xref: /linux/tools/perf/builtin-timechart.c (revision 9f2c9170934eace462499ba0bfe042cc72900173)
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 #include <traceevent/event-parse.h>
42 
43 #ifdef LACKS_OPEN_MEMSTREAM_PROTOTYPE
44 FILE *open_memstream(char **ptr, size_t *sizeloc);
45 #endif
46 
47 #define SUPPORT_OLD_POWER_EVENTS 1
48 #define PWR_EVENT_EXIT -1
49 
50 struct per_pid;
51 struct power_event;
52 struct wake_event;
53 
54 struct timechart {
55 	struct perf_tool	tool;
56 	struct per_pid		*all_data;
57 	struct power_event	*power_events;
58 	struct wake_event	*wake_events;
59 	int			proc_num;
60 	unsigned int		numcpus;
61 	u64			min_freq,	/* Lowest CPU frequency seen */
62 				max_freq,	/* Highest CPU frequency seen */
63 				turbo_frequency,
64 				first_time, last_time;
65 	bool			power_only,
66 				tasks_only,
67 				with_backtrace,
68 				topology;
69 	bool			force;
70 	/* IO related settings */
71 	bool			io_only,
72 				skip_eagain;
73 	u64			io_events;
74 	u64			min_time,
75 				merge_dist;
76 };
77 
78 struct per_pidcomm;
79 struct cpu_sample;
80 struct io_sample;
81 
82 /*
83  * Datastructure layout:
84  * We keep an list of "pid"s, matching the kernels notion of a task struct.
85  * Each "pid" entry, has a list of "comm"s.
86  *	this is because we want to track different programs different, while
87  *	exec will reuse the original pid (by design).
88  * Each comm has a list of samples that will be used to draw
89  * final graph.
90  */
91 
92 struct per_pid {
93 	struct per_pid *next;
94 
95 	int		pid;
96 	int		ppid;
97 
98 	u64		start_time;
99 	u64		end_time;
100 	u64		total_time;
101 	u64		total_bytes;
102 	int		display;
103 
104 	struct per_pidcomm *all;
105 	struct per_pidcomm *current;
106 };
107 
108 
109 struct per_pidcomm {
110 	struct per_pidcomm *next;
111 
112 	u64		start_time;
113 	u64		end_time;
114 	u64		total_time;
115 	u64		max_bytes;
116 	u64		total_bytes;
117 
118 	int		Y;
119 	int		display;
120 
121 	long		state;
122 	u64		state_since;
123 
124 	char		*comm;
125 
126 	struct cpu_sample *samples;
127 	struct io_sample  *io_samples;
128 };
129 
130 struct sample_wrapper {
131 	struct sample_wrapper *next;
132 
133 	u64		timestamp;
134 	unsigned char	data[];
135 };
136 
137 #define TYPE_NONE	0
138 #define TYPE_RUNNING	1
139 #define TYPE_WAITING	2
140 #define TYPE_BLOCKED	3
141 
142 struct cpu_sample {
143 	struct cpu_sample *next;
144 
145 	u64 start_time;
146 	u64 end_time;
147 	int type;
148 	int cpu;
149 	const char *backtrace;
150 };
151 
152 enum {
153 	IOTYPE_READ,
154 	IOTYPE_WRITE,
155 	IOTYPE_SYNC,
156 	IOTYPE_TX,
157 	IOTYPE_RX,
158 	IOTYPE_POLL,
159 };
160 
161 struct io_sample {
162 	struct io_sample *next;
163 
164 	u64 start_time;
165 	u64 end_time;
166 	u64 bytes;
167 	int type;
168 	int fd;
169 	int err;
170 	int merges;
171 };
172 
173 #define CSTATE 1
174 #define PSTATE 2
175 
176 struct power_event {
177 	struct power_event *next;
178 	int type;
179 	int state;
180 	u64 start_time;
181 	u64 end_time;
182 	int cpu;
183 };
184 
185 struct wake_event {
186 	struct wake_event *next;
187 	int waker;
188 	int wakee;
189 	u64 time;
190 	const char *backtrace;
191 };
192 
193 struct process_filter {
194 	char			*name;
195 	int			pid;
196 	struct process_filter	*next;
197 };
198 
199 static struct process_filter *process_filter;
200 
201 
202 static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
203 {
204 	struct per_pid *cursor = tchart->all_data;
205 
206 	while (cursor) {
207 		if (cursor->pid == pid)
208 			return cursor;
209 		cursor = cursor->next;
210 	}
211 	cursor = zalloc(sizeof(*cursor));
212 	assert(cursor != NULL);
213 	cursor->pid = pid;
214 	cursor->next = tchart->all_data;
215 	tchart->all_data = cursor;
216 	return cursor;
217 }
218 
219 static struct per_pidcomm *create_pidcomm(struct per_pid *p)
220 {
221 	struct per_pidcomm *c;
222 
223 	c = zalloc(sizeof(*c));
224 	if (!c)
225 		return NULL;
226 	p->current = c;
227 	c->next = p->all;
228 	p->all = c;
229 	return c;
230 }
231 
232 static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
233 {
234 	struct per_pid *p;
235 	struct per_pidcomm *c;
236 	p = find_create_pid(tchart, pid);
237 	c = p->all;
238 	while (c) {
239 		if (c->comm && strcmp(c->comm, comm) == 0) {
240 			p->current = c;
241 			return;
242 		}
243 		if (!c->comm) {
244 			c->comm = strdup(comm);
245 			p->current = c;
246 			return;
247 		}
248 		c = c->next;
249 	}
250 	c = create_pidcomm(p);
251 	assert(c != NULL);
252 	c->comm = strdup(comm);
253 }
254 
255 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
256 {
257 	struct per_pid *p, *pp;
258 	p = find_create_pid(tchart, pid);
259 	pp = find_create_pid(tchart, ppid);
260 	p->ppid = ppid;
261 	if (pp->current && pp->current->comm && !p->current)
262 		pid_set_comm(tchart, pid, pp->current->comm);
263 
264 	p->start_time = timestamp;
265 	if (p->current && !p->current->start_time) {
266 		p->current->start_time = timestamp;
267 		p->current->state_since = timestamp;
268 	}
269 }
270 
271 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
272 {
273 	struct per_pid *p;
274 	p = find_create_pid(tchart, pid);
275 	p->end_time = timestamp;
276 	if (p->current)
277 		p->current->end_time = timestamp;
278 }
279 
280 static void pid_put_sample(struct timechart *tchart, int pid, int type,
281 			   unsigned int cpu, u64 start, u64 end,
282 			   const char *backtrace)
283 {
284 	struct per_pid *p;
285 	struct per_pidcomm *c;
286 	struct cpu_sample *sample;
287 
288 	p = find_create_pid(tchart, pid);
289 	c = p->current;
290 	if (!c) {
291 		c = create_pidcomm(p);
292 		assert(c != NULL);
293 	}
294 
295 	sample = zalloc(sizeof(*sample));
296 	assert(sample != NULL);
297 	sample->start_time = start;
298 	sample->end_time = end;
299 	sample->type = type;
300 	sample->next = c->samples;
301 	sample->cpu = cpu;
302 	sample->backtrace = backtrace;
303 	c->samples = sample;
304 
305 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
306 		c->total_time += (end-start);
307 		p->total_time += (end-start);
308 	}
309 
310 	if (c->start_time == 0 || c->start_time > start)
311 		c->start_time = start;
312 	if (p->start_time == 0 || p->start_time > start)
313 		p->start_time = start;
314 }
315 
316 #define MAX_CPUS 4096
317 
318 static u64 cpus_cstate_start_times[MAX_CPUS];
319 static int cpus_cstate_state[MAX_CPUS];
320 static u64 cpus_pstate_start_times[MAX_CPUS];
321 static u64 cpus_pstate_state[MAX_CPUS];
