1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * builtin-timechart.c - make an svg timechart of system activity
4  *
5  * (C) Copyright 2009 Intel Corporation
6  *
7  * Authors:
8  *     Arjan van de Ven <arjan@linux.intel.com>
9  */
10 
11 #include <errno.h>
12 #include <inttypes.h>
13 
14 #include "builtin.h"
15 #include "util/color.h"
16 #include <linux/list.h>
17 #include "util/evlist.h" // for struct evsel_str_handler
18 #include "util/evsel.h"
19 #include <linux/kernel.h>
20 #include <linux/rbtree.h>
21 #include <linux/time64.h>
22 #include <linux/zalloc.h>
23 #include "util/symbol.h"
24 #include "util/thread.h"
25 #include "util/callchain.h"
26 
27 #include "perf.h"
28 #include "util/header.h"
29 #include <subcmd/pager.h>
30 #include <subcmd/parse-options.h>
31 #include "util/parse-events.h"
32 #include "util/event.h"
33 #include "util/session.h"
34 #include "util/svghelper.h"
35 #include "util/tool.h"
36 #include "util/data.h"
37 #include "util/debug.h"
38 
39 #ifdef LACKS_OPEN_MEMSTREAM_PROTOTYPE
40 FILE *open_memstream(char **ptr, size_t *sizeloc);
41 #endif
42 
43 #define SUPPORT_OLD_POWER_EVENTS 1
44 #define PWR_EVENT_EXIT -1
45 
46 struct per_pid;
47 struct power_event;
48 struct wake_event;
49 
50 struct timechart {
51 	struct perf_tool	tool;
52 	struct per_pid		*all_data;
53 	struct power_event	*power_events;
54 	struct wake_event	*wake_events;
55 	int			proc_num;
56 	unsigned int		numcpus;
57 	u64			min_freq,	/* Lowest CPU frequency seen */
58 				max_freq,	/* Highest CPU frequency seen */
59 				turbo_frequency,
60 				first_time, last_time;
61 	bool			power_only,
62 				tasks_only,
63 				with_backtrace,
64 				topology;
65 	bool			force;
66 	/* IO related settings */
67 	bool			io_only,
68 				skip_eagain;
69 	u64			io_events;
70 	u64			min_time,
71 				merge_dist;
72 };
73 
74 struct per_pidcomm;
75 struct cpu_sample;
76 struct io_sample;
77 
78 /*
79  * Datastructure layout:
80  * We keep an list of "pid"s, matching the kernels notion of a task struct.
81  * Each "pid" entry, has a list of "comm"s.
82  *	this is because we want to track different programs different, while
83  *	exec will reuse the original pid (by design).
84  * Each comm has a list of samples that will be used to draw
85  * final graph.
86  */
87 
88 struct per_pid {
89 	struct per_pid *next;
90 
91 	int		pid;
92 	int		ppid;
93 
94 	u64		start_time;
95 	u64		end_time;
96 	u64		total_time;
97 	u64		total_bytes;
98 	int		display;
99 
100 	struct per_pidcomm *all;
101 	struct per_pidcomm *current;
102 };
103 
104 
105 struct per_pidcomm {
106 	struct per_pidcomm *next;
107 
108 	u64		start_time;
109 	u64		end_time;
110 	u64		total_time;
111 	u64		max_bytes;
112 	u64		total_bytes;
113 
114 	int		Y;
115 	int		display;
116 
117 	long		state;
118 	u64		state_since;
119 
120 	char		*comm;
121 
122 	struct cpu_sample *samples;
123 	struct io_sample  *io_samples;
124 };
125 
126 struct sample_wrapper {
127 	struct sample_wrapper *next;
128 
129 	u64		timestamp;
130 	unsigned char	data[0];
131 };
132 
133 #define TYPE_NONE	0
134 #define TYPE_RUNNING	1
135 #define TYPE_WAITING	2
136 #define TYPE_BLOCKED	3
137 
138 struct cpu_sample {
139 	struct cpu_sample *next;
140 
141 	u64 start_time;
142 	u64 end_time;
143 	int type;
144 	int cpu;
145 	const char *backtrace;
146 };
147 
148 enum {
149 	IOTYPE_READ,
150 	IOTYPE_WRITE,
151 	IOTYPE_SYNC,
152 	IOTYPE_TX,
153 	IOTYPE_RX,
154 	IOTYPE_POLL,
155 };
156 
157 struct io_sample {
158 	struct io_sample *next;
159 
160 	u64 start_time;
161 	u64 end_time;
162 	u64 bytes;
163 	int type;
164 	int fd;
165 	int err;
166 	int merges;
167 };
168 
169 #define CSTATE 1
170 #define PSTATE 2
171 
172 struct power_event {
173 	struct power_event *next;
174 	int type;
175 	int state;
176 	u64 start_time;
177 	u64 end_time;
178 	int cpu;
179 };
180 
181 struct wake_event {
182 	struct wake_event *next;
183 	int waker;
184 	int wakee;
185 	u64 time;
186 	const char *backtrace;
187 };
188 
189 struct process_filter {
190 	char			*name;
191 	int			pid;
192 	struct process_filter	*next;
193 };
194 
195 static struct process_filter *process_filter;
196 
197 
198 static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
199 {
200 	struct per_pid *cursor = tchart->all_data;
201 
202 	while (cursor) {
203 		if (cursor->pid == pid)
204 			return cursor;
205 		cursor = cursor->next;
206 	}
207 	cursor = zalloc(sizeof(*cursor));
208 	assert(cursor != NULL);
209 	cursor->pid = pid;
210 	cursor->next = tchart->all_data;
211 	tchart->all_data = cursor;
212 	return cursor;
213 }
214 
215 static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
216 {
217 	struct per_pid *p;
218 	struct per_pidcomm *c;
219 	p = find_create_pid(tchart, pid);
220 	c = p->all;
221 	while (c) {
222 		if (c->comm && strcmp(c->comm, comm) == 0) {
223 			p->current = c;
224 			return;
225 		}
226 		if (!c->comm) {
227 			c->comm = strdup(comm);
228 			p->current = c;
229 			return;
230 		}
231 		c = c->next;
232 	}
233 	c = zalloc(sizeof(*c));
234 	assert(c != NULL);
235 	c->comm = strdup(comm);
236 	p->current = c;
237 	c->next = p->all;
238 	p->all = c;
239 }
240 
241 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
242 {
243 	struct per_pid *p, *pp;
244 	p = find_create_pid(tchart, pid);
245 	pp = find_create_pid(tchart, ppid);
246 	p->ppid = ppid;
247 	if (pp->current && pp->current->comm && !p->current)
248 		pid_set_comm(tchart, pid, pp->current->comm);
249 
250 	p->start_time = timestamp;
251 	if (p->current && !p->current->start_time) {
252 		p->current->start_time = timestamp;
253 		p->current->state_since = timestamp;
254 	}
255 }
256 
257 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
258 {
259 	struct per_pid *p;
260 	p = find_create_pid(tchart, pid);
261 	p->end_time = timestamp;
262 	if (p->current)
263 		p->current->end_time = timestamp;
264 }
265 
266 static void pid_put_sample(struct timechart *tchart, int pid, int type,
267 			   unsigned int cpu, u64 start, u64 end,
268 			   const char *backtrace)
269 {
270 	struct per_pid *p;
271 	struct per_pidcomm *c;
272 	struct cpu_sample *sample;
273 
274 	p = find_create_pid(tchart, pid);
275 	c = p->current;
276 	if (!c) {
277 		c = zalloc(sizeof(*c));
278 		assert(c != NULL);
279 		p->current = c;
280 		c->next = p->all;
281 		p->all = c;
282 	}
283 
284 	sample = zalloc(sizeof(*sample));
285 	assert(sample != NULL);
286 	sample->start_time = start;
287 	sample->end_time = end;
288 	sample->type = type;
289 	sample->next = c->samples;
290 	sample->cpu = cpu;
291 	sample->backtrace = backtrace;
292 	c->samples = sample;
293 
294 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
295 		c->total_time += (end-start);
