xref: /linux/tools/perf/builtin-timechart.c (revision a77ecea7ced2fef7cc0a8ad0323542f781ad9788)
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 <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 ? strdup(backtrace) : NULL;
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(const 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(const 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(const 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 ? strdup(backtrace) : NULL;
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 /*
493  * Returns a malloc'd backtrace string built via open_memstream, or NULL
494  * on error.  Caller must free() the returned pointer.
495  */
496 static char *cat_backtrace(union perf_event *event,
497 			   struct perf_sample *sample,
498 			   struct machine *machine)
499 {
500 	struct addr_location al;
501 	unsigned int i;
502 	char *p = NULL;
503 	size_t p_len;
504 	u8 cpumode = PERF_RECORD_MISC_USER;
505 	struct ip_callchain *chain = sample->callchain;
506 	FILE *f = open_memstream(&p, &p_len);
507 	bool corrupted = false;
508 
509 	if (!f) {
510 		perror("open_memstream error");
511 		return NULL;
512 	}
513 
514 	addr_location__init(&al);
515 	if (!chain)
516 		goto exit;
517 
518 	if (machine__resolve(machine, &al, sample) < 0) {
519 		pr_err("problem processing %s (%u) event at offset %#" PRIx64 ", skipping it.\n",
520 		       perf_event__name(event->header.type), event->header.type,
521 		       sample->file_offset);
522 		goto exit;
523 	}
524 
525 	for (i = 0; i < chain->nr; i++) {
526 		u64 ip;
527 		struct addr_location tal;
528 
529 		if (callchain_param.order == ORDER_CALLEE)
530 			ip = chain->ips[i];
531 		else
532 			ip = chain->ips[chain->nr - i - 1];
533 
534 		if (ip >= PERF_CONTEXT_MAX) {
535 			switch (ip) {
536 			case PERF_CONTEXT_HV:
537 				cpumode = PERF_RECORD_MISC_HYPERVISOR;
538 				break;
539 			case PERF_CONTEXT_KERNEL:
540 				cpumode = PERF_RECORD_MISC_KERNEL;
541 				break;
542 			case PERF_CONTEXT_USER:
543 				cpumode = PERF_RECORD_MISC_USER;
544 				break;
545 			default:
546 				pr_debug("invalid callchain context: %" PRId64 "\n", (s64) ip);
547 				corrupted = true;
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 	/*
565 	 * fclose() on an open_memstream always sets p to a valid buffer,
566 	 * even if nothing was written — see open_memstream(3).  So p is
567 	 * never NULL after fclose and we need the flag to discard it.
568 	 */
569 	fclose(f);
570 	if (corrupted)
571 		zfree(&p);
572 
573 	return p;
574 }
575 
576 typedef int (*tracepoint_handler)(struct timechart *tchart,
577 				  struct perf_sample *sample,
578 				  const char *backtrace);
579 
580 static int process_sample_event(const struct perf_tool *tool,
581 				union perf_event *event,
582 				struct perf_sample *sample,
583 				struct machine *machine)
584 {
585 	struct timechart *tchart = container_of(tool, struct timechart, tool);
586 	struct evsel *evsel = sample->evsel;
587 	int ret = 0;
588 
589 	if (evsel->core.attr.sample_type & PERF_SAMPLE_TIME) {
590 		if (!tchart->first_time || tchart->first_time > sample->time)
591 			tchart->first_time = sample->time;
592 		if (tchart->last_time < sample->time)
593 			tchart->last_time = sample->time;
594 	}
595 
596 	if (evsel->handler != NULL) {
597 		tracepoint_handler f = evsel->handler;
598 		char *backtrace = cat_backtrace(event, sample, machine);
599 
600 		ret = f(tchart, sample, backtrace);
601 		free(backtrace);
602 	}
603 
604 	return ret;
605 }
606 
607 static int
608 process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
609 			struct perf_sample *sample,
610 			const char *backtrace __maybe_unused)
611 {
612 	u32 state  = perf_sample__intval(sample, "state");
613 	u32 cpu_id = perf_sample__intval(sample, "cpu_id");
614 
615 	/* perf.data is untrusted input — cpu_id may be corrupted */
616 	if (cpu_id >= MAX_CPUS) {
617 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu_id %u\n",
618 			 sample->file_offset, cpu_id);
619 		return -1;
620 	}
621 	if (state == (u32)PWR_EVENT_EXIT)
622 		c_state_end(tchart, cpu_id, sample->time);
623 	else
624 		c_state_start(cpu_id, sample->time, state);
625 	return 0;
626 }
627 
628 static int
629 process_sample_cpu_frequency(struct timechart *tchart,
630 			     struct perf_sample *sample,
631 			     const char *backtrace __maybe_unused)
632 {
633 	u32 state  = perf_sample__intval(sample, "state");
634 	u32 cpu_id = perf_sample__intval(sample, "cpu_id");
635 
636 	/* perf.data is untrusted input — cpu_id may be corrupted */
637 	if (cpu_id >= MAX_CPUS) {
638 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu_id %u\n",
639 			 sample->file_offset, cpu_id);
640 		return -1;
641 	}
642 	p_state_change(tchart, cpu_id, sample->time, state);
643 	return 0;
644 }
645 
646 static int
647 process_sample_sched_wakeup(struct timechart *tchart,
648 			    struct perf_sample *sample,
649 			    const char *backtrace)
650 {
651 	u8 flags  = perf_sample__intval(sample, "common_flags");
652 	int waker = perf_sample__intval(sample, "common_pid");
653 	int wakee = perf_sample__intval(sample, "pid");
654 
655 	/* perf.data is untrusted input — CPU may be absent or corrupted */
656 	if (sample->cpu >= MAX_CPUS) {
657 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu %u\n",
658 			 sample->file_offset, sample->cpu);
659 		return -1;
660 	}
661 	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
662 	return 0;
663 }
664 
665 static int
666 process_sample_sched_switch(struct timechart *tchart,
667 			    struct perf_sample *sample,
668 			    const char *backtrace)
669 {
670 	int prev_pid   = perf_sample__intval(sample, "prev_pid");
671 	int next_pid   = perf_sample__intval(sample, "next_pid");
672 	u64 prev_state = perf_sample__intval(sample, "prev_state");
673 
674 	/* perf.data is untrusted input — CPU may be absent or corrupted */
