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 #include "util/string2.h"
39 #include "util/tracepoint.h"
40 #include <linux/err.h>
41 
42 #ifdef LACKS_OPEN_MEMSTREAM_PROTOTYPE
43 FILE *open_memstream(char **ptr, size_t *sizeloc);
44 #endif
45 
46 #define SUPPORT_OLD_POWER_EVENTS 1
47 #define PWR_EVENT_EXIT -1
48 
49 struct per_pid;
50 struct power_event;
51 struct wake_event;
52 
53 struct timechart {
54 	struct perf_tool	tool;
55 	struct per_pid		*all_data;
56 	struct power_event	*power_events;
57 	struct wake_event	*wake_events;
58 	int			proc_num;
59 	unsigned int		numcpus;
60 	u64			min_freq,	/* Lowest CPU frequency seen */
61 				max_freq,	/* Highest CPU frequency seen */
62 				turbo_frequency,
63 				first_time, last_time;
64 	bool			power_only,
65 				tasks_only,
66 				with_backtrace,
67 				topology;
68 	bool			force;
69 	/* IO related settings */
70 	bool			io_only,
71 				skip_eagain;
72 	u64			io_events;
73 	u64			min_time,
74 				merge_dist;
75 };
76 
77 struct per_pidcomm;
78 struct cpu_sample;
79 struct io_sample;
80 
81 /*
82  * Datastructure layout:
83  * We keep an list of "pid"s, matching the kernels notion of a task struct.
84  * Each "pid" entry, has a list of "comm"s.
85  *	this is because we want to track different programs different, while
86  *	exec will reuse the original pid (by design).
87  * Each comm has a list of samples that will be used to draw
88  * final graph.
89  */
90 
91 struct per_pid {
92 	struct per_pid *next;
93 
94 	int		pid;
95 	int		ppid;
96 
97 	u64		start_time;
98 	u64		end_time;
99 	u64		total_time;
100 	u64		total_bytes;
101 	int		display;
102 
103 	struct per_pidcomm *all;
104 	struct per_pidcomm *current;
105 };
106 
107 
108 struct per_pidcomm {
109 	struct per_pidcomm *next;
110 
111 	u64		start_time;
112 	u64		end_time;
113 	u64		total_time;
114 	u64		max_bytes;
115 	u64		total_bytes;
116 
117 	int		Y;
118 	int		display;
119 
120 	long		state;
121 	u64		state_since;
122 
123 	char		*comm;
124 
125 	struct cpu_sample *samples;
126 	struct io_sample  *io_samples;
127 };
128 
129 struct sample_wrapper {
130 	struct sample_wrapper *next;
131 
132 	u64		timestamp;
133 	unsigned char	data[];
134 };
135 
136 #define TYPE_NONE	0
137 #define TYPE_RUNNING	1
138 #define TYPE_WAITING	2
139 #define TYPE_BLOCKED	3
140 
141 struct cpu_sample {
142 	struct cpu_sample *next;
143 
144 	u64 start_time;
145 	u64 end_time;
146 	int type;
147 	int cpu;
148 	const char *backtrace;
149 };
150 
151 enum {
152 	IOTYPE_READ,
153 	IOTYPE_WRITE,
154 	IOTYPE_SYNC,
155 	IOTYPE_TX,
156 	IOTYPE_RX,
157 	IOTYPE_POLL,
158 };
159 
160 struct io_sample {
161 	struct io_sample *next;
162 
163 	u64 start_time;
164 	u64 end_time;
165 	u64 bytes;
166 	int type;
167 	int fd;
168 	int err;
169 	int merges;
170 };
171 
172 #define CSTATE 1
173 #define PSTATE 2
174 
175 struct power_event {
176 	struct power_event *next;
177 	int type;
178 	int state;
179 	u64 start_time;
180 	u64 end_time;
181 	int cpu;
182 };
183 
184 struct wake_event {
185 	struct wake_event *next;
186 	int waker;
187 	int wakee;
188 	u64 time;
189 	const char *backtrace;
190 };
191 
192 struct process_filter {
193 	char			*name;
194 	int			pid;
195 	struct process_filter	*next;
196 };
197 
198 static struct process_filter *process_filter;
199 
200 
201 static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
202 {
203 	struct per_pid *cursor = tchart->all_data;
204 
205 	while (cursor) {
206 		if (cursor->pid == pid)
207 			return cursor;
208 		cursor = cursor->next;
209 	}
210 	cursor = zalloc(sizeof(*cursor));
211 	assert(cursor != NULL);
212 	cursor->pid = pid;
213 	cursor->next = tchart->all_data;
214 	tchart->all_data = cursor;
215 	return cursor;
216 }
217 
218 static struct per_pidcomm *create_pidcomm(struct per_pid *p)
219 {
220 	struct per_pidcomm *c;
221 
222 	c = zalloc(sizeof(*c));
223 	if (!c)
224 		return NULL;
225 	p->current = c;
226 	c->next = p->all;
227 	p->all = c;
228 	return c;
229 }
230 
231 static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
232 {
233 	struct per_pid *p;
234 	struct per_pidcomm *c;
235 	p = find_create_pid(tchart, pid);
236 	c = p->all;
237 	while (c) {
238 		if (c->comm && strcmp(c->comm, comm) == 0) {
239 			p->current = c;
240 			return;
241 		}
242 		if (!c->comm) {
243 			c->comm = strdup(comm);
244 			p->current = c;
245 			return;
246 		}
247 		c = c->next;
248 	}
249 	c = create_pidcomm(p);
250 	assert(c != NULL);
251 	c->comm = strdup(comm);
252 }
253 
254 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
255 {
256 	struct per_pid *p, *pp;
257 	p = find_create_pid(tchart, pid);
258 	pp = find_create_pid(tchart, ppid);
259 	p->ppid = ppid;
260 	if (pp->current && pp->current->comm && !p->current)
261 		pid_set_comm(tchart, pid, pp->current->comm);
262 
263 	p->start_time = timestamp;
264 	if (p->current && !p->current->start_time) {
265 		p->current->start_time = timestamp;
266 		p->current->state_since = timestamp;
267 	}
268 }
269 
270 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
271 {
272 	struct per_pid *p;
273 	p = find_create_pid(tchart, pid);
274 	p->end_time = timestamp;
275 	if (p->current)
276 		p->current->end_time = timestamp;
277 }
278 
279 static void pid_put_sample(struct timechart *tchart, int pid, int type,
280 			   unsigned int cpu, u64 start, u64 end,
281 			   const char *backtrace)
282 {
283 	struct per_pid *p;
284 	struct per_pidcomm *c;
285 	struct cpu_sample *sample;
286 
287 	p = find_create_pid(tchart, pid);
288 	c = p->current;
289 	if (!c) {
290 		c = create_pidcomm(p);
291 		assert(c != NULL);
292 	}
293 
294 	sample = zalloc(sizeof(*sample));
295 	assert(sample != NULL);
296 	sample->start_time = start;
297 	sample->end_time = end;
298 	sample->type = type;
299 	sample->next = c->samples;
300 	sample->cpu = cpu;
301 	sample->backtrace = backtrace;
302 	c->samples = sample;
303 
304 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
305 		c->total_time += (end-start);
306 		p->total_time += (end-start);
307 	}
308 
309 	if (c->start_time == 0 || c->start_time > start)
310 		c->start_time = start;
311 	if (p->start_time == 0 || p->start_time > start)
312 		p->start_time = start;
313 }
314 
315 #define MAX_CPUS 4096
