xref: /linux/tools/perf/builtin-timechart.c (revision 63307d015b91e626c97bb82e88054af3d0b74643)
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 #include <traceevent/event-parse.h>
14 
15 #include "builtin.h"
16 
17 #include "util/util.h"
18 
19 #include "util/color.h"
20 #include <linux/list.h>
21 #include "util/cache.h"
22 #include "util/evlist.h"
23 #include "util/evsel.h"
24 #include <linux/kernel.h>
25 #include <linux/rbtree.h>
26 #include <linux/time64.h>
27 #include "util/symbol.h"
28 #include "util/thread.h"
29 #include "util/callchain.h"
30 
31 #include "perf.h"
32 #include "util/header.h"
33 #include <subcmd/parse-options.h>
34 #include "util/parse-events.h"
35 #include "util/event.h"
36 #include "util/session.h"
37 #include "util/svghelper.h"
38 #include "util/tool.h"
39 #include "util/data.h"
40 #include "util/debug.h"
41 
42 #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[0];
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 void pid_set_comm(struct timechart *tchart, int pid, char *comm)
219 {
220 	struct per_pid *p;
221 	struct per_pidcomm *c;
222 	p = find_create_pid(tchart, pid);
223 	c = p->all;
224 	while (c) {
225 		if (c->comm && strcmp(c->comm, comm) == 0) {
226 			p->current = c;
227 			return;
228 		}
229 		if (!c->comm) {
230 			c->comm = strdup(comm);
231 			p->current = c;
232 			return;
233 		}
234 		c = c->next;
235 	}
236 	c = zalloc(sizeof(*c));
237 	assert(c != NULL);
238 	c->comm = strdup(comm);
239 	p->current = c;
240 	c->next = p->all;
241 	p->all = c;
242 }
243 
244 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
245 {
246 	struct per_pid *p, *pp;
247 	p = find_create_pid(tchart, pid);
248 	pp = find_create_pid(tchart, ppid);
249 	p->ppid = ppid;
250 	if (pp->current && pp->current->comm && !p->current)
251 		pid_set_comm(tchart, pid, pp->current->comm);
252 
253 	p->start_time = timestamp;
254 	if (p->current && !p->current->start_time) {
255 		p->current->start_time = timestamp;
256 		p->current->state_since = timestamp;
257 	}
258 }
259 
260 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
261 {
262 	struct per_pid *p;
263 	p = find_create_pid(tchart, pid);
264 	p->end_time = timestamp;
265 	if (p->current)
266 		p->current->end_time = timestamp;
267 }
268 
269 static void pid_put_sample(struct timechart *tchart, int pid, int type,
270 			   unsigned int cpu, u64 start, u64 end,
271 			   const char *backtrace)
272 {
273 	struct per_pid *p;
274 	struct per_pidcomm *c;
275 	struct cpu_sample *sample;
276 
277 	p = find_create_pid(tchart, pid);
278 	c = p->current;
279 	if (!c) {
280 		c = zalloc(sizeof(*c));
281 		assert(c != NULL);
282 		p->current = c;
283 		c->next = p->all;
284 		p->all = c;
285 	}
286 
287 	sample = zalloc(sizeof(*sample));
288 	assert(sample != NULL);
289 	sample->start_time = start;
290 	sample->end_time = end;
291 	sample->type = type;
292 	sample->next = c->samples;
293 	sample->cpu = cpu;
294 	sample->backtrace = backtrace;
295 	c->samples = sample;
296 
297 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
298 		c->total_time += (end-start);
299 		p->total_time += (end-start);
300 	}
301 
302 	if (c->start_time == 0 || c->start_time > start)
303 		c->start_time = start;
304 	if (p->start_time == 0 || p->start_time > start)
305 		p->start_time = start;
306 }
307 
308 #define MAX_CPUS 4096
309 
310 static u64 cpus_cstate_start_times[MAX_CPUS];
311 static int cpus_cstate_state[MAX_CPUS];
312 static u64 cpus_pstate_start_times[MAX_CPUS];
313 static u64 cpus_pstate_state[MAX_CPUS];
314 
315 static int process_comm_event(struct perf_tool *tool,
316 			      union perf_event *event,
317 			      struct perf_sample *sample __maybe_unused,
318 			      struct machine *machine __maybe_unused)
319 {
320 	struct timechart *tchart = container_of(tool, struct timechart, tool);
321 	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
322 	return 0;
323 }
324 
325 static int process_fork_event(struct perf_tool *tool,
326 			      union perf_event *event,
327 			      struct perf_sample *sample __maybe_unused,
328 			      struct machine *machine __maybe_unused)
329 {
330 	struct timechart *tchart = container_of(tool, struct timechart, tool);
331 	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
332 	return 0;
333 }
334 
335 static int process_exit_event(struct perf_tool *tool,
336 			      union perf_event *event,
337 			      struct perf_sample *sample __maybe_unused,
338 			      struct machine *machine __maybe_unused)
339 {
340 	struct timechart *tchart = container_of(tool, struct timechart, tool);
341 	pid_exit(tchart, event->fork.pid, event->fork.time);
342 	return 0;
343 }
344 
345 #ifdef SUPPORT_OLD_POWER_EVENTS
346 static int use_old_power_events;
347 #endif
348 
349 static void c_state_start(int cpu, u64 timestamp, int state)
350 {
351 	cpus_cstate_start_times[cpu] = timestamp;
