xref: /linux/tools/perf/builtin-kwork.c (revision 6e7fd890f1d6ac83805409e9c346240de2705584)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * builtin-kwork.c
4  *
5  * Copyright (c) 2022  Huawei Inc,  Yang Jihong <yangjihong1@huawei.com>
6  */
7 
8 #include "builtin.h"
9 
10 #include "util/data.h"
11 #include "util/evlist.h"
12 #include "util/evsel.h"
13 #include "util/header.h"
14 #include "util/kwork.h"
15 #include "util/debug.h"
16 #include "util/session.h"
17 #include "util/symbol.h"
18 #include "util/thread.h"
19 #include "util/string2.h"
20 #include "util/callchain.h"
21 #include "util/evsel_fprintf.h"
22 #include "util/util.h"
23 
24 #include <subcmd/pager.h>
25 #include <subcmd/parse-options.h>
26 #include <traceevent/event-parse.h>
27 
28 #include <errno.h>
29 #include <inttypes.h>
30 #include <signal.h>
31 #include <linux/err.h>
32 #include <linux/time64.h>
33 #include <linux/zalloc.h>
34 
35 /*
36  * report header elements width
37  */
38 #define PRINT_CPU_WIDTH 4
39 #define PRINT_COUNT_WIDTH 9
40 #define PRINT_RUNTIME_WIDTH 10
41 #define PRINT_LATENCY_WIDTH 10
42 #define PRINT_TIMESTAMP_WIDTH 17
43 #define PRINT_KWORK_NAME_WIDTH 30
44 #define RPINT_DECIMAL_WIDTH 3
45 #define PRINT_BRACKETPAIR_WIDTH 2
46 #define PRINT_TIME_UNIT_SEC_WIDTH 2
47 #define PRINT_TIME_UNIT_MESC_WIDTH 3
48 #define PRINT_PID_WIDTH 7
49 #define PRINT_TASK_NAME_WIDTH 16
50 #define PRINT_CPU_USAGE_WIDTH 6
51 #define PRINT_CPU_USAGE_DECIMAL_WIDTH 2
52 #define PRINT_CPU_USAGE_HIST_WIDTH 30
53 #define PRINT_RUNTIME_HEADER_WIDTH (PRINT_RUNTIME_WIDTH + PRINT_TIME_UNIT_MESC_WIDTH)
54 #define PRINT_LATENCY_HEADER_WIDTH (PRINT_LATENCY_WIDTH + PRINT_TIME_UNIT_MESC_WIDTH)
55 #define PRINT_TIMEHIST_CPU_WIDTH (PRINT_CPU_WIDTH + PRINT_BRACKETPAIR_WIDTH)
56 #define PRINT_TIMESTAMP_HEADER_WIDTH (PRINT_TIMESTAMP_WIDTH + PRINT_TIME_UNIT_SEC_WIDTH)
57 
58 struct sort_dimension {
59 	const char      *name;
60 	int             (*cmp)(struct kwork_work *l, struct kwork_work *r);
61 	struct          list_head list;
62 };
63 
64 static int id_cmp(struct kwork_work *l, struct kwork_work *r)
65 {
66 	if (l->cpu > r->cpu)
67 		return 1;
68 	if (l->cpu < r->cpu)
69 		return -1;
70 
71 	if (l->id > r->id)
72 		return 1;
73 	if (l->id < r->id)
74 		return -1;
75 
76 	return 0;
77 }
78 
79 static int count_cmp(struct kwork_work *l, struct kwork_work *r)
80 {
81 	if (l->nr_atoms > r->nr_atoms)
82 		return 1;
83 	if (l->nr_atoms < r->nr_atoms)
84 		return -1;
85 
86 	return 0;
87 }
88 
89 static int runtime_cmp(struct kwork_work *l, struct kwork_work *r)
90 {
91 	if (l->total_runtime > r->total_runtime)
92 		return 1;
93 	if (l->total_runtime < r->total_runtime)
94 		return -1;
95 
96 	return 0;
97 }
98 
99 static int max_runtime_cmp(struct kwork_work *l, struct kwork_work *r)
100 {
101 	if (l->max_runtime > r->max_runtime)
102 		return 1;
103 	if (l->max_runtime < r->max_runtime)
104 		return -1;
105 
106 	return 0;
107 }
108 
109 static int avg_latency_cmp(struct kwork_work *l, struct kwork_work *r)
110 {
111 	u64 avgl, avgr;
112 
113 	if (!r->nr_atoms)
114 		return 1;
115 	if (!l->nr_atoms)
116 		return -1;
117 
118 	avgl = l->total_latency / l->nr_atoms;
119 	avgr = r->total_latency / r->nr_atoms;
120 
121 	if (avgl > avgr)
122 		return 1;
123 	if (avgl < avgr)
124 		return -1;
125 
126 	return 0;
127 }
128 
129 static int max_latency_cmp(struct kwork_work *l, struct kwork_work *r)
130 {
131 	if (l->max_latency > r->max_latency)
132 		return 1;
133 	if (l->max_latency < r->max_latency)
134 		return -1;
135 
136 	return 0;
137 }
138 
139 static int cpu_usage_cmp(struct kwork_work *l, struct kwork_work *r)
140 {
141 	if (l->cpu_usage > r->cpu_usage)
142 		return 1;
143 	if (l->cpu_usage < r->cpu_usage)
144 		return -1;
145 
146 	return 0;
147 }
148 
149 static int id_or_cpu_r_cmp(struct kwork_work *l, struct kwork_work *r)
150 {
151 	if (l->id < r->id)
152 		return 1;
153 	if (l->id > r->id)
154 		return -1;
155 
156 	if (l->id != 0)
157 		return 0;
158 
159 	if (l->cpu < r->cpu)
160 		return 1;
161 	if (l->cpu > r->cpu)
162 		return -1;
163 
164 	return 0;
165 }
166 
167 static int sort_dimension__add(struct perf_kwork *kwork __maybe_unused,
168 			       const char *tok, struct list_head *list)
169 {
170 	size_t i;
171 	static struct sort_dimension max_sort_dimension = {
172 		.name = "max",
173 		.cmp  = max_runtime_cmp,
174 	};
175 	static struct sort_dimension id_sort_dimension = {
176 		.name = "id",
177 		.cmp  = id_cmp,
178 	};
179 	static struct sort_dimension runtime_sort_dimension = {
180 		.name = "runtime",
181 		.cmp  = runtime_cmp,
182 	};
183 	static struct sort_dimension count_sort_dimension = {
184 		.name = "count",
185 		.cmp  = count_cmp,
186 	};
187 	static struct sort_dimension avg_sort_dimension = {
188 		.name = "avg",
189 		.cmp  = avg_latency_cmp,
190 	};
191 	static struct sort_dimension rate_sort_dimension = {
192 		.name = "rate",
193 		.cmp  = cpu_usage_cmp,
194 	};
195 	static struct sort_dimension tid_sort_dimension = {
196 		.name = "tid",
197 		.cmp  = id_or_cpu_r_cmp,
198 	};
199 	struct sort_dimension *available_sorts[] = {
200 		&id_sort_dimension,
201 		&max_sort_dimension,
202 		&count_sort_dimension,
203 		&runtime_sort_dimension,
204 		&avg_sort_dimension,
205 		&rate_sort_dimension,
206 		&tid_sort_dimension,
207 	};
208 
209 	if (kwork->report == KWORK_REPORT_LATENCY)
210 		max_sort_dimension.cmp = max_latency_cmp;
211 
212 	for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
213 		if (!strcmp(available_sorts[i]->name, tok)) {
214 			list_add_tail(&available_sorts[i]->list, list);
215 			return 0;
216 		}
217 	}
218 
219 	return -1;
220 }
221 
222 static void setup_sorting(struct perf_kwork *kwork,
223 			  const struct option *options,
224 			  const char * const usage_msg[])
225 {
226 	char *tmp, *tok, *str = strdup(kwork->sort_order);
227 
228 	for (tok = strtok_r(str, ", ", &tmp);
229 	     tok; tok = strtok_r(NULL, ", ", &tmp)) {
230 		if (sort_dimension__add(kwork, tok, &kwork->sort_list) < 0)
231 			usage_with_options_msg(usage_msg, options,
232 					       "Unknown --sort key: `%s'", tok);
233 	}
234 
235 	pr_debug("Sort order: %s\n", kwork->sort_order);
236 	free(str);
237 }
238 
239 static struct kwork_atom *atom_new(struct perf_kwork *kwork,
240 				   struct perf_sample *sample)
241 {
242 	unsigned long i;
243 	struct kwork_atom_page *page;
244 	struct kwork_atom *atom = NULL;
245 
246 	list_for_each_entry(page, &kwork->atom_page_list, list) {
247 		if (!bitmap_full(page->bitmap, NR_ATOM_PER_PAGE)) {
248 			i = find_first_zero_bit(page->bitmap, NR_ATOM_PER_PAGE);
249 			BUG_ON(i >= NR_ATOM_PER_PAGE);
250 			atom = &page->atoms[i];
251 			goto found_atom;
252 		}
253 	}
254 
255 	/*
256 	 * new page
257 	 */
258 	page = zalloc(sizeof(*page));
259 	if (page == NULL) {
260 		pr_err("Failed to zalloc kwork atom page\n");
261 		return NULL;
262 	}
263 
264 	i = 0;
265 	atom = &page->atoms[0];
266 	list_add_tail(&page->list, &kwork->atom_page_list);
267 
268 found_atom:
269 	__set_bit(i, page->bitmap);
270 	atom->time = sample->time;
271 	atom->prev = NULL;
272 	atom->page_addr = page;
273 	atom->bit_inpage = i;
274 	return atom;
275 }
276 
277 static void atom_free(struct kwork_atom *atom)
278 {
279 	if (atom->prev != NULL)
280 		atom_free(atom->prev);
281 
282 	__clear_bit(atom->bit_inpage,
283 		    ((struct kwork_atom_page *)atom->page_addr)->bitmap);
284 }
285 
286 static void atom_del(struct kwork_atom *atom)
287 {
288 	list_del(&atom->list);
289 	atom_free(atom);
290 }
291 
292 static int work_cmp(struct list_head *list,
293 		    struct kwork_work *l, struct kwork_work *r)
294 {
295 	int ret = 0;
296 	struct sort_dimension *sort;
297 
298 	BUG_ON(list_empty(list));
299 
300 	list_for_each_entry(sort, list, list) {
301 		ret = sort->cmp(l, r);
302 		if (ret)
303 			return ret;
304 	}
305 
306 	return ret;
307 }
308 
309 static struct kwork_work *work_search(struct rb_root_cached *root,
310 				      struct kwork_work *key,
311 				      struct list_head *sort_list)
312 {
313 	int cmp;
314 	struct kwork_work *work;
315 	struct rb_node *node = root->rb_root.rb_node;
316 
317 	while (node) {
318 		work = container_of(node, struct kwork_work, node);
319 		cmp = work_cmp(sort_list, key, work);
320 		if (cmp > 0)
321 			node = node->rb_left;
322 		else if (cmp < 0)
323 			node = node->rb_right;
324 		else {
325 			if (work->name == NULL)
326 				work->name = key->name;
327 			return work;
328 		}
329 	}
330 	return NULL;
331 }
332 
333 static void work_insert(struct rb_root_cached *root,
334 			struct kwork_work *key, struct list_head *sort_list)
335 {
336 	int cmp;
337 	bool leftmost = true;
338 	struct kwork_work *cur;
339 	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
340 
341 	while (*new) {
342 		cur = container_of(*new, struct kwork_work, node);
343 		parent = *new;
344 		cmp = work_cmp(sort_list, key, cur);
345 
346 		if (cmp > 0)
347 			new = &((*new)->rb_left);
348 		else {
349 			new = &((*new)->rb_right);
350 			leftmost = false;
351 		}
352 	}
353 
354 	rb_link_node(&key->node, parent, new);
355 	rb_insert_color_cached(&key->node, root, leftmost);
356 }
357 
358 static struct kwork_work *work_new(struct kwork_work *key)
359 {
360 	int i;
361 	struct kwork_work *work = zalloc(sizeof(*work));
362 
363 	if (work == NULL) {
364 		pr_err("Failed to zalloc kwork work\n");
365 		return NULL;
366 	}
367 
368 	for (i = 0; i < KWORK_TRACE_MAX; i++)
369 		INIT_LIST_HEAD(&work->atom_list[i]);
370 
371 	work->id = key->id;
372 	work->cpu = key->cpu;
373 	work->name = key->name;
374 	work->class = key->class;
375 	return work;
376 }
377 
378 static struct kwork_work *work_findnew(struct rb_root_cached *root,
379 				       struct kwork_work *key,
380 				       struct list_head *sort_list)
381 {
382 	struct kwork_work *work = work_search(root, key, sort_list);
383 
384 	if (work != NULL)
385 		return work;
386 
387 	work = work_new(key);
388 	if (work)
389 		work_insert(root, work, sort_list);
390 
391 	return work;
392 }
393 
394 static void profile_update_timespan(struct perf_kwork *kwork,
395 				    struct perf_sample *sample)
396 {
397 	if (!kwork->summary)
398 		return;
399 
400 	if ((kwork->timestart == 0) || (kwork->timestart > sample->time))
401 		kwork->timestart = sample->time;
402 
403 	if (kwork->timeend < sample->time)
404 		kwork->timeend = sample->time;
405 }
406 
407 static bool profile_name_match(struct perf_kwork *kwork,
408 			       struct kwork_work *work)
409 {
410 	if (kwork->profile_name && work->name &&
411 	    (strcmp(work->name, kwork->profile_name) != 0)) {
412 		return false;
413 	}
414 
415 	return true;
416 }
417 
418 static bool profile_event_match(struct perf_kwork *kwork,
419 				struct kwork_work *work,
420 				struct perf_sample *sample)
421 {
422 	int cpu = work->cpu;
423 	u64 time = sample->time;
424 	struct perf_time_interval *ptime = &kwork->ptime;
425 
426 	if ((kwork->cpu_list != NULL) && !test_bit(cpu, kwork->cpu_bitmap))
427 		return false;
428 
429 	if (((ptime->start != 0) && (ptime->start > time)) ||
430 	    ((ptime->end != 0) && (ptime->end < time)))
431 		return false;
432 
433 	/*
434 	 * report top needs to collect the runtime of all tasks to
435 	 * calculate the load of each core.
