1 /*
2  * builtin-timechart.c - make an svg timechart of system activity
3  *
4  * (C) Copyright 2009 Intel Corporation
5  *
6  * Authors:
7  *     Arjan van de Ven <arjan@linux.intel.com>
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * as published by the Free Software Foundation; version 2
12  * of the License.
13  */
14 
15 #include <traceevent/event-parse.h>
16 
17 #include "builtin.h"
18 
19 #include "util/util.h"
20 
21 #include "util/color.h"
22 #include <linux/list.h>
23 #include "util/cache.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include <linux/rbtree.h>
27 #include "util/symbol.h"
28 #include "util/callchain.h"
29 #include "util/strlist.h"
30 
31 #include "perf.h"
32 #include "util/header.h"
33 #include "util/parse-options.h"
34 #include "util/parse-events.h"
35 #include "util/event.h"
36 #include "util/session.h"
37 #include "util/svghelper.h"
38 #include "util/tool.h"
39 #include "util/data.h"
40 #include "util/debug.h"
41 
42 #define SUPPORT_OLD_POWER_EVENTS 1
43 #define PWR_EVENT_EXIT -1
44 
45 struct per_pid;
46 struct power_event;
47 struct wake_event;
48 
49 struct timechart {
50 	struct perf_tool	tool;
51 	struct per_pid		*all_data;
52 	struct power_event	*power_events;
53 	struct wake_event	*wake_events;
54 	int			proc_num;
55 	unsigned int		numcpus;
56 	u64			min_freq,	/* Lowest CPU frequency seen */
57 				max_freq,	/* Highest CPU frequency seen */
58 				turbo_frequency,
59 				first_time, last_time;
60 	bool			power_only,
61 				tasks_only,
62 				with_backtrace,
63 				topology;
64 	/* IO related settings */
65 	u64			io_events;
66 	bool			io_only,
67 				skip_eagain;
68 	u64			min_time,
69 				merge_dist;
70 	bool			force;
71 };
72 
73 struct per_pidcomm;
74 struct cpu_sample;
75 struct io_sample;
76 
77 /*
78  * Datastructure layout:
79  * We keep an list of "pid"s, matching the kernels notion of a task struct.
80  * Each "pid" entry, has a list of "comm"s.
81  *	this is because we want to track different programs different, while
82  *	exec will reuse the original pid (by design).
83  * Each comm has a list of samples that will be used to draw
84  * final graph.
85  */
86 
87 struct per_pid {
88 	struct per_pid *next;
89 
90 	int		pid;
91 	int		ppid;
92 
93 	u64		start_time;
94 	u64		end_time;
95 	u64		total_time;
96 	u64		total_bytes;
97 	int		display;
98 
99 	struct per_pidcomm *all;
100 	struct per_pidcomm *current;
101 };
102 
103 
104 struct per_pidcomm {
105 	struct per_pidcomm *next;
106 
107 	u64		start_time;
108 	u64		end_time;
109 	u64		total_time;
110 	u64		max_bytes;
111 	u64		total_bytes;
112 
113 	int		Y;
114 	int		display;
115 
116 	long		state;
117 	u64		state_since;
118 
119 	char		*comm;
120 
121 	struct cpu_sample *samples;
122 	struct io_sample  *io_samples;
123 };
124 
125 struct sample_wrapper {
126 	struct sample_wrapper *next;
127 
128 	u64		timestamp;
129 	unsigned char	data[0];
130 };
131 
132 #define TYPE_NONE	0
133 #define TYPE_RUNNING	1
134 #define TYPE_WAITING	2
135 #define TYPE_BLOCKED	3
136 
137 struct cpu_sample {
138 	struct cpu_sample *next;
139 
140 	u64 start_time;
141 	u64 end_time;
142 	int type;
143 	int cpu;
144 	const char *backtrace;
145 };
146 
147 enum {
148 	IOTYPE_READ,
149 	IOTYPE_WRITE,
150 	IOTYPE_SYNC,
151 	IOTYPE_TX,
152 	IOTYPE_RX,
153 	IOTYPE_POLL,
154 };
155 
156 struct io_sample {
157 	struct io_sample *next;
158 
159 	u64 start_time;
160 	u64 end_time;
161 	u64 bytes;
162 	int type;
163 	int fd;
164 	int err;
165 	int merges;
166 };
167 
168 #define CSTATE 1
169 #define PSTATE 2
170 
171 struct power_event {
172 	struct power_event *next;
173 	int type;
174 	int state;
175 	u64 start_time;
176 	u64 end_time;
177 	int cpu;
178 };
179 
180 struct wake_event {
181 	struct wake_event *next;
182 	int waker;
183 	int wakee;
184 	u64 time;
185 	const char *backtrace;
186 };
187 
188 struct process_filter {
189 	char			*name;
190 	int			pid;
191 	struct process_filter	*next;
192 };
193 
194 static struct process_filter *process_filter;
195 
196 
197 static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
198 {
199 	struct per_pid *cursor = tchart->all_data;
200 
201 	while (cursor) {
202 		if (cursor->pid == pid)
203 			return cursor;
204 		cursor = cursor->next;
205 	}
206 	cursor = zalloc(sizeof(*cursor));
207 	assert(cursor != NULL);
208 	cursor->pid = pid;
209 	cursor->next = tchart->all_data;
210 	tchart->all_data = cursor;
211 	return cursor;
212 }
213 
214 static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
215 {
216 	struct per_pid *p;
217 	struct per_pidcomm *c;
218 	p = find_create_pid(tchart, pid);
219 	c = p->all;
220 	while (c) {
221 		if (c->comm && strcmp(c->comm, comm) == 0) {
222 			p->current = c;
223 			return;
224 		}
225 		if (!c->comm) {
226 			c->comm = strdup(comm);
227 			p->current = c;
228 			return;
229 		}
230 		c = c->next;
231 	}
232 	c = zalloc(sizeof(*c));
233 	assert(c != NULL);
234 	c->comm = strdup(comm);
235 	p->current = c;
236 	c->next = p->all;
237 	p->all = c;
238 }
239 
240 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
241 {
242 	struct per_pid *p, *pp;
243 	p = find_create_pid(tchart, pid);
244 	pp = find_create_pid(tchart, ppid);
245 	p->ppid = ppid;
246 	if (pp->current && pp->current->comm && !p->current)
247 		pid_set_comm(tchart, pid, pp->current->comm);
248 
249 	p->start_time = timestamp;
250 	if (p->current && !p->current->start_time) {
251 		p->current->start_time = timestamp;
252 		p->current->state_since = timestamp;
253 	}
254 }
255 
256 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
257 {
258 	struct per_pid *p;
259 	p = find_create_pid(tchart, pid);
260 	p->end_time = timestamp;
261 	if (p->current)
262 		p->current->end_time = timestamp;
263 }
264 
265 static void pid_put_sample(struct timechart *tchart, int pid, int type,
266 			   unsigned int cpu, u64 start, u64 end,
267 			   const char *backtrace)
268 {
269 	struct per_pid *p;
270 	struct per_pidcomm *c;
271 	struct cpu_sample *sample;
272 
273 	p = find_create_pid(tchart, pid);
274 	c = p->current;
275 	if (!c) {
276 		c = zalloc(sizeof(*c));
277 		assert(c != NULL);
278 		p->current = c;
279 		c->next = p->all;
280 		p->all = c;
281 	}
282 
283 	sample = zalloc(sizeof(*sample));
284 	assert(sample != NULL);
285 	sample->start_time = start;
286 	sample->end_time = end;
287 	sample->type = type;
288 	sample->next = c->samples;
289 	sample->cpu = cpu;
290 	sample->backtrace = backtrace;
291 	c->samples = sample;
292 
293 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
294 		c->total_time += (end-start);
295 		p->total_time += (end-start);
296 	}
297 
