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 "util/header.h"
28 #include <subcmd/pager.h>
29 #include <subcmd/parse-options.h>
30 #include "util/parse-events.h"
31 #include "util/event.h"
32 #include "util/session.h"
33 #include "util/svghelper.h"
34 #include "util/tool.h"
35 #include "util/data.h"
36 #include "util/debug.h"
37 #include "util/string2.h"
38 #include "util/tracepoint.h"
39 #include "util/util.h"
40 #include <linux/err.h>
41 #include <traceevent/event-parse.h>
42 
43 #ifdef LACKS_OPEN_MEMSTREAM_PROTOTYPE
44 FILE *open_memstream(char **ptr, size_t *sizeloc);
45 #endif
46 
47 #define SUPPORT_OLD_POWER_EVENTS 1
48 #define PWR_EVENT_EXIT -1
49 
50 struct per_pid;
51 struct power_event;
52 struct wake_event;
53 
54 struct timechart {
55 	struct perf_tool	tool;
56 	struct per_pid		*all_data;
57 	struct power_event	*power_events;
58 	struct wake_event	*wake_events;
59 	int			proc_num;
60 	unsigned int		numcpus;
61 	u64			min_freq,	/* Lowest CPU frequency seen */
62 				max_freq,	/* Highest CPU frequency seen */
63 				turbo_frequency,
64 				first_time, last_time;
65 	bool			power_only,
66 				tasks_only,
67 				with_backtrace,
68 				topology;
69 	bool			force;
70 	/* IO related settings */
71 	bool			io_only,
72 				skip_eagain;
73 	u64			io_events;
74 	u64			min_time,
75 				merge_dist;
76 };
77 
78 struct per_pidcomm;
79 struct cpu_sample;
80 struct io_sample;
81 
82 /*
83  * Datastructure layout:
84  * We keep an list of "pid"s, matching the kernels notion of a task struct.
85  * Each "pid" entry, has a list of "comm"s.
86  *	this is because we want to track different programs different, while
87  *	exec will reuse the original pid (by design).
88  * Each comm has a list of samples that will be used to draw
89  * final graph.
90  */
91 
92 struct per_pid {
93 	struct per_pid *next;
94 
95 	int		pid;
96 	int		ppid;
97 
98 	u64		start_time;
99 	u64		end_time;
100 	u64		total_time;
101 	u64		total_bytes;
102 	int		display;
103 
104 	struct per_pidcomm *all;
105 	struct per_pidcomm *current;
106 };
107 
108 
109 struct per_pidcomm {
110 	struct per_pidcomm *next;
111 
112 	u64		start_time;
113 	u64		end_time;
114 	u64		total_time;
115 	u64		max_bytes;
116 	u64		total_bytes;
117 
118 	int		Y;
119 	int		display;
120 
121 	long		state;
122 	u64		state_since;
123 
124 	char		*comm;
125 
126 	struct cpu_sample *samples;
127 	struct io_sample  *io_samples;
128 };
129 
130 struct sample_wrapper {
131 	struct sample_wrapper *next;
132 
133 	u64		timestamp;
134 	unsigned char	data[];
135 };
136 
137 #define TYPE_NONE	0
138 #define TYPE_RUNNING	1
139 #define TYPE_WAITING	2
140 #define TYPE_BLOCKED	3
141 
142 struct cpu_sample {
143 	struct cpu_sample *next;
144 
145 	u64 start_time;
146 	u64 end_time;
147 	int type;
148 	int cpu;
149 	const char *backtrace;
150 };
151 
152 enum {
153 	IOTYPE_READ,
154 	IOTYPE_WRITE,
155 	IOTYPE_SYNC,
156 	IOTYPE_TX,
157 	IOTYPE_RX,
158 	IOTYPE_POLL,
159 };
160 
161 struct io_sample {
162 	struct io_sample *next;
163 
164 	u64 start_time;
165 	u64 end_time;
166 	u64 bytes;
167 	int type;
168 	int fd;
169 	int err;
170 	int merges;
171 };
172 
173 #define CSTATE 1
174 #define PSTATE 2
175 
176 struct power_event {
177 	struct power_event *next;
178 	int type;
179 	int state;
180 	u64 start_time;
181 	u64 end_time;
182 	int cpu;
183 };
184 
185 struct wake_event {
186 	struct wake_event *next;
187 	int waker;
188 	int wakee;
189 	u64 time;
190 	const char *backtrace;
191 };
192 
193 struct process_filter {
194 	char			*name;
195 	int			pid;
196 	struct process_filter	*next;
197 };
198 
199 static struct process_filter *process_filter;
200 
201 
202 static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
203 {
204 	struct per_pid *cursor = tchart->all_data;
205 
206 	while (cursor) {
207 		if (cursor->pid == pid)
208 			return cursor;
209 		cursor = cursor->next;
210 	}
211 	cursor = zalloc(sizeof(*cursor));
212 	assert(cursor != NULL);
213 	cursor->pid = pid;
214 	cursor->next = tchart->all_data;
215 	tchart->all_data = cursor;
216 	return cursor;
217 }
218 
219 static struct per_pidcomm *create_pidcomm(struct per_pid *p)
220 {
221 	struct per_pidcomm *c;
222 
223 	c = zalloc(sizeof(*c));
224 	if (!c)
225 		return NULL;
226 	p->current = c;
227 	c->next = p->all;
228 	p->all = c;
229 	return c;
230 }
231 
232 static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
233 {
234 	struct per_pid *p;
235 	struct per_pidcomm *c;
236 	p = find_create_pid(tchart, pid);
237 	c = p->all;
238 	while (c) {
239 		if (c->comm && strcmp(c->comm, comm) == 0) {
240 			p->current = c;
241 			return;
242 		}
243 		if (!c->comm) {
244 			c->comm = strdup(comm);
245 			p->current = c;
246 			return;
247 		}
248 		c = c->next;
249 	}
250 	c = create_pidcomm(p);
251 	assert(c != NULL);
252 	c->comm = strdup(comm);
253 }
254 
255 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
256 {
257 	struct per_pid *p, *pp;
258 	p = find_create_pid(tchart, pid);
259 	pp = find_create_pid(tchart, ppid);
260 	p->ppid = ppid;
261 	if (pp->current && pp->current->comm && !p->current)
262 		pid_set_comm(tchart, pid, pp->current->comm);
263 
264 	p->start_time = timestamp;
265 	if (p->current && !p->current->start_time) {
266 		p->current->start_time = timestamp;
267 		p->current->state_since = timestamp;
268 	}
269 }
270 
271 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
272 {
273 	struct per_pid *p;
274 	p = find_create_pid(tchart, pid);
275 	p->end_time = timestamp;
276 	if (p->current)
277 		p->current->end_time = timestamp;
278 }
279 
280 static void pid_put_sample(struct timechart *tchart, int pid, int type,
281 			   unsigned int cpu, u64 start, u64 end,
282 			   const char *backtrace)
283 {
284 	struct per_pid *p;
285 	struct per_pidcomm *c;
286 	struct cpu_sample *sample;
287 
288 	p = find_create_pid(tchart, pid);
289 	c = p->current;
290 	if (!c) {
291 		c = create_pidcomm(p);
292 		assert(c != NULL);
293 	}
294 
295 	sample = zalloc(sizeof(*sample));
296 	assert(sample != NULL);
297 	sample->start_time = start;
298 	sample->end_time = end;
299 	sample->type = type;
300 	sample->next = c->samples;
301 	sample->cpu = cpu;
302 	sample->backtrace = backtrace;
303 	c->samples = sample;
304 
305 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
306 		c->total_time += (end-start);
307 		p->total_time += (end-start);
308 	}
309 
310 	if (c->start_time == 0 || c->start_time > start)
311 		c->start_time = start;
312 	if (p->start_time == 0 || p->start_time > start)
313 		p->start_time = start;
314 }
315 
316 #define MAX_CPUS 4096
317 
318 static u64 *cpus_cstate_start_times;