296 		p->total_time += (end-start);
297 	}
298 
299 	if (c->start_time == 0 || c->start_time > start)
300 		c->start_time = start;
301 	if (p->start_time == 0 || p->start_time > start)
302 		p->start_time = start;
303 }
304 
305 #define MAX_CPUS 4096
306 
307 static u64 cpus_cstate_start_times[MAX_CPUS];
308 static int cpus_cstate_state[MAX_CPUS];
309 static u64 cpus_pstate_start_times[MAX_CPUS];
310 static u64 cpus_pstate_state[MAX_CPUS];
311 
312 static int process_comm_event(struct perf_tool *tool,
313 			      union perf_event *event,
314 			      struct perf_sample *sample __maybe_unused,
315 			      struct machine *machine __maybe_unused)
316 {
317 	struct timechart *tchart = container_of(tool, struct timechart, tool);
318 	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
319 	return 0;
320 }
321 
322 static int process_fork_event(struct perf_tool *tool,
323 			      union perf_event *event,
324 			      struct perf_sample *sample __maybe_unused,
325 			      struct machine *machine __maybe_unused)
326 {
327 	struct timechart *tchart = container_of(tool, struct timechart, tool);
328 	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
329 	return 0;
330 }
331 
332 static int process_exit_event(struct perf_tool *tool,
333 			      union perf_event *event,
334 			      struct perf_sample *sample __maybe_unused,
335 			      struct machine *machine __maybe_unused)
336 {
337 	struct timechart *tchart = container_of(tool, struct timechart, tool);
338 	pid_exit(tchart, event->fork.pid, event->fork.time);
339 	return 0;
340 }
341 
342 #ifdef SUPPORT_OLD_POWER_EVENTS
343 static int use_old_power_events;
344 #endif
345 
346 static void c_state_start(int cpu, u64 timestamp, int state)
347 {
348 	cpus_cstate_start_times[cpu] = timestamp;
349 	cpus_cstate_state[cpu] = state;
350 }
351 
352 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
353 {
354 	struct power_event *pwr = zalloc(sizeof(*pwr));
355 
356 	if (!pwr)
357 		return;
358 
359 	pwr->state = cpus_cstate_state[cpu];
360 	pwr->start_time = cpus_cstate_start_times[cpu];
361 	pwr->end_time = timestamp;
362 	pwr->cpu = cpu;
363 	pwr->type = CSTATE;
364 	pwr->next = tchart->power_events;
365 
366 	tchart->power_events = pwr;
367 }
368 
369 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
370 {
371 	struct power_event *pwr;
372 
373 	if (new_freq > 8000000) /* detect invalid data */
374 		return;
375 
376 	pwr = zalloc(sizeof(*pwr));
377 	if (!pwr)
378 		return;
379 
380 	pwr->state = cpus_pstate_state[cpu];
381 	pwr->start_time = cpus_pstate_start_times[cpu];
382 	pwr->end_time = timestamp;
383 	pwr->cpu = cpu;
384 	pwr->type = PSTATE;
385 	pwr->next = tchart->power_events;
386 
387 	if (!pwr->start_time)
388 		pwr->start_time = tchart->first_time;
389 
390 	tchart->power_events = pwr;
391 
392 	cpus_pstate_state[cpu] = new_freq;
393 	cpus_pstate_start_times[cpu] = timestamp;
394 
395 	if ((u64)new_freq > tchart->max_freq)
396 		tchart->max_freq = new_freq;
397 
398 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
399 		tchart->min_freq = new_freq;
400 
401 	if (new_freq == tchart->max_freq - 1000)
402 		tchart->turbo_frequency = tchart->max_freq;
403 }
404 
405 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
406 			 int waker, int wakee, u8 flags, const char *backtrace)
407 {
408 	struct per_pid *p;
409 	struct wake_event *we = zalloc(sizeof(*we));
410 
411 	if (!we)
412 		return;
413 
414 	we->time = timestamp;
415 	we->waker = waker;
416 	we->backtrace = backtrace;
417 
418 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
419 		we->waker = -1;
420 
421 	we->wakee = wakee;
422 	we->next = tchart->wake_events;
423 	tchart->wake_events = we;
424 	p = find_create_pid(tchart, we->wakee);
425 
426 	if (p && p->current && p->current->state == TYPE_NONE) {
427 		p->current->state_since = timestamp;
428 		p->current->state = TYPE_WAITING;
429 	}
430 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
431 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
432 			       p->current->state_since, timestamp, NULL);
433 		p->current->state_since = timestamp;
434 		p->current->state = TYPE_WAITING;
435 	}
436 }
437 
438 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
439 			 int prev_pid, int next_pid, u64 prev_state,
440 			 const char *backtrace)
441 {
442 	struct per_pid *p = NULL, *prev_p;
443 
444 	prev_p = find_create_pid(tchart, prev_pid);
445 
446 	p = find_create_pid(tchart, next_pid);
447 
448 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
449 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
450 			       prev_p->current->state_since, timestamp,
451 			       backtrace);
452 	if (p && p->current) {
453 		if (p->current->state != TYPE_NONE)
454 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
455 				       p->current->state_since, timestamp,
456 				       backtrace);
457 
458 		p->current->state_since = timestamp;
459 		p->current->state = TYPE_RUNNING;
460 	}
461 
462 	if (prev_p->current) {
463 		prev_p->current->state = TYPE_NONE;
464 		prev_p->current->state_since = timestamp;
465 		if (prev_state & 2)
466 			prev_p->current->state = TYPE_BLOCKED;
467 		if (prev_state == 0)
468 			prev_p->current->state = TYPE_WAITING;
469 	}
470 }
471 
472 static const char *cat_backtrace(union perf_event *event,
473 				 struct perf_sample *sample,
474 				 struct machine *machine)
475 {
476 	struct addr_location al;
477 	unsigned int i;
478 	char *p = NULL;
479 	size_t p_len;
480 	u8 cpumode = PERF_RECORD_MISC_USER;
481 	struct addr_location tal;
482 	struct ip_callchain *chain = sample->callchain;
483 	FILE *f = open_memstream(&p, &p_len);
484 
485 	if (!f) {
486 		perror("open_memstream error");
487 		return NULL;
488 	}
489 
490 	if (!chain)
491 		goto exit;
492 
493 	if (machine__resolve(machine, &al, sample) < 0) {
494 		fprintf(stderr, "problem processing %d event, skipping it.\n",
495 			event->header.type);
496 		goto exit;
497 	}
498 
499 	for (i = 0; i < chain->nr; i++) {
500 		u64 ip;
501 
502 		if (callchain_param.order == ORDER_CALLEE)
503 			ip = chain->ips[i];
504 		else
505 			ip = chain->ips[chain->nr - i - 1];
506 
507 		if (ip >= PERF_CONTEXT_MAX) {
508 			switch (ip) {
509 			case PERF_CONTEXT_HV:
510 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
511 				break;
512 			case PERF_CONTEXT_KERNEL:
513 				cpumode = PERF_RECORD_MISC_KERNEL;
514 				break;
515 			case PERF_CONTEXT_USER:
516 				cpumode = PERF_RECORD_MISC_USER;
517 				break;
518 			default:
519 				pr_debug("invalid callchain context: "
520 					 "%"PRId64"\n", (s64) ip);
521 
522 				/*
523 				 * It seems the callchain is corrupted.
524 				 * Discard all.
