1 /*
2  * builtin-timechart.c - make an svg timechart of system activity
3  *
4  * (C) Copyright 2009 Intel Corporation
5  *
6  * Authors:
7  *     Arjan van de Ven <arjan@linux.intel.com>
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * as published by the Free Software Foundation; version 2
12  * of the License.
13  */
14 
15 #include <errno.h>
16 #include <inttypes.h>
17 #include <traceevent/event-parse.h>
18 
19 #include "builtin.h"
20 
21 #include "util/util.h"
22 
23 #include "util/color.h"
24 #include <linux/list.h>
25 #include "util/cache.h"
26 #include "util/evlist.h"
27 #include "util/evsel.h"
28 #include <linux/kernel.h>
29 #include <linux/rbtree.h>
30 #include <linux/time64.h>
31 #include "util/symbol.h"
32 #include "util/thread.h"
33 #include "util/callchain.h"
34 
35 #include "perf.h"
36 #include "util/header.h"
37 #include <subcmd/parse-options.h>
38 #include "util/parse-events.h"
39 #include "util/event.h"
40 #include "util/session.h"
41 #include "util/svghelper.h"
42 #include "util/tool.h"
43 #include "util/data.h"
44 #include "util/debug.h"
45 
46 #ifdef LACKS_OPEN_MEMSTREAM_PROTOTYPE
47 FILE *open_memstream(char **ptr, size_t *sizeloc);
48 #endif
49 
50 #define SUPPORT_OLD_POWER_EVENTS 1
51 #define PWR_EVENT_EXIT -1
52 
53 struct per_pid;
54 struct power_event;
55 struct wake_event;
56 
57 struct timechart {
58 	struct perf_tool	tool;
59 	struct per_pid		*all_data;
60 	struct power_event	*power_events;
61 	struct wake_event	*wake_events;
62 	int			proc_num;
63 	unsigned int		numcpus;
64 	u64			min_freq,	/* Lowest CPU frequency seen */
65 				max_freq,	/* Highest CPU frequency seen */
66 				turbo_frequency,
67 				first_time, last_time;
68 	bool			power_only,
69 				tasks_only,
70 				with_backtrace,
71 				topology;
72 	bool			force;
73 	/* IO related settings */
74 	bool			io_only,
75 				skip_eagain;
76 	u64			io_events;
77 	u64			min_time,
78 				merge_dist;
79 };
80 
81 struct per_pidcomm;
82 struct cpu_sample;
83 struct io_sample;
84 
85 /*
86  * Datastructure layout:
87  * We keep an list of "pid"s, matching the kernels notion of a task struct.
88  * Each "pid" entry, has a list of "comm"s.
89  *	this is because we want to track different programs different, while
90  *	exec will reuse the original pid (by design).
91  * Each comm has a list of samples that will be used to draw
92  * final graph.
93  */
94 
95 struct per_pid {
96 	struct per_pid *next;
97 
98 	int		pid;
99 	int		ppid;
100 
101 	u64		start_time;
102 	u64		end_time;
103 	u64		total_time;
104 	u64		total_bytes;
105 	int		display;
106 
107 	struct per_pidcomm *all;
108 	struct per_pidcomm *current;
109 };
110 
111 
112 struct per_pidcomm {
113 	struct per_pidcomm *next;
114 
115 	u64		start_time;
116 	u64		end_time;
117 	u64		total_time;
118 	u64		max_bytes;
119 	u64		total_bytes;
120 
121 	int		Y;
122 	int		display;
123 
124 	long		state;
125 	u64		state_since;
126 
127 	char		*comm;
128 
129 	struct cpu_sample *samples;
130 	struct io_sample  *io_samples;
131 };
132 
133 struct sample_wrapper {
134 	struct sample_wrapper *next;
135 
136 	u64		timestamp;
137 	unsigned char	data[0];
138 };
139 
140 #define TYPE_NONE	0
141 #define TYPE_RUNNING	1
142 #define TYPE_WAITING	2
143 #define TYPE_BLOCKED	3
144 
145 struct cpu_sample {
146 	struct cpu_sample *next;
147 
148 	u64 start_time;
149 	u64 end_time;
150 	int type;
151 	int cpu;
152 	const char *backtrace;
153 };
154 
155 enum {
156 	IOTYPE_READ,
157 	IOTYPE_WRITE,
158 	IOTYPE_SYNC,
159 	IOTYPE_TX,
160 	IOTYPE_RX,
161 	IOTYPE_POLL,
162 };
163 
164 struct io_sample {
165 	struct io_sample *next;
166 
167 	u64 start_time;
168 	u64 end_time;
169 	u64 bytes;
170 	int type;
171 	int fd;
172 	int err;
173 	int merges;
174 };
175 
176 #define CSTATE 1
177 #define PSTATE 2
178 
179 struct power_event {
180 	struct power_event *next;
181 	int type;
182 	int state;
183 	u64 start_time;
184 	u64 end_time;
185 	int cpu;
186 };
187 
188 struct wake_event {
189 	struct wake_event *next;
190 	int waker;
191 	int wakee;
192 	u64 time;
193 	const char *backtrace;
194 };
195 
196 struct process_filter {
197 	char			*name;
198 	int			pid;
199 	struct process_filter	*next;
200 };
201 
202 static struct process_filter *process_filter;
203 
204 
205 static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
206 {
207 	struct per_pid *cursor = tchart->all_data;
208 
209 	while (cursor) {
210 		if (cursor->pid == pid)
211 			return cursor;
212 		cursor = cursor->next;
213 	}
214 	cursor = zalloc(sizeof(*cursor));
215 	assert(cursor != NULL);
216 	cursor->pid = pid;
217 	cursor->next = tchart->all_data;
218 	tchart->all_data = cursor;
219 	return cursor;
220 }
221 
222 static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
223 {
224 	struct per_pid *p;
225 	struct per_pidcomm *c;
226 	p = find_create_pid(tchart, pid);
227 	c = p->all;
228 	while (c) {
229 		if (c->comm && strcmp(c->comm, comm) == 0) {
230 			p->current = c;
231 			return;
232 		}
233 		if (!c->comm) {
234 			c->comm = strdup(comm);
235 			p->current = c;
236 			return;
237 		}
238 		c = c->next;
239 	}
240 	c = zalloc(sizeof(*c));
241 	assert(c != NULL);
242 	c->comm = strdup(comm);
243 	p->current = c;
244 	c->next = p->all;
245 	p->all = c;
246 }
247 
248 static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
249 {
250 	struct per_pid *p, *pp;
251 	p = find_create_pid(tchart, pid);
252 	pp = find_create_pid(tchart, ppid);
253 	p->ppid = ppid;
254 	if (pp->current && pp->current->comm && !p->current)
255 		pid_set_comm(tchart, pid, pp->current->comm);
256 
257 	p->start_time = timestamp;
258 	if (p->current && !p->current->start_time) {
259 		p->current->start_time = timestamp;
260 		p->current->state_since = timestamp;
261 	}
262 }
263 
264 static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
265 {
266 	struct per_pid *p;
267 	p = find_create_pid(tchart, pid);
268 	p->end_time = timestamp;
269 	if (p->current)
270 		p->current->end_time = timestamp;
271 }
272 
273 static void pid_put_sample(struct timechart *tchart, int pid, int type,
274 			   unsigned int cpu, u64 start, u64 end,
275 			   const char *backtrace)
276 {
277 	struct per_pid *p;
278 	struct per_pidcomm *c;
279 	struct cpu_sample *sample;
280 
281 	p = find_create_pid(tchart, pid);
282 	c = p->current;
283 	if (!c) {
284 		c = zalloc(sizeof(*c));
285 		assert(c != NULL);
286 		p->current = c;
287 		c->next = p->all;
288 		p->all = c;
289 	}
290 
291 	sample = zalloc(sizeof(*sample));
292 	assert(sample != NULL);
293 	sample->start_time = start;
294 	sample->end_time = end;
295 	sample->type = type;
296 	sample->next = c->samples;
297 	sample->cpu = cpu;
298 	sample->backtrace = backtrace;
299 	c->samples = sample;
300 
301 	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