298 	if (c->start_time == 0 || c->start_time > start)
299 		c->start_time = start;
300 	if (p->start_time == 0 || p->start_time > start)
301 		p->start_time = start;
302 }
303 
304 #define MAX_CPUS 4096
305 
306 static u64 cpus_cstate_start_times[MAX_CPUS];
307 static int cpus_cstate_state[MAX_CPUS];
308 static u64 cpus_pstate_start_times[MAX_CPUS];
309 static u64 cpus_pstate_state[MAX_CPUS];
310 
311 static int process_comm_event(struct perf_tool *tool,
312 			      union perf_event *event,
313 			      struct perf_sample *sample __maybe_unused,
314 			      struct machine *machine __maybe_unused)
315 {
316 	struct timechart *tchart = container_of(tool, struct timechart, tool);
317 	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
318 	return 0;
319 }
320 
321 static int process_fork_event(struct perf_tool *tool,
322 			      union perf_event *event,
323 			      struct perf_sample *sample __maybe_unused,
324 			      struct machine *machine __maybe_unused)
325 {
326 	struct timechart *tchart = container_of(tool, struct timechart, tool);
327 	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
328 	return 0;
329 }
330 
331 static int process_exit_event(struct perf_tool *tool,
332 			      union perf_event *event,
333 			      struct perf_sample *sample __maybe_unused,
334 			      struct machine *machine __maybe_unused)
335 {
336 	struct timechart *tchart = container_of(tool, struct timechart, tool);
337 	pid_exit(tchart, event->fork.pid, event->fork.time);
338 	return 0;
339 }
340 
341 #ifdef SUPPORT_OLD_POWER_EVENTS
342 static int use_old_power_events;
343 #endif
344 
345 static void c_state_start(int cpu, u64 timestamp, int state)
346 {
347 	cpus_cstate_start_times[cpu] = timestamp;
348 	cpus_cstate_state[cpu] = state;
349 }
350 
351 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
352 {
353 	struct power_event *pwr = zalloc(sizeof(*pwr));
354 
355 	if (!pwr)
356 		return;
357 
358 	pwr->state = cpus_cstate_state[cpu];
359 	pwr->start_time = cpus_cstate_start_times[cpu];
360 	pwr->end_time = timestamp;
361 	pwr->cpu = cpu;
362 	pwr->type = CSTATE;
363 	pwr->next = tchart->power_events;
364 
365 	tchart->power_events = pwr;
366 }
367 
368 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
369 {
370 	struct power_event *pwr;
371 
372 	if (new_freq > 8000000) /* detect invalid data */
373 		return;
374 
375 	pwr = zalloc(sizeof(*pwr));
376 	if (!pwr)
377 		return;
378 
379 	pwr->state = cpus_pstate_state[cpu];
380 	pwr->start_time = cpus_pstate_start_times[cpu];
381 	pwr->end_time = timestamp;
382 	pwr->cpu = cpu;
383 	pwr->type = PSTATE;
384 	pwr->next = tchart->power_events;
385 
386 	if (!pwr->start_time)
387 		pwr->start_time = tchart->first_time;
388 
389 	tchart->power_events = pwr;
390 
391 	cpus_pstate_state[cpu] = new_freq;
392 	cpus_pstate_start_times[cpu] = timestamp;
393 
394 	if ((u64)new_freq > tchart->max_freq)
395 		tchart->max_freq = new_freq;
396 
397 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
398 		tchart->min_freq = new_freq;
399 
400 	if (new_freq == tchart->max_freq - 1000)
401 		tchart->turbo_frequency = tchart->max_freq;
402 }
403 
404 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
405 			 int waker, int wakee, u8 flags, const char *backtrace)
406 {
407 	struct per_pid *p;
408 	struct wake_event *we = zalloc(sizeof(*we));
409 
410 	if (!we)
411 		return;
412 
413 	we->time = timestamp;
414 	we->waker = waker;
415 	we->backtrace = backtrace;
416 
417 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
418 		we->waker = -1;
419 
420 	we->wakee = wakee;
421 	we->next = tchart->wake_events;
422 	tchart->wake_events = we;
423 	p = find_create_pid(tchart, we->wakee);
424 
425 	if (p && p->current && p->current->state == TYPE_NONE) {
426 		p->current->state_since = timestamp;
427 		p->current->state = TYPE_WAITING;
428 	}
429 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
430 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
431 			       p->current->state_since, timestamp, NULL);
432 		p->current->state_since = timestamp;
433 		p->current->state = TYPE_WAITING;
434 	}
435 }
436 
437 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
438 			 int prev_pid, int next_pid, u64 prev_state,
439 			 const char *backtrace)
440 {
441 	struct per_pid *p = NULL, *prev_p;
442 
443 	prev_p = find_create_pid(tchart, prev_pid);
444 
445 	p = find_create_pid(tchart, next_pid);
446 
447 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
448 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
449 			       prev_p->current->state_since, timestamp,
450 			       backtrace);
451 	if (p && p->current) {
452 		if (p->current->state != TYPE_NONE)
453 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
454 				       p->current->state_since, timestamp,
455 				       backtrace);
456 
457 		p->current->state_since = timestamp;
458 		p->current->state = TYPE_RUNNING;
459 	}
460 
461 	if (prev_p->current) {
462 		prev_p->current->state = TYPE_NONE;
463 		prev_p->current->state_since = timestamp;
464 		if (prev_state & 2)
465 			prev_p->current->state = TYPE_BLOCKED;
466 		if (prev_state == 0)
467 			prev_p->current->state = TYPE_WAITING;
468 	}
469 }
470 
471 static const char *cat_backtrace(union perf_event *event,
472 				 struct perf_sample *sample,
473 				 struct machine *machine)
474 {
475 	struct addr_location al;
476 	unsigned int i;
477 	char *p = NULL;
478 	size_t p_len;
479 	u8 cpumode = PERF_RECORD_MISC_USER;
480 	struct addr_location tal;
481 	struct ip_callchain *chain = sample->callchain;
482 	FILE *f = open_memstream(&p, &p_len);
483 
484 	if (!f) {
485 		perror("open_memstream error");
486 		return NULL;
487 	}
488 
489 	if (!chain)
490 		goto exit;
491 
492 	if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
493 		fprintf(stderr, "problem processing %d event, skipping it.\n",
494 			event->header.type);
495 		goto exit;
496 	}
497 
498 	for (i = 0; i < chain->nr; i++) {
499 		u64 ip;
500 
501 		if (callchain_param.order == ORDER_CALLEE)
502 			ip = chain->ips[i];
503 		else
504 			ip = chain->ips[chain->nr - i - 1];
505 
506 		if (ip >= PERF_CONTEXT_MAX) {
507 			switch (ip) {
508 			case PERF_CONTEXT_HV:
509 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
510 				break;
511 			case PERF_CONTEXT_KERNEL:
512 				cpumode = PERF_RECORD_MISC_KERNEL;
513 				break;
514 			case PERF_CONTEXT_USER:
515 				cpumode = PERF_RECORD_MISC_USER;
516 				break;
517 			default:
518 				pr_debug("invalid callchain context: "
519 					 "%"PRId64"\n", (s64) ip);
520 
521 				/*
522 				 * It seems the callchain is corrupted.
523 				 * Discard all.
