id2sid(struct evlist * evlist,u64 id)646 struct perf_sample_id *evlist__id2sid(struct evlist *evlist, u64 id)
647 {
648 	struct hlist_head *head;
649 	struct perf_sample_id *sid;
650 	int hash;
651 
652 	hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
653 	head = &evlist->core.heads[hash];
654 
655 	hlist_for_each_entry(sid, head, node)
656 		if (sid->id == id)
657 			return sid;
658 
659 	return NULL;
660 }
661 
evlist__id2evsel(struct evlist * evlist,u64 id)662 struct evsel *evlist__id2evsel(struct evlist *evlist, u64 id)
663 {
664 	struct perf_sample_id *sid;
665 
666 	if (evlist->core.nr_entries == 1 || !id)
667 		return evlist__first(evlist);
668 
669 	sid = evlist__id2sid(evlist, id);
670 	if (sid)
671 		return container_of(sid->evsel, struct evsel, core);
672 
673 	if (!evlist__sample_id_all(evlist))
674 		return evlist__first(evlist);
675 
676 	return NULL;
677 }
678 
evlist__id2evsel_strict(struct evlist * evlist,u64 id)679 struct evsel *evlist__id2evsel_strict(struct evlist *evlist, u64 id)
680 {
681 	struct perf_sample_id *sid;
682 
683 	if (!id)
684 		return NULL;
685 
686 	sid = evlist__id2sid(evlist, id);
687 	if (sid)
688 		return container_of(sid->evsel, struct evsel, core);
689 
690 	return NULL;
691 }
692 
evlist__event2id(struct evlist * evlist,union perf_event * event,u64 * id)693 static int evlist__event2id(struct evlist *evlist, union perf_event *event, u64 *id)
694 {
695 	const __u64 *array = event->sample.array;
696 	ssize_t n;
697 
698 	n = (event->header.size - sizeof(event->header)) >> 3;
699 
700 	if (event->header.type == PERF_RECORD_SAMPLE) {
701 		if (evlist->id_pos >= n)
702 			return -1;
703 		*id = array[evlist->id_pos];
704 	} else {
705 		if (evlist->is_pos > n)
706 			return -1;
707 		n -= evlist->is_pos;
708 		*id = array[n];
709 	}
710 	return 0;
711 }
712 
evlist__event2evsel(struct evlist * evlist,union perf_event * event)713 struct evsel *evlist__event2evsel(struct evlist *evlist, union perf_event *event)
714 {
715 	struct evsel *first = evlist__first(evlist);
716 	struct hlist_head *head;
717 	struct perf_sample_id *sid;
718 	int hash;
719 	u64 id;
720 
721 	if (evlist->core.nr_entries == 1)
722 		return first;
723 
724 	if (!first->core.attr.sample_id_all &&
725 	    event->header.type != PERF_RECORD_SAMPLE)
726 		return first;
727 
728 	if (evlist__event2id(evlist, event, &id))
729 		return NULL;
730 
731 	/* Synthesized events have an id of zero */
732 	if (!id)
733 		return first;
734 
735 	hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
736 	head = &evlist->core.heads[hash];
737 
738 	hlist_for_each_entry(sid, head, node) {
739 		if (sid->id == id)
740 			return container_of(sid->evsel, struct evsel, core);
741 	}
742 	return NULL;
743 }
744 
evlist__set_paused(struct evlist * evlist,bool value)745 static int evlist__set_paused(struct evlist *evlist, bool value)
746 {
747 	int i;
748 
749 	if (!evlist->overwrite_mmap)
750 		return 0;
751 
752 	for (i = 0; i < evlist->core.nr_mmaps; i++) {
753 		int fd = evlist->overwrite_mmap[i].core.fd;
754 		int err;
755 
756 		if (fd < 0)
757 			continue;
758 		err = ioctl(fd, PERF_EVENT_IOC_PAUSE_OUTPUT, value ? 1 : 0);
759 		if (err)
760 			return err;
761 	}
762 	return 0;
763 }
764 
evlist__pause(struct evlist * evlist)765 static int evlist__pause(struct evlist *evlist)
766 {
767 	return evlist__set_paused(evlist, true);
768 }
769 
evlist__resume(struct evlist * evlist)770 static int evlist__resume(struct evlist *evlist)
771 {
772 	return evlist__set_paused(evlist, false);
773 }
774 
evlist__munmap_nofree(struct evlist * evlist)775 static void evlist__munmap_nofree(struct evlist *evlist)
776 {
777 	int i;
778 
779 	if (evlist->mmap)
780 		for (i = 0; i < evlist->core.nr_mmaps; i++)
781 			perf_mmap__munmap(&evlist->mmap[i].core);
782 
783 	if (evlist->overwrite_mmap)
784 		for (i = 0; i < evlist->core.nr_mmaps; i++)
785 			perf_mmap__munmap(&evlist->overwrite_mmap[i].core);
786 }
787 
evlist__munmap(struct evlist * evlist)788 void evlist__munmap(struct evlist *evlist)
789 {
790 	evlist__munmap_nofree(evlist);
791 	zfree(&evlist->mmap);
792 	zfree(&evlist->overwrite_mmap);
793 }
794 
perf_mmap__unmap_cb(struct perf_mmap * map)795 static void perf_mmap__unmap_cb(struct perf_mmap *map)
796 {
797 	struct mmap *m = container_of(map, struct mmap, core);
798 
799 	mmap__munmap(m);
800 }
801 
evlist__alloc_mmap(struct evlist * evlist,bool overwrite)802 static struct mmap *evlist__alloc_mmap(struct evlist *evlist,
803 				       bool overwrite)
804 {
805 	int i;
806 	struct mmap *map;
807 
808 	map = zalloc(evlist->core.nr_mmaps * sizeof(struct mmap));
809 	if (!map)
810 		return NULL;
811 
812 	for (i = 0; i < evlist->core.nr_mmaps; i++) {
813 		struct perf_mmap *prev = i ? &map[i - 1].core : NULL;
814 
815 		/*
816 		 * When the perf_mmap() call is made we grab one refcount, plus
817 		 * one extra to let perf_mmap__consume() get the last
818 		 * events after all real references (perf_mmap__get()) are
819 		 * dropped.
820 		 *
821 		 * Each PERF_EVENT_IOC_SET_OUTPUT points to this mmap and
822 		 * thus does perf_mmap__get() on it.
823 		 */
824 		perf_mmap__init(&map[i].core, prev, overwrite, perf_mmap__unmap_cb);
825 	}
826 
827 	return map;
828 }
829 
830 static void
perf_evlist__mmap_cb_idx(struct perf_evlist * _evlist,struct perf_evsel * _evsel,struct perf_mmap_param * _mp,int idx)831 perf_evlist__mmap_cb_idx(struct perf_evlist *_evlist,
832 			 struct perf_evsel *_evsel,
833 			 struct perf_mmap_param *_mp,
834 			 int idx)
835 {
836 	struct evlist *evlist = container_of(_evlist, struct evlist, core);
837 	struct mmap_params *mp = container_of(_mp, struct mmap_params, core);
838 	struct evsel *evsel = container_of(_evsel, struct evsel, core);
839 
840 	auxtrace_mmap_params__set_idx(&mp->auxtrace_mp, evlist, evsel, idx);
841 }
842 
843 static struct perf_mmap*
perf_evlist__mmap_cb_get(struct perf_evlist * _evlist,bool overwrite,int idx)844 perf_evlist__mmap_cb_get(struct perf_evlist *_evlist, bool overwrite, int idx)
845 {
846 	struct evlist *evlist = container_of(_evlist, struct evlist, core);
847 	struct mmap *maps;
848 
849 	maps = overwrite ? evlist->overwrite_mmap : evlist->mmap;
850 
851 	if (!maps) {
852 		maps = evlist__alloc_mmap(evlist, overwrite);
853 		if (!maps)
854 			return NULL;
855 
856 		if (overwrite) {
857 			evlist->overwrite_mmap = maps;
858 			if (evlist->bkw_mmap_state == BKW_MMAP_NOTREADY)
859 				evlist__toggle_bkw_mmap(evlist, BKW_MMAP_RUNNING);
860 		} else {
861 			evlist->mmap = maps;
862 		}
863 	}
864 
865 	return &maps[idx].core;
866 }
867 
868 static int
perf_evlist__mmap_cb_mmap(struct perf_mmap * _map,struct perf_mmap_param * _mp,int output,struct perf_cpu cpu)869 perf_evlist__mmap_cb_mmap(struct perf_mmap *_map, struct perf_mmap_param *_mp,
870 			  int output, struct perf_cpu cpu)
871 {
872 	struct mmap *map = container_of(_map, struct mmap, core);
873 	struct mmap_params *mp = container_of(_mp, struct mmap_params, core);
874 
875 	return mmap__mmap(map, mp, output, cpu);
876 }
877 
perf_event_mlock_kb_in_pages(void)878 unsigned long perf_event_mlock_kb_in_pages(void)
879 {
880 	unsigned long pages;
881 	int max;
882 
883 	if (sysctl__read_int("kernel/perf_event_mlock_kb", &max) < 0) {
884 		/*
885 		 * Pick a once upon a time good value, i.e. things look
886 		 * strange since we can't read a sysctl value, but lets not
887 		 * die yet...