322 
323 static int process_comm_event(struct perf_tool *tool,
324 			      union perf_event *event,
325 			      struct perf_sample *sample __maybe_unused,
326 			      struct machine *machine __maybe_unused)
327 {
328 	struct timechart *tchart = container_of(tool, struct timechart, tool);
329 	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
330 	return 0;
331 }
332 
333 static int process_fork_event(struct perf_tool *tool,
334 			      union perf_event *event,
335 			      struct perf_sample *sample __maybe_unused,
336 			      struct machine *machine __maybe_unused)
337 {
338 	struct timechart *tchart = container_of(tool, struct timechart, tool);
339 	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
340 	return 0;
341 }
342 
343 static int process_exit_event(struct perf_tool *tool,
344 			      union perf_event *event,
345 			      struct perf_sample *sample __maybe_unused,
346 			      struct machine *machine __maybe_unused)
347 {
348 	struct timechart *tchart = container_of(tool, struct timechart, tool);
349 	pid_exit(tchart, event->fork.pid, event->fork.time);
350 	return 0;
351 }
352 
353 #ifdef SUPPORT_OLD_POWER_EVENTS
354 static int use_old_power_events;
355 #endif
356 
357 static void c_state_start(int cpu, u64 timestamp, int state)
358 {
359 	cpus_cstate_start_times[cpu] = timestamp;
360 	cpus_cstate_state[cpu] = state;
361 }
362 
363 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
364 {
365 	struct power_event *pwr = zalloc(sizeof(*pwr));
366 
367 	if (!pwr)
368 		return;
369 
370 	pwr->state = cpus_cstate_state[cpu];
371 	pwr->start_time = cpus_cstate_start_times[cpu];
372 	pwr->end_time = timestamp;
373 	pwr->cpu = cpu;
374 	pwr->type = CSTATE;
375 	pwr->next = tchart->power_events;
376 
377 	tchart->power_events = pwr;
378 }
379 
380 static struct power_event *p_state_end(struct timechart *tchart, int cpu,
381 					u64 timestamp)
382 {
383 	struct power_event *pwr = zalloc(sizeof(*pwr));
384 
385 	if (!pwr)
386 		return NULL;
387 
388 	pwr->state = cpus_pstate_state[cpu];
389 	pwr->start_time = cpus_pstate_start_times[cpu];
390 	pwr->end_time = timestamp;
391 	pwr->cpu = cpu;
392 	pwr->type = PSTATE;
393 	pwr->next = tchart->power_events;
394 	if (!pwr->start_time)
395 		pwr->start_time = tchart->first_time;
396 
397 	tchart->power_events = pwr;
398 	return pwr;
399 }
400 
401 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
402 {
403 	struct power_event *pwr;
404 
405 	if (new_freq > 8000000) /* detect invalid data */
406 		return;
407 
408 	pwr = p_state_end(tchart, cpu, timestamp);
409 	if (!pwr)
410 		return;
411 
412 	cpus_pstate_state[cpu] = new_freq;
413 	cpus_pstate_start_times[cpu] = timestamp;
414 
415 	if ((u64)new_freq > tchart->max_freq)
416 		tchart->max_freq = new_freq;
417 
418 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
419 		tchart->min_freq = new_freq;
420 
421 	if (new_freq == tchart->max_freq - 1000)
422 		tchart->turbo_frequency = tchart->max_freq;
423 }
424 
425 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
426 			 int waker, int wakee, u8 flags, const char *backtrace)
427 {
428 	struct per_pid *p;
429 	struct wake_event *we = zalloc(sizeof(*we));
430 
431 	if (!we)
432 		return;
433 
434 	we->time = timestamp;
435 	we->waker = waker;
436 	we->backtrace = backtrace;
437 
438 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
439 		we->waker = -1;
440 
441 	we->wakee = wakee;
442 	we->next = tchart->wake_events;
443 	tchart->wake_events = we;
444 	p = find_create_pid(tchart, we->wakee);
445 
446 	if (p && p->current && p->current->state == TYPE_NONE) {
447 		p->current->state_since = timestamp;
448 		p->current->state = TYPE_WAITING;
449 	}
450 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
451 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
452 			       p->current->state_since, timestamp, NULL);
453 		p->current->state_since = timestamp;
454 		p->current->state = TYPE_WAITING;
455 	}
456 }
457 
458 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
459 			 int prev_pid, int next_pid, u64 prev_state,
460 			 const char *backtrace)
461 {
462 	struct per_pid *p = NULL, *prev_p;
463 
464 	prev_p = find_create_pid(tchart, prev_pid);
465 
466 	p = find_create_pid(tchart, next_pid);
467 
468 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
469 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
470 			       prev_p->current->state_since, timestamp,
471 			       backtrace);
472 	if (p && p->current) {
473 		if (p->current->state != TYPE_NONE)
474 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
475 				       p->current->state_since, timestamp,
476 				       backtrace);
477 
478 		p->current->state_since = timestamp;
479 		p->current->state = TYPE_RUNNING;
480 	}
481 
482 	if (prev_p->current) {
483 		prev_p->current->state = TYPE_NONE;
484 		prev_p->current->state_since = timestamp;
485 		if (prev_state & 2)
486 			prev_p->current->state = TYPE_BLOCKED;
487 		if (prev_state == 0)
488 			prev_p->current->state = TYPE_WAITING;
489 	}
490 }
491 
492 static const char *cat_backtrace(union perf_event *event,
493 				 struct perf_sample *sample,
494 				 struct machine *machine)
495 {
496 	struct addr_location al;
497 	unsigned int i;
498 	char *p = NULL;
499 	size_t p_len;
500 	u8 cpumode = PERF_RECORD_MISC_USER;
501 	struct addr_location tal;
502 	struct ip_callchain *chain = sample->callchain;
503 	FILE *f = open_memstream(&p, &p_len);
504 
505 	if (!f) {
506 		perror("open_memstream error");
507 		return NULL;
508 	}
509 
510 	if (!chain)
511 		goto exit;
512 
513 	if (machine__resolve(machine, &al, sample) < 0) {
514 		fprintf(stderr, "problem processing %d event, skipping it.\n",
515 			event->header.type);
516 		goto exit;
517 	}
518 
519 	for (i = 0; i < chain->nr; i++) {
520 		u64 ip;
521 
522 		if (callchain_param.order == ORDER_CALLEE)
523 			ip = chain->ips[i];
524 		else
525 			ip = chain->ips[chain->nr - i - 1];
526 
527 		if (ip >= PERF_CONTEXT_MAX) {
528 			switch (ip) {
529 			case PERF_CONTEXT_HV:
530 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
531 				break;
532 			case PERF_CONTEXT_KERNEL:
533 				cpumode = PERF_RECORD_MISC_KERNEL;
534 				break;
535 			case PERF_CONTEXT_USER:
536 				cpumode = PERF_RECORD_MISC_USER;
537 				break;
538 			default:
539 				pr_debug("invalid callchain context: "
540 					 "%"PRId64"\n", (s64) ip);
541 
542 				/*
543 				 * It seems the callchain is corrupted.
544 				 * Discard all.