524 				 */
525 				zfree(&p);
526 				goto exit;
527 			}
528 			continue;
529 		}
530 
531 		tal.filtered = 0;
532 		thread__find_addr_location(al.thread, cpumode,
533 					   MAP__FUNCTION, ip, &tal);
534 
535 		if (tal.sym)
536 			fprintf(f, "..... %016" PRIx64 " %s\n", ip,
537 				tal.sym->name);
538 		else
539 			fprintf(f, "..... %016" PRIx64 "\n", ip);
540 	}
541 
542 exit:
543 	fclose(f);
544 
545 	return p;
546 }
547 
548 typedef int (*tracepoint_handler)(struct timechart *tchart,
549 				  struct perf_evsel *evsel,
550 				  struct perf_sample *sample,
551 				  const char *backtrace);
552 
553 static int process_sample_event(struct perf_tool *tool,
554 				union perf_event *event,
555 				struct perf_sample *sample,
556 				struct perf_evsel *evsel,
557 				struct machine *machine)
558 {
559 	struct timechart *tchart = container_of(tool, struct timechart, tool);
560 
561 	if (evsel->attr.sample_type & PERF_SAMPLE_TIME) {
562 		if (!tchart->first_time || tchart->first_time > sample->time)
563 			tchart->first_time = sample->time;
564 		if (tchart->last_time < sample->time)
565 			tchart->last_time = sample->time;
566 	}
567 
568 	if (evsel->handler != NULL) {
569 		tracepoint_handler f = evsel->handler;
570 		return f(tchart, evsel, sample,
571 			 cat_backtrace(event, sample, machine));
572 	}
573 
574 	return 0;
575 }
576 
577 static int
578 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
579 			struct perf_evsel *evsel,
580 			struct perf_sample *sample,
581 			const char *backtrace __maybe_unused)
582 {
583 	u32 state = perf_evsel__intval(evsel, sample, "state");
584 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
585 
586 	if (state == (u32)PWR_EVENT_EXIT)
587 		c_state_end(tchart, cpu_id, sample->time);
588 	else
589 		c_state_start(cpu_id, sample->time, state);
590 	return 0;
591 }
592 
593 static int
594 process_sample_cpu_frequency(struct timechart *tchart,
595 			     struct perf_evsel *evsel,
596 			     struct perf_sample *sample,
597 			     const char *backtrace __maybe_unused)
598 {
599 	u32 state = perf_evsel__intval(evsel, sample, "state");
600 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
601 
602 	p_state_change(tchart, cpu_id, sample->time, state);
603 	return 0;
604 }
605 
606 static int
607 process_sample_sched_wakeup(struct timechart *tchart,
608 			    struct perf_evsel *evsel,
609 			    struct perf_sample *sample,
610 			    const char *backtrace)
611 {
612 	u8 flags = perf_evsel__intval(evsel, sample, "common_flags");
613 	int waker = perf_evsel__intval(evsel, sample, "common_pid");
614 	int wakee = perf_evsel__intval(evsel, sample, "pid");
615 
616 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
617 	return 0;
618 }
619 
620 static int
621 process_sample_sched_switch(struct timechart *tchart,
622 			    struct perf_evsel *evsel,
623 			    struct perf_sample *sample,
624 			    const char *backtrace)
625 {
626 	int prev_pid = perf_evsel__intval(evsel, sample, "prev_pid");
627 	int next_pid = perf_evsel__intval(evsel, sample, "next_pid");
628 	u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
629 
630 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
631 		     prev_state, backtrace);
632 	return 0;
633 }
634 
635 #ifdef SUPPORT_OLD_POWER_EVENTS
636 static int
637 process_sample_power_start(struct timechart *tchart __maybe_unused,
638 			   struct perf_evsel *evsel,
639 			   struct perf_sample *sample,
640 			   const char *backtrace __maybe_unused)
641 {
642 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
643 	u64 value = perf_evsel__intval(evsel, sample, "value");
644 
645 	c_state_start(cpu_id, sample->time, value);
646 	return 0;
647 }
648 
649 static int
650 process_sample_power_end(struct timechart *tchart,
651 			 struct perf_evsel *evsel __maybe_unused,
652 			 struct perf_sample *sample,
653 			 const char *backtrace __maybe_unused)
654 {
655 	c_state_end(tchart, sample->cpu, sample->time);
656 	return 0;
657 }
658 
659 static int
660 process_sample_power_frequency(struct timechart *tchart,
661 			       struct perf_evsel *evsel,
662 			       struct perf_sample *sample,
663 			       const char *backtrace __maybe_unused)
664 {
665 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
666 	u64 value = perf_evsel__intval(evsel, sample, "value");
667 
668 	p_state_change(tchart, cpu_id, sample->time, value);
669 	return 0;
670 }
671 #endif /* SUPPORT_OLD_POWER_EVENTS */
672 
673 /*
674  * After the last sample we need to wrap up the current C/P state
675  * and close out each CPU for these.
676  */
677 static void end_sample_processing(struct timechart *tchart)
678 {
679 	u64 cpu;
680 	struct power_event *pwr;
681 
682 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
683 		/* C state */
684 #if 0
685 		pwr = zalloc(sizeof(*pwr));
686 		if (!pwr)
687 			return;
688 
689 		pwr->state = cpus_cstate_state[cpu];
690 		pwr->start_time = cpus_cstate_start_times[cpu];
691 		pwr->end_time = tchart->last_time;
692 		pwr->cpu = cpu;
693 		pwr->type = CSTATE;
694 		pwr->next = tchart->power_events;
695 
696 		tchart->power_events = pwr;
697 #endif
698 		/* P state */
699 
700 		pwr = zalloc(sizeof(*pwr));
701 		if (!pwr)
702 			return;
703 
704 		pwr->state = cpus_pstate_state[cpu];
705 		pwr->start_time = cpus_pstate_start_times[cpu];
706 		pwr->end_time = tchart->last_time;
707 		pwr->cpu = cpu;
708 		pwr->type = PSTATE;
709 		pwr->next = tchart->power_events;
710 
711 		if (!pwr->start_time)
712 			pwr->start_time = tchart->first_time;
713 		if (!pwr->state)
714 			pwr->state = tchart->min_freq;
715 		tchart->power_events = pwr;
716 	}
717 }
718 
719 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
720 			       u64 start, int fd)
721 {
722 	struct per_pid *p = find_create_pid(tchart, pid);
723 	struct per_pidcomm *c = p->current;
724 	struct io_sample *sample;
725 	struct io_sample *prev;
726 
727 	if (!c) {
728 		c = zalloc(sizeof(*c));
729 		if (!c)
730 			return -ENOMEM;
731 		p->current = c;
732 		c->next = p->all;
733 		p->all = c;
734 	}
735 
736 	prev = c->io_samples;
737 
738 	if (prev && prev->start_time && !prev->end_time) {
739 		pr_warning("Skip invalid start event: "
740 			   "previous event already started!\n");
741 
742 		/* remove previous event that has been started,
743 		 * we are not sure we will ever get an end for it */
744 		c->io_samples = prev->next;
745 		free(prev);
746 		return 0;
747 	}
748 
749 	sample = zalloc(sizeof(*sample));
750 	if (!sample)
751 		return -ENOMEM;
752 	sample->start_time = start;
753 	sample->type = type;
754 	sample->fd = fd;
755 	sample->next = c->io_samples;
756 	c->io_samples = sample;
757 
758 	if (c->start_time == 0 || c->start_time > start)
759 		c->start_time = start;
760 
761 	return 0;
762 }
763 
764 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
765 			     u64 end, long ret)
766 {
767 	struct per_pid *p = find_create_pid(tchart, pid);
768 	struct per_pidcomm *c = p->current;
769 	struct io_sample *sample, *prev;
770 
771 	if (!c) {
772 		pr_warning("Invalid pidcomm!\n");
773 		return -1;
774 	}
775 