888 		 */
889 		max = 512;
890 	} else {
891 		max -= (page_size / 1024);
892 	}
893 
894 	pages = (max * 1024) / page_size;
895 	if (!is_power_of_2(pages))
896 		pages = rounddown_pow_of_two(pages);
897 
898 	return pages;
899 }
900 
evlist__mmap_size(unsigned long pages)901 size_t evlist__mmap_size(unsigned long pages)
902 {
903 	if (pages == UINT_MAX)
904 		pages = perf_event_mlock_kb_in_pages();
905 	else if (!is_power_of_2(pages))
906 		return 0;
907 
908 	return (pages + 1) * page_size;
909 }
910 
parse_pages_arg(const char * str,unsigned long min,unsigned long max)911 static long parse_pages_arg(const char *str, unsigned long min,
912 			    unsigned long max)
913 {
914 	unsigned long pages, val;
915 	static struct parse_tag tags[] = {
916 		{ .tag  = 'B', .mult = 1       },
917 		{ .tag  = 'K', .mult = 1 << 10 },
918 		{ .tag  = 'M', .mult = 1 << 20 },
919 		{ .tag  = 'G', .mult = 1 << 30 },
920 		{ .tag  = 0 },
921 	};
922 
923 	if (str == NULL)
924 		return -EINVAL;
925 
926 	val = parse_tag_value(str, tags);
927 	if (val != (unsigned long) -1) {
928 		/* we got file size value */
929 		pages = PERF_ALIGN(val, page_size) / page_size;
930 	} else {
931 		/* we got pages count value */
932 		char *eptr;
933 		pages = strtoul(str, &eptr, 10);
934 		if (*eptr != '\0')
935 			return -EINVAL;
936 	}
937 
938 	if (pages == 0 && min == 0) {
939 		/* leave number of pages at 0 */
940 	} else if (!is_power_of_2(pages)) {
941 		char buf[100];
942 
943 		/* round pages up to next power of 2 */
944 		pages = roundup_pow_of_two(pages);
945 		if (!pages)
946 			return -EINVAL;
947 
948 		unit_number__scnprintf(buf, sizeof(buf), pages * page_size);
949 		pr_info("rounding mmap pages size to %s (%lu pages)\n",
950 			buf, pages);
951 	}
952 
953 	if (pages > max)
954 		return -EINVAL;
955 
956 	return pages;
957 }
958 
__evlist__parse_mmap_pages(unsigned int * mmap_pages,const char * str)959 int __evlist__parse_mmap_pages(unsigned int *mmap_pages, const char *str)
960 {
961 	unsigned long max = UINT_MAX;
962 	long pages;
963 
964 	if (max > SIZE_MAX / page_size)
965 		max = SIZE_MAX / page_size;
966 
967 	pages = parse_pages_arg(str, 1, max);
968 	if (pages < 0) {
969 		pr_err("Invalid argument for --mmap_pages/-m\n");
970 		return -1;
971 	}
972 
973 	*mmap_pages = pages;
974 	return 0;
975 }
976 
evlist__parse_mmap_pages(const struct option * opt,const char * str,int unset __maybe_unused)977 int evlist__parse_mmap_pages(const struct option *opt, const char *str, int unset __maybe_unused)
978 {
979 	return __evlist__parse_mmap_pages(opt->value, str);
980 }
981 
982 /**
983  * evlist__mmap_ex - Create mmaps to receive events.
984  * @evlist: list of events
985  * @pages: map length in pages
986  * @overwrite: overwrite older events?
987  * @auxtrace_pages - auxtrace map length in pages
988  * @auxtrace_overwrite - overwrite older auxtrace data?
989  *
990  * If @overwrite is %false the user needs to signal event consumption using
991  * perf_mmap__write_tail().  Using evlist__mmap_read() does this
992  * automatically.
993  *
994  * Similarly, if @auxtrace_overwrite is %false the user needs to signal data
995  * consumption using auxtrace_mmap__write_tail().
996  *
997  * Return: %0 on success, negative error code otherwise.
998  */
evlist__mmap_ex(struct evlist * evlist,unsigned int pages,unsigned int auxtrace_pages,bool auxtrace_overwrite,int nr_cblocks,int affinity,int flush,int comp_level)999 int evlist__mmap_ex(struct evlist *evlist, unsigned int pages,
1000 			 unsigned int auxtrace_pages,
1001 			 bool auxtrace_overwrite, int nr_cblocks, int affinity, int flush,
1002 			 int comp_level)
1003 {
1004 	/*
1005 	 * Delay setting mp.prot: set it before calling perf_mmap__mmap.
1006 	 * Its value is decided by evsel's write_backward.
1007 	 * So &mp should not be passed through const pointer.
1008 	 */
1009 	struct mmap_params mp = {
1010 		.nr_cblocks	= nr_cblocks,
1011 		.affinity	= affinity,
1012 		.flush		= flush,
1013 		.comp_level	= comp_level
1014 	};
1015 	struct perf_evlist_mmap_ops ops = {
1016 		.idx  = perf_evlist__mmap_cb_idx,
1017 		.get  = perf_evlist__mmap_cb_get,
1018 		.mmap = perf_evlist__mmap_cb_mmap,
1019 	};
1020 
1021 	evlist->core.mmap_len = evlist__mmap_size(pages);
1022 	pr_debug("mmap size %zuB\n", evlist->core.mmap_len);
1023 
1024 	auxtrace_mmap_params__init(&mp.auxtrace_mp, evlist->core.mmap_len,
1025 				   auxtrace_pages, auxtrace_overwrite);
1026 
1027 	return perf_evlist__mmap_ops(&evlist->core, &ops, &mp.core);
1028 }
1029 
evlist__mmap(struct evlist * evlist,unsigned int pages)1030 int evlist__mmap(struct evlist *evlist, unsigned int pages)
1031 {
1032 	return evlist__mmap_ex(evlist, pages, 0, false, 0, PERF_AFFINITY_SYS, 1, 0);
1033 }
1034 
evlist__create_maps(struct evlist * evlist,struct target * target)1035 int evlist__create_maps(struct evlist *evlist, struct target *target)
1036 {
1037 	bool all_threads = (target->per_thread && target->system_wide);
1038 	struct perf_cpu_map *cpus;
1039 	struct perf_thread_map *threads;
1040 
1041 	/*
1042 	 * If specify '-a' and '--per-thread' to perf record, perf record
1043 	 * will override '--per-thread'. target->per_thread = false and
1044 	 * target->system_wide = true.
1045 	 *
1046 	 * If specify '--per-thread' only to perf record,
1047 	 * target->per_thread = true and target->system_wide = false.
1048 	 *
1049 	 * So target->per_thread && target->system_wide is false.
1050 	 * For perf record, thread_map__new_str doesn't call
1051 	 * thread_map__new_all_cpus. That will keep perf record's
1052 	 * current behavior.
1053 	 *
1054 	 * For perf stat, it allows the case that target->per_thread and
1055 	 * target->system_wide are all true. It means to collect system-wide
1056 	 * per-thread data. thread_map__new_str will call
1057 	 * thread_map__new_all_cpus to enumerate all threads.
1058 	 */
1059 	threads = thread_map__new_str(target->pid, target->tid, target->uid,
1060 				      all_threads);
1061 
1062 	if (!threads)
1063 		return -1;
1064 
1065 	if (target__uses_dummy_map(target))
1066 		cpus = perf_cpu_map__dummy_new();
1067 	else
1068 		cpus = perf_cpu_map__new(target->cpu_list);
1069 
1070 	if (!cpus)
1071 		goto out_delete_threads;
1072 
1073 	evlist->core.has_user_cpus = !!target->cpu_list;
1074 
1075 	perf_evlist__set_maps(&evlist->core, cpus, threads);
1076 
1077 	/* as evlist now has references, put count here */
1078 	perf_cpu_map__put(cpus);
1079 	perf_thread_map__put(threads);
1080 
1081 	return 0;
1082 
1083 out_delete_threads:
1084 	perf_thread_map__put(threads);
1085 	return -1;
1086 }
1087 
evlist__apply_filters(struct evlist * evlist,struct evsel ** err_evsel)1088 int evlist__apply_filters(struct evlist *evlist, struct evsel **err_evsel)
1089 {
1090 	struct evsel *evsel;
1091 	int err = 0;
1092 
1093 	evlist__for_each_entry(evlist, evsel) {
1094 		/*
1095 		 * filters only work for tracepoint event, which doesn't have cpu limit.
1096 		 * So evlist and evsel should always be same.
1097 		 */
1098 		if (evsel->filter) {
1099 			err = perf_evsel__apply_filter(&evsel->core, evsel->filter);
1100 			if (err) {
1101 				*err_evsel = evsel;
1102 				break;
1103 			}
1104 		}
1105 
1106 		/*
1107 		 * non-tracepoint events can have BPF filters.
