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 <traceevent/event-parse.h>
16
17#include "builtin.h"
18
19#include "util/util.h"
20
21#include "util/color.h"
22#include <linux/list.h>
23#include "util/cache.h"
24#include "util/evlist.h"
25#include "util/evsel.h"
26#include <linux/rbtree.h>
27#include "util/symbol.h"
28#include "util/callchain.h"
29#include "util/strlist.h"
30
31#include "perf.h"
32#include "util/header.h"
33#include "util/parse-options.h"
34#include "util/parse-events.h"
35#include "util/event.h"
36#include "util/session.h"
37#include "util/svghelper.h"
38#include "util/tool.h"
39#include "util/data.h"
40#include "util/debug.h"
41
42#define SUPPORT_OLD_POWER_EVENTS 1
43#define PWR_EVENT_EXIT -1
44
45struct per_pid;
46struct power_event;
47struct wake_event;
48
49struct timechart {
50	struct perf_tool	tool;
51	struct per_pid		*all_data;
52	struct power_event	*power_events;
53	struct wake_event	*wake_events;
54	int			proc_num;
55	unsigned int		numcpus;
56	u64			min_freq,	/* Lowest CPU frequency seen */
57				max_freq,	/* Highest CPU frequency seen */
58				turbo_frequency,
59				first_time, last_time;
60	bool			power_only,
61				tasks_only,
62				with_backtrace,
63				topology;
64	bool			force;
65	/* IO related settings */
66	bool			io_only,
67				skip_eagain;
68	u64			io_events;
69	u64			min_time,
70				merge_dist;
71};
72
73struct per_pidcomm;
74struct cpu_sample;
75struct io_sample;
76
77/*
78 * Datastructure layout:
79 * We keep an list of "pid"s, matching the kernels notion of a task struct.
80 * Each "pid" entry, has a list of "comm"s.
81 *	this is because we want to track different programs different, while
82 *	exec will reuse the original pid (by design).
83 * Each comm has a list of samples that will be used to draw
84 * final graph.
85 */
86
87struct per_pid {
88	struct per_pid *next;
89
90	int		pid;
91	int		ppid;
92
93	u64		start_time;
94	u64		end_time;
95	u64		total_time;
96	u64		total_bytes;
97	int		display;
98
99	struct per_pidcomm *all;
100	struct per_pidcomm *current;
101};
102
103
104struct per_pidcomm {
105	struct per_pidcomm *next;
106
107	u64		start_time;
108	u64		end_time;
109	u64		total_time;
110	u64		max_bytes;
111	u64		total_bytes;
112
113	int		Y;
114	int		display;
115
116	long		state;
117	u64		state_since;
118
119	char		*comm;
120
121	struct cpu_sample *samples;
122	struct io_sample  *io_samples;
123};
124
125struct sample_wrapper {
126	struct sample_wrapper *next;
127
128	u64		timestamp;
129	unsigned char	data[0];
130};
131
132#define TYPE_NONE	0
133#define TYPE_RUNNING	1
134#define TYPE_WAITING	2
135#define TYPE_BLOCKED	3
136
137struct cpu_sample {
138	struct cpu_sample *next;
139
140	u64 start_time;
141	u64 end_time;
142	int type;
143	int cpu;
144	const char *backtrace;
145};
146
147enum {
148	IOTYPE_READ,
149	IOTYPE_WRITE,
150	IOTYPE_SYNC,
151	IOTYPE_TX,
152	IOTYPE_RX,
153	IOTYPE_POLL,
154};
155
156struct io_sample {
157	struct io_sample *next;
158
159	u64 start_time;
160	u64 end_time;
161	u64 bytes;
162	int type;
163	int fd;
164	int err;
165	int merges;
166};
167
168#define CSTATE 1
169#define PSTATE 2
170
171struct power_event {
172	struct power_event *next;
173	int type;
174	int state;
175	u64 start_time;
176	u64 end_time;
177	int cpu;
178};
179
180struct wake_event {
181	struct wake_event *next;
182	int waker;
183	int wakee;
184	u64 time;
185	const char *backtrace;
186};
187
188struct process_filter {
189	char			*name;
190	int			pid;
191	struct process_filter	*next;
192};
193
194static struct process_filter *process_filter;
195
196
197static struct per_pid *find_create_pid(struct timechart *tchart, int pid)
198{
199	struct per_pid *cursor = tchart->all_data;
200
201	while (cursor) {
202		if (cursor->pid == pid)
203			return cursor;
204		cursor = cursor->next;
205	}
206	cursor = zalloc(sizeof(*cursor));
207	assert(cursor != NULL);
208	cursor->pid = pid;