352 	cpus_cstate_state[cpu] = state;
353 }
354 
355 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
356 {
357 	struct power_event *pwr = zalloc(sizeof(*pwr));
358 
359 	if (!pwr)
360 		return;
361 
362 	pwr->state = cpus_cstate_state[cpu];
363 	pwr->start_time = cpus_cstate_start_times[cpu];
364 	pwr->end_time = timestamp;
365 	pwr->cpu = cpu;
366 	pwr->type = CSTATE;
367 	pwr->next = tchart->power_events;
368 
369 	tchart->power_events = pwr;
370 }
371 
372 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
373 {
374 	struct power_event *pwr;
375 
376 	if (new_freq > 8000000) /* detect invalid data */
377 		return;
378 
379 	pwr = zalloc(sizeof(*pwr));
380 	if (!pwr)
381 		return;
382 
383 	pwr->state = cpus_pstate_state[cpu];
384 	pwr->start_time = cpus_pstate_start_times[cpu];
385 	pwr->end_time = timestamp;
386 	pwr->cpu = cpu;
387 	pwr->type = PSTATE;
388 	pwr->next = tchart->power_events;
389 
390 	if (!pwr->start_time)
391 		pwr->start_time = tchart->first_time;
392 
393 	tchart->power_events = pwr;
394 
395 	cpus_pstate_state[cpu] = new_freq;
396 	cpus_pstate_start_times[cpu] = timestamp;
397 
398 	if ((u64)new_freq > tchart->max_freq)
399 		tchart->max_freq = new_freq;
400 
401 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
402 		tchart->min_freq = new_freq;
403 
404 	if (new_freq == tchart->max_freq - 1000)
405 		tchart->turbo_frequency = tchart->max_freq;
406 }
407 
408 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
409 			 int waker, int wakee, u8 flags, const char *backtrace)
410 {
411 	struct per_pid *p;
412 	struct wake_event *we = zalloc(sizeof(*we));
413 
414 	if (!we)
415 		return;
416 
417 	we->time = timestamp;
418 	we->waker = waker;
419 	we->backtrace = backtrace;
420 
421 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
422 		we->waker = -1;
423 
424 	we->wakee = wakee;
425 	we->next = tchart->wake_events;
426 	tchart->wake_events = we;
427 	p = find_create_pid(tchart, we->wakee);
428 
429 	if (p && p->current && p->current->state == TYPE_NONE) {
430 		p->current->state_since = timestamp;
431 		p->current->state = TYPE_WAITING;
432 	}
433 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
434 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
435 			       p->current->state_since, timestamp, NULL);
436 		p->current->state_since = timestamp;
437 		p->current->state = TYPE_WAITING;
438 	}
439 }
440 
441 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
442 			 int prev_pid, int next_pid, u64 prev_state,
443 			 const char *backtrace)
444 {
445 	struct per_pid *p = NULL, *prev_p;
446 
447 	prev_p = find_create_pid(tchart, prev_pid);
448 
449 	p = find_create_pid(tchart, next_pid);
450 
451 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
452 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
453 			       prev_p->current->state_since, timestamp,
454 			       backtrace);
455 	if (p && p->current) {
456 		if (p->current->state != TYPE_NONE)
457 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
458 				       p->current->state_since, timestamp,
459 				       backtrace);
460 
461 		p->current->state_since = timestamp;
462 		p->current->state = TYPE_RUNNING;
463 	}
464 
465 	if (prev_p->current) {
466 		prev_p->current->state = TYPE_NONE;
467 		prev_p->current->state_since = timestamp;
468 		if (prev_state & 2)
469 			prev_p->current->state = TYPE_BLOCKED;
470 		if (prev_state == 0)
471 			prev_p->current->state = TYPE_WAITING;
472 	}
473 }
474 
475 static const char *cat_backtrace(union perf_event *event,
476 				 struct perf_sample *sample,
477 				 struct machine *machine)
478 {
479 	struct addr_location al;
480 	unsigned int i;
481 	char *p = NULL;
482 	size_t p_len;
483 	u8 cpumode = PERF_RECORD_MISC_USER;
484 	struct addr_location tal;
485 	struct ip_callchain *chain = sample->callchain;
486 	FILE *f = open_memstream(&p, &p_len);
487 
488 	if (!f) {
489 		perror("open_memstream error");
490 		return NULL;
491 	}
492 
493 	if (!chain)
494 		goto exit;
495 
496 	if (machine__resolve(machine, &al, sample) < 0) {
497 		fprintf(stderr, "problem processing %d event, skipping it.\n",
498 			event->header.type);
499 		goto exit;
500 	}
501 
502 	for (i = 0; i < chain->nr; i++) {
503 		u64 ip;
504 
505 		if (callchain_param.order == ORDER_CALLEE)
506 			ip = chain->ips[i];
507 		else
508 			ip = chain->ips[chain->nr - i - 1];
509 
510 		if (ip >= PERF_CONTEXT_MAX) {
511 			switch (ip) {
512 			case PERF_CONTEXT_HV:
513 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
514 				break;
515 			case PERF_CONTEXT_KERNEL:
516 				cpumode = PERF_RECORD_MISC_KERNEL;
517 				break;
518 			case PERF_CONTEXT_USER:
519 				cpumode = PERF_RECORD_MISC_USER;
520 				break;
521 			default:
522 				pr_debug("invalid callchain context: "
523 					 "%"PRId64"\n", (s64) ip);
524 
525 				/*
526 				 * It seems the callchain is corrupted.