525 				 */
526 				zfree(&p);
527 				goto exit_put;
528 			}
529 			continue;
530 		}
531 
532 		tal.filtered = 0;
533 		if (thread__find_symbol(al.thread, cpumode, ip, &tal))
534 			fprintf(f, "..... %016" PRIx64 " %s\n", ip, tal.sym->name);
535 		else
536 			fprintf(f, "..... %016" PRIx64 "\n", ip);
537 	}
538 exit_put:
539 	addr_location__put(&al);
540 exit:
541 	fclose(f);
542 
543 	return p;
544 }
545 
546 typedef int (*tracepoint_handler)(struct timechart *tchart,
547 				  struct evsel *evsel,
548 				  struct perf_sample *sample,
549 				  const char *backtrace);
550 
551 static int process_sample_event(struct perf_tool *tool,
552 				union perf_event *event,
553 				struct perf_sample *sample,
554 				struct evsel *evsel,
555 				struct machine *machine)
556 {
557 	struct timechart *tchart = container_of(tool, struct timechart, tool);
558 
559 	if (evsel->core.attr.sample_type & PERF_SAMPLE_TIME) {
560 		if (!tchart->first_time || tchart->first_time > sample->time)
561 			tchart->first_time = sample->time;
562 		if (tchart->last_time < sample->time)
563 			tchart->last_time = sample->time;
564 	}
565 
566 	if (evsel->handler != NULL) {
567 		tracepoint_handler f = evsel->handler;
568 		return f(tchart, evsel, sample,
569 			 cat_backtrace(event, sample, machine));
570 	}
571 
572 	return 0;
573 }
574 
575 static int
576 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
577 			struct evsel *evsel,
578 			struct perf_sample *sample,
579 			const char *backtrace __maybe_unused)
580 {
581 	u32 state = perf_evsel__intval(evsel, sample, "state");
582 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
583 
584 	if (state == (u32)PWR_EVENT_EXIT)
585 		c_state_end(tchart, cpu_id, sample->time);
586 	else
587 		c_state_start(cpu_id, sample->time, state);
588 	return 0;
589 }
590 
591 static int
592 process_sample_cpu_frequency(struct timechart *tchart,
593 			     struct evsel *evsel,
594 			     struct perf_sample *sample,
595 			     const char *backtrace __maybe_unused)
596 {
597 	u32 state = perf_evsel__intval(evsel, sample, "state");
598 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
599 
600 	p_state_change(tchart, cpu_id, sample->time, state);
601 	return 0;
602 }
603 
604 static int
605 process_sample_sched_wakeup(struct timechart *tchart,
606 			    struct evsel *evsel,
607 			    struct perf_sample *sample,
608 			    const char *backtrace)
609 {
610 	u8 flags = perf_evsel__intval(evsel, sample, "common_flags");
611 	int waker = perf_evsel__intval(evsel, sample, "common_pid");
612 	int wakee = perf_evsel__intval(evsel, sample, "pid");
613 
614 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
615 	return 0;
616 }
617 
618 static int
619 process_sample_sched_switch(struct timechart *tchart,
620 			    struct evsel *evsel,
621 			    struct perf_sample *sample,
622 			    const char *backtrace)
623 {
624 	int prev_pid = perf_evsel__intval(evsel, sample, "prev_pid");
625 	int next_pid = perf_evsel__intval(evsel, sample, "next_pid");
626 	u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
627 
628 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
629 		     prev_state, backtrace);
630 	return 0;
631 }
632 
633 #ifdef SUPPORT_OLD_POWER_EVENTS
634 static int
635 process_sample_power_start(struct timechart *tchart __maybe_unused,
636 			   struct evsel *evsel,
637 			   struct perf_sample *sample,
638 			   const char *backtrace __maybe_unused)
639 {
640 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
641 	u64 value = perf_evsel__intval(evsel, sample, "value");
642 
643 	c_state_start(cpu_id, sample->time, value);
644 	return 0;
645 }
646 
647 static int
648 process_sample_power_end(struct timechart *tchart,
649 			 struct evsel *evsel __maybe_unused,
650 			 struct perf_sample *sample,
651 			 const char *backtrace __maybe_unused)
652 {
653 	c_state_end(tchart, sample->cpu, sample->time);
654 	return 0;
655 }
656 
657 static int
658 process_sample_power_frequency(struct timechart *tchart,
659 			       struct evsel *evsel,
660 			       struct perf_sample *sample,
661 			       const char *backtrace __maybe_unused)
662 {
663 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
664 	u64 value = perf_evsel__intval(evsel, sample, "value");
665 
666 	p_state_change(tchart, cpu_id, sample->time, value);
667 	return 0;
668 }
669 #endif /* SUPPORT_OLD_POWER_EVENTS */
670 
671 /*
672  * After the last sample we need to wrap up the current C/P state
673  * and close out each CPU for these.
674  */
675 static void end_sample_processing(struct timechart *tchart)
676 {
677 	u64 cpu;
678 	struct power_event *pwr;
679 
680 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
681 		/* C state */
682 #if 0
683 		pwr = zalloc(sizeof(*pwr));
684 		if (!pwr)
685 			return;
686 
687 		pwr->state = cpus_cstate_state[cpu];
688 		pwr->start_time = cpus_cstate_start_times[cpu];
689 		pwr->end_time = tchart->last_time;
690 		pwr->cpu = cpu;
691 		pwr->type = CSTATE;
692 		pwr->next = tchart->power_events;
693 
694 		tchart->power_events = pwr;
695 #endif
696 		/* P state */
697 
698 		pwr = zalloc(sizeof(*pwr));
699 		if (!pwr)
700 			return;
701 
702 		pwr->state = cpus_pstate_state[cpu];
703 		pwr->start_time = cpus_pstate_start_times[cpu];
704 		pwr->end_time = tchart->last_time;
705 		pwr->cpu = cpu;
706 		pwr->type = PSTATE;
707 		pwr->next = tchart->power_events;
708 
709 		if (!pwr->start_time)
710 			pwr->start_time = tchart->first_time;
711 		if (!pwr->state)
712 			pwr->state = tchart->min_freq;
713 		tchart->power_events = pwr;
714 	}
715 }
716 
717 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
718 			       u64 start, int fd)
719 {
720 	struct per_pid *p = find_create_pid(tchart, pid);
721 	struct per_pidcomm *c = p->current;
722 	struct io_sample *sample;
723 	struct io_sample *prev;
724 
725 	if (!c) {
726 		c = zalloc(sizeof(*c));
727 		if (!c)
728 			return -ENOMEM;
729 		p->current = c;
730 		c->next = p->all;
731 		p->all = c;
732 	}
733 
734 	prev = c->io_samples;
735 
736 	if (prev && prev->start_time && !prev->end_time) {
737 		pr_warning("Skip invalid start event: "
738 			   "previous event already started!\n");
739 
740 		/* remove previous event that has been started,
741 		 * we are not sure we will ever get an end for it */
742 		c->io_samples = prev->next;
743 		free(prev);
744 		return 0;
745 	}
746 
747 	sample = zalloc(sizeof(*sample));
748 	if (!sample)
749 		return -ENOMEM;
750 	sample->start_time = start;
751 	sample->type = type;
752 	sample->fd = fd;
753 	sample->next = c->io_samples;
754 	c->io_samples = sample;
755 
756 	if (c->start_time == 0 || c->start_time > start)
757 		c->start_time = start;
758 
759 	return 0;
760 }
761 
762 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
763 			     u64 end, long ret)
764 {
765 	struct per_pid *p = find_create_pid(tchart, pid);
766 	struct per_pidcomm *c = p->current;
767 	struct io_sample *sample, *prev;
768 
769 	if (!c) {
770 		pr_warning("Invalid pidcomm!\n");
771 		return -1;
772 	}
773 