675 	if (sample->cpu >= MAX_CPUS) {
676 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu %u\n",
677 			 sample->file_offset, sample->cpu);
678 		return -1;
679 	}
680 	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
681 		     prev_state, backtrace);
682 	return 0;
683 }
684 
685 #ifdef SUPPORT_OLD_POWER_EVENTS
686 static int
687 process_sample_power_start(struct timechart *tchart __maybe_unused,
688 			   struct perf_sample *sample,
689 			   const char *backtrace __maybe_unused)
690 {
691 	u64 cpu_id = perf_sample__intval(sample, "cpu_id");
692 	u64 value  = perf_sample__intval(sample, "value");
693 
694 	/* perf.data is untrusted input — cpu_id may be corrupted */
695 	if (cpu_id >= MAX_CPUS) {
696 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu_id %llu\n",
697 			 sample->file_offset, (unsigned long long)cpu_id);
698 		return -1;
699 	}
700 	c_state_start(cpu_id, sample->time, value);
701 	return 0;
702 }
703 
704 static int
705 process_sample_power_end(struct timechart *tchart,
706 			 struct perf_sample *sample,
707 			 const char *backtrace __maybe_unused)
708 {
709 	/* perf.data is untrusted input — CPU may be absent or corrupted */
710 	if (sample->cpu >= MAX_CPUS) {
711 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu %u\n",
712 			 sample->file_offset, sample->cpu);
713 		return -1;
714 	}
715 	c_state_end(tchart, sample->cpu, sample->time);
716 	return 0;
717 }
718 
719 static int
720 process_sample_power_frequency(struct timechart *tchart,
721 			       struct perf_sample *sample,
722 			       const char *backtrace __maybe_unused)
723 {
724 	u64 cpu_id = perf_sample__intval(sample, "cpu_id");
725 	u64 value  = perf_sample__intval(sample, "value");
726 
727 	/* perf.data is untrusted input — cpu_id may be corrupted */
728 	if (cpu_id >= MAX_CPUS) {
729 		pr_debug("at offset %#" PRIx64 ": out-of-bounds cpu_id %llu\n",
730 			 sample->file_offset, (unsigned long long)cpu_id);
731 		return -1;
732 	}
733 	p_state_change(tchart, cpu_id, sample->time, value);
734 	return 0;
735 }
736 #endif /* SUPPORT_OLD_POWER_EVENTS */
737 
738 /*
739  * After the last sample we need to wrap up the current C/P state
740  * and close out each CPU for these.
741  */
742 static void end_sample_processing(struct timechart *tchart)
743 {
744 	for (u64 cpu = 0; cpu < tchart->numcpus; cpu++) {
745 		struct power_event *pwr;
746 
747 		/* C state */
748 #if 0
749 		pwr = zalloc(sizeof(*pwr));
750 		if (!pwr)
751 			return;
752 
753 		pwr->state = cpus_cstate_state[cpu];
754 		pwr->start_time = cpus_cstate_start_times[cpu];
755 		pwr->end_time = tchart->last_time;
756 		pwr->cpu = cpu;
757 		pwr->type = CSTATE;
758 		pwr->next = tchart->power_events;
759 
760 		tchart->power_events = pwr;
761 #endif
762 		/* P state */
763 
764 		pwr = p_state_end(tchart, cpu, tchart->last_time);
765 		if (!pwr)
766 			return;
767 
768 		if (!pwr->state)
769 			pwr->state = tchart->min_freq;
770 	}
771 }
772 
773 static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
774 			       u64 start, int fd)
775 {
776 	struct per_pid *p = find_create_pid(tchart, pid);
777 	struct per_pidcomm *c = p->current;
778 	struct io_sample *sample;
779 	struct io_sample *prev;
780 
781 	if (!c) {
782 		c = create_pidcomm(p);
783 		if (!c)
784 			return -ENOMEM;
785 	}
786 
787 	prev = c->io_samples;
788 
789 	if (prev && prev->start_time && !prev->end_time) {
790 		pr_warning("Skip invalid start event: "
791 			   "previous event already started!\n");
792 
793 		/* remove previous event that has been started,
794 		 * we are not sure we will ever get an end for it */
795 		c->io_samples = prev->next;
796 		free(prev);
797 		return 0;
798 	}
799 
800 	sample = zalloc(sizeof(*sample));
801 	if (!sample)
802 		return -ENOMEM;
803 	sample->start_time = start;
804 	sample->type = type;
805 	sample->fd = fd;
806 	sample->next = c->io_samples;
807 	c->io_samples = sample;
808 
809 	if (c->start_time == 0 || c->start_time > start)
810 		c->start_time = start;
811 
812 	return 0;
813 }
814 
815 static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
816 			     u64 end, long ret)
817 {
818 	struct per_pid *p = find_create_pid(tchart, pid);
819 	struct per_pidcomm *c = p->current;
820 	struct io_sample *sample, *prev;
821 
822 	if (!c) {
823 		pr_warning("Invalid pidcomm!\n");
824 		return -1;
825 	}
826 
827 	sample = c->io_samples;
828 
829 	if (!sample) /* skip partially captured events */
830 		return 0;
831 
832 	if (sample->end_time) {
833 		pr_warning("Skip invalid end event: "
834 			   "previous event already ended!\n");
835 		return 0;
836 	}
837 
838 	if (sample->type != type) {
839 		pr_warning("Skip invalid end event: invalid event type!\n");
840 		return 0;
841 	}
842 
843 	sample->end_time = end;
844 	prev = sample->next;
845 
846 	/* we want to be able to see small and fast transfers, so make them
847 	 * at least min_time long, but don't overlap them */
848 	if (sample->end_time - sample->start_time < tchart->min_time)
849 		sample->end_time = sample->start_time + tchart->min_time;
850 	if (prev && sample->start_time < prev->end_time) {
851 		if (prev->err) /* try to make errors more visible */
852 			sample->start_time = prev->end_time;
853 		else
854 			prev->end_time = sample->start_time;
855 	}
856 
857 	if (ret < 0) {
858 		sample->err = ret;
859 	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
860 		   type == IOTYPE_TX || type == IOTYPE_RX) {
861 
862 		if ((u64)ret > c->max_bytes)
863 			c->max_bytes = ret;
864 
865 		c->total_bytes += ret;
866 		p->total_bytes += ret;
867 		sample->bytes = ret;
868 	}
869 
870 	/* merge two requests to make svg smaller and render-friendly */
871 	if (prev &&
872 	    prev->type == sample->type &&
873 	    prev->err == sample->err &&
874 	    prev->fd == sample->fd &&