776 	sample = c->io_samples;
777 
778 	if (!sample) /* skip partially captured events */
779 		return 0;
780 
781 	if (sample->end_time) {
782 		pr_warning("Skip invalid end event: "
783 			   "previous event already ended!\n");
784 		return 0;
785 	}
786 
787 	if (sample->type != type) {
788 		pr_warning("Skip invalid end event: invalid event type!\n");
789 		return 0;
790 	}
791 
792 	sample->end_time = end;
793 	prev = sample->next;
794 
795 	/* we want to be able to see small and fast transfers, so make them
796 	 * at least min_time long, but don't overlap them */
797 	if (sample->end_time - sample->start_time < tchart->min_time)
798 		sample->end_time = sample->start_time + tchart->min_time;
799 	if (prev && sample->start_time < prev->end_time) {
800 		if (prev->err) /* try to make errors more visible */
801 			sample->start_time = prev->end_time;
802 		else
803 			prev->end_time = sample->start_time;
804 	}
805 
806 	if (ret < 0) {
807 		sample->err = ret;
808 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
809 		   type == IOTYPE_TX || type == IOTYPE_RX) {
810 
811 		if ((u64)ret > c->max_bytes)
812 			c->max_bytes = ret;
813 
814 		c->total_bytes += ret;
815 		p->total_bytes += ret;
816 		sample->bytes = ret;
817 	}
818 
819 	/* merge two requests to make svg smaller and render-friendly */
820 	if (prev &&
821 	    prev->type == sample->type &&
822 	    prev->err == sample->err &&
823 	    prev->fd == sample->fd &&
824 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
825 
826 		sample->bytes += prev->bytes;
827 		sample->merges += prev->merges + 1;
828 
829 		sample->start_time = prev->start_time;
830 		sample->next = prev->next;
831 		free(prev);
832 
833 		if (!sample->err && sample->bytes > c->max_bytes)
834 			c->max_bytes = sample->bytes;
835 	}
836 
837 	tchart->io_events++;
838 
839 	return 0;
840 }
841 
842 static int
843 process_enter_read(struct timechart *tchart,
844 		   struct perf_evsel *evsel,
845 		   struct perf_sample *sample)
846 {
847 	long fd = perf_evsel__intval(evsel, sample, "fd");
848 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
849 				   sample->time, fd);
850 }
851 
852 static int
853 process_exit_read(struct timechart *tchart,
854 		  struct perf_evsel *evsel,
855 		  struct perf_sample *sample)
856 {
857 	long ret = perf_evsel__intval(evsel, sample, "ret");
858 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
859 				 sample->time, ret);
860 }
861 
862 static int
863 process_enter_write(struct timechart *tchart,
864 		    struct perf_evsel *evsel,
865 		    struct perf_sample *sample)
866 {
867 	long fd = perf_evsel__intval(evsel, sample, "fd");
868 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
869 				   sample->time, fd);
870 }
871 
872 static int
873 process_exit_write(struct timechart *tchart,
874 		   struct perf_evsel *evsel,
875 		   struct perf_sample *sample)
876 {
877 	long ret = perf_evsel__intval(evsel, sample, "ret");
878 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
879 				 sample->time, ret);
880 }
881 
882 static int
883 process_enter_sync(struct timechart *tchart,
884 		   struct perf_evsel *evsel,
885 		   struct perf_sample *sample)
886 {
887 	long fd = perf_evsel__intval(evsel, sample, "fd");
888 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
889 				   sample->time, fd);
890 }
891 
892 static int
893 process_exit_sync(struct timechart *tchart,
894 		  struct perf_evsel *evsel,
895 		  struct perf_sample *sample)
896 {
897 	long ret = perf_evsel__intval(evsel, sample, "ret");
898 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
899 				 sample->time, ret);
900 }
901 
902 static int
903 process_enter_tx(struct timechart *tchart,
904 		 struct perf_evsel *evsel,
905 		 struct perf_sample *sample)
906 {
907 	long fd = perf_evsel__intval(evsel, sample, "fd");
908 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
909 				   sample->time, fd);
910 }
911 
912 static int
913 process_exit_tx(struct timechart *tchart,
914 		struct perf_evsel *evsel,
915 		struct perf_sample *sample)
916 {
917 	long ret = perf_evsel__intval(evsel, sample, "ret");
918 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
919 				 sample->time, ret);
920 }
921 
922 static int
923 process_enter_rx(struct timechart *tchart,
924 		 struct perf_evsel *evsel,
925 		 struct perf_sample *sample)
926 {
927 	long fd = perf_evsel__intval(evsel, sample, "fd");
928 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
929 				   sample->time, fd);
930 }
931 
932 static int
933 process_exit_rx(struct timechart *tchart,
934 		struct perf_evsel *evsel,
935 		struct perf_sample *sample)
936 {
937 	long ret = perf_evsel__intval(evsel, sample, "ret");
938 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
939 				 sample->time, ret);
940 }
941 
942 static int
943 process_enter_poll(struct timechart *tchart,
944 		   struct perf_evsel *evsel,
945 		   struct perf_sample *sample)
946 {
947 	long fd = perf_evsel__intval(evsel, sample, "fd");
948 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
949 				   sample->time, fd);
950 }
951 
952 static int
953 process_exit_poll(struct timechart *tchart,
954 		  struct perf_evsel *evsel,
955 		  struct perf_sample *sample)
956 {
957 	long ret = perf_evsel__intval(evsel, sample, "ret");
958 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
959 				 sample->time, ret);
960 }
961 
962 /*
963  * Sort the pid datastructure
964  */
965 static void sort_pids(struct timechart *tchart)
966 {
967 	struct per_pid *new_list, *p, *cursor, *prev;
968 	/* sort by ppid first, then by pid, lowest to highest */
969 
970 	new_list = NULL;
971 
972 	while (tchart->all_data) {
973 		p = tchart->all_data;
974 		tchart->all_data = p->next;
975 		p->next = NULL;
976 
977 		if (new_list == NULL) {
978 			new_list = p;
979 			p->next = NULL;
980 			continue;
981 		}
982 		prev = NULL;
983 		cursor = new_list;
984 		while (cursor) {
985 			if (cursor->ppid > p->ppid ||
986 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
987 				/* must insert before */
988 				if (prev) {
989 					p->next = prev->next;
990 					prev->next = p;
991 					cursor = NULL;
992 					continue;
993 				} else {
994 					p->next = new_list;
995 					new_list = p;
996 					cursor = NULL;
997 					continue;
998 				}
999 			}
1000 
1001 			prev = cursor;
1002 			cursor = cursor->next;
1003 			if (!cursor)
1004 				prev->next = p;
1005 		}
1006 	}
1007 	tchart->all_data = new_list;
1008 }
1009 
1010 
1011 static void draw_c_p_states(struct timechart *tchart)
1012 {
1013 	struct power_event *pwr;
1014 	pwr = tchart->power_events;
1015 
1016 	/*
1017 	 * two pass drawing so that the P state bars are on top of the C state blocks
1018 	 */
1019 	while (pwr) {
1020 		if (pwr->type == CSTATE)
1021 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1022 		pwr = pwr->next;
1023 	}
1024 
1025 	pwr = tchart->power_events;
1026 	while (pwr) {
1027 		if (pwr->type == PSTATE) {
1028 			if (!pwr->state)
1029 				pwr->state = tchart->min_freq;
1030 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1031 		}