774 	sample = c->io_samples;
775 
776 	if (!sample) /* skip partially captured events */
777 		return 0;
778 
779 	if (sample->end_time) {
780 		pr_warning("Skip invalid end event: "
781 			   "previous event already ended!\n");
782 		return 0;
783 	}
784 
785 	if (sample->type != type) {
786 		pr_warning("Skip invalid end event: invalid event type!\n");
787 		return 0;
788 	}
789 
790 	sample->end_time = end;
791 	prev = sample->next;
792 
793 	/* we want to be able to see small and fast transfers, so make them
794 	 * at least min_time long, but don't overlap them */
795 	if (sample->end_time - sample->start_time < tchart->min_time)
796 		sample->end_time = sample->start_time + tchart->min_time;
797 	if (prev && sample->start_time < prev->end_time) {
798 		if (prev->err) /* try to make errors more visible */
799 			sample->start_time = prev->end_time;
800 		else
801 			prev->end_time = sample->start_time;
802 	}
803 
804 	if (ret < 0) {
805 		sample->err = ret;
806 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
807 		   type == IOTYPE_TX || type == IOTYPE_RX) {
808 
809 		if ((u64)ret > c->max_bytes)
810 			c->max_bytes = ret;
811 
812 		c->total_bytes += ret;
813 		p->total_bytes += ret;
814 		sample->bytes = ret;
815 	}
816 
817 	/* merge two requests to make svg smaller and render-friendly */
818 	if (prev &&
819 	    prev->type == sample->type &&
820 	    prev->err == sample->err &&
821 	    prev->fd == sample->fd &&
822 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
823 
824 		sample->bytes += prev->bytes;
825 		sample->merges += prev->merges + 1;
826 
827 		sample->start_time = prev->start_time;
828 		sample->next = prev->next;
829 		free(prev);
830 
831 		if (!sample->err && sample->bytes > c->max_bytes)
832 			c->max_bytes = sample->bytes;
833 	}
834 
835 	tchart->io_events++;
836 
837 	return 0;
838 }
839 
840 static int
841 process_enter_read(struct timechart *tchart,
842 		   struct evsel *evsel,
843 		   struct perf_sample *sample)
844 {
845 	long fd = perf_evsel__intval(evsel, sample, "fd");
846 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
847 				   sample->time, fd);
848 }
849 
850 static int
851 process_exit_read(struct timechart *tchart,
852 		  struct evsel *evsel,
853 		  struct perf_sample *sample)
854 {
855 	long ret = perf_evsel__intval(evsel, sample, "ret");
856 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
857 				 sample->time, ret);
858 }
859 
860 static int
861 process_enter_write(struct timechart *tchart,
862 		    struct evsel *evsel,
863 		    struct perf_sample *sample)
864 {
865 	long fd = perf_evsel__intval(evsel, sample, "fd");
866 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
867 				   sample->time, fd);
868 }
869 
870 static int
871 process_exit_write(struct timechart *tchart,
872 		   struct evsel *evsel,
873 		   struct perf_sample *sample)
874 {
875 	long ret = perf_evsel__intval(evsel, sample, "ret");
876 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
877 				 sample->time, ret);
878 }
879 
880 static int
881 process_enter_sync(struct timechart *tchart,
882 		   struct evsel *evsel,
883 		   struct perf_sample *sample)
884 {
885 	long fd = perf_evsel__intval(evsel, sample, "fd");
886 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
887 				   sample->time, fd);
888 }
889 
890 static int
891 process_exit_sync(struct timechart *tchart,
892 		  struct evsel *evsel,
893 		  struct perf_sample *sample)
894 {
895 	long ret = perf_evsel__intval(evsel, sample, "ret");
896 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
897 				 sample->time, ret);
898 }
899 
900 static int
901 process_enter_tx(struct timechart *tchart,
902 		 struct evsel *evsel,
903 		 struct perf_sample *sample)
904 {
905 	long fd = perf_evsel__intval(evsel, sample, "fd");
906 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
907 				   sample->time, fd);
908 }
909 
910 static int
911 process_exit_tx(struct timechart *tchart,
912 		struct evsel *evsel,
913 		struct perf_sample *sample)
914 {
915 	long ret = perf_evsel__intval(evsel, sample, "ret");
916 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
917 				 sample->time, ret);
918 }
919 
920 static int
921 process_enter_rx(struct timechart *tchart,
922 		 struct evsel *evsel,
923 		 struct perf_sample *sample)
924 {
925 	long fd = perf_evsel__intval(evsel, sample, "fd");
926 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
927 				   sample->time, fd);
928 }
929 
930 static int
931 process_exit_rx(struct timechart *tchart,
932 		struct evsel *evsel,
933 		struct perf_sample *sample)
934 {
935 	long ret = perf_evsel__intval(evsel, sample, "ret");
936 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
937 				 sample->time, ret);
938 }
939 
940 static int
941 process_enter_poll(struct timechart *tchart,
942 		   struct evsel *evsel,
943 		   struct perf_sample *sample)
944 {
945 	long fd = perf_evsel__intval(evsel, sample, "fd");
946 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
947 				   sample->time, fd);
948 }
949 
950 static int
951 process_exit_poll(struct timechart *tchart,
952 		  struct evsel *evsel,
953 		  struct perf_sample *sample)
954 {
955 	long ret = perf_evsel__intval(evsel, sample, "ret");
956 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
957 				 sample->time, ret);
958 }
959 
960 /*
961  * Sort the pid datastructure
962  */
963 static void sort_pids(struct timechart *tchart)
964 {
965 	struct per_pid *new_list, *p, *cursor, *prev;
966 	/* sort by ppid first, then by pid, lowest to highest */
967 
968 	new_list = NULL;
969 
970 	while (tchart->all_data) {
971 		p = tchart->all_data;
972 		tchart->all_data = p->next;
973 		p->next = NULL;
974 
975 		if (new_list == NULL) {
976 			new_list = p;
977 			p->next = NULL;
978 			continue;
979 		}
980 		prev = NULL;
981 		cursor = new_list;
982 		while (cursor) {
983 			if (cursor->ppid > p->ppid ||
984 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
985 				/* must insert before */
986 				if (prev) {
987 					p->next = prev->next;
988 					prev->next = p;
989 					cursor = NULL;
990 					continue;
991 				} else {
992 					p->next = new_list;
993 					new_list = p;
994 					cursor = NULL;
995 					continue;
996 				}
997 			}
998 
999 			prev = cursor;
1000 			cursor = cursor->next;
1001 			if (!cursor)
1002 				prev->next = p;
1003 		}
1004 	}
1005 	tchart->all_data = new_list;
1006 }
1007 
1008 
1009 static void draw_c_p_states(struct timechart *tchart)
1010 {
1011 	struct power_event *pwr;
1012 	pwr = tchart->power_events;
1013 
1014 	/*
1015 	 * two pass drawing so that the P state bars are on top of the C state blocks
1016 	 */
1017 	while (pwr) {
1018 		if (pwr->type == CSTATE)
1019 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1020 		pwr = pwr->next;
1021 	}
1022 
1023 	pwr = tchart->power_events;
1024 	while (pwr) {
1025 		if (pwr->type == PSTATE) {
1026 			if (!pwr->state)
1027 				pwr->state = tchart->min_freq;
1028 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1029 		}