875 	    prev->end_time + tchart->merge_dist >= sample->start_time) {
876 
877 		sample->bytes += prev->bytes;
878 		sample->merges += prev->merges + 1;
879 
880 		sample->start_time = prev->start_time;
881 		sample->next = prev->next;
882 		free(prev);
883 
884 		if (!sample->err && sample->bytes > c->max_bytes)
885 			c->max_bytes = sample->bytes;
886 	}
887 
888 	tchart->io_events++;
889 
890 	return 0;
891 }
892 
893 static int
894 process_enter_read(struct timechart *tchart,
895 		   struct perf_sample *sample,
896 		   const char *backtrace __maybe_unused)
897 {
898 	long fd = perf_sample__intval(sample, "fd");
899 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
900 				   sample->time, fd);
901 }
902 
903 static int
904 process_exit_read(struct timechart *tchart,
905 		  struct perf_sample *sample,
906 		  const char *backtrace __maybe_unused)
907 {
908 	long ret = perf_sample__intval(sample, "ret");
909 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
910 				 sample->time, ret);
911 }
912 
913 static int
914 process_enter_write(struct timechart *tchart,
915 		    struct perf_sample *sample,
916 		    const char *backtrace __maybe_unused)
917 {
918 	long fd = perf_sample__intval(sample, "fd");
919 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
920 				   sample->time, fd);
921 }
922 
923 static int
924 process_exit_write(struct timechart *tchart,
925 		   struct perf_sample *sample,
926 		   const char *backtrace __maybe_unused)
927 {
928 	long ret = perf_sample__intval(sample, "ret");
929 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
930 				 sample->time, ret);
931 }
932 
933 static int
934 process_enter_sync(struct timechart *tchart,
935 		   struct perf_sample *sample,
936 		   const char *backtrace __maybe_unused)
937 {
938 	long fd = perf_sample__intval(sample, "fd");
939 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
940 				   sample->time, fd);
941 }
942 
943 static int
944 process_exit_sync(struct timechart *tchart,
945 		  struct perf_sample *sample,
946 		  const char *backtrace __maybe_unused)
947 {
948 	long ret = perf_sample__intval(sample, "ret");
949 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
950 				 sample->time, ret);
951 }
952 
953 static int
954 process_enter_tx(struct timechart *tchart,
955 		 struct perf_sample *sample,
956 		 const char *backtrace __maybe_unused)
957 {
958 	long fd = perf_sample__intval(sample, "fd");
959 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
960 				   sample->time, fd);
961 }
962 
963 static int
964 process_exit_tx(struct timechart *tchart,
965 		struct perf_sample *sample,
966 		const char *backtrace __maybe_unused)
967 {
968 	long ret = perf_sample__intval(sample, "ret");
969 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
970 				 sample->time, ret);
971 }
972 
973 static int
974 process_enter_rx(struct timechart *tchart,
975 		 struct perf_sample *sample,
976 		 const char *backtrace __maybe_unused)
977 {
978 	long fd = perf_sample__intval(sample, "fd");
979 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
980 				   sample->time, fd);
981 }
982 
983 static int
984 process_exit_rx(struct timechart *tchart,
985 		struct perf_sample *sample,
986 		const char *backtrace __maybe_unused)
987 {
988 	long ret = perf_sample__intval(sample, "ret");
989 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
990 				 sample->time, ret);
991 }
992 
993 static int
994 process_enter_poll(struct timechart *tchart,
995 		   struct perf_sample *sample,
996 		   const char *backtrace __maybe_unused)
997 {
998 	long fd = perf_sample__intval(sample, "fd");
999 	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
1000 				   sample->time, fd);
1001 }
1002 
1003 static int
1004 process_exit_poll(struct timechart *tchart,
1005 		  struct perf_sample *sample,
1006 		  const char *backtrace __maybe_unused)
1007 {
1008 	long ret = perf_sample__intval(sample, "ret");
1009 	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
1010 				 sample->time, ret);
1011 }
1012 
1013 /*
1014  * Sort the pid datastructure
1015  */
1016 static void sort_pids(struct timechart *tchart)
1017 {
1018 	struct per_pid *new_list, *p, *cursor, *prev;
1019 	/* sort by ppid first, then by pid, lowest to highest */
1020 
1021 	new_list = NULL;
1022 
1023 	while (tchart->all_data) {
1024 		p = tchart->all_data;
1025 		tchart->all_data = p->next;
1026 		p->next = NULL;
1027 
1028 		if (new_list == NULL) {
1029 			new_list = p;
1030 			p->next = NULL;
1031 			continue;
1032 		}
1033 		prev = NULL;
1034 		cursor = new_list;
1035 		while (cursor) {
1036 			if (cursor->ppid > p->ppid ||
1037 				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
1038 				/* must insert before */
1039 				if (prev) {
1040 					p->next = prev->next;
1041 					prev->next = p;
1042 					cursor = NULL;
1043 					continue;
1044 				} else {
1045 					p->next = new_list;
1046 					new_list = p;
1047 					cursor = NULL;
1048 					continue;
1049 				}
1050 			}
1051 
1052 			prev = cursor;
1053 			cursor = cursor->next;
1054 			if (!cursor)
1055 				prev->next = p;
1056 		}
1057 	}
1058 	tchart->all_data = new_list;
1059 }
1060 
1061 
1062 static void draw_c_p_states(struct timechart *tchart)
1063 {
1064 	struct power_event *pwr;
1065 	pwr = tchart->power_events;
1066 
1067 	/*
1068 	 * two pass drawing so that the P state bars are on top of the C state blocks
1069 	 */
1070 	while (pwr) {
1071 		if (pwr->type == CSTATE)
1072 			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1073 		pwr = pwr->next;
1074 	}
1075 
1076 	pwr = tchart->power_events;
1077 	while (pwr) {
1078 		if (pwr->type == PSTATE) {
1079 			if (!pwr->state)
1080 				pwr->state = tchart->min_freq;
1081 			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1082 		}
1083 		pwr = pwr->next;
1084 	}
1085 }
1086 
1087 static void draw_wakeups(struct timechart *tchart)