527 				 * Discard all.
528 				 */
529 				zfree(&p);
530 				goto exit_put;
531 			}
532 			continue;
533 		}
534 
535 		tal.filtered = 0;
536 		if (thread__find_symbol(al.thread, cpumode, ip, &tal))
537 			fprintf(f, "..... %016" PRIx64 " %s\n", ip, tal.sym->name);
538 		else
539 			fprintf(f, "..... %016" PRIx64 "\n", ip);
540 	}
541 exit_put:
542 	addr_location__put(&al);
543 exit:
544 	fclose(f);
545 
546 	return p;
547 }
548 
549 typedef int (*tracepoint_handler)(struct timechart *tchart,
550 				  struct perf_evsel *evsel,
551 				  struct perf_sample *sample,
552 				  const char *backtrace);
553 
554 static int process_sample_event(struct perf_tool *tool,
555 				union perf_event *event,
556 				struct perf_sample *sample,
557 				struct perf_evsel *evsel,
558 				struct machine *machine)
559 {
560 	struct timechart *tchart = container_of(tool, struct timechart, tool);
561 
562 	if (evsel->attr.sample_type & PERF_SAMPLE_TIME) {
563 		if (!tchart->first_time || tchart->first_time > sample->time)
564 			tchart->first_time = sample->time;
565 		if (tchart->last_time < sample->time)
566 			tchart->last_time = sample->time;
567 	}
568 
569 	if (evsel->handler != NULL) {
570 		tracepoint_handler f = evsel->handler;
571 		return f(tchart, evsel, sample,
572 			 cat_backtrace(event, sample, machine));
573 	}
574 
575 	return 0;
576 }
577 
578 static int
579 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
580 			struct perf_evsel *evsel,
581 			struct perf_sample *sample,
582 			const char *backtrace __maybe_unused)
583 {
584 	u32 state = perf_evsel__intval(evsel, sample, "state");
585 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
586 
587 	if (state == (u32)PWR_EVENT_EXIT)
588 		c_state_end(tchart, cpu_id, sample->time);
589 	else
590 		c_state_start(cpu_id, sample->time, state);
591 	return 0;
592 }
593 
594 static int
595 process_sample_cpu_frequency(struct timechart *tchart,
596 			     struct perf_evsel *evsel,
597 			     struct perf_sample *sample,
598 			     const char *backtrace __maybe_unused)
599 {
600 	u32 state = perf_evsel__intval(evsel, sample, "state");
601 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
602 
603 	p_state_change(tchart, cpu_id, sample->time, state);
604 	return 0;
605 }
606 
607 static int
608 process_sample_sched_wakeup(struct timechart *tchart,
609 			    struct perf_evsel *evsel,
610 			    struct perf_sample *sample,
611 			    const char *backtrace)
612 {
613 	u8 flags = perf_evsel__intval(evsel, sample, "common_flags");
614 	int waker = perf_evsel__intval(evsel, sample, "common_pid");
615 	int wakee = perf_evsel__intval(evsel, sample, "pid");
616 
617 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
618 	return 0;
619 }
620 
621 static int
622 process_sample_sched_switch(struct timechart *tchart,
623 			    struct perf_evsel *evsel,
624 			    struct perf_sample *sample,
625 			    const char *backtrace)
626 {
627 	int prev_pid = perf_evsel__intval(evsel, sample, "prev_pid");
628 	int next_pid = perf_evsel__intval(evsel, sample, "next_pid");
629 	u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
630 
631 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
632 		     prev_state, backtrace);
633 	return 0;
634 }
635 
636 #ifdef SUPPORT_OLD_POWER_EVENTS
637 static int
638 process_sample_power_start(struct timechart *tchart __maybe_unused,
639 			   struct perf_evsel *evsel,
640 			   struct perf_sample *sample,
641 			   const char *backtrace __maybe_unused)
642 {
643 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
644 	u64 value = perf_evsel__intval(evsel, sample, "value");
645 
646 	c_state_start(cpu_id, sample->time, value);
647 	return 0;
648 }
649 
650 static int
651 process_sample_power_end(struct timechart *tchart,
652 			 struct perf_evsel *evsel __maybe_unused,
653 			 struct perf_sample *sample,
654 			 const char *backtrace __maybe_unused)
655 {
656 	c_state_end(tchart, sample->cpu, sample->time);
657 	return 0;
658 }
659 
660 static int
661 process_sample_power_frequency(struct timechart *tchart,
662 			       struct perf_evsel *evsel,
663 			       struct perf_sample *sample,
664 			       const char *backtrace __maybe_unused)
665 {
666 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
667 	u64 value = perf_evsel__intval(evsel, sample, "value");
668 
669 	p_state_change(tchart, cpu_id, sample->time, value);
670 	return 0;
671 }
672 #endif /* SUPPORT_OLD_POWER_EVENTS */
673 
674 /*
675  * After the last sample we need to wrap up the current C/P state
676  * and close out each CPU for these.