1108 		 */
1109 		if (!list_empty(&evsel->bpf_filters)) {
1110 			err = perf_bpf_filter__prepare(evsel);
1111 			if (err) {
1112 				*err_evsel = evsel;
1113 				break;
1114 			}
1115 		}
1116 	}
1117 
1118 	return err;
1119 }
1120 
evlist__set_tp_filter(struct evlist * evlist,const char * filter)1121 int evlist__set_tp_filter(struct evlist *evlist, const char *filter)
1122 {
1123 	struct evsel *evsel;
1124 	int err = 0;
1125 
1126 	if (filter == NULL)
1127 		return -1;
1128 
1129 	evlist__for_each_entry(evlist, evsel) {
1130 		if (evsel->core.attr.type != PERF_TYPE_TRACEPOINT)
1131 			continue;
1132 
1133 		err = evsel__set_filter(evsel, filter);
1134 		if (err)
1135 			break;
1136 	}
1137 
1138 	return err;
1139 }
1140 
evlist__append_tp_filter(struct evlist * evlist,const char * filter)1141 int evlist__append_tp_filter(struct evlist *evlist, const char *filter)
1142 {
1143 	struct evsel *evsel;
1144 	int err = 0;
1145 
1146 	if (filter == NULL)
1147 		return -1;
1148 
1149 	evlist__for_each_entry(evlist, evsel) {
1150 		if (evsel->core.attr.type != PERF_TYPE_TRACEPOINT)
1151 			continue;
1152 
1153 		err = evsel__append_tp_filter(evsel, filter);
1154 		if (err)
1155 			break;
1156 	}
1157 
1158 	return err;
1159 }
1160 
asprintf__tp_filter_pids(size_t npids,pid_t * pids)1161 char *asprintf__tp_filter_pids(size_t npids, pid_t *pids)
1162 {
1163 	char *filter;
1164 	size_t i;
1165 
1166 	for (i = 0; i < npids; ++i) {
1167 		if (i == 0) {
1168 			if (asprintf(&filter, "common_pid != %d", pids[i]) < 0)
1169 				return NULL;
1170 		} else {
1171 			char *tmp;
1172 
1173 			if (asprintf(&tmp, "%s && common_pid != %d", filter, pids[i]) < 0)
1174 				goto out_free;
1175 
1176 			free(filter);
1177 			filter = tmp;
1178 		}
1179 	}
1180 
1181 	return filter;
1182 out_free:
1183 	free(filter);
1184 	return NULL;
1185 }
1186 
evlist__set_tp_filter_pids(struct evlist * evlist,size_t npids,pid_t * pids)1187 int evlist__set_tp_filter_pids(struct evlist *evlist, size_t npids, pid_t *pids)
1188 {
1189 	char *filter = asprintf__tp_filter_pids(npids, pids);
1190 	int ret = evlist__set_tp_filter(evlist, filter);
1191 
1192 	free(filter);
1193 	return ret;
1194 }
1195 
evlist__set_tp_filter_pid(struct evlist * evlist,pid_t pid)1196 int evlist__set_tp_filter_pid(struct evlist *evlist, pid_t pid)
1197 {
1198 	return evlist__set_tp_filter_pids(evlist, 1, &pid);
1199 }
1200 
evlist__append_tp_filter_pids(struct evlist * evlist,size_t npids,pid_t * pids)1201 int evlist__append_tp_filter_pids(struct evlist *evlist, size_t npids, pid_t *pids)
1202 {
1203 	char *filter = asprintf__tp_filter_pids(npids, pids);
1204 	int ret = evlist__append_tp_filter(evlist, filter);
1205 
1206 	free(filter);
1207 	return ret;
1208 }
1209 
evlist__append_tp_filter_pid(struct evlist * evlist,pid_t pid)1210 int evlist__append_tp_filter_pid(struct evlist *evlist, pid_t pid)
1211 {
1212 	return evlist__append_tp_filter_pids(evlist, 1, &pid);
1213 }
1214 
evlist__valid_sample_type(struct evlist * evlist)1215 bool evlist__valid_sample_type(struct evlist *evlist)
1216 {
1217 	struct evsel *pos;
1218 
1219 	if (evlist->core.nr_entries == 1)
1220 		return true;
1221 
1222 	if (evlist->id_pos < 0 || evlist->is_pos < 0)
1223 		return false;
1224 
1225 	evlist__for_each_entry(evlist, pos) {
1226 		if (pos->id_pos != evlist->id_pos ||
1227 		    pos->is_pos != evlist->is_pos)
1228 			return false;
1229 	}
1230 
1231 	return true;
1232 }
1233 
__evlist__combined_sample_type(struct evlist * evlist)1234 u64 __evlist__combined_sample_type(struct evlist *evlist)
1235 {
1236 	struct evsel *evsel;
1237 
1238 	if (evlist->combined_sample_type)
1239 		return evlist->combined_sample_type;
1240 
1241 	evlist__for_each_entry(evlist, evsel)
1242 		evlist->combined_sample_type |= evsel->core.attr.sample_type;
1243 
1244 	return evlist->combined_sample_type;
1245 }
1246 
evlist__combined_sample_type(struct evlist * evlist)1247 u64 evlist__combined_sample_type(struct evlist *evlist)
1248 {
1249 	evlist->combined_sample_type = 0;
1250 	return __evlist__combined_sample_type(evlist);
1251 }
1252 
evlist__combined_branch_type(struct evlist * evlist)1253 u64 evlist__combined_branch_type(struct evlist *evlist)
1254 {
1255 	struct evsel *evsel;
1256 	u64 branch_type = 0;
1257 
1258 	evlist__for_each_entry(evlist, evsel)
1259 		branch_type |= evsel->core.attr.branch_sample_type;
1260 	return branch_type;
1261 }
1262 
evlist__valid_read_format(struct evlist * evlist)1263 bool evlist__valid_read_format(struct evlist *evlist)
1264 {
1265 	struct evsel *first = evlist__first(evlist), *pos = first;
1266 	u64 read_format = first->core.attr.read_format;
1267 	u64 sample_type = first->core.attr.sample_type;
1268 
1269 	evlist__for_each_entry(evlist, pos) {
1270 		if (read_format != pos->core.attr.read_format) {
1271 			pr_debug("Read format differs %#" PRIx64 " vs %#" PRIx64 "\n",
1272 				 read_format, (u64)pos->core.attr.read_format);
1273 		}
1274 	}
1275 
1276 	/* PERF_SAMPLE_READ implies PERF_FORMAT_ID. */
1277 	if ((sample_type & PERF_SAMPLE_READ) &&
1278 	    !(read_format & PERF_FORMAT_ID)) {
1279 		return false;
1280 	}
1281 
1282 	return true;
1283 }
1284 
evlist__id_hdr_size(struct evlist * evlist)1285 u16 evlist__id_hdr_size(struct evlist *evlist)
1286 {
1287 	struct evsel *first = evlist__first(evlist);
1288 
1289 	return first->core.attr.sample_id_all ? evsel__id_hdr_size(first) : 0;
1290 }
1291 
evlist__valid_sample_id_all(struct evlist * evlist)1292 bool evlist__valid_sample_id_all(struct evlist *evlist)
1293 {
1294 	struct evsel *first = evlist__first(evlist), *pos = first;
1295 
1296 	evlist__for_each_entry_continue(evlist, pos) {
1297 		if (first->core.attr.sample_id_all != pos->core.attr.sample_id_all)
1298 			return false;
1299 	}
1300 
1301 	return true;
1302 }
1303 
evlist__sample_id_all(struct evlist * evlist)1304 bool evlist__sample_id_all(struct evlist *evlist)
1305 {
1306 	struct evsel *first = evlist__first(evlist);
1307 	return first->core.attr.sample_id_all;
1308 }
1309 
evlist__set_selected(struct evlist * evlist,struct evsel * evsel)1310 void evlist__set_selected(struct evlist *evlist, struct evsel *evsel)
1311 {
1312 	evlist->selected = evsel;
1313 }
1314 
evlist__close(struct evlist * evlist)1315 void evlist__close(struct evlist *evlist)
1316 {
1317 	struct evsel *evsel;
1318 	struct evlist_cpu_iterator evlist_cpu_itr;
1319 	struct affinity affinity;
1320 
1321 	/*
1322 	 * With perf record core.user_requested_cpus is usually NULL.
1323 	 * Use the old method to handle this for now.
1324 	 */
1325 	if (!evlist->core.user_requested_cpus ||
1326 	    cpu_map__is_dummy(evlist->core.user_requested_cpus)) {
1327 		evlist__for_each_entry_reverse(evlist, evsel)
1328 			evsel__close(evsel);
1329 		return;
1330 	}
1331 
1332 	if (affinity__setup(&affinity) < 0)
1333 		return;
1334 
1335 	evlist__for_each_cpu(evlist_cpu_itr, evlist, &affinity) {
1336 		perf_evsel__close_cpu(&evlist_cpu_itr.evsel->core,
1337 				      evlist_cpu_itr.cpu_map_idx);
1338 	}
1339 
1340 	affinity__cleanup(&affinity);
1341 	evlist__for_each_entry_reverse(evlist, evsel) {
1342 		perf_evsel__free_fd(&evsel->core);
1343 		perf_evsel__free_id(&evsel->core);
1344 	}
1345 	perf_evlist__reset_id_hash(&evlist->core);
1346 }
1347 
evlist__create_syswide_maps(struct evlist * evlist)1348 static int evlist__create_syswide_maps(struct evlist *evlist)
1349 {
1350 	struct perf_cpu_map *cpus;
1351 	struct perf_thread_map *threads;
1352 
1353 	/*
1354 	 * Try reading /sys/devices/system/cpu/online to get
1355 	 * an all cpus map.