209	cursor->next = tchart->all_data;
210	tchart->all_data = cursor;
211	return cursor;
212}
213
214static void pid_set_comm(struct timechart *tchart, int pid, char *comm)
215{
216	struct per_pid *p;
217	struct per_pidcomm *c;
218	p = find_create_pid(tchart, pid);
219	c = p->all;
220	while (c) {
221		if (c->comm && strcmp(c->comm, comm) == 0) {
222			p->current = c;
223			return;
224		}
225		if (!c->comm) {
226			c->comm = strdup(comm);
227			p->current = c;
228			return;
229		}
230		c = c->next;
231	}
232	c = zalloc(sizeof(*c));
233	assert(c != NULL);
234	c->comm = strdup(comm);
235	p->current = c;
236	c->next = p->all;
237	p->all = c;
238}
239
240static void pid_fork(struct timechart *tchart, int pid, int ppid, u64 timestamp)
241{
242	struct per_pid *p, *pp;
243	p = find_create_pid(tchart, pid);
244	pp = find_create_pid(tchart, ppid);
245	p->ppid = ppid;
246	if (pp->current && pp->current->comm && !p->current)
247		pid_set_comm(tchart, pid, pp->current->comm);
248
249	p->start_time = timestamp;
250	if (p->current && !p->current->start_time) {
251		p->current->start_time = timestamp;
252		p->current->state_since = timestamp;
253	}
254}
255
256static void pid_exit(struct timechart *tchart, int pid, u64 timestamp)
257{
258	struct per_pid *p;
259	p = find_create_pid(tchart, pid);
260	p->end_time = timestamp;
261	if (p->current)
262		p->current->end_time = timestamp;
263}
264
265static void pid_put_sample(struct timechart *tchart, int pid, int type,
266			   unsigned int cpu, u64 start, u64 end,
267			   const char *backtrace)
268{
269	struct per_pid *p;
270	struct per_pidcomm *c;
271	struct cpu_sample *sample;
272
273	p = find_create_pid(tchart, pid);
274	c = p->current;
275	if (!c) {
276		c = zalloc(sizeof(*c));
277		assert(c != NULL);
278		p->current = c;
279		c->next = p->all;
280		p->all = c;
281	}
282
283	sample = zalloc(sizeof(*sample));
284	assert(sample != NULL);
285	sample->start_time = start;
286	sample->end_time = end;
287	sample->type = type;
288	sample->next = c->samples;
289	sample->cpu = cpu;
290	sample->backtrace = backtrace;
291	c->samples = sample;
292
293	if (sample->type == TYPE_RUNNING && end > start && start > 0) {
294		c->total_time += (end-start);
295		p->total_time += (end-start);
296	}
297
298	if (c->start_time == 0 || c->start_time > start)
299		c->start_time = start;
300	if (p->start_time == 0 || p->start_time > start)
301		p->start_time = start;
302}
303
304#define MAX_CPUS 4096
305
306static u64 cpus_cstate_start_times[MAX_CPUS];
307static int cpus_cstate_state[MAX_CPUS];
308static u64 cpus_pstate_start_times[MAX_CPUS];
309static u64 cpus_pstate_state[MAX_CPUS];
310
311static int process_comm_event(struct perf_tool *tool,
312			      union perf_event *event,
313			      struct perf_sample *sample __maybe_unused,
314			      struct machine *machine __maybe_unused)
315{
316	struct timechart *tchart = container_of(tool, struct timechart, tool);
317	pid_set_comm(tchart, event->comm.tid, event->comm.comm);
318	return 0;
319}
320
321static int process_fork_event(struct perf_tool *tool,
322			      union perf_event *event,
323			      struct perf_sample *sample __maybe_unused,
324			      struct machine *machine __maybe_unused)
325{
326	struct timechart *tchart = container_of(tool, struct timechart, tool);
327	pid_fork(tchart, event->fork.pid, event->fork.ppid, event->fork.time);
328	return 0;
329}
330
331static int process_exit_event(struct perf_tool *tool,
332			      union perf_event *event,
333			      struct perf_sample *sample __maybe_unused,
334			      struct machine *machine __maybe_unused)
335{
336	struct timechart *tchart = container_of(tool, struct timechart, tool);
337	pid_exit(tchart, event->fork.pid, event->fork.time);
338	return 0;
339}
340
341#ifdef SUPPORT_OLD_POWER_EVENTS
342static int use_old_power_events;
343#endif
344