545 				 */
546 				zfree(&p);
547 				goto exit_put;
548 			}
549 			continue;
550 		}
551 
552 		tal.filtered = 0;
553 		if (thread__find_symbol(al.thread, cpumode, ip, &tal))
554 			fprintf(f, "..... %016" PRIx64 " %s\n", ip, tal.sym->name);
555 		else
556 			fprintf(f, "..... %016" PRIx64 "\n", ip);
557 	}
558 exit_put:
559 	addr_location__put(&al);
560 exit:
561 	fclose(f);
562 
563 	return p;
564 }
565 
566 typedef int (*tracepoint_handler)(struct timechart *tchart,
567 				  struct evsel *evsel,
568 				  struct perf_sample *sample,
569 				  const char *backtrace);
570 
571 static int process_sample_event(struct perf_tool *tool,
572 				union perf_event *event,
573 				struct perf_sample *sample,
574 				struct evsel *evsel,
575 				struct machine *machine)
576 {
577 	struct timechart *tchart = container_of(tool, struct timechart, tool);
578 
579 	if (evsel->core.attr.sample_type & PERF_SAMPLE_TIME) {
580 		if (!tchart->first_time || tchart->first_time > sample->time)
581 			tchart->first_time = sample->time;
582 		if (tchart->last_time < sample->time)
583 			tchart->last_time = sample->time;
584 	}
585 
586 	if (evsel->handler != NULL) {
587 		tracepoint_handler f = evsel->handler;
588 		return f(tchart, evsel, sample,
589 			 cat_backtrace(event, sample, machine));
590 	}
591 
592 	return 0;
593 }
594 
595 static int
596 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
597 			struct evsel *evsel,
598 			struct perf_sample *sample,
599 			const char *backtrace __maybe_unused)
600 {
601 	u32 state  = evsel__intval(evsel, sample, "state");
602 	u32 cpu_id = evsel__intval(evsel, sample, "cpu_id");
603 
604 	if (state == (u32)PWR_EVENT_EXIT)
605 		c_state_end(tchart, cpu_id, sample->time);
606 	else
607 		c_state_start(cpu_id, sample->time, state);
608 	return 0;
609 }
610 
611 static int
612 process_sample_cpu_frequency(struct timechart *tchart,
613 			     struct evsel *evsel,
614 			     struct perf_sample *sample,
615 			     const char *backtrace __maybe_unused)
616 {
617 	u32 state  = evsel__intval(evsel, sample, "state");
618 	u32 cpu_id = evsel__intval(evsel, sample, "cpu_id");
619 
620 	p_state_change(tchart, cpu_id, sample->time, state);
621 	return 0;
622 }
623 
624 static int
625 process_sample_sched_wakeup(struct timechart *tchart,
626 			    struct evsel *evsel,
627 			    struct perf_sample *sample,
628 			    const char *backtrace)
629 {
630 	u8 flags  = evsel__intval(evsel, sample, "common_flags");
631 	int waker = evsel__intval(evsel, sample, "common_pid");
632 	int wakee = evsel__intval(evsel, sample, "pid");
633 
634 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
635 	return 0;
636 }
637 
638 static int
639 process_sample_sched_switch(struct timechart *tchart,
640 			    struct evsel *evsel,
641 			    struct perf_sample *sample,
642 			    const char *backtrace)
643 {
644 	int prev_pid   = evsel__intval(evsel, sample, "prev_pid");
645 	int next_pid   = evsel__intval(evsel, sample, "next_pid");
646 	u64 prev_state = evsel__intval(evsel, sample, "prev_state");
647 
648 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
649 		     prev_state, backtrace);
650 	return 0;
651 }
652 
653 #ifdef SUPPORT_OLD_POWER_EVENTS
654 static int
655 process_sample_power_start(struct timechart *tchart __maybe_unused,
656 			   struct evsel *evsel,
657 			   struct perf_sample *sample,
658 			   const char *backtrace __maybe_unused)
659 {
660 	u64 cpu_id = evsel__intval(evsel, sample, "cpu_id");
661 	u64 value  = evsel__intval(evsel, sample, "value");
662 
663 	c_state_start(cpu_id, sample->time, value);
664 	return 0;
665 }
666 
667 static int
668 process_sample_power_end(struct timechart *tchart,
669 			 struct evsel *evsel __maybe_unused,
670 			 struct perf_sample *sample,
671 			 const char *backtrace __maybe_unused)
672 {
673 	c_state_end(tchart, sample->cpu, sample->time);
674 	return 0;
675 }
676 
677 static int
678 process_sample_power_frequency(struct timechart *tchart,
679 			       struct evsel *evsel,
680 			       struct perf_sample *sample,
681 			       const char *backtrace __maybe_unused)
682 {
683 	u64 cpu_id = evsel__intval(evsel, sample, "cpu_id");
684 	u64 value  = evsel__intval(evsel, sample, "value");
685 
686 	p_state_change(tchart, cpu_id, sample->time, value);
687 	return 0;
688 }
689 #endif /* SUPPORT_OLD_POWER_EVENTS */
690 
691 /*
692  * After the last sample we need to wrap up the current C/P state
693  * and close out each CPU for these.
694  */
695 static void end_sample_processing(struct timechart *tchart)
696 {
697 	u64 cpu;
698 	struct power_event *pwr;
699 
700 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
701 		/* C state */
702 #if 0
703 		pwr = zalloc(sizeof(*pwr));
704 		if (!pwr)
705 			return;
706 
707 		pwr->state = cpus_cstate_state[cpu];
708 		pwr->start_time = cpus_cstate_start_times[cpu];
709 		pwr->end_time = tchart->last_time;
710 		pwr->cpu = cpu;
711 		pwr->type = CSTATE;
712 		pwr->next = tchart->power_events;
713 
714 		tchart->power_events = pwr;
715 #endif
716 		/* P state */
717 
718 		pwr = p_state_end(tchart, cpu, tchart->last_time);
719 		if (!pwr)
720 			return;
721 
722 		if (!pwr->state)
723 			pwr->state = tchart->min_freq;
724 	}
725 }
726 
727 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
728 			       u64 start, int fd)
729 {
730 	struct per_pid *p = find_create_pid(tchart, pid);
731 	struct per_pidcomm *c = p->current;
732 	struct io_sample *sample;
733 	struct io_sample *prev;
734 
735 	if (!c) {
736 		c = create_pidcomm(p);
737 		if (!c)
738 			return -ENOMEM;
739 	}
740 
741 	prev = c->io_samples;
742 
743 	if (prev && prev->start_time && !prev->end_time) {
744 		pr_warning("Skip invalid start event: "
745 			   "previous event already started!\n");
746 
747 		/* remove previous event that has been started,
748 		 * we are not sure we will ever get an end for it */
749 		c->io_samples = prev->next;
750 		free(prev);
751 		return 0;
752 	}
753 
754 	sample = zalloc(sizeof(*sample));
755 	if (!sample)
756 		return -ENOMEM;
757 	sample->start_time = start;
758 	sample->type = type;
759 	sample->fd = fd;
760 	sample->next = c->io_samples;
761 	c->io_samples = sample;
762 
763 	if (c->start_time == 0 || c->start_time > start)
764 		c->start_time = start;
765 
766 	return 0;
767 }
768 
769 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
770 			     u64 end, long ret)
771 {
772 	struct per_pid *p = find_create_pid(tchart, pid);
773 	struct per_pidcomm *c = p->current;
774 	struct io_sample *sample, *prev;
775 
776 	if (!c) {
777 		pr_warning("Invalid pidcomm!\n");
778 		return -1;
779 	}
780 
781 	sample = c->io_samples;
782 
783 	if (!sample) /* skip partially captured events */
784 		return 0;
785 
786 	if (sample->end_time) {
787 		pr_warning("Skip invalid end event: "
788 			   "previous event already ended!\n");
789 		return 0;
790 	}
791 
792 	if (sample->type != type) {
793 		pr_warning("Skip invalid end event: invalid event type!\n");