1356 	 *
1357 	 * FIXME: -ENOMEM is the best we can do here, the cpu_map
1358 	 * code needs an overhaul to properly forward the
1359 	 * error, and we may not want to do that fallback to a
1360 	 * default cpu identity map :-\
1361 	 */
1362 	cpus = perf_cpu_map__new(NULL);
1363 	if (!cpus)
1364 		goto out;
1365 
1366 	threads = perf_thread_map__new_dummy();
1367 	if (!threads)
1368 		goto out_put;
1369 
1370 	perf_evlist__set_maps(&evlist->core, cpus, threads);
1371 
1372 	perf_thread_map__put(threads);
1373 out_put:
1374 	perf_cpu_map__put(cpus);
1375 out:
1376 	return -ENOMEM;
1377 }
1378 
evlist__open(struct evlist * evlist)1379 int evlist__open(struct evlist *evlist)
1380 {
1381 	struct evsel *evsel;
1382 	int err;
1383 
1384 	/*
1385 	 * Default: one fd per CPU, all threads, aka systemwide
1386 	 * as sys_perf_event_open(cpu = -1, thread = -1) is EINVAL
1387 	 */
1388 	if (evlist->core.threads == NULL && evlist->core.user_requested_cpus == NULL) {
1389 		err = evlist__create_syswide_maps(evlist);
1390 		if (err < 0)
1391 			goto out_err;
1392 	}
1393 
1394 	evlist__update_id_pos(evlist);
1395 
1396 	evlist__for_each_entry(evlist, evsel) {
1397 		err = evsel__open(evsel, evsel->core.cpus, evsel->core.threads);
1398 		if (err < 0)
1399 			goto out_err;
1400 	}
1401 
1402 	return 0;
1403 out_err:
1404 	evlist__close(evlist);
1405 	errno = -err;
1406 	return err;
1407 }
1408 
evlist__prepare_workload(struct evlist * evlist,struct target * target,const char * argv[],bool pipe_output,void (* exec_error)(int signo,siginfo_t * info,void * ucontext))1409 int evlist__prepare_workload(struct evlist *evlist, struct target *target, const char *argv[],
1410 			     bool pipe_output, void (*exec_error)(int signo, siginfo_t *info, void *ucontext))
1411 {
1412 	int child_ready_pipe[2], go_pipe[2];
1413 	char bf;
1414 
1415 	if (pipe(child_ready_pipe) < 0) {
1416 		perror("failed to create 'ready' pipe");
1417 		return -1;
1418 	}
1419 
1420 	if (pipe(go_pipe) < 0) {
1421 		perror("failed to create 'go' pipe");
1422 		goto out_close_ready_pipe;
1423 	}
1424 
1425 	evlist->workload.pid = fork();
1426 	if (evlist->workload.pid < 0) {
1427 		perror("failed to fork");
1428 		goto out_close_pipes;
1429 	}
1430 
1431 	if (!evlist->workload.pid) {
1432 		int ret;
1433 
1434 		if (pipe_output)
1435 			dup2(2, 1);
1436 
1437 		signal(SIGTERM, SIG_DFL);
1438 
1439 		close(child_ready_pipe[0]);
1440 		close(go_pipe[1]);
1441 		fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC);
1442 
1443 		/*
1444 		 * Change the name of this process not to confuse --exclude-perf users
1445 		 * that sees 'perf' in the window up to the execvp() and thinks that
1446 		 * perf samples are not being excluded.
1447 		 */
1448 		prctl(PR_SET_NAME, "perf-exec");
1449 
1450 		/*
1451 		 * Tell the parent we're ready to go
1452 		 */
1453 		close(child_ready_pipe[1]);
1454 
1455 		/*
1456 		 * Wait until the parent tells us to go.
1457 		 */
1458 		ret = read(go_pipe[0], &bf, 1);
1459 		/*
1460 		 * The parent will ask for the execvp() to be performed by
1461 		 * writing exactly one byte, in workload.cork_fd, usually via
1462 		 * evlist__start_workload().
1463 		 *
1464 		 * For cancelling the workload without actually running it,
1465 		 * the parent will just close workload.cork_fd, without writing
1466 		 * anything, i.e. read will return zero and we just exit()
1467 		 * here.
1468 		 */
1469 		if (ret != 1) {
1470 			if (ret == -1)
1471 				perror("unable to read pipe");
1472 			exit(ret);
1473 		}
1474 
1475 		execvp(argv[0], (char **)argv);
1476 
1477 		if (exec_error) {
1478 			union sigval val;
1479 
1480 			val.sival_int = errno;
1481 			if (sigqueue(getppid(), SIGUSR1, val))
1482 				perror(argv[0]);
1483 		} else
1484 			perror(argv[0]);
1485 		exit(-1);
1486 	}
1487 
1488 	if (exec_error) {
1489 		struct sigaction act = {
1490 			.sa_flags     = SA_SIGINFO,
1491 			.sa_sigaction = exec_error,
1492 		};
1493 		sigaction(SIGUSR1, &act, NULL);
1494 	}
1495 
1496 	if (target__none(target)) {
1497 		if (evlist->core.threads == NULL) {
1498 			fprintf(stderr, "FATAL: evlist->threads need to be set at this point (%s:%d).\n",
1499 				__func__, __LINE__);
1500 			goto out_close_pipes;
1501 		}
1502 		perf_thread_map__set_pid(evlist->core.threads, 0, evlist->workload.pid);
1503 	}
1504 
1505 	close(child_ready_pipe[1]);
1506 	close(go_pipe[0]);
1507 	/*
1508 	 * wait for child to settle
1509 	 */
1510 	if (read(child_ready_pipe[0], &bf, 1) == -1) {
1511 		perror("unable to read pipe");
1512 		goto out_close_pipes;
1513 	}
1514 
1515 	fcntl(go_pipe[1], F_SETFD, FD_CLOEXEC);
1516 	evlist->workload.cork_fd = go_pipe[1];
1517 	close(child_ready_pipe[0]);
1518 	return 0;
1519 
1520 out_close_pipes:
1521 	close(go_pipe[0]);
1522 	close(go_pipe[1]);
1523 out_close_ready_pipe:
1524 	close(child_ready_pipe[0]);
1525 	close(child_ready_pipe[1]);
1526 	return -1;
1527 }
1528 
evlist__start_workload(struct evlist * evlist)1529 int evlist__start_workload(struct evlist *evlist)
1530 {
1531 	if (evlist->workload.cork_fd > 0) {
1532 		char bf = 0;
1533 		int ret;
1534 		/*
1535 		 * Remove the cork, let it rip!