345static void c_state_start(int cpu, u64 timestamp, int state)
346{
347	cpus_cstate_start_times[cpu] = timestamp;
348	cpus_cstate_state[cpu] = state;
349}
350
351static void c_state_end(struct timechart *tchart, int cpu, u64 timestamp)
352{
353	struct power_event *pwr = zalloc(sizeof(*pwr));
354
355	if (!pwr)
356		return;
357
358	pwr->state = cpus_cstate_state[cpu];
359	pwr->start_time = cpus_cstate_start_times[cpu];
360	pwr->end_time = timestamp;
361	pwr->cpu = cpu;
362	pwr->type = CSTATE;
363	pwr->next = tchart->power_events;
364
365	tchart->power_events = pwr;
366}
367
368static void p_state_change(struct timechart *tchart, int cpu, u64 timestamp, u64 new_freq)
369{
370	struct power_event *pwr;
371
372	if (new_freq > 8000000) /* detect invalid data */
373		return;
374
375	pwr = zalloc(sizeof(*pwr));
376	if (!pwr)
377		return;
378
379	pwr->state = cpus_pstate_state[cpu];
380	pwr->start_time = cpus_pstate_start_times[cpu];
381	pwr->end_time = timestamp;
382	pwr->cpu = cpu;
383	pwr->type = PSTATE;
384	pwr->next = tchart->power_events;
385
386	if (!pwr->start_time)
387		pwr->start_time = tchart->first_time;
388
389	tchart->power_events = pwr;
390
391	cpus_pstate_state[cpu] = new_freq;
392	cpus_pstate_start_times[cpu] = timestamp;
393
394	if ((u64)new_freq > tchart->max_freq)
395		tchart->max_freq = new_freq;
396
397	if (new_freq < tchart->min_freq || tchart->min_freq == 0)
398		tchart->min_freq = new_freq;
399
400	if (new_freq == tchart->max_freq - 1000)
401		tchart->turbo_frequency = tchart->max_freq;
402}
403
404static void sched_wakeup(struct timechart *tchart, int cpu, u64 timestamp,
405			 int waker, int wakee, u8 flags, const char *backtrace)
406{
407	struct per_pid *p;
408	struct wake_event *we = zalloc(sizeof(*we));
409
410	if (!we)
411		return;
412
413	we->time = timestamp;
414	we->waker = waker;
415	we->backtrace = backtrace;
416
417	if ((flags & TRACE_FLAG_HARDIRQ) || (flags & TRACE_FLAG_SOFTIRQ))
418		we->waker = -1;
419
420	we->wakee = wakee;
421	we->next = tchart->wake_events;
422	tchart->wake_events = we;
423	p = find_create_pid(tchart, we->wakee);
424
425	if (p && p->current && p->current->state == TYPE_NONE) {
426		p->current->state_since = timestamp;
427		p->current->state = TYPE_WAITING;
428	}
429	if (p && p->current && p->current->state == TYPE_BLOCKED) {
430		pid_put_sample(tchart, p->pid, p->current->state, cpu,
431			       p->current->state_since, timestamp, NULL);
432		p->current->state_since = timestamp;
433		p->current->state = TYPE_WAITING;
434	}
435}
436
437static void sched_switch(struct timechart *tchart, int cpu, u64 timestamp,
438			 int prev_pid, int next_pid, u64 prev_state,
439			 const char *backtrace)
440{
441	struct per_pid *p = NULL, *prev_p;
442
443	prev_p = find_create_pid(tchart, prev_pid);
444
445	p = find_create_pid(tchart, next_pid);
446
447	if (prev_p->current && prev_p->current->state != TYPE_NONE)
448		pid_put_sample(tchart, prev_pid, TYPE_RUNNING, cpu,
449			       prev_p->current->state_since, timestamp,
450			       backtrace);
451	if (p && p->current) {
452		if (p->current->state != TYPE_NONE)
453			pid_put_sample(tchart, next_pid, p->current->state, cpu,
454				       p->current->state_since, timestamp,
455				       backtrace);
456
457		p->current->state_since = timestamp;
458		p->current->state = TYPE_RUNNING;
459	}
460
461	if (prev_p->current) {
462		prev_p->current->state = TYPE_NONE;
463		prev_p->current->state_since = timestamp;
464		if (prev_state & 2)
465			prev_p->current->state = TYPE_BLOCKED;
466		if (prev_state == 0)
467			prev_p->current->state = TYPE_WAITING;
468	}
469}
470
471static const char *cat_backtrace(union perf_event *event,
472				 struct perf_sample *sample,
473				 struct machine *machine)
474{
475	struct addr_location al;
476	unsigned int i;
477	char *p = NULL;
478	size_t p_len;
479	u8 cpumode = PERF_RECORD_MISC_USER;
480	struct addr_location tal;