1088 {
1089 	struct wake_event *we;
1090 	struct per_pid *p;
1091 	struct per_pidcomm *c;
1092 
1093 	we = tchart->wake_events;
1094 	while (we) {
1095 		int from = 0, to = 0;
1096 		char *task_from = NULL, *task_to = NULL;
1097 
1098 		/* locate the column of the waker and wakee */
1099 		p = tchart->all_data;
1100 		while (p) {
1101 			if (p->pid == we->waker || p->pid == we->wakee) {
1102 				c = p->all;
1103 				while (c) {
1104 					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1105 						if (p->pid == we->waker && !from) {
1106 							from = c->Y;
1107 							task_from = strdup(c->comm);
1108 						}
1109 						if (p->pid == we->wakee && !to) {
1110 							to = c->Y;
1111 							task_to = strdup(c->comm);
1112 						}
1113 					}
1114 					c = c->next;
1115 				}
1116 				c = p->all;
1117 				while (c) {
1118 					if (p->pid == we->waker && !from) {
1119 						from = c->Y;
1120 						task_from = strdup(c->comm);
1121 					}
1122 					if (p->pid == we->wakee && !to) {
1123 						to = c->Y;
1124 						task_to = strdup(c->comm);
1125 					}
1126 					c = c->next;
1127 				}
1128 			}
1129 			p = p->next;
1130 		}
1131 
1132 		if (!task_from) {
1133 			task_from = malloc(40);
1134 			sprintf(task_from, "[%i]", we->waker);
1135 		}
1136 		if (!task_to) {
1137 			task_to = malloc(40);
1138 			sprintf(task_to, "[%i]", we->wakee);
1139 		}
1140 
1141 		if (we->waker == -1)
1142 			svg_interrupt(we->time, to, we->backtrace);
1143 		else if (from && to && abs(from - to) == 1)
1144 			svg_wakeline(we->time, from, to, we->backtrace);
1145 		else
1146 			svg_partial_wakeline(we->time, from, task_from, to,
1147 					     task_to, we->backtrace);
1148 		we = we->next;
1149 
1150 		free(task_from);
1151 		free(task_to);
1152 	}
1153 }
1154 
1155 static void draw_cpu_usage(struct timechart *tchart)
1156 {
1157 	struct per_pid *p;
1158 	struct per_pidcomm *c;
1159 	struct cpu_sample *sample;
1160 	p = tchart->all_data;
1161 	while (p) {
1162 		c = p->all;
1163 		while (c) {
1164 			sample = c->samples;
1165 			while (sample) {
1166 				if (sample->type == TYPE_RUNNING) {
1167 					svg_process(sample->cpu,
1168 						    sample->start_time,
1169 						    sample->end_time,
1170 						    p->pid,
1171 						    c->comm,
1172 						    sample->backtrace);
1173 				}
1174 
1175 				sample = sample->next;
1176 			}
1177 			c = c->next;
1178 		}
1179 		p = p->next;
1180 	}
1181 }
1182 
1183 static void draw_io_bars(struct timechart *tchart)
1184 {
1185 	const char *suf;
1186 	double bytes;
1187 	char comm[256];
1188 	struct per_pid *p;
1189 	struct per_pidcomm *c;
1190 	struct io_sample *sample;
1191 	int Y = 1;
1192 
1193 	p = tchart->all_data;
1194 	while (p) {
1195 		c = p->all;
1196 		while (c) {
1197 			if (!c->display) {
1198 				c->Y = 0;
1199 				c = c->next;
1200 				continue;
1201 			}
1202 
1203 			svg_box(Y, c->start_time, c->end_time, "process3");
1204 			for (sample = c->io_samples; sample; sample = sample->next) {
1205 				double h = (double)sample->bytes / c->max_bytes;
1206 
1207 				if (tchart->skip_eagain &&
1208 				    sample->err == -EAGAIN)
1209 					continue;
1210 
1211 				if (sample->err)
1212 					h = 1;
1213 
1214 				if (sample->type == IOTYPE_SYNC)
1215 					svg_fbox(Y,
1216 						sample->start_time,
1217 						sample->end_time,
1218 						1,
1219 						sample->err ? "error" : "sync",
1220 						sample->fd,
1221 						sample->err,
1222 						sample->merges);
1223 				else if (sample->type == IOTYPE_POLL)
1224 					svg_fbox(Y,
1225 						sample->start_time,
1226 						sample->end_time,
1227 						1,
1228 						sample->err ? "error" : "poll",
1229 						sample->fd,
1230 						sample->err,
1231 						sample->merges);
1232 				else if (sample->type == IOTYPE_READ)
1233 					svg_ubox(Y,
1234 						sample->start_time,
1235 						sample->end_time,
1236 						h,
1237 						sample->err ? "error" : "disk",
1238 						sample->fd,
1239 						sample->err,
1240 						sample->merges);
1241 				else if (sample->type == IOTYPE_WRITE)
1242 					svg_lbox(Y,
1243 						sample->start_time,
1244 						sample->end_time,
1245 						h,
1246 						sample->err ? "error" : "disk",
1247 						sample->fd,
1248 						sample->err,
1249 						sample->merges);
1250 				else if (sample->type == IOTYPE_RX)
1251 					svg_ubox(Y,
1252 						sample->start_time,
1253 						sample->end_time,
1254 						h,
1255 						sample->err ? "error" : "net",
1256 						sample->fd,
1257 						sample->err,
1258 						sample->merges);
1259 				else if (sample->type == IOTYPE_TX)
1260 					svg_lbox(Y,
1261 						sample->start_time,
1262 						sample->end_time,
1263 						h,
1264 						sample->err ? "error" : "net",
1265 						sample->fd,
1266 						sample->err,
1267 						sample->merges);
1268 			}
1269 
1270 			suf = "";
1271 			bytes = c->total_bytes;
1272 			if (bytes > 1024) {
1273 				bytes = bytes / 1024;
1274 				suf = "K";
1275 			}
1276 			if (bytes > 1024) {
1277 				bytes = bytes / 1024;
1278 				suf = "M";
1279 			}
1280 			if (bytes > 1024) {
1281 				bytes = bytes / 1024;
1282 				suf = "G";
1283 			}
1284 
1285 
1286 			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1287 			svg_text(Y, c->start_time, comm);
1288 
1289 			c->Y = Y;
1290 			Y++;
1291 			c = c->next;
1292 		}
1293 		p = p->next;
1294 	}
1295 }
1296 
1297 static void draw_process_bars(struct timechart *tchart)
1298 {
1299 	struct per_pid *p;
1300 	struct per_pidcomm *c;
1301 	struct cpu_sample *sample;
1302 	int Y = 0;
1303 
1304 	Y = 2 * tchart->numcpus + 2;
1305 
1306 	p = tchart->all_data;
1307 	while (p) {
1308 		c = p->all;
1309 		while (c) {
1310 			if (!c->display) {
1311 				c->Y = 0;
1312 				c = c->next;
1313 				continue;
1314 			}
1315 
1316 			svg_box(Y, c->start_time, c->end_time, "process");
1317 			sample = c->samples;
1318 			while (sample) {
1319 				if (sample->type == TYPE_RUNNING)