1032 		pwr = pwr->next;
1033 	}
1034 }
1035 
1036 static void draw_wakeups(struct timechart *tchart)
1037 {
1038 	struct wake_event *we;
1039 	struct per_pid *p;
1040 	struct per_pidcomm *c;
1041 
1042 	we = tchart->wake_events;
1043 	while (we) {
1044 		int from = 0, to = 0;
1045 		char *task_from = NULL, *task_to = NULL;
1046 
1047 		/* locate the column of the waker and wakee */
1048 		p = tchart->all_data;
1049 		while (p) {
1050 			if (p->pid == we->waker || p->pid == we->wakee) {
1051 				c = p->all;
1052 				while (c) {
1053 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1054 						if (p->pid == we->waker && !from) {
1055 							from = c->Y;
1056 							task_from = strdup(c->comm);
1057 						}
1058 						if (p->pid == we->wakee && !to) {
1059 							to = c->Y;
1060 							task_to = strdup(c->comm);
1061 						}
1062 					}
1063 					c = c->next;
1064 				}
1065 				c = p->all;
1066 				while (c) {
1067 					if (p->pid == we->waker && !from) {
1068 						from = c->Y;
1069 						task_from = strdup(c->comm);
1070 					}
1071 					if (p->pid == we->wakee && !to) {
1072 						to = c->Y;
1073 						task_to = strdup(c->comm);
1074 					}
1075 					c = c->next;
1076 				}
1077 			}
1078 			p = p->next;
1079 		}
1080 
1081 		if (!task_from) {
1082 			task_from = malloc(40);
1083 			sprintf(task_from, "[%i]", we->waker);
1084 		}
1085 		if (!task_to) {
1086 			task_to = malloc(40);
1087 			sprintf(task_to, "[%i]", we->wakee);
1088 		}
1089 
1090 		if (we->waker == -1)
1091 			svg_interrupt(we->time, to, we->backtrace);
1092 		else if (from && to && abs(from - to) == 1)
1093 			svg_wakeline(we->time, from, to, we->backtrace);
1094 		else
1095 			svg_partial_wakeline(we->time, from, task_from, to,
1096 					     task_to, we->backtrace);
1097 		we = we->next;
1098 
1099 		free(task_from);
1100 		free(task_to);
1101 	}
1102 }
1103 
1104 static void draw_cpu_usage(struct timechart *tchart)
1105 {
1106 	struct per_pid *p;
1107 	struct per_pidcomm *c;
1108 	struct cpu_sample *sample;
1109 	p = tchart->all_data;
1110 	while (p) {
1111 		c = p->all;
1112 		while (c) {
1113 			sample = c->samples;
1114 			while (sample) {
1115 				if (sample->type == TYPE_RUNNING) {
1116 					svg_process(sample->cpu,
1117 						    sample->start_time,
1118 						    sample->end_time,
1119 						    p->pid,
1120 						    c->comm,
1121 						    sample->backtrace);
1122 				}
1123 
1124 				sample = sample->next;
1125 			}
1126 			c = c->next;
1127 		}
1128 		p = p->next;
1129 	}
1130 }
1131 
1132 static void draw_io_bars(struct timechart *tchart)
1133 {
1134 	const char *suf;
1135 	double bytes;
1136 	char comm[256];
1137 	struct per_pid *p;
1138 	struct per_pidcomm *c;
1139 	struct io_sample *sample;
1140 	int Y = 1;
1141 
1142 	p = tchart->all_data;
1143 	while (p) {
1144 		c = p->all;
1145 		while (c) {
1146 			if (!c->display) {
1147 				c->Y = 0;
1148 				c = c->next;
1149 				continue;
1150 			}
1151 
1152 			svg_box(Y, c->start_time, c->end_time, "process3");
1153 			sample = c->io_samples;
1154 			for (sample = c->io_samples; sample; sample = sample->next) {
1155 				double h = (double)sample->bytes / c->max_bytes;
1156 
1157 				if (tchart->skip_eagain &&
1158 				    sample->err == -EAGAIN)
1159 					continue;
1160 
1161 				if (sample->err)
1162 					h = 1;
1163 
1164 				if (sample->type == IOTYPE_SYNC)
1165 					svg_fbox(Y,
1166 						sample->start_time,
1167 						sample->end_time,
1168 						1,
1169 						sample->err ? "error" : "sync",
1170 						sample->fd,
1171 						sample->err,
1172 						sample->merges);
1173 				else if (sample->type == IOTYPE_POLL)
1174 					svg_fbox(Y,
1175 						sample->start_time,
1176 						sample->end_time,
1177 						1,
1178 						sample->err ? "error" : "poll",
1179 						sample->fd,
1180 						sample->err,
1181 						sample->merges);
1182 				else if (sample->type == IOTYPE_READ)
1183 					svg_ubox(Y,
1184 						sample->start_time,
1185 						sample->end_time,
1186 						h,
1187 						sample->err ? "error" : "disk",
1188 						sample->fd,
1189 						sample->err,
1190 						sample->merges);
1191 				else if (sample->type == IOTYPE_WRITE)
1192 					svg_lbox(Y,
1193 						sample->start_time,
1194 						sample->end_time,
1195 						h,
1196 						sample->err ? "error" : "disk",
1197 						sample->fd,
1198 						sample->err,
1199 						sample->merges);
1200 				else if (sample->type == IOTYPE_RX)
1201 					svg_ubox(Y,
1202 						sample->start_time,
1203 						sample->end_time,
1204 						h,
1205 						sample->err ? "error" : "net",
1206 						sample->fd,
1207 						sample->err,
1208 						sample->merges);
1209 				else if (sample->type == IOTYPE_TX)
1210 					svg_lbox(Y,
1211 						sample->start_time,
1212 						sample->end_time,
1213 						h,
1214 						sample->err ? "error" : "net",
1215 						sample->fd,
1216 						sample->err,
1217 						sample->merges);
1218 			}
1219 
1220 			suf = "";
1221 			bytes = c->total_bytes;
1222 			if (bytes > 1024) {
1223 				bytes = bytes / 1024;
1224 				suf = "K";
1225 			}
1226 			if (bytes > 1024) {
1227 				bytes = bytes / 1024;
1228 				suf = "M";
1229 			}
1230 			if (bytes > 1024) {
1231 				bytes = bytes / 1024;
1232 				suf = "G";
1233 			}
1234 
1235 
1236 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1237 			svg_text(Y, c->start_time, comm);
1238 
1239 			c->Y = Y;
1240 			Y++;
1241 			c = c->next;
1242 		}
1243 		p = p->next;
1244 	}
1245 }
1246 
1247 static void draw_process_bars(struct timechart *tchart)
1248 {
1249 	struct per_pid *p;
1250 	struct per_pidcomm *c;
1251 	struct cpu_sample *sample;
1252 	int Y = 0;
1253 
1254 	Y = 2 * tchart->numcpus + 2;
1255 
1256 	p = tchart->all_data;
1257 	while (p) {
1258 		c = p->all;
1259 		while (c) {
1260 			if (!c->display) {
1261 				c->Y = 0;
1262 				c = c->next;
1263 				continue;
1264 			}
1265 
1266 			svg_box(Y, c->start_time, c->end_time, "process");
1267 			sample = c->samples;
1268 			while (sample) {
1269 				if (sample->type == TYPE_RUNNING)
1270 					svg_running(Y, sample->cpu,
1271 						    sample->start_time,
1272 						    sample->end_time,
1273 						    sample->backtrace);
1274 				if (sample->type == TYPE_BLOCKED)
1275 					svg_blocked(Y, sample->cpu,
1276 						    sample->start_time,
1277 						    sample->end_time,
1278 						    sample->backtrace);
1279 				if (sample->type == TYPE_WAITING)
1280 					svg_waiting(Y, sample->cpu,
1281 						    sample->start_time,
1282 						    sample->end_time,
1283 						    sample->backtrace);
1284 				sample = sample->next;
1285 			}
1286 
1287 			if (c->comm) {
1288 				char comm[256];
1289 				if (c->total_time > 5000000000) /* 5 seconds */
1290 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / 1000000000.0);
1291 				else
1292 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / 1000000.0);
1293 
1294 				svg_text(Y, c->start_time, comm);
1295 			}
1296 			c->Y = Y;
1297 			Y++;
1298 			c = c->next;
1299 		}
1300 		p = p->next;
1301 	}
1302 }
1303 
1304 static void add_process_filter(const char *string)
1305 {
1306 	int pid = strtoull(string, NULL, 10);