319 static int *cpus_cstate_state;
320 static u64 *cpus_pstate_start_times;
321 static u64 *cpus_pstate_state;
322 
323 static int process_comm_event(struct perf_tool *tool,
324 			      union perf_event *event,
325 			      struct perf_sample *sample __maybe_unused,
326 			      struct machine *machine __maybe_unused)
327 {
328 	struct timechart *tchart = container_of(tool, struct timechart, tool);
329 	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
330 	return 0;
331 }
332 
333 static int process_fork_event(struct perf_tool *tool,
334 			      union perf_event *event,
335 			      struct perf_sample *sample __maybe_unused,
336 			      struct machine *machine __maybe_unused)
337 {
338 	struct timechart *tchart = container_of(tool, struct timechart, tool);
339 	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
340 	return 0;
341 }
342 
343 static int process_exit_event(struct perf_tool *tool,
344 			      union perf_event *event,
345 			      struct perf_sample *sample __maybe_unused,
346 			      struct machine *machine __maybe_unused)
347 {
348 	struct timechart *tchart = container_of(tool, struct timechart, tool);
349 	pid_exit(tchart, event->fork.pid, event->fork.time);
350 	return 0;
351 }
352 
353 #ifdef SUPPORT_OLD_POWER_EVENTS
354 static int use_old_power_events;
355 #endif
356 
357 static void c_state_start(int cpu, u64 timestamp, int state)
358 {
359 	cpus_cstate_start_times[cpu] = timestamp;
360 	cpus_cstate_state[cpu] = state;
361 }
362 
363 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
364 {
365 	struct power_event *pwr = zalloc(sizeof(*pwr));
366 
367 	if (!pwr)
368 		return;
369 
370 	pwr->state = cpus_cstate_state[cpu];
371 	pwr->start_time = cpus_cstate_start_times[cpu];
372 	pwr->end_time = timestamp;
373 	pwr->cpu = cpu;
374 	pwr->type = CSTATE;
375 	pwr->next = tchart->power_events;
376 
377 	tchart->power_events = pwr;
378 }
379 
380 static struct power_event *p_state_end(struct timechart *tchart, int cpu,
381 					u64 timestamp)
382 {
383 	struct power_event *pwr = zalloc(sizeof(*pwr));
384 
385 	if (!pwr)
386 		return NULL;
387 
388 	pwr->state = cpus_pstate_state[cpu];
389 	pwr->start_time = cpus_pstate_start_times[cpu];
390 	pwr->end_time = timestamp;
391 	pwr->cpu = cpu;
392 	pwr->type = PSTATE;
393 	pwr->next = tchart->power_events;
394 	if (!pwr->start_time)
395 		pwr->start_time = tchart->first_time;
396 
397 	tchart->power_events = pwr;
398 	return pwr;
399 }
400 
401 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
402 {
403 	struct power_event *pwr;
404 
405 	if (new_freq > 8000000) /* detect invalid data */
406 		return;
407 
408 	pwr = p_state_end(tchart, cpu, timestamp);
409 	if (!pwr)
410 		return;
411 
412 	cpus_pstate_state[cpu] = new_freq;
413 	cpus_pstate_start_times[cpu] = timestamp;
414 
415 	if ((u64)new_freq > tchart->max_freq)
416 		tchart->max_freq = new_freq;
417 
418 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
419 		tchart->min_freq = new_freq;
420 
421 	if (new_freq == tchart->max_freq - 1000)
422 		tchart->turbo_frequency = tchart->max_freq;
423 }
424 
425 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
426 			 int waker, int wakee, u8 flags, const char *backtrace)
427 {
428 	struct per_pid *p;
429 	struct wake_event *we = zalloc(sizeof(*we));
430 
431 	if (!we)
432 		return;
433 
434 	we->time = timestamp;
435 	we->waker = waker;
436 	we->backtrace = backtrace;
437 
438 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
439 		we->waker = -1;
440 
441 	we->wakee = wakee;
442 	we->next = tchart->wake_events;
443 	tchart->wake_events = we;
444 	p = find_create_pid(tchart, we->wakee);
445 
446 	if (p && p->current && p->current->state == TYPE_NONE) {
447 		p->current->state_since = timestamp;
448 		p->current->state = TYPE_WAITING;
449 	}
450 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
451 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
452 			       p->current->state_since, timestamp, NULL);
453 		p->current->state_since = timestamp;
454 		p->current->state = TYPE_WAITING;
455 	}
456 }
457 
458 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
459 			 int prev_pid, int next_pid, u64 prev_state,
460 			 const char *backtrace)
461 {
462 	struct per_pid *p = NULL, *prev_p;
463 
464 	prev_p = find_create_pid(tchart, prev_pid);
465 
466 	p = find_create_pid(tchart, next_pid);
467 
468 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
469 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
470 			       prev_p->current->state_since, timestamp,
471 			       backtrace);
472 	if (p && p->current) {
473 		if (p->current->state != TYPE_NONE)
474 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
475 				       p->current->state_since, timestamp,
476 				       backtrace);
477 
478 		p->current->state_since = timestamp;
479 		p->current->state = TYPE_RUNNING;
480 	}
481 
482 	if (prev_p->current) {
483 		prev_p->current->state = TYPE_NONE;
484 		prev_p->current->state_since = timestamp;
485 		if (prev_state & 2)
486 			prev_p->current->state = TYPE_BLOCKED;
487 		if (prev_state == 0)
488 			prev_p->current->state = TYPE_WAITING;
489 	}
490 }
491 
492 static const char *cat_backtrace(union perf_event *event,
493 				 struct perf_sample *sample,
494 				 struct machine *machine)
495 {
496 	struct addr_location al;
497 	unsigned int i;
498 	char *p = NULL;
499 	size_t p_len;
500 	u8 cpumode = PERF_RECORD_MISC_USER;
501 	struct ip_callchain *chain = sample->callchain;
502 	FILE *f = open_memstream(&p, &p_len);
503 
504 	if (!f) {
505 		perror("open_memstream error");
506 		return NULL;
507 	}
508 
509 	addr_location__init(&al);
510 	if (!chain)
511 		goto exit;
512 
513 	if (machine__resolve(machine, &al, sample) < 0) {
514 		fprintf(stderr, "problem processing %d event, skipping it.\n",
515 			event->header.type);
516 		goto exit;
517 	}
518 
519 	for (i = 0; i < chain->nr; i++) {
520 		u64 ip;
521 		struct addr_location tal;
522 
523 		if (callchain_param.order == ORDER_CALLEE)
524 			ip = chain->ips[i];
525 		else
526 			ip = chain->ips[chain->nr - i - 1];
527 
528 		if (ip >= PERF_CONTEXT_MAX) {
529 			switch (ip) {
530 			case PERF_CONTEXT_HV:
531 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
532 				break;
533 			case PERF_CONTEXT_KERNEL:
534 				cpumode = PERF_RECORD_MISC_KERNEL;
535 				break;
536 			case PERF_CONTEXT_USER:
537 				cpumode = PERF_RECORD_MISC_USER;
538 				break;
539 			default:
540 				pr_debug("invalid callchain context: "
541 					 "%"PRId64"\n", (s64) ip);
542 
543 				/*
544 				 * It seems the callchain is corrupted.
545 				 * Discard all.