794 		return 0;
795 	}
796 
797 	sample->end_time = end;
798 	prev = sample->next;
799 
800 	/* we want to be able to see small and fast transfers, so make them
801 	 * at least min_time long, but don't overlap them */
802 	if (sample->end_time - sample->start_time < tchart->min_time)
803 		sample->end_time = sample->start_time + tchart->min_time;
804 	if (prev && sample->start_time < prev->end_time) {
805 		if (prev->err) /* try to make errors more visible */
806 			sample->start_time = prev->end_time;
807 		else
808 			prev->end_time = sample->start_time;
809 	}
810 
811 	if (ret < 0) {
812 		sample->err = ret;
813 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
814 		   type == IOTYPE_TX || type == IOTYPE_RX) {
815 
816 		if ((u64)ret > c->max_bytes)
817 			c->max_bytes = ret;
818 
819 		c->total_bytes += ret;
820 		p->total_bytes += ret;
821 		sample->bytes = ret;
822 	}
823 
824 	/* merge two requests to make svg smaller and render-friendly */
825 	if (prev &&
826 	    prev->type == sample->type &&
827 	    prev->err == sample->err &&
828 	    prev->fd == sample->fd &&
829 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
830 
831 		sample->bytes += prev->bytes;
832 		sample->merges += prev->merges + 1;
833 
834 		sample->start_time = prev->start_time;
835 		sample->next = prev->next;
836 		free(prev);
837 
838 		if (!sample->err && sample->bytes > c->max_bytes)
839 			c->max_bytes = sample->bytes;
840 	}
841 
842 	tchart->io_events++;
843 
844 	return 0;
845 }
846 
847 static int
848 process_enter_read(struct timechart *tchart,
849 		   struct evsel *evsel,
850 		   struct perf_sample *sample)
851 {
852 	long fd = evsel__intval(evsel, sample, "fd");
853 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
854 				   sample->time, fd);
855 }
856 
857 static int
858 process_exit_read(struct timechart *tchart,
859 		  struct evsel *evsel,
860 		  struct perf_sample *sample)
861 {
862 	long ret = evsel__intval(evsel, sample, "ret");
863 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
864 				 sample->time, ret);
865 }
866 
867 static int
868 process_enter_write(struct timechart *tchart,
869 		    struct evsel *evsel,
870 		    struct perf_sample *sample)
871 {
872 	long fd = evsel__intval(evsel, sample, "fd");
873 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
874 				   sample->time, fd);
875 }
876 
877 static int
878 process_exit_write(struct timechart *tchart,
879 		   struct evsel *evsel,
880 		   struct perf_sample *sample)
881 {
882 	long ret = evsel__intval(evsel, sample, "ret");
883 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
884 				 sample->time, ret);
885 }
886 
887 static int
888 process_enter_sync(struct timechart *tchart,
889 		   struct evsel *evsel,
890 		   struct perf_sample *sample)
891 {
892 	long fd = evsel__intval(evsel, sample, "fd");
893 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
894 				   sample->time, fd);
895 }
896 
897 static int
898 process_exit_sync(struct timechart *tchart,
899 		  struct evsel *evsel,
900 		  struct perf_sample *sample)
901 {
902 	long ret = evsel__intval(evsel, sample, "ret");
903 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
904 				 sample->time, ret);
905 }
906 
907 static int
908 process_enter_tx(struct timechart *tchart,
909 		 struct evsel *evsel,
910 		 struct perf_sample *sample)
911 {
912 	long fd = evsel__intval(evsel, sample, "fd");
913 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
914 				   sample->time, fd);
915 }
916 
917 static int
918 process_exit_tx(struct timechart *tchart,
919 		struct evsel *evsel,
920 		struct perf_sample *sample)
921 {
922 	long ret = evsel__intval(evsel, sample, "ret");
923 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
924 				 sample->time, ret);
925 }
926 
927 static int
928 process_enter_rx(struct timechart *tchart,
929 		 struct evsel *evsel,
930 		 struct perf_sample *sample)
931 {
932 	long fd = evsel__intval(evsel, sample, "fd");
933 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
934 				   sample->time, fd);
935 }
936 
937 static int
938 process_exit_rx(struct timechart *tchart,
939 		struct evsel *evsel,
940 		struct perf_sample *sample)
941 {
942 	long ret = evsel__intval(evsel, sample, "ret");
943 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
944 				 sample->time, ret);
945 }
946 
947 static int
948 process_enter_poll(struct timechart *tchart,
949 		   struct evsel *evsel,
950 		   struct perf_sample *sample)
951 {
952 	long fd = evsel__intval(evsel, sample, "fd");
953 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
954 				   sample->time, fd);
955 }
956 
957 static int
958 process_exit_poll(struct timechart *tchart,
959 		  struct evsel *evsel,
960 		  struct perf_sample *sample)
961 {
962 	long ret = evsel__intval(evsel, sample, "ret");
963 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
964 				 sample->time, ret);
965 }
966 
967 /*
968  * Sort the pid datastructure
969  */
970 static void sort_pids(struct timechart *tchart)
971 {
972 	struct per_pid *new_list, *p, *cursor, *prev;
973 	/* sort by ppid first, then by pid, lowest to highest */
974 
975 	new_list = NULL;
976 
977 	while (tchart->all_data) {
978 		p = tchart->all_data;
979 		tchart->all_data = p->next;
980 		p->next = NULL;
981 
982 		if (new_list == NULL) {
983 			new_list = p;
984 			p->next = NULL;
985 			continue;
986 		}
987 		prev = NULL;
988 		cursor = new_list;
989 		while (cursor) {
990 			if (cursor->ppid > p->ppid ||
991 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
992 				/* must insert before */
993 				if (prev) {
994 					p->next = prev->next;
995 					prev->next = p;
996 					cursor = NULL;
997 					continue;
998 				} else {
999 					p->next = new_list;
1000 					new_list = p;
1001 					cursor = NULL;
1002 					continue;
1003 				}
1004 			}
1005 
1006 			prev = cursor;
1007 			cursor = cursor->next;
1008 			if (!cursor)
1009 				prev->next = p;
1010 		}
1011 	}
1012 	tchart->all_data = new_list;
1013 }
1014 
1015 
1016 static void draw_c_p_states(struct timechart *tchart)
1017 {
1018 	struct power_event *pwr;
1019 	pwr = tchart->power_events;
1020 
1021 	/*
1022 	 * two pass drawing so that the P state bars are on top of the C state blocks
1023 	 */
1024 	while (pwr) {
1025 		if (pwr->type == CSTATE)
1026 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1027 		pwr = pwr->next;
1028 	}
1029 
1030 	pwr = tchart->power_events;
1031 	while (pwr) {
1032 		if (pwr->type == PSTATE) {
1033 			if (!pwr->state)
1034 				pwr->state = tchart->min_freq;
1035 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1036 		}
1037 		pwr = pwr->next;
1038 	}
1039 }
1040 
1041 static void draw_wakeups(struct timechart *tchart)
1042 {
1043 	struct wake_event *we;
1044 	struct per_pid *p;
1045 	struct per_pidcomm *c;
1046 
1047 	we = tchart->wake_events;
1048 	while (we) {
1049 		int from = 0, to = 0;
1050 		char *task_from = NULL, *task_to = NULL;