316 
317 static u64 cpus_cstate_start_times[MAX_CPUS];
318 static int cpus_cstate_state[MAX_CPUS];
319 static u64 cpus_pstate_start_times[MAX_CPUS];
320 static u64 cpus_pstate_state[MAX_CPUS];
321 
322 static int process_comm_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_set_comm(tchart, event->comm.tid, event->comm.comm);
329 	return 0;
330 }
331 
332 static int process_fork_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_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
339 	return 0;
340 }
341 
342 static int process_exit_event(struct perf_tool *tool,
343 			      union perf_event *event,
344 			      struct perf_sample *sample __maybe_unused,
345 			      struct machine *machine __maybe_unused)
346 {
347 	struct timechart *tchart = container_of(tool, struct timechart, tool);
348 	pid_exit(tchart, event->fork.pid, event->fork.time);
349 	return 0;
350 }
351 
352 #ifdef SUPPORT_OLD_POWER_EVENTS
353 static int use_old_power_events;
354 #endif
355 
356 static void c_state_start(int cpu, u64 timestamp, int state)
357 {
358 	cpus_cstate_start_times[cpu] = timestamp;
359 	cpus_cstate_state[cpu] = state;
360 }
361 
362 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
363 {
364 	struct power_event *pwr = zalloc(sizeof(*pwr));
365 
366 	if (!pwr)
367 		return;
368 
369 	pwr->state = cpus_cstate_state[cpu];
370 	pwr->start_time = cpus_cstate_start_times[cpu];
371 	pwr->end_time = timestamp;
372 	pwr->cpu = cpu;
373 	pwr->type = CSTATE;
374 	pwr->next = tchart->power_events;
375 
376 	tchart->power_events = pwr;
377 }
378 
379 static struct power_event *p_state_end(struct timechart *tchart, int cpu,
380 					u64 timestamp)
381 {
382 	struct power_event *pwr = zalloc(sizeof(*pwr));
383 
384 	if (!pwr)
385 		return NULL;
386 
387 	pwr->state = cpus_pstate_state[cpu];
388 	pwr->start_time = cpus_pstate_start_times[cpu];
389 	pwr->end_time = timestamp;
390 	pwr->cpu = cpu;
391 	pwr->type = PSTATE;
392 	pwr->next = tchart->power_events;
393 	if (!pwr->start_time)
394 		pwr->start_time = tchart->first_time;
395 
396 	tchart->power_events = pwr;
397 	return pwr;
398 }
399 
400 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
401 {
402 	struct power_event *pwr;
403 
404 	if (new_freq > 8000000) /* detect invalid data */
405 		return;
406 
407 	pwr = p_state_end(tchart, cpu, timestamp);
408 	if (!pwr)
409 		return;
410 
411 	cpus_pstate_state[cpu] = new_freq;
412 	cpus_pstate_start_times[cpu] = timestamp;
413 
414 	if ((u64)new_freq > tchart->max_freq)
415 		tchart->max_freq = new_freq;
416 
417 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
418 		tchart->min_freq = new_freq;
419 
420 	if (new_freq == tchart->max_freq - 1000)
421 		tchart->turbo_frequency = tchart->max_freq;
422 }
423 
424 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
425 			 int waker, int wakee, u8 flags, const char *backtrace)
426 {
427 	struct per_pid *p;
428 	struct wake_event *we = zalloc(sizeof(*we));
429 
430 	if (!we)
431 		return;
432 
433 	we->time = timestamp;
434 	we->waker = waker;
435 	we->backtrace = backtrace;
436 
437 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
438 		we->waker = -1;
439 
440 	we->wakee = wakee;
441 	we->next = tchart->wake_events;
442 	tchart->wake_events = we;
443 	p = find_create_pid(tchart, we->wakee);
444 
445 	if (p && p->current && p->current->state == TYPE_NONE) {
446 		p->current->state_since = timestamp;
447 		p->current->state = TYPE_WAITING;
448 	}
449 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
450 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
451 			       p->current->state_since, timestamp, NULL);
452 		p->current->state_since = timestamp;
453 		p->current->state = TYPE_WAITING;
454 	}
455 }
456 
457 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
458 			 int prev_pid, int next_pid, u64 prev_state,
459 			 const char *backtrace)
460 {
461 	struct per_pid *p = NULL, *prev_p;
462 
463 	prev_p = find_create_pid(tchart, prev_pid);
464 
465 	p = find_create_pid(tchart, next_pid);
466 
467 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
468 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
469 			       prev_p->current->state_since, timestamp,
470 			       backtrace);
471 	if (p && p->current) {
472 		if (p->current->state != TYPE_NONE)
473 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
474 				       p->current->state_since, timestamp,
475 				       backtrace);
476 
477 		p->current->state_since = timestamp;
478 		p->current->state = TYPE_RUNNING;
479 	}
480 
481 	if (prev_p->current) {
482 		prev_p->current->state = TYPE_NONE;
483 		prev_p->current->state_since = timestamp;
484 		if (prev_state & 2)
485 			prev_p->current->state = TYPE_BLOCKED;
486 		if (prev_state == 0)
487 			prev_p->current->state = TYPE_WAITING;
488 	}
489 }
490 
491 static const char *cat_backtrace(union perf_event *event,
492 				 struct perf_sample *sample,
493 				 struct machine *machine)
494 {
495 	struct addr_location al;
496 	unsigned int i;
497 	char *p = NULL;
498 	size_t p_len;
499 	u8 cpumode = PERF_RECORD_MISC_USER;
500 	struct addr_location tal;
501 	struct ip_callchain *chain = sample->callchain;
502 	FILE *f = open_memstream(&p, &p_len);
503 
504 	if (!f) {
505 		perror("open_memstream error");
506 		return NULL;
507 	}
508 
509 	if (!chain)
510 		goto exit;
511 
512 	if (machine__resolve(machine, &al, sample) < 0) {
513 		fprintf(stderr, "problem processing %d event, skipping it.\n",
514 			event->header.type);
515 		goto exit;
516 	}
517 
518 	for (i = 0; i < chain->nr; i++) {
519 		u64 ip;
520 
521 		if (callchain_param.order == ORDER_CALLEE)
522 			ip = chain->ips[i];
523 		else
524 			ip = chain->ips[chain->nr - i - 1];
525 
526 		if (ip >= PERF_CONTEXT_MAX) {
527 			switch (ip) {
528 			case PERF_CONTEXT_HV:
529 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
530 				break;
531 			case PERF_CONTEXT_KERNEL:
532 				cpumode = PERF_RECORD_MISC_KERNEL;
533 				break;
534 			case PERF_CONTEXT_USER:
535 				cpumode = PERF_RECORD_MISC_USER;
536 				break;
537 			default:
538 				pr_debug("invalid callchain context: "
539 					 "%"PRId64"\n", (s64) ip);
540 
541 				/*
542 				 * It seems the callchain is corrupted.