1536 		 */
1537 		ret = write(evlist->workload.cork_fd, &bf, 1);
1538 		if (ret < 0)
1539 			perror("unable to write to pipe");
1540 
1541 		close(evlist->workload.cork_fd);
1542 		return ret;
1543 	}
1544 
1545 	return 0;
1546 }
1547 
evlist__parse_sample(struct evlist * evlist,union perf_event * event,struct perf_sample * sample)1548 int evlist__parse_sample(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1549 {
1550 	struct evsel *evsel = evlist__event2evsel(evlist, event);
1551 	int ret;
1552 
1553 	if (!evsel)
1554 		return -EFAULT;
1555 	ret = evsel__parse_sample(evsel, event, sample);
1556 	if (ret)
1557 		return ret;
1558 	if (perf_guest && sample->id) {
1559 		struct perf_sample_id *sid = evlist__id2sid(evlist, sample->id);
1560 
1561 		if (sid) {
1562 			sample->machine_pid = sid->machine_pid;
1563 			sample->vcpu = sid->vcpu.cpu;
1564 		}
1565 	}
1566 	return 0;
1567 }
1568 
evlist__parse_sample_timestamp(struct evlist * evlist,union perf_event * event,u64 * timestamp)1569 int evlist__parse_sample_timestamp(struct evlist *evlist, union perf_event *event, u64 *timestamp)
1570 {
1571 	struct evsel *evsel = evlist__event2evsel(evlist, event);
1572 
1573 	if (!evsel)
1574 		return -EFAULT;
1575 	return evsel__parse_sample_timestamp(evsel, event, timestamp);
1576 }
1577 
evlist__strerror_open(struct evlist * evlist,int err,char * buf,size_t size)1578 int evlist__strerror_open(struct evlist *evlist, int err, char *buf, size_t size)
1579 {
1580 	int printed, value;
1581 	char sbuf[STRERR_BUFSIZE], *emsg = str_error_r(err, sbuf, sizeof(sbuf));
1582 
1583 	switch (err) {
1584 	case EACCES:
1585 	case EPERM:
1586 		printed = scnprintf(buf, size,
1587 				    "Error:\t%s.\n"
1588 				    "Hint:\tCheck /proc/sys/kernel/perf_event_paranoid setting.", emsg);
1589 
1590 		value = perf_event_paranoid();
1591 
1592 		printed += scnprintf(buf + printed, size - printed, "\nHint:\t");
1593 
1594 		if (value >= 2) {
1595 			printed += scnprintf(buf + printed, size - printed,
1596 					     "For your workloads it needs to be <= 1\nHint:\t");
1597 		}
1598 		printed += scnprintf(buf + printed, size - printed,
1599 				     "For system wide tracing it needs to be set to -1.\n");
1600 
1601 		printed += scnprintf(buf + printed, size - printed,
1602 				    "Hint:\tTry: 'sudo sh -c \"echo -1 > /proc/sys/kernel/perf_event_paranoid\"'\n"
1603 				    "Hint:\tThe current value is %d.", value);
1604 		break;
1605 	case EINVAL: {
1606 		struct evsel *first = evlist__first(evlist);
1607 		int max_freq;
1608 
1609 		if (sysctl__read_int("kernel/perf_event_max_sample_rate", &max_freq) < 0)
1610 			goto out_default;
1611 
1612 		if (first->core.attr.sample_freq < (u64)max_freq)
1613 			goto out_default;
1614 
1615 		printed = scnprintf(buf, size,
1616 				    "Error:\t%s.\n"
1617 				    "Hint:\tCheck /proc/sys/kernel/perf_event_max_sample_rate.\n"
1618 				    "Hint:\tThe current value is %d and %" PRIu64 " is being requested.",
1619 				    emsg, max_freq, first->core.attr.sample_freq);
1620 		break;
1621 	}
1622 	default:
1623 out_default:
1624 		scnprintf(buf, size, "%s", emsg);
1625 		break;
1626 	}
1627 
1628 	return 0;
1629 }
1630 
evlist__strerror_mmap(struct evlist * evlist,int err,char * buf,size_t size)1631 int evlist__strerror_mmap(struct evlist *evlist, int err, char *buf, size_t size)
1632 {
1633 	char sbuf[STRERR_BUFSIZE], *emsg = str_error_r(err, sbuf, sizeof(sbuf));
1634 	int pages_attempted = evlist->core.mmap_len / 1024, pages_max_per_user, printed = 0;
1635 
1636 	switch (err) {
1637 	case EPERM:
1638 		sysctl__read_int("kernel/perf_event_mlock_kb", &pages_max_per_user);
1639 		printed += scnprintf(buf + printed, size - printed,
1640 				     "Error:\t%s.\n"
1641 				     "Hint:\tCheck /proc/sys/kernel/perf_event_mlock_kb (%d kB) setting.\n"
1642 				     "Hint:\tTried using %zd kB.\n",
1643 				     emsg, pages_max_per_user, pages_attempted);
1644 
1645 		if (pages_attempted >= pages_max_per_user) {
1646 			printed += scnprintf(buf + printed, size - printed,
1647 					     "Hint:\tTry 'sudo sh -c \"echo %d > /proc/sys/kernel/perf_event_mlock_kb\"', or\n",
1648 					     pages_max_per_user + pages_attempted);
1649 		}
1650 
1651 		printed += scnprintf(buf + printed, size - printed,
1652 				     "Hint:\tTry using a smaller -m/--mmap-pages value.");
1653 		break;
1654 	default:
1655 		scnprintf(buf, size, "%s", emsg);
1656 		break;
1657 	}
1658 
1659 	return 0;
1660 }
1661 
evlist__to_front(struct evlist * evlist,struct evsel * move_evsel)1662 void evlist__to_front(struct evlist *evlist, struct evsel *move_evsel)
1663 {
1664 	struct evsel *evsel, *n;
1665 	LIST_HEAD(move);
1666 
1667 	if (move_evsel == evlist__first(evlist))
1668 		return;
1669 
1670 	evlist__for_each_entry_safe(evlist, n, evsel) {
1671 		if (evsel__leader(evsel) == evsel__leader(move_evsel))
1672 			list_move_tail(&evsel->core.node, &move);
1673 	}
1674 
1675 	list_splice(&move, &evlist->core.entries);
1676 }
1677 
evlist__get_tracking_event(struct evlist * evlist)1678 struct evsel *evlist__get_tracking_event(struct evlist *evlist)
1679 {
1680 	struct evsel *evsel;
1681 
1682 	evlist__for_each_entry(evlist, evsel) {
1683 		if (evsel->tracking)
1684 			return evsel;
1685 	}
1686 
1687 	return evlist__first(evlist);
1688 }
1689 
evlist__set_tracking_event(struct evlist * evlist,struct evsel * tracking_evsel)1690 void evlist__set_tracking_event(struct evlist *evlist, struct evsel *tracking_evsel)
1691 {
1692 	struct evsel *evsel;
1693 
1694 	if (tracking_evsel->tracking)
1695 		return;
1696 
1697 	evlist__for_each_entry(evlist, evsel) {
1698 		if (evsel != tracking_evsel)
1699 			evsel->tracking = false;
1700 	}
1701 
1702 	tracking_evsel->tracking = true;
1703 }
1704 
evlist__find_evsel_by_str(struct evlist * evlist,const char * str)1705 struct evsel *evlist__find_evsel_by_str(struct evlist *evlist, const char *str)
1706 {
1707 	struct evsel *evsel;
1708 
1709 	evlist__for_each_entry(evlist, evsel) {
1710 		if (!evsel->name)
1711 			continue;
1712 		if (evsel__name_is(evsel, str))
1713 			return evsel;
1714 	}
1715 
1716 	return NULL;
1717 }
1718 
evlist__toggle_bkw_mmap(struct evlist * evlist,enum bkw_mmap_state state)1719 void evlist__toggle_bkw_mmap(struct evlist *evlist, enum bkw_mmap_state state)
1720 {
1721 	enum bkw_mmap_state old_state = evlist->bkw_mmap_state;
1722 	enum action {
1723 		NONE,
1724 		PAUSE,
1725 		RESUME,
1726 	} action = NONE;
1727 
1728 	if (!evlist->overwrite_mmap)
1729 		return;
1730 
1731 	switch (old_state) {
1732 	case BKW_MMAP_NOTREADY: {
1733 		if (state != BKW_MMAP_RUNNING)
1734 			goto state_err;
1735 		break;
1736 	}
1737 	case BKW_MMAP_RUNNING: {
1738 		if (state != BKW_MMAP_DATA_PENDING)
1739 			goto state_err;
1740 		action = PAUSE;
1741 		break;
1742 	}
1743 	case BKW_MMAP_DATA_PENDING: {
1744 		if (state != BKW_MMAP_EMPTY)
1745 			goto state_err;
1746 		break;
1747 	}
1748 	case BKW_MMAP_EMPTY: {
1749 		if (state != BKW_MMAP_RUNNING)
1750 			goto state_err;
1751 		action = RESUME;
1752 		break;
1753 	}
1754 	default:
1755 		WARN_ONCE(1, "Shouldn't get there\n");
1756 	}
1757 
1758 	evlist->bkw_mmap_state = state;
1759 
1760 	switch (action) {
1761 	case PAUSE:
1762 		evlist__pause(evlist);
1763 		break;
1764 	case RESUME:
1765 		evlist__resume(evlist);
1766 		break;
1767 	case NONE:
1768 	default:
1769 		break;
1770 	}
1771 
1772 state_err:
1773 	return;
1774 }
1775 
evlist__exclude_kernel(struct evlist * evlist)1776 bool evlist__exclude_kernel(struct evlist *evlist)
1777 {
1778 	struct evsel *evsel;
1779 
1780 	evlist__for_each_entry(evlist, evsel) {
1781 		if (!evsel->core.attr.exclude_kernel)
1782 			return false;
1783 	}
1784 
1785 	return true;
1786 }
1787 
1788 /*
1789  * Events in data file are not collect in groups, but we still want
1790  * the group display. Set the artificial group and set the leader's
1791  * forced_leader flag to notify the display code.
1792  */
evlist__force_leader(struct evlist * evlist)1793 void evlist__force_leader(struct evlist *evlist)
1794 {
1795 	if (evlist__nr_groups(evlist) == 0) {
1796 		struct evsel *leader = evlist__first(evlist);
1797 
1798 		evlist__set_leader(evlist);
1799 		leader->forced_leader = true;
1800 	}
1801 }
1802 
evlist__reset_weak_group(struct evlist * evsel_list,struct evsel * evsel,bool close)1803 struct evsel *evlist__reset_weak_group(struct evlist *evsel_list, struct evsel *evsel, bool close)
1804 {
1805 	struct evsel *c2, *leader;
1806 	bool is_open = true;
1807 
1808 	leader = evsel__leader(evsel);
1809 
1810 	pr_debug("Weak group for %s/%d failed\n",
1811 			leader->name, leader->core.nr_members);
1812 
1813 	/*
1814 	 * for_each_group_member doesn't work here because it doesn't
1815 	 * include the first entry.
1816 	 */
1817 	evlist__for_each_entry(evsel_list, c2) {
1818 		if (c2 == evsel)
1819 			is_open = false;
1820 		if (evsel__has_leader(c2, leader)) {
1821 			if (is_open && close)
1822 				perf_evsel__close(&c2->core);
1823 			/*
1824 			 * We want to close all members of the group and reopen
1825 			 * them. Some events, like Intel topdown, require being
1826 			 * in a group and so keep these in the group.
1827 			 */
1828 			evsel__remove_from_group(c2, leader);
1829 
1830 			/*
1831 			 * Set this for all former members of the group
1832 			 * to indicate they get reopened.