436 	 */
437 	if ((kwork->report != KWORK_REPORT_TOP) &&
438 	    !profile_name_match(kwork, work)) {
439 		return false;
440 	}
441 
442 	profile_update_timespan(kwork, sample);
443 	return true;
444 }
445 
446 static int work_push_atom(struct perf_kwork *kwork,
447 			  struct kwork_class *class,
448 			  enum kwork_trace_type src_type,
449 			  enum kwork_trace_type dst_type,
450 			  struct evsel *evsel,
451 			  struct perf_sample *sample,
452 			  struct machine *machine,
453 			  struct kwork_work **ret_work,
454 			  bool overwrite)
455 {
456 	struct kwork_atom *atom, *dst_atom, *last_atom;
457 	struct kwork_work *work, key;
458 
459 	BUG_ON(class->work_init == NULL);
460 	class->work_init(kwork, class, &key, src_type, evsel, sample, machine);
461 
462 	atom = atom_new(kwork, sample);
463 	if (atom == NULL)
464 		return -1;
465 
466 	work = work_findnew(&class->work_root, &key, &kwork->cmp_id);
467 	if (work == NULL) {
468 		atom_free(atom);
469 		return -1;
470 	}
471 
472 	if (!profile_event_match(kwork, work, sample)) {
473 		atom_free(atom);
474 		return 0;
475 	}
476 
477 	if (dst_type < KWORK_TRACE_MAX) {
478 		dst_atom = list_last_entry_or_null(&work->atom_list[dst_type],
479 						   struct kwork_atom, list);
480 		if (dst_atom != NULL) {
481 			atom->prev = dst_atom;
482 			list_del(&dst_atom->list);
483 		}
484 	}
485 
486 	if (ret_work != NULL)
487 		*ret_work = work;
488 
489 	if (overwrite) {
490 		last_atom = list_last_entry_or_null(&work->atom_list[src_type],
491 						    struct kwork_atom, list);
492 		if (last_atom) {
493 			atom_del(last_atom);
494 
495 			kwork->nr_skipped_events[src_type]++;
496 			kwork->nr_skipped_events[KWORK_TRACE_MAX]++;
497 		}
498 	}
499 
500 	list_add_tail(&atom->list, &work->atom_list[src_type]);
501 
502 	return 0;
503 }
504 
505 static struct kwork_atom *work_pop_atom(struct perf_kwork *kwork,
506 					struct kwork_class *class,
507 					enum kwork_trace_type src_type,
508 					enum kwork_trace_type dst_type,
509 					struct evsel *evsel,
510 					struct perf_sample *sample,
511 					struct machine *machine,
512 					struct kwork_work **ret_work)
513 {
514 	struct kwork_atom *atom, *src_atom;
515 	struct kwork_work *work, key;
516 
517 	BUG_ON(class->work_init == NULL);
518 	class->work_init(kwork, class, &key, src_type, evsel, sample, machine);
519 
520 	work = work_findnew(&class->work_root, &key, &kwork->cmp_id);
521 	if (ret_work != NULL)
522 		*ret_work = work;
523 
524 	if (work == NULL)
525 		return NULL;
526 
527 	if (!profile_event_match(kwork, work, sample))
528 		return NULL;
529 
530 	atom = list_last_entry_or_null(&work->atom_list[dst_type],
531 				       struct kwork_atom, list);
532 	if (atom != NULL)
533 		return atom;
534 
535 	src_atom = atom_new(kwork, sample);
536 	if (src_atom != NULL)
537 		list_add_tail(&src_atom->list, &work->atom_list[src_type]);
538 	else {
539 		if (ret_work != NULL)
540 			*ret_work = NULL;
541 	}
542 
543 	return NULL;
544 }
545 
546 static struct kwork_work *find_work_by_id(struct rb_root_cached *root,
547 					  u64 id, int cpu)
548 {
549 	struct rb_node *next;
550 	struct kwork_work *work;
551 
552 	next = rb_first_cached(root);
553 	while (next) {
554 		work = rb_entry(next, struct kwork_work, node);
555 		if ((cpu != -1 && work->id == id && work->cpu == cpu) ||
556 		    (cpu == -1 && work->id == id))
557 			return work;
558 
559 		next = rb_next(next);
560 	}
561 
562 	return NULL;
563 }
564 
565 static struct kwork_class *get_kwork_class(struct perf_kwork *kwork,
566 					   enum kwork_class_type type)
567 {
568 	struct kwork_class *class;
569 
570 	list_for_each_entry(class, &kwork->class_list, list) {
571 		if (class->type == type)
572 			return class;
573 	}
574 
575 	return NULL;
576 }
577 
578 static void report_update_exit_event(struct kwork_work *work,
579 				     struct kwork_atom *atom,
580 				     struct perf_sample *sample)
581 {
582 	u64 delta;
583 	u64 exit_time = sample->time;
584 	u64 entry_time = atom->time;
585 
586 	if ((entry_time != 0) && (exit_time >= entry_time)) {
587 		delta = exit_time - entry_time;
588 		if ((delta > work->max_runtime) ||
589 		    (work->max_runtime == 0)) {
590 			work->max_runtime = delta;
591 			work->max_runtime_start = entry_time;
592 			work->max_runtime_end = exit_time;
593 		}
594 		work->total_runtime += delta;
595 		work->nr_atoms++;
596 	}
597 }
598 
599 static int report_entry_event(struct perf_kwork *kwork,
600 			      struct kwork_class *class,
601 			      struct evsel *evsel,
602 			      struct perf_sample *sample,
603 			      struct machine *machine)
604 {
605 	return work_push_atom(kwork, class, KWORK_TRACE_ENTRY,
606 			      KWORK_TRACE_MAX, evsel, sample,
607 			      machine, NULL, true);
608 }
609 
610 static int report_exit_event(struct perf_kwork *kwork,
611 			     struct kwork_class *class,
612 			     struct evsel *evsel,
613 			     struct perf_sample *sample,
614 			     struct machine *machine)
615 {
616 	struct kwork_atom *atom = NULL;
617 	struct kwork_work *work = NULL;
618 
619 	atom = work_pop_atom(kwork, class, KWORK_TRACE_EXIT,
620 			     KWORK_TRACE_ENTRY, evsel, sample,
621 			     machine, &work);
622 	if (work == NULL)
623 		return -1;
624 
625 	if (atom != NULL) {
626 		report_update_exit_event(work, atom, sample);
627 		atom_del(atom);
628 	}
629 
630 	return 0;
631 }
632 
633 static void latency_update_entry_event(struct kwork_work *work,
634 				       struct kwork_atom *atom,
635 				       struct perf_sample *sample)
636 {
637 	u64 delta;
638 	u64 entry_time = sample->time;
639 	u64 raise_time = atom->time;
640 
641 	if ((raise_time != 0) && (entry_time >= raise_time)) {
642 		delta = entry_time - raise_time;
643 		if ((delta > work->max_latency) ||
644 		    (work->max_latency == 0)) {
645 			work->max_latency = delta;
646 			work->max_latency_start = raise_time;
647 			work->max_latency_end = entry_time;
648 		}
649 		work->total_latency += delta;
650 		work->nr_atoms++;
651 	}
652 }
653 
654 static int latency_raise_event(struct perf_kwork *kwork,
655 			       struct kwork_class *class,
656 			       struct evsel *evsel,
657 			       struct perf_sample *sample,
658 			       struct machine *machine)
659 {
660 	return work_push_atom(kwork, class, KWORK_TRACE_RAISE,
661 			      KWORK_TRACE_MAX, evsel, sample,
662 			      machine, NULL, true);
663 }
664 
665 static int latency_entry_event(struct perf_kwork *kwork,
666 			       struct kwork_class *class,
667 			       struct evsel *evsel,
668 			       struct perf_sample *sample,
669 			       struct machine *machine)
670 {
671 	struct kwork_atom *atom = NULL;
672 	struct kwork_work *work = NULL;
673 
674 	atom = work_pop_atom(kwork, class, KWORK_TRACE_ENTRY,
675 			     KWORK_TRACE_RAISE, evsel, sample,
676 			     machine, &work);
677 	if (work == NULL)
678 		return -1;
679 
680 	if (atom != NULL) {
681 		latency_update_entry_event(work, atom, sample);
682 		atom_del(atom);
683 	}
684 
685 	return 0;
686 }
687 
688 static void timehist_save_callchain(struct perf_kwork *kwork,
689 				    struct perf_sample *sample,
690 				    struct evsel *evsel,
691 				    struct machine *machine)
692 {
693 	struct symbol *sym;
694 	struct thread *thread;
695 	struct callchain_cursor_node *node;
696 	struct callchain_cursor *cursor;
697 
698 	if (!kwork->show_callchain || sample->callchain == NULL)
699 		return;
700 
701 	/* want main thread for process - has maps */
702 	thread = machine__findnew_thread(machine, sample->pid, sample->pid);
703 	if (thread == NULL) {
704 		pr_debug("Failed to get thread for pid %d\n", sample->pid);
705 		return;
706 	}
707 
708 	cursor = get_tls_callchain_cursor();
709 
710 	if (thread__resolve_callchain(thread, cursor, evsel, sample,
711 				      NULL, NULL, kwork->max_stack + 2) != 0) {
712 		pr_debug("Failed to resolve callchain, skipping\n");
713 		goto out_put;
714 	}
715 
716 	callchain_cursor_commit(cursor);
717 
718 	while (true) {
719 		node = callchain_cursor_current(cursor);
720 		if (node == NULL)
721 			break;
722 
723 		sym = node->ms.sym;
724 		if (sym) {
725 			if (!strcmp(sym->name, "__softirqentry_text_start") ||
726 			    !strcmp(sym->name, "__do_softirq"))
727 				sym->ignore = 1;
728 		}
729 
730 		callchain_cursor_advance(cursor);
731 	}
732 
733 out_put:
734 	thread__put(thread);
735 }
736 
737 static void timehist_print_event(struct perf_kwork *kwork,
738 				 struct kwork_work *work,
739 				 struct kwork_atom *atom,
740 				 struct perf_sample *sample,
741 				 struct addr_location *al)
742 {
743 	char entrytime[32], exittime[32];
744 	char kwork_name[PRINT_KWORK_NAME_WIDTH];
745 
746 	/*
747 	 * runtime start
748 	 */
749 	timestamp__scnprintf_usec(atom->time,
750 				  entrytime, sizeof(entrytime));
751 	printf(" %*s ", PRINT_TIMESTAMP_WIDTH, entrytime);
752 
753 	/*
754 	 * runtime end
755 	 */
756 	timestamp__scnprintf_usec(sample->time,
757 				  exittime, sizeof(exittime));
758 	printf(" %*s ", PRINT_TIMESTAMP_WIDTH, exittime);
759 
760 	/*
761 	 * cpu
762 	 */
763 	printf(" [%0*d] ", PRINT_CPU_WIDTH, work->cpu);
764 
765 	/*
766 	 * kwork name
767 	 */
768 	if (work->class && work->class->work_name) {
769 		work->class->work_name(work, kwork_name,