1051 
1052 		/* locate the column of the waker and wakee */
1053 		p = tchart->all_data;
1054 		while (p) {
1055 			if (p->pid == we->waker || p->pid == we->wakee) {
1056 				c = p->all;
1057 				while (c) {
1058 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1059 						if (p->pid == we->waker && !from) {
1060 							from = c->Y;
1061 							task_from = strdup(c->comm);
1062 						}
1063 						if (p->pid == we->wakee && !to) {
1064 							to = c->Y;
1065 							task_to = strdup(c->comm);
1066 						}
1067 					}
1068 					c = c->next;
1069 				}
1070 				c = p->all;
1071 				while (c) {
1072 					if (p->pid == we->waker && !from) {
1073 						from = c->Y;
1074 						task_from = strdup(c->comm);
1075 					}
1076 					if (p->pid == we->wakee && !to) {
1077 						to = c->Y;
1078 						task_to = strdup(c->comm);
1079 					}
1080 					c = c->next;
1081 				}
1082 			}
1083 			p = p->next;
1084 		}
1085 
1086 		if (!task_from) {
1087 			task_from = malloc(40);
1088 			sprintf(task_from, "[%i]", we->waker);
1089 		}
1090 		if (!task_to) {
1091 			task_to = malloc(40);
1092 			sprintf(task_to, "[%i]", we->wakee);
1093 		}
1094 
1095 		if (we->waker == -1)
1096 			svg_interrupt(we->time, to, we->backtrace);
1097 		else if (from && to && abs(from - to) == 1)
1098 			svg_wakeline(we->time, from, to, we->backtrace);
1099 		else
1100 			svg_partial_wakeline(we->time, from, task_from, to,
1101 					     task_to, we->backtrace);
1102 		we = we->next;
1103 
1104 		free(task_from);
1105 		free(task_to);
1106 	}
1107 }
1108 
1109 static void draw_cpu_usage(struct timechart *tchart)
1110 {
1111 	struct per_pid *p;
1112 	struct per_pidcomm *c;
1113 	struct cpu_sample *sample;
1114 	p = tchart->all_data;
1115 	while (p) {
1116 		c = p->all;
1117 		while (c) {
1118 			sample = c->samples;
1119 			while (sample) {
1120 				if (sample->type == TYPE_RUNNING) {
1121 					svg_process(sample->cpu,
1122 						    sample->start_time,
1123 						    sample->end_time,
1124 						    p->pid,
1125 						    c->comm,
1126 						    sample->backtrace);
1127 				}
1128 
1129 				sample = sample->next;
1130 			}
1131 			c = c->next;
1132 		}
1133 		p = p->next;
1134 	}
1135 }
1136 
1137 static void draw_io_bars(struct timechart *tchart)
1138 {
1139 	const char *suf;
1140 	double bytes;
1141 	char comm[256];
1142 	struct per_pid *p;
1143 	struct per_pidcomm *c;
1144 	struct io_sample *sample;
1145 	int Y = 1;
1146 
1147 	p = tchart->all_data;
1148 	while (p) {
1149 		c = p->all;
1150 		while (c) {
1151 			if (!c->display) {
1152 				c->Y = 0;
1153 				c = c->next;
1154 				continue;
1155 			}
1156 
1157 			svg_box(Y, c->start_time, c->end_time, "process3");
1158 			sample = c->io_samples;
1159 			for (sample = c->io_samples; sample; sample = sample->next) {
1160 				double h = (double)sample->bytes / c->max_bytes;
1161 
1162 				if (tchart->skip_eagain &&
1163 				    sample->err == -EAGAIN)
1164 					continue;
1165 
1166 				if (sample->err)
1167 					h = 1;
1168 
1169 				if (sample->type == IOTYPE_SYNC)
1170 					svg_fbox(Y,
1171 						sample->start_time,
1172 						sample->end_time,
1173 						1,
1174 						sample->err ? "error" : "sync",
1175 						sample->fd,
1176 						sample->err,
1177 						sample->merges);
1178 				else if (sample->type == IOTYPE_POLL)
1179 					svg_fbox(Y,
1180 						sample->start_time,
1181 						sample->end_time,
1182 						1,
1183 						sample->err ? "error" : "poll",
1184 						sample->fd,
1185 						sample->err,
1186 						sample->merges);
1187 				else if (sample->type == IOTYPE_READ)
1188 					svg_ubox(Y,
1189 						sample->start_time,
1190 						sample->end_time,
1191 						h,
1192 						sample->err ? "error" : "disk",
1193 						sample->fd,
1194 						sample->err,
1195 						sample->merges);
1196 				else if (sample->type == IOTYPE_WRITE)
1197 					svg_lbox(Y,
1198 						sample->start_time,
1199 						sample->end_time,
1200 						h,
1201 						sample->err ? "error" : "disk",
1202 						sample->fd,
1203 						sample->err,
1204 						sample->merges);
1205 				else if (sample->type == IOTYPE_RX)
1206 					svg_ubox(Y,
1207 						sample->start_time,
1208 						sample->end_time,
1209 						h,
1210 						sample->err ? "error" : "net",
1211 						sample->fd,
1212 						sample->err,
1213 						sample->merges);
1214 				else if (sample->type == IOTYPE_TX)
1215 					svg_lbox(Y,
1216 						sample->start_time,
1217 						sample->end_time,
1218 						h,
1219 						sample->err ? "error" : "net",
1220 						sample->fd,
1221 						sample->err,
1222 						sample->merges);
1223 			}
1224 
1225 			suf = "";
1226 			bytes = c->total_bytes;
1227 			if (bytes > 1024) {
1228 				bytes = bytes / 1024;
1229 				suf = "K";
1230 			}
1231 			if (bytes > 1024) {
1232 				bytes = bytes / 1024;
1233 				suf = "M";
1234 			}
1235 			if (bytes > 1024) {
1236 				bytes = bytes / 1024;
1237 				suf = "G";
1238 			}
1239 
1240 
1241 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1242 			svg_text(Y, c->start_time, comm);
1243 
1244 			c->Y = Y;
1245 			Y++;
1246 			c = c->next;
1247 		}
1248 		p = p->next;
1249 	}
1250 }
1251 
1252 static void draw_process_bars(struct timechart *tchart)
1253 {
1254 	struct per_pid *p;
1255 	struct per_pidcomm *c;
1256 	struct cpu_sample *sample;
1257 	int Y = 0;
1258 
1259 	Y = 2 * tchart->numcpus + 2;
1260 
1261 	p = tchart->all_data;
1262 	while (p) {
1263 		c = p->all;
1264 		while (c) {
1265 			if (!c->display) {
1266 				c->Y = 0;
1267 				c = c->next;
1268 				continue;
1269 			}
1270 
1271 			svg_box(Y, c->start_time, c->end_time, "process");
1272 			sample = c->samples;
1273 			while (sample) {
1274 				if (sample->type == TYPE_RUNNING)
1275 					svg_running(Y, sample->cpu,
1276 						    sample->start_time,
1277 						    sample->end_time,
1278 						    sample->backtrace);
1279 				if (sample->type == TYPE_BLOCKED)
1280 					svg_blocked(Y, sample->cpu,
1281 						    sample->start_time,
1282 						    sample->end_time,
1283 						    sample->backtrace);
1284 				if (sample->type == TYPE_WAITING)
1285 					svg_waiting(Y, sample->cpu,
1286 						    sample->start_time,
1287 						    sample->end_time,
1288 						    sample->backtrace);
1289 				sample = sample->next;
1290 			}
1291 
1292 			if (c->comm) {
1293 				char comm[256];
1294 				if (c->total_time > 5000000000) /* 5 seconds */
1295 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1296 				else
1297 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1298 
1299 				svg_text(Y, c->start_time, comm);
1300 			}
1301 			c->Y = Y;
1302 			Y++;
1303 			c = c->next;
1304 		}
1305 		p = p->next;
1306 	}
1307 }
1308 
1309 static void add_process_filter(const char *string)
1310 {
1311 	int pid = strtoull(string, NULL, 10);
1312 	struct process_filter *filt = malloc(sizeof(*filt));
1313 
1314 	if (!filt)
1315 		return;
1316 
1317 	filt->name = strdup(string);
1318 	filt->pid  = pid;
1319 	filt->next = process_filter;
1320 
1321 	process_filter = filt;
1322 }
1323 