677  */
678 static void end_sample_processing(struct timechart *tchart)
679 {
680 	u64 cpu;
681 	struct power_event *pwr;
682 
683 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
684 		/* C state */
685 #if 0
686 		pwr = zalloc(sizeof(*pwr));
687 		if (!pwr)
688 			return;
689 
690 		pwr->state = cpus_cstate_state[cpu];
691 		pwr->start_time = cpus_cstate_start_times[cpu];
692 		pwr->end_time = tchart->last_time;
693 		pwr->cpu = cpu;
694 		pwr->type = CSTATE;
695 		pwr->next = tchart->power_events;
696 
697 		tchart->power_events = pwr;
698 #endif
699 		/* P state */
700 
701 		pwr = zalloc(sizeof(*pwr));
702 		if (!pwr)
703 			return;
704 
705 		pwr->state = cpus_pstate_state[cpu];
706 		pwr->start_time = cpus_pstate_start_times[cpu];
707 		pwr->end_time = tchart->last_time;
708 		pwr->cpu = cpu;
709 		pwr->type = PSTATE;
710 		pwr->next = tchart->power_events;
711 
712 		if (!pwr->start_time)
713 			pwr->start_time = tchart->first_time;
714 		if (!pwr->state)
715 			pwr->state = tchart->min_freq;
716 		tchart->power_events = pwr;
717 	}
718 }
719 
720 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
721 			       u64 start, int fd)
722 {
723 	struct per_pid *p = find_create_pid(tchart, pid);
724 	struct per_pidcomm *c = p->current;
725 	struct io_sample *sample;
726 	struct io_sample *prev;
727 
728 	if (!c) {
729 		c = zalloc(sizeof(*c));
730 		if (!c)
731 			return -ENOMEM;
732 		p->current = c;
733 		c->next = p->all;
734 		p->all = c;
735 	}
736 
737 	prev = c->io_samples;
738 
739 	if (prev && prev->start_time && !prev->end_time) {
740 		pr_warning("Skip invalid start event: "
741 			   "previous event already started!\n");
742 
743 		/* remove previous event that has been started,
744 		 * we are not sure we will ever get an end for it */
745 		c->io_samples = prev->next;
746 		free(prev);
747 		return 0;
748 	}
749 
750 	sample = zalloc(sizeof(*sample));
751 	if (!sample)
752 		return -ENOMEM;
753 	sample->start_time = start;
754 	sample->type = type;
755 	sample->fd = fd;
756 	sample->next = c->io_samples;
757 	c->io_samples = sample;
758 
759 	if (c->start_time == 0 || c->start_time > start)
760 		c->start_time = start;
761 
762 	return 0;
763 }
764 
765 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
766 			     u64 end, long ret)
767 {
768 	struct per_pid *p = find_create_pid(tchart, pid);
769 	struct per_pidcomm *c = p->current;
770 	struct io_sample *sample, *prev;
771 
772 	if (!c) {
773 		pr_warning("Invalid pidcomm!\n");
774 		return -1;
775 	}
776 
777 	sample = c->io_samples;
778 
779 	if (!sample) /* skip partially captured events */
780 		return 0;
781 
782 	if (sample->end_time) {
783 		pr_warning("Skip invalid end event: "
784 			   "previous event already ended!\n");
785 		return 0;
786 	}
787 
788 	if (sample->type != type) {
789 		pr_warning("Skip invalid end event: invalid event type!\n");
790 		return 0;
791 	}
792 
793 	sample->end_time = end;
794 	prev = sample->next;
795 
796 	/* we want to be able to see small and fast transfers, so make them
797 	 * at least min_time long, but don't overlap them */
798 	if (sample->end_time - sample->start_time < tchart->min_time)
799 		sample->end_time = sample->start_time + tchart->min_time;
800 	if (prev && sample->start_time < prev->end_time) {
801 		if (prev->err) /* try to make errors more visible */
802 			sample->start_time = prev->end_time;
803 		else
804 			prev->end_time = sample->start_time;
805 	}
806 
807 	if (ret < 0) {
808 		sample->err = ret;
809 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
810 		   type == IOTYPE_TX || type == IOTYPE_RX) {
811 
812 		if ((u64)ret > c->max_bytes)
813 			c->max_bytes = ret;
814 
815 		c->total_bytes += ret;
816 		p->total_bytes += ret;
817 		sample->bytes = ret;
818 	}
819 
820 	/* merge two requests to make svg smaller and render-friendly */
821 	if (prev &&
822 	    prev->type == sample->type &&
823 	    prev->err == sample->err &&
824 	    prev->fd == sample->fd &&
825 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
826 
827 		sample->bytes += prev->bytes;
828 		sample->merges += prev->merges + 1;
829 
830 		sample->start_time = prev->start_time;
831 		sample->next = prev->next;
832 		free(prev);
833 
834 		if (!sample->err && sample->bytes > c->max_bytes)
835 			c->max_bytes = sample->bytes;
836 	}
837 
838 	tchart->io_events++;
839 
840 	return 0;
841 }
842 
843 static int
844 process_enter_read(struct timechart *tchart,