770 				       PRINT_KWORK_NAME_WIDTH);
771 		printf(" %-*s ", PRINT_KWORK_NAME_WIDTH, kwork_name);
772 	} else
773 		printf(" %-*s ", PRINT_KWORK_NAME_WIDTH, "");
774 
775 	/*
776 	 *runtime
777 	 */
778 	printf(" %*.*f ",
779 	       PRINT_RUNTIME_WIDTH, RPINT_DECIMAL_WIDTH,
780 	       (double)(sample->time - atom->time) / NSEC_PER_MSEC);
781 
782 	/*
783 	 * delaytime
784 	 */
785 	if (atom->prev != NULL)
786 		printf(" %*.*f ", PRINT_LATENCY_WIDTH, RPINT_DECIMAL_WIDTH,
787 		       (double)(atom->time - atom->prev->time) / NSEC_PER_MSEC);
788 	else
789 		printf(" %*s ", PRINT_LATENCY_WIDTH, " ");
790 
791 	/*
792 	 * callchain
793 	 */
794 	if (kwork->show_callchain) {
795 		struct callchain_cursor *cursor = get_tls_callchain_cursor();
796 
797 		if (cursor == NULL)
798 			return;
799 
800 		printf(" ");
801 
802 		sample__fprintf_sym(sample, al, 0,
803 				    EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE |
804 				    EVSEL__PRINT_CALLCHAIN_ARROW |
805 				    EVSEL__PRINT_SKIP_IGNORED,
806 				    cursor, symbol_conf.bt_stop_list,
807 				    stdout);
808 	}
809 
810 	printf("\n");
811 }
812 
813 static int timehist_raise_event(struct perf_kwork *kwork,
814 				struct kwork_class *class,
815 				struct evsel *evsel,
816 				struct perf_sample *sample,
817 				struct machine *machine)
818 {
819 	return work_push_atom(kwork, class, KWORK_TRACE_RAISE,
820 			      KWORK_TRACE_MAX, evsel, sample,
821 			      machine, NULL, true);
822 }
823 
824 static int timehist_entry_event(struct perf_kwork *kwork,
825 				struct kwork_class *class,
826 				struct evsel *evsel,
827 				struct perf_sample *sample,
828 				struct machine *machine)
829 {
830 	int ret;
831 	struct kwork_work *work = NULL;
832 
833 	ret = work_push_atom(kwork, class, KWORK_TRACE_ENTRY,
834 			     KWORK_TRACE_RAISE, evsel, sample,
835 			     machine, &work, true);
836 	if (ret)
837 		return ret;
838 
839 	if (work != NULL)
840 		timehist_save_callchain(kwork, sample, evsel, machine);
841 
842 	return 0;
843 }
844 
845 static int timehist_exit_event(struct perf_kwork *kwork,
846 			       struct kwork_class *class,
847 			       struct evsel *evsel,
848 			       struct perf_sample *sample,
849 			       struct machine *machine)
850 {
851 	struct kwork_atom *atom = NULL;
852 	struct kwork_work *work = NULL;
853 	struct addr_location al;
854 	int ret = 0;
855 
856 	addr_location__init(&al);
857 	if (machine__resolve(machine, &al, sample) < 0) {
858 		pr_debug("Problem processing event, skipping it\n");
859 		ret = -1;
860 		goto out;
861 	}
862 
863 	atom = work_pop_atom(kwork, class, KWORK_TRACE_EXIT,
864 			     KWORK_TRACE_ENTRY, evsel, sample,
865 			     machine, &work);
866 	if (work == NULL) {
867 		ret = -1;
868 		goto out;
869 	}
870 
871 	if (atom != NULL) {
872 		work->nr_atoms++;
873 		timehist_print_event(kwork, work, atom, sample, &al);
874 		atom_del(atom);
875 	}
876 
877 out:
878 	addr_location__exit(&al);
879 	return ret;
880 }
881 
882 static void top_update_runtime(struct kwork_work *work,
883 			       struct kwork_atom *atom,
884 			       struct perf_sample *sample)
885 {
886 	u64 delta;
887 	u64 exit_time = sample->time;
888 	u64 entry_time = atom->time;
889 
890 	if ((entry_time != 0) && (exit_time >= entry_time)) {
891 		delta = exit_time - entry_time;
892 		work->total_runtime += delta;
893 	}
894 }
895 
896 static int top_entry_event(struct perf_kwork *kwork,
897 			   struct kwork_class *class,
898 			   struct evsel *evsel,
899 			   struct perf_sample *sample,
900 			   struct machine *machine)
901 {
902 	return work_push_atom(kwork, class, KWORK_TRACE_ENTRY,
903 			      KWORK_TRACE_MAX, evsel, sample,
904 			      machine, NULL, true);
905 }
906 
907 static int top_exit_event(struct perf_kwork *kwork,
908 			  struct kwork_class *class,
909 			  struct evsel *evsel,
910 			  struct perf_sample *sample,
911 			  struct machine *machine)
912 {
913 	struct kwork_work *work, *sched_work;
914 	struct kwork_class *sched_class;
915 	struct kwork_atom *atom;
916 
917 	atom = work_pop_atom(kwork, class, KWORK_TRACE_EXIT,
918 			     KWORK_TRACE_ENTRY, evsel, sample,
919 			     machine, &work);
920 	if (!work)
921 		return -1;
922 
923 	if (atom) {
924 		sched_class = get_kwork_class(kwork, KWORK_CLASS_SCHED);
925 		if (sched_class) {
926 			sched_work = find_work_by_id(&sched_class->work_root,
927 						     work->id, work->cpu);
928 			if (sched_work)
929 				top_update_runtime(work, atom, sample);
930 		}
931 		atom_del(atom);
932 	}
933 
934 	return 0;
935 }
936 
937 static int top_sched_switch_event(struct perf_kwork *kwork,
938 				  struct kwork_class *class,
939 				  struct evsel *evsel,
940 				  struct perf_sample *sample,
941 				  struct machine *machine)
942 {
943 	struct kwork_atom *atom;
944 	struct kwork_work *work;
945 
946 	atom = work_pop_atom(kwork, class, KWORK_TRACE_EXIT,
947 			     KWORK_TRACE_ENTRY, evsel, sample,
948 			     machine, &work);
949 	if (!work)
950 		return -1;
951 
952 	if (atom) {
953 		top_update_runtime(work, atom, sample);
954 		atom_del(atom);
955 	}
956 
957 	return top_entry_event(kwork, class, evsel, sample, machine);
958 }
959 
960 static struct kwork_class kwork_irq;
961 static int process_irq_handler_entry_event(struct perf_tool *tool,
962 					   struct evsel *evsel,
963 					   struct perf_sample *sample,
964 					   struct machine *machine)
965 {
966 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
967 
968 	if (kwork->tp_handler->entry_event)
969 		return kwork->tp_handler->entry_event(kwork, &kwork_irq,
970 						      evsel, sample, machine);
971 	return 0;
972 }
973 
974 static int process_irq_handler_exit_event(struct perf_tool *tool,
975 					  struct evsel *evsel,
976 					  struct perf_sample *sample,
977 					  struct machine *machine)
978 {
979 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
980 
981 	if (kwork->tp_handler->exit_event)
982 		return kwork->tp_handler->exit_event(kwork, &kwork_irq,
983 						     evsel, sample, machine);
984 	return 0;
985 }
986 
987 const struct evsel_str_handler irq_tp_handlers[] = {
988 	{ "irq:irq_handler_entry", process_irq_handler_entry_event, },
989 	{ "irq:irq_handler_exit",  process_irq_handler_exit_event,  },
990 };
991 
992 static int irq_class_init(struct kwork_class *class,
993 			  struct perf_session *session)
994 {
995 	if (perf_session__set_tracepoints_handlers(session, irq_tp_handlers)) {
996 		pr_err("Failed to set irq tracepoints handlers\n");
997 		return -1;
998 	}
999 
1000 	class->work_root = RB_ROOT_CACHED;
1001 	return 0;
1002 }
1003 
1004 static void irq_work_init(struct perf_kwork *kwork,
1005 			  struct kwork_class *class,
1006 			  struct kwork_work *work,
1007 			  enum kwork_trace_type src_type __maybe_unused,
1008 			  struct evsel *evsel,
1009 			  struct perf_sample *sample,
1010 			  struct machine *machine __maybe_unused)
1011 {
1012 	work->class = class;
1013 	work->cpu = sample->cpu;
1014 
1015 	if (kwork->report == KWORK_REPORT_TOP) {
1016 		work->id = evsel__intval_common(evsel, sample, "common_pid");
1017 		work->name = NULL;
1018 	} else {
1019 		work->id = evsel__intval(evsel, sample, "irq");
1020 		work->name = evsel__strval(evsel, sample, "name");
1021 	}
1022 }
1023 
1024 static void irq_work_name(struct kwork_work *work, char *buf, int len)
1025 {
1026 	snprintf(buf, len, "%s:%" PRIu64 "", work->name, work->id);
1027 }
1028 
1029 static struct kwork_class kwork_irq = {
1030 	.name           = "irq",
1031 	.type           = KWORK_CLASS_IRQ,
1032 	.nr_tracepoints = 2,
1033 	.tp_handlers    = irq_tp_handlers,
1034 	.class_init     = irq_class_init,
1035 	.work_init      = irq_work_init,
1036 	.work_name      = irq_work_name,
1037 };
1038 
1039 static struct kwork_class kwork_softirq;
1040 static int process_softirq_raise_event(struct perf_tool *tool,
1041 				       struct evsel *evsel,
1042 				       struct perf_sample *sample,
1043 				       struct machine *machine)
1044 {
1045 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1046 
1047 	if (kwork->tp_handler->raise_event)
1048 		return kwork->tp_handler->raise_event(kwork, &kwork_softirq,
1049 						      evsel, sample, machine);
1050 
1051 	return 0;
1052 }
1053 
1054 static int process_softirq_entry_event(struct perf_tool *tool,
1055 				       struct evsel *evsel,
1056 				       struct perf_sample *sample,
1057 				       struct machine *machine)
1058 {
1059 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1060 
1061 	if (kwork->tp_handler->entry_event)
1062 		return kwork->tp_handler->entry_event(kwork, &kwork_softirq,
1063 						      evsel, sample, machine);
1064 
1065 	return 0;
1066 }
1067 
1068 static int process_softirq_exit_event(struct perf_tool *tool,
1069 				      struct evsel *evsel,
1070 				      struct perf_sample *sample,
1071 				      struct machine *machine)
1072 {
1073 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1074 
1075 	if (kwork->tp_handler->exit_event)
1076 		return kwork->tp_handler->exit_event(kwork, &kwork_softirq,
1077 						     evsel, sample, machine);
1078 
1079 	return 0;
1080 }
1081 
1082 const struct evsel_str_handler softirq_tp_handlers[] = {
1083 	{ "irq:softirq_raise", process_softirq_raise_event, },