1324 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1325 {
1326 	struct process_filter *filt;
1327 	if (!process_filter)
1328 		return 1;
1329 
1330 	filt = process_filter;
1331 	while (filt) {
1332 		if (filt->pid && p->pid == filt->pid)
1333 			return 1;
1334 		if (strcmp(filt->name, c->comm) == 0)
1335 			return 1;
1336 		filt = filt->next;
1337 	}
1338 	return 0;
1339 }
1340 
1341 static int determine_display_tasks_filtered(struct timechart *tchart)
1342 {
1343 	struct per_pid *p;
1344 	struct per_pidcomm *c;
1345 	int count = 0;
1346 
1347 	p = tchart->all_data;
1348 	while (p) {
1349 		p->display = 0;
1350 		if (p->start_time == 1)
1351 			p->start_time = tchart->first_time;
1352 
1353 		/* no exit marker, task kept running to the end */
1354 		if (p->end_time == 0)
1355 			p->end_time = tchart->last_time;
1356 
1357 		c = p->all;
1358 
1359 		while (c) {
1360 			c->display = 0;
1361 
1362 			if (c->start_time == 1)
1363 				c->start_time = tchart->first_time;
1364 
1365 			if (passes_filter(p, c)) {
1366 				c->display = 1;
1367 				p->display = 1;
1368 				count++;
1369 			}
1370 
1371 			if (c->end_time == 0)
1372 				c->end_time = tchart->last_time;
1373 
1374 			c = c->next;
1375 		}
1376 		p = p->next;
1377 	}
1378 	return count;
1379 }
1380 
1381 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1382 {
1383 	struct per_pid *p;
1384 	struct per_pidcomm *c;
1385 	int count = 0;
1386 
1387 	p = tchart->all_data;
1388 	while (p) {
1389 		p->display = 0;
1390 		if (p->start_time == 1)
1391 			p->start_time = tchart->first_time;
1392 
1393 		/* no exit marker, task kept running to the end */
1394 		if (p->end_time == 0)
1395 			p->end_time = tchart->last_time;
1396 		if (p->total_time >= threshold)
1397 			p->display = 1;
1398 
1399 		c = p->all;
1400 
1401 		while (c) {
1402 			c->display = 0;
1403 
1404 			if (c->start_time == 1)
1405 				c->start_time = tchart->first_time;
1406 
1407 			if (c->total_time >= threshold) {
1408 				c->display = 1;
1409 				count++;
1410 			}
1411 
1412 			if (c->end_time == 0)
1413 				c->end_time = tchart->last_time;
1414 
1415 			c = c->next;
1416 		}
1417 		p = p->next;
1418 	}
1419 	return count;
1420 }
1421 
1422 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1423 {
1424 	struct per_pid *p;
1425 	struct per_pidcomm *c;
1426 	int count = 0;
1427 
1428 	p = timechart->all_data;
1429 	while (p) {
1430 		/* no exit marker, task kept running to the end */
1431 		if (p->end_time == 0)
1432 			p->end_time = timechart->last_time;
1433 
1434 		c = p->all;
1435 
1436 		while (c) {
1437 			c->display = 0;
1438 
1439 			if (c->total_bytes >= threshold) {
1440 				c->display = 1;
1441 				count++;
1442 			}
1443 
1444 			if (c->end_time == 0)
1445 				c->end_time = timechart->last_time;
1446 
1447 			c = c->next;
1448 		}
1449 		p = p->next;
1450 	}
1451 	return count;
1452 }
1453 
1454 #define BYTES_THRESH (1 * 1024 * 1024)
1455 #define TIME_THRESH 10000000
1456 
1457 static void write_svg_file(struct timechart *tchart, const char *filename)
1458 {
1459 	u64 i;
1460 	int count;
1461 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1462 
1463 	if (tchart->power_only)
1464 		tchart->proc_num = 0;
1465 
1466 	/* We'd like to show at least proc_num tasks;
1467 	 * be less picky if we have fewer */
1468 	do {
1469 		if (process_filter)
1470 			count = determine_display_tasks_filtered(tchart);
1471 		else if (tchart->io_events)
1472 			count = determine_display_io_tasks(tchart, thresh);
1473 		else
1474 			count = determine_display_tasks(tchart, thresh);
1475 		thresh /= 10;
1476 	} while (!process_filter && thresh && count < tchart->proc_num);
1477 
1478 	if (!tchart->proc_num)
1479 		count = 0;
1480 
1481 	if (tchart->io_events) {
1482 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1483 
1484 		svg_time_grid(0.5);
1485 		svg_io_legenda();
1486 
1487 		draw_io_bars(tchart);
1488 	} else {
1489 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1490 
1491 		svg_time_grid(0);
1492 
1493 		svg_legenda();
1494 
1495 		for (i = 0; i < tchart->numcpus; i++)
1496 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1497 
1498 		draw_cpu_usage(tchart);
1499 		if (tchart->proc_num)
1500 			draw_process_bars(tchart);
1501 		if (!tchart->tasks_only)
1502 			draw_c_p_states(tchart);
1503 		if (tchart->proc_num)
1504 			draw_wakeups(tchart);
1505 	}
1506 
1507 	svg_close();
1508 }
1509 
1510 static int process_header(struct perf_file_section *section __maybe_unused,
1511 			  struct perf_header *ph,
1512 			  int feat,
1513 			  int fd __maybe_unused,
1514 			  void *data)
1515 {
1516 	struct timechart *tchart = data;
1517 
1518 	switch (feat) {
1519 	case HEADER_NRCPUS:
1520 		tchart->numcpus = ph->env.nr_cpus_avail;
1521 		break;
1522 
1523 	case HEADER_CPU_TOPOLOGY:
1524 		if (!tchart->topology)
1525 			break;
1526 
1527 		if (svg_build_topology_map(&ph->env))
1528 			fprintf(stderr, "problem building topology\n");
1529 		break;
1530 
1531 	default:
1532 		break;
1533 	}
1534 
1535 	return 0;
1536 }
1537 
1538 static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1539 {
1540 	const struct evsel_str_handler power_tracepoints[] = {
1541 		{ "power:cpu_idle",		process_sample_cpu_idle },
1542 		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1543 		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1544 		{ "sched:sched_switch",		process_sample_sched_switch },
1545 #ifdef SUPPORT_OLD_POWER_EVENTS
1546 		{ "power:power_start",		process_sample_power_start },
1547 		{ "power:power_end",		process_sample_power_end },
1548 		{ "power:power_frequency",	process_sample_power_frequency },
1549 #endif
1550 
1551 		{ "syscalls:sys_enter_read",		process_enter_read },
1552 		{ "syscalls:sys_enter_pread64",		process_enter_read },
1553 		{ "syscalls:sys_enter_readv",		process_enter_read },
1554 		{ "syscalls:sys_enter_preadv",		process_enter_read },
1555 		{ "syscalls:sys_enter_write",		process_enter_write },
1556 		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1557 		{ "syscalls:sys_enter_writev",		process_enter_write },
1558 		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1559 		{ "syscalls:sys_enter_sync",		process_enter_sync },
1560 		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1561 		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1562 		{ "syscalls:sys_enter_msync",		process_enter_sync },
1563 		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1564 		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1565 		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1566 		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1567 		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1568 		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1569 		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1570 		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1571 		{ "syscalls:sys_enter_poll",		process_enter_poll },