543 				 * Discard all.
544 				 */
545 				zfree(&p);
546 				goto exit_put;
547 			}
548 			continue;
549 		}
550 
551 		tal.filtered = 0;
552 		if (thread__find_symbol(al.thread, cpumode, ip, &tal))
553 			fprintf(f, "..... %016" PRIx64 " %s\n", ip, tal.sym->name);
554 		else
555 			fprintf(f, "..... %016" PRIx64 "\n", ip);
556 	}
557 exit_put:
558 	addr_location__put(&al);
559 exit:
560 	fclose(f);
561 
562 	return p;
563 }
564 
565 typedef int (*tracepoint_handler)(struct timechart *tchart,
566 				  struct evsel *evsel,
567 				  struct perf_sample *sample,
568 				  const char *backtrace);
569 
570 static int process_sample_event(struct perf_tool *tool,
571 				union perf_event *event,
572 				struct perf_sample *sample,
573 				struct evsel *evsel,
574 				struct machine *machine)
575 {
576 	struct timechart *tchart = container_of(tool, struct timechart, tool);
577 
578 	if (evsel->core.attr.sample_type & PERF_SAMPLE_TIME) {
579 		if (!tchart->first_time || tchart->first_time > sample->time)
580 			tchart->first_time = sample->time;
581 		if (tchart->last_time < sample->time)
582 			tchart->last_time = sample->time;
583 	}
584 
585 	if (evsel->handler != NULL) {
586 		tracepoint_handler f = evsel->handler;
587 		return f(tchart, evsel, sample,
588 			 cat_backtrace(event, sample, machine));
589 	}
590 
591 	return 0;
592 }
593 
594 static int
595 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
596 			struct evsel *evsel,
597 			struct perf_sample *sample,
598 			const char *backtrace __maybe_unused)
599 {
600 	u32 state  = evsel__intval(evsel, sample, "state");
601 	u32 cpu_id = evsel__intval(evsel, sample, "cpu_id");
602 
603 	if (state == (u32)PWR_EVENT_EXIT)
604 		c_state_end(tchart, cpu_id, sample->time);
605 	else
606 		c_state_start(cpu_id, sample->time, state);
607 	return 0;
608 }
609 
610 static int
611 process_sample_cpu_frequency(struct timechart *tchart,
612 			     struct evsel *evsel,
613 			     struct perf_sample *sample,
614 			     const char *backtrace __maybe_unused)
615 {
616 	u32 state  = evsel__intval(evsel, sample, "state");
617 	u32 cpu_id = evsel__intval(evsel, sample, "cpu_id");
618 
619 	p_state_change(tchart, cpu_id, sample->time, state);
620 	return 0;
621 }
622 
623 static int
624 process_sample_sched_wakeup(struct timechart *tchart,
625 			    struct evsel *evsel,
626 			    struct perf_sample *sample,
627 			    const char *backtrace)
628 {
629 	u8 flags  = evsel__intval(evsel, sample, "common_flags");
630 	int waker = evsel__intval(evsel, sample, "common_pid");
631 	int wakee = evsel__intval(evsel, sample, "pid");
632 
633 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
634 	return 0;
635 }
636 
637 static int
638 process_sample_sched_switch(struct timechart *tchart,
639 			    struct evsel *evsel,
640 			    struct perf_sample *sample,
641 			    const char *backtrace)
642 {
643 	int prev_pid   = evsel__intval(evsel, sample, "prev_pid");
644 	int next_pid   = evsel__intval(evsel, sample, "next_pid");
645 	u64 prev_state = evsel__intval(evsel, sample, "prev_state");
646 
647 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
648 		     prev_state, backtrace);
649 	return 0;
650 }
651 
652 #ifdef SUPPORT_OLD_POWER_EVENTS
653 static int
654 process_sample_power_start(struct timechart *tchart __maybe_unused,
655 			   struct evsel *evsel,
656 			   struct perf_sample *sample,
657 			   const char *backtrace __maybe_unused)
658 {
659 	u64 cpu_id = evsel__intval(evsel, sample, "cpu_id");
660 	u64 value  = evsel__intval(evsel, sample, "value");
661 
662 	c_state_start(cpu_id, sample->time, value);
663 	return 0;
664 }
665 
666 static int
667 process_sample_power_end(struct timechart *tchart,
668 			 struct evsel *evsel __maybe_unused,
669 			 struct perf_sample *sample,
670 			 const char *backtrace __maybe_unused)
671 {
672 	c_state_end(tchart, sample->cpu, sample->time);
673 	return 0;
674 }
675 
676 static int
677 process_sample_power_frequency(struct timechart *tchart,
678 			       struct evsel *evsel,
679 			       struct perf_sample *sample,
680 			       const char *backtrace __maybe_unused)
681 {
682 	u64 cpu_id = evsel__intval(evsel, sample, "cpu_id");
683 	u64 value  = evsel__intval(evsel, sample, "value");
684 
685 	p_state_change(tchart, cpu_id, sample->time, value);
686 	return 0;
687 }
688 #endif /* SUPPORT_OLD_POWER_EVENTS */
689 
690 /*
691  * After the last sample we need to wrap up the current C/P state
692  * and close out each CPU for these.