302 		c->total_time += (end-start);
303 		p->total_time += (end-start);
304 	}
305 
306 	if (c->start_time == 0 || c->start_time > start)
307 		c->start_time = start;
308 	if (p->start_time == 0 || p->start_time > start)
309 		p->start_time = start;
310 }
311 
312 #define MAX_CPUS 4096
313 
314 static u64 cpus_cstate_start_times[MAX_CPUS];
315 static int cpus_cstate_state[MAX_CPUS];
316 static u64 cpus_pstate_start_times[MAX_CPUS];
317 static u64 cpus_pstate_state[MAX_CPUS];
318 
319 static int process_comm_event(struct perf_tool *tool,
320 			      union perf_event *event,
321 			      struct perf_sample *sample __maybe_unused,
322 			      struct machine *machine __maybe_unused)
323 {
324 	struct timechart *tchart = container_of(tool, struct timechart, tool);
325 	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
326 	return 0;
327 }
328 
329 static int process_fork_event(struct perf_tool *tool,
330 			      union perf_event *event,
331 			      struct perf_sample *sample __maybe_unused,
332 			      struct machine *machine __maybe_unused)
333 {
334 	struct timechart *tchart = container_of(tool, struct timechart, tool);
335 	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
336 	return 0;
337 }
338 
339 static int process_exit_event(struct perf_tool *tool,
340 			      union perf_event *event,
341 			      struct perf_sample *sample __maybe_unused,
342 			      struct machine *machine __maybe_unused)
343 {
344 	struct timechart *tchart = container_of(tool, struct timechart, tool);
345 	pid_exit(tchart, event->fork.pid, event->fork.time);
346 	return 0;
347 }
348 
349 #ifdef SUPPORT_OLD_POWER_EVENTS
350 static int use_old_power_events;
351 #endif
352 
353 static void c_state_start(int cpu, u64 timestamp, int state)
354 {
355 	cpus_cstate_start_times[cpu] = timestamp;
356 	cpus_cstate_state[cpu] = state;
357 }
358 
359 static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
360 {
361 	struct power_event *pwr = zalloc(sizeof(*pwr));
362 
363 	if (!pwr)
364 		return;
365 
366 	pwr->state = cpus_cstate_state[cpu];
367 	pwr->start_time = cpus_cstate_start_times[cpu];
368 	pwr->end_time = timestamp;
369 	pwr->cpu = cpu;
370 	pwr->type = CSTATE;
371 	pwr->next = tchart->power_events;
372 
373 	tchart->power_events = pwr;
374 }
375 
376 static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
377 {
378 	struct power_event *pwr;
379 
380 	if (new_freq > 8000000) /* detect invalid data */
381 		return;
382 
383 	pwr = zalloc(sizeof(*pwr));
384 	if (!pwr)
385 		return;
386 
387 	pwr->state = cpus_pstate_state[cpu];
388 	pwr->start_time = cpus_pstate_start_times[cpu];
389 	pwr->end_time = timestamp;
390 	pwr->cpu = cpu;
391 	pwr->type = PSTATE;
392 	pwr->next = tchart->power_events;
393 
394 	if (!pwr->start_time)
395 		pwr->start_time = tchart->first_time;
396 
397 	tchart->power_events = pwr;
398 
399 	cpus_pstate_state[cpu] = new_freq;
400 	cpus_pstate_start_times[cpu] = timestamp;
401 
402 	if ((u64)new_freq > tchart->max_freq)
403 		tchart->max_freq = new_freq;
404 
405 	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
406 		tchart->min_freq = new_freq;
407 
408 	if (new_freq == tchart->max_freq - 1000)
409 		tchart->turbo_frequency = tchart->max_freq;
410 }
411 
412 static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
413 			 int waker, int wakee, u8 flags, const char *backtrace)
414 {
415 	struct per_pid *p;
416 	struct wake_event *we = zalloc(sizeof(*we));
417 
418 	if (!we)
419 		return;
420 
421 	we->time = timestamp;
422 	we->waker = waker;
423 	we->backtrace = backtrace;
424 
425 	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
426 		we->waker = -1;
427 
428 	we->wakee = wakee;
429 	we->next = tchart->wake_events;
430 	tchart->wake_events = we;
431 	p = find_create_pid(tchart, we->wakee);
432 
433 	if (p && p->current && p->current->state == TYPE_NONE) {
434 		p->current->state_since = timestamp;
435 		p->current->state = TYPE_WAITING;
436 	}
437 	if (p && p->current && p->current->state == TYPE_BLOCKED) {
438 		pid_put_sample(tchart, p->pid, p->current->state, cpu,
439 			       p->current->state_since, timestamp, NULL);
440 		p->current->state_since = timestamp;
441 		p->current->state = TYPE_WAITING;
442 	}
443 }
444 
445 static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
446 			 int prev_pid, int next_pid, u64 prev_state,
447 			 const char *backtrace)
448 {
449 	struct per_pid *p = NULL, *prev_p;
450 
451 	prev_p = find_create_pid(tchart, prev_pid);
452 
453 	p = find_create_pid(tchart, next_pid);
454 
455 	if (prev_p->current && prev_p->current->state != TYPE_NONE)
456 		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
457 			       prev_p->current->state_since, timestamp,
458 			       backtrace);
459 	if (p && p->current) {
460 		if (p->current->state != TYPE_NONE)
461 			pid_put_sample(tchart, next_pid, p->current->state, cpu,
462 				       p->current->state_since, timestamp,
463 				       backtrace);
464 
465 		p->current->state_since = timestamp;
466 		p->current->state = TYPE_RUNNING;
467 	}
468 
469 	if (prev_p->current) {
470 		prev_p->current->state = TYPE_NONE;
471 		prev_p->current->state_since = timestamp;
472 		if (prev_state & 2)
473 			prev_p->current->state = TYPE_BLOCKED;
474 		if (prev_state == 0)
475 			prev_p->current->state = TYPE_WAITING;
476 	}
477 }
478 
479 static const char *cat_backtrace(union perf_event *event,
480 				 struct perf_sample *sample,
481 				 struct machine *machine)
482 {
483 	struct addr_location al;
484 	unsigned int i;
485 	char *p = NULL;
486 	size_t p_len;
487 	u8 cpumode = PERF_RECORD_MISC_USER;
488 	struct addr_location tal;
489 	struct ip_callchain *chain = sample->callchain;
490 	FILE *f = open_memstream(&p, &p_len);
491 
492 	if (!f) {
493 		perror("open_memstream error");
494 		return NULL;
495 	}
496 
497 	if (!chain)
498 		goto exit;
499 
500 	if (machine__resolve(machine, &al, sample) < 0) {
501 		fprintf(stderr, "problem processing %d event, skipping it.\n",
502 			event->header.type);
503 		goto exit;
504 	}
505 
506 	for (i = 0; i < chain->nr; i++) {
507 		u64 ip;
508 
509 		if (callchain_param.order == ORDER_CALLEE)
510 			ip = chain->ips[i];
511 		else
512 			ip = chain->ips[chain->nr - i - 1];
513 
514 		if (ip >= PERF_CONTEXT_MAX) {
515 			switch (ip) {
516 			case PERF_CONTEXT_HV:
517 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
518 				break;
519 			case PERF_CONTEXT_KERNEL:
520 				cpumode = PERF_RECORD_MISC_KERNEL;
521 				break;
522 			case PERF_CONTEXT_USER:
523 				cpumode = PERF_RECORD_MISC_USER;
524 				break;
525 			default:
526 				pr_debug("invalid callchain context: "
527 					 "%"PRId64"\n", (s64) ip);
528 
529 				/*
530 				 * It seems the callchain is corrupted.