546 				 */
547 				zfree(&p);
548 				goto exit;
549 			}
550 			continue;
551 		}
552 
553 		addr_location__init(&tal);
554 		tal.filtered = 0;
555 		if (thread__find_symbol(al.thread, cpumode, ip, &tal))
556 			fprintf(f, "..... %016" PRIx64 " %s\n", ip, tal.sym->name);
557 		else
558 			fprintf(f, "..... %016" PRIx64 "\n", ip);
559 
560 		addr_location__exit(&tal);
561 	}
562 exit:
563 	addr_location__exit(&al);
564 	fclose(f);
565 
566 	return p;
567 }
568 
569 typedef int (*tracepoint_handler)(struct timechart *tchart,
570 				  struct evsel *evsel,
571 				  struct perf_sample *sample,
572 				  const char *backtrace);
573 
574 static int process_sample_event(struct perf_tool *tool,
575 				union perf_event *event,
576 				struct perf_sample *sample,
577 				struct evsel *evsel,
578 				struct machine *machine)
579 {
580 	struct timechart *tchart = container_of(tool, struct timechart, tool);
581 
582 	if (evsel->core.attr.sample_type & PERF_SAMPLE_TIME) {
583 		if (!tchart->first_time || tchart->first_time > sample->time)
584 			tchart->first_time = sample->time;
585 		if (tchart->last_time < sample->time)
586 			tchart->last_time = sample->time;
587 	}
588 
589 	if (evsel->handler != NULL) {
590 		tracepoint_handler f = evsel->handler;
591 		return f(tchart, evsel, sample,
592 			 cat_backtrace(event, sample, machine));
593 	}
594 
595 	return 0;
596 }
597 
598 static int
599 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
600 			struct evsel *evsel,
601 			struct perf_sample *sample,
602 			const char *backtrace __maybe_unused)
603 {
604 	u32 state  = evsel__intval(evsel, sample, "state");
605 	u32 cpu_id = evsel__intval(evsel, sample, "cpu_id");
606 
607 	if (state == (u32)PWR_EVENT_EXIT)
608 		c_state_end(tchart, cpu_id, sample->time);
609 	else
610 		c_state_start(cpu_id, sample->time, state);
611 	return 0;
612 }
613 
614 static int
615 process_sample_cpu_frequency(struct timechart *tchart,
616 			     struct evsel *evsel,
617 			     struct perf_sample *sample,
618 			     const char *backtrace __maybe_unused)
619 {
620 	u32 state  = evsel__intval(evsel, sample, "state");
621 	u32 cpu_id = evsel__intval(evsel, sample, "cpu_id");
622 
623 	p_state_change(tchart, cpu_id, sample->time, state);
624 	return 0;
625 }
626 
627 static int
628 process_sample_sched_wakeup(struct timechart *tchart,
629 			    struct evsel *evsel,
630 			    struct perf_sample *sample,
631 			    const char *backtrace)
632 {
633 	u8 flags  = evsel__intval(evsel, sample, "common_flags");
634 	int waker = evsel__intval(evsel, sample, "common_pid");
635 	int wakee = evsel__intval(evsel, sample, "pid");
636 
637 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
638 	return 0;
639 }
640 
641 static int
642 process_sample_sched_switch(struct timechart *tchart,
643 			    struct evsel *evsel,
644 			    struct perf_sample *sample,
645 			    const char *backtrace)
646 {
647 	int prev_pid   = evsel__intval(evsel, sample, "prev_pid");
648 	int next_pid   = evsel__intval(evsel, sample, "next_pid");
649 	u64 prev_state = evsel__intval(evsel, sample, "prev_state");
650 
651 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
652 		     prev_state, backtrace);
653 	return 0;
654 }
655 
656 #ifdef SUPPORT_OLD_POWER_EVENTS
657 static int
658 process_sample_power_start(struct timechart *tchart __maybe_unused,
659 			   struct evsel *evsel,
660 			   struct perf_sample *sample,
661 			   const char *backtrace __maybe_unused)
662 {
663 	u64 cpu_id = evsel__intval(evsel, sample, "cpu_id");
664 	u64 value  = evsel__intval(evsel, sample, "value");
665 
666 	c_state_start(cpu_id, sample->time, value);
667 	return 0;
668 }
669 
670 static int
671 process_sample_power_end(struct timechart *tchart,
672 			 struct evsel *evsel __maybe_unused,
673 			 struct perf_sample *sample,
674 			 const char *backtrace __maybe_unused)
675 {
676 	c_state_end(tchart, sample->cpu, sample->time);
677 	return 0;
678 }
679 
680 static int
681 process_sample_power_frequency(struct timechart *tchart,
682 			       struct evsel *evsel,
683 			       struct perf_sample *sample,
684 			       const char *backtrace __maybe_unused)
685 {
686 	u64 cpu_id = evsel__intval(evsel, sample, "cpu_id");
687 	u64 value  = evsel__intval(evsel, sample, "value");
688 
689 	p_state_change(tchart, cpu_id, sample->time, value);
690 	return 0;
691 }
692 #endif /* SUPPORT_OLD_POWER_EVENTS */
693 
694 /*
695  * After the last sample we need to wrap up the current C/P state
696  * and close out each CPU for these.
697  */
698 static void end_sample_processing(struct timechart *tchart)
699 {
700 	u64 cpu;
701 	struct power_event *pwr;
702 
703 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
704 		/* C state */
705 #if 0
706 		pwr = zalloc(sizeof(*pwr));
707 		if (!pwr)
708 			return;
709 
710 		pwr->state = cpus_cstate_state[cpu];
711 		pwr->start_time = cpus_cstate_start_times[cpu];
712 		pwr->end_time = tchart->last_time;
713 		pwr->cpu = cpu;
714 		pwr->type = CSTATE;
715 		pwr->next = tchart->power_events;
716 
717 		tchart->power_events = pwr;
718 #endif
719 		/* P state */
720 
721 		pwr = p_state_end(tchart, cpu, tchart->last_time);
722 		if (!pwr)
723 			return;
724 
725 		if (!pwr->state)
726 			pwr->state = tchart->min_freq;
727 	}
728 }
729 
730 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
731 			       u64 start, int fd)
732 {
733 	struct per_pid *p = find_create_pid(tchart, pid);
734 	struct per_pidcomm *c = p->current;
735 	struct io_sample *sample;
736 	struct io_sample *prev;
737 
738 	if (!c) {
739 		c = create_pidcomm(p);
740 		if (!c)
741 			return -ENOMEM;
742 	}
743 
744 	prev = c->io_samples;
745 
746 	if (prev && prev->start_time && !prev->end_time) {
747 		pr_warning("Skip invalid start event: "
748 			   "previous event already started!\n");
749 
750 		/* remove previous event that has been started,
751 		 * we are not sure we will ever get an end for it */
752 		c->io_samples = prev->next;
753 		free(prev);
754 		return 0;
755 	}
756 
757 	sample = zalloc(sizeof(*sample));
758 	if (!sample)
759 		return -ENOMEM;
760 	sample->start_time = start;
761 	sample->type = type;
762 	sample->fd = fd;
763 	sample->next = c->io_samples;
764 	c->io_samples = sample;
765 
766 	if (c->start_time == 0 || c->start_time > start)
767 		c->start_time = start;
768 
769 	return 0;
770 }
771 
772 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
773 			     u64 end, long ret)
774 {
775 	struct per_pid *p = find_create_pid(tchart, pid);
776 	struct per_pidcomm *c = p->current;
777 	struct io_sample *sample, *prev;
778 
779 	if (!c) {
780 		pr_warning("Invalid pidcomm!\n");
781 		return -1;
782 	}
783 
784 	sample = c->io_samples;
785 
786 	if (!sample) /* skip partially captured events */
787 		return 0;
788 
789 	if (sample->end_time) {
790 		pr_warning("Skip invalid end event: "
791 			   "previous event already ended!\n");
792 		return 0;
793 	}
794 
795 	if (sample->type != type) {
796 		pr_warning("Skip invalid end event: invalid event type!\n");
797 		return 0;
798 	}
799 