693  */
694 static void end_sample_processing(struct timechart *tchart)
695 {
696 	u64 cpu;
697 	struct power_event *pwr;
698 
699 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
700 		/* C state */
701 #if 0
702 		pwr = zalloc(sizeof(*pwr));
703 		if (!pwr)
704 			return;
705 
706 		pwr->state = cpus_cstate_state[cpu];
707 		pwr->start_time = cpus_cstate_start_times[cpu];
708 		pwr->end_time = tchart->last_time;
709 		pwr->cpu = cpu;
710 		pwr->type = CSTATE;
711 		pwr->next = tchart->power_events;
712 
713 		tchart->power_events = pwr;
714 #endif
715 		/* P state */
716 
717 		pwr = p_state_end(tchart, cpu, tchart->last_time);
718 		if (!pwr)
719 			return;
720 
721 		if (!pwr->state)
722 			pwr->state = tchart->min_freq;
723 	}
724 }
725 
726 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
727 			       u64 start, int fd)
728 {
729 	struct per_pid *p = find_create_pid(tchart, pid);
730 	struct per_pidcomm *c = p->current;
731 	struct io_sample *sample;
732 	struct io_sample *prev;
733 
734 	if (!c) {
735 		c = create_pidcomm(p);
736 		if (!c)
737 			return -ENOMEM;
738 	}
739 
740 	prev = c->io_samples;
741 
742 	if (prev && prev->start_time && !prev->end_time) {
743 		pr_warning("Skip invalid start event: "
744 			   "previous event already started!\n");
745 
746 		/* remove previous event that has been started,
747 		 * we are not sure we will ever get an end for it */
748 		c->io_samples = prev->next;
749 		free(prev);
750 		return 0;
751 	}
752 
753 	sample = zalloc(sizeof(*sample));
754 	if (!sample)
755 		return -ENOMEM;
756 	sample->start_time = start;
757 	sample->type = type;
758 	sample->fd = fd;
759 	sample->next = c->io_samples;
760 	c->io_samples = sample;
761 
762 	if (c->start_time == 0 || c->start_time > start)
763 		c->start_time = start;
764 
765 	return 0;
766 }
767 
768 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
769 			     u64 end, long ret)
770 {
771 	struct per_pid *p = find_create_pid(tchart, pid);
772 	struct per_pidcomm *c = p->current;
773 	struct io_sample *sample, *prev;
774 
775 	if (!c) {
776 		pr_warning("Invalid pidcomm!\n");
777 		return -1;
778 	}
779 
780 	sample = c->io_samples;
781 
782 	if (!sample) /* skip partially captured events */
783 		return 0;
784 
785 	if (sample->end_time) {
786 		pr_warning("Skip invalid end event: "
787 			   "previous event already ended!\n");
788 		return 0;
789 	}
790 
791 	if (sample->type != type) {
792 		pr_warning("Skip invalid end event: invalid event type!\n");
793 		return 0;
794 	}
795 
796 	sample->end_time = end;
797 	prev = sample->next;
798 
799 	/* we want to be able to see small and fast transfers, so make them
800 	 * at least min_time long, but don't overlap them */
801 	if (sample->end_time - sample->start_time < tchart->min_time)
802 		sample->end_time = sample->start_time + tchart->min_time;
803 	if (prev && sample->start_time < prev->end_time) {
804 		if (prev->err) /* try to make errors more visible */
805 			sample->start_time = prev->end_time;
806 		else
807 			prev->end_time = sample->start_time;
808 	}
809 
810 	if (ret < 0) {
811 		sample->err = ret;
812 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
813 		   type == IOTYPE_TX || type == IOTYPE_RX) {
814 
815 		if ((u64)ret > c->max_bytes)
816 			c->max_bytes = ret;
817 
818 		c->total_bytes += ret;
819 		p->total_bytes += ret;
820 		sample->bytes = ret;
821 	}
822 
823 	/* merge two requests to make svg smaller and render-friendly */
824 	if (prev &&
825 	    prev->type == sample->type &&
826 	    prev->err == sample->err &&
827 	    prev->fd == sample->fd &&
828 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
829 
830 		sample->bytes += prev->bytes;
831 		sample->merges += prev->merges + 1;
832 
833 		sample->start_time = prev->start_time;
834 		sample->next = prev->next;
835 		free(prev);
836 
837 		if (!sample->err && sample->bytes > c->max_bytes)
838 			c->max_bytes = sample->bytes;
839 	}
840 
841 	tchart->io_events++;
842 
843 	return 0;
844 }
845 
846 static int
847 process_enter_read(struct timechart *tchart,
848 		   struct evsel *evsel,
849 		   struct perf_sample *sample)
850 {
851 	long fd = evsel__intval(evsel, sample, "fd");
852 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
853 				   sample->time, fd);
854 }
855 
856 static int
857 process_exit_read(struct timechart *tchart,
858 		  struct evsel *evsel,
859 		  struct perf_sample *sample)
860 {
861 	long ret = evsel__intval(evsel, sample, "ret");
862 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
863 				 sample->time, ret);
864 }
865 
866 static int
867 process_enter_write(struct timechart *tchart,
868 		    struct evsel *evsel,
869 		    struct perf_sample *sample)
870 {
871 	long fd = evsel__intval(evsel, sample, "fd");
872 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
873 				   sample->time, fd);
874 }
875 
876 static int
877 process_exit_write(struct timechart *tchart,
878 		   struct evsel *evsel,
879 		   struct perf_sample *sample)
880 {
881 	long ret = evsel__intval(evsel, sample, "ret");
882 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
883 				 sample->time, ret);
884 }
885 
886 static int
887 process_enter_sync(struct timechart *tchart,
888 		   struct evsel *evsel,
889 		   struct perf_sample *sample)
890 {
891 	long fd = evsel__intval(evsel, sample, "fd");
892 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
893 				   sample->time, fd);
894 }
895 
896 static int
897 process_exit_sync(struct timechart *tchart,
898 		  struct evsel *evsel,
899 		  struct perf_sample *sample)
900 {
901 	long ret = evsel__intval(evsel, sample, "ret");
902 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
903 				 sample->time, ret);
904 }
905 
906 static int
907 process_enter_tx(struct timechart *tchart,
908 		 struct evsel *evsel,
909 		 struct perf_sample *sample)
910 {
911 	long fd = evsel__intval(evsel, sample, "fd");
912 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
913 				   sample->time, fd);
914 }
915 
916 static int
917 process_exit_tx(struct timechart *tchart,
918 		struct evsel *evsel,
919 		struct perf_sample *sample)
920 {
921 	long ret = evsel__intval(evsel, sample, "ret");
922 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
923 				 sample->time, ret);
924 }
925 
926 static int
927 process_enter_rx(struct timechart *tchart,
928 		 struct evsel *evsel,
929 		 struct perf_sample *sample)
930 {
931 	long fd = evsel__intval(evsel, sample, "fd");
932 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
933 				   sample->time, fd);