1833 			 */
1834 			c2->reset_group = true;
1835 		}
1836 	}
1837 	/* Reset the leader count if all entries were removed. */
1838 	if (leader->core.nr_members == 1)
1839 		leader->core.nr_members = 0;
1840 	return leader;
1841 }
1842 
evlist__parse_control_fifo(const char * str,int * ctl_fd,int * ctl_fd_ack,bool * ctl_fd_close)1843 static int evlist__parse_control_fifo(const char *str, int *ctl_fd, int *ctl_fd_ack, bool *ctl_fd_close)
1844 {
1845 	char *s, *p;
1846 	int ret = 0, fd;
1847 
1848 	if (strncmp(str, "fifo:", 5))
1849 		return -EINVAL;
1850 
1851 	str += 5;
1852 	if (!*str || *str == ',')
1853 		return -EINVAL;
1854 
1855 	s = strdup(str);
1856 	if (!s)
1857 		return -ENOMEM;
1858 
1859 	p = strchr(s, ',');
1860 	if (p)
1861 		*p = '\0';
1862 
1863 	/*
1864 	 * O_RDWR avoids POLLHUPs which is necessary to allow the other
1865 	 * end of a FIFO to be repeatedly opened and closed.
1866 	 */
1867 	fd = open(s, O_RDWR | O_NONBLOCK | O_CLOEXEC);
1868 	if (fd < 0) {
1869 		pr_err("Failed to open '%s'\n", s);
1870 		ret = -errno;
1871 		goto out_free;
1872 	}
1873 	*ctl_fd = fd;
1874 	*ctl_fd_close = true;
1875 
1876 	if (p && *++p) {
1877 		/* O_RDWR | O_NONBLOCK means the other end need not be open */
1878 		fd = open(p, O_RDWR | O_NONBLOCK | O_CLOEXEC);
1879 		if (fd < 0) {
1880 			pr_err("Failed to open '%s'\n", p);
1881 			ret = -errno;
1882 			goto out_free;
1883 		}
1884 		*ctl_fd_ack = fd;
1885 	}
1886 
1887 out_free:
1888 	free(s);
1889 	return ret;
1890 }
1891 
evlist__parse_control(const char * str,int * ctl_fd,int * ctl_fd_ack,bool * ctl_fd_close)1892 int evlist__parse_control(const char *str, int *ctl_fd, int *ctl_fd_ack, bool *ctl_fd_close)
1893 {
1894 	char *comma = NULL, *endptr = NULL;
1895 
1896 	*ctl_fd_close = false;
1897 
1898 	if (strncmp(str, "fd:", 3))
1899 		return evlist__parse_control_fifo(str, ctl_fd, ctl_fd_ack, ctl_fd_close);
1900 
1901 	*ctl_fd = strtoul(&str[3], &endptr, 0);
1902 	if (endptr == &str[3])
1903 		return -EINVAL;
1904 
1905 	comma = strchr(str, ',');
1906 	if (comma) {
1907 		if (endptr != comma)
1908 			return -EINVAL;
1909 
1910 		*ctl_fd_ack = strtoul(comma + 1, &endptr, 0);
1911 		if (endptr == comma + 1 || *endptr != '\0')
1912 			return -EINVAL;
1913 	}
1914 
1915 	return 0;
1916 }
1917 
evlist__close_control(int ctl_fd,int ctl_fd_ack,bool * ctl_fd_close)1918 void evlist__close_control(int ctl_fd, int ctl_fd_ack, bool *ctl_fd_close)
1919 {
1920 	if (*ctl_fd_close) {
1921 		*ctl_fd_close = false;
1922 		close(ctl_fd);
1923 		if (ctl_fd_ack >= 0)
1924 			close(ctl_fd_ack);
1925 	}
1926 }
1927 
evlist__initialize_ctlfd(struct evlist * evlist,int fd,int ack)1928 int evlist__initialize_ctlfd(struct evlist *evlist, int fd, int ack)
1929 {
1930 	if (fd == -1) {
1931 		pr_debug("Control descriptor is not initialized\n");
1932 		return 0;
1933 	}
1934 
1935 	evlist->ctl_fd.pos = perf_evlist__add_pollfd(&evlist->core, fd, NULL, POLLIN,
1936 						     fdarray_flag__nonfilterable |
1937 						     fdarray_flag__non_perf_event);
1938 	if (evlist->ctl_fd.pos < 0) {
1939 		evlist->ctl_fd.pos = -1;
1940 		pr_err("Failed to add ctl fd entry: %m\n");
1941 		return -1;
1942 	}
1943 
1944 	evlist->ctl_fd.fd = fd;
1945 	evlist->ctl_fd.ack = ack;
1946 
1947 	return 0;
1948 }
1949 
evlist__ctlfd_initialized(struct evlist * evlist)1950 bool evlist__ctlfd_initialized(struct evlist *evlist)
1951 {
1952 	return evlist->ctl_fd.pos >= 0;
1953 }
1954 
evlist__finalize_ctlfd(struct evlist * evlist)1955 int evlist__finalize_ctlfd(struct evlist *evlist)
1956 {
1957 	struct pollfd *entries = evlist->core.pollfd.entries;
1958 
1959 	if (!evlist__ctlfd_initialized(evlist))
1960 		return 0;
1961 
1962 	entries[evlist->ctl_fd.pos].fd = -1;
1963 	entries[evlist->ctl_fd.pos].events = 0;
1964 	entries[evlist->ctl_fd.pos].revents = 0;
1965 
1966 	evlist->ctl_fd.pos = -1;
1967 	evlist->ctl_fd.ack = -1;
1968 	evlist->ctl_fd.fd = -1;
1969 
1970 	return 0;
1971 }
1972 
evlist__ctlfd_recv(struct evlist * evlist,enum evlist_ctl_cmd * cmd,char * cmd_data,size_t data_size)1973 static int evlist__ctlfd_recv(struct evlist *evlist, enum evlist_ctl_cmd *cmd,
1974 			      char *cmd_data, size_t data_size)
1975 {
1976 	int err;
1977 	char c;
1978 	size_t bytes_read = 0;
1979 
1980 	*cmd = EVLIST_CTL_CMD_UNSUPPORTED;
1981 	memset(cmd_data, 0, data_size);
1982 	data_size--;
1983 
1984 	do {
1985 		err = read(evlist->ctl_fd.fd, &c, 1);
1986 		if (err > 0) {
1987 			if (c == '\n' || c == '\0')
1988 				break;
1989 			cmd_data[bytes_read++] = c;
1990 			if (bytes_read == data_size)
1991 				break;
1992 			continue;
1993 		} else if (err == -1) {
1994 			if (errno == EINTR)
1995 				continue;
1996 			if (errno == EAGAIN || errno == EWOULDBLOCK)
1997 				err = 0;
1998 			else
1999 				pr_err("Failed to read from ctlfd %d: %m\n", evlist->ctl_fd.fd);
2000 		}
2001 		break;
2002 	} while (1);
2003 
2004 	pr_debug("Message from ctl_fd: \"%s%s\"\n", cmd_data,
2005 		 bytes_read == data_size ? "" : c == '\n' ? "\\n" : "\\0");
2006 
2007 	if (bytes_read > 0) {
2008 		if (!strncmp(cmd_data, EVLIST_CTL_CMD_ENABLE_TAG,
2009 			     (sizeof(EVLIST_CTL_CMD_ENABLE_TAG)-1))) {
2010 			*cmd = EVLIST_CTL_CMD_ENABLE;
2011 		} else if (!strncmp(cmd_data, EVLIST_CTL_CMD_DISABLE_TAG,
2012 				    (sizeof(EVLIST_CTL_CMD_DISABLE_TAG)-1))) {
2013 			*cmd = EVLIST_CTL_CMD_DISABLE;
2014 		} else if (!strncmp(cmd_data, EVLIST_CTL_CMD_SNAPSHOT_TAG,
2015 				    (sizeof(EVLIST_CTL_CMD_SNAPSHOT_TAG)-1))) {
2016 			*cmd = EVLIST_CTL_CMD_SNAPSHOT;
2017 			pr_debug("is snapshot\n");
2018 		} else if (!strncmp(cmd_data, EVLIST_CTL_CMD_EVLIST_TAG,
2019 				    (sizeof(EVLIST_CTL_CMD_EVLIST_TAG)-1))) {
2020 			*cmd = EVLIST_CTL_CMD_EVLIST;
2021 		} else if (!strncmp(cmd_data, EVLIST_CTL_CMD_STOP_TAG,
2022 				    (sizeof(EVLIST_CTL_CMD_STOP_TAG)-1))) {
2023 			*cmd = EVLIST_CTL_CMD_STOP;
2024 		} else if (!strncmp(cmd_data, EVLIST_CTL_CMD_PING_TAG,
2025 				    (sizeof(EVLIST_CTL_CMD_PING_TAG)-1))) {
2026 			*cmd = EVLIST_CTL_CMD_PING;
2027 		}
2028 	}
2029 
2030 	return bytes_read ? (int)bytes_read : err;
2031 }
2032 
evlist__ctlfd_ack(struct evlist * evlist)2033 int evlist__ctlfd_ack(struct evlist *evlist)
2034 {
2035 	int err;
2036 
2037 	if (evlist->ctl_fd.ack == -1)
2038 		return 0;
2039 
2040 	err = write(evlist->ctl_fd.ack, EVLIST_CTL_CMD_ACK_TAG,