1307 	struct process_filter *filt = malloc(sizeof(*filt));
1308 
1309 	if (!filt)
1310 		return;
1311 
1312 	filt->name = strdup(string);
1313 	filt->pid  = pid;
1314 	filt->next = process_filter;
1315 
1316 	process_filter = filt;
1317 }
1318 
1319 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1320 {
1321 	struct process_filter *filt;
1322 	if (!process_filter)
1323 		return 1;
1324 
1325 	filt = process_filter;
1326 	while (filt) {
1327 		if (filt->pid && p->pid == filt->pid)
1328 			return 1;
1329 		if (strcmp(filt->name, c->comm) == 0)
1330 			return 1;
1331 		filt = filt->next;
1332 	}
1333 	return 0;
1334 }
1335 
1336 static int determine_display_tasks_filtered(struct timechart *tchart)
1337 {
1338 	struct per_pid *p;
1339 	struct per_pidcomm *c;
1340 	int count = 0;
1341 
1342 	p = tchart->all_data;
1343 	while (p) {
1344 		p->display = 0;
1345 		if (p->start_time == 1)
1346 			p->start_time = tchart->first_time;
1347 
1348 		/* no exit marker, task kept running to the end */
1349 		if (p->end_time == 0)
1350 			p->end_time = tchart->last_time;
1351 
1352 		c = p->all;
1353 
1354 		while (c) {
1355 			c->display = 0;
1356 
1357 			if (c->start_time == 1)
1358 				c->start_time = tchart->first_time;
1359 
1360 			if (passes_filter(p, c)) {
1361 				c->display = 1;
1362 				p->display = 1;
1363 				count++;
1364 			}
1365 
1366 			if (c->end_time == 0)
1367 				c->end_time = tchart->last_time;
1368 
1369 			c = c->next;
1370 		}
1371 		p = p->next;
1372 	}
1373 	return count;
1374 }
1375 
1376 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1377 {
1378 	struct per_pid *p;
1379 	struct per_pidcomm *c;
1380 	int count = 0;
1381 
1382 	p = tchart->all_data;
1383 	while (p) {
1384 		p->display = 0;
1385 		if (p->start_time == 1)
1386 			p->start_time = tchart->first_time;
1387 
1388 		/* no exit marker, task kept running to the end */
1389 		if (p->end_time == 0)
1390 			p->end_time = tchart->last_time;
1391 		if (p->total_time >= threshold)
1392 			p->display = 1;
1393 
1394 		c = p->all;
1395 
1396 		while (c) {
1397 			c->display = 0;
1398 
1399 			if (c->start_time == 1)
1400 				c->start_time = tchart->first_time;
1401 
1402 			if (c->total_time >= threshold) {
1403 				c->display = 1;
1404 				count++;
1405 			}
1406 
1407 			if (c->end_time == 0)
1408 				c->end_time = tchart->last_time;
1409 
1410 			c = c->next;
1411 		}
1412 		p = p->next;
1413 	}
1414 	return count;
1415 }
1416 
1417 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1418 {
1419 	struct per_pid *p;
1420 	struct per_pidcomm *c;
1421 	int count = 0;
1422 
1423 	p = timechart->all_data;
1424 	while (p) {
1425 		/* no exit marker, task kept running to the end */
1426 		if (p->end_time == 0)
1427 			p->end_time = timechart->last_time;
1428 
1429 		c = p->all;
1430 
1431 		while (c) {
1432 			c->display = 0;
1433 
1434 			if (c->total_bytes >= threshold) {
1435 				c->display = 1;
1436 				count++;
1437 			}
1438 
1439 			if (c->end_time == 0)
1440 				c->end_time = timechart->last_time;
1441 
1442 			c = c->next;
1443 		}
1444 		p = p->next;
1445 	}
1446 	return count;
1447 }
1448 
1449 #define BYTES_THRESH (1 * 1024 * 1024)
1450 #define TIME_THRESH 10000000
1451 
1452 static void write_svg_file(struct timechart *tchart, const char *filename)
1453 {
1454 	u64 i;
1455 	int count;
1456 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1457 
1458 	if (tchart->power_only)
1459 		tchart->proc_num = 0;
1460 
1461 	/* We'd like to show at least proc_num tasks;
1462 	 * be less picky if we have fewer */
1463 	do {
1464 		if (process_filter)
1465 			count = determine_display_tasks_filtered(tchart);
1466 		else if (tchart->io_events)
1467 			count = determine_display_io_tasks(tchart, thresh);
1468 		else
1469 			count = determine_display_tasks(tchart, thresh);
1470 		thresh /= 10;
1471 	} while (!process_filter && thresh && count < tchart->proc_num);
1472 
1473 	if (!tchart->proc_num)
1474 		count = 0;
1475 
1476 	if (tchart->io_events) {
1477 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1478 
1479 		svg_time_grid(0.5);
1480 		svg_io_legenda();
1481 
1482 		draw_io_bars(tchart);
1483 	} else {
1484 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1485 
1486 		svg_time_grid(0);
1487 
1488 		svg_legenda();
1489 
1490 		for (i = 0; i < tchart->numcpus; i++)
1491 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1492 
1493 		draw_cpu_usage(tchart);
1494 		if (tchart->proc_num)
1495 			draw_process_bars(tchart);
1496 		if (!tchart->tasks_only)
1497 			draw_c_p_states(tchart);
1498 		if (tchart->proc_num)
1499 			draw_wakeups(tchart);
1500 	}
1501 
1502 	svg_close();
1503 }
1504 
1505 static int process_header(struct perf_file_section *section __maybe_unused,
1506 			  struct perf_header *ph,
1507 			  int feat,
1508 			  int fd __maybe_unused,
1509 			  void *data)
1510 {
1511 	struct timechart *tchart = data;
1512 
1513 	switch (feat) {
1514 	case HEADER_NRCPUS:
1515 		tchart->numcpus = ph->env.nr_cpus_avail;
1516 		break;
1517 
1518 	case HEADER_CPU_TOPOLOGY:
1519 		if (!tchart->topology)
1520 			break;
1521 
1522 		if (svg_build_topology_map(ph->env.sibling_cores,
1523 					   ph->env.nr_sibling_cores,
1524 					   ph->env.sibling_threads,
1525 					   ph->env.nr_sibling_threads))
1526 			fprintf(stderr, "problem building topology\n");
1527 		break;
1528 
1529 	default:
1530 		break;
1531 	}
1532 
1533 	return 0;
1534 }
1535 
1536 static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1537 {
1538 	const struct perf_evsel_str_handler power_tracepoints[] = {
1539 		{ "power:cpu_idle",		process_sample_cpu_idle },
1540 		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1541 		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1542 		{ "sched:sched_switch",		process_sample_sched_switch },
1543 #ifdef SUPPORT_OLD_POWER_EVENTS
1544 		{ "power:power_start",		process_sample_power_start },
1545 		{ "power:power_end",		process_sample_power_end },
1546 		{ "power:power_frequency",	process_sample_power_frequency },
1547 #endif
1548 
1549 		{ "syscalls:sys_enter_read",		process_enter_read },
1550 		{ "syscalls:sys_enter_pread64",		process_enter_read },
1551 		{ "syscalls:sys_enter_readv",		process_enter_read },
1552 		{ "syscalls:sys_enter_preadv",		process_enter_read },
1553 		{ "syscalls:sys_enter_write",		process_enter_write },
1554 		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1555 		{ "syscalls:sys_enter_writev",		process_enter_write },
1556 		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1557 		{ "syscalls:sys_enter_sync",		process_enter_sync },
1558 		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1559 		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1560 		{ "syscalls:sys_enter_msync",		process_enter_sync },
1561 		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1562 		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1563 		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1564 		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1565 		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1566 		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1567 		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1568 		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1569 		{ "syscalls:sys_enter_poll",		process_enter_poll },