845 		   struct perf_evsel *evsel,
846 		   struct perf_sample *sample)
847 {
848 	long fd = perf_evsel__intval(evsel, sample, "fd");
849 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
850 				   sample->time, fd);
851 }
852 
853 static int
854 process_exit_read(struct timechart *tchart,
855 		  struct perf_evsel *evsel,
856 		  struct perf_sample *sample)
857 {
858 	long ret = perf_evsel__intval(evsel, sample, "ret");
859 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
860 				 sample->time, ret);
861 }
862 
863 static int
864 process_enter_write(struct timechart *tchart,
865 		    struct perf_evsel *evsel,
866 		    struct perf_sample *sample)
867 {
868 	long fd = perf_evsel__intval(evsel, sample, "fd");
869 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
870 				   sample->time, fd);
871 }
872 
873 static int
874 process_exit_write(struct timechart *tchart,
875 		   struct perf_evsel *evsel,
876 		   struct perf_sample *sample)
877 {
878 	long ret = perf_evsel__intval(evsel, sample, "ret");
879 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
880 				 sample->time, ret);
881 }
882 
883 static int
884 process_enter_sync(struct timechart *tchart,
885 		   struct perf_evsel *evsel,
886 		   struct perf_sample *sample)
887 {
888 	long fd = perf_evsel__intval(evsel, sample, "fd");
889 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
890 				   sample->time, fd);
891 }
892 
893 static int
894 process_exit_sync(struct timechart *tchart,
895 		  struct perf_evsel *evsel,
896 		  struct perf_sample *sample)
897 {
898 	long ret = perf_evsel__intval(evsel, sample, "ret");
899 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
900 				 sample->time, ret);
901 }
902 
903 static int
904 process_enter_tx(struct timechart *tchart,
905 		 struct perf_evsel *evsel,
906 		 struct perf_sample *sample)
907 {
908 	long fd = perf_evsel__intval(evsel, sample, "fd");
909 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
910 				   sample->time, fd);
911 }
912 
913 static int
914 process_exit_tx(struct timechart *tchart,
915 		struct perf_evsel *evsel,
916 		struct perf_sample *sample)
917 {
918 	long ret = perf_evsel__intval(evsel, sample, "ret");
919 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
920 				 sample->time, ret);
921 }
922 
923 static int
924 process_enter_rx(struct timechart *tchart,
925 		 struct perf_evsel *evsel,
926 		 struct perf_sample *sample)
927 {
928 	long fd = perf_evsel__intval(evsel, sample, "fd");
929 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
930 				   sample->time, fd);
931 }
932 
933 static int
934 process_exit_rx(struct timechart *tchart,
935 		struct perf_evsel *evsel,
936 		struct perf_sample *sample)
937 {
938 	long ret = perf_evsel__intval(evsel, sample, "ret");
939 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
940 				 sample->time, ret);
941 }
942 
943 static int
944 process_enter_poll(struct timechart *tchart,
945 		   struct perf_evsel *evsel,
946 		   struct perf_sample *sample)
947 {
948 	long fd = perf_evsel__intval(evsel, sample, "fd");
949 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
950 				   sample->time, fd);
951 }
952 
953 static int
954 process_exit_poll(struct timechart *tchart,
955 		  struct perf_evsel *evsel,
956 		  struct perf_sample *sample)
957 {
958 	long ret = perf_evsel__intval(evsel, sample, "ret");
959 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
960 				 sample->time, ret);
961 }
962 
963 /*
964  * Sort the pid datastructure
965  */
966 static void sort_pids(struct timechart *tchart)
967 {
968 	struct per_pid *new_list, *p, *cursor, *prev;
969 	/* sort by ppid first, then by pid, lowest to highest */
970 
971 	new_list = NULL;
972 
973 	while (tchart->all_data) {
974 		p = tchart->all_data;
975 		tchart->all_data = p->next;
976 		p->next = NULL;
977 
978 		if (new_list == NULL) {
979 			new_list = p;
980 			p->next = NULL;
981 			continue;
982 		}
983 		prev = NULL;
984 		cursor = new_list;
985 		while (cursor) {
986 			if (cursor->ppid > p->ppid ||
987 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
988 				/* must insert before */
989 				if (prev) {
990 					p->next = prev->next;
991 					prev->next = p;
992 					cursor = NULL;
993 					continue;
994 				} else {
995 					p->next = new_list;
996 					new_list = p;
997 					cursor = NULL;
998 					continue;
999 				}
1000 			}
1001 
1002 			prev = cursor;
1003 			cursor = cursor->next;
1004 			if (!cursor)