481	struct ip_callchain *chain = sample->callchain;
482	FILE *f = open_memstream(&p, &p_len);
483
484	if (!f) {
485		perror("open_memstream error");
486		return NULL;
487	}
488
489	if (!chain)
490		goto exit;
491
492	if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
493		fprintf(stderr, "problem processing %d event, skipping it.\n",
494			event->header.type);
495		goto exit;
496	}
497
498	for (i = 0; i < chain->nr; i++) {
499		u64 ip;
500
501		if (callchain_param.order == ORDER_CALLEE)
502			ip = chain->ips[i];
503		else
504			ip = chain->ips[chain->nr - i - 1];
505
506		if (ip >= PERF_CONTEXT_MAX) {
507			switch (ip) {
508			case PERF_CONTEXT_HV:
509				cpumode = PERF_RECORD_MISC_HYPERVISOR;
510				break;
511			case PERF_CONTEXT_KERNEL:
512				cpumode = PERF_RECORD_MISC_KERNEL;
513				break;
514			case PERF_CONTEXT_USER:
515				cpumode = PERF_RECORD_MISC_USER;
516				break;
517			default:
518				pr_debug("invalid callchain context: "
519					 "%"PRId64"\n", (s64) ip);
520
521				/*
522				 * It seems the callchain is corrupted.
523				 * Discard all.
524				 */
525				zfree(&p);
526				goto exit_put;
527			}
528			continue;
529		}
530
531		tal.filtered = 0;
532		thread__find_addr_location(al.thread, cpumode,
533					   MAP__FUNCTION, ip, &tal);
534
535		if (tal.sym)
536			fprintf(f, "..... %016" PRIx64 " %s\n", ip,
537				tal.sym->name);
538		else
539			fprintf(f, "..... %016" PRIx64 "\n", ip);
540	}
541exit_put:
542	addr_location__put(&al);
543exit:
544	fclose(f);
545
546	return p;
547}
548
549typedef int (*tracepoint_handler)(struct timechart *tchart,
550				  struct perf_evsel *evsel,
551				  struct perf_sample *sample,
552				  const char *backtrace);
553
554static int process_sample_event(struct perf_tool *tool,
555				union perf_event *event,
556				struct perf_sample *sample,
557				struct perf_evsel *evsel,
558				struct machine *machine)
559{
560	struct timechart *tchart = container_of(tool, struct timechart, tool);
561
562	if (evsel->attr.sample_type & PERF_SAMPLE_TIME) {
563		if (!tchart->first_time || tchart->first_time > sample->time)
564			tchart->first_time = sample->time;
565		if (tchart->last_time < sample->time)
566			tchart->last_time = sample->time;
567	}
568
569	if (evsel->handler != NULL) {
570		tracepoint_handler f = evsel->handler;
571		return f(tchart, evsel, sample,
572			 cat_backtrace(event, sample, machine));
573	}
574
575	return 0;
576}
577
578static int
579process_sample_cpu_idle(struct timechart *tchart __maybe_unused,
580			struct perf_evsel *evsel,
581			struct perf_sample *sample,
582			const char *backtrace __maybe_unused)
583{
584	u32 state = perf_evsel__intval(evsel, sample, "state");
585	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
586
587	if (state == (u32)PWR_EVENT_EXIT)
588		c_state_end(tchart, cpu_id, sample->time);
589	else
590		c_state_start(cpu_id, sample->time, state);
591	return 0;
592}
593
594static int
595process_sample_cpu_frequency(struct timechart *tchart,
596			     struct perf_evsel *evsel,
597			     struct perf_sample *sample,
598			     const char *backtrace __maybe_unused)
599{
600	u32 state = perf_evsel__intval(evsel, sample, "state");
601	u32 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
602
603	p_state_change(tchart, cpu_id, sample->time, state);
604	return 0;
605}
606
607static int
608process_sample_sched_wakeup(struct timechart *tchart,
609			    struct perf_evsel *evsel,
610			    struct perf_sample *sample,
611			    const char *backtrace)
612{
613	u8 flags = perf_evsel__intval(evsel, sample, "common_flags");
614	int waker = perf_evsel__intval(evsel, sample, "common_pid");
615	int wakee = perf_evsel__intval(evsel, sample, "pid");
616
617	sched_wakeup(tchart, sample->cpu, sample->time, waker, wakee, flags, backtrace);
618	return 0;
619}
620
621static int
622process_sample_sched_switch(struct timechart *tchart,