934 }
935 
936 static int
937 process_exit_rx(struct timechart *tchart,
938 		struct evsel *evsel,
939 		struct perf_sample *sample)
940 {
941 	long ret = evsel__intval(evsel, sample, "ret");
942 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
943 				 sample->time, ret);
944 }
945 
946 static int
947 process_enter_poll(struct timechart *tchart,
948 		   struct evsel *evsel,
949 		   struct perf_sample *sample)
950 {
951 	long fd = evsel__intval(evsel, sample, "fd");
952 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
953 				   sample->time, fd);
954 }
955 
956 static int
957 process_exit_poll(struct timechart *tchart,
958 		  struct evsel *evsel,
959 		  struct perf_sample *sample)
960 {
961 	long ret = evsel__intval(evsel, sample, "ret");
962 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
963 				 sample->time, ret);
964 }
965 
966 /*
967  * Sort the pid datastructure
968  */
969 static void sort_pids(struct timechart *tchart)
970 {
971 	struct per_pid *new_list, *p, *cursor, *prev;
972 	/* sort by ppid first, then by pid, lowest to highest */
973 
974 	new_list = NULL;
975 
976 	while (tchart->all_data) {
977 		p = tchart->all_data;
978 		tchart->all_data = p->next;
979 		p->next = NULL;
980 
981 		if (new_list == NULL) {
982 			new_list = p;
983 			p->next = NULL;
984 			continue;
985 		}
986 		prev = NULL;
987 		cursor = new_list;
988 		while (cursor) {
989 			if (cursor->ppid > p->ppid ||
990 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
991 				/* must insert before */
992 				if (prev) {
993 					p->next = prev->next;
994 					prev->next = p;
995 					cursor = NULL;
996 					continue;
997 				} else {
998 					p->next = new_list;
999 					new_list = p;
1000 					cursor = NULL;
1001 					continue;
1002 				}
1003 			}
1004 
1005 			prev = cursor;
1006 			cursor = cursor->next;
1007 			if (!cursor)
1008 				prev->next = p;
1009 		}
1010 	}
1011 	tchart->all_data = new_list;
1012 }
1013 
1014 
1015 static void draw_c_p_states(struct timechart *tchart)
1016 {
1017 	struct power_event *pwr;
1018 	pwr = tchart->power_events;
1019 
1020 	/*
1021 	 * two pass drawing so that the P state bars are on top of the C state blocks
1022 	 */
1023 	while (pwr) {
1024 		if (pwr->type == CSTATE)
1025 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1026 		pwr = pwr->next;
1027 	}
1028 
1029 	pwr = tchart->power_events;
1030 	while (pwr) {
1031 		if (pwr->type == PSTATE) {
1032 			if (!pwr->state)
1033 				pwr->state = tchart->min_freq;
1034 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1035 		}
1036 		pwr = pwr->next;
1037 	}
1038 }
1039 
1040 static void draw_wakeups(struct timechart *tchart)
1041 {
1042 	struct wake_event *we;
1043 	struct per_pid *p;
1044 	struct per_pidcomm *c;
1045 
1046 	we = tchart->wake_events;
1047 	while (we) {
1048 		int from = 0, to = 0;
1049 		char *task_from = NULL, *task_to = NULL;
1050 
1051 		/* locate the column of the waker and wakee */
1052 		p = tchart->all_data;
1053 		while (p) {
1054 			if (p->pid == we->waker || p->pid == we->wakee) {
1055 				c = p->all;
1056 				while (c) {
1057 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1058 						if (p->pid == we->waker && !from) {
1059 							from = c->Y;
1060 							task_from = strdup(c->comm);
1061 						}
1062 						if (p->pid == we->wakee && !to) {
1063 							to = c->Y;
1064 							task_to = strdup(c->comm);
1065 						}
1066 					}
1067 					c = c->next;
1068 				}
1069 				c = p->all;
1070 				while (c) {
1071 					if (p->pid == we->waker && !from) {
1072 						from = c->Y;
1073 						task_from = strdup(c->comm);
1074 					}
1075 					if (p->pid == we->wakee && !to) {
1076 						to = c->Y;
1077 						task_to = strdup(c->comm);
1078 					}
1079 					c = c->next;
1080 				}
1081 			}
1082 			p = p->next;
1083 		}
1084 
1085 		if (!task_from) {
1086 			task_from = malloc(40);
1087 			sprintf(task_from, "[%i]", we->waker);
1088 		}
1089 		if (!task_to) {
1090 			task_to = malloc(40);
1091 			sprintf(task_to, "[%i]", we->wakee);
1092 		}
1093 
1094 		if (we->waker == -1)
1095 			svg_interrupt(we->time, to, we->backtrace);
1096 		else if (from && to && abs(from - to) == 1)
1097 			svg_wakeline(we->time, from, to, we->backtrace);
1098 		else
1099 			svg_partial_wakeline(we->time, from, task_from, to,
1100 					     task_to, we->backtrace);
1101 		we = we->next;
1102 
1103 		free(task_from);
1104 		free(task_to);
1105 	}
1106 }
1107 
1108 static void draw_cpu_usage(struct timechart *tchart)
1109 {
1110 	struct per_pid *p;
1111 	struct per_pidcomm *c;
1112 	struct cpu_sample *sample;
1113 	p = tchart->all_data;
1114 	while (p) {
1115 		c = p->all;
1116 		while (c) {
1117 			sample = c->samples;
1118 			while (sample) {
1119 				if (sample->type == TYPE_RUNNING) {
1120 					svg_process(sample->cpu,
1121 						    sample->start_time,
1122 						    sample->end_time,
1123 						    p->pid,
1124 						    c->comm,
1125 						    sample->backtrace);
1126 				}
1127 
1128 				sample = sample->next;
1129 			}
1130 			c = c->next;
1131 		}
1132 		p = p->next;
1133 	}
1134 }
1135 
1136 static void draw_io_bars(struct timechart *tchart)
1137 {
1138 	const char *suf;
1139 	double bytes;
1140 	char comm[256];
1141 	struct per_pid *p;
1142 	struct per_pidcomm *c;
1143 	struct io_sample *sample;
1144 	int Y = 1;
1145 
1146 	p = tchart->all_data;
1147 	while (p) {
1148 		c = p->all;
1149 		while (c) {
1150 			if (!c->display) {
1151 				c->Y = 0;
1152 				c = c->next;
1153 				continue;
1154 			}
1155 
1156 			svg_box(Y, c->start_time, c->end_time, "process3");
1157 			sample = c->io_samples;
1158 			for (sample = c->io_samples; sample; sample = sample->next) {
1159 				double h = (double)sample->bytes / c->max_bytes;
1160 
1161 				if (tchart->skip_eagain &&
1162 				    sample->err == -EAGAIN)
1163 					continue;
1164 
1165 				if (sample->err)
1166 					h = 1;
1167 
1168 				if (sample->type == IOTYPE_SYNC)
1169 					svg_fbox(Y,
1170 						sample->start_time,
1171 						sample->end_time,
1172 						1,
1173 						sample->err ? "error" : "sync",
1174 						sample->fd,
1175 						sample->err,
1176 						sample->merges);
1177 				else if (sample->type == IOTYPE_POLL)
1178 					svg_fbox(Y,
1179 						sample->start_time,
1180 						sample->end_time,
1181 						1,
1182 						sample->err ? "error" : "poll",
1183 						sample->fd,
1184 						sample->err,
1185 						sample->merges);
1186 				else if (sample->type == IOTYPE_READ)
1187 					svg_ubox(Y,
1188 						sample->start_time,
1189 						sample->end_time,
1190 						h,