1005 				prev->next = p;
1006 		}
1007 	}
1008 	tchart->all_data = new_list;
1009 }
1010 
1011 
1012 static void draw_c_p_states(struct timechart *tchart)
1013 {
1014 	struct power_event *pwr;
1015 	pwr = tchart->power_events;
1016 
1017 	/*
1018 	 * two pass drawing so that the P state bars are on top of the C state blocks
1019 	 */
1020 	while (pwr) {
1021 		if (pwr->type == CSTATE)
1022 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1023 		pwr = pwr->next;
1024 	}
1025 
1026 	pwr = tchart->power_events;
1027 	while (pwr) {
1028 		if (pwr->type == PSTATE) {
1029 			if (!pwr->state)
1030 				pwr->state = tchart->min_freq;
1031 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1032 		}
1033 		pwr = pwr->next;
1034 	}
1035 }
1036 
1037 static void draw_wakeups(struct timechart *tchart)
1038 {
1039 	struct wake_event *we;
1040 	struct per_pid *p;
1041 	struct per_pidcomm *c;
1042 
1043 	we = tchart->wake_events;
1044 	while (we) {
1045 		int from = 0, to = 0;
1046 		char *task_from = NULL, *task_to = NULL;
1047 
1048 		/* locate the column of the waker and wakee */
1049 		p = tchart->all_data;
1050 		while (p) {
1051 			if (p->pid == we->waker || p->pid == we->wakee) {
1052 				c = p->all;
1053 				while (c) {
1054 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1055 						if (p->pid == we->waker && !from) {
1056 							from = c->Y;
1057 							task_from = strdup(c->comm);
1058 						}
1059 						if (p->pid == we->wakee && !to) {
1060 							to = c->Y;
1061 							task_to = strdup(c->comm);
1062 						}
1063 					}
1064 					c = c->next;
1065 				}
1066 				c = p->all;
1067 				while (c) {
1068 					if (p->pid == we->waker && !from) {
1069 						from = c->Y;
1070 						task_from = strdup(c->comm);
1071 					}
1072 					if (p->pid == we->wakee && !to) {
1073 						to = c->Y;
1074 						task_to = strdup(c->comm);
1075 					}
1076 					c = c->next;
1077 				}
1078 			}
1079 			p = p->next;
1080 		}
1081 
1082 		if (!task_from) {
1083 			task_from = malloc(40);
1084 			sprintf(task_from, "[%i]", we->waker);
1085 		}
1086 		if (!task_to) {
1087 			task_to = malloc(40);
1088 			sprintf(task_to, "[%i]", we->wakee);
1089 		}
1090 
1091 		if (we->waker == -1)
1092 			svg_interrupt(we->time, to, we->backtrace);
1093 		else if (from && to && abs(from - to) == 1)
1094 			svg_wakeline(we->time, from, to, we->backtrace);
1095 		else
1096 			svg_partial_wakeline(we->time, from, task_from, to,
1097 					     task_to, we->backtrace);
1098 		we = we->next;
1099 
1100 		free(task_from);
1101 		free(task_to);
1102 	}
1103 }
1104 
1105 static void draw_cpu_usage(struct timechart *tchart)
1106 {
1107 	struct per_pid *p;
1108 	struct per_pidcomm *c;
1109 	struct cpu_sample *sample;
1110 	p = tchart->all_data;
1111 	while (p) {
1112 		c = p->all;
1113 		while (c) {
1114 			sample = c->samples;
1115 			while (sample) {
1116 				if (sample->type == TYPE_RUNNING) {
1117 					svg_process(sample->cpu,
1118 						    sample->start_time,
1119 						    sample->end_time,
1120 						    p->pid,
1121 						    c->comm,
1122 						    sample->backtrace);
1123 				}
1124 
1125 				sample = sample->next;
1126 			}
1127 			c = c->next;
1128 		}
1129 		p = p->next;
1130 	}
1131 }
1132 
1133 static void draw_io_bars(struct timechart *tchart)
1134 {
1135 	const char *suf;
1136 	double bytes;
1137 	char comm[256];
1138 	struct per_pid *p;
1139 	struct per_pidcomm *c;
1140 	struct io_sample *sample;
1141 	int Y = 1;
1142 
1143 	p = tchart->all_data;
1144 	while (p) {
1145 		c = p->all;
1146 		while (c) {
1147 			if (!c->display) {
1148 				c->Y = 0;
1149 				c = c->next;
1150 				continue;
1151 			}
1152 
1153 			svg_box(Y, c->start_time, c->end_time, "process3");
1154 			sample = c->io_samples;
1155 			for (sample = c->io_samples; sample; sample = sample->next) {
1156 				double h = (double)sample->bytes / c->max_bytes;
1157 
1158 				if (tchart->skip_eagain &&
1159 				    sample->err == -EAGAIN)
1160 					continue;
1161 
1162 				if (sample->err)
1163 					h = 1;
1164 
1165 				if (sample->type == IOTYPE_SYNC)
1166 					svg_fbox(Y,
1167 						sample->start_time,
1168 						sample->end_time,
1169 						1,
1170 						sample->err ? "error" : "sync",
1171 						sample->fd,
1172 						sample->err,
1173 						sample->merges);
1174 				else if (sample->type == IOTYPE_POLL)
1175 					svg_fbox(Y,
1176 						sample->start_time,
1177 						sample->end_time,
1178 						1,
1179 						sample->err ? "error" : "poll",