2041 		    sizeof(EVLIST_CTL_CMD_ACK_TAG));
2042 	if (err == -1)
2043 		pr_err("failed to write to ctl_ack_fd %d: %m\n", evlist->ctl_fd.ack);
2044 
2045 	return err;
2046 }
2047 
get_cmd_arg(char * cmd_data,size_t cmd_size,char ** arg)2048 static int get_cmd_arg(char *cmd_data, size_t cmd_size, char **arg)
2049 {
2050 	char *data = cmd_data + cmd_size;
2051 
2052 	/* no argument */
2053 	if (!*data)
2054 		return 0;
2055 
2056 	/* there's argument */
2057 	if (*data == ' ') {
2058 		*arg = data + 1;
2059 		return 1;
2060 	}
2061 
2062 	/* malformed */
2063 	return -1;
2064 }
2065 
evlist__ctlfd_enable(struct evlist * evlist,char * cmd_data,bool enable)2066 static int evlist__ctlfd_enable(struct evlist *evlist, char *cmd_data, bool enable)
2067 {
2068 	struct evsel *evsel;
2069 	char *name;
2070 	int err;
2071 
2072 	err = get_cmd_arg(cmd_data,
2073 			  enable ? sizeof(EVLIST_CTL_CMD_ENABLE_TAG) - 1 :
2074 				   sizeof(EVLIST_CTL_CMD_DISABLE_TAG) - 1,
2075 			  &name);
2076 	if (err < 0) {
2077 		pr_info("failed: wrong command\n");
2078 		return -1;
2079 	}
2080 
2081 	if (err) {
2082 		evsel = evlist__find_evsel_by_str(evlist, name);
2083 		if (evsel) {
2084 			if (enable)
2085 				evlist__enable_evsel(evlist, name);
2086 			else
2087 				evlist__disable_evsel(evlist, name);
2088 			pr_info("Event %s %s\n", evsel->name,
2089 				enable ? "enabled" : "disabled");
2090 		} else {
2091 			pr_info("failed: can't find '%s' event\n", name);
2092 		}
2093 	} else {
2094 		if (enable) {
2095 			evlist__enable(evlist);
2096 			pr_info(EVLIST_ENABLED_MSG);
2097 		} else {
2098 			evlist__disable(evlist);
2099 			pr_info(EVLIST_DISABLED_MSG);
2100 		}
2101 	}
2102 
2103 	return 0;
2104 }
2105 
evlist__ctlfd_list(struct evlist * evlist,char * cmd_data)2106 static int evlist__ctlfd_list(struct evlist *evlist, char *cmd_data)
2107 {
2108 	struct perf_attr_details details = { .verbose = false, };
2109 	struct evsel *evsel;
2110 	char *arg;
2111 	int err;
2112 
2113 	err = get_cmd_arg(cmd_data,
2114 			  sizeof(EVLIST_CTL_CMD_EVLIST_TAG) - 1,
2115 			  &arg);
2116 	if (err < 0) {
2117 		pr_info("failed: wrong command\n");
2118 		return -1;
2119 	}
2120 
2121 	if (err) {
2122 		if (!strcmp(arg, "-v")) {
2123 			details.verbose = true;
2124 		} else if (!strcmp(arg, "-g")) {
2125 			details.event_group = true;
2126 		} else if (!strcmp(arg, "-F")) {
2127 			details.freq = true;
2128 		} else {
2129 			pr_info("failed: wrong command\n");
2130 			return -1;
2131 		}
2132 	}
2133 
2134 	evlist__for_each_entry(evlist, evsel)
2135 		evsel__fprintf(evsel, &details, stderr);
2136 
2137 	return 0;
2138 }
2139 
evlist__ctlfd_process(struct evlist * evlist,enum evlist_ctl_cmd * cmd)2140 int evlist__ctlfd_process(struct evlist *evlist, enum evlist_ctl_cmd *cmd)
2141 {
2142 	int err = 0;
2143 	char cmd_data[EVLIST_CTL_CMD_MAX_LEN];
2144 	int ctlfd_pos = evlist->ctl_fd.pos;
2145 	struct pollfd *entries = evlist->core.pollfd.entries;
2146 
2147 	if (!evlist__ctlfd_initialized(evlist) || !entries[ctlfd_pos].revents)
2148 		return 0;
2149 
2150 	if (entries[ctlfd_pos].revents & POLLIN) {
2151 		err = evlist__ctlfd_recv(evlist, cmd, cmd_data,
2152 					 EVLIST_CTL_CMD_MAX_LEN);
2153 		if (err > 0) {
2154 			switch (*cmd) {
2155 			case EVLIST_CTL_CMD_ENABLE:
2156 			case EVLIST_CTL_CMD_DISABLE:
2157 				err = evlist__ctlfd_enable(evlist, cmd_data,
2158 							   *cmd == EVLIST_CTL_CMD_ENABLE);
2159 				break;
2160 			case EVLIST_CTL_CMD_EVLIST:
2161 				err = evlist__ctlfd_list(evlist, cmd_data);
2162 				break;
2163 			case EVLIST_CTL_CMD_SNAPSHOT:
2164 			case EVLIST_CTL_CMD_STOP:
2165 			case EVLIST_CTL_CMD_PING:
2166 				break;
2167 			case EVLIST_CTL_CMD_ACK:
2168 			case EVLIST_CTL_CMD_UNSUPPORTED:
2169 			default:
2170 				pr_debug("ctlfd: unsupported %d\n", *cmd);
2171 				break;
2172 			}
2173 			if (!(*cmd == EVLIST_CTL_CMD_ACK || *cmd == EVLIST_CTL_CMD_UNSUPPORTED ||
2174 			      *cmd == EVLIST_CTL_CMD_SNAPSHOT))
2175 				evlist__ctlfd_ack(evlist);
2176 		}
2177 	}
2178 
2179 	if (entries[ctlfd_pos].revents & (POLLHUP | POLLERR))
2180 		evlist__finalize_ctlfd(evlist);
2181 	else
2182 		entries[ctlfd_pos].revents = 0;
2183 
2184 	return err;
2185 }
2186 
2187 /**
2188  * struct event_enable_time - perf record -D/--delay single time range.
2189  * @start: start of time range to enable events in milliseconds
2190  * @end: end of time range to enable events in milliseconds
2191  *
2192  * N.B. this structure is also accessed as an array of int.
2193  */
2194 struct event_enable_time {
2195 	int	start;
2196 	int	end;
2197 };
2198 
parse_event_enable_time(const char * str,struct event_enable_time * range,bool first)2199 static int parse_event_enable_time(const char *str, struct event_enable_time *range, bool first)
2200 {
2201 	const char *fmt = first ? "%u - %u %n" : " , %u - %u %n";
2202 	int ret, start, end, n;
2203 
2204 	ret = sscanf(str, fmt, &start, &end, &n);
2205 	if (ret != 2 || end <= start)
2206 		return -EINVAL;
2207 	if (range) {
2208 		range->start = start;
2209 		range->end = end;
2210 	}
2211 	return n;
2212 }
2213 
parse_event_enable_times(const char * str,struct event_enable_time * range)2214 static ssize_t parse_event_enable_times(const char *str, struct event_enable_time *range)
2215 {
2216 	int incr = !!range;
2217 	bool first = true;
2218 	ssize_t ret, cnt;
2219 
2220 	for (cnt = 0; *str; cnt++) {
2221 		ret = parse_event_enable_time(str, range, first);
2222 		if (ret < 0)
2223 			return ret;
2224 		/* Check no overlap */
2225 		if (!first && range && range->start <= range[-1].end)
2226 			return -EINVAL;
2227 		str += ret;
2228 		range += incr;
2229 		first = false;
2230 	}
2231 	return cnt;
2232 }
2233 
2234 /**
2235  * struct event_enable_timer - control structure for perf record -D/--delay.
2236  * @evlist: event list
2237  * @times: time ranges that events are enabled (N.B. this is also accessed as an
2238  *         array of int)
2239  * @times_cnt: number of time ranges
2240  * @timerfd: timer file descriptor
2241  * @pollfd_pos: position in @evlist array of file descriptors to poll (fdarray)
2242  * @times_step: current position in (int *)@times)[],
2243  *              refer event_enable_timer__process()
2244  *
2245  * Note, this structure is only used when there are time ranges, not when there
2246  * is only an initial delay.
