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