1084 	{ "irq:softirq_entry", process_softirq_entry_event, },
1085 	{ "irq:softirq_exit",  process_softirq_exit_event,  },
1086 };
1087 
1088 static int softirq_class_init(struct kwork_class *class,
1089 			      struct perf_session *session)
1090 {
1091 	if (perf_session__set_tracepoints_handlers(session,
1092 						   softirq_tp_handlers)) {
1093 		pr_err("Failed to set softirq tracepoints handlers\n");
1094 		return -1;
1095 	}
1096 
1097 	class->work_root = RB_ROOT_CACHED;
1098 	return 0;
1099 }
1100 
1101 static char *evsel__softirq_name(struct evsel *evsel, u64 num)
1102 {
1103 	char *name = NULL;
1104 	bool found = false;
1105 	struct tep_print_flag_sym *sym = NULL;
1106 	struct tep_print_arg *args = evsel->tp_format->print_fmt.args;
1107 
1108 	if ((args == NULL) || (args->next == NULL))
1109 		return NULL;
1110 
1111 	/* skip softirq field: "REC->vec" */
1112 	for (sym = args->next->symbol.symbols; sym != NULL; sym = sym->next) {
1113 		if ((eval_flag(sym->value) == (unsigned long long)num) &&
1114 		    (strlen(sym->str) != 0)) {
1115 			found = true;
1116 			break;
1117 		}
1118 	}
1119 
1120 	if (!found)
1121 		return NULL;
1122 
1123 	name = strdup(sym->str);
1124 	if (name == NULL) {
1125 		pr_err("Failed to copy symbol name\n");
1126 		return NULL;
1127 	}
1128 	return name;
1129 }
1130 
1131 static void softirq_work_init(struct perf_kwork *kwork,
1132 			      struct kwork_class *class,
1133 			      struct kwork_work *work,
1134 			      enum kwork_trace_type src_type __maybe_unused,
1135 			      struct evsel *evsel,
1136 			      struct perf_sample *sample,
1137 			      struct machine *machine __maybe_unused)
1138 {
1139 	u64 num;
1140 
1141 	work->class = class;
1142 	work->cpu = sample->cpu;
1143 
1144 	if (kwork->report == KWORK_REPORT_TOP) {
1145 		work->id = evsel__intval_common(evsel, sample, "common_pid");
1146 		work->name = NULL;
1147 	} else {
1148 		num = evsel__intval(evsel, sample, "vec");
1149 		work->id = num;
1150 		work->name = evsel__softirq_name(evsel, num);
1151 	}
1152 }
1153 
1154 static void softirq_work_name(struct kwork_work *work, char *buf, int len)
1155 {
1156 	snprintf(buf, len, "(s)%s:%" PRIu64 "", work->name, work->id);
1157 }
1158 
1159 static struct kwork_class kwork_softirq = {
1160 	.name           = "softirq",
1161 	.type           = KWORK_CLASS_SOFTIRQ,
1162 	.nr_tracepoints = 3,
1163 	.tp_handlers    = softirq_tp_handlers,
1164 	.class_init     = softirq_class_init,
1165 	.work_init      = softirq_work_init,
1166 	.work_name      = softirq_work_name,
1167 };
1168 
1169 static struct kwork_class kwork_workqueue;
1170 static int process_workqueue_activate_work_event(struct perf_tool *tool,
1171 						 struct evsel *evsel,
1172 						 struct perf_sample *sample,
1173 						 struct machine *machine)
1174 {
1175 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1176 
1177 	if (kwork->tp_handler->raise_event)
1178 		return kwork->tp_handler->raise_event(kwork, &kwork_workqueue,
1179 						    evsel, sample, machine);
1180 
1181 	return 0;
1182 }
1183 
1184 static int process_workqueue_execute_start_event(struct perf_tool *tool,
1185 						 struct evsel *evsel,
1186 						 struct perf_sample *sample,
1187 						 struct machine *machine)
1188 {
1189 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1190 
1191 	if (kwork->tp_handler->entry_event)
1192 		return kwork->tp_handler->entry_event(kwork, &kwork_workqueue,
1193 						    evsel, sample, machine);
1194 
1195 	return 0;
1196 }
1197 
1198 static int process_workqueue_execute_end_event(struct perf_tool *tool,
1199 					       struct evsel *evsel,
1200 					       struct perf_sample *sample,
1201 					       struct machine *machine)
1202 {
1203 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1204 
1205 	if (kwork->tp_handler->exit_event)
1206 		return kwork->tp_handler->exit_event(kwork, &kwork_workqueue,
1207 						   evsel, sample, machine);
1208 
1209 	return 0;
1210 }
1211 
1212 const struct evsel_str_handler workqueue_tp_handlers[] = {
1213 	{ "workqueue:workqueue_activate_work", process_workqueue_activate_work_event, },
1214 	{ "workqueue:workqueue_execute_start", process_workqueue_execute_start_event, },
1215 	{ "workqueue:workqueue_execute_end",   process_workqueue_execute_end_event,   },
1216 };
1217 
1218 static int workqueue_class_init(struct kwork_class *class,
1219 				struct perf_session *session)
1220 {
1221 	if (perf_session__set_tracepoints_handlers(session,
1222 						   workqueue_tp_handlers)) {
1223 		pr_err("Failed to set workqueue tracepoints handlers\n");
1224 		return -1;
1225 	}
1226 
1227 	class->work_root = RB_ROOT_CACHED;
1228 	return 0;
1229 }
1230 
1231 static void workqueue_work_init(struct perf_kwork *kwork __maybe_unused,
1232 				struct kwork_class *class,
1233 				struct kwork_work *work,
1234 				enum kwork_trace_type src_type __maybe_unused,
1235 				struct evsel *evsel,
1236 				struct perf_sample *sample,
1237 				struct machine *machine)
1238 {
1239 	char *modp = NULL;
1240 	unsigned long long function_addr = evsel__intval(evsel,
1241 							 sample, "function");
1242 
1243 	work->class = class;
1244 	work->cpu = sample->cpu;
1245 	work->id = evsel__intval(evsel, sample, "work");
1246 	work->name = function_addr == 0 ? NULL :
1247 		machine__resolve_kernel_addr(machine, &function_addr, &modp);
1248 }
1249 
1250 static void workqueue_work_name(struct kwork_work *work, char *buf, int len)
1251 {
1252 	if (work->name != NULL)
1253 		snprintf(buf, len, "(w)%s", work->name);
1254 	else
1255 		snprintf(buf, len, "(w)0x%" PRIx64, work->id);
1256 }
1257 
1258 static struct kwork_class kwork_workqueue = {
1259 	.name           = "workqueue",
1260 	.type           = KWORK_CLASS_WORKQUEUE,
1261 	.nr_tracepoints = 3,
1262 	.tp_handlers    = workqueue_tp_handlers,
1263 	.class_init     = workqueue_class_init,
1264 	.work_init      = workqueue_work_init,
1265 	.work_name      = workqueue_work_name,
1266 };
1267 
1268 static struct kwork_class kwork_sched;
1269 static int process_sched_switch_event(struct perf_tool *tool,
1270 				      struct evsel *evsel,
1271 				      struct perf_sample *sample,
1272 				      struct machine *machine)
1273 {
1274 	struct perf_kwork *kwork = container_of(tool, struct perf_kwork, tool);
1275 
1276 	if (kwork->tp_handler->sched_switch_event)
1277 		return kwork->tp_handler->sched_switch_event(kwork, &kwork_sched,
1278 							     evsel, sample, machine);
1279 	return 0;
1280 }
1281 
1282 const struct evsel_str_handler sched_tp_handlers[] = {
1283 	{ "sched:sched_switch",  process_sched_switch_event, },
1284 };
1285 
1286 static int sched_class_init(struct kwork_class *class,
1287 			    struct perf_session *session)
1288 {
1289 	if (perf_session__set_tracepoints_handlers(session,
1290 						   sched_tp_handlers)) {
1291 		pr_err("Failed to set sched tracepoints handlers\n");
1292 		return -1;
1293 	}
1294 
1295 	class->work_root = RB_ROOT_CACHED;
1296 	return 0;
1297 }
1298 
1299 static void sched_work_init(struct perf_kwork *kwork __maybe_unused,
1300 			    struct kwork_class *class,
1301 			    struct kwork_work *work,
1302 			    enum kwork_trace_type src_type,
1303 			    struct evsel *evsel,
1304 			    struct perf_sample *sample,
1305 			    struct machine *machine __maybe_unused)
1306 {
1307 	work->class = class;
1308 	work->cpu = sample->cpu;
1309 
1310 	if (src_type == KWORK_TRACE_EXIT) {
1311 		work->id = evsel__intval(evsel, sample, "prev_pid");
1312 		work->name = strdup(evsel__strval(evsel, sample, "prev_comm"));
1313 	} else if (src_type == KWORK_TRACE_ENTRY) {
1314 		work->id = evsel__intval(evsel, sample, "next_pid");
1315 		work->name = strdup(evsel__strval(evsel, sample, "next_comm"));
1316 	}
1317 }
1318 
1319 static void sched_work_name(struct kwork_work *work, char *buf, int len)
1320 {
1321 	snprintf(buf, len, "%s", work->name);
1322 }
1323 
1324 static struct kwork_class kwork_sched = {
1325 	.name		= "sched",
1326 	.type		= KWORK_CLASS_SCHED,
1327 	.nr_tracepoints	= ARRAY_SIZE(sched_tp_handlers),
1328 	.tp_handlers	= sched_tp_handlers,
1329 	.class_init	= sched_class_init,
1330 	.work_init	= sched_work_init,
1331 	.work_name	= sched_work_name,
1332 };
1333 
1334 static struct kwork_class *kwork_class_supported_list[KWORK_CLASS_MAX] = {
1335 	[KWORK_CLASS_IRQ]       = &kwork_irq,
1336 	[KWORK_CLASS_SOFTIRQ]   = &kwork_softirq,
1337 	[KWORK_CLASS_WORKQUEUE] = &kwork_workqueue,
1338 	[KWORK_CLASS_SCHED]     = &kwork_sched,
1339 };
1340 
1341 static void print_separator(int len)
1342 {
1343 	printf(" %.*s\n", len, graph_dotted_line);
1344 }
1345 
1346 static int report_print_work(struct perf_kwork *kwork, struct kwork_work *work)
1347 {
1348 	int ret = 0;
1349 	char kwork_name[PRINT_KWORK_NAME_WIDTH];
1350 	char max_runtime_start[32], max_runtime_end[32];
1351 	char max_latency_start[32], max_latency_end[32];
1352 
1353 	printf(" ");
1354 
1355 	/*
1356 	 * kwork name
1357 	 */
1358 	if (work->class && work->class->work_name) {
1359 		work->class->work_name(work, kwork_name,
1360 				       PRINT_KWORK_NAME_WIDTH);
1361 		ret += printf(" %-*s |", PRINT_KWORK_NAME_WIDTH, kwork_name);
1362 	} else {
1363 		ret += printf(" %-*s |", PRINT_KWORK_NAME_WIDTH, "");
1364 	}
1365 
1366 	/*