531 				 * Discard all.
532 				 */
533 				zfree(&p);
534 				goto exit_put;
535 			}
536 			continue;
537 		}
538 
539 		tal.filtered = 0;
540 		if (thread__find_symbol(al.thread, cpumode, ip, &tal))
541 			fprintf(f, "..... %016" PRIx64 " %s\n", ip, tal.sym->name);
542 		else
543 			fprintf(f, "..... %016" PRIx64 "\n", ip);
544 	}
545 exit_put:
546 	addr_location__put(&al);
547 exit:
548 	fclose(f);
549 
550 	return p;
551 }
552 
553 typedef int (*tracepoint_handler)(struct timechart *tchart,
554 				  struct perf_evsel *evsel,
555 				  struct perf_sample *sample,
556 				  const char *backtrace);
557 
558 static int process_sample_event(struct perf_tool *tool,
559 				union perf_event *event,
560 				struct perf_sample *sample,
561 				struct perf_evsel *evsel,
562 				struct machine *machine)
563 {
564 	struct timechart *tchart = container_of(tool, struct timechart, tool);
565 
566 	if (evsel->attr.sample_type & PERF_SAMPLE_TIME) {
567 		if (!tchart->first_time || tchart->first_time > sample->time)
568 			tchart->first_time = sample->time;
569 		if (tchart->last_time < sample->time)
570 			tchart->last_time = sample->time;
571 	}
572 
573 	if (evsel->handler != NULL) {
574 		tracepoint_handler f = evsel->handler;
575 		return f(tchart, evsel, sample,
576 			 cat_backtrace(event, sample, machine));
577 	}
578 
579 	return 0;
580 }
581 
582 static int
583 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
584 			struct perf_evsel *evsel,
585 			struct perf_sample *sample,
586 			const char *backtrace __maybe_unused)
587 {
588 	u32 state = perf_evsel__intval(evsel, sample, "state");
589 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
590 
591 	if (state == (u32)PWR_EVENT_EXIT)
592 		c_state_end(tchart, cpu_id, sample->time);
593 	else
594 		c_state_start(cpu_id, sample->time, state);
595 	return 0;
596 }
597 
598 static int
599 process_sample_cpu_frequency(struct timechart *tchart,
600 			     struct perf_evsel *evsel,
601 			     struct perf_sample *sample,
602 			     const char *backtrace __maybe_unused)
603 {
604 	u32 state = perf_evsel__intval(evsel, sample, "state");
605 	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
606 
607 	p_state_change(tchart, cpu_id, sample->time, state);
608 	return 0;
609 }
610 
611 static int
612 process_sample_sched_wakeup(struct timechart *tchart,
613 			    struct perf_evsel *evsel,
614 			    struct perf_sample *sample,
615 			    const char *backtrace)
616 {
617 	u8 flags = perf_evsel__intval(evsel, sample, "common_flags");
618 	int waker = perf_evsel__intval(evsel, sample, "common_pid");
619 	int wakee = perf_evsel__intval(evsel, sample, "pid");
620 
621 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
622 	return 0;
623 }
624 
625 static int
626 process_sample_sched_switch(struct timechart *tchart,
627 			    struct perf_evsel *evsel,
628 			    struct perf_sample *sample,
629 			    const char *backtrace)
630 {
631 	int prev_pid = perf_evsel__intval(evsel, sample, "prev_pid");
632 	int next_pid = perf_evsel__intval(evsel, sample, "next_pid");
633 	u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
634 
635 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
636 		     prev_state, backtrace);
637 	return 0;
638 }
639 
640 #ifdef SUPPORT_OLD_POWER_EVENTS
641 static int
642 process_sample_power_start(struct timechart *tchart __maybe_unused,
643 			   struct perf_evsel *evsel,
644 			   struct perf_sample *sample,
645 			   const char *backtrace __maybe_unused)
646 {
647 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
648 	u64 value = perf_evsel__intval(evsel, sample, "value");
649 
650 	c_state_start(cpu_id, sample->time, value);
651 	return 0;
652 }
653 
654 static int
655 process_sample_power_end(struct timechart *tchart,
656 			 struct perf_evsel *evsel __maybe_unused,
657 			 struct perf_sample *sample,
658 			 const char *backtrace __maybe_unused)
659 {
660 	c_state_end(tchart, sample->cpu, sample->time);
661 	return 0;
662 }
663 
664 static int
665 process_sample_power_frequency(struct timechart *tchart,
666 			       struct perf_evsel *evsel,
667 			       struct perf_sample *sample,
668 			       const char *backtrace __maybe_unused)
669 {
670 	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
671 	u64 value = perf_evsel__intval(evsel, sample, "value");
672 
673 	p_state_change(tchart, cpu_id, sample->time, value);
674 	return 0;
675 }
676 #endif /* SUPPORT_OLD_POWER_EVENTS */
677 
678 /*
679  * After the last sample we need to wrap up the current C/P state
680  * and close out each CPU for these.