800 	sample->end_time = end;
801 	prev = sample->next;
802 
803 	/* we want to be able to see small and fast transfers, so make them
804 	 * at least min_time long, but don't overlap them */
805 	if (sample->end_time - sample->start_time < tchart->min_time)
806 		sample->end_time = sample->start_time + tchart->min_time;
807 	if (prev && sample->start_time < prev->end_time) {
808 		if (prev->err) /* try to make errors more visible */
809 			sample->start_time = prev->end_time;
810 		else
811 			prev->end_time = sample->start_time;
812 	}
813 
814 	if (ret < 0) {
815 		sample->err = ret;
816 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
817 		   type == IOTYPE_TX || type == IOTYPE_RX) {
818 
819 		if ((u64)ret > c->max_bytes)
820 			c->max_bytes = ret;
821 
822 		c->total_bytes += ret;
823 		p->total_bytes += ret;
824 		sample->bytes = ret;
825 	}
826 
827 	/* merge two requests to make svg smaller and render-friendly */
828 	if (prev &&
829 	    prev->type == sample->type &&
830 	    prev->err == sample->err &&
831 	    prev->fd == sample->fd &&
832 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
833 
834 		sample->bytes += prev->bytes;
835 		sample->merges += prev->merges + 1;
836 
837 		sample->start_time = prev->start_time;
838 		sample->next = prev->next;
839 		free(prev);
840 
841 		if (!sample->err && sample->bytes > c->max_bytes)
842 			c->max_bytes = sample->bytes;
843 	}
844 
845 	tchart->io_events++;
846 
847 	return 0;
848 }
849 
850 static int
851 process_enter_read(struct timechart *tchart,
852 		   struct evsel *evsel,
853 		   struct perf_sample *sample)
854 {
855 	long fd = evsel__intval(evsel, sample, "fd");
856 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
857 				   sample->time, fd);
858 }
859 
860 static int
861 process_exit_read(struct timechart *tchart,
862 		  struct evsel *evsel,
863 		  struct perf_sample *sample)
864 {
865 	long ret = evsel__intval(evsel, sample, "ret");
866 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
867 				 sample->time, ret);
868 }
869 
870 static int
871 process_enter_write(struct timechart *tchart,
872 		    struct evsel *evsel,
873 		    struct perf_sample *sample)
874 {
875 	long fd = evsel__intval(evsel, sample, "fd");
876 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
877 				   sample->time, fd);
878 }
879 
880 static int
881 process_exit_write(struct timechart *tchart,
882 		   struct evsel *evsel,
883 		   struct perf_sample *sample)
884 {
885 	long ret = evsel__intval(evsel, sample, "ret");
886 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
887 				 sample->time, ret);
888 }
889 
890 static int
891 process_enter_sync(struct timechart *tchart,
892 		   struct evsel *evsel,
893 		   struct perf_sample *sample)
894 {
895 	long fd = evsel__intval(evsel, sample, "fd");
896 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
897 				   sample->time, fd);
898 }
899 
900 static int
901 process_exit_sync(struct timechart *tchart,
902 		  struct evsel *evsel,
903 		  struct perf_sample *sample)
904 {
905 	long ret = evsel__intval(evsel, sample, "ret");
906 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
907 				 sample->time, ret);
908 }
909 
910 static int
911 process_enter_tx(struct timechart *tchart,
912 		 struct evsel *evsel,
913 		 struct perf_sample *sample)
914 {
915 	long fd = evsel__intval(evsel, sample, "fd");
916 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
917 				   sample->time, fd);
918 }
919 
920 static int
921 process_exit_tx(struct timechart *tchart,
922 		struct evsel *evsel,
923 		struct perf_sample *sample)
924 {
925 	long ret = evsel__intval(evsel, sample, "ret");
926 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
927 				 sample->time, ret);
928 }
929 
930 static int
931 process_enter_rx(struct timechart *tchart,
932 		 struct evsel *evsel,
933 		 struct perf_sample *sample)
934 {
935 	long fd = evsel__intval(evsel, sample, "fd");
936 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
937 				   sample->time, fd);
938 }
939 
940 static int
941 process_exit_rx(struct timechart *tchart,
942 		struct evsel *evsel,
943 		struct perf_sample *sample)
944 {
945 	long ret = evsel__intval(evsel, sample, "ret");
946 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
947 				 sample->time, ret);
948 }
949 
950 static int
951 process_enter_poll(struct timechart *tchart,
952 		   struct evsel *evsel,
953 		   struct perf_sample *sample)
954 {
955 	long fd = evsel__intval(evsel, sample, "fd");
956 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
957 				   sample->time, fd);
958 }
959 
960 static int
961 process_exit_poll(struct timechart *tchart,
962 		  struct evsel *evsel,
963 		  struct perf_sample *sample)
964 {
965 	long ret = evsel__intval(evsel, sample, "ret");
966 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
967 				 sample->time, ret);
968 }
969 
970 /*
971  * Sort the pid datastructure
972  */
973 static void sort_pids(struct timechart *tchart)
974 {
975 	struct per_pid *new_list, *p, *cursor, *prev;
976 	/* sort by ppid first, then by pid, lowest to highest */
977 
978 	new_list = NULL;
979 
980 	while (tchart->all_data) {
981 		p = tchart->all_data;
982 		tchart->all_data = p->next;
983 		p->next = NULL;
984 
985 		if (new_list == NULL) {
986 			new_list = p;
987 			p->next = NULL;
988 			continue;
989 		}
990 		prev = NULL;
991 		cursor = new_list;
992 		while (cursor) {
993 			if (cursor->ppid > p->ppid ||
994 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
995 				/* must insert before */
996 				if (prev) {
997 					p->next = prev->next;
998 					prev->next = p;
999 					cursor = NULL;
1000 					continue;
1001 				} else {
1002 					p->next = new_list;
1003 					new_list = p;
1004 					cursor = NULL;
1005 					continue;
1006 				}
1007 			}
1008 
1009 			prev = cursor;
1010 			cursor = cursor->next;
1011 			if (!cursor)
1012 				prev->next = p;
1013 		}
1014 	}
1015 	tchart->all_data = new_list;
1016 }
1017 
1018 
1019 static void draw_c_p_states(struct timechart *tchart)
1020 {
1021 	struct power_event *pwr;
1022 	pwr = tchart->power_events;
1023 
1024 	/*
1025 	 * two pass drawing so that the P state bars are on top of the C state blocks
1026 	 */
1027 	while (pwr) {
1028 		if (pwr->type == CSTATE)
1029 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1030 		pwr = pwr->next;
1031 	}
1032 
1033 	pwr = tchart->power_events;
1034 	while (pwr) {
1035 		if (pwr->type == PSTATE) {
1036 			if (!pwr->state)
1037 				pwr->state = tchart->min_freq;
1038 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1039 		}
1040 		pwr = pwr->next;
1041 	}
1042 }
1043 
1044 static void draw_wakeups(struct timechart *tchart)
1045 {
1046 	struct wake_event *we;
1047 	struct per_pid *p;
1048 	struct per_pidcomm *c;
1049 
1050 	we = tchart->wake_events;
1051 	while (we) {
1052 		int from = 0, to = 0;
1053 		char *task_from = NULL, *task_to = NULL;
1054 
1055 		/* locate the column of the waker and wakee */
1056 		p = tchart->all_data;
1057 		while (p) {
1058 			if (p->pid == we->waker || p->pid == we->wakee) {