1367 	 * cpu
1368 	 */
1369 	ret += printf(" %0*d |", PRINT_CPU_WIDTH, work->cpu);
1370 
1371 	/*
1372 	 * total runtime
1373 	 */
1374 	if (kwork->report == KWORK_REPORT_RUNTIME) {
1375 		ret += printf(" %*.*f ms |",
1376 			      PRINT_RUNTIME_WIDTH, RPINT_DECIMAL_WIDTH,
1377 			      (double)work->total_runtime / NSEC_PER_MSEC);
1378 	} else if (kwork->report == KWORK_REPORT_LATENCY) { // avg delay
1379 		ret += printf(" %*.*f ms |",
1380 			      PRINT_LATENCY_WIDTH, RPINT_DECIMAL_WIDTH,
1381 			      (double)work->total_latency /
1382 			      work->nr_atoms / NSEC_PER_MSEC);
1383 	}
1384 
1385 	/*
1386 	 * count
1387 	 */
1388 	ret += printf(" %*" PRIu64 " |", PRINT_COUNT_WIDTH, work->nr_atoms);
1389 
1390 	/*
1391 	 * max runtime, max runtime start, max runtime end
1392 	 */
1393 	if (kwork->report == KWORK_REPORT_RUNTIME) {
1394 		timestamp__scnprintf_usec(work->max_runtime_start,
1395 					  max_runtime_start,
1396 					  sizeof(max_runtime_start));
1397 		timestamp__scnprintf_usec(work->max_runtime_end,
1398 					  max_runtime_end,
1399 					  sizeof(max_runtime_end));
1400 		ret += printf(" %*.*f ms | %*s s | %*s s |",
1401 			      PRINT_RUNTIME_WIDTH, RPINT_DECIMAL_WIDTH,
1402 			      (double)work->max_runtime / NSEC_PER_MSEC,
1403 			      PRINT_TIMESTAMP_WIDTH, max_runtime_start,
1404 			      PRINT_TIMESTAMP_WIDTH, max_runtime_end);
1405 	}
1406 	/*
1407 	 * max delay, max delay start, max delay end
1408 	 */
1409 	else if (kwork->report == KWORK_REPORT_LATENCY) {
1410 		timestamp__scnprintf_usec(work->max_latency_start,
1411 					  max_latency_start,
1412 					  sizeof(max_latency_start));
1413 		timestamp__scnprintf_usec(work->max_latency_end,
1414 					  max_latency_end,
1415 					  sizeof(max_latency_end));
1416 		ret += printf(" %*.*f ms | %*s s | %*s s |",
1417 			      PRINT_LATENCY_WIDTH, RPINT_DECIMAL_WIDTH,
1418 			      (double)work->max_latency / NSEC_PER_MSEC,
1419 			      PRINT_TIMESTAMP_WIDTH, max_latency_start,
1420 			      PRINT_TIMESTAMP_WIDTH, max_latency_end);
1421 	}
1422 
1423 	printf("\n");
1424 	return ret;
1425 }
1426 
1427 static int report_print_header(struct perf_kwork *kwork)
1428 {
1429 	int ret;
1430 
1431 	printf("\n ");
1432 	ret = printf(" %-*s | %-*s |",
1433 		     PRINT_KWORK_NAME_WIDTH, "Kwork Name",
1434 		     PRINT_CPU_WIDTH, "Cpu");
1435 
1436 	if (kwork->report == KWORK_REPORT_RUNTIME) {
1437 		ret += printf(" %-*s |",
1438 			      PRINT_RUNTIME_HEADER_WIDTH, "Total Runtime");
1439 	} else if (kwork->report == KWORK_REPORT_LATENCY) {
1440 		ret += printf(" %-*s |",
1441 			      PRINT_LATENCY_HEADER_WIDTH, "Avg delay");
1442 	}
1443 
1444 	ret += printf(" %-*s |", PRINT_COUNT_WIDTH, "Count");
1445 
1446 	if (kwork->report == KWORK_REPORT_RUNTIME) {
1447 		ret += printf(" %-*s | %-*s | %-*s |",
1448 			      PRINT_RUNTIME_HEADER_WIDTH, "Max runtime",
1449 			      PRINT_TIMESTAMP_HEADER_WIDTH, "Max runtime start",
1450 			      PRINT_TIMESTAMP_HEADER_WIDTH, "Max runtime end");
1451 	} else if (kwork->report == KWORK_REPORT_LATENCY) {
1452 		ret += printf(" %-*s | %-*s | %-*s |",
1453 			      PRINT_LATENCY_HEADER_WIDTH, "Max delay",
1454 			      PRINT_TIMESTAMP_HEADER_WIDTH, "Max delay start",
1455 			      PRINT_TIMESTAMP_HEADER_WIDTH, "Max delay end");
1456 	}
1457 
1458 	printf("\n");
1459 	print_separator(ret);
1460 	return ret;
1461 }
1462 
1463 static void timehist_print_header(void)
1464 {
1465 	/*
1466 	 * header row
1467 	 */
1468 	printf(" %-*s  %-*s  %-*s  %-*s  %-*s  %-*s\n",
1469 	       PRINT_TIMESTAMP_WIDTH, "Runtime start",
1470 	       PRINT_TIMESTAMP_WIDTH, "Runtime end",
1471 	       PRINT_TIMEHIST_CPU_WIDTH, "Cpu",
1472 	       PRINT_KWORK_NAME_WIDTH, "Kwork name",
1473 	       PRINT_RUNTIME_WIDTH, "Runtime",
1474 	       PRINT_RUNTIME_WIDTH, "Delaytime");
1475 
1476 	/*
1477 	 * units row
1478 	 */
1479 	printf(" %-*s  %-*s  %-*s  %-*s  %-*s  %-*s\n",
1480 	       PRINT_TIMESTAMP_WIDTH, "",
1481 	       PRINT_TIMESTAMP_WIDTH, "",
1482 	       PRINT_TIMEHIST_CPU_WIDTH, "",
1483 	       PRINT_KWORK_NAME_WIDTH, "(TYPE)NAME:NUM",
1484 	       PRINT_RUNTIME_WIDTH, "(msec)",
1485 	       PRINT_RUNTIME_WIDTH, "(msec)");
1486 
1487 	/*
1488 	 * separator
1489 	 */
1490 	printf(" %.*s  %.*s  %.*s  %.*s  %.*s  %.*s\n",
1491 	       PRINT_TIMESTAMP_WIDTH, graph_dotted_line,
1492 	       PRINT_TIMESTAMP_WIDTH, graph_dotted_line,
1493 	       PRINT_TIMEHIST_CPU_WIDTH, graph_dotted_line,
1494 	       PRINT_KWORK_NAME_WIDTH, graph_dotted_line,
1495 	       PRINT_RUNTIME_WIDTH, graph_dotted_line,
1496 	       PRINT_RUNTIME_WIDTH, graph_dotted_line);
1497 }
1498 
1499 static void print_summary(struct perf_kwork *kwork)
1500 {
1501 	u64 time = kwork->timeend - kwork->timestart;
1502 
1503 	printf("  Total count            : %9" PRIu64 "\n", kwork->all_count);
1504 	printf("  Total runtime   (msec) : %9.3f (%.3f%% load average)\n",
1505 	       (double)kwork->all_runtime / NSEC_PER_MSEC,
1506 	       time == 0 ? 0 : (double)kwork->all_runtime / time);
1507 	printf("  Total time span (msec) : %9.3f\n",
1508 	       (double)time / NSEC_PER_MSEC);
1509 }
1510 
1511 static unsigned long long nr_list_entry(struct list_head *head)
1512 {
1513 	struct list_head *pos;
1514 	unsigned long long n = 0;
1515 
1516 	list_for_each(pos, head)
1517 		n++;
1518 
1519 	return n;
1520 }
1521 
1522 static void print_skipped_events(struct perf_kwork *kwork)
1523 {
1524 	int i;
1525 	const char *const kwork_event_str[] = {
1526 		[KWORK_TRACE_RAISE] = "raise",
1527 		[KWORK_TRACE_ENTRY] = "entry",
1528 		[KWORK_TRACE_EXIT]  = "exit",
1529 	};
1530 
1531 	if ((kwork->nr_skipped_events[KWORK_TRACE_MAX] != 0) &&
1532 	    (kwork->nr_events != 0)) {
1533 		printf("  INFO: %.3f%% skipped events (%" PRIu64 " including ",
1534 		       (double)kwork->nr_skipped_events[KWORK_TRACE_MAX] /
1535 		       (double)kwork->nr_events * 100.0,
1536 		       kwork->nr_skipped_events[KWORK_TRACE_MAX]);
1537 
1538 		for (i = 0; i < KWORK_TRACE_MAX; i++) {
1539 			printf("%" PRIu64 " %s%s",
1540 			       kwork->nr_skipped_events[i],
1541 			       kwork_event_str[i],
1542 			       (i == KWORK_TRACE_MAX - 1) ? ")\n" : ", ");
1543 		}
1544 	}
1545 
1546 	if (verbose > 0)
1547 		printf("  INFO: use %lld atom pages\n",
1548 		       nr_list_entry(&kwork->atom_page_list));
1549 }
1550 
1551 static void print_bad_events(struct perf_kwork *kwork)
1552 {
1553 	if ((kwork->nr_lost_events != 0) && (kwork->nr_events != 0)) {
1554 		printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
1555 		       (double)kwork->nr_lost_events /
1556 		       (double)kwork->nr_events * 100.0,
1557 		       kwork->nr_lost_events, kwork->nr_events,
1558 		       kwork->nr_lost_chunks);
1559 	}
1560 }
1561 
1562 const char *graph_load = "||||||||||||||||||||||||||||||||||||||||||||||||";
1563 const char *graph_idle = "                                                ";
1564 static void top_print_per_cpu_load(struct perf_kwork *kwork)
1565 {
1566 	int i, load_width;
1567 	u64 total, load, load_ratio;
1568 	struct kwork_top_stat *stat = &kwork->top_stat;
1569 
1570 	for (i = 0; i < MAX_NR_CPUS; i++) {
1571 		total = stat->cpus_runtime[i].total;
1572 		load = stat->cpus_runtime[i].load;
1573 		if (test_bit(i, stat->all_cpus_bitmap) && total) {
1574 			load_ratio = load * 10000 / total;
1575 			load_width = PRINT_CPU_USAGE_HIST_WIDTH *
1576 				load_ratio / 10000;
1577 
1578 			printf("%%Cpu%-*d[%.*s%.*s %*.*f%%]\n",
1579 			       PRINT_CPU_WIDTH, i,
1580 			       load_width, graph_load,
1581 			       PRINT_CPU_USAGE_HIST_WIDTH - load_width,
1582 			       graph_idle,
1583 			       PRINT_CPU_USAGE_WIDTH,
1584 			       PRINT_CPU_USAGE_DECIMAL_WIDTH,
1585 			       (double)load_ratio / 100);
1586 		}
1587 	}
1588 }
1589 
1590 static void top_print_cpu_usage(struct perf_kwork *kwork)
1591 {
1592 	struct kwork_top_stat *stat = &kwork->top_stat;
1593 	u64 idle_time = stat->cpus_runtime[MAX_NR_CPUS].idle;
1594 	u64 hardirq_time = stat->cpus_runtime[MAX_NR_CPUS].irq;
1595 	u64 softirq_time = stat->cpus_runtime[MAX_NR_CPUS].softirq;
1596 	int cpus_nr = bitmap_weight(stat->all_cpus_bitmap, MAX_NR_CPUS);
1597 	u64 cpus_total_time = stat->cpus_runtime[MAX_NR_CPUS].total;
1598 
1599 	printf("Total  : %*.*f ms, %d cpus\n",
1600 	       PRINT_RUNTIME_WIDTH, RPINT_DECIMAL_WIDTH,
1601 	       (double)cpus_total_time / NSEC_PER_MSEC,
1602 	       cpus_nr);
1603 
1604 	printf("%%Cpu(s): %*.*f%% id, %*.*f%% hi, %*.*f%% si\n",
1605 	       PRINT_CPU_USAGE_WIDTH, PRINT_CPU_USAGE_DECIMAL_WIDTH,
1606 	       cpus_total_time ? (double)idle_time * 100 / cpus_total_time : 0,
1607 
1608 	       PRINT_CPU_USAGE_WIDTH, PRINT_CPU_USAGE_DECIMAL_WIDTH,
1609 	       cpus_total_time ? (double)hardirq_time * 100 / cpus_total_time : 0,
1610 
1611 	       PRINT_CPU_USAGE_WIDTH, PRINT_CPU_USAGE_DECIMAL_WIDTH,
1612 	       cpus_total_time ? (double)softirq_time * 100 / cpus_total_time : 0);
1613 
1614 	top_print_per_cpu_load(kwork);
1615 }
1616 
1617 static void top_print_header(struct perf_kwork *kwork __maybe_unused)
1618 {
1619 	int ret;
1620 
1621 	printf("\n ");
1622 	ret = printf(" %*s %s%*s%s %*s  %*s  %-*s",
1623 		     PRINT_PID_WIDTH, "PID",
1624 
1625 		     kwork->use_bpf ? " " : "",