1030 		pwr = pwr->next;
1031 	}
1032 }
1033 
1034 static void draw_wakeups(struct timechart *tchart)
1035 {
1036 	struct wake_event *we;
1037 	struct per_pid *p;
1038 	struct per_pidcomm *c;
1039 
1040 	we = tchart->wake_events;
1041 	while (we) {
1042 		int from = 0, to = 0;
1043 		char *task_from = NULL, *task_to = NULL;
1044 
1045 		/* locate the column of the waker and wakee */
1046 		p = tchart->all_data;
1047 		while (p) {
1048 			if (p->pid == we->waker || p->pid == we->wakee) {
1049 				c = p->all;
1050 				while (c) {
1051 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1052 						if (p->pid == we->waker && !from) {
1053 							from = c->Y;
1054 							task_from = strdup(c->comm);
1055 						}
1056 						if (p->pid == we->wakee && !to) {
1057 							to = c->Y;
1058 							task_to = strdup(c->comm);
1059 						}
1060 					}
1061 					c = c->next;
1062 				}
1063 				c = p->all;
1064 				while (c) {
1065 					if (p->pid == we->waker && !from) {
1066 						from = c->Y;
1067 						task_from = strdup(c->comm);
1068 					}
1069 					if (p->pid == we->wakee && !to) {
1070 						to = c->Y;
1071 						task_to = strdup(c->comm);
1072 					}
1073 					c = c->next;
1074 				}
1075 			}
1076 			p = p->next;
1077 		}
1078 
1079 		if (!task_from) {
1080 			task_from = malloc(40);
1081 			sprintf(task_from, "[%i]", we->waker);
1082 		}
1083 		if (!task_to) {
1084 			task_to = malloc(40);
1085 			sprintf(task_to, "[%i]", we->wakee);
1086 		}
1087 
1088 		if (we->waker == -1)
1089 			svg_interrupt(we->time, to, we->backtrace);
1090 		else if (from && to && abs(from - to) == 1)
1091 			svg_wakeline(we->time, from, to, we->backtrace);
1092 		else
1093 			svg_partial_wakeline(we->time, from, task_from, to,
1094 					     task_to, we->backtrace);
1095 		we = we->next;
1096 
1097 		free(task_from);
1098 		free(task_to);
1099 	}
1100 }
1101 
1102 static void draw_cpu_usage(struct timechart *tchart)
1103 {
1104 	struct per_pid *p;
1105 	struct per_pidcomm *c;
1106 	struct cpu_sample *sample;
1107 	p = tchart->all_data;
1108 	while (p) {
1109 		c = p->all;
1110 		while (c) {
1111 			sample = c->samples;
1112 			while (sample) {
1113 				if (sample->type == TYPE_RUNNING) {
1114 					svg_process(sample->cpu,
1115 						    sample->start_time,
1116 						    sample->end_time,
1117 						    p->pid,
1118 						    c->comm,
1119 						    sample->backtrace);
1120 				}
1121 
1122 				sample = sample->next;
1123 			}
1124 			c = c->next;
1125 		}
1126 		p = p->next;
1127 	}
1128 }
1129 
1130 static void draw_io_bars(struct timechart *tchart)
1131 {
1132 	const char *suf;
1133 	double bytes;
1134 	char comm[256];
1135 	struct per_pid *p;
1136 	struct per_pidcomm *c;
1137 	struct io_sample *sample;
1138 	int Y = 1;
1139 
1140 	p = tchart->all_data;
1141 	while (p) {
1142 		c = p->all;
1143 		while (c) {
1144 			if (!c->display) {
1145 				c->Y = 0;
1146 				c = c->next;
1147 				continue;
1148 			}
1149 
1150 			svg_box(Y, c->start_time, c->end_time, "process3");
1151 			sample = c->io_samples;
1152 			for (sample = c->io_samples; sample; sample = sample->next) {
1153 				double h = (double)sample->bytes / c->max_bytes;
1154 
1155 				if (tchart->skip_eagain &&
1156 				    sample->err == -EAGAIN)
1157 					continue;
1158 
1159 				if (sample->err)
1160 					h = 1;
1161 
1162 				if (sample->type == IOTYPE_SYNC)
1163 					svg_fbox(Y,
1164 						sample->start_time,
1165 						sample->end_time,
1166 						1,
1167 						sample->err ? "error" : "sync",
1168 						sample->fd,
1169 						sample->err,
1170 						sample->merges);
1171 				else if (sample->type == IOTYPE_POLL)
1172 					svg_fbox(Y,
1173 						sample->start_time,
1174 						sample->end_time,
1175 						1,
1176 						sample->err ? "error" : "poll",
1177 						sample->fd,
1178 						sample->err,
1179 						sample->merges);
1180 				else if (sample->type == IOTYPE_READ)
1181 					svg_ubox(Y,
1182 						sample->start_time,
1183 						sample->end_time,
1184 						h,
1185 						sample->err ? "error" : "disk",
1186 						sample->fd,
1187 						sample->err,
1188 						sample->merges);
1189 				else if (sample->type == IOTYPE_WRITE)
1190 					svg_lbox(Y,
1191 						sample->start_time,
1192 						sample->end_time,
1193 						h,
1194 						sample->err ? "error" : "disk",
1195 						sample->fd,
1196 						sample->err,
1197 						sample->merges);
1198 				else if (sample->type == IOTYPE_RX)
1199 					svg_ubox(Y,
1200 						sample->start_time,
1201 						sample->end_time,
1202 						h,
1203 						sample->err ? "error" : "net",
1204 						sample->fd,
1205 						sample->err,
1206 						sample->merges);
1207 				else if (sample->type == IOTYPE_TX)
1208 					svg_lbox(Y,
1209 						sample->start_time,
1210 						sample->end_time,
1211 						h,
1212 						sample->err ? "error" : "net",
1213 						sample->fd,
1214 						sample->err,
1215 						sample->merges);
1216 			}
1217 
1218 			suf = "";
1219 			bytes = c->total_bytes;
1220 			if (bytes > 1024) {
1221 				bytes = bytes / 1024;
1222 				suf = "K";
1223 			}
1224 			if (bytes > 1024) {
1225 				bytes = bytes / 1024;
1226 				suf = "M";
1227 			}
1228 			if (bytes > 1024) {
1229 				bytes = bytes / 1024;
1230 				suf = "G";
1231 			}
1232 
1233 
1234 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1235 			svg_text(Y, c->start_time, comm);
1236 
1237 			c->Y = Y;
1238 			Y++;
1239 			c = c->next;
1240 		}
1241 		p = p->next;
1242 	}
1243 }
1244 
1245 static void draw_process_bars(struct timechart *tchart)
1246 {
1247 	struct per_pid *p;
1248 	struct per_pidcomm *c;
1249 	struct cpu_sample *sample;
1250 	int Y = 0;
1251 
1252 	Y = 2 * tchart->numcpus + 2;
1253 
1254 	p = tchart->all_data;
1255 	while (p) {
1256 		c = p->all;
1257 		while (c) {
1258 			if (!c->display) {
1259 				c->Y = 0;
1260 				c = c->next;
1261 				continue;
1262 			}
1263 
1264 			svg_box(Y, c->start_time, c->end_time, "process");
1265 			sample = c->samples;
1266 			while (sample) {
1267 				if (sample->type == TYPE_RUNNING)
1268 					svg_running(Y, sample->cpu,
1269 						    sample->start_time,
1270 						    sample->end_time,
1271 						    sample->backtrace);
1272 				if (sample->type == TYPE_BLOCKED)
1273 					svg_blocked(Y, sample->cpu,
1274 						    sample->start_time,
1275 						    sample->end_time,
1276 						    sample->backtrace);
1277 				if (sample->type == TYPE_WAITING)
1278 					svg_waiting(Y, sample->cpu,
1279 						    sample->start_time,
1280 						    sample->end_time,
1281 						    sample->backtrace);
1282 				sample = sample->next;
1283 			}
1284 
1285 			if (c->comm) {
1286 				char comm[256];
1287 				if (c->total_time > 5000000000) /* 5 seconds */
1288 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1289 				else
1290 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1291 
1292 				svg_text(Y, c->start_time, comm);
1293 			}
1294 			c->Y = Y;
1295 			Y++;
1296 			c = c->next;
1297 		}
1298 		p = p->next;
1299 	}
1300 }
1301 
1302 static void add_process_filter(const char *string)
1303 {
1304 	int pid = strtoull(string, NULL, 10);