681  */
682 static void end_sample_processing(struct timechart *tchart)
683 {
684 	u64 cpu;
685 	struct power_event *pwr;
686 
687 	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
688 		/* C state */
689 #if 0
690 		pwr = zalloc(sizeof(*pwr));
691 		if (!pwr)
692 			return;
693 
694 		pwr->state = cpus_cstate_state[cpu];
695 		pwr->start_time = cpus_cstate_start_times[cpu];
696 		pwr->end_time = tchart->last_time;
697 		pwr->cpu = cpu;
698 		pwr->type = CSTATE;
699 		pwr->next = tchart->power_events;
700 
701 		tchart->power_events = pwr;
702 #endif
703 		/* P state */
704 
705 		pwr = zalloc(sizeof(*pwr));
706 		if (!pwr)
707 			return;
708 
709 		pwr->state = cpus_pstate_state[cpu];
710 		pwr->start_time = cpus_pstate_start_times[cpu];
711 		pwr->end_time = tchart->last_time;
712 		pwr->cpu = cpu;
713 		pwr->type = PSTATE;
714 		pwr->next = tchart->power_events;
715 
716 		if (!pwr->start_time)
717 			pwr->start_time = tchart->first_time;
718 		if (!pwr->state)
719 			pwr->state = tchart->min_freq;
720 		tchart->power_events = pwr;
721 	}
722 }
723 
724 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
725 			       u64 start, int fd)
726 {
727 	struct per_pid *p = find_create_pid(tchart, pid);
728 	struct per_pidcomm *c = p->current;
729 	struct io_sample *sample;
730 	struct io_sample *prev;
731 
732 	if (!c) {
733 		c = zalloc(sizeof(*c));
734 		if (!c)
735 			return -ENOMEM;
736 		p->current = c;
737 		c->next = p->all;
738 		p->all = c;
739 	}
740 
741 	prev = c->io_samples;
742 
743 	if (prev && prev->start_time && !prev->end_time) {
744 		pr_warning("Skip invalid start event: "
745 			   "previous event already started!\n");
746 
747 		/* remove previous event that has been started,
748 		 * we are not sure we will ever get an end for it */
749 		c->io_samples = prev->next;
750 		free(prev);
751 		return 0;
752 	}
753 
754 	sample = zalloc(sizeof(*sample));
755 	if (!sample)
756 		return -ENOMEM;
757 	sample->start_time = start;
758 	sample->type = type;
759 	sample->fd = fd;
760 	sample->next = c->io_samples;
761 	c->io_samples = sample;
762 
763 	if (c->start_time == 0 || c->start_time > start)
764 		c->start_time = start;
765 
766 	return 0;
767 }
768 
769 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
770 			     u64 end, long ret)
771 {
772 	struct per_pid *p = find_create_pid(tchart, pid);
773 	struct per_pidcomm *c = p->current;
774 	struct io_sample *sample, *prev;
775 
776 	if (!c) {
777 		pr_warning("Invalid pidcomm!\n");
778 		return -1;
779 	}
780 
781 	sample = c->io_samples;
782 
783 	if (!sample) /* skip partially captured events */
784 		return 0;
785 
786 	if (sample->end_time) {
787 		pr_warning("Skip invalid end event: "
788 			   "previous event already ended!\n");
789 		return 0;
790 	}
791 
792 	if (sample->type != type) {
793 		pr_warning("Skip invalid end event: invalid event type!\n");
794 		return 0;
795 	}
796 
797 	sample->end_time = end;
798 	prev = sample->next;
799 
800 	/* we want to be able to see small and fast transfers, so make them
801 	 * at least min_time long, but don't overlap them */
802 	if (sample->end_time - sample->start_time < tchart->min_time)
803 		sample->end_time = sample->start_time + tchart->min_time;
804 	if (prev && sample->start_time < prev->end_time) {
805 		if (prev->err) /* try to make errors more visible */
806 			sample->start_time = prev->end_time;
807 		else
808 			prev->end_time = sample->start_time;
809 	}
810 
811 	if (ret < 0) {
812 		sample->err = ret;
813 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
814 		   type == IOTYPE_TX || type == IOTYPE_RX) {
815 
816 		if ((u64)ret > c->max_bytes)
817 			c->max_bytes = ret;
818 
819 		c->total_bytes += ret;
820 		p->total_bytes += ret;
821 		sample->bytes = ret;
822 	}
823 
824 	/* merge two requests to make svg smaller and render-friendly */
825 	if (prev &&
826 	    prev->type == sample->type &&
827 	    prev->err == sample->err &&
828 	    prev->fd == sample->fd &&
829 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
830 
831 		sample->bytes += prev->bytes;
832 		sample->merges += prev->merges + 1;
833 
834 		sample->start_time = prev->start_time;
835 		sample->next = prev->next;
836 		free(prev);
837 
838 		if (!sample->err && sample->bytes > c->max_bytes)
839 			c->max_bytes = sample->bytes;
840 	}
841 
842 	tchart->io_events++;
843 
844 	return 0;
845 }
846 
847 static int
848 process_enter_read(struct timechart *tchart,
849 		   struct perf_evsel *evsel,
850 		   struct perf_sample *sample)
851 {
852 	long fd = perf_evsel__intval(evsel, sample, "fd");
853 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
854 				   sample->time, fd);
855 }
856 
857 static int
858 process_exit_read(struct timechart *tchart,
859 		  struct perf_evsel *evsel,
860 		  struct perf_sample *sample)
861 {
862 	long ret = perf_evsel__intval(evsel, sample, "ret");
863 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
864 				 sample->time, ret);
865 }
866 
867 static int
868 process_enter_write(struct timechart *tchart,
869 		    struct perf_evsel *evsel,
870 		    struct perf_sample *sample)
871 {
872 	long fd = perf_evsel__intval(evsel, sample, "fd");
873 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
874 				   sample->time, fd);
875 }
876 
877 static int
878 process_exit_write(struct timechart *tchart,
879 		   struct perf_evsel *evsel,
880 		   struct perf_sample *sample)
881 {
882 	long ret = perf_evsel__intval(evsel, sample, "ret");
883 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
884 				 sample->time, ret);
885 }
886 
887 static int
888 process_enter_sync(struct timechart *tchart,
889 		   struct perf_evsel *evsel,
890 		   struct perf_sample *sample)
891 {
892 	long fd = perf_evsel__intval(evsel, sample, "fd");
893 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
894 				   sample->time, fd);
895 }
896 
897 static int
898 process_exit_sync(struct timechart *tchart,
899 		  struct perf_evsel *evsel,
900 		  struct perf_sample *sample)
901 {
902 	long ret = perf_evsel__intval(evsel, sample, "ret");
903 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
904 				 sample->time, ret);
905 }
906 
907 static int
908 process_enter_tx(struct timechart *tchart,
909 		 struct perf_evsel *evsel,
910 		 struct perf_sample *sample)
911 {
912 	long fd = perf_evsel__intval(evsel, sample, "fd");
913 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
914 				   sample->time, fd);
915 }
916 
917 static int
918 process_exit_tx(struct timechart *tchart,
919 		struct perf_evsel *evsel,
920 		struct perf_sample *sample)
921 {
922 	long ret = perf_evsel__intval(evsel, sample, "ret");
923 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
924 				 sample->time, ret);
925 }
926 
927 static int
928 process_enter_rx(struct timechart *tchart,
929 		 struct perf_evsel *evsel,