1180 						sample->fd,
1181 						sample->err,
1182 						sample->merges);
1183 				else if (sample->type == IOTYPE_READ)
1184 					svg_ubox(Y,
1185 						sample->start_time,
1186 						sample->end_time,
1187 						h,
1188 						sample->err ? "error" : "disk",
1189 						sample->fd,
1190 						sample->err,
1191 						sample->merges);
1192 				else if (sample->type == IOTYPE_WRITE)
1193 					svg_lbox(Y,
1194 						sample->start_time,
1195 						sample->end_time,
1196 						h,
1197 						sample->err ? "error" : "disk",
1198 						sample->fd,
1199 						sample->err,
1200 						sample->merges);
1201 				else if (sample->type == IOTYPE_RX)
1202 					svg_ubox(Y,
1203 						sample->start_time,
1204 						sample->end_time,
1205 						h,
1206 						sample->err ? "error" : "net",
1207 						sample->fd,
1208 						sample->err,
1209 						sample->merges);
1210 				else if (sample->type == IOTYPE_TX)
1211 					svg_lbox(Y,
1212 						sample->start_time,
1213 						sample->end_time,
1214 						h,
1215 						sample->err ? "error" : "net",
1216 						sample->fd,
1217 						sample->err,
1218 						sample->merges);
1219 			}
1220 
1221 			suf = "";
1222 			bytes = c->total_bytes;
1223 			if (bytes > 1024) {
1224 				bytes = bytes / 1024;
1225 				suf = "K";
1226 			}
1227 			if (bytes > 1024) {
1228 				bytes = bytes / 1024;
1229 				suf = "M";
1230 			}
1231 			if (bytes > 1024) {
1232 				bytes = bytes / 1024;
1233 				suf = "G";
1234 			}
1235 
1236 
1237 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1238 			svg_text(Y, c->start_time, comm);
1239 
1240 			c->Y = Y;
1241 			Y++;
1242 			c = c->next;
1243 		}
1244 		p = p->next;
1245 	}
1246 }
1247 
1248 static void draw_process_bars(struct timechart *tchart)
1249 {
1250 	struct per_pid *p;
1251 	struct per_pidcomm *c;
1252 	struct cpu_sample *sample;
1253 	int Y = 0;
1254 
1255 	Y = 2 * tchart->numcpus + 2;
1256 
1257 	p = tchart->all_data;
1258 	while (p) {
1259 		c = p->all;
1260 		while (c) {
1261 			if (!c->display) {
1262 				c->Y = 0;
1263 				c = c->next;
1264 				continue;
1265 			}
1266 
1267 			svg_box(Y, c->start_time, c->end_time, "process");
1268 			sample = c->samples;
1269 			while (sample) {
1270 				if (sample->type == TYPE_RUNNING)
1271 					svg_running(Y, sample->cpu,
1272 						    sample->start_time,
1273 						    sample->end_time,
1274 						    sample->backtrace);
1275 				if (sample->type == TYPE_BLOCKED)
1276 					svg_blocked(Y, sample->cpu,
1277 						    sample->start_time,
1278 						    sample->end_time,
1279 						    sample->backtrace);
1280 				if (sample->type == TYPE_WAITING)
1281 					svg_waiting(Y, sample->cpu,
1282 						    sample->start_time,
1283 						    sample->end_time,
1284 						    sample->backtrace);
1285 				sample = sample->next;
1286 			}
1287 
1288 			if (c->comm) {
1289 				char comm[256];
1290 				if (c->total_time > 5000000000) /* 5 seconds */
1291 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1292 				else
1293 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1294 
1295 				svg_text(Y, c->start_time, comm);
1296 			}
1297 			c->Y = Y;
1298 			Y++;
1299 			c = c->next;
1300 		}
1301 		p = p->next;
1302 	}
1303 }
1304 
1305 static void add_process_filter(const char *string)
1306 {
1307 	int pid = strtoull(string, NULL, 10);
1308 	struct process_filter *filt = malloc(sizeof(*filt));
1309 
1310 	if (!filt)
1311 		return;
1312 
1313 	filt->name = strdup(string);
1314 	filt->pid  = pid;
1315 	filt->next = process_filter;
1316 
1317 	process_filter = filt;
1318 }
1319 
1320 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1321 {
1322 	struct process_filter *filt;
1323 	if (!process_filter)
1324 		return 1;
1325 
1326 	filt = process_filter;
1327 	while (filt) {
1328 		if (filt->pid && p->pid == filt->pid)
1329 			return 1;
1330 		if (strcmp(filt->name, c->comm) == 0)
1331 			return 1;
1332 		filt = filt->next;
1333 	}
1334 	return 0;
1335 }
1336 
1337 static int determine_display_tasks_filtered(struct timechart *tchart)
1338 {
1339 	struct per_pid *p;
1340 	struct per_pidcomm *c;
1341 	int count = 0;
1342 
1343 	p = tchart->all_data;
1344 	while (p) {
1345 		p->display = 0;
1346 		if (p->start_time == 1)
1347 			p->start_time = tchart->first_time;
1348 
1349 		/* no exit marker, task kept running to the end */
1350 		if (p->end_time == 0)
1351 			p->end_time = tchart->last_time;
1352 
1353 		c = p->all;
1354 
1355 		while (c) {
1356 			c->display = 0;
1357 
1358 			if (c->start_time == 1)