1305 	struct process_filter *filt = malloc(sizeof(*filt));
1306 
1307 	if (!filt)
1308 		return;
1309 
1310 	filt->name = strdup(string);
1311 	filt->pid  = pid;
1312 	filt->next = process_filter;
1313 
1314 	process_filter = filt;
1315 }
1316 
1317 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1318 {
1319 	struct process_filter *filt;
1320 	if (!process_filter)
1321 		return 1;
1322 
1323 	filt = process_filter;
1324 	while (filt) {
1325 		if (filt->pid && p->pid == filt->pid)
1326 			return 1;
1327 		if (strcmp(filt->name, c->comm) == 0)
1328 			return 1;
1329 		filt = filt->next;
1330 	}
1331 	return 0;
1332 }
1333 
1334 static int determine_display_tasks_filtered(struct timechart *tchart)
1335 {
1336 	struct per_pid *p;
1337 	struct per_pidcomm *c;
1338 	int count = 0;
1339 
1340 	p = tchart->all_data;
1341 	while (p) {
1342 		p->display = 0;
1343 		if (p->start_time == 1)
1344 			p->start_time = tchart->first_time;
1345 
1346 		/* no exit marker, task kept running to the end */
1347 		if (p->end_time == 0)
1348 			p->end_time = tchart->last_time;
1349 
1350 		c = p->all;
1351 
1352 		while (c) {
1353 			c->display = 0;
1354 
1355 			if (c->start_time == 1)
1356 				c->start_time = tchart->first_time;
1357 
1358 			if (passes_filter(p, c)) {
1359 				c->display = 1;
1360 				p->display = 1;
1361 				count++;
1362 			}
1363 
1364 			if (c->end_time == 0)
1365 				c->end_time = tchart->last_time;
1366 
1367 			c = c->next;
1368 		}
1369 		p = p->next;
1370 	}
1371 	return count;
1372 }
1373 
1374 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1375 {
1376 	struct per_pid *p;
1377 	struct per_pidcomm *c;
1378 	int count = 0;
1379 
1380 	p = tchart->all_data;
1381 	while (p) {
1382 		p->display = 0;
1383 		if (p->start_time == 1)
1384 			p->start_time = tchart->first_time;
1385 
1386 		/* no exit marker, task kept running to the end */
1387 		if (p->end_time == 0)
1388 			p->end_time = tchart->last_time;
1389 		if (p->total_time >= threshold)
1390 			p->display = 1;
1391 
1392 		c = p->all;
1393 
1394 		while (c) {
1395 			c->display = 0;
1396 
1397 			if (c->start_time == 1)
1398 				c->start_time = tchart->first_time;
1399 
1400 			if (c->total_time >= threshold) {
1401 				c->display = 1;
1402 				count++;
1403 			}
1404 
1405 			if (c->end_time == 0)
1406 				c->end_time = tchart->last_time;
1407 
1408 			c = c->next;
1409 		}
1410 		p = p->next;
1411 	}
1412 	return count;
1413 }
1414 
1415 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1416 {
1417 	struct per_pid *p;
1418 	struct per_pidcomm *c;
1419 	int count = 0;
1420 
1421 	p = timechart->all_data;
1422 	while (p) {
1423 		/* no exit marker, task kept running to the end */
1424 		if (p->end_time == 0)
1425 			p->end_time = timechart->last_time;
1426 
1427 		c = p->all;
1428 
1429 		while (c) {
1430 			c->display = 0;
1431 
1432 			if (c->total_bytes >= threshold) {
1433 				c->display = 1;
1434 				count++;
1435 			}
1436 
1437 			if (c->end_time == 0)
1438 				c->end_time = timechart->last_time;
1439 
1440 			c = c->next;
1441 		}
1442 		p = p->next;
1443 	}
1444 	return count;
1445 }
1446 
1447 #define BYTES_THRESH (1 * 1024 * 1024)
1448 #define TIME_THRESH 10000000
1449 
1450 static void write_svg_file(struct timechart *tchart, const char *filename)
1451 {
1452 	u64 i;
1453 	int count;
1454 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1455 
1456 	if (tchart->power_only)
1457 		tchart->proc_num = 0;
1458 
1459 	/* We'd like to show at least proc_num tasks;
1460 	 * be less picky if we have fewer */
1461 	do {
1462 		if (process_filter)
1463 			count = determine_display_tasks_filtered(tchart);
1464 		else if (tchart->io_events)
1465 			count = determine_display_io_tasks(tchart, thresh);
1466 		else
1467 			count = determine_display_tasks(tchart, thresh);
1468 		thresh /= 10;
1469 	} while (!process_filter && thresh && count < tchart->proc_num);
1470 
1471 	if (!tchart->proc_num)
1472 		count = 0;
1473 
1474 	if (tchart->io_events) {
1475 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1476 
1477 		svg_time_grid(0.5);
1478 		svg_io_legenda();
1479 
1480 		draw_io_bars(tchart);
1481 	} else {
1482 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1483 
1484 		svg_time_grid(0);
1485 
1486 		svg_legenda();
1487 
1488 		for (i = 0; i < tchart->numcpus; i++)
1489 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1490 
1491 		draw_cpu_usage(tchart);
1492 		if (tchart->proc_num)
1493 			draw_process_bars(tchart);
1494 		if (!tchart->tasks_only)
1495 			draw_c_p_states(tchart);
1496 		if (tchart->proc_num)
1497 			draw_wakeups(tchart);
1498 	}
1499 
1500 	svg_close();
1501 }
1502 
1503 static int process_header(struct perf_file_section *section __maybe_unused,
1504 			  struct perf_header *ph,
1505 			  int feat,
1506 			  int fd __maybe_unused,
1507 			  void *data)
1508 {
1509 	struct timechart *tchart = data;
1510 
1511 	switch (feat) {
1512 	case HEADER_NRCPUS:
1513 		tchart->numcpus = ph->env.nr_cpus_avail;
1514 		break;
1515 
1516 	case HEADER_CPU_TOPOLOGY:
1517 		if (!tchart->topology)
1518 			break;
1519 
1520 		if (svg_build_topology_map(&ph->env))
1521 			fprintf(stderr, "problem building topology\n");
1522 		break;
1523 
1524 	default:
1525 		break;
1526 	}
1527 
1528 	return 0;
1529 }
1530 
1531 static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1532 {
1533 	const struct evsel_str_handler power_tracepoints[] = {
1534 		{ "power:cpu_idle",		process_sample_cpu_idle },
1535 		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1536 		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1537 		{ "sched:sched_switch",		process_sample_sched_switch },
1538 #ifdef SUPPORT_OLD_POWER_EVENTS
1539 		{ "power:power_start",		process_sample_power_start },
1540 		{ "power:power_end",		process_sample_power_end },
1541 		{ "power:power_frequency",	process_sample_power_frequency },
1542 #endif
1543 
1544 		{ "syscalls:sys_enter_read",		process_enter_read },
1545 		{ "syscalls:sys_enter_pread64",		process_enter_read },
1546 		{ "syscalls:sys_enter_readv",		process_enter_read },
1547 		{ "syscalls:sys_enter_preadv",		process_enter_read },
1548 		{ "syscalls:sys_enter_write",		process_enter_write },
1549 		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1550 		{ "syscalls:sys_enter_writev",		process_enter_write },
1551 		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1552 		{ "syscalls:sys_enter_sync",		process_enter_sync },
1553 		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1554 		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1555 		{ "syscalls:sys_enter_msync",		process_enter_sync },
1556 		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1557 		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1558 		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1559 		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1560 		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1561 		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1562 		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1563 		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1564 		{ "syscalls:sys_enter_poll",		process_enter_poll },