930 		 struct perf_sample *sample)
931 {
932 	long fd = perf_evsel__intval(evsel, sample, "fd");
933 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
934 				   sample->time, fd);
935 }
936 
937 static int
938 process_exit_rx(struct timechart *tchart,
939 		struct perf_evsel *evsel,
940 		struct perf_sample *sample)
941 {
942 	long ret = perf_evsel__intval(evsel, sample, "ret");
943 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
944 				 sample->time, ret);
945 }
946 
947 static int
948 process_enter_poll(struct timechart *tchart,
949 		   struct perf_evsel *evsel,
950 		   struct perf_sample *sample)
951 {
952 	long fd = perf_evsel__intval(evsel, sample, "fd");
953 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
954 				   sample->time, fd);
955 }
956 
957 static int
958 process_exit_poll(struct timechart *tchart,
959 		  struct perf_evsel *evsel,
960 		  struct perf_sample *sample)
961 {
962 	long ret = perf_evsel__intval(evsel, sample, "ret");
963 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
964 				 sample->time, ret);
965 }
966 
967 /*
968  * Sort the pid datastructure
969  */
970 static void sort_pids(struct timechart *tchart)
971 {
972 	struct per_pid *new_list, *p, *cursor, *prev;
973 	/* sort by ppid first, then by pid, lowest to highest */
974 
975 	new_list = NULL;
976 
977 	while (tchart->all_data) {
978 		p = tchart->all_data;
979 		tchart->all_data = p->next;
980 		p->next = NULL;
981 
982 		if (new_list == NULL) {
983 			new_list = p;
984 			p->next = NULL;
985 			continue;
986 		}
987 		prev = NULL;
988 		cursor = new_list;
989 		while (cursor) {
990 			if (cursor->ppid > p->ppid ||
991 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
992 				/* must insert before */
993 				if (prev) {
994 					p->next = prev->next;
995 					prev->next = p;
996 					cursor = NULL;
997 					continue;
998 				} else {
999 					p->next = new_list;
1000 					new_list = p;
1001 					cursor = NULL;
1002 					continue;
1003 				}
1004 			}
1005 
1006 			prev = cursor;
1007 			cursor = cursor->next;
1008 			if (!cursor)
1009 				prev->next = p;
1010 		}
1011 	}
1012 	tchart->all_data = new_list;
1013 }
1014 
1015 
1016 static void draw_c_p_states(struct timechart *tchart)
1017 {
1018 	struct power_event *pwr;
1019 	pwr = tchart->power_events;
1020 
1021 	/*
1022 	 * two pass drawing so that the P state bars are on top of the C state blocks
1023 	 */
1024 	while (pwr) {
1025 		if (pwr->type == CSTATE)
1026 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1027 		pwr = pwr->next;
1028 	}
1029 
1030 	pwr = tchart->power_events;
1031 	while (pwr) {
1032 		if (pwr->type == PSTATE) {
1033 			if (!pwr->state)
1034 				pwr->state = tchart->min_freq;
1035 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1036 		}
1037 		pwr = pwr->next;
1038 	}
1039 }
1040 
1041 static void draw_wakeups(struct timechart *tchart)
1042 {
1043 	struct wake_event *we;
1044 	struct per_pid *p;
1045 	struct per_pidcomm *c;
1046 
1047 	we = tchart->wake_events;
1048 	while (we) {
1049 		int from = 0, to = 0;
1050 		char *task_from = NULL, *task_to = NULL;
1051 
1052 		/* locate the column of the waker and wakee */
1053 		p = tchart->all_data;
1054 		while (p) {
1055 			if (p->pid == we->waker || p->pid == we->wakee) {
1056 				c = p->all;
1057 				while (c) {
1058 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1059 						if (p->pid == we->waker && !from) {
1060 							from = c->Y;
1061 							task_from = strdup(c->comm);
1062 						}
1063 						if (p->pid == we->wakee && !to) {
1064 							to = c->Y;
1065 							task_to = strdup(c->comm);
1066 						}
1067 					}
1068 					c = c->next;
1069 				}
1070 				c = p->all;
1071 				while (c) {
1072 					if (p->pid == we->waker && !from) {
1073 						from = c->Y;
1074 						task_from = strdup(c->comm);
1075 					}
1076 					if (p->pid == we->wakee && !to) {
1077 						to = c->Y;
1078 						task_to = strdup(c->comm);
1079 					}
1080 					c = c->next;
1081 				}
1082 			}
1083 			p = p->next;
1084 		}
1085 
1086 		if (!task_from) {
1087 			task_from = malloc(40);
1088 			sprintf(task_from, "[%i]", we->waker);
1089 		}
1090 		if (!task_to) {
1091 			task_to = malloc(40);
1092 			sprintf(task_to, "[%i]", we->wakee);
1093 		}
1094 
1095 		if (we->waker == -1)
1096 			svg_interrupt(we->time, to, we->backtrace);
1097 		else if (from && to && abs(from - to) == 1)
1098 			svg_wakeline(we->time, from, to, we->backtrace);
1099 		else
1100 			svg_partial_wakeline(we->time, from, task_from, to,
1101 					     task_to, we->backtrace);
1102 		we = we->next;
1103 
1104 		free(task_from);
1105 		free(task_to);
1106 	}
1107 }
1108 
1109 static void draw_cpu_usage(struct timechart *tchart)
1110 {
1111 	struct per_pid *p;
1112 	struct per_pidcomm *c;
1113 	struct cpu_sample *sample;
1114 	p = tchart->all_data;
1115 	while (p) {
1116 		c = p->all;
1117 		while (c) {
1118 			sample = c->samples;
1119 			while (sample) {
1120 				if (sample->type == TYPE_RUNNING) {
1121 					svg_process(sample->cpu,
1122 						    sample->start_time,
1123 						    sample->end_time,
1124 						    p->pid,
1125 						    c->comm,
1126 						    sample->backtrace);
1127 				}
1128 
1129 				sample = sample->next;
1130 			}
1131 			c = c->next;
1132 		}
1133 		p = p->next;
1134 	}
1135 }
1136 
1137 static void draw_io_bars(struct timechart *tchart)
1138 {
1139 	const char *suf;
1140 	double bytes;
1141 	char comm[256];
1142 	struct per_pid *p;
1143 	struct per_pidcomm *c;
1144 	struct io_sample *sample;
1145 	int Y = 1;
1146 
1147 	p = tchart->all_data;
1148 	while (p) {
1149 		c = p->all;
1150 		while (c) {
1151 			if (!c->display) {
1152 				c->Y = 0;
1153 				c = c->next;
1154 				continue;
1155 			}
1156 
1157 			svg_box(Y, c->start_time, c->end_time, "process3");
1158 			sample = c->io_samples;
1159 			for (sample = c->io_samples; sample; sample = sample->next) {
1160 				double h = (double)sample->bytes / c->max_bytes;
1161 
1162 				if (tchart->skip_eagain &&
1163 				    sample->err == -EAGAIN)
1164 					continue;
1165 
1166 				if (sample->err)
1167 					h = 1;
1168 
1169 				if (sample->type == IOTYPE_SYNC)
1170 					svg_fbox(Y,
1171 						sample->start_time,
1172 						sample->end_time,
1173 						1,
1174 						sample->err ? "error" : "sync",
1175 						sample->fd,
1176 						sample->err,
1177 						sample->merges);
1178 				else if (sample->type == IOTYPE_POLL)
1179 					svg_fbox(Y,
1180 						sample->start_time,
1181 						sample->end_time,
1182 						1,
1183 						sample->err ? "error" : "poll",
1184 						sample->fd,
1185 						sample->err,
1186 						sample->merges);
1187 				else if (sample->type == IOTYPE_READ)
1188 					svg_ubox(Y,
1189 						sample->start_time,
1190 						sample->end_time,
1191 						h,
1192 						sample->err ? "error" : "disk",
1193 						sample->fd,