1572 		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1573 		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1574 		{ "syscalls:sys_enter_select",		process_enter_poll },
1575 
1576 		{ "syscalls:sys_exit_read",		process_exit_read },
1577 		{ "syscalls:sys_exit_pread64",		process_exit_read },
1578 		{ "syscalls:sys_exit_readv",		process_exit_read },
1579 		{ "syscalls:sys_exit_preadv",		process_exit_read },
1580 		{ "syscalls:sys_exit_write",		process_exit_write },
1581 		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1582 		{ "syscalls:sys_exit_writev",		process_exit_write },
1583 		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1584 		{ "syscalls:sys_exit_sync",		process_exit_sync },
1585 		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1586 		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1587 		{ "syscalls:sys_exit_msync",		process_exit_sync },
1588 		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1589 		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1590 		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1591 		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1592 		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1593 		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1594 		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1595 		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1596 		{ "syscalls:sys_exit_poll",		process_exit_poll },
1597 		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1598 		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1599 		{ "syscalls:sys_exit_select",		process_exit_poll },
1600 	};
1601 	struct perf_data data = {
1602 		.path  = input_name,
1603 		.mode  = PERF_DATA_MODE_READ,
1604 		.force = tchart->force,
1605 	};
1606 
1607 	struct perf_session *session = perf_session__new(&data, &tchart->tool);
1608 	int ret = -EINVAL;
1609 
1610 	if (IS_ERR(session))
1611 		return PTR_ERR(session);
1612 
1613 	symbol__init(&session->header.env);
1614 
1615 	(void)perf_header__process_sections(&session->header,
1616 					    perf_data__fd(session->data),
1617 					    tchart,
1618 					    process_header);
1619 
1620 	if (!perf_session__has_traces(session, "timechart record"))
1621 		goto out_delete;
1622 
1623 	if (perf_session__set_tracepoints_handlers(session,
1624 						   power_tracepoints)) {
1625 		pr_err("Initializing session tracepoint handlers failed\n");
1626 		goto out_delete;
1627 	}
1628 
1629 	ret = perf_session__process_events(session);
1630 	if (ret)
1631 		goto out_delete;
1632 
1633 	end_sample_processing(tchart);
1634 
1635 	sort_pids(tchart);
1636 
1637 	write_svg_file(tchart, output_name);
1638 
1639 	pr_info("Written %2.1f seconds of trace to %s.\n",
1640 		(tchart->last_time - tchart->first_time) / (double)NSEC_PER_SEC, output_name);
1641 out_delete:
1642 	perf_session__delete(session);
1643 	return ret;
1644 }
1645 
1646 static int timechart__io_record(int argc, const char **argv)
1647 {
1648 	unsigned int rec_argc, i;
1649 	const char **rec_argv;
1650 	const char **p;
1651 	char *filter = NULL;
1652 
1653 	const char * const common_args[] = {
1654 		"record", "-a", "-R", "-c", "1",
1655 	};
1656 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1657 
1658 	const char * const disk_events[] = {
1659 		"syscalls:sys_enter_read",
1660 		"syscalls:sys_enter_pread64",
1661 		"syscalls:sys_enter_readv",
1662 		"syscalls:sys_enter_preadv",
1663 		"syscalls:sys_enter_write",
1664 		"syscalls:sys_enter_pwrite64",
1665 		"syscalls:sys_enter_writev",
1666 		"syscalls:sys_enter_pwritev",
1667 		"syscalls:sys_enter_sync",
1668 		"syscalls:sys_enter_sync_file_range",
1669 		"syscalls:sys_enter_fsync",
1670 		"syscalls:sys_enter_msync",
1671 
1672 		"syscalls:sys_exit_read",
1673 		"syscalls:sys_exit_pread64",
1674 		"syscalls:sys_exit_readv",
1675 		"syscalls:sys_exit_preadv",
1676 		"syscalls:sys_exit_write",
1677 		"syscalls:sys_exit_pwrite64",
1678 		"syscalls:sys_exit_writev",
1679 		"syscalls:sys_exit_pwritev",
1680 		"syscalls:sys_exit_sync",
1681 		"syscalls:sys_exit_sync_file_range",
1682 		"syscalls:sys_exit_fsync",
1683 		"syscalls:sys_exit_msync",
1684 	};
1685 	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1686 
1687 	const char * const net_events[] = {
1688 		"syscalls:sys_enter_recvfrom",
1689 		"syscalls:sys_enter_recvmmsg",
1690 		"syscalls:sys_enter_recvmsg",
1691 		"syscalls:sys_enter_sendto",
1692 		"syscalls:sys_enter_sendmsg",
1693 		"syscalls:sys_enter_sendmmsg",
1694 
1695 		"syscalls:sys_exit_recvfrom",
1696 		"syscalls:sys_exit_recvmmsg",
1697 		"syscalls:sys_exit_recvmsg",
1698 		"syscalls:sys_exit_sendto",
1699 		"syscalls:sys_exit_sendmsg",
1700 		"syscalls:sys_exit_sendmmsg",
1701 	};
1702 	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1703 
1704 	const char * const poll_events[] = {
1705 		"syscalls:sys_enter_epoll_pwait",
1706 		"syscalls:sys_enter_epoll_wait",
1707 		"syscalls:sys_enter_poll",
1708 		"syscalls:sys_enter_ppoll",
1709 		"syscalls:sys_enter_pselect6",
1710 		"syscalls:sys_enter_select",
1711 
1712 		"syscalls:sys_exit_epoll_pwait",
1713 		"syscalls:sys_exit_epoll_wait",
1714 		"syscalls:sys_exit_poll",
1715 		"syscalls:sys_exit_ppoll",
1716 		"syscalls:sys_exit_pselect6",
1717 		"syscalls:sys_exit_select",
1718 	};
1719 	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1720 
1721 	rec_argc = common_args_nr +
1722 		disk_events_nr * 4 +
1723 		net_events_nr * 4 +
1724 		poll_events_nr * 4 +
1725 		argc;
1726 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1727 
1728 	if (rec_argv == NULL)
1729 		return -ENOMEM;
1730 
1731 	if (asprintf(&filter, "common_pid != %d", getpid()) < 0) {
1732 		free(rec_argv);
1733 		return -ENOMEM;
1734 	}
1735 
1736 	p = rec_argv;
1737 	for (i = 0; i < common_args_nr; i++)
1738 		*p++ = strdup(common_args[i]);
1739 
1740 	for (i = 0; i < disk_events_nr; i++) {
1741 		if (!is_valid_tracepoint(disk_events[i])) {
1742 			rec_argc -= 4;
1743 			continue;
1744 		}
1745 
1746 		*p++ = "-e";
1747 		*p++ = strdup(disk_events[i]);
1748 		*p++ = "--filter";
1749 		*p++ = filter;
1750 	}
1751 	for (i = 0; i < net_events_nr; i++) {
1752 		if (!is_valid_tracepoint(net_events[i])) {
1753 			rec_argc -= 4;
1754 			continue;
1755 		}
1756 
1757 		*p++ = "-e";
1758 		*p++ = strdup(net_events[i]);
1759 		*p++ = "--filter";
1760 		*p++ = filter;
1761 	}
1762 	for (i = 0; i < poll_events_nr; i++) {
1763 		if (!is_valid_tracepoint(poll_events[i])) {
1764 			rec_argc -= 4;
1765 			continue;
1766 		}
1767 
1768 		*p++ = "-e";
1769 		*p++ = strdup(poll_events[i]);
1770 		*p++ = "--filter";
1771 		*p++ = filter;
1772 	}
1773 
1774 	for (i = 0; i < (unsigned int)argc; i++)
1775 		*p++ = argv[i];
1776 
1777 	return cmd_record(rec_argc, rec_argv);
1778 }
1779 
1780 
1781 static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1782 {
1783 	unsigned int rec_argc, i, j;
1784 	const char **rec_argv;
1785 	const char **p;
1786 	unsigned int record_elems;
1787 
1788 	const char * const common_args[] = {