1626 		     kwork->use_bpf ? PRINT_PID_WIDTH : 0,
1627 		     kwork->use_bpf ? "SPID" : "",
1628 		     kwork->use_bpf ? " " : "",
1629 
1630 		     PRINT_CPU_USAGE_WIDTH, "%CPU",
1631 		     PRINT_RUNTIME_HEADER_WIDTH + RPINT_DECIMAL_WIDTH, "RUNTIME",
1632 		     PRINT_TASK_NAME_WIDTH, "COMMAND");
1633 	printf("\n ");
1634 	print_separator(ret);
1635 }
1636 
1637 static int top_print_work(struct perf_kwork *kwork __maybe_unused, struct kwork_work *work)
1638 {
1639 	int ret = 0;
1640 
1641 	printf(" ");
1642 
1643 	/*
1644 	 * pid
1645 	 */
1646 	ret += printf(" %*" PRIu64 " ", PRINT_PID_WIDTH, work->id);
1647 
1648 	/*
1649 	 * tgid
1650 	 */
1651 	if (kwork->use_bpf)
1652 		ret += printf(" %*d ", PRINT_PID_WIDTH, work->tgid);
1653 
1654 	/*
1655 	 * cpu usage
1656 	 */
1657 	ret += printf(" %*.*f ",
1658 		      PRINT_CPU_USAGE_WIDTH, PRINT_CPU_USAGE_DECIMAL_WIDTH,
1659 		      (double)work->cpu_usage / 100);
1660 
1661 	/*
1662 	 * total runtime
1663 	 */
1664 	ret += printf(" %*.*f ms ",
1665 		      PRINT_RUNTIME_WIDTH + RPINT_DECIMAL_WIDTH, RPINT_DECIMAL_WIDTH,
1666 		      (double)work->total_runtime / NSEC_PER_MSEC);
1667 
1668 	/*
1669 	 * command
1670 	 */
1671 	if (kwork->use_bpf)
1672 		ret += printf(" %s%s%s",
1673 			      work->is_kthread ? "[" : "",
1674 			      work->name,
1675 			      work->is_kthread ? "]" : "");
1676 	else
1677 		ret += printf(" %-*s", PRINT_TASK_NAME_WIDTH, work->name);
1678 
1679 	printf("\n");
1680 	return ret;
1681 }
1682 
1683 static void work_sort(struct perf_kwork *kwork,
1684 		      struct kwork_class *class, struct rb_root_cached *root)
1685 {
1686 	struct rb_node *node;
1687 	struct kwork_work *data;
1688 
1689 	pr_debug("Sorting %s ...\n", class->name);
1690 	for (;;) {
1691 		node = rb_first_cached(root);
1692 		if (!node)
1693 			break;
1694 
1695 		rb_erase_cached(node, root);
1696 		data = rb_entry(node, struct kwork_work, node);
1697 		work_insert(&kwork->sorted_work_root,
1698 			       data, &kwork->sort_list);
1699 	}
1700 }
1701 
1702 static void perf_kwork__sort(struct perf_kwork *kwork)
1703 {
1704 	struct kwork_class *class;
1705 
1706 	list_for_each_entry(class, &kwork->class_list, list)
1707 		work_sort(kwork, class, &class->work_root);
1708 }
1709 
1710 static int perf_kwork__check_config(struct perf_kwork *kwork,
1711 				    struct perf_session *session)
1712 {
1713 	int ret;
1714 	struct evsel *evsel;
1715 	struct kwork_class *class;
1716 
1717 	static struct trace_kwork_handler report_ops = {
1718 		.entry_event = report_entry_event,
1719 		.exit_event  = report_exit_event,
1720 	};
1721 	static struct trace_kwork_handler latency_ops = {
1722 		.raise_event = latency_raise_event,
1723 		.entry_event = latency_entry_event,
1724 	};
1725 	static struct trace_kwork_handler timehist_ops = {
1726 		.raise_event = timehist_raise_event,
1727 		.entry_event = timehist_entry_event,
1728 		.exit_event  = timehist_exit_event,
1729 	};
1730 	static struct trace_kwork_handler top_ops = {
1731 		.entry_event        = timehist_entry_event,
1732 		.exit_event         = top_exit_event,
1733 		.sched_switch_event = top_sched_switch_event,
1734 	};
1735 
1736 	switch (kwork->report) {
1737 	case KWORK_REPORT_RUNTIME:
1738 		kwork->tp_handler = &report_ops;
1739 		break;
1740 	case KWORK_REPORT_LATENCY:
1741 		kwork->tp_handler = &latency_ops;
1742 		break;
1743 	case KWORK_REPORT_TIMEHIST:
1744 		kwork->tp_handler = &timehist_ops;
1745 		break;
1746 	case KWORK_REPORT_TOP:
1747 		kwork->tp_handler = &top_ops;
1748 		break;
1749 	default:
1750 		pr_debug("Invalid report type %d\n", kwork->report);
1751 		return -1;
1752 	}
1753 
1754 	list_for_each_entry(class, &kwork->class_list, list)
1755 		if ((class->class_init != NULL) &&
1756 		    (class->class_init(class, session) != 0))
1757 			return -1;
1758 
1759 	if (kwork->cpu_list != NULL) {
1760 		ret = perf_session__cpu_bitmap(session,
1761 					       kwork->cpu_list,
1762 					       kwork->cpu_bitmap);
1763 		if (ret < 0) {
1764 			pr_err("Invalid cpu bitmap\n");
1765 			return -1;
1766 		}
1767 	}
1768 
1769 	if (kwork->time_str != NULL) {
1770 		ret = perf_time__parse_str(&kwork->ptime, kwork->time_str);
1771 		if (ret != 0) {
1772 			pr_err("Invalid time span\n");
1773 			return -1;
1774 		}
1775 	}
1776 
1777 	list_for_each_entry(evsel, &session->evlist->core.entries, core.node) {
1778 		if (kwork->show_callchain && !evsel__has_callchain(evsel)) {
1779 			pr_debug("Samples do not have callchains\n");
1780 			kwork->show_callchain = 0;
1781 			symbol_conf.use_callchain = 0;
1782 		}
1783 	}
1784 
1785 	return 0;
1786 }
1787 
1788 static int perf_kwork__read_events(struct perf_kwork *kwork)
1789 {
1790 	int ret = -1;
1791 	struct perf_session *session = NULL;
1792 
1793 	struct perf_data data = {
1794 		.path  = input_name,
1795 		.mode  = PERF_DATA_MODE_READ,
1796 		.force = kwork->force,
1797 	};
1798 
1799 	session = perf_session__new(&data, &kwork->tool);
1800 	if (IS_ERR(session)) {
1801 		pr_debug("Error creating perf session\n");
1802 		return PTR_ERR(session);
1803 	}
1804 
1805 	symbol__init(&session->header.env);
1806 
1807 	if (perf_kwork__check_config(kwork, session) != 0)
1808 		goto out_delete;
1809 
1810 	if (session->tevent.pevent &&
1811 	    tep_set_function_resolver(session->tevent.pevent,
1812 				      machine__resolve_kernel_addr,
1813 				      &session->machines.host) < 0) {
1814 		pr_err("Failed to set libtraceevent function resolver\n");
1815 		goto out_delete;
1816 	}
1817 
1818 	if (kwork->report == KWORK_REPORT_TIMEHIST)
1819 		timehist_print_header();
1820 
1821 	ret = perf_session__process_events(session);
1822 	if (ret) {
1823 		pr_debug("Failed to process events, error %d\n", ret);
1824 		goto out_delete;
1825 	}
1826 
1827 	kwork->nr_events      = session->evlist->stats.nr_events[0];
1828 	kwork->nr_lost_events = session->evlist->stats.total_lost;
1829 	kwork->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST];
1830 
1831 out_delete:
1832 	perf_session__delete(session);
1833 	return ret;
1834 }
1835 
1836 static void process_skipped_events(struct perf_kwork *kwork,
1837 				   struct kwork_work *work)
1838 {
1839 	int i;
1840 	unsigned long long count;
1841 
1842 	for (i = 0; i < KWORK_TRACE_MAX; i++) {
1843 		count = nr_list_entry(&work->atom_list[i]);
1844 		kwork->nr_skipped_events[i] += count;
1845 		kwork->nr_skipped_events[KWORK_TRACE_MAX] += count;
1846 	}
1847 }
1848 
1849 struct kwork_work *perf_kwork_add_work(struct perf_kwork *kwork,
1850 				       struct kwork_class *class,
1851 				       struct kwork_work *key)
1852 {
1853 	struct kwork_work *work = NULL;
1854 
1855 	work = work_new(key);
1856 	if (work == NULL)
1857 		return NULL;
1858 
1859 	work_insert(&class->work_root, work, &kwork->cmp_id);
1860 	return work;
1861 }
1862 
1863 static void sig_handler(int sig)
1864 {
1865 	/*
1866 	 * Simply capture termination signal so that
1867 	 * the program can continue after pause returns
1868 	 */
1869 	pr_debug("Capture signal %d\n", sig);
1870 }
1871 
1872 static int perf_kwork__report_bpf(struct perf_kwork *kwork)
1873 {
1874 	int ret;
1875 
1876 	signal(SIGINT, sig_handler);
1877 	signal(SIGTERM, sig_handler);
1878 
1879 	ret = perf_kwork__trace_prepare_bpf(kwork);
1880 	if (ret)
1881 		return -1;
1882 
1883 	printf("Starting trace, Hit <Ctrl+C> to stop and report\n");
1884 
1885 	perf_kwork__trace_start();
1886 
1887 	/*
1888 	 * a simple pause, wait here for stop signal
1889 	 */
1890 	pause();
1891 
1892 	perf_kwork__trace_finish();
1893 
1894 	perf_kwork__report_read_bpf(kwork);
1895 
1896 	perf_kwork__report_cleanup_bpf();
1897 
1898 	return 0;
1899 }
1900 
1901 static int perf_kwork__report(struct perf_kwork *kwork)
1902 {
1903 	int ret;
1904 	struct rb_node *next;
1905 	struct kwork_work *work;
1906 
1907 	if (kwork->use_bpf)
1908 		ret = perf_kwork__report_bpf(kwork);
1909 	else
1910 		ret = perf_kwork__read_events(kwork);
1911 
1912 	if (ret != 0)
1913 		return -1;
1914 
1915 	perf_kwork__sort(kwork);
1916 
1917 	setup_pager();
1918 
1919 	ret = report_print_header(kwork);
1920 	next = rb_first_cached(&kwork->sorted_work_root);
1921 	while (next) {
1922 		work = rb_entry(next, struct kwork_work, node);
1923 		process_skipped_events(kwork, work);
1924 
1925 		if (work->nr_atoms != 0) {
1926 			report_print_work(kwork, work);
1927 			if (kwork->summary) {
1928 				kwork->all_runtime += work->total_runtime;
1929 				kwork->all_count += work->nr_atoms;
1930 			}
1931 		}
1932 		next = rb_next(next);
1933 	}
1934 	print_separator(ret);
1935 
1936 	if (kwork->summary) {
1937 		print_summary(kwork);
1938 		print_separator(ret);
1939 	}
1940 
1941 	print_bad_events(kwork);
1942 	print_skipped_events(kwork);
1943 	printf("\n");
1944 
1945 	return 0;
1946 }
1947 
1948 typedef int (*tracepoint_handler)(struct perf_tool *tool,
1949 				  struct evsel *evsel,
1950 				  struct perf_sample *sample,
1951 				  struct machine *machine);
1952 
1953 static int perf_kwork__process_tracepoint_sample(struct perf_tool *tool,
1954 						 union perf_event *event __maybe_unused,
1955 						 struct perf_sample *sample,
1956 						 struct evsel *evsel,
1957 						 struct machine *machine)
1958 {
1959 	int err = 0;
1960 