623			    struct perf_evsel *evsel,
624			    struct perf_sample *sample,
625			    const char *backtrace)
626{
627	int prev_pid = perf_evsel__intval(evsel, sample, "prev_pid");
628	int next_pid = perf_evsel__intval(evsel, sample, "next_pid");
629	u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
630
631	sched_switch(tchart, sample->cpu, sample->time, prev_pid, next_pid,
632		     prev_state, backtrace);
633	return 0;
634}
635
636#ifdef SUPPORT_OLD_POWER_EVENTS
637static int
638process_sample_power_start(struct timechart *tchart __maybe_unused,
639			   struct perf_evsel *evsel,
640			   struct perf_sample *sample,
641			   const char *backtrace __maybe_unused)
642{
643	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
644	u64 value = perf_evsel__intval(evsel, sample, "value");
645
646	c_state_start(cpu_id, sample->time, value);
647	return 0;
648}
649
650static int
651process_sample_power_end(struct timechart *tchart,
652			 struct perf_evsel *evsel __maybe_unused,
653			 struct perf_sample *sample,
654			 const char *backtrace __maybe_unused)
655{
656	c_state_end(tchart, sample->cpu, sample->time);
657	return 0;
658}
659
660static int
661process_sample_power_frequency(struct timechart *tchart,
662			       struct perf_evsel *evsel,
663			       struct perf_sample *sample,
664			       const char *backtrace __maybe_unused)
665{
666	u64 cpu_id = perf_evsel__intval(evsel, sample, "cpu_id");
667	u64 value = perf_evsel__intval(evsel, sample, "value");
668
669	p_state_change(tchart, cpu_id, sample->time, value);
670	return 0;
671}
672#endif /* SUPPORT_OLD_POWER_EVENTS */
673
674/*
675 * After the last sample we need to wrap up the current C/P state
676 * and close out each CPU for these.
677 */
678static void end_sample_processing(struct timechart *tchart)
679{
680	u64 cpu;
681	struct power_event *pwr;
682
683	for (cpu = 0; cpu <= tchart->numcpus; cpu++) {
684		/* C state */
685#if 0
686		pwr = zalloc(sizeof(*pwr));
687		if (!pwr)
688			return;
689
690		pwr->state = cpus_cstate_state[cpu];
691		pwr->start_time = cpus_cstate_start_times[cpu];
692		pwr->end_time = tchart->last_time;
693		pwr->cpu = cpu;
694		pwr->type = CSTATE;
695		pwr->next = tchart->power_events;
696
697		tchart->power_events = pwr;
698#endif
699		/* P state */
700
701		pwr = zalloc(sizeof(*pwr));
702		if (!pwr)
703			return;
704
705		pwr->state = cpus_pstate_state[cpu];
706		pwr->start_time = cpus_pstate_start_times[cpu];
707		pwr->end_time = tchart->last_time;
708		pwr->cpu = cpu;
709		pwr->type = PSTATE;
710		pwr->next = tchart->power_events;
711
712		if (!pwr->start_time)
713			pwr->start_time = tchart->first_time;
714		if (!pwr->state)
715			pwr->state = tchart->min_freq;
716		tchart->power_events = pwr;
717	}
718}
719
720static int pid_begin_io_sample(struct timechart *tchart, int pid, int type,
721			       u64 start, int fd)
722{
723	struct per_pid *p = find_create_pid(tchart, pid);
724	struct per_pidcomm *c = p->current;
725	struct io_sample *sample;
726	struct io_sample *prev;
727
728	if (!c) {
729		c = zalloc(sizeof(*c));
730		if (!c)
731			return -ENOMEM;
732		p->current = c;
733		c->next = p->all;
734		p->all = c;
735	}
736
737	prev = c->io_samples;
738
739	if (prev && prev->start_time && !prev->end_time) {
740		pr_warning("Skip invalid start event: "
741			   "previous event already started!\n");
742
743		/* remove previous event that has been started,
744		 * we are not sure we will ever get an end for it */
745		c->io_samples = prev->next;
746		free(prev);
747		return 0;
748	}
749
750	sample = zalloc(sizeof(*sample));
751	if (!sample)
752		return -ENOMEM;
753	sample->start_time = start;
754	sample->type = type;
755	sample->fd = fd;
756	sample->next = c->io_samples;
757	c->io_samples = sample;
758
759	if (c->start_time == 0 || c->start_time > start)
760		c->start_time = start;
761
762	return 0;
763}