1570 		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1571 		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1572 		{ "syscalls:sys_enter_select",		process_enter_poll },
1573 
1574 		{ "syscalls:sys_exit_read",		process_exit_read },
1575 		{ "syscalls:sys_exit_pread64",		process_exit_read },
1576 		{ "syscalls:sys_exit_readv",		process_exit_read },
1577 		{ "syscalls:sys_exit_preadv",		process_exit_read },
1578 		{ "syscalls:sys_exit_write",		process_exit_write },
1579 		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1580 		{ "syscalls:sys_exit_writev",		process_exit_write },
1581 		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1582 		{ "syscalls:sys_exit_sync",		process_exit_sync },
1583 		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1584 		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1585 		{ "syscalls:sys_exit_msync",		process_exit_sync },
1586 		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1587 		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1588 		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1589 		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1590 		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1591 		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1592 		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1593 		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1594 		{ "syscalls:sys_exit_poll",		process_exit_poll },
1595 		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1596 		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1597 		{ "syscalls:sys_exit_select",		process_exit_poll },
1598 	};
1599 	struct perf_data_file file = {
1600 		.path = input_name,
1601 		.mode = PERF_DATA_MODE_READ,
1602 		.force = tchart->force,
1603 	};
1604 
1605 	struct perf_session *session = perf_session__new(&file, false,
1606 							 &tchart->tool);
1607 	int ret = -EINVAL;
1608 
1609 	if (session == NULL)
1610 		return -1;
1611 
1612 	symbol__init(&session->header.env);
1613 
1614 	(void)perf_header__process_sections(&session->header,
1615 					    perf_data_file__fd(session->file),
1616 					    tchart,
1617 					    process_header);
1618 
1619 	if (!perf_session__has_traces(session, "timechart record"))
1620 		goto out_delete;
1621 
1622 	if (perf_session__set_tracepoints_handlers(session,
1623 						   power_tracepoints)) {
1624 		pr_err("Initializing session tracepoint handlers failed\n");
1625 		goto out_delete;
1626 	}
1627 
1628 	ret = perf_session__process_events(session);
1629 	if (ret)
1630 		goto out_delete;
1631 
1632 	end_sample_processing(tchart);
1633 
1634 	sort_pids(tchart);
1635 
1636 	write_svg_file(tchart, output_name);
1637 
1638 	pr_info("Written %2.1f seconds of trace to %s.\n",
1639 		(tchart->last_time - tchart->first_time) / 1000000000.0, output_name);
1640 out_delete:
1641 	perf_session__delete(session);
1642 	return ret;
1643 }
1644 
1645 static int timechart__io_record(int argc, const char **argv)
1646 {
1647 	unsigned int rec_argc, i;
1648 	const char **rec_argv;
1649 	const char **p;
1650 	char *filter = NULL;
1651 
1652 	const char * const common_args[] = {
1653 		"record", "-a", "-R", "-c", "1",
1654 	};
1655 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1656 
1657 	const char * const disk_events[] = {
1658 		"syscalls:sys_enter_read",
1659 		"syscalls:sys_enter_pread64",
1660 		"syscalls:sys_enter_readv",
1661 		"syscalls:sys_enter_preadv",
1662 		"syscalls:sys_enter_write",
1663 		"syscalls:sys_enter_pwrite64",
1664 		"syscalls:sys_enter_writev",
1665 		"syscalls:sys_enter_pwritev",
1666 		"syscalls:sys_enter_sync",
1667 		"syscalls:sys_enter_sync_file_range",
1668 		"syscalls:sys_enter_fsync",
1669 		"syscalls:sys_enter_msync",
1670 
1671 		"syscalls:sys_exit_read",
1672 		"syscalls:sys_exit_pread64",
1673 		"syscalls:sys_exit_readv",
1674 		"syscalls:sys_exit_preadv",
1675 		"syscalls:sys_exit_write",
1676 		"syscalls:sys_exit_pwrite64",
1677 		"syscalls:sys_exit_writev",
1678 		"syscalls:sys_exit_pwritev",
1679 		"syscalls:sys_exit_sync",
1680 		"syscalls:sys_exit_sync_file_range",
1681 		"syscalls:sys_exit_fsync",
1682 		"syscalls:sys_exit_msync",
1683 	};
1684 	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1685 
1686 	const char * const net_events[] = {
1687 		"syscalls:sys_enter_recvfrom",
1688 		"syscalls:sys_enter_recvmmsg",
1689 		"syscalls:sys_enter_recvmsg",
1690 		"syscalls:sys_enter_sendto",
1691 		"syscalls:sys_enter_sendmsg",
1692 		"syscalls:sys_enter_sendmmsg",
1693 
1694 		"syscalls:sys_exit_recvfrom",
1695 		"syscalls:sys_exit_recvmmsg",
1696 		"syscalls:sys_exit_recvmsg",
1697 		"syscalls:sys_exit_sendto",
1698 		"syscalls:sys_exit_sendmsg",
1699 		"syscalls:sys_exit_sendmmsg",
1700 	};
1701 	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1702 
1703 	const char * const poll_events[] = {
1704 		"syscalls:sys_enter_epoll_pwait",
1705 		"syscalls:sys_enter_epoll_wait",
1706 		"syscalls:sys_enter_poll",
1707 		"syscalls:sys_enter_ppoll",
1708 		"syscalls:sys_enter_pselect6",
1709 		"syscalls:sys_enter_select",
1710 
1711 		"syscalls:sys_exit_epoll_pwait",
1712 		"syscalls:sys_exit_epoll_wait",
1713 		"syscalls:sys_exit_poll",
1714 		"syscalls:sys_exit_ppoll",
1715 		"syscalls:sys_exit_pselect6",
1716 		"syscalls:sys_exit_select",
1717 	};
1718 	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1719 
1720 	rec_argc = common_args_nr +
1721 		disk_events_nr * 4 +
1722 		net_events_nr * 4 +
1723 		poll_events_nr * 4 +
1724 		argc;
1725 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1726 
1727 	if (rec_argv == NULL)
1728 		return -ENOMEM;
1729 
1730 	if (asprintf(&filter, "common_pid != %d", getpid()) < 0)
1731 		return -ENOMEM;
1732 
1733 	p = rec_argv;
1734 	for (i = 0; i < common_args_nr; i++)
1735 		*p++ = strdup(common_args[i]);
1736 
1737 	for (i = 0; i < disk_events_nr; i++) {
1738 		if (!is_valid_tracepoint(disk_events[i])) {
1739 			rec_argc -= 4;
1740 			continue;
1741 		}
1742 
1743 		*p++ = "-e";
1744 		*p++ = strdup(disk_events[i]);
1745 		*p++ = "--filter";
1746 		*p++ = filter;
1747 	}
1748 	for (i = 0; i < net_events_nr; i++) {
1749 		if (!is_valid_tracepoint(net_events[i])) {
1750 			rec_argc -= 4;
1751 			continue;
1752 		}
1753 
1754 		*p++ = "-e";
1755 		*p++ = strdup(net_events[i]);
1756 		*p++ = "--filter";
1757 		*p++ = filter;
1758 	}
1759 	for (i = 0; i < poll_events_nr; i++) {
1760 		if (!is_valid_tracepoint(poll_events[i])) {
1761 			rec_argc -= 4;
1762 			continue;
1763 		}
1764 
1765 		*p++ = "-e";
1766 		*p++ = strdup(poll_events[i]);
1767 		*p++ = "--filter";
1768 		*p++ = filter;
1769 	}
1770 
1771 	for (i = 0; i < (unsigned int)argc; i++)
1772 		*p++ = argv[i];
1773 
1774 	return cmd_record(rec_argc, rec_argv, NULL);
1775 }
1776 
1777 
1778 static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1779 {
1780 	unsigned int rec_argc, i, j;
1781 	const char **rec_argv;