1961 	if (evsel->handler != NULL) {
1962 		tracepoint_handler f = evsel->handler;
1963 
1964 		err = f(tool, evsel, sample, machine);
1965 	}
1966 
1967 	return err;
1968 }
1969 
1970 static int perf_kwork__timehist(struct perf_kwork *kwork)
1971 {
1972 	/*
1973 	 * event handlers for timehist option
1974 	 */
1975 	kwork->tool.comm	 = perf_event__process_comm;
1976 	kwork->tool.exit	 = perf_event__process_exit;
1977 	kwork->tool.fork	 = perf_event__process_fork;
1978 	kwork->tool.attr	 = perf_event__process_attr;
1979 	kwork->tool.tracing_data = perf_event__process_tracing_data;
1980 	kwork->tool.build_id	 = perf_event__process_build_id;
1981 	kwork->tool.ordered_events = true;
1982 	kwork->tool.ordering_requires_timestamps = true;
1983 	symbol_conf.use_callchain = kwork->show_callchain;
1984 
1985 	if (symbol__validate_sym_arguments()) {
1986 		pr_err("Failed to validate sym arguments\n");
1987 		return -1;
1988 	}
1989 
1990 	setup_pager();
1991 
1992 	return perf_kwork__read_events(kwork);
1993 }
1994 
1995 static void top_calc_total_runtime(struct perf_kwork *kwork)
1996 {
1997 	struct kwork_class *class;
1998 	struct kwork_work *work;
1999 	struct rb_node *next;
2000 	struct kwork_top_stat *stat = &kwork->top_stat;
2001 
2002 	class = get_kwork_class(kwork, KWORK_CLASS_SCHED);
2003 	if (!class)
2004 		return;
2005 
2006 	next = rb_first_cached(&class->work_root);
2007 	while (next) {
2008 		work = rb_entry(next, struct kwork_work, node);
2009 		BUG_ON(work->cpu >= MAX_NR_CPUS);
2010 		stat->cpus_runtime[work->cpu].total += work->total_runtime;
2011 		stat->cpus_runtime[MAX_NR_CPUS].total += work->total_runtime;
2012 		next = rb_next(next);
2013 	}
2014 }
2015 
2016 static void top_calc_idle_time(struct perf_kwork *kwork,
2017 				struct kwork_work *work)
2018 {
2019 	struct kwork_top_stat *stat = &kwork->top_stat;
2020 
2021 	if (work->id == 0) {
2022 		stat->cpus_runtime[work->cpu].idle += work->total_runtime;
2023 		stat->cpus_runtime[MAX_NR_CPUS].idle += work->total_runtime;
2024 	}
2025 }
2026 
2027 static void top_calc_irq_runtime(struct perf_kwork *kwork,
2028 				 enum kwork_class_type type,
2029 				 struct kwork_work *work)
2030 {
2031 	struct kwork_top_stat *stat = &kwork->top_stat;
2032 
2033 	if (type == KWORK_CLASS_IRQ) {
2034 		stat->cpus_runtime[work->cpu].irq += work->total_runtime;
2035 		stat->cpus_runtime[MAX_NR_CPUS].irq += work->total_runtime;
2036 	} else if (type == KWORK_CLASS_SOFTIRQ) {
2037 		stat->cpus_runtime[work->cpu].softirq += work->total_runtime;
2038 		stat->cpus_runtime[MAX_NR_CPUS].softirq += work->total_runtime;
2039 	}
2040 }
2041 
2042 static void top_subtract_irq_runtime(struct perf_kwork *kwork,
2043 				     struct kwork_work *work)
2044 {
2045 	struct kwork_class *class;
2046 	struct kwork_work *data;
2047 	unsigned int i;
2048 	int irq_class_list[] = {KWORK_CLASS_IRQ, KWORK_CLASS_SOFTIRQ};
2049 
2050 	for (i = 0; i < ARRAY_SIZE(irq_class_list); i++) {
2051 		class = get_kwork_class(kwork, irq_class_list[i]);
2052 		if (!class)
2053 			continue;
2054 
2055 		data = find_work_by_id(&class->work_root,
2056 				       work->id, work->cpu);
2057 		if (!data)
2058 			continue;
2059 
2060 		if (work->total_runtime > data->total_runtime) {
2061 			work->total_runtime -= data->total_runtime;
2062 			top_calc_irq_runtime(kwork, irq_class_list[i], data);
2063 		}
2064 	}
2065 }
2066 
2067 static void top_calc_cpu_usage(struct perf_kwork *kwork)
2068 {
2069 	struct kwork_class *class;
2070 	struct kwork_work *work;
2071 	struct rb_node *next;
2072 	struct kwork_top_stat *stat = &kwork->top_stat;
2073 
2074 	class = get_kwork_class(kwork, KWORK_CLASS_SCHED);
2075 	if (!class)
2076 		return;
2077 
2078 	next = rb_first_cached(&class->work_root);
2079 	while (next) {
2080 		work = rb_entry(next, struct kwork_work, node);
2081 
2082 		if (work->total_runtime == 0)
2083 			goto next;
2084 
2085 		__set_bit(work->cpu, stat->all_cpus_bitmap);
2086 
2087 		top_subtract_irq_runtime(kwork, work);
2088 
2089 		work->cpu_usage = work->total_runtime * 10000 /
2090 			stat->cpus_runtime[work->cpu].total;
2091 
2092 		top_calc_idle_time(kwork, work);
2093 next:
2094 		next = rb_next(next);
2095 	}
2096 }
2097 
2098 static void top_calc_load_runtime(struct perf_kwork *kwork,
2099 				  struct kwork_work *work)
2100 {
2101 	struct kwork_top_stat *stat = &kwork->top_stat;
2102 
2103 	if (work->id != 0) {
2104 		stat->cpus_runtime[work->cpu].load += work->total_runtime;
2105 		stat->cpus_runtime[MAX_NR_CPUS].load += work->total_runtime;
2106 	}
2107 }
2108 
2109 static void top_merge_tasks(struct perf_kwork *kwork)
2110 {
2111 	struct kwork_work *merged_work, *data;
2112 	struct kwork_class *class;
2113 	struct rb_node *node;
2114 	int cpu;
2115 	struct rb_root_cached merged_root = RB_ROOT_CACHED;
2116 
2117 	class = get_kwork_class(kwork, KWORK_CLASS_SCHED);
2118 	if (!class)
2119 		return;
2120 
2121 	for (;;) {
2122 		node = rb_first_cached(&class->work_root);
2123 		if (!node)
2124 			break;
2125 
2126 		rb_erase_cached(node, &class->work_root);
2127 		data = rb_entry(node, struct kwork_work, node);
2128 
2129 		if (!profile_name_match(kwork, data))
2130 			continue;
2131 
2132 		cpu = data->cpu;
2133 		merged_work = find_work_by_id(&merged_root, data->id,
2134 					      data->id == 0 ? cpu : -1);
2135 		if (!merged_work) {
2136 			work_insert(&merged_root, data, &kwork->cmp_id);
2137 		} else {
2138 			merged_work->total_runtime += data->total_runtime;
2139 			merged_work->cpu_usage += data->cpu_usage;
2140 		}
2141 
2142 		top_calc_load_runtime(kwork, data);
2143 	}
2144 
2145 	work_sort(kwork, class, &merged_root);
2146 }
2147 
2148 static void perf_kwork__top_report(struct perf_kwork *kwork)
2149 {
2150 	struct kwork_work *work;
2151 	struct rb_node *next;
2152 
2153 	printf("\n");
2154 
2155 	top_print_cpu_usage(kwork);
2156 	top_print_header(kwork);
2157 	next = rb_first_cached(&kwork->sorted_work_root);
2158 	while (next) {
2159 		work = rb_entry(next, struct kwork_work, node);
2160 		process_skipped_events(kwork, work);
2161 
2162 		if (work->total_runtime == 0)
2163 			goto next;
2164 
2165 		top_print_work(kwork, work);
2166 
2167 next:
2168 		next = rb_next(next);
2169 	}
2170 
2171 	printf("\n");
2172 }
2173 
2174 static int perf_kwork__top_bpf(struct perf_kwork *kwork)
2175 {
2176 	int ret;
2177 
2178 	signal(SIGINT, sig_handler);
2179 	signal(SIGTERM, sig_handler);
2180 
2181 	ret = perf_kwork__top_prepare_bpf(kwork);
2182 	if (ret)
2183 		return -1;
2184 
2185 	printf("Starting trace, Hit <Ctrl+C> to stop and report\n");
2186 
2187 	perf_kwork__top_start();
2188 
2189 	/*
2190 	 * a simple pause, wait here for stop signal
2191 	 */
2192 	pause();
2193 
2194 	perf_kwork__top_finish();
2195 
2196 	perf_kwork__top_read_bpf(kwork);
2197 
2198 	perf_kwork__top_cleanup_bpf();
2199 
2200 	return 0;
2201 
2202 }
2203 
2204 static int perf_kwork__top(struct perf_kwork *kwork)
2205 {
2206 	struct __top_cpus_runtime *cpus_runtime;
2207 	int ret = 0;
2208 
2209 	cpus_runtime = zalloc(sizeof(struct __top_cpus_runtime) * (MAX_NR_CPUS + 1));
2210 	if (!cpus_runtime)
2211 		return -1;
2212 
2213 	kwork->top_stat.cpus_runtime = cpus_runtime;
2214 	bitmap_zero(kwork->top_stat.all_cpus_bitmap, MAX_NR_CPUS);
2215 
2216 	if (kwork->use_bpf)
2217 		ret = perf_kwork__top_bpf(kwork);
2218 	else
2219 		ret = perf_kwork__read_events(kwork);
2220 
2221 	if (ret)
2222 		goto out;
2223 
2224 	top_calc_total_runtime(kwork);
2225 	top_calc_cpu_usage(kwork);
2226 	top_merge_tasks(kwork);
2227 
2228 	setup_pager();
2229 
2230 	perf_kwork__top_report(kwork);
2231 
2232 out:
2233 	zfree(&kwork->top_stat.cpus_runtime);
2234 	return ret;
2235 }
2236 
2237 static void setup_event_list(struct perf_kwork *kwork,
2238 			     const struct option *options,
2239 			     const char * const usage_msg[])
2240 {
2241 	int i;
2242 	struct kwork_class *class;
2243 	char *tmp, *tok, *str;
2244 
2245 	/*
2246 	 * set default events list if not specified
2247 	 */
2248 	if (kwork->event_list_str == NULL)
2249 		kwork->event_list_str = "irq, softirq, workqueue";
2250 
2251 	str = strdup(kwork->event_list_str);
2252 	for (tok = strtok_r(str, ", ", &tmp);
2253 	     tok; tok = strtok_r(NULL, ", ", &tmp)) {
2254 		for (i = 0; i < KWORK_CLASS_MAX; i++) {
2255 			class = kwork_class_supported_list[i];
2256 			if (strcmp(tok, class->name) == 0) {
2257 				list_add_tail(&class->list, &kwork->class_list);
2258 				break;
2259 			}
2260 		}
2261 		if (i == KWORK_CLASS_MAX) {
2262 			usage_with_options_msg(usage_msg, options,
2263 					       "Unknown --event key: `%s'", tok);
2264 		}
2265 	}
2266 	free(str);
2267 
2268 	pr_debug("Config event list:");
2269 	list_for_each_entry(class, &kwork->class_list, list)
2270 		pr_debug(" %s", class->name);
2271 	pr_debug("\n");
2272 }
2273 
2274 static int perf_kwork__record(struct perf_kwork *kwork,
2275 			      int argc, const char **argv)
2276 {
2277 	const char **rec_argv;
2278 	unsigned int rec_argc, i, j;
2279 	struct kwork_class *class;
2280 
2281 	const char *const record_args[] = {
2282 		"record",
2283 		"-a",
2284 		"-R",
2285 		"-m", "1024",
2286 		"-c", "1",
2287 	};
2288 
2289 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
2290 