1191 						sample->err ? "error" : "disk",
1192 						sample->fd,
1193 						sample->err,
1194 						sample->merges);
1195 				else if (sample->type == IOTYPE_WRITE)
1196 					svg_lbox(Y,
1197 						sample->start_time,
1198 						sample->end_time,
1199 						h,
1200 						sample->err ? "error" : "disk",
1201 						sample->fd,
1202 						sample->err,
1203 						sample->merges);
1204 				else if (sample->type == IOTYPE_RX)
1205 					svg_ubox(Y,
1206 						sample->start_time,
1207 						sample->end_time,
1208 						h,
1209 						sample->err ? "error" : "net",
1210 						sample->fd,
1211 						sample->err,
1212 						sample->merges);
1213 				else if (sample->type == IOTYPE_TX)
1214 					svg_lbox(Y,
1215 						sample->start_time,
1216 						sample->end_time,
1217 						h,
1218 						sample->err ? "error" : "net",
1219 						sample->fd,
1220 						sample->err,
1221 						sample->merges);
1222 			}
1223 
1224 			suf = "";
1225 			bytes = c->total_bytes;
1226 			if (bytes > 1024) {
1227 				bytes = bytes / 1024;
1228 				suf = "K";
1229 			}
1230 			if (bytes > 1024) {
1231 				bytes = bytes / 1024;
1232 				suf = "M";
1233 			}
1234 			if (bytes > 1024) {
1235 				bytes = bytes / 1024;
1236 				suf = "G";
1237 			}
1238 
1239 
1240 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1241 			svg_text(Y, c->start_time, comm);
1242 
1243 			c->Y = Y;
1244 			Y++;
1245 			c = c->next;
1246 		}
1247 		p = p->next;
1248 	}
1249 }
1250 
1251 static void draw_process_bars(struct timechart *tchart)
1252 {
1253 	struct per_pid *p;
1254 	struct per_pidcomm *c;
1255 	struct cpu_sample *sample;
1256 	int Y = 0;
1257 
1258 	Y = 2 * tchart->numcpus + 2;
1259 
1260 	p = tchart->all_data;
1261 	while (p) {
1262 		c = p->all;
1263 		while (c) {
1264 			if (!c->display) {
1265 				c->Y = 0;
1266 				c = c->next;
1267 				continue;
1268 			}
1269 
1270 			svg_box(Y, c->start_time, c->end_time, "process");
1271 			sample = c->samples;
1272 			while (sample) {
1273 				if (sample->type == TYPE_RUNNING)
1274 					svg_running(Y, sample->cpu,
1275 						    sample->start_time,
1276 						    sample->end_time,
1277 						    sample->backtrace);
1278 				if (sample->type == TYPE_BLOCKED)
1279 					svg_blocked(Y, sample->cpu,
1280 						    sample->start_time,
1281 						    sample->end_time,
1282 						    sample->backtrace);
1283 				if (sample->type == TYPE_WAITING)
1284 					svg_waiting(Y, sample->cpu,
1285 						    sample->start_time,
1286 						    sample->end_time,
1287 						    sample->backtrace);
1288 				sample = sample->next;
1289 			}
1290 
1291 			if (c->comm) {
1292 				char comm[256];
1293 				if (c->total_time > 5000000000) /* 5 seconds */
1294 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1295 				else
1296 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1297 
1298 				svg_text(Y, c->start_time, comm);
1299 			}
1300 			c->Y = Y;
1301 			Y++;
1302 			c = c->next;
1303 		}
1304 		p = p->next;
1305 	}
1306 }
1307 
1308 static void add_process_filter(const char *string)
1309 {
1310 	int pid = strtoull(string, NULL, 10);
1311 	struct process_filter *filt = malloc(sizeof(*filt));
1312 
1313 	if (!filt)
1314 		return;
1315 
1316 	filt->name = strdup(string);
1317 	filt->pid  = pid;
1318 	filt->next = process_filter;
1319 
1320 	process_filter = filt;
1321 }
1322 
1323 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1324 {
1325 	struct process_filter *filt;
1326 	if (!process_filter)
1327 		return 1;
1328 
1329 	filt = process_filter;
1330 	while (filt) {
1331 		if (filt->pid && p->pid == filt->pid)
1332 			return 1;
1333 		if (strcmp(filt->name, c->comm) == 0)
1334 			return 1;
1335 		filt = filt->next;
1336 	}
1337 	return 0;
1338 }
1339 
1340 static int determine_display_tasks_filtered(struct timechart *tchart)
1341 {
1342 	struct per_pid *p;
1343 	struct per_pidcomm *c;
1344 	int count = 0;
1345 
1346 	p = tchart->all_data;
1347 	while (p) {
1348 		p->display = 0;
1349 		if (p->start_time == 1)
1350 			p->start_time = tchart->first_time;
1351 
1352 		/* no exit marker, task kept running to the end */
1353 		if (p->end_time == 0)
1354 			p->end_time = tchart->last_time;
1355 
1356 		c = p->all;
1357 
1358 		while (c) {
1359 			c->display = 0;
1360 
1361 			if (c->start_time == 1)
1362 				c->start_time = tchart->first_time;
1363 
1364 			if (passes_filter(p, c)) {
1365 				c->display = 1;
1366 				p->display = 1;
1367 				count++;
1368 			}
1369 
1370 			if (c->end_time == 0)
1371 				c->end_time = tchart->last_time;
1372 
1373 			c = c->next;
1374 		}
1375 		p = p->next;
1376 	}
1377 	return count;
1378 }
1379 
1380 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1381 {
1382 	struct per_pid *p;
1383 	struct per_pidcomm *c;
1384 	int count = 0;
1385 
1386 	p = tchart->all_data;
1387 	while (p) {
1388 		p->display = 0;
1389 		if (p->start_time == 1)
1390 			p->start_time = tchart->first_time;
1391 
1392 		/* no exit marker, task kept running to the end */
1393 		if (p->end_time == 0)
1394 			p->end_time = tchart->last_time;
1395 		if (p->total_time >= threshold)
1396 			p->display = 1;
1397 
1398 		c = p->all;
1399 
1400 		while (c) {
1401 			c->display = 0;
1402 
1403 			if (c->start_time == 1)
1404 				c->start_time = tchart->first_time;
1405 
1406 			if (c->total_time >= threshold) {
1407 				c->display = 1;
1408 				count++;
1409 			}
1410 
1411 			if (c->end_time == 0)
1412 				c->end_time = tchart->last_time;
1413 
1414 			c = c->next;
1415 		}
1416 		p = p->next;
1417 	}
1418 	return count;
1419 }
1420 
1421 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1422 {
1423 	struct per_pid *p;
1424 	struct per_pidcomm *c;
1425 	int count = 0;
1426 
1427 	p = timechart->all_data;
1428 	while (p) {
1429 		/* no exit marker, task kept running to the end */
1430 		if (p->end_time == 0)
1431 			p->end_time = timechart->last_time;
1432 
1433 		c = p->all;
1434 
1435 		while (c) {
1436 			c->display = 0;
1437 
1438 			if (c->total_bytes >= threshold) {
1439 				c->display = 1;
1440 				count++;
1441 			}
1442 
1443 			if (c->end_time == 0)
1444 				c->end_time = timechart->last_time;
1445 
1446 			c = c->next;
1447 		}
1448 		p = p->next;
1449 	}
1450 	return count;
1451 }
1452 
1453 #define BYTES_THRESH (1 * 1024 * 1024)
1454 #define TIME_THRESH 10000000
1455 
1456 static void write_svg_file(struct timechart *tchart, const char *filename)
1457 {
1458 	u64 i;
1459 	int count;
1460 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1461 
1462 	if (tchart->power_only)
1463 		tchart->proc_num = 0;
1464 
1465 	/* We'd like to show at least proc_num tasks;