1194 						sample->err,
1195 						sample->merges);
1196 				else if (sample->type == IOTYPE_WRITE)
1197 					svg_lbox(Y,
1198 						sample->start_time,
1199 						sample->end_time,
1200 						h,
1201 						sample->err ? "error" : "disk",
1202 						sample->fd,
1203 						sample->err,
1204 						sample->merges);
1205 				else if (sample->type == IOTYPE_RX)
1206 					svg_ubox(Y,
1207 						sample->start_time,
1208 						sample->end_time,
1209 						h,
1210 						sample->err ? "error" : "net",
1211 						sample->fd,
1212 						sample->err,
1213 						sample->merges);
1214 				else if (sample->type == IOTYPE_TX)
1215 					svg_lbox(Y,
1216 						sample->start_time,
1217 						sample->end_time,
1218 						h,
1219 						sample->err ? "error" : "net",
1220 						sample->fd,
1221 						sample->err,
1222 						sample->merges);
1223 			}
1224 
1225 			suf = "";
1226 			bytes = c->total_bytes;
1227 			if (bytes > 1024) {
1228 				bytes = bytes / 1024;
1229 				suf = "K";
1230 			}
1231 			if (bytes > 1024) {
1232 				bytes = bytes / 1024;
1233 				suf = "M";
1234 			}
1235 			if (bytes > 1024) {
1236 				bytes = bytes / 1024;
1237 				suf = "G";
1238 			}
1239 
1240 
1241 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1242 			svg_text(Y, c->start_time, comm);
1243 
1244 			c->Y = Y;
1245 			Y++;
1246 			c = c->next;
1247 		}
1248 		p = p->next;
1249 	}
1250 }
1251 
1252 static void draw_process_bars(struct timechart *tchart)
1253 {
1254 	struct per_pid *p;
1255 	struct per_pidcomm *c;
1256 	struct cpu_sample *sample;
1257 	int Y = 0;
1258 
1259 	Y = 2 * tchart->numcpus + 2;
1260 
1261 	p = tchart->all_data;
1262 	while (p) {
1263 		c = p->all;
1264 		while (c) {
1265 			if (!c->display) {
1266 				c->Y = 0;
1267 				c = c->next;
1268 				continue;
1269 			}
1270 
1271 			svg_box(Y, c->start_time, c->end_time, "process");
1272 			sample = c->samples;
1273 			while (sample) {
1274 				if (sample->type == TYPE_RUNNING)
1275 					svg_running(Y, sample->cpu,
1276 						    sample->start_time,
1277 						    sample->end_time,
1278 						    sample->backtrace);
1279 				if (sample->type == TYPE_BLOCKED)
1280 					svg_blocked(Y, sample->cpu,
1281 						    sample->start_time,
1282 						    sample->end_time,
1283 						    sample->backtrace);
1284 				if (sample->type == TYPE_WAITING)
1285 					svg_waiting(Y, sample->cpu,
1286 						    sample->start_time,
1287 						    sample->end_time,
1288 						    sample->backtrace);
1289 				sample = sample->next;
1290 			}
1291 
1292 			if (c->comm) {
1293 				char comm[256];
1294 				if (c->total_time > 5000000000) /* 5 seconds */
1295 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1296 				else
1297 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1298 
1299 				svg_text(Y, c->start_time, comm);
1300 			}
1301 			c->Y = Y;
1302 			Y++;
1303 			c = c->next;
1304 		}
1305 		p = p->next;
1306 	}
1307 }
1308 
1309 static void add_process_filter(const char *string)
1310 {
1311 	int pid = strtoull(string, NULL, 10);
1312 	struct process_filter *filt = malloc(sizeof(*filt));
1313 
1314 	if (!filt)
1315 		return;
1316 
1317 	filt->name = strdup(string);
1318 	filt->pid  = pid;
1319 	filt->next = process_filter;
1320 
1321 	process_filter = filt;
1322 }
1323 
1324 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1325 {
1326 	struct process_filter *filt;
1327 	if (!process_filter)
1328 		return 1;
1329 
1330 	filt = process_filter;
1331 	while (filt) {
1332 		if (filt->pid && p->pid == filt->pid)
1333 			return 1;
1334 		if (strcmp(filt->name, c->comm) == 0)
1335 			return 1;
1336 		filt = filt->next;
1337 	}
1338 	return 0;
1339 }
1340 
1341 static int determine_display_tasks_filtered(struct timechart *tchart)
1342 {
1343 	struct per_pid *p;
1344 	struct per_pidcomm *c;
1345 	int count = 0;
1346 
1347 	p = tchart->all_data;
1348 	while (p) {
1349 		p->display = 0;
1350 		if (p->start_time == 1)
1351 			p->start_time = tchart->first_time;
1352 
1353 		/* no exit marker, task kept running to the end */
1354 		if (p->end_time == 0)
1355 			p->end_time = tchart->last_time;
1356 
1357 		c = p->all;
1358 
1359 		while (c) {
1360 			c->display = 0;
1361 
1362 			if (c->start_time == 1)
1363 				c->start_time = tchart->first_time;
1364 
1365 			if (passes_filter(p, c)) {
1366 				c->display = 1;
1367 				p->display = 1;
1368 				count++;
1369 			}
1370 
1371 			if (c->end_time == 0)
1372 				c->end_time = tchart->last_time;
1373 
1374 			c = c->next;
1375 		}
1376 		p = p->next;
1377 	}
1378 	return count;
1379 }
1380 
1381 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1382 {
1383 	struct per_pid *p;
1384 	struct per_pidcomm *c;
1385 	int count = 0;
1386 
1387 	p = tchart->all_data;
1388 	while (p) {
1389 		p->display = 0;
1390 		if (p->start_time == 1)
1391 			p->start_time = tchart->first_time;
1392 
1393 		/* no exit marker, task kept running to the end */
1394 		if (p->end_time == 0)
1395 			p->end_time = tchart->last_time;
1396 		if (p->total_time >= threshold)
1397 			p->display = 1;
1398 
1399 		c = p->all;
1400 
1401 		while (c) {
1402 			c->display = 0;
1403 
1404 			if (c->start_time == 1)
1405 				c->start_time = tchart->first_time;
1406 
1407 			if (c->total_time >= threshold) {
1408 				c->display = 1;
1409 				count++;
1410 			}
1411 
1412 			if (c->end_time == 0)
1413 				c->end_time = tchart->last_time;
1414 
1415 			c = c->next;
1416 		}
1417 		p = p->next;
1418 	}
1419 	return count;
1420 }
1421 
1422 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1423 {
1424 	struct per_pid *p;
1425 	struct per_pidcomm *c;
1426 	int count = 0;
1427 
1428 	p = timechart->all_data;
1429 	while (p) {
1430 		/* no exit marker, task kept running to the end */
1431 		if (p->end_time == 0)
1432 			p->end_time = timechart->last_time;
1433 
1434 		c = p->all;
1435 
1436 		while (c) {
1437 			c->display = 0;
1438 
1439 			if (c->total_bytes >= threshold) {
1440 				c->display = 1;
1441 				count++;
1442 			}
1443 
1444 			if (c->end_time == 0)
1445 				c->end_time = timechart->last_time;
1446 
1447 			c = c->next;
1448 		}
1449 		p = p->next;
1450 	}
1451 	return count;
1452 }
1453 
1454 #define BYTES_THRESH (1 * 1024 * 1024)
1455 #define TIME_THRESH 10000000
1456 
1457 static void write_svg_file(struct timechart *tchart, const char *filename)
1458 {
1459 	u64 i;
1460 	int count;
1461 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1462 
1463 	if (tchart->power_only)
1464 		tchart->proc_num = 0;
1465 
1466 	/* We'd like to show at least proc_num tasks;
1467 	 * be less picky if we have fewer */
1468 	do {
1469 		if (process_filter)
1470 			count = determine_display_tasks_filtered(tchart);
1471 		else if (tchart->io_events)
1472 			count = determine_display_io_tasks(tchart, thresh);
1473 		else
1474 			count = determine_display_tasks(tchart, thresh);
1475 		thresh /= 10;