1320 					svg_running(Y, sample->cpu,
1321 						    sample->start_time,
1322 						    sample->end_time,
1323 						    sample->backtrace);
1324 				if (sample->type == TYPE_BLOCKED)
1325 					svg_blocked(Y, sample->cpu,
1326 						    sample->start_time,
1327 						    sample->end_time,
1328 						    sample->backtrace);
1329 				if (sample->type == TYPE_WAITING)
1330 					svg_waiting(Y, sample->cpu,
1331 						    sample->start_time,
1332 						    sample->end_time,
1333 						    sample->backtrace);
1334 				sample = sample->next;
1335 			}
1336 
1337 			if (c->comm) {
1338 				char comm[256];
1339 				if (c->total_time > 5000000000) /* 5 seconds */
1340 					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / (double)NSEC_PER_SEC);
1341 				else
1342 					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / (double)NSEC_PER_MSEC);
1343 
1344 				svg_text(Y, c->start_time, comm);
1345 			}
1346 			c->Y = Y;
1347 			Y++;
1348 			c = c->next;
1349 		}
1350 		p = p->next;
1351 	}
1352 }
1353 
1354 static void add_process_filter(const char *string)
1355 {
1356 	int pid = strtoull(string, NULL, 10);
1357 	struct process_filter *filt = malloc(sizeof(*filt));
1358 
1359 	if (!filt)
1360 		return;
1361 
1362 	filt->name = strdup(string);
1363 	filt->pid  = pid;
1364 	filt->next = process_filter;
1365 
1366 	process_filter = filt;
1367 }
1368 
1369 static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1370 {
1371 	struct process_filter *filt;
1372 	if (!process_filter)
1373 		return 1;
1374 
1375 	filt = process_filter;
1376 	while (filt) {
1377 		if (filt->pid && p->pid == filt->pid)
1378 			return 1;
1379 		if (strcmp(filt->name, c->comm) == 0)
1380 			return 1;
1381 		filt = filt->next;
1382 	}
1383 	return 0;
1384 }
1385 
1386 static int determine_display_tasks_filtered(struct timechart *tchart)
1387 {
1388 	struct per_pid *p;
1389 	struct per_pidcomm *c;
1390 	int count = 0;
1391 
1392 	p = tchart->all_data;
1393 	while (p) {
1394 		p->display = 0;
1395 		if (p->start_time == 1)
1396 			p->start_time = tchart->first_time;
1397 
1398 		/* no exit marker, task kept running to the end */
1399 		if (p->end_time == 0)
1400 			p->end_time = tchart->last_time;
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 (passes_filter(p, c)) {
1411 				c->display = 1;
1412 				p->display = 1;
1413 				count++;
1414 			}
1415 
1416 			if (c->end_time == 0)
1417 				c->end_time = tchart->last_time;
1418 
1419 			c = c->next;
1420 		}
1421 		p = p->next;
1422 	}
1423 	return count;
1424 }
1425 
1426 static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1427 {
1428 	struct per_pid *p;
1429 	struct per_pidcomm *c;
1430 	int count = 0;
1431 
1432 	p = tchart->all_data;
1433 	while (p) {
1434 		p->display = 0;
1435 		if (p->start_time == 1)
1436 			p->start_time = tchart->first_time;
1437 
1438 		/* no exit marker, task kept running to the end */
1439 		if (p->end_time == 0)
1440 			p->end_time = tchart->last_time;
1441 		if (p->total_time >= threshold)
1442 			p->display = 1;
1443 
1444 		c = p->all;
1445 
1446 		while (c) {
1447 			c->display = 0;
1448 
1449 			if (c->start_time == 1)
1450 				c->start_time = tchart->first_time;
1451 
1452 			if (c->total_time >= threshold) {
1453 				c->display = 1;
1454 				count++;
1455 			}
1456 
1457 			if (c->end_time == 0)
1458 				c->end_time = tchart->last_time;
1459 
1460 			c = c->next;
1461 		}
1462 		p = p->next;
1463 	}
1464 	return count;
1465 }
1466 
1467 static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1468 {
1469 	struct per_pid *p;
1470 	struct per_pidcomm *c;
1471 	int count = 0;
1472 
1473 	p = timechart->all_data;
1474 	while (p) {
1475 		/* no exit marker, task kept running to the end */
1476 		if (p->end_time == 0)
1477 			p->end_time = timechart->last_time;
1478 
1479 		c = p->all;
1480 
1481 		while (c) {
1482 			c->display = 0;
1483 
1484 			if (c->total_bytes >= threshold) {
1485 				c->display = 1;
1486 				count++;
1487 			}
1488 
1489 			if (c->end_time == 0)
1490 				c->end_time = timechart->last_time;
1491 
1492 			c = c->next;
1493 		}
1494 		p = p->next;
1495 	}
1496 	return count;
1497 }
1498 
1499 #define BYTES_THRESH (1 * 1024 * 1024)
1500 #define TIME_THRESH 10000000
1501 
1502 static void write_svg_file(struct timechart *tchart, const char *filename)
1503 {
1504 	u64 i;
1505 	int count;
1506 	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1507 
1508 	if (tchart->power_only)
1509 		tchart->proc_num = 0;
1510 
1511 	/* We'd like to show at least proc_num tasks;
1512 	 * be less picky if we have fewer */
1513 	do {
1514 		if (process_filter)
1515 			count = determine_display_tasks_filtered(tchart);
1516 		else if (tchart->io_events)
1517 			count = determine_display_io_tasks(tchart, thresh);
1518 		else
1519 			count = determine_display_tasks(tchart, thresh);
1520 		thresh /= 10;
1521 	} while (!process_filter && thresh && count < tchart->proc_num);
1522 
1523 	if (!tchart->proc_num)
1524 		count = 0;
1525 
1526 	if (tchart->io_events) {
1527 		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1528 
1529 		svg_time_grid(0.5);
1530 		svg_io_legenda();
1531 
1532 		draw_io_bars(tchart);
1533 	} else {
1534 		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1535 
1536 		svg_time_grid(0);
1537 
1538 		svg_legenda();
1539 
1540 		for (i = 0; i < tchart->numcpus; i++)
1541 			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1542 
1543 		draw_cpu_usage(tchart);
1544 		if (tchart->proc_num)
1545 			draw_process_bars(tchart);
1546 		if (!tchart->tasks_only)
1547 			draw_c_p_states(tchart);
1548 		if (tchart->proc_num)
1549 			draw_wakeups(tchart);
1550 	}
1551 
1552 	svg_close();
1553 }
1554 
1555 static int process_header(struct perf_file_section *section __maybe_unused,
1556 			  struct perf_header *ph,
1557 			  int feat,
1558 			  int fd __maybe_unused,