1466 	 * be less picky if we have fewer */
1467 	do {
1468 		if (process_filter)
1469 			count = determine_display_tasks_filtered(tchart);
1470 		else if (tchart->io_events)
1471 			count = determine_display_io_tasks(tchart, thresh);
1472 		else
1473 			count = determine_display_tasks(tchart, thresh);
1474 		thresh /= 10;
1475 	} while (!process_filter && thresh && count < tchart->proc_num);
1476 
1477 	if (!tchart->proc_num)
1478 		count = 0;
1479 
1480 	if (tchart->io_events) {
1481 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1482 
1483 		svg_time_grid(0.5);
1484 		svg_io_legenda();
1485 
1486 		draw_io_bars(tchart);
1487 	} else {
1488 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1489 
1490 		svg_time_grid(0);
1491 
1492 		svg_legenda();
1493 
1494 		for (i = 0; i < tchart->numcpus; i++)
1495 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1496 
1497 		draw_cpu_usage(tchart);
1498 		if (tchart->proc_num)
1499 			draw_process_bars(tchart);
1500 		if (!tchart->tasks_only)
1501 			draw_c_p_states(tchart);
1502 		if (tchart->proc_num)
1503 			draw_wakeups(tchart);
1504 	}
1505 
1506 	svg_close();
1507 }
1508 
1509 static int process_header(struct perf_file_section *section __maybe_unused,
1510 			  struct perf_header *ph,
1511 			  int feat,
1512 			  int fd __maybe_unused,
1513 			  void *data)
1514 {
1515 	struct timechart *tchart = data;
1516 
1517 	switch (feat) {
1518 	case HEADER_NRCPUS:
1519 		tchart->numcpus = ph->env.nr_cpus_avail;
1520 		break;
1521 
1522 	case HEADER_CPU_TOPOLOGY:
1523 		if (!tchart->topology)
1524 			break;
1525 
1526 		if (svg_build_topology_map(&ph->env))
1527 			fprintf(stderr, "problem building topology\n");
1528 		break;
1529 
1530 	default:
1531 		break;
1532 	}
1533 
1534 	return 0;
1535 }
1536 
1537 static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1538 {
1539 	const struct evsel_str_handler power_tracepoints[] = {
1540 		{ "power:cpu_idle",		process_sample_cpu_idle },
1541 		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1542 		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1543 		{ "sched:sched_switch",		process_sample_sched_switch },
1544 #ifdef SUPPORT_OLD_POWER_EVENTS
1545 		{ "power:power_start",		process_sample_power_start },
1546 		{ "power:power_end",		process_sample_power_end },
1547 		{ "power:power_frequency",	process_sample_power_frequency },
1548 #endif
1549 
1550 		{ "syscalls:sys_enter_read",		process_enter_read },
1551 		{ "syscalls:sys_enter_pread64",		process_enter_read },
1552 		{ "syscalls:sys_enter_readv",		process_enter_read },
1553 		{ "syscalls:sys_enter_preadv",		process_enter_read },
1554 		{ "syscalls:sys_enter_write",		process_enter_write },
1555 		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1556 		{ "syscalls:sys_enter_writev",		process_enter_write },
1557 		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1558 		{ "syscalls:sys_enter_sync",		process_enter_sync },
1559 		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1560 		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1561 		{ "syscalls:sys_enter_msync",		process_enter_sync },
1562 		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1563 		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1564 		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1565 		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1566 		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1567 		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1568 		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1569 		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1570 		{ "syscalls:sys_enter_poll",		process_enter_poll },
1571 		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1572 		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1573 		{ "syscalls:sys_enter_select",		process_enter_poll },
1574 
1575 		{ "syscalls:sys_exit_read",		process_exit_read },
1576 		{ "syscalls:sys_exit_pread64",		process_exit_read },
1577 		{ "syscalls:sys_exit_readv",		process_exit_read },
1578 		{ "syscalls:sys_exit_preadv",		process_exit_read },
1579 		{ "syscalls:sys_exit_write",		process_exit_write },
1580 		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1581 		{ "syscalls:sys_exit_writev",		process_exit_write },
1582 		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1583 		{ "syscalls:sys_exit_sync",		process_exit_sync },
1584 		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1585 		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1586 		{ "syscalls:sys_exit_msync",		process_exit_sync },
1587 		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1588 		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1589 		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1590 		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1591 		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1592 		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1593 		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1594 		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1595 		{ "syscalls:sys_exit_poll",		process_exit_poll },
1596 		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1597 		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1598 		{ "syscalls:sys_exit_select",		process_exit_poll },
1599 	};
1600 	struct perf_data data = {
1601 		.path  = input_name,
1602 		.mode  = PERF_DATA_MODE_READ,
1603 		.force = tchart->force,
1604 	};
1605 
1606 	struct perf_session *session = perf_session__new(&data, &tchart->tool);
1607 	int ret = -EINVAL;
1608 
1609 	if (IS_ERR(session))
1610 		return PTR_ERR(session);
1611 
1612 	symbol__init(&session->header.env);
1613 
1614 	(void)perf_header__process_sections(&session->header,
1615 					    perf_data__fd(session->data),
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) / (double)NSEC_PER_SEC, 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 		free(rec_argv);
1732 		return -ENOMEM;
1733 	}
1734 
1735 	p = rec_argv;
1736 	for (i = 0; i < common_args_nr; i++)
1737 		*p++ = strdup(common_args[i]);
1738 
1739 	for (i = 0; i < disk_events_nr; i++) {
1740 		if (!is_valid_tracepoint(disk_events[i])) {
1741 			rec_argc -= 4;
1742 			continue;
1743 		}
1744 
1745 		*p++ = "-e";
1746 		*p++ = strdup(disk_events[i]);
1747 		*p++ = "--filter";
1748 		*p++ = filter;
1749 	}
1750 	for (i = 0; i < net_events_nr; i++) {
1751 		if (!is_valid_tracepoint(net_events[i])) {
1752 			rec_argc -= 4;
1753 			continue;
1754 		}
1755 
1756 		*p++ = "-e";
1757 		*p++ = strdup(net_events[i]);
1758 		*p++ = "--filter";
1759 		*p++ = filter;