1789 		"record", "-a", "-R", "-c", "1",
1790 	};
1791 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1792 
1793 	const char * const backtrace_args[] = {
1794 		"-g",
1795 	};
1796 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1797 
1798 	const char * const power_args[] = {
1799 		"-e", "power:cpu_frequency",
1800 		"-e", "power:cpu_idle",
1801 	};
1802 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1803 
1804 	const char * const old_power_args[] = {
1805 #ifdef SUPPORT_OLD_POWER_EVENTS
1806 		"-e", "power:power_start",
1807 		"-e", "power:power_end",
1808 		"-e", "power:power_frequency",
1809 #endif
1810 	};
1811 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1812 
1813 	const char * const tasks_args[] = {
1814 		"-e", "sched:sched_wakeup",
1815 		"-e", "sched:sched_switch",
1816 	};
1817 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1818 
1819 #ifdef SUPPORT_OLD_POWER_EVENTS
1820 	if (!is_valid_tracepoint("power:cpu_idle") &&
1821 	    is_valid_tracepoint("power:power_start")) {
1822 		use_old_power_events = 1;
1823 		power_args_nr = 0;
1824 	} else {
1825 		old_power_args_nr = 0;
1826 	}
1827 #endif
1828 
1829 	if (tchart->power_only)
1830 		tasks_args_nr = 0;
1831 
1832 	if (tchart->tasks_only) {
1833 		power_args_nr = 0;
1834 		old_power_args_nr = 0;
1835 	}
1836 
1837 	if (!tchart->with_backtrace)
1838 		backtrace_args_no = 0;
1839 
1840 	record_elems = common_args_nr + tasks_args_nr +
1841 		power_args_nr + old_power_args_nr + backtrace_args_no;
1842 
1843 	rec_argc = record_elems + argc;
1844 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1845 
1846 	if (rec_argv == NULL)
1847 		return -ENOMEM;
1848 
1849 	p = rec_argv;
1850 	for (i = 0; i < common_args_nr; i++)
1851 		*p++ = strdup(common_args[i]);
1852 
1853 	for (i = 0; i < backtrace_args_no; i++)
1854 		*p++ = strdup(backtrace_args[i]);
1855 
1856 	for (i = 0; i < tasks_args_nr; i++)
1857 		*p++ = strdup(tasks_args[i]);
1858 
1859 	for (i = 0; i < power_args_nr; i++)
1860 		*p++ = strdup(power_args[i]);
1861 
1862 	for (i = 0; i < old_power_args_nr; i++)
1863 		*p++ = strdup(old_power_args[i]);
1864 
1865 	for (j = 0; j < (unsigned int)argc; j++)
1866 		*p++ = argv[j];
1867 
1868 	return cmd_record(rec_argc, rec_argv);
1869 }
1870 
1871 static int
1872 parse_process(const struct option *opt __maybe_unused, const char *arg,
1873 	      int __maybe_unused unset)
1874 {
1875 	if (arg)
1876 		add_process_filter(arg);
1877 	return 0;
1878 }
1879 
1880 static int
1881 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1882 		int __maybe_unused unset)
1883 {
1884 	unsigned long duration = strtoul(arg, NULL, 0);
1885 
1886 	if (svg_highlight || svg_highlight_name)
1887 		return -1;
1888 
1889 	if (duration)
1890 		svg_highlight = duration;
1891 	else
1892 		svg_highlight_name = strdup(arg);
1893 
1894 	return 0;
1895 }
1896 
1897 static int
1898 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1899 {
1900 	char unit = 'n';
1901 	u64 *value = opt->value;
1902 
1903 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1904 		switch (unit) {
1905 		case 'm':
1906 			*value *= NSEC_PER_MSEC;
1907 			break;
1908 		case 'u':
1909 			*value *= NSEC_PER_USEC;
1910 			break;
1911 		case 'n':
1912 			break;
1913 		default:
1914 			return -1;
1915 		}
1916 	}
1917 
1918 	return 0;
1919 }
1920 
1921 int cmd_timechart(int argc, const char **argv)
1922 {
1923 	struct timechart tchart = {
1924 		.tool = {
1925 			.comm		 = process_comm_event,
1926 			.fork		 = process_fork_event,
1927 			.exit		 = process_exit_event,
1928 			.sample		 = process_sample_event,
1929 			.ordered_events	 = true,
1930 		},
1931 		.proc_num = 15,
1932 		.min_time = NSEC_PER_MSEC,
1933 		.merge_dist = 1000,
1934 	};
1935 	const char *output_name = "output.svg";
1936 	const struct option timechart_common_options[] = {
1937 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1938 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only, "output processes data only"),
1939 	OPT_END()
1940 	};
1941 	const struct option timechart_options[] = {
1942 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1943 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1944 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1945 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1946 		      "highlight tasks. Pass duration in ns or process name.",
1947 		       parse_highlight),
1948 	OPT_CALLBACK('p', "process", NULL, "process",
1949 		      "process selector. Pass a pid or process name.",
1950 		       parse_process),
1951 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1952 		     "Look for files with symbols relative to this directory",
1953 		     symbol__config_symfs),
1954 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1955 		    "min. number of tasks to print"),
1956 	OPT_BOOLEAN('t', "topology", &tchart.topology,
1957 		    "sort CPUs according to topology"),
1958 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1959 		    "skip EAGAIN errors"),
1960 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1961 		     "all IO faster than min-time will visually appear longer",
1962 		     parse_time),
1963 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1964 		     "merge events that are merge-dist us apart",
1965 		     parse_time),
1966 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1967 	OPT_PARENT(timechart_common_options),
1968 	};
1969 	const char * const timechart_subcommands[] = { "record", NULL };
1970 	const char *timechart_usage[] = {
1971 		"perf timechart [<options>] {record}",
1972 		NULL
1973 	};
1974 	const struct option timechart_record_options[] = {
1975 	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
1976 		    "record only IO data"),
1977 	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
1978 	OPT_PARENT(timechart_common_options),
1979 	};
1980 	const char * const timechart_record_usage[] = {
1981 		"perf timechart record [<options>]",
1982 		NULL
1983 	};
1984 	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
1985 			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1986 
1987 	if (tchart.power_only && tchart.tasks_only) {
1988 		pr_err("-P and -T options cannot be used at the same time.\n");
1989 		return -1;
1990 	}
1991 
1992 	if (argc && strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
1993 		argc = parse_options(argc, argv, timechart_record_options,
1994 				     timechart_record_usage,
1995 				     PARSE_OPT_STOP_AT_NON_OPTION);
1996 
1997 		if (tchart.power_only && tchart.tasks_only) {
1998 			pr_err("-P and -T options cannot be used at the same time.\n");
1999 			return -1;
2000 		}
2001 
2002 		if (tchart.io_only)
2003 			return timechart__io_record(argc, argv);
2004 		else
2005 			return timechart__record(&tchart, argc, argv);
2006 	} else if (argc)
2007 		usage_with_options(timechart_usage, timechart_options);
2008 
2009 	setup_pager();
2010 
2011 	return __cmd_timechart(&tchart, output_name);
2012 }
2013