764
765static int pid_end_io_sample(struct timechart *tchart, int pid, int type,
766			     u64 end, long ret)
767{
768	struct per_pid *p = find_create_pid(tchart, pid);
769	struct per_pidcomm *c = p->current;
770	struct io_sample *sample, *prev;
771
772	if (!c) {
773		pr_warning("Invalid pidcomm!\n");
774		return -1;
775	}
776
777	sample = c->io_samples;
778
779	if (!sample) /* skip partially captured events */
780		return 0;
781
782	if (sample->end_time) {
783		pr_warning("Skip invalid end event: "
784			   "previous event already ended!\n");
785		return 0;
786	}
787
788	if (sample->type != type) {
789		pr_warning("Skip invalid end event: invalid event type!\n");
790		return 0;
791	}
792
793	sample->end_time = end;
794	prev = sample->next;
795
796	/* we want to be able to see small and fast transfers, so make them
797	 * at least min_time long, but don't overlap them */
798	if (sample->end_time - sample->start_time < tchart->min_time)
799		sample->end_time = sample->start_time + tchart->min_time;
800	if (prev && sample->start_time < prev->end_time) {
801		if (prev->err) /* try to make errors more visible */
802			sample->start_time = prev->end_time;
803		else
804			prev->end_time = sample->start_time;
805	}
806
807	if (ret < 0) {
808		sample->err = ret;
809	} else if (type == IOTYPE_READ || type == IOTYPE_WRITE ||
810		   type == IOTYPE_TX || type == IOTYPE_RX) {
811
812		if ((u64)ret > c->max_bytes)
813			c->max_bytes = ret;
814
815		c->total_bytes += ret;
816		p->total_bytes += ret;
817		sample->bytes = ret;
818	}
819
820	/* merge two requests to make svg smaller and render-friendly */
821	if (prev &&
822	    prev->type == sample->type &&
823	    prev->err == sample->err &&
824	    prev->fd == sample->fd &&
825	    prev->end_time + tchart->merge_dist >= sample->start_time) {
826
827		sample->bytes += prev->bytes;
828		sample->merges += prev->merges + 1;
829
830		sample->start_time = prev->start_time;
831		sample->next = prev->next;
832		free(prev);
833
834		if (!sample->err && sample->bytes > c->max_bytes)
835			c->max_bytes = sample->bytes;
836	}
837
838	tchart->io_events++;
839
840	return 0;
841}
842
843static int
844process_enter_read(struct timechart *tchart,
845		   struct perf_evsel *evsel,
846		   struct perf_sample *sample)
847{
848	long fd = perf_evsel__intval(evsel, sample, "fd");
849	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_READ,
850				   sample->time, fd);
851}
852
853static int
854process_exit_read(struct timechart *tchart,
855		  struct perf_evsel *evsel,
856		  struct perf_sample *sample)
857{
858	long ret = perf_evsel__intval(evsel, sample, "ret");
859	return pid_end_io_sample(tchart, sample->tid, IOTYPE_READ,
860				 sample->time, ret);
861}
862
863static int
864process_enter_write(struct timechart *tchart,
865		    struct perf_evsel *evsel,
866		    struct perf_sample *sample)
867{
868	long fd = perf_evsel__intval(evsel, sample, "fd");
869	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_WRITE,
870				   sample->time, fd);
871}
872
873static int
874process_exit_write(struct timechart *tchart,
875		   struct perf_evsel *evsel,
876		   struct perf_sample *sample)
877{
878	long ret = perf_evsel__intval(evsel, sample, "ret");
879	return pid_end_io_sample(tchart, sample->tid, IOTYPE_WRITE,
880				 sample->time, ret);
881}
882
883static int
884process_enter_sync(struct timechart *tchart,
885		   struct perf_evsel *evsel,
886		   struct perf_sample *sample)
887{
888	long fd = perf_evsel__intval(evsel, sample, "fd");
889	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_SYNC,
890				   sample->time, fd);
891}
892
893static int
894process_exit_sync(struct timechart *tchart,
895		  struct perf_evsel *evsel,
896		  struct perf_sample *sample)
897{
898	long ret = perf_evsel__intval(evsel, sample, "ret");
899	return pid_end_io_sample(tchart, sample->tid, IOTYPE_SYNC