1760 	}
1761 	for (i = 0; i < poll_events_nr; i++) {
1762 		if (!is_valid_tracepoint(poll_events[i])) {
1763 			rec_argc -= 4;
1764 			continue;
1765 		}
1766 
1767 		*p++ = "-e";
1768 		*p++ = strdup(poll_events[i]);
1769 		*p++ = "--filter";
1770 		*p++ = filter;
1771 	}
1772 
1773 	for (i = 0; i < (unsigned int)argc; i++)
1774 		*p++ = argv[i];
1775 
1776 	return cmd_record(rec_argc, rec_argv);
1777 }
1778 
1779 
1780 static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1781 {
1782 	unsigned int rec_argc, i, j;
1783 	const char **rec_argv;
1784 	const char **p;
1785 	unsigned int record_elems;
1786 
1787 	const char * const common_args[] = {
1788 		"record", "-a", "-R", "-c", "1",
1789 	};
1790 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1791 
1792 	const char * const backtrace_args[] = {
1793 		"-g",
1794 	};
1795 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1796 
1797 	const char * const power_args[] = {
1798 		"-e", "power:cpu_frequency",
1799 		"-e", "power:cpu_idle",
1800 	};
1801 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1802 
1803 	const char * const old_power_args[] = {
1804 #ifdef SUPPORT_OLD_POWER_EVENTS
1805 		"-e", "power:power_start",
1806 		"-e", "power:power_end",
1807 		"-e", "power:power_frequency",
1808 #endif
1809 	};
1810 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1811 
1812 	const char * const tasks_args[] = {
1813 		"-e", "sched:sched_wakeup",
1814 		"-e", "sched:sched_switch",
1815 	};
1816 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1817 
1818 #ifdef SUPPORT_OLD_POWER_EVENTS
1819 	if (!is_valid_tracepoint("power:cpu_idle") &&
1820 	    is_valid_tracepoint("power:power_start")) {
1821 		use_old_power_events = 1;
1822 		power_args_nr = 0;
1823 	} else {
1824 		old_power_args_nr = 0;
1825 	}
1826 #endif
1827 
1828 	if (tchart->power_only)
1829 		tasks_args_nr = 0;
1830 
1831 	if (tchart->tasks_only) {
1832 		power_args_nr = 0;
1833 		old_power_args_nr = 0;
1834 	}
1835 
1836 	if (!tchart->with_backtrace)
1837 		backtrace_args_no = 0;
1838 
1839 	record_elems = common_args_nr + tasks_args_nr +
1840 		power_args_nr + old_power_args_nr + backtrace_args_no;
1841 
1842 	rec_argc = record_elems + argc;
1843 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1844 
1845 	if (rec_argv == NULL)
1846 		return -ENOMEM;
1847 
1848 	p = rec_argv;
1849 	for (i = 0; i < common_args_nr; i++)
1850 		*p++ = strdup(common_args[i]);
1851 
1852 	for (i = 0; i < backtrace_args_no; i++)
1853 		*p++ = strdup(backtrace_args[i]);
1854 
1855 	for (i = 0; i < tasks_args_nr; i++)
1856 		*p++ = strdup(tasks_args[i]);
1857 
1858 	for (i = 0; i < power_args_nr; i++)
1859 		*p++ = strdup(power_args[i]);
1860 
1861 	for (i = 0; i < old_power_args_nr; i++)
1862 		*p++ = strdup(old_power_args[i]);
1863 
1864 	for (j = 0; j < (unsigned int)argc; j++)
1865 		*p++ = argv[j];
1866 
1867 	return cmd_record(rec_argc, rec_argv);
1868 }
1869 
1870 static int
1871 parse_process(const struct option *opt __maybe_unused, const char *arg,
1872 	      int __maybe_unused unset)
1873 {
1874 	if (arg)
1875 		add_process_filter(arg);
1876 	return 0;
1877 }
1878 
1879 static int
1880 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1881 		int __maybe_unused unset)
1882 {
1883 	unsigned long duration = strtoul(arg, NULL, 0);
1884 
1885 	if (svg_highlight || svg_highlight_name)
1886 		return -1;
1887 
1888 	if (duration)
1889 		svg_highlight = duration;
1890 	else
1891 		svg_highlight_name = strdup(arg);
1892 
1893 	return 0;
1894 }
1895 
1896 static int
1897 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1898 {
1899 	char unit = 'n';
1900 	u64 *value = opt->value;
1901 
1902 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1903 		switch (unit) {
1904 		case 'm':
1905 			*value *= NSEC_PER_MSEC;
1906 			break;
1907 		case 'u':
1908 			*value *= NSEC_PER_USEC;
1909 			break;
1910 		case 'n':
1911 			break;
1912 		default:
1913 			return -1;
1914 		}
1915 	}
1916 
1917 	return 0;
1918 }
1919 
1920 int cmd_timechart(int argc, const char **argv)
1921 {
1922 	struct timechart tchart = {
1923 		.tool = {
1924 			.comm		 = process_comm_event,
1925 			.fork		 = process_fork_event,
1926 			.exit		 = process_exit_event,
1927 			.sample		 = process_sample_event,
1928 			.ordered_events	 = true,
1929 		},
1930 		.proc_num = 15,
1931 		.min_time = NSEC_PER_MSEC,
1932 		.merge_dist = 1000,
1933 	};
1934 	const char *output_name = "output.svg";
1935 	const struct option timechart_common_options[] = {
1936 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1937 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only, "output processes data only"),
1938 	OPT_END()
1939 	};
1940 	const struct option timechart_options[] = {
1941 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1942 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1943 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1944 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1945 		      "highlight tasks. Pass duration in ns or process name.",
1946 		       parse_highlight),
1947 	OPT_CALLBACK('p', "process", NULL, "process",
1948 		      "process selector. Pass a pid or process name.",
1949 		       parse_process),
1950 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1951 		     "Look for files with symbols relative to this directory",
1952 		     symbol__config_symfs),
1953 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1954 		    "min. number of tasks to print"),
1955 	OPT_BOOLEAN('t', "topology", &tchart.topology,
1956 		    "sort CPUs according to topology"),
1957 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1958 		    "skip EAGAIN errors"),
1959 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1960 		     "all IO faster than min-time will visually appear longer",
1961 		     parse_time),
1962 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1963 		     "merge events that are merge-dist us apart",
1964 		     parse_time),
1965 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1966 	OPT_PARENT(timechart_common_options),
1967 	};
1968 	const char * const timechart_subcommands[] = { "record", NULL };
1969 	const char *timechart_usage[] = {
1970 		"perf timechart [<options>] {record}",
1971 		NULL
1972 	};
1973 	const struct option timechart_record_options[] = {
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_PARENT(timechart_common_options),
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 && strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
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