1565 		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1566 		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1567 		{ "syscalls:sys_enter_select",		process_enter_poll },
1568 
1569 		{ "syscalls:sys_exit_read",		process_exit_read },
1570 		{ "syscalls:sys_exit_pread64",		process_exit_read },
1571 		{ "syscalls:sys_exit_readv",		process_exit_read },
1572 		{ "syscalls:sys_exit_preadv",		process_exit_read },
1573 		{ "syscalls:sys_exit_write",		process_exit_write },
1574 		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1575 		{ "syscalls:sys_exit_writev",		process_exit_write },
1576 		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1577 		{ "syscalls:sys_exit_sync",		process_exit_sync },
1578 		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1579 		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1580 		{ "syscalls:sys_exit_msync",		process_exit_sync },
1581 		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1582 		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1583 		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1584 		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1585 		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1586 		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1587 		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1588 		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1589 		{ "syscalls:sys_exit_poll",		process_exit_poll },
1590 		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1591 		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1592 		{ "syscalls:sys_exit_select",		process_exit_poll },
1593 	};
1594 	struct perf_data data = {
1595 		.path  = input_name,
1596 		.mode  = PERF_DATA_MODE_READ,
1597 		.force = tchart->force,
1598 	};
1599 
1600 	struct perf_session *session = perf_session__new(&data, false,
1601 							 &tchart->tool);
1602 	int ret = -EINVAL;
1603 
1604 	if (session == NULL)
1605 		return -1;
1606 
1607 	symbol__init(&session->header.env);
1608 
1609 	(void)perf_header__process_sections(&session->header,
1610 					    perf_data__fd(session->data),
1611 					    tchart,
1612 					    process_header);
1613 
1614 	if (!perf_session__has_traces(session, "timechart record"))
1615 		goto out_delete;
1616 
1617 	if (perf_session__set_tracepoints_handlers(session,
1618 						   power_tracepoints)) {
1619 		pr_err("Initializing session tracepoint handlers failed\n");
1620 		goto out_delete;
1621 	}
1622 
1623 	ret = perf_session__process_events(session);
1624 	if (ret)
1625 		goto out_delete;
1626 
1627 	end_sample_processing(tchart);
1628 
1629 	sort_pids(tchart);
1630 
1631 	write_svg_file(tchart, output_name);
1632 
1633 	pr_info("Written %2.1f seconds of trace to %s.\n",
1634 		(tchart->last_time - tchart->first_time) / (double)NSEC_PER_SEC, output_name);
1635 out_delete:
1636 	perf_session__delete(session);
1637 	return ret;
1638 }
1639 
1640 static int timechart__io_record(int argc, const char **argv)
1641 {
1642 	unsigned int rec_argc, i;
1643 	const char **rec_argv;
1644 	const char **p;
1645 	char *filter = NULL;
1646 
1647 	const char * const common_args[] = {
1648 		"record", "-a", "-R", "-c", "1",
1649 	};
1650 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1651 
1652 	const char * const disk_events[] = {
1653 		"syscalls:sys_enter_read",
1654 		"syscalls:sys_enter_pread64",
1655 		"syscalls:sys_enter_readv",
1656 		"syscalls:sys_enter_preadv",
1657 		"syscalls:sys_enter_write",
1658 		"syscalls:sys_enter_pwrite64",
1659 		"syscalls:sys_enter_writev",
1660 		"syscalls:sys_enter_pwritev",
1661 		"syscalls:sys_enter_sync",
1662 		"syscalls:sys_enter_sync_file_range",
1663 		"syscalls:sys_enter_fsync",
1664 		"syscalls:sys_enter_msync",
1665 
1666 		"syscalls:sys_exit_read",
1667 		"syscalls:sys_exit_pread64",
1668 		"syscalls:sys_exit_readv",
1669 		"syscalls:sys_exit_preadv",
1670 		"syscalls:sys_exit_write",
1671 		"syscalls:sys_exit_pwrite64",
1672 		"syscalls:sys_exit_writev",
1673 		"syscalls:sys_exit_pwritev",
1674 		"syscalls:sys_exit_sync",
1675 		"syscalls:sys_exit_sync_file_range",
1676 		"syscalls:sys_exit_fsync",
1677 		"syscalls:sys_exit_msync",
1678 	};
1679 	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1680 
1681 	const char * const net_events[] = {
1682 		"syscalls:sys_enter_recvfrom",
1683 		"syscalls:sys_enter_recvmmsg",
1684 		"syscalls:sys_enter_recvmsg",
1685 		"syscalls:sys_enter_sendto",
1686 		"syscalls:sys_enter_sendmsg",
1687 		"syscalls:sys_enter_sendmmsg",
1688 
1689 		"syscalls:sys_exit_recvfrom",
1690 		"syscalls:sys_exit_recvmmsg",
1691 		"syscalls:sys_exit_recvmsg",
1692 		"syscalls:sys_exit_sendto",
1693 		"syscalls:sys_exit_sendmsg",
1694 		"syscalls:sys_exit_sendmmsg",
1695 	};
1696 	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1697 
1698 	const char * const poll_events[] = {
1699 		"syscalls:sys_enter_epoll_pwait",
1700 		"syscalls:sys_enter_epoll_wait",
1701 		"syscalls:sys_enter_poll",
1702 		"syscalls:sys_enter_ppoll",
1703 		"syscalls:sys_enter_pselect6",
1704 		"syscalls:sys_enter_select",
1705 
1706 		"syscalls:sys_exit_epoll_pwait",
1707 		"syscalls:sys_exit_epoll_wait",
1708 		"syscalls:sys_exit_poll",
1709 		"syscalls:sys_exit_ppoll",
1710 		"syscalls:sys_exit_pselect6",
1711 		"syscalls:sys_exit_select",
1712 	};
1713 	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1714 
1715 	rec_argc = common_args_nr +
1716 		disk_events_nr * 4 +
1717 		net_events_nr * 4 +
1718 		poll_events_nr * 4 +
1719 		argc;
1720 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1721 
1722 	if (rec_argv == NULL)
1723 		return -ENOMEM;
1724 
1725 	if (asprintf(&filter, "common_pid != %d", getpid()) < 0) {
1726 		free(rec_argv);
1727 		return -ENOMEM;
1728 	}
1729 
1730 	p = rec_argv;
1731 	for (i = 0; i < common_args_nr; i++)
1732 		*p++ = strdup(common_args[i]);
1733 
1734 	for (i = 0; i < disk_events_nr; i++) {
1735 		if (!is_valid_tracepoint(disk_events[i])) {
1736 			rec_argc -= 4;
1737 			continue;
1738 		}
1739 
1740 		*p++ = "-e";
1741 		*p++ = strdup(disk_events[i]);
1742 		*p++ = "--filter";
1743 		*p++ = filter;
1744 	}
1745 	for (i = 0; i < net_events_nr; i++) {
1746 		if (!is_valid_tracepoint(net_events[i])) {
1747 			rec_argc -= 4;
1748 			continue;
1749 		}
1750 
1751 		*p++ = "-e";
1752 		*p++ = strdup(net_events[i]);
1753 		*p++ = "--filter";
1754 		*p++ = filter;
1755 	}
1756 	for (i = 0; i < poll_events_nr; i++) {
1757 		if (!is_valid_tracepoint(poll_events[i])) {
1758 			rec_argc -= 4;
1759 			continue;
1760 		}
1761 
1762 		*p++ = "-e";
1763 		*p++ = strdup(poll_events[i]);
1764 		*p++ = "--filter";
1765 		*p++ = filter;
1766 	}
1767 
1768 	for (i = 0; i < (unsigned int)argc; i++)
1769 		*p++ = argv[i];
1770 
1771 	return cmd_record(rec_argc, rec_argv);
1772 }
1773 
1774 
1775 static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1776 {
1777 	unsigned int rec_argc, i, j;
1778 	const char **rec_argv;