1476 	} while (!process_filter && thresh && count < tchart->proc_num);
1477 
1478 	if (!tchart->proc_num)
1479 		count = 0;
1480 
1481 	if (tchart->io_events) {
1482 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1483 
1484 		svg_time_grid(0.5);
1485 		svg_io_legenda();
1486 
1487 		draw_io_bars(tchart);
1488 	} else {
1489 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1490 
1491 		svg_time_grid(0);
1492 
1493 		svg_legenda();
1494 
1495 		for (i = 0; i < tchart->numcpus; i++)
1496 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1497 
1498 		draw_cpu_usage(tchart);
1499 		if (tchart->proc_num)
1500 			draw_process_bars(tchart);
1501 		if (!tchart->tasks_only)
1502 			draw_c_p_states(tchart);
1503 		if (tchart->proc_num)
1504 			draw_wakeups(tchart);
1505 	}
1506 
1507 	svg_close();
1508 }
1509 
1510 static int process_header(struct perf_file_section *section __maybe_unused,
1511 			  struct perf_header *ph,
1512 			  int feat,
1513 			  int fd __maybe_unused,
1514 			  void *data)
1515 {
1516 	struct timechart *tchart = data;
1517 
1518 	switch (feat) {
1519 	case HEADER_NRCPUS:
1520 		tchart->numcpus = ph->env.nr_cpus_avail;
1521 		break;
1522 
1523 	case HEADER_CPU_TOPOLOGY:
1524 		if (!tchart->topology)
1525 			break;
1526 
1527 		if (svg_build_topology_map(ph->env.sibling_cores,
1528 					   ph->env.nr_sibling_cores,
1529 					   ph->env.sibling_threads,
1530 					   ph->env.nr_sibling_threads))
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 perf_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, false,
1611 							 &tchart->tool);
1612 	int ret = -EINVAL;
1613 
1614 	if (session == NULL)
1615 		return -1;
1616 
1617 	symbol__init(&session->header.env);
1618 
1619 	(void)perf_header__process_sections(&session->header,
1620 					    perf_data__fd(session->data),
1621 					    tchart,
1622 					    process_header);
1623 
1624 	if (!perf_session__has_traces(session, "timechart record"))
1625 		goto out_delete;
1626 
1627 	if (perf_session__set_tracepoints_handlers(session,
1628 						   power_tracepoints)) {
1629 		pr_err("Initializing session tracepoint handlers failed\n");
1630 		goto out_delete;
1631 	}
1632 
1633 	ret = perf_session__process_events(session);
1634 	if (ret)
1635 		goto out_delete;
1636 
1637 	end_sample_processing(tchart);
1638 
1639 	sort_pids(tchart);
1640 
1641 	write_svg_file(tchart, output_name);
1642 
1643 	pr_info("Written %2.1f seconds of trace to %s.\n",
1644 		(tchart->last_time - tchart->first_time) / (double)NSEC_PER_SEC, output_name);
1645 out_delete:
1646 	perf_session__delete(session);
1647 	return ret;
1648 }
1649 
1650 static int timechart__io_record(int argc, const char **argv)
1651 {
1652 	unsigned int rec_argc, i;
1653 	const char **rec_argv;
1654 	const char **p;
1655 	char *filter = NULL;
1656 
1657 	const char * const common_args[] = {
1658 		"record", "-a", "-R", "-c", "1",
1659 	};
1660 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1661 
1662 	const char * const disk_events[] = {
1663 		"syscalls:sys_enter_read",
1664 		"syscalls:sys_enter_pread64",
1665 		"syscalls:sys_enter_readv",
1666 		"syscalls:sys_enter_preadv",
1667 		"syscalls:sys_enter_write",
1668 		"syscalls:sys_enter_pwrite64",
1669 		"syscalls:sys_enter_writev",
1670 		"syscalls:sys_enter_pwritev",
1671 		"syscalls:sys_enter_sync",
1672 		"syscalls:sys_enter_sync_file_range",
1673 		"syscalls:sys_enter_fsync",
1674 		"syscalls:sys_enter_msync",
1675 
1676 		"syscalls:sys_exit_read",
1677 		"syscalls:sys_exit_pread64",
1678 		"syscalls:sys_exit_readv",
1679 		"syscalls:sys_exit_preadv",
1680 		"syscalls:sys_exit_write",
1681 		"syscalls:sys_exit_pwrite64",
1682 		"syscalls:sys_exit_writev",
1683 		"syscalls:sys_exit_pwritev",
1684 		"syscalls:sys_exit_sync",
1685 		"syscalls:sys_exit_sync_file_range",
1686 		"syscalls:sys_exit_fsync",
1687 		"syscalls:sys_exit_msync",
1688 	};
1689 	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1690 
1691 	const char * const net_events[] = {
1692 		"syscalls:sys_enter_recvfrom",
1693 		"syscalls:sys_enter_recvmmsg",
1694 		"syscalls:sys_enter_recvmsg",
1695 		"syscalls:sys_enter_sendto",
1696 		"syscalls:sys_enter_sendmsg",
1697 		"syscalls:sys_enter_sendmmsg",
1698 
1699 		"syscalls:sys_exit_recvfrom",
1700 		"syscalls:sys_exit_recvmmsg",
1701 		"syscalls:sys_exit_recvmsg",
1702 		"syscalls:sys_exit_sendto",
1703 		"syscalls:sys_exit_sendmsg",
1704 		"syscalls:sys_exit_sendmmsg",
1705 	};
1706 	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1707 
1708 	const char * const poll_events[] = {
1709 		"syscalls:sys_enter_epoll_pwait",
1710 		"syscalls:sys_enter_epoll_wait",
1711 		"syscalls:sys_enter_poll",
1712 		"syscalls:sys_enter_ppoll",
1713 		"syscalls:sys_enter_pselect6",
1714 		"syscalls:sys_enter_select",
1715 
1716 		"syscalls:sys_exit_epoll_pwait",
1717 		"syscalls:sys_exit_epoll_wait",
1718 		"syscalls:sys_exit_poll",
1719 		"syscalls:sys_exit_ppoll",
1720 		"syscalls:sys_exit_pselect6",
1721 		"syscalls:sys_exit_select",
1722 	};
1723 	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1724 
1725 	rec_argc = common_args_nr +
1726 		disk_events_nr * 4 +
1727 		net_events_nr * 4 +
1728 		poll_events_nr * 4 +
1729 		argc;
1730 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1731 
1732 	if (rec_argv == NULL)
1733 		return -ENOMEM;
1734 
1735 	if (asprintf(&filter, "common_pid != %d", getpid()) < 0) {
1736 		free(rec_argv);
1737 		return -ENOMEM;
1738 	}
1739 
1740 	p = rec_argv;
1741 	for (i = 0; i < common_args_nr; i++)
1742 		*p++ = strdup(common_args[i]);
1743 
1744 	for (i = 0; i < disk_events_nr; i++) {
1745 		if (!is_valid_tracepoint(disk_events[i])) {
1746 			rec_argc -= 4;
1747 			continue;
1748 		}
1749 
1750 		*p++ = "-e";
1751 		*p++ = strdup(disk_events[i]);
1752 		*p++ = "--filter";
1753 		*p++ = filter;
1754 	}
1755 	for (i = 0; i < net_events_nr; i++) {
1756 		if (!is_valid_tracepoint(net_events[i])) {
1757 			rec_argc -= 4;
1758 			continue;
1759 		}
1760 
1761 		*p++ = "-e";
1762 		*p++ = strdup(net_events[i]);
1763 		*p++ = "--filter";
1764 		*p++ = filter;
1765 	}
1766 	for (i = 0; i < poll_events_nr; i++) {
1767 		if (!is_valid_tracepoint(poll_events[i])) {
1768 			rec_argc -= 4;
1769 			continue;
1770 		}
1771 
1772 		*p++ = "-e";
1773 		*p++ = strdup(poll_events[i]);
1774 		*p++ = "--filter";
1775 		*p++ = filter;
1776 	}
1777 
1778 	for (i = 0; i < (unsigned int)argc; i++)
1779 		*p++ = argv[i];
1780 
1781 	return cmd_record(rec_argc, rec_argv);
1782 }
1783 
1784 
1785 static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1786 {
1787 	unsigned int rec_argc, i, j;
1788 	const char **rec_argv;
1789 	const char **p;
1790 	unsigned int record_elems;
1791 
1792 	const char * const common_args[] = {
1793 		"record", "-a", "-R", "-c", "1",
1794 	};
1795 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1796 