1559 			  void *data)
1560 {
1561 	struct timechart *tchart = data;
1562 
1563 	switch (feat) {
1564 	case HEADER_NRCPUS:
1565 		tchart->numcpus = ph->env.nr_cpus_avail;
1566 		if (tchart->numcpus > MAX_CPUS)
1567 			tchart->numcpus = MAX_CPUS;
1568 		break;
1569 
1570 	case HEADER_CPU_TOPOLOGY:
1571 		if (!tchart->topology)
1572 			break;
1573 
1574 		if (svg_build_topology_map(&ph->env))
1575 			fprintf(stderr, "problem building topology\n");
1576 		break;
1577 
1578 	default:
1579 		break;
1580 	}
1581 
1582 	return 0;
1583 }
1584 
1585 static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1586 {
1587 	const struct evsel_str_handler power_tracepoints[] = {
1588 		{ "power:cpu_idle",		process_sample_cpu_idle },
1589 		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1590 		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1591 		{ "sched:sched_switch",		process_sample_sched_switch },
1592 #ifdef SUPPORT_OLD_POWER_EVENTS
1593 		{ "power:power_start",		process_sample_power_start },
1594 		{ "power:power_end",		process_sample_power_end },
1595 		{ "power:power_frequency",	process_sample_power_frequency },
1596 #endif
1597 
1598 		{ "syscalls:sys_enter_read",		process_enter_read },
1599 		{ "syscalls:sys_enter_pread64",		process_enter_read },
1600 		{ "syscalls:sys_enter_readv",		process_enter_read },
1601 		{ "syscalls:sys_enter_preadv",		process_enter_read },
1602 		{ "syscalls:sys_enter_write",		process_enter_write },
1603 		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1604 		{ "syscalls:sys_enter_writev",		process_enter_write },
1605 		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1606 		{ "syscalls:sys_enter_sync",		process_enter_sync },
1607 		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1608 		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1609 		{ "syscalls:sys_enter_msync",		process_enter_sync },
1610 		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1611 		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1612 		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1613 		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1614 		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1615 		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1616 		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1617 		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1618 		{ "syscalls:sys_enter_poll",		process_enter_poll },
1619 		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1620 		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1621 		{ "syscalls:sys_enter_select",		process_enter_poll },
1622 
1623 		{ "syscalls:sys_exit_read",		process_exit_read },
1624 		{ "syscalls:sys_exit_pread64",		process_exit_read },
1625 		{ "syscalls:sys_exit_readv",		process_exit_read },
1626 		{ "syscalls:sys_exit_preadv",		process_exit_read },
1627 		{ "syscalls:sys_exit_write",		process_exit_write },
1628 		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1629 		{ "syscalls:sys_exit_writev",		process_exit_write },
1630 		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1631 		{ "syscalls:sys_exit_sync",		process_exit_sync },
1632 		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1633 		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1634 		{ "syscalls:sys_exit_msync",		process_exit_sync },
1635 		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1636 		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1637 		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1638 		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1639 		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1640 		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1641 		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1642 		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1643 		{ "syscalls:sys_exit_poll",		process_exit_poll },
1644 		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1645 		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1646 		{ "syscalls:sys_exit_select",		process_exit_poll },
1647 	};
1648 	struct perf_data data = {
1649 		.path  = input_name,
1650 		.mode  = PERF_DATA_MODE_READ,
1651 		.force = tchart->force,
1652 	};
1653 	struct perf_session *session;
1654 	int ret = -EINVAL;
1655 
1656 	perf_tool__init(&tchart->tool, /*ordered_events=*/true);
1657 	tchart->tool.comm		 = process_comm_event;
1658 	tchart->tool.fork		 = process_fork_event;
1659 	tchart->tool.exit		 = process_exit_event;
1660 	tchart->tool.sample		 = process_sample_event;
1661 
1662 	session = perf_session__new(&data, &tchart->tool);
1663 	if (IS_ERR(session))
1664 		return PTR_ERR(session);
1665 
1666 	symbol__init(perf_session__env(session));
1667 
1668 	(void)perf_header__process_sections(&session->header,
1669 					    perf_data__fd(session->data),
1670 					    tchart,
1671 					    process_header);
1672 
1673 	if (!perf_session__has_traces(session, "timechart record"))
1674 		goto out_delete;
1675 
1676 	if (perf_session__set_tracepoints_handlers(session,
1677 						   power_tracepoints)) {
1678 		pr_err("Initializing session tracepoint handlers failed\n");
1679 		goto out_delete;
1680 	}
1681 
1682 	ret = perf_session__process_events(session);
1683 	if (ret)
1684 		goto out_delete;
1685 
1686 	end_sample_processing(tchart);
1687 
1688 	sort_pids(tchart);
1689 
1690 	write_svg_file(tchart, output_name);
1691 
1692 	pr_info("Written %2.1f seconds of trace to %s.\n",
1693 		(tchart->last_time - tchart->first_time) / (double)NSEC_PER_SEC, output_name);
1694 out_delete:
1695 	perf_session__delete(session);
1696 	return ret;
1697 }
1698 
1699 static int timechart__io_record(int argc, const char **argv, const char *output_data)
1700 {
1701 	unsigned int rec_argc, i;
1702 	const char **rec_argv;
1703 	const char **p;
1704 	char *filter = NULL;
1705 
1706 	const char * const common_args[] = {