2247  */
2248 struct event_enable_timer {
2249 	struct evlist *evlist;
2250 	struct event_enable_time *times;
2251 	size_t	times_cnt;
2252 	int	timerfd;
2253 	int	pollfd_pos;
2254 	size_t	times_step;
2255 };
2256 
str_to_delay(const char * str)2257 static int str_to_delay(const char *str)
2258 {
2259 	char *endptr;
2260 	long d;
2261 
2262 	d = strtol(str, &endptr, 10);
2263 	if (*endptr || d > INT_MAX || d < -1)
2264 		return 0;
2265 	return d;
2266 }
2267 
evlist__parse_event_enable_time(struct evlist * evlist,struct record_opts * opts,const char * str,int unset)2268 int evlist__parse_event_enable_time(struct evlist *evlist, struct record_opts *opts,
2269 				    const char *str, int unset)
2270 {
2271 	enum fdarray_flags flags = fdarray_flag__nonfilterable | fdarray_flag__non_perf_event;
2272 	struct event_enable_timer *eet;
2273 	ssize_t times_cnt;
2274 	ssize_t ret;
2275 	int err;
2276 
2277 	if (unset)
2278 		return 0;
2279 
2280 	opts->target.initial_delay = str_to_delay(str);
2281 	if (opts->target.initial_delay)
2282 		return 0;
2283 
2284 	ret = parse_event_enable_times(str, NULL);
2285 	if (ret < 0)
2286 		return ret;
2287 
2288 	times_cnt = ret;
2289 	if (times_cnt == 0)
2290 		return -EINVAL;
2291 
2292 	eet = zalloc(sizeof(*eet));
2293 	if (!eet)
2294 		return -ENOMEM;
2295 
2296 	eet->times = calloc(times_cnt, sizeof(*eet->times));
2297 	if (!eet->times) {
2298 		err = -ENOMEM;
2299 		goto free_eet;
2300 	}
2301 
2302 	if (parse_event_enable_times(str, eet->times) != times_cnt) {
2303 		err = -EINVAL;
2304 		goto free_eet_times;
2305 	}
2306 
2307 	eet->times_cnt = times_cnt;
2308 
2309 	eet->timerfd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
2310 	if (eet->timerfd == -1) {
2311 		err = -errno;
2312 		pr_err("timerfd_create failed: %s\n", strerror(errno));
2313 		goto free_eet_times;
2314 	}
2315 
2316 	eet->pollfd_pos = perf_evlist__add_pollfd(&evlist->core, eet->timerfd, NULL, POLLIN, flags);
2317 	if (eet->pollfd_pos < 0) {
2318 		err = eet->pollfd_pos;
2319 		goto close_timerfd;
2320 	}
2321 
2322 	eet->evlist = evlist;
2323 	evlist->eet = eet;
2324 	opts->target.initial_delay = eet->times[0].start;
2325 
2326 	return 0;
2327 
2328 close_timerfd:
2329 	close(eet->timerfd);
2330 free_eet_times:
2331 	zfree(&eet->times);
2332 free_eet:
2333 	free(eet);
2334 	return err;
2335 }
2336 
event_enable_timer__set_timer(struct event_enable_timer * eet,int ms)2337 static int event_enable_timer__set_timer(struct event_enable_timer *eet, int ms)
2338 {
2339 	struct itimerspec its = {
2340 		.it_value.tv_sec = ms / MSEC_PER_SEC,
2341 		.it_value.tv_nsec = (ms % MSEC_PER_SEC) * NSEC_PER_MSEC,
2342 	};
2343 	int err = 0;
2344 
2345 	if (timerfd_settime(eet->timerfd, 0, &its, NULL) < 0) {
2346 		err = -errno;
2347 		pr_err("timerfd_settime failed: %s\n", strerror(errno));
2348 	}
2349 	return err;
2350 }
2351 
event_enable_timer__start(struct event_enable_timer * eet)2352 int event_enable_timer__start(struct event_enable_timer *eet)
2353 {
2354 	int ms;
2355 
2356 	if (!eet)
2357 		return 0;
2358 
2359 	ms = eet->times[0].end - eet->times[0].start;
2360 	eet->times_step = 1;
2361 
2362 	return event_enable_timer__set_timer(eet, ms);
2363 }
2364 
event_enable_timer__process(struct event_enable_timer * eet)2365 int event_enable_timer__process(struct event_enable_timer *eet)
2366 {
2367 	struct pollfd *entries;
2368 	short revents;
2369 
2370 	if (!eet)
2371 		return 0;
2372 
2373 	entries = eet->evlist->core.pollfd.entries;
2374 	revents = entries[eet->pollfd_pos].revents;
2375 	entries[eet->pollfd_pos].revents = 0;
2376 
2377 	if (revents & POLLIN) {
2378 		size_t step = eet->times_step;
2379 		size_t pos = step / 2;
2380 
2381 		if (step & 1) {
2382 			evlist__disable_non_dummy(eet->evlist);
2383 			pr_info(EVLIST_DISABLED_MSG);
2384 			if (pos >= eet->times_cnt - 1) {
2385 				/* Disarm timer */
2386 				event_enable_timer__set_timer(eet, 0);
2387 				return 1; /* Stop */
2388 			}
2389 		} else {
2390 			evlist__enable_non_dummy(eet->evlist);
2391 			pr_info(EVLIST_ENABLED_MSG);
2392 		}
2393 
2394 		step += 1;
2395 		pos = step / 2;
2396 
2397 		if (pos < eet->times_cnt) {
2398 			int *times = (int *)eet->times; /* Accessing 'times' as array of int */
2399 			int ms = times[step] - times[step - 1];
2400 
2401 			eet->times_step = step;
2402 			return event_enable_timer__set_timer(eet, ms);
2403 		}
2404 	}
2405 
2406 	return 0;
2407 }
2408 
event_enable_timer__exit(struct event_enable_timer ** ep)2409 void event_enable_timer__exit(struct event_enable_timer **ep)
2410 {
2411 	if (!ep || !*ep)
2412 		return;
2413 	zfree(&(*ep)->times);
2414 	zfree(ep);
2415 }
2416 
evlist__find_evsel(struct evlist * evlist,int idx)2417 struct evsel *evlist__find_evsel(struct evlist *evlist, int idx)
2418 {
2419 	struct evsel *evsel;
2420 
2421 	evlist__for_each_entry(evlist, evsel) {
2422 		if (evsel->core.idx == idx)
2423 			return evsel;
2424 	}
2425 	return NULL;
2426 }
2427 
evlist__scnprintf_evsels(struct evlist * evlist,size_t size,char * bf)2428 int evlist__scnprintf_evsels(struct evlist *evlist, size_t size, char *bf)
2429 {
2430 	struct evsel *evsel;
2431 	int printed = 0;
2432 
2433 	evlist__for_each_entry(evlist, evsel) {
2434 		if (evsel__is_dummy_event(evsel))
2435 			continue;
2436 		if (size > (strlen(evsel__name(evsel)) + (printed ? 2 : 1))) {
2437 			printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "," : "", evsel__name(evsel));
2438 		} else {
2439 			printed += scnprintf(bf + printed, size - printed, "%s...", printed ? "," : "");
2440 			break;
2441 		}
2442 	}
2443 
2444 	return printed;
2445 }
2446 
evlist__check_mem_load_aux(struct evlist * evlist)2447 void evlist__check_mem_load_aux(struct evlist *evlist)
2448 {
2449 	struct evsel *leader, *evsel, *pos;
2450 
2451 	/*
2452 	 * For some platforms, the 'mem-loads' event is required to use
2453 	 * together with 'mem-loads-aux' within a group and 'mem-loads-aux'
2454 	 * must be the group leader. Now we disable this group before reporting
2455 	 * because 'mem-loads-aux' is just an auxiliary event. It doesn't carry
2456 	 * any valid memory load information.
2457 	 */
2458 	evlist__for_each_entry(evlist, evsel) {
2459 		leader = evsel__leader(evsel);
2460 		if (leader == evsel)
2461 			continue;
2462 
2463 		if (leader->name && strstr(leader->name, "mem-loads-aux")) {
2464 			for_each_group_evsel(pos, leader) {
2465 				evsel__set_leader(pos, pos);
2466 				pos->core.nr_members = 0;
2467 			}
2468 		}
2469 	}
2470 }
2471 
2472 /**
2473  * evlist__warn_user_requested_cpus() - Check each evsel against requested CPUs
2474  *     and warn if the user CPU list is inapplicable for the event's PMU's
2475  *     CPUs. Not core PMUs list a CPU in sysfs, but this may be overwritten by a
2476  *     user requested CPU and so any online CPU is applicable. Core PMUs handle
2477  *     events on the CPUs in their list and otherwise the event isn't supported.
2478  * @evlist: The list of events being checked.
2479  * @cpu_list: The user provided list of CPUs.
2480  */
evlist__warn_user_requested_cpus(struct evlist * evlist,const char * cpu_list)2481 void evlist__warn_user_requested_cpus(struct evlist *evlist, const char *cpu_list)
2482 {
2483 	struct perf_cpu_map *user_requested_cpus;
2484 	struct evsel *pos;
2485 
2486 	if (!cpu_list)
2487 		return;
2488 
2489 	user_requested_cpus = perf_cpu_map__new(cpu_list);
2490 	if (!user_requested_cpus)
2491 		return;
2492 
2493 	evlist__for_each_entry(evlist, pos) {
2494 		struct perf_cpu_map *intersect, *to_test;
2495 		const struct perf_pmu *pmu = evsel__find_pmu(pos);
2496 
2497 		to_test = pmu && pmu->is_core ? pmu->cpus : cpu_map__online();
2498 		intersect = perf_cpu_map__intersect(to_test, user_requested_cpus);
2499 		if (!perf_cpu_map__equal(intersect, user_requested_cpus)) {
2500 			char buf[128];
2501 
2502 			cpu_map__snprint(to_test, buf, sizeof(buf));
2503 			pr_warning("WARNING: A requested CPU in '%s' is not supported by PMU '%s' (CPUs %s) for event '%s'\n",
2504 				cpu_list, pmu ? pmu->name : "cpu", buf, evsel__name(pos));
2505 		}
2506 		perf_cpu_map__put(intersect);
2507 	}
2508 	perf_cpu_map__put(user_requested_cpus);
2509 }
2510