1059 				c = p->all;
1060 				while (c) {
1061 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1062 						if (p->pid == we->waker && !from) {
1063 							from = c->Y;
1064 							task_from = strdup(c->comm);
1065 						}
1066 						if (p->pid == we->wakee && !to) {
1067 							to = c->Y;
1068 							task_to = strdup(c->comm);
1069 						}
1070 					}
1071 					c = c->next;
1072 				}
1073 				c = p->all;
1074 				while (c) {
1075 					if (p->pid == we->waker && !from) {
1076 						from = c->Y;
1077 						task_from = strdup(c->comm);
1078 					}
1079 					if (p->pid == we->wakee && !to) {
1080 						to = c->Y;
1081 						task_to = strdup(c->comm);
1082 					}
1083 					c = c->next;
1084 				}
1085 			}
1086 			p = p->next;
1087 		}
1088 
1089 		if (!task_from) {
1090 			task_from = malloc(40);
1091 			sprintf(task_from, "[%i]", we->waker);
1092 		}
1093 		if (!task_to) {
1094 			task_to = malloc(40);
1095 			sprintf(task_to, "[%i]", we->wakee);
1096 		}
1097 
1098 		if (we->waker == -1)
1099 			svg_interrupt(we->time, to, we->backtrace);
1100 		else if (from && to && abs(from - to) == 1)
1101 			svg_wakeline(we->time, from, to, we->backtrace);
1102 		else
1103 			svg_partial_wakeline(we->time, from, task_from, to,
1104 					     task_to, we->backtrace);
1105 		we = we->next;
1106 
1107 		free(task_from);
1108 		free(task_to);
1109 	}
1110 }
1111 
1112 static void draw_cpu_usage(struct timechart *tchart)
1113 {
1114 	struct per_pid *p;
1115 	struct per_pidcomm *c;
1116 	struct cpu_sample *sample;
1117 	p = tchart->all_data;
1118 	while (p) {
1119 		c = p->all;
1120 		while (c) {
1121 			sample = c->samples;
1122 			while (sample) {
1123 				if (sample->type == TYPE_RUNNING) {
1124 					svg_process(sample->cpu,
1125 						    sample->start_time,
1126 						    sample->end_time,
1127 						    p->pid,
1128 						    c->comm,
1129 						    sample->backtrace);
1130 				}
1131 
1132 				sample = sample->next;
1133 			}
1134 			c = c->next;
1135 		}
1136 		p = p->next;
1137 	}
1138 }
1139 
1140 static void draw_io_bars(struct timechart *tchart)
1141 {
1142 	const char *suf;
1143 	double bytes;
1144 	char comm[256];
1145 	struct per_pid *p;
1146 	struct per_pidcomm *c;
1147 	struct io_sample *sample;
1148 	int Y = 1;
1149 
1150 	p = tchart->all_data;
1151 	while (p) {
1152 		c = p->all;
1153 		while (c) {
1154 			if (!c->display) {
1155 				c->Y = 0;
1156 				c = c->next;
1157 				continue;
1158 			}
1159 
1160 			svg_box(Y, c->start_time, c->end_time, "process3");
1161 			sample = c->io_samples;
1162 			for (sample = c->io_samples; sample; sample = sample->next) {
1163 				double h = (double)sample->bytes / c->max_bytes;
1164 
1165 				if (tchart->skip_eagain &&
1166 				    sample->err == -EAGAIN)
1167 					continue;
1168 
1169 				if (sample->err)
1170 					h = 1;
1171 
1172 				if (sample->type == IOTYPE_SYNC)
1173 					svg_fbox(Y,
1174 						sample->start_time,
1175 						sample->end_time,
1176 						1,
1177 						sample->err ? "error" : "sync",
1178 						sample->fd,
1179 						sample->err,
1180 						sample->merges);
1181 				else if (sample->type == IOTYPE_POLL)
1182 					svg_fbox(Y,
1183 						sample->start_time,
1184 						sample->end_time,
1185 						1,
1186 						sample->err ? "error" : "poll",
1187 						sample->fd,
1188 						sample->err,
1189 						sample->merges);
1190 				else if (sample->type == IOTYPE_READ)
1191 					svg_ubox(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_WRITE)
1200 					svg_lbox(Y,
1201 						sample->start_time,
1202 						sample->end_time,
1203 						h,
1204 						sample->err ? "error" : "disk",
1205 						sample->fd,
1206 						sample->err,
1207 						sample->merges);
1208 				else if (sample->type == IOTYPE_RX)
1209 					svg_ubox(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 				else if (sample->type == IOTYPE_TX)
1218 					svg_lbox(Y,
1219 						sample->start_time,
1220 						sample->end_time,
1221 						h,
1222 						sample->err ? "error" : "net",
1223 						sample->fd,
1224 						sample->err,
1225 						sample->merges);
1226 			}
1227 
1228 			suf = "";
1229 			bytes = c->total_bytes;
1230 			if (bytes > 1024) {
1231 				bytes = bytes / 1024;
1232 				suf = "K";
1233 			}
1234 			if (bytes > 1024) {
1235 				bytes = bytes / 1024;
1236 				suf = "M";
1237 			}
1238 			if (bytes > 1024) {
1239 				bytes = bytes / 1024;
1240 				suf = "G";
1241 			}
1242 
1243 
1244 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1245 			svg_text(Y, c->start_time, comm);
1246 
1247 			c->Y = Y;
1248 			Y++;
1249 			c = c->next;
1250 		}
1251 		p = p->next;
1252 	}
1253 }
1254 
1255 static void draw_process_bars(struct timechart *tchart)
1256 {
1257 	struct per_pid *p;
1258 	struct per_pidcomm *c;
1259 	struct cpu_sample *sample;
1260 	int Y = 0;
1261 
1262 	Y = 2 * tchart->numcpus + 2;
1263 
1264 	p = tchart->all_data;
1265 	while (p) {
1266 		c = p->all;
1267 		while (c) {
1268 			if (!c->display) {
1269 				c->Y = 0;
1270 				c = c->next;
1271 				continue;
1272 			}
1273 
1274 			svg_box(Y, c->start_time, c->end_time, "process");
1275 			sample = c->samples;
1276 			while (sample) {
1277 				if (sample->type == TYPE_RUNNING)
1278 					svg_running(Y, sample->cpu,
1279 						    sample->start_time,
1280 						    sample->end_time,
1281 						    sample->backtrace);
1282 				if (sample->type == TYPE_BLOCKED)
1283 					svg_blocked(Y, sample->cpu,
1284 						    sample->start_time,
1285 						    sample->end_time,
1286 						    sample->backtrace);
1287 				if (sample->type == TYPE_WAITING)
1288 					svg_waiting(Y, sample->cpu,
1289 						    sample->start_time,
1290 						    sample->end_time,
1291 						    sample->backtrace);
1292 				sample = sample->next;
1293 			}
1294 
1295 			if (c->comm) {
1296 				char comm[256];
1297 				if (c->total_time > 5000000000) /* 5 seconds */
1298 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1299 				else
1300 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1301 
1302 				svg_text(Y, c->start_time, comm);
1303 			}
1304 			c->Y = Y;
1305 			Y++;
1306 			c = c->next;
1307 		}
1308 		p = p->next;
1309 	}
1310 }
1311 
1312 static void add_process_filter(const char *string)
1313 {
1314 	int pid = strtoull(string, NULL, 10);
1315 	struct process_filter *filt = malloc(sizeof(*filt));
1316 
1317 	if (!filt)
1318 		return;
1319 
1320 	filt->name = strdup(string);
1321 	filt->pid  = pid;
1322 	filt->next = process_filter;
1323 
1324 	process_filter = filt;
1325 }
1326 
1327 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1328 {
1329 	struct process_filter *filt;
1330 	if (!process_filter)
1331 		return 1;
1332 
1333 	filt = process_filter;
1334 	while (filt) {
1335 		if (filt->pid && p->pid == filt->pid)
1336 			return 1;
1337 		if (strcmp(filt->name, c->comm) == 0)
1338 			return 1;
1339 		filt = filt->next;
1340 	}
1341 	return 0;
1342 }
1343 
1344 static int determine_display_tasks_filtered(struct timechart *tchart)
1345 {
1346 	struct per_pid *p;
1347 	struct per_pidcomm *c;
1348 	int count = 0;
1349 
1350 	p = tchart->all_data;