1707 		"record", "-a", "-R", "-c", "1", "-o", output_data,
1708 	};
1709 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1710 
1711 	const char * const disk_events[] = {
1712 		"syscalls:sys_enter_read",
1713 		"syscalls:sys_enter_pread64",
1714 		"syscalls:sys_enter_readv",
1715 		"syscalls:sys_enter_preadv",
1716 		"syscalls:sys_enter_write",
1717 		"syscalls:sys_enter_pwrite64",
1718 		"syscalls:sys_enter_writev",
1719 		"syscalls:sys_enter_pwritev",
1720 		"syscalls:sys_enter_sync",
1721 		"syscalls:sys_enter_sync_file_range",
1722 		"syscalls:sys_enter_fsync",
1723 		"syscalls:sys_enter_msync",
1724 
1725 		"syscalls:sys_exit_read",
1726 		"syscalls:sys_exit_pread64",
1727 		"syscalls:sys_exit_readv",
1728 		"syscalls:sys_exit_preadv",
1729 		"syscalls:sys_exit_write",
1730 		"syscalls:sys_exit_pwrite64",
1731 		"syscalls:sys_exit_writev",
1732 		"syscalls:sys_exit_pwritev",
1733 		"syscalls:sys_exit_sync",
1734 		"syscalls:sys_exit_sync_file_range",
1735 		"syscalls:sys_exit_fsync",
1736 		"syscalls:sys_exit_msync",
1737 	};
1738 	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1739 
1740 	const char * const net_events[] = {
1741 		"syscalls:sys_enter_recvfrom",
1742 		"syscalls:sys_enter_recvmmsg",
1743 		"syscalls:sys_enter_recvmsg",
1744 		"syscalls:sys_enter_sendto",
1745 		"syscalls:sys_enter_sendmsg",
1746 		"syscalls:sys_enter_sendmmsg",
1747 
1748 		"syscalls:sys_exit_recvfrom",
1749 		"syscalls:sys_exit_recvmmsg",
1750 		"syscalls:sys_exit_recvmsg",
1751 		"syscalls:sys_exit_sendto",
1752 		"syscalls:sys_exit_sendmsg",
1753 		"syscalls:sys_exit_sendmmsg",
1754 	};
1755 	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1756 
1757 	const char * const poll_events[] = {
1758 		"syscalls:sys_enter_epoll_pwait",
1759 		"syscalls:sys_enter_epoll_wait",
1760 		"syscalls:sys_enter_poll",
1761 		"syscalls:sys_enter_ppoll",
1762 		"syscalls:sys_enter_pselect6",
1763 		"syscalls:sys_enter_select",
1764 
1765 		"syscalls:sys_exit_epoll_pwait",
1766 		"syscalls:sys_exit_epoll_wait",
1767 		"syscalls:sys_exit_poll",
1768 		"syscalls:sys_exit_ppoll",
1769 		"syscalls:sys_exit_pselect6",
1770 		"syscalls:sys_exit_select",
1771 	};
1772 	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1773 
1774 	rec_argc = common_args_nr +
1775 		disk_events_nr * 4 +
1776 		net_events_nr * 4 +
1777 		poll_events_nr * 4 +
1778 		argc;
1779 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1780 
1781 	if (rec_argv == NULL)
1782 		return -ENOMEM;
1783 
1784 	if (asprintf(&filter, "common_pid != %d", getpid()) < 0) {
1785 		free(rec_argv);
1786 		return -ENOMEM;
1787 	}
1788 
1789 	p = rec_argv;
1790 	for (i = 0; i < common_args_nr; i++)
1791 		*p++ = strdup(common_args[i]);
1792 
1793 	for (i = 0; i < disk_events_nr; i++) {
1794 		if (!is_valid_tracepoint(disk_events[i])) {
1795 			rec_argc -= 4;
1796 			continue;
1797 		}
1798 
1799 		*p++ = "-e";
1800 		*p++ = strdup(disk_events[i]);
1801 		*p++ = "--filter";
1802 		*p++ = filter;
1803 	}
1804 	for (i = 0; i < net_events_nr; i++) {
1805 		if (!is_valid_tracepoint(net_events[i])) {
1806 			rec_argc -= 4;
1807 			continue;
1808 		}
1809 
1810 		*p++ = "-e";
1811 		*p++ = strdup(net_events[i]);
1812 		*p++ = "--filter";
1813 		*p++ = filter;
1814 	}
1815 	for (i = 0; i < poll_events_nr; i++) {
1816 		if (!is_valid_tracepoint(poll_events[i])) {
1817 			rec_argc -= 4;
1818 			continue;
1819 		}
1820 
1821 		*p++ = "-e";
1822 		*p++ = strdup(poll_events[i]);
1823 		*p++ = "--filter";
1824 		*p++ = filter;
1825 	}
1826 
1827 	for (i = 0; i < (unsigned int)argc; i++)
1828 		*p++ = argv[i];
1829 
1830 	return cmd_record(rec_argc, rec_argv);
1831 }
1832 
1833 
1834 static int timechart__record(struct timechart *tchart, int argc, const char **argv,
1835 			     const char *output_data)
1836 {
1837 	unsigned int rec_argc, i, j;
1838 	const char **rec_argv;
1839 	const char **p;
1840 	unsigned int record_elems;
1841 
1842 	const char * const common_args[] = {
1843 		"record", "-a", "-R", "-c", "1", "-o", output_data,
1844 	};
1845 	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1846 
1847 	const char * const backtrace_args[] = {
1848 		"-g",
1849 	};
1850 	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1851 
1852 	const char * const power_args[] = {
1853 		"-e", "power:cpu_frequency",
1854 		"-e", "power:cpu_idle",
1855 	};
1856 	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1857 
1858 	const char * const old_power_args[] = {
1859 #ifdef SUPPORT_OLD_POWER_EVENTS
1860 		"-e", "power:power_start",
1861 		"-e", "power:power_end",
1862 		"-e", "power:power_frequency",
1863 #endif
1864 	};
1865 	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1866 
1867 	const char * const tasks_args[] = {
1868 		"-e", "sched:sched_wakeup",
1869 		"-e", "sched:sched_switch",
1870 	};
1871 	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1872 
1873 #ifdef SUPPORT_OLD_POWER_EVENTS
1874 	if (!is_valid_tracepoint("power:cpu_idle") &&
1875 	    is_valid_tracepoint("power:power_start")) {
1876 		use_old_power_events = 1;
1877 		power_args_nr = 0;
1878 	} else {
1879 		old_power_args_nr = 0;
1880 	}
1881 #endif
1882 
1883 	if (tchart->power_only)
1884 		tasks_args_nr = 0;
1885 
1886 	if (tchart->tasks_only) {
1887 		power_args_nr = 0;
1888 		old_power_args_nr = 0;
1889 	}
1890 
1891 	if (!tchart->with_backtrace)
1892 		backtrace_args_no = 0;
1893 
1894 	record_elems = common_args_nr + tasks_args_nr +
1895 		power_args_nr + old_power_args_nr + backtrace_args_no;
1896 
1897 	rec_argc = record_elems + argc;
1898 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1899 
1900 	if (rec_argv == NULL)
1901 		return -ENOMEM;
1902 
1903 	p = rec_argv;
1904 	for (i = 0; i < common_args_nr; i++)
1905 		*p++ = strdup(common_args[i]);
1906 
1907 	for (i = 0; i < backtrace_args_no; i++)
1908 		*p++ = strdup(backtrace_args[i]);
1909 
1910 	for (i = 0; i < tasks_args_nr; i++)