1782 	const char **p;
1783 	unsigned int record_elems;
1784 
1785 	const char * const common_args[] = {
1786 		"record", "-a", "-R", "-c", "1",
1787 	};
1788 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1789 
1790 	const char * const backtrace_args[] = {
1791 		"-g",
1792 	};
1793 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1794 
1795 	const char * const power_args[] = {
1796 		"-e", "power:cpu_frequency",
1797 		"-e", "power:cpu_idle",
1798 	};
1799 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1800 
1801 	const char * const old_power_args[] = {
1802 #ifdef SUPPORT_OLD_POWER_EVENTS
1803 		"-e", "power:power_start",
1804 		"-e", "power:power_end",
1805 		"-e", "power:power_frequency",
1806 #endif
1807 	};
1808 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1809 
1810 	const char * const tasks_args[] = {
1811 		"-e", "sched:sched_wakeup",
1812 		"-e", "sched:sched_switch",
1813 	};
1814 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1815 
1816 #ifdef SUPPORT_OLD_POWER_EVENTS
1817 	if (!is_valid_tracepoint("power:cpu_idle") &&
1818 	    is_valid_tracepoint("power:power_start")) {
1819 		use_old_power_events = 1;
1820 		power_args_nr = 0;
1821 	} else {
1822 		old_power_args_nr = 0;
1823 	}
1824 #endif
1825 
1826 	if (tchart->power_only)
1827 		tasks_args_nr = 0;
1828 
1829 	if (tchart->tasks_only) {
1830 		power_args_nr = 0;
1831 		old_power_args_nr = 0;
1832 	}
1833 
1834 	if (!tchart->with_backtrace)
1835 		backtrace_args_no = 0;
1836 
1837 	record_elems = common_args_nr + tasks_args_nr +
1838 		power_args_nr + old_power_args_nr + backtrace_args_no;
1839 
1840 	rec_argc = record_elems + argc;
1841 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1842 
1843 	if (rec_argv == NULL)
1844 		return -ENOMEM;
1845 
1846 	p = rec_argv;
1847 	for (i = 0; i < common_args_nr; i++)
1848 		*p++ = strdup(common_args[i]);
1849 
1850 	for (i = 0; i < backtrace_args_no; i++)
1851 		*p++ = strdup(backtrace_args[i]);
1852 
1853 	for (i = 0; i < tasks_args_nr; i++)
1854 		*p++ = strdup(tasks_args[i]);
1855 
1856 	for (i = 0; i < power_args_nr; i++)
1857 		*p++ = strdup(power_args[i]);
1858 
1859 	for (i = 0; i < old_power_args_nr; i++)
1860 		*p++ = strdup(old_power_args[i]);
1861 
1862 	for (j = 0; j < (unsigned int)argc; j++)
1863 		*p++ = argv[j];
1864 
1865 	return cmd_record(rec_argc, rec_argv, NULL);
1866 }
1867 
1868 static int
1869 parse_process(const struct option *opt __maybe_unused, const char *arg,
1870 	      int __maybe_unused unset)
1871 {
1872 	if (arg)
1873 		add_process_filter(arg);
1874 	return 0;
1875 }
1876 
1877 static int
1878 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1879 		int __maybe_unused unset)
1880 {
1881 	unsigned long duration = strtoul(arg, NULL, 0);
1882 
1883 	if (svg_highlight || svg_highlight_name)
1884 		return -1;
1885 
1886 	if (duration)
1887 		svg_highlight = duration;
1888 	else
1889 		svg_highlight_name = strdup(arg);
1890 
1891 	return 0;
1892 }
1893 
1894 static int
1895 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1896 {
1897 	char unit = 'n';
1898 	u64 *value = opt->value;
1899 
1900 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1901 		switch (unit) {
1902 		case 'm':
1903 			*value *= 1000000;
1904 			break;
1905 		case 'u':
1906 			*value *= 1000;
1907 			break;
1908 		case 'n':
1909 			break;
1910 		default:
1911 			return -1;
1912 		}
1913 	}
1914 
1915 	return 0;
1916 }
1917 
1918 int cmd_timechart(int argc, const char **argv,
1919 		  const char *prefix __maybe_unused)
1920 {
1921 	struct timechart tchart = {
1922 		.tool = {
1923 			.comm		 = process_comm_event,
1924 			.fork		 = process_fork_event,
1925 			.exit		 = process_exit_event,
1926 			.sample		 = process_sample_event,
1927 			.ordered_events	 = true,
1928 		},
1929 		.proc_num = 15,
1930 		.min_time = 1000000,
1931 		.merge_dist = 1000,
1932 	};
1933 	const char *output_name = "output.svg";
1934 	const struct option timechart_options[] = {
1935 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1936 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1937 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1938 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1939 		      "highlight tasks. Pass duration in ns or process name.",
1940 		       parse_highlight),
1941 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1942 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only,
1943 		    "output processes data only"),
1944 	OPT_CALLBACK('p', "process", NULL, "process",
1945 		      "process selector. Pass a pid or process name.",
1946 		       parse_process),
1947 	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
1948 		    "Look for files with symbols relative to this directory"),
1949 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1950 		    "min. number of tasks to print"),
1951 	OPT_BOOLEAN('t', "topology", &tchart.topology,
1952 		    "sort CPUs according to topology"),
1953 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1954 		    "skip EAGAIN errors"),
1955 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1956 		     "all IO faster than min-time will visually appear longer",
1957 		     parse_time),
1958 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1959 		     "merge events that are merge-dist us apart",
1960 		     parse_time),
1961 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1962 	OPT_END()
1963 	};
1964 	const char * const timechart_subcommands[] = { "record", NULL };
1965 	const char *timechart_usage[] = {
1966 		"perf timechart [<options>] {record}",
1967 		NULL
1968 	};
1969 
1970 	const struct option timechart_record_options[] = {
1971 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1972 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only,
1973 		    "output processes data only"),
1974 	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
1975 		    "record only IO data"),
1976 	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
1977 	OPT_END()
1978 	};
1979 	const char * const timechart_record_usage[] = {
1980 		"perf timechart record [<options>]",
1981 		NULL
1982 	};
1983 	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
1984 			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1985 
1986 	if (tchart.power_only && tchart.tasks_only) {
1987 		pr_err("-P and -T options cannot be used at the same time.\n");
1988 		return -1;
1989 	}
1990 
1991 	if (argc && !strncmp(argv[0], "rec", 3)) {
1992 		argc = parse_options(argc, argv, timechart_record_options,
1993 				     timechart_record_usage,
1994 				     PARSE_OPT_STOP_AT_NON_OPTION);
1995 
1996 		if (tchart.power_only && tchart.tasks_only) {
1997 			pr_err("-P and -T options cannot be used at the same time.\n");
1998 			return -1;
1999 		}
2000 
2001 		if (tchart.io_only)
2002 			return timechart__io_record(argc, argv);
2003 		else
2004 			return timechart__record(&tchart, argc, argv);
2005 	} else if (argc)
2006 		usage_with_options(timechart_usage, timechart_options);
2007 
2008 	setup_pager();
2009 
2010 	return __cmd_timechart(&tchart, output_name);
2011 }
2012