1351 	while (p) {
1352 		p->display = 0;
1353 		if (p->start_time == 1)
1354 			p->start_time = tchart->first_time;
1355 
1356 		/* no exit marker, task kept running to the end */
1357 		if (p->end_time == 0)
1358 			p->end_time = tchart->last_time;
1359 
1360 		c = p->all;
1361 
1362 		while (c) {
1363 			c->display = 0;
1364 
1365 			if (c->start_time == 1)
1366 				c->start_time = tchart->first_time;
1367 
1368 			if (passes_filter(p, c)) {
1369 				c->display = 1;
1370 				p->display = 1;
1371 				count++;
1372 			}
1373 
1374 			if (c->end_time == 0)
1375 				c->end_time = tchart->last_time;
1376 
1377 			c = c->next;
1378 		}
1379 		p = p->next;
1380 	}
1381 	return count;
1382 }
1383 
1384 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1385 {
1386 	struct per_pid *p;
1387 	struct per_pidcomm *c;
1388 	int count = 0;
1389 
1390 	p = tchart->all_data;
1391 	while (p) {
1392 		p->display = 0;
1393 		if (p->start_time == 1)
1394 			p->start_time = tchart->first_time;
1395 
1396 		/* no exit marker, task kept running to the end */
1397 		if (p->end_time == 0)
1398 			p->end_time = tchart->last_time;
1399 		if (p->total_time >= threshold)
1400 			p->display = 1;
1401 
1402 		c = p->all;
1403 
1404 		while (c) {
1405 			c->display = 0;
1406 
1407 			if (c->start_time == 1)
1408 				c->start_time = tchart->first_time;
1409 
1410 			if (c->total_time >= threshold) {
1411 				c->display = 1;
1412 				count++;
1413 			}
1414 
1415 			if (c->end_time == 0)
1416 				c->end_time = tchart->last_time;
1417 
1418 			c = c->next;
1419 		}
1420 		p = p->next;
1421 	}
1422 	return count;
1423 }
1424 
1425 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1426 {
1427 	struct per_pid *p;
1428 	struct per_pidcomm *c;
1429 	int count = 0;
1430 
1431 	p = timechart->all_data;
1432 	while (p) {
1433 		/* no exit marker, task kept running to the end */
1434 		if (p->end_time == 0)
1435 			p->end_time = timechart->last_time;
1436 
1437 		c = p->all;
1438 
1439 		while (c) {
1440 			c->display = 0;
1441 
1442 			if (c->total_bytes >= threshold) {
1443 				c->display = 1;
1444 				count++;
1445 			}
1446 
1447 			if (c->end_time == 0)
1448 				c->end_time = timechart->last_time;
1449 
1450 			c = c->next;
1451 		}
1452 		p = p->next;
1453 	}
1454 	return count;
1455 }
1456 
1457 #define BYTES_THRESH (1 * 1024 * 1024)
1458 #define TIME_THRESH 10000000
1459 
1460 static void write_svg_file(struct timechart *tchart, const char *filename)
1461 {
1462 	u64 i;
1463 	int count;
1464 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1465 
1466 	if (tchart->power_only)
1467 		tchart->proc_num = 0;
1468 
1469 	/* We'd like to show at least proc_num tasks;
1470 	 * be less picky if we have fewer */
1471 	do {
1472 		if (process_filter)
1473 			count = determine_display_tasks_filtered(tchart);
1474 		else if (tchart->io_events)
1475 			count = determine_display_io_tasks(tchart, thresh);
1476 		else
1477 			count = determine_display_tasks(tchart, thresh);
1478 		thresh /= 10;
1479 	} while (!process_filter && thresh && count < tchart->proc_num);
1480 
1481 	if (!tchart->proc_num)
1482 		count = 0;
1483 
1484 	if (tchart->io_events) {
1485 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1486 
1487 		svg_time_grid(0.5);
1488 		svg_io_legenda();
1489 
1490 		draw_io_bars(tchart);
1491 	} else {
1492 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1493 
1494 		svg_time_grid(0);
1495 
1496 		svg_legenda();
1497 
1498 		for (i = 0; i < tchart->numcpus; i++)
1499 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1500 
1501 		draw_cpu_usage(tchart);
1502 		if (tchart->proc_num)
1503 			draw_process_bars(tchart);
1504 		if (!tchart->tasks_only)
1505 			draw_c_p_states(tchart);
1506 		if (tchart->proc_num)
1507 			draw_wakeups(tchart);
1508 	}
1509 
1510 	svg_close();
1511 }
1512 
1513 static int process_header(struct perf_file_section *section __maybe_unused,
1514 			  struct perf_header *ph,
1515 			  int feat,
1516 			  int fd __maybe_unused,
1517 			  void *data)
1518 {
1519 	struct timechart *tchart = data;
1520 
1521 	switch (feat) {
1522 	case HEADER_NRCPUS:
1523 		tchart->numcpus = ph->env.nr_cpus_avail;
1524 		break;
1525 
1526 	case HEADER_CPU_TOPOLOGY:
1527 		if (!tchart->topology)
1528 			break;
1529 
1530 		if (svg_build_topology_map(&ph->env))
1531 			fprintf(stderr, "problem building topology\n");
1532 		break;
1533 
1534 	default:
1535 		break;
1536 	}
1537 
1538 	return 0;
1539 }
1540 
1541 static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1542 {
1543 	const struct evsel_str_handler power_tracepoints[] = {
1544 		{ "power:cpu_idle",		process_sample_cpu_idle },
1545 		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1546 		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1547 		{ "sched:sched_switch",		process_sample_sched_switch },
1548 #ifdef SUPPORT_OLD_POWER_EVENTS
1549 		{ "power:power_start",		process_sample_power_start },
1550 		{ "power:power_end",		process_sample_power_end },
1551 		{ "power:power_frequency",	process_sample_power_frequency },
1552 #endif
1553 
1554 		{ "syscalls:sys_enter_read",		process_enter_read },
1555 		{ "syscalls:sys_enter_pread64",		process_enter_read },
1556 		{ "syscalls:sys_enter_readv",		process_enter_read },
1557 		{ "syscalls:sys_enter_preadv",		process_enter_read },
1558 		{ "syscalls:sys_enter_write",		process_enter_write },
1559 		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1560 		{ "syscalls:sys_enter_writev",		process_enter_write },
1561 		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1562 		{ "syscalls:sys_enter_sync",		process_enter_sync },
1563 		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1564 		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1565 		{ "syscalls:sys_enter_msync",		process_enter_sync },
1566 		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1567 		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1568 		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1569 		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1570 		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1571 		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1572 		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1573 		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1574 		{ "syscalls:sys_enter_poll",		process_enter_poll },
1575 		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1576 		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1577 		{ "syscalls:sys_enter_select",		process_enter_poll },
1578 
1579 		{ "syscalls:sys_exit_read",		process_exit_read },
1580 		{ "syscalls:sys_exit_pread64",		process_exit_read },
1581 		{ "syscalls:sys_exit_readv",		process_exit_read },
1582 		{ "syscalls:sys_exit_preadv",		process_exit_read },
1583 		{ "syscalls:sys_exit_write",		process_exit_write },
1584 		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1585 		{ "syscalls:sys_exit_writev",		process_exit_write },
1586 		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1587 		{ "syscalls:sys_exit_sync",		process_exit_sync },