1911 		*p++ = strdup(tasks_args[i]);
1912 
1913 	for (i = 0; i < power_args_nr; i++)
1914 		*p++ = strdup(power_args[i]);
1915 
1916 	for (i = 0; i < old_power_args_nr; i++)
1917 		*p++ = strdup(old_power_args[i]);
1918 
1919 	for (j = 0; j < (unsigned int)argc; j++)
1920 		*p++ = argv[j];
1921 
1922 	return cmd_record(rec_argc, rec_argv);
1923 }
1924 
1925 static int
1926 parse_process(const struct option *opt __maybe_unused, const char *arg,
1927 	      int __maybe_unused unset)
1928 {
1929 	if (arg)
1930 		add_process_filter(arg);
1931 	return 0;
1932 }
1933 
1934 static int
1935 parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1936 		int __maybe_unused unset)
1937 {
1938 	unsigned long duration = strtoul(arg, NULL, 0);
1939 
1940 	if (svg_highlight || svg_highlight_name)
1941 		return -1;
1942 
1943 	if (duration)
1944 		svg_highlight = duration;
1945 	else
1946 		svg_highlight_name = strdup(arg);
1947 
1948 	return 0;
1949 }
1950 
1951 static int
1952 parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1953 {
1954 	char unit = 'n';
1955 	u64 *value = opt->value;
1956 
1957 	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1958 		switch (unit) {
1959 		case 'm':
1960 			*value *= NSEC_PER_MSEC;
1961 			break;
1962 		case 'u':
1963 			*value *= NSEC_PER_USEC;
1964 			break;
1965 		case 'n':
1966 			break;
1967 		default:
1968 			return -1;
1969 		}
1970 	}
1971 
1972 	return 0;
1973 }
1974 
1975 int cmd_timechart(int argc, const char **argv)
1976 {
1977 	struct timechart tchart = {
1978 		.proc_num = 15,
1979 		.min_time = NSEC_PER_MSEC,
1980 		.merge_dist = 1000,
1981 	};
1982 	const char *output_name = "output.svg";
1983 	const char *output_record_data = "perf.data";
1984 	const struct option timechart_common_options[] = {
1985 	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1986 	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only, "output processes data only"),
1987 	OPT_END()
1988 	};
1989 	const struct option timechart_options[] = {
1990 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1991 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1992 	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1993 	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1994 		      "highlight tasks. Pass duration in ns or process name.",
1995 		       parse_highlight),
1996 	OPT_CALLBACK('p', "process", NULL, "process",
1997 		      "process selector. Pass a pid or process name.",
1998 		       parse_process),
1999 	OPT_CALLBACK(0, "symfs", NULL, "directory[,layout]", SYMFS_HELP,
2000 		     symbol__config_symfs),
2001 	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
2002 		    "min. number of tasks to print"),
2003 	OPT_BOOLEAN('t', "topology", &tchart.topology,
2004 		    "sort CPUs according to topology"),
2005 	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
2006 		    "skip EAGAIN errors"),
2007 	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
2008 		     "all IO faster than min-time will visually appear longer",
2009 		     parse_time),
2010 	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
2011 		     "merge events that are merge-dist us apart",
2012 		     parse_time),
2013 	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
2014 	OPT_PARENT(timechart_common_options),
2015 	};
2016 	const char * const timechart_subcommands[] = { "record", NULL };
2017 	const char *timechart_usage[] = {
2018 		"perf timechart [<options>] {record}",
2019 		NULL
2020 	};
2021 	const struct option timechart_record_options[] = {
2022 	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
2023 		    "record only IO data"),
2024 	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
2025 	OPT_STRING('o', "output", &output_record_data, "file", "output data file name"),
2026 	OPT_PARENT(timechart_common_options),
2027 	};
2028 	const char * const timechart_record_usage[] = {
2029 		"perf timechart record [<options>]",
2030 		NULL
2031 	};
2032 	int ret;
2033 
2034 	cpus_cstate_start_times = calloc(MAX_CPUS, sizeof(*cpus_cstate_start_times));
2035 	if (!cpus_cstate_start_times)
2036 		return -ENOMEM;
2037 	cpus_cstate_state = calloc(MAX_CPUS, sizeof(*cpus_cstate_state));
2038 	if (!cpus_cstate_state) {
2039 		ret = -ENOMEM;
2040 		goto out;
2041 	}
2042 	cpus_pstate_start_times = calloc(MAX_CPUS, sizeof(*cpus_pstate_start_times));
2043 	if (!cpus_pstate_start_times) {
2044 		ret = -ENOMEM;
2045 		goto out;
2046 	}
2047 	cpus_pstate_state = calloc(MAX_CPUS, sizeof(*cpus_pstate_state));
2048 	if (!cpus_pstate_state) {
2049 		ret = -ENOMEM;
2050 		goto out;
2051 	}
2052 
2053 	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
2054 			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
2055 
2056 	if (tchart.power_only && tchart.tasks_only) {
2057 		pr_err("-P and -T options cannot be used at the same time.\n");
2058 		ret = -1;
2059 		goto out;
2060 	}
2061 
2062 	if (argc && strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
2063 		argc = parse_options(argc, argv, timechart_record_options,
2064 				     timechart_record_usage,
2065 				     PARSE_OPT_STOP_AT_NON_OPTION);
2066 
2067 		if (tchart.power_only && tchart.tasks_only) {
2068 			pr_err("-P and -T options cannot be used at the same time.\n");
2069 			ret = -1;
2070 			goto out;
2071 		}
2072 
2073 		if (tchart.io_only)
2074 			ret = timechart__io_record(argc, argv, output_record_data);
2075 		else
2076 			ret = timechart__record(&tchart, argc, argv, output_record_data);
2077 		goto out;
2078 	} else if (argc)
2079 		usage_with_options(timechart_usage, timechart_options);
2080 
2081 	setup_pager();
2082 
2083 	ret = __cmd_timechart(&tchart, output_name);
2084 out:
2085 	zfree(&cpus_cstate_start_times);
2086 	zfree(&cpus_cstate_state);
2087 	zfree(&cpus_pstate_start_times);
2088 	zfree(&cpus_pstate_state);
2089 	return ret;
2090 }
2091