1359 				c->start_time = tchart->first_time;
1360 
1361 			if (passes_filter(p, c)) {
1362 				c->display = 1;
1363 				p->display = 1;
1364 				count++;
1365 			}
1366 
1367 			if (c->end_time == 0)
1368 				c->end_time = tchart->last_time;
1369 
1370 			c = c->next;
1371 		}
1372 		p = p->next;
1373 	}
1374 	return count;
1375 }
1376 
1377 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1378 {
1379 	struct per_pid *p;
1380 	struct per_pidcomm *c;
1381 	int count = 0;
1382 
1383 	p = tchart->all_data;
1384 	while (p) {
1385 		p->display = 0;
1386 		if (p->start_time == 1)
1387 			p->start_time = tchart->first_time;
1388 
1389 		/* no exit marker, task kept running to the end */
1390 		if (p->end_time == 0)
1391 			p->end_time = tchart->last_time;
1392 		if (p->total_time >= threshold)
1393 			p->display = 1;
1394 
1395 		c = p->all;
1396 
1397 		while (c) {
1398 			c->display = 0;
1399 
1400 			if (c->start_time == 1)
1401 				c->start_time = tchart->first_time;
1402 
1403 			if (c->total_time >= threshold) {
1404 				c->display = 1;
1405 				count++;
1406 			}
1407 
1408 			if (c->end_time == 0)
1409 				c->end_time = tchart->last_time;
1410 
1411 			c = c->next;
1412 		}
1413 		p = p->next;
1414 	}
1415 	return count;
1416 }
1417 
1418 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1419 {
1420 	struct per_pid *p;
1421 	struct per_pidcomm *c;
1422 	int count = 0;
1423 
1424 	p = timechart->all_data;
1425 	while (p) {
1426 		/* no exit marker, task kept running to the end */
1427 		if (p->end_time == 0)
1428 			p->end_time = timechart->last_time;
1429 
1430 		c = p->all;
1431 
1432 		while (c) {
1433 			c->display = 0;
1434 
1435 			if (c->total_bytes >= threshold) {
1436 				c->display = 1;
1437 				count++;
1438 			}
1439 
1440 			if (c->end_time == 0)
1441 				c->end_time = timechart->last_time;
1442 
1443 			c = c->next;
1444 		}
1445 		p = p->next;
1446 	}
1447 	return count;
1448 }
1449 
1450 #define BYTES_THRESH (1 * 1024 * 1024)
1451 #define TIME_THRESH 10000000
1452 
1453 static void write_svg_file(struct timechart *tchart, const char *filename)
1454 {
1455 	u64 i;
1456 	int count;
1457 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1458 
1459 	if (tchart->power_only)
1460 		tchart->proc_num = 0;
1461 
1462 	/* We'd like to show at least proc_num tasks;
1463 	 * be less picky if we have fewer */
1464 	do {
1465 		if (process_filter)
1466 			count = determine_display_tasks_filtered(tchart);
1467 		else if (tchart->io_events)
1468 			count = determine_display_io_tasks(tchart, thresh);
1469 		else
1470 			count = determine_display_tasks(tchart, thresh);
1471 		thresh /= 10;
1472 	} while (!process_filter && thresh && count < tchart->proc_num);
1473 
1474 	if (!tchart->proc_num)
1475 		count = 0;
1476 
1477 	if (tchart->io_events) {
1478 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1479 
1480 		svg_time_grid(0.5);
1481 		svg_io_legenda();
1482 
1483 		draw_io_bars(tchart);
1484 	} else {
1485 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1486 
1487 		svg_time_grid(0);
1488 
1489 		svg_legenda();
1490 
1491 		for (i = 0; i < tchart->numcpus; i++)
1492 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1493 
1494 		draw_cpu_usage(tchart);
1495 		if (tchart->proc_num)
1496 			draw_process_bars(tchart);
1497 		if (!tchart->tasks_only)
1498 			draw_c_p_states(tchart);
1499 		if (tchart->proc_num)
1500 			draw_wakeups(tchart);
1501 	}
1502 
1503 	svg_close();
1504 }
1505 
1506 static int process_header(struct perf_file_section *section __maybe_unused,
1507 			  struct perf_header *ph,
1508 			  int feat,
1509 			  int fd __maybe_unused,
1510 			  void *data)
1511 {
1512 	struct timechart *tchart = data;
1513 
1514 	switch (feat) {
1515 	case HEADER_NRCPUS:
1516 		tchart->numcpus = ph->env.nr_cpus_avail;
1517 		break;
1518 
1519 	case HEADER_CPU_TOPOLOGY:
1520 		if (!tchart->topology)
1521 			break;
1522 
1523 		if (svg_build_topology_map(ph->env.sibling_cores,
1524 					   ph->env.nr_sibling_cores,
1525 					   ph->env.sibling_threads,
1526 					   ph->env.nr_sibling_threads))
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 perf_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, false,
1607 							 &tchart->tool);
1608 	int ret = -EINVAL;
1609 
1610 	if (session == NULL)
1611 		return -1;
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 && !strncmp(argv[0], "rec", 3)) {
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