1588 		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1589 		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1590 		{ "syscalls:sys_exit_msync",		process_exit_sync },
1591 		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1592 		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1593 		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1594 		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1595 		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1596 		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1597 		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1598 		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1599 		{ "syscalls:sys_exit_poll",		process_exit_poll },
1600 		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1601 		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1602 		{ "syscalls:sys_exit_select",		process_exit_poll },
1603 	};
1604 	struct perf_data data = {
1605 		.path  = input_name,
1606 		.mode  = PERF_DATA_MODE_READ,
1607 		.force = tchart->force,
1608 	};
1609 
1610 	struct perf_session *session = perf_session__new(&data, &tchart->tool);
1611 	int ret = -EINVAL;
1612 
1613 	if (IS_ERR(session))
1614 		return PTR_ERR(session);
1615 
1616 	symbol__init(&session->header.env);
1617 
1618 	(void)perf_header__process_sections(&session->header,
1619 					    perf_data__fd(session->data),
1620 					    tchart,
1621 					    process_header);
1622 
1623 	if (!perf_session__has_traces(session, "timechart record"))
1624 		goto out_delete;
1625 
1626 	if (perf_session__set_tracepoints_handlers(session,
1627 						   power_tracepoints)) {
1628 		pr_err("Initializing session tracepoint handlers failed\n");
1629 		goto out_delete;
1630 	}
1631 
1632 	ret = perf_session__process_events(session);
1633 	if (ret)
1634 		goto out_delete;
1635 
1636 	end_sample_processing(tchart);
1637 
1638 	sort_pids(tchart);
1639 
1640 	write_svg_file(tchart, output_name);
1641 
1642 	pr_info("Written %2.1f seconds of trace to %s.\n",
1643 		(tchart->last_time - tchart->first_time) / (double)NSEC_PER_SEC, output_name);
1644 out_delete:
1645 	perf_session__delete(session);
1646 	return ret;
1647 }
1648 
1649 static int timechart__io_record(int argc, const char **argv)
1650 {
1651 	unsigned int rec_argc, i;
1652 	const char **rec_argv;
1653 	const char **p;
1654 	char *filter = NULL;
1655 
1656 	const char * const common_args[] = {
1657 		"record", "-a", "-R", "-c", "1",
1658 	};
1659 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1660 
1661 	const char * const disk_events[] = {
1662 		"syscalls:sys_enter_read",
1663 		"syscalls:sys_enter_pread64",
1664 		"syscalls:sys_enter_readv",
1665 		"syscalls:sys_enter_preadv",
1666 		"syscalls:sys_enter_write",
1667 		"syscalls:sys_enter_pwrite64",
1668 		"syscalls:sys_enter_writev",
1669 		"syscalls:sys_enter_pwritev",
1670 		"syscalls:sys_enter_sync",
1671 		"syscalls:sys_enter_sync_file_range",
1672 		"syscalls:sys_enter_fsync",
1673 		"syscalls:sys_enter_msync",
1674 
1675 		"syscalls:sys_exit_read",
1676 		"syscalls:sys_exit_pread64",
1677 		"syscalls:sys_exit_readv",
1678 		"syscalls:sys_exit_preadv",
1679 		"syscalls:sys_exit_write",
1680 		"syscalls:sys_exit_pwrite64",
1681 		"syscalls:sys_exit_writev",
1682 		"syscalls:sys_exit_pwritev",
1683 		"syscalls:sys_exit_sync",
1684 		"syscalls:sys_exit_sync_file_range",
1685 		"syscalls:sys_exit_fsync",
1686 		"syscalls:sys_exit_msync",
1687 	};
1688 	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1689 
1690 	const char * const net_events[] = {
1691 		"syscalls:sys_enter_recvfrom",
1692 		"syscalls:sys_enter_recvmmsg",
1693 		"syscalls:sys_enter_recvmsg",
1694 		"syscalls:sys_enter_sendto",
1695 		"syscalls:sys_enter_sendmsg",
1696 		"syscalls:sys_enter_sendmmsg",
1697 
1698 		"syscalls:sys_exit_recvfrom",
1699 		"syscalls:sys_exit_recvmmsg",
1700 		"syscalls:sys_exit_recvmsg",
1701 		"syscalls:sys_exit_sendto",
1702 		"syscalls:sys_exit_sendmsg",
1703 		"syscalls:sys_exit_sendmmsg",
1704 	};
1705 	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1706 
1707 	const char * const poll_events[] = {
1708 		"syscalls:sys_enter_epoll_pwait",
1709 		"syscalls:sys_enter_epoll_wait",
1710 		"syscalls:sys_enter_poll",
1711 		"syscalls:sys_enter_ppoll",
1712 		"syscalls:sys_enter_pselect6",
1713 		"syscalls:sys_enter_select",
1714 
1715 		"syscalls:sys_exit_epoll_pwait",
1716 		"syscalls:sys_exit_epoll_wait",
1717 		"syscalls:sys_exit_poll",
1718 		"syscalls:sys_exit_ppoll",
1719 		"syscalls:sys_exit_pselect6",
1720 		"syscalls:sys_exit_select",
1721 	};
1722 	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1723 
1724 	rec_argc = common_args_nr +
1725 		disk_events_nr * 4 +
1726 		net_events_nr * 4 +
1727 		poll_events_nr * 4 +
1728 		argc;
1729 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1730 
1731 	if (rec_argv == NULL)
1732 		return -ENOMEM;
1733 
1734 	if (asprintf(&filter, "common_pid != %d", getpid()) < 0) {
1735 		free(rec_argv);
1736 		return -ENOMEM;
1737 	}
1738 
1739 	p = rec_argv;
1740 	for (i = 0; i < common_args_nr; i++)
1741 		*p++ = strdup(common_args[i]);
1742 
1743 	for (i = 0; i < disk_events_nr; i++) {
1744 		if (!is_valid_tracepoint(disk_events[i])) {
1745 			rec_argc -= 4;
1746 			continue;
1747 		}
1748 
1749 		*p++ = "-e";
1750 		*p++ = strdup(disk_events[i]);
1751 		*p++ = "--filter";
1752 		*p++ = filter;
1753 	}
1754 	for (i = 0; i < net_events_nr; i++) {
1755 		if (!is_valid_tracepoint(net_events[i])) {
1756 			rec_argc -= 4;
1757 			continue;
1758 		}
1759 
1760 		*p++ = "-e";
1761 		*p++ = strdup(net_events[i]);
1762 		*p++ = "--filter";
1763 		*p++ = filter;
1764 	}
1765 	for (i = 0; i < poll_events_nr; i++) {
1766 		if (!is_valid_tracepoint(poll_events[i])) {
1767 			rec_argc -= 4;
1768 			continue;
1769 		}
1770 
1771 		*p++ = "-e";
1772 		*p++ = strdup(poll_events[i]);
1773 		*p++ = "--filter";
1774 		*p++ = filter;
1775 	}
1776 
1777 	for (i = 0; i < (unsigned int)argc; i++)
1778 		*p++ = argv[i];
1779 
1780 	return cmd_record(rec_argc, rec_argv);
1781 }
1782 
1783 
1784 static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1785 {
1786 	unsigned int rec_argc, i, j;
1787 	const char **rec_argv;
1788 	const char **p;
1789 	unsigned int record_elems;
1790 
1791 	const char * const common_args[] = {
1792 		"record", "-a", "-R", "-c", "1",
1793 	};
1794 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1795 
1796 	const char * const backtrace_args[] = {
1797 		"-g",
1798 	};
1799 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1800 
1801 	const char * const power_args[] = {
1802 		"-e", "power:cpu_frequency",
1803 		"-e", "power:cpu_idle",
1804 	};
1805 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1806 
1807 	const char * const old_power_args[] = {
1808 #ifdef SUPPORT_OLD_POWER_EVENTS
1809 		"-e", "power:power_start",
1810 		"-e", "power:power_end",
1811 		"-e", "power:power_frequency",
1812 #endif
1813 	};
1814 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1815 
1816 	const char * const tasks_args[] = {
1817 		"-e", "sched:sched_wakeup",