1779 	const char **p;
1780 	unsigned int record_elems;
1781 
1782 	const char * const common_args[] = {
1783 		"record", "-a", "-R", "-c", "1",
1784 	};
1785 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1786 
1787 	const char * const backtrace_args[] = {
1788 		"-g",
1789 	};
1790 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1791 
1792 	const char * const power_args[] = {
1793 		"-e", "power:cpu_frequency",
1794 		"-e", "power:cpu_idle",
1795 	};
1796 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1797 
1798 	const char * const old_power_args[] = {
1799 #ifdef SUPPORT_OLD_POWER_EVENTS
1800 		"-e", "power:power_start",
1801 		"-e", "power:power_end",
1802 		"-e", "power:power_frequency",
1803 #endif
1804 	};
1805 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1806 
1807 	const char * const tasks_args[] = {
1808 		"-e", "sched:sched_wakeup",
1809 		"-e", "sched:sched_switch",
1810 	};
1811 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1812 
1813 #ifdef SUPPORT_OLD_POWER_EVENTS
1814 	if (!is_valid_tracepoint("power:cpu_idle") &&
1815 	    is_valid_tracepoint("power:power_start")) {
1816 		use_old_power_events = 1;
1817 		power_args_nr = 0;
1818 	} else {
1819 		old_power_args_nr = 0;
1820 	}
1821 #endif
1822 
1823 	if (tchart->power_only)
1824 		tasks_args_nr = 0;
1825 
1826 	if (tchart->tasks_only) {
1827 		power_args_nr = 0;
1828 		old_power_args_nr = 0;
1829 	}
1830 
1831 	if (!tchart->with_backtrace)
1832 		backtrace_args_no = 0;
1833 
1834 	record_elems = common_args_nr + tasks_args_nr +
1835 		power_args_nr + old_power_args_nr + backtrace_args_no;
1836 
1837 	rec_argc = record_elems + argc;
1838 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1839 
1840 	if (rec_argv == NULL)
1841 		return -ENOMEM;
1842 
1843 	p = rec_argv;
1844 	for (i = 0; i < common_args_nr; i++)
1845 		*p++ = strdup(common_args[i]);
1846 
1847 	for (i = 0; i < backtrace_args_no; i++)
1848 		*p++ = strdup(backtrace_args[i]);
1849 
1850 	for (i = 0; i < tasks_args_nr; i++)
1851 		*p++ = strdup(tasks_args[i]);
1852 
1853 	for (i = 0; i < power_args_nr; i++)
1854 		*p++ = strdup(power_args[i]);
1855 
1856 	for (i = 0; i < old_power_args_nr; i++)
1857 		*p++ = strdup(old_power_args[i]);
1858 
1859 	for (j = 0; j < (unsigned int)argc; j++)
1860 		*p++ = argv[j];
1861 
1862 	return cmd_record(rec_argc, rec_argv);
1863 }
1864 
1865 static int
1866 parse_process(const struct option *opt __maybe_unused, const char *arg,
1867 	      int __maybe_unused unset)
1868 {
1869 	if (arg)
1870 		add_process_filter(arg);
1871 	return 0;
1872 }
1873 
1874 static int
1875 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1876 		int __maybe_unused unset)
1877 {
1878 	unsigned long duration = strtoul(arg, NULL, 0);
1879 
1880 	if (svg_highlight || svg_highlight_name)
1881 		return -1;
1882 
1883 	if (duration)
1884 		svg_highlight = duration;
1885 	else
1886 		svg_highlight_name = strdup(arg);
1887 
1888 	return 0;
1889 }
1890 
1891 static int
1892 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1893 {
1894 	char unit = 'n';
1895 	u64 *value = opt->value;
1896 
1897 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1898 		switch (unit) {
1899 		case 'm':
1900 			*value *= NSEC_PER_MSEC;
1901 			break;
1902 		case 'u':
1903 			*value *= NSEC_PER_USEC;
1904 			break;
1905 		case 'n':
1906 			break;
1907 		default:
1908 			return -1;
1909 		}
1910 	}
1911 
1912 	return 0;
1913 }
1914 
1915 int cmd_timechart(int argc, const char **argv)
1916 {
1917 	struct timechart tchart = {
1918 		.tool = {
1919 			.comm		 = process_comm_event,
1920 			.fork		 = process_fork_event,
1921 			.exit		 = process_exit_event,
1922 			.sample		 = process_sample_event,
1923 			.ordered_events	 = true,
1924 		},
1925 		.proc_num = 15,
1926 		.min_time = NSEC_PER_MSEC,
1927 		.merge_dist = 1000,
1928 	};
1929 	const char *output_name = "output.svg";
1930 	const struct option timechart_common_options[] = {
1931 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1932 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only, "output processes data only"),
1933 	OPT_END()
1934 	};
1935 	const struct option timechart_options[] = {
1936 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1937 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1938 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1939 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1940 		      "highlight tasks. Pass duration in ns or process name.",
1941 		       parse_highlight),
1942 	OPT_CALLBACK('p', "process", NULL, "process",
1943 		      "process selector. Pass a pid or process name.",
1944 		       parse_process),
1945 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1946 		     "Look for files with symbols relative to this directory",
1947 		     symbol__config_symfs),
1948 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1949 		    "min. number of tasks to print"),
1950 	OPT_BOOLEAN('t', "topology", &tchart.topology,
1951 		    "sort CPUs according to topology"),
1952 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1953 		    "skip EAGAIN errors"),
1954 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1955 		     "all IO faster than min-time will visually appear longer",
1956 		     parse_time),
1957 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1958 		     "merge events that are merge-dist us apart",
1959 		     parse_time),
1960 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1961 	OPT_PARENT(timechart_common_options),
1962 	};
1963 	const char * const timechart_subcommands[] = { "record", NULL };
1964 	const char *timechart_usage[] = {
1965 		"perf timechart [<options>] {record}",
1966 		NULL
1967 	};
1968 	const struct option timechart_record_options[] = {
1969 	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
1970 		    "record only IO data"),
1971 	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
1972 	OPT_PARENT(timechart_common_options),
1973 	};
1974 	const char * const timechart_record_usage[] = {
1975 		"perf timechart record [<options>]",
1976 		NULL
1977 	};
1978 	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
1979 			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1980 
1981 	if (tchart.power_only && tchart.tasks_only) {
1982 		pr_err("-P and -T options cannot be used at the same time.\n");
1983 		return -1;
1984 	}
1985 
1986 	if (argc && !strncmp(argv[0], "rec", 3)) {
1987 		argc = parse_options(argc, argv, timechart_record_options,
1988 				     timechart_record_usage,
1989 				     PARSE_OPT_STOP_AT_NON_OPTION);
1990 
1991 		if (tchart.power_only && tchart.tasks_only) {
1992 			pr_err("-P and -T options cannot be used at the same time.\n");
1993 			return -1;
1994 		}
1995 
1996 		if (tchart.io_only)
1997 			return timechart__io_record(argc, argv);
1998 		else
1999 			return timechart__record(&tchart, argc, argv);
2000 	} else if (argc)
2001 		usage_with_options(timechart_usage, timechart_options);
2002 
2003 	setup_pager();
2004 
2005 	return __cmd_timechart(&tchart, output_name);
2006 }
2007