1797 	const char * const backtrace_args[] = {
1798 		"-g",
1799 	};
1800 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1801 
1802 	const char * const power_args[] = {
1803 		"-e", "power:cpu_frequency",
1804 		"-e", "power:cpu_idle",
1805 	};
1806 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1807 
1808 	const char * const old_power_args[] = {
1809 #ifdef SUPPORT_OLD_POWER_EVENTS
1810 		"-e", "power:power_start",
1811 		"-e", "power:power_end",
1812 		"-e", "power:power_frequency",
1813 #endif
1814 	};
1815 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1816 
1817 	const char * const tasks_args[] = {
1818 		"-e", "sched:sched_wakeup",
1819 		"-e", "sched:sched_switch",
1820 	};
1821 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1822 
1823 #ifdef SUPPORT_OLD_POWER_EVENTS
1824 	if (!is_valid_tracepoint("power:cpu_idle") &&
1825 	    is_valid_tracepoint("power:power_start")) {
1826 		use_old_power_events = 1;
1827 		power_args_nr = 0;
1828 	} else {
1829 		old_power_args_nr = 0;
1830 	}
1831 #endif
1832 
1833 	if (tchart->power_only)
1834 		tasks_args_nr = 0;
1835 
1836 	if (tchart->tasks_only) {
1837 		power_args_nr = 0;
1838 		old_power_args_nr = 0;
1839 	}
1840 
1841 	if (!tchart->with_backtrace)
1842 		backtrace_args_no = 0;
1843 
1844 	record_elems = common_args_nr + tasks_args_nr +
1845 		power_args_nr + old_power_args_nr + backtrace_args_no;
1846 
1847 	rec_argc = record_elems + argc;
1848 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1849 
1850 	if (rec_argv == NULL)
1851 		return -ENOMEM;
1852 
1853 	p = rec_argv;
1854 	for (i = 0; i < common_args_nr; i++)
1855 		*p++ = strdup(common_args[i]);
1856 
1857 	for (i = 0; i < backtrace_args_no; i++)
1858 		*p++ = strdup(backtrace_args[i]);
1859 
1860 	for (i = 0; i < tasks_args_nr; i++)
1861 		*p++ = strdup(tasks_args[i]);
1862 
1863 	for (i = 0; i < power_args_nr; i++)
1864 		*p++ = strdup(power_args[i]);
1865 
1866 	for (i = 0; i < old_power_args_nr; i++)
1867 		*p++ = strdup(old_power_args[i]);
1868 
1869 	for (j = 0; j < (unsigned int)argc; j++)
1870 		*p++ = argv[j];
1871 
1872 	return cmd_record(rec_argc, rec_argv);
1873 }
1874 
1875 static int
1876 parse_process(const struct option *opt __maybe_unused, const char *arg,
1877 	      int __maybe_unused unset)
1878 {
1879 	if (arg)
1880 		add_process_filter(arg);
1881 	return 0;
1882 }
1883 
1884 static int
1885 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1886 		int __maybe_unused unset)
1887 {
1888 	unsigned long duration = strtoul(arg, NULL, 0);
1889 
1890 	if (svg_highlight || svg_highlight_name)
1891 		return -1;
1892 
1893 	if (duration)
1894 		svg_highlight = duration;
1895 	else
1896 		svg_highlight_name = strdup(arg);
1897 
1898 	return 0;
1899 }
1900 
1901 static int
1902 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1903 {
1904 	char unit = 'n';
1905 	u64 *value = opt->value;
1906 
1907 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1908 		switch (unit) {
1909 		case 'm':
1910 			*value *= NSEC_PER_MSEC;
1911 			break;
1912 		case 'u':
1913 			*value *= NSEC_PER_USEC;
1914 			break;
1915 		case 'n':
1916 			break;
1917 		default:
1918 			return -1;
1919 		}
1920 	}
1921 
1922 	return 0;
1923 }
1924 
1925 int cmd_timechart(int argc, const char **argv)
1926 {
1927 	struct timechart tchart = {
1928 		.tool = {
1929 			.comm		 = process_comm_event,
1930 			.fork		 = process_fork_event,
1931 			.exit		 = process_exit_event,
1932 			.sample		 = process_sample_event,
1933 			.ordered_events	 = true,
1934 		},
1935 		.proc_num = 15,
1936 		.min_time = NSEC_PER_MSEC,
1937 		.merge_dist = 1000,
1938 	};
1939 	const char *output_name = "output.svg";
1940 	const struct option timechart_common_options[] = {
1941 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1942 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only, "output processes data only"),
1943 	OPT_END()
1944 	};
1945 	const struct option timechart_options[] = {
1946 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1947 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1948 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1949 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1950 		      "highlight tasks. Pass duration in ns or process name.",
1951 		       parse_highlight),
1952 	OPT_CALLBACK('p', "process", NULL, "process",
1953 		      "process selector. Pass a pid or process name.",
1954 		       parse_process),
1955 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1956 		     "Look for files with symbols relative to this directory",
1957 		     symbol__config_symfs),
1958 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1959 		    "min. number of tasks to print"),
1960 	OPT_BOOLEAN('t', "topology", &tchart.topology,
1961 		    "sort CPUs according to topology"),
1962 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1963 		    "skip EAGAIN errors"),
1964 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1965 		     "all IO faster than min-time will visually appear longer",
1966 		     parse_time),
1967 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1968 		     "merge events that are merge-dist us apart",
1969 		     parse_time),
1970 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1971 	OPT_PARENT(timechart_common_options),
1972 	};
1973 	const char * const timechart_subcommands[] = { "record", NULL };
1974 	const char *timechart_usage[] = {
1975 		"perf timechart [<options>] {record}",
1976 		NULL
1977 	};
1978 	const struct option timechart_record_options[] = {
1979 	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
1980 		    "record only IO data"),
1981 	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
1982 	OPT_PARENT(timechart_common_options),
1983 	};
1984 	const char * const timechart_record_usage[] = {
1985 		"perf timechart record [<options>]",
1986 		NULL
1987 	};
1988 	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
1989 			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1990 
1991 	if (tchart.power_only && tchart.tasks_only) {
1992 		pr_err("-P and -T options cannot be used at the same time.\n");
1993 		return -1;
1994 	}
1995 
1996 	if (argc && !strncmp(argv[0], "rec", 3)) {
1997 		argc = parse_options(argc, argv, timechart_record_options,
1998 				     timechart_record_usage,
1999 				     PARSE_OPT_STOP_AT_NON_OPTION);
2000 
2001 		if (tchart.power_only && tchart.tasks_only) {
2002 			pr_err("-P and -T options cannot be used at the same time.\n");
2003 			return -1;
2004 		}
2005 
2006 		if (tchart.io_only)
2007 			return timechart__io_record(argc, argv);
2008 		else
2009 			return timechart__record(&tchart, argc, argv);
2010 	} else if (argc)
2011 		usage_with_options(timechart_usage, timechart_options);
2012 
2013 	setup_pager();
2014 
2015 	return __cmd_timechart(&tchart, output_name);
2016 }
2017