1818 		"-e", "sched:sched_switch",
1819 	};
1820 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1821 
1822 #ifdef SUPPORT_OLD_POWER_EVENTS
1823 	if (!is_valid_tracepoint("power:cpu_idle") &&
1824 	    is_valid_tracepoint("power:power_start")) {
1825 		use_old_power_events = 1;
1826 		power_args_nr = 0;
1827 	} else {
1828 		old_power_args_nr = 0;
1829 	}
1830 #endif
1831 
1832 	if (tchart->power_only)
1833 		tasks_args_nr = 0;
1834 
1835 	if (tchart->tasks_only) {
1836 		power_args_nr = 0;
1837 		old_power_args_nr = 0;
1838 	}
1839 
1840 	if (!tchart->with_backtrace)
1841 		backtrace_args_no = 0;
1842 
1843 	record_elems = common_args_nr + tasks_args_nr +
1844 		power_args_nr + old_power_args_nr + backtrace_args_no;
1845 
1846 	rec_argc = record_elems + argc;
1847 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1848 
1849 	if (rec_argv == NULL)
1850 		return -ENOMEM;
1851 
1852 	p = rec_argv;
1853 	for (i = 0; i < common_args_nr; i++)
1854 		*p++ = strdup(common_args[i]);
1855 
1856 	for (i = 0; i < backtrace_args_no; i++)
1857 		*p++ = strdup(backtrace_args[i]);
1858 
1859 	for (i = 0; i < tasks_args_nr; i++)
1860 		*p++ = strdup(tasks_args[i]);
1861 
1862 	for (i = 0; i < power_args_nr; i++)
1863 		*p++ = strdup(power_args[i]);
1864 
1865 	for (i = 0; i < old_power_args_nr; i++)
1866 		*p++ = strdup(old_power_args[i]);
1867 
1868 	for (j = 0; j < (unsigned int)argc; j++)
1869 		*p++ = argv[j];
1870 
1871 	return cmd_record(rec_argc, rec_argv);
1872 }
1873 
1874 static int
1875 parse_process(const struct option *opt __maybe_unused, const char *arg,
1876 	      int __maybe_unused unset)
1877 {
1878 	if (arg)
1879 		add_process_filter(arg);
1880 	return 0;
1881 }
1882 
1883 static int
1884 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1885 		int __maybe_unused unset)
1886 {
1887 	unsigned long duration = strtoul(arg, NULL, 0);
1888 
1889 	if (svg_highlight || svg_highlight_name)
1890 		return -1;
1891 
1892 	if (duration)
1893 		svg_highlight = duration;
1894 	else
1895 		svg_highlight_name = strdup(arg);
1896 
1897 	return 0;
1898 }
1899 
1900 static int
1901 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1902 {
1903 	char unit = 'n';
1904 	u64 *value = opt->value;
1905 
1906 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1907 		switch (unit) {
1908 		case 'm':
1909 			*value *= NSEC_PER_MSEC;
1910 			break;
1911 		case 'u':
1912 			*value *= NSEC_PER_USEC;
1913 			break;
1914 		case 'n':
1915 			break;
1916 		default:
1917 			return -1;
1918 		}
1919 	}
1920 
1921 	return 0;
1922 }
1923 
1924 int cmd_timechart(int argc, const char **argv)
1925 {
1926 	struct timechart tchart = {
1927 		.tool = {
1928 			.comm		 = process_comm_event,
1929 			.fork		 = process_fork_event,
1930 			.exit		 = process_exit_event,
1931 			.sample		 = process_sample_event,
1932 			.ordered_events	 = true,
1933 		},
1934 		.proc_num = 15,
1935 		.min_time = NSEC_PER_MSEC,
1936 		.merge_dist = 1000,
1937 	};
1938 	const char *output_name = "output.svg";
1939 	const struct option timechart_common_options[] = {
1940 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1941 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only, "output processes data only"),
1942 	OPT_END()
1943 	};
1944 	const struct option timechart_options[] = {
1945 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1946 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1947 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1948 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1949 		      "highlight tasks. Pass duration in ns or process name.",
1950 		       parse_highlight),
1951 	OPT_CALLBACK('p', "process", NULL, "process",
1952 		      "process selector. Pass a pid or process name.",
1953 		       parse_process),
1954 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1955 		     "Look for files with symbols relative to this directory",
1956 		     symbol__config_symfs),
1957 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1958 		    "min. number of tasks to print"),
1959 	OPT_BOOLEAN('t', "topology", &tchart.topology,
1960 		    "sort CPUs according to topology"),
1961 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1962 		    "skip EAGAIN errors"),
1963 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1964 		     "all IO faster than min-time will visually appear longer",
1965 		     parse_time),
1966 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1967 		     "merge events that are merge-dist us apart",
1968 		     parse_time),
1969 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1970 	OPT_PARENT(timechart_common_options),
1971 	};
1972 	const char * const timechart_subcommands[] = { "record", NULL };
1973 	const char *timechart_usage[] = {
1974 		"perf timechart [<options>] {record}",
1975 		NULL
1976 	};
1977 	const struct option timechart_record_options[] = {
1978 	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
1979 		    "record only IO data"),
1980 	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
1981 	OPT_PARENT(timechart_common_options),
1982 	};
1983 	const char * const timechart_record_usage[] = {
1984 		"perf timechart record [<options>]",
1985 		NULL
1986 	};
1987 	int ret;
1988 
1989 	cpus_cstate_start_times = calloc(MAX_CPUS, sizeof(*cpus_cstate_start_times));
1990 	if (!cpus_cstate_start_times)
1991 		return -ENOMEM;
1992 	cpus_cstate_state = calloc(MAX_CPUS, sizeof(*cpus_cstate_state));
1993 	if (!cpus_cstate_state) {
1994 		ret = -ENOMEM;
1995 		goto out;
1996 	}
1997 	cpus_pstate_start_times = calloc(MAX_CPUS, sizeof(*cpus_pstate_start_times));
1998 	if (!cpus_pstate_start_times) {
1999 		ret = -ENOMEM;
2000 		goto out;
2001 	}
2002 	cpus_pstate_state = calloc(MAX_CPUS, sizeof(*cpus_pstate_state));
2003 	if (!cpus_pstate_state) {
2004 		ret = -ENOMEM;
2005 		goto out;
2006 	}
2007 
2008 	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
2009 			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
2010 
2011 	if (tchart.power_only && tchart.tasks_only) {
2012 		pr_err("-P and -T options cannot be used at the same time.\n");
2013 		ret = -1;
2014 		goto out;
2015 	}
2016 
2017 	if (argc && strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
2018 		argc = parse_options(argc, argv, timechart_record_options,
2019 				     timechart_record_usage,
2020 				     PARSE_OPT_STOP_AT_NON_OPTION);
2021 
2022 		if (tchart.power_only && tchart.tasks_only) {
2023 			pr_err("-P and -T options cannot be used at the same time.\n");
2024 			ret = -1;
2025 			goto out;
2026 		}
2027 
2028 		if (tchart.io_only)
2029 			ret = timechart__io_record(argc, argv);
2030 		else
2031 			ret = timechart__record(&tchart, argc, argv);
2032 		goto out;
2033 	} else if (argc)
2034 		usage_with_options(timechart_usage, timechart_options);
2035 
2036 	setup_pager();
2037 
2038 	ret = __cmd_timechart(&tchart, output_name);
2039 out:
2040 	zfree(&cpus_cstate_start_times);
2041 	zfree(&cpus_cstate_state);
2042 	zfree(&cpus_pstate_start_times);
2043 	zfree(&cpus_pstate_state);
2044 	return ret;
2045 }
2046