2291 	list_for_each_entry(class, &kwork->class_list, list)
2292 		rec_argc += 2 * class->nr_tracepoints;
2293 
2294 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
2295 	if (rec_argv == NULL)
2296 		return -ENOMEM;
2297 
2298 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
2299 		rec_argv[i] = strdup(record_args[i]);
2300 
2301 	list_for_each_entry(class, &kwork->class_list, list) {
2302 		for (j = 0; j < class->nr_tracepoints; j++) {
2303 			rec_argv[i++] = strdup("-e");
2304 			rec_argv[i++] = strdup(class->tp_handlers[j].name);
2305 		}
2306 	}
2307 
2308 	for (j = 1; j < (unsigned int)argc; j++, i++)
2309 		rec_argv[i] = argv[j];
2310 
2311 	BUG_ON(i != rec_argc);
2312 
2313 	pr_debug("record comm: ");
2314 	for (j = 0; j < rec_argc; j++)
2315 		pr_debug("%s ", rec_argv[j]);
2316 	pr_debug("\n");
2317 
2318 	return cmd_record(i, rec_argv);
2319 }
2320 
2321 int cmd_kwork(int argc, const char **argv)
2322 {
2323 	static struct perf_kwork kwork = {
2324 		.class_list          = LIST_HEAD_INIT(kwork.class_list),
2325 		.tool = {
2326 			.mmap		= perf_event__process_mmap,
2327 			.mmap2		= perf_event__process_mmap2,
2328 			.sample		= perf_kwork__process_tracepoint_sample,
2329 			.ordered_events = true,
2330 		},
2331 		.atom_page_list      = LIST_HEAD_INIT(kwork.atom_page_list),
2332 		.sort_list           = LIST_HEAD_INIT(kwork.sort_list),
2333 		.cmp_id              = LIST_HEAD_INIT(kwork.cmp_id),
2334 		.sorted_work_root    = RB_ROOT_CACHED,
2335 		.tp_handler          = NULL,
2336 		.profile_name        = NULL,
2337 		.cpu_list            = NULL,
2338 		.time_str            = NULL,
2339 		.force               = false,
2340 		.event_list_str      = NULL,
2341 		.summary             = false,
2342 		.sort_order          = NULL,
2343 		.show_callchain      = false,
2344 		.max_stack           = 5,
2345 		.timestart           = 0,
2346 		.timeend             = 0,
2347 		.nr_events           = 0,
2348 		.nr_lost_chunks      = 0,
2349 		.nr_lost_events      = 0,
2350 		.all_runtime         = 0,
2351 		.all_count           = 0,
2352 		.nr_skipped_events   = { 0 },
2353 	};
2354 	static const char default_report_sort_order[] = "runtime, max, count";
2355 	static const char default_latency_sort_order[] = "avg, max, count";
2356 	static const char default_top_sort_order[] = "rate, runtime";
2357 	const struct option kwork_options[] = {
2358 	OPT_INCR('v', "verbose", &verbose,
2359 		 "be more verbose (show symbol address, etc)"),
2360 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
2361 		    "dump raw trace in ASCII"),
2362 	OPT_STRING('k', "kwork", &kwork.event_list_str, "kwork",
2363 		   "list of kwork to profile (irq, softirq, workqueue, sched, etc)"),
2364 	OPT_BOOLEAN('f', "force", &kwork.force, "don't complain, do it"),
2365 	OPT_END()
2366 	};
2367 	const struct option report_options[] = {
2368 	OPT_STRING('s', "sort", &kwork.sort_order, "key[,key2...]",
2369 		   "sort by key(s): runtime, max, count"),
2370 	OPT_STRING('C', "cpu", &kwork.cpu_list, "cpu",
2371 		   "list of cpus to profile"),
2372 	OPT_STRING('n', "name", &kwork.profile_name, "name",
2373 		   "event name to profile"),
2374 	OPT_STRING(0, "time", &kwork.time_str, "str",
2375 		   "Time span for analysis (start,stop)"),
2376 	OPT_STRING('i', "input", &input_name, "file",
2377 		   "input file name"),
2378 	OPT_BOOLEAN('S', "with-summary", &kwork.summary,
2379 		    "Show summary with statistics"),
2380 #ifdef HAVE_BPF_SKEL
2381 	OPT_BOOLEAN('b', "use-bpf", &kwork.use_bpf,
2382 		    "Use BPF to measure kwork runtime"),
2383 #endif
2384 	OPT_PARENT(kwork_options)
2385 	};
2386 	const struct option latency_options[] = {
2387 	OPT_STRING('s', "sort", &kwork.sort_order, "key[,key2...]",
2388 		   "sort by key(s): avg, max, count"),
2389 	OPT_STRING('C', "cpu", &kwork.cpu_list, "cpu",
2390 		   "list of cpus to profile"),
2391 	OPT_STRING('n', "name", &kwork.profile_name, "name",
2392 		   "event name to profile"),
2393 	OPT_STRING(0, "time", &kwork.time_str, "str",
2394 		   "Time span for analysis (start,stop)"),
2395 	OPT_STRING('i', "input", &input_name, "file",
2396 		   "input file name"),
2397 #ifdef HAVE_BPF_SKEL
2398 	OPT_BOOLEAN('b', "use-bpf", &kwork.use_bpf,
2399 		    "Use BPF to measure kwork latency"),
2400 #endif
2401 	OPT_PARENT(kwork_options)
2402 	};
2403 	const struct option timehist_options[] = {
2404 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
2405 		   "file", "vmlinux pathname"),
2406 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
2407 		   "file", "kallsyms pathname"),
2408 	OPT_BOOLEAN('g', "call-graph", &kwork.show_callchain,
2409 		    "Display call chains if present"),
2410 	OPT_UINTEGER(0, "max-stack", &kwork.max_stack,
2411 		   "Maximum number of functions to display backtrace."),
2412 	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
2413 		    "Look for files with symbols relative to this directory"),
2414 	OPT_STRING(0, "time", &kwork.time_str, "str",
2415 		   "Time span for analysis (start,stop)"),
2416 	OPT_STRING('C', "cpu", &kwork.cpu_list, "cpu",
2417 		   "list of cpus to profile"),
2418 	OPT_STRING('n', "name", &kwork.profile_name, "name",
2419 		   "event name to profile"),
2420 	OPT_STRING('i', "input", &input_name, "file",
2421 		   "input file name"),
2422 	OPT_PARENT(kwork_options)
2423 	};
2424 	const struct option top_options[] = {
2425 	OPT_STRING('s', "sort", &kwork.sort_order, "key[,key2...]",
2426 		   "sort by key(s): rate, runtime, tid"),
2427 	OPT_STRING('C', "cpu", &kwork.cpu_list, "cpu",
2428 		   "list of cpus to profile"),
2429 	OPT_STRING('n', "name", &kwork.profile_name, "name",
2430 		   "event name to profile"),
2431 	OPT_STRING(0, "time", &kwork.time_str, "str",
2432 		   "Time span for analysis (start,stop)"),
2433 	OPT_STRING('i', "input", &input_name, "file",
2434 		   "input file name"),
2435 #ifdef HAVE_BPF_SKEL
2436 	OPT_BOOLEAN('b', "use-bpf", &kwork.use_bpf,
2437 		    "Use BPF to measure task cpu usage"),
2438 #endif
2439 	OPT_PARENT(kwork_options)
2440 	};
2441 	const char *kwork_usage[] = {
2442 		NULL,
2443 		NULL
2444 	};
2445 	const char * const report_usage[] = {
2446 		"perf kwork report [<options>]",
2447 		NULL
2448 	};
2449 	const char * const latency_usage[] = {
2450 		"perf kwork latency [<options>]",
2451 		NULL
2452 	};
2453 	const char * const timehist_usage[] = {
2454 		"perf kwork timehist [<options>]",
2455 		NULL
2456 	};
2457 	const char * const top_usage[] = {
2458 		"perf kwork top [<options>]",
2459 		NULL
2460 	};
2461 	const char *const kwork_subcommands[] = {
2462 		"record", "report", "latency", "timehist", "top", NULL
2463 	};
2464 
2465 	argc = parse_options_subcommand(argc, argv, kwork_options,
2466 					kwork_subcommands, kwork_usage,
2467 					PARSE_OPT_STOP_AT_NON_OPTION);
2468 	if (!argc)
2469 		usage_with_options(kwork_usage, kwork_options);
2470 
2471 	sort_dimension__add(&kwork, "id", &kwork.cmp_id);
2472 
2473 	if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
2474 		setup_event_list(&kwork, kwork_options, kwork_usage);
2475 		return perf_kwork__record(&kwork, argc, argv);
2476 	} else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
2477 		kwork.sort_order = default_report_sort_order;
2478 		if (argc > 1) {
2479 			argc = parse_options(argc, argv, report_options, report_usage, 0);
2480 			if (argc)
2481 				usage_with_options(report_usage, report_options);
2482 		}
2483 		kwork.report = KWORK_REPORT_RUNTIME;
2484 		setup_sorting(&kwork, report_options, report_usage);
2485 		setup_event_list(&kwork, kwork_options, kwork_usage);
2486 		return perf_kwork__report(&kwork);
2487 	} else if (strlen(argv[0]) > 2 && strstarts("latency", argv[0])) {
2488 		kwork.sort_order = default_latency_sort_order;
2489 		if (argc > 1) {
2490 			argc = parse_options(argc, argv, latency_options, latency_usage, 0);
2491 			if (argc)
2492 				usage_with_options(latency_usage, latency_options);
2493 		}
2494 		kwork.report = KWORK_REPORT_LATENCY;
2495 		setup_sorting(&kwork, latency_options, latency_usage);
2496 		setup_event_list(&kwork, kwork_options, kwork_usage);
2497 		return perf_kwork__report(&kwork);
2498 	} else if (strlen(argv[0]) > 2 && strstarts("timehist", argv[0])) {
2499 		if (argc > 1) {
2500 			argc = parse_options(argc, argv, timehist_options, timehist_usage, 0);
2501 			if (argc)
2502 				usage_with_options(timehist_usage, timehist_options);
2503 		}
2504 		kwork.report = KWORK_REPORT_TIMEHIST;
2505 		setup_event_list(&kwork, kwork_options, kwork_usage);
2506 		return perf_kwork__timehist(&kwork);
2507 	} else if (strlen(argv[0]) > 2 && strstarts("top", argv[0])) {
2508 		kwork.sort_order = default_top_sort_order;
2509 		if (argc > 1) {
2510 			argc = parse_options(argc, argv, top_options, top_usage, 0);
2511 			if (argc)
2512 				usage_with_options(top_usage, top_options);
2513 		}
2514 		kwork.report = KWORK_REPORT_TOP;
2515 		if (!kwork.event_list_str)
2516 			kwork.event_list_str = "sched, irq, softirq";
2517 		setup_event_list(&kwork, kwork_options, kwork_usage);
2518 		setup_sorting(&kwork, top_options, top_usage);
2519 		return perf_kwork__top(&kwork);
2520 	} else
2521 		usage_with_options(kwork_usage, kwork_options);
2522 
2523 	return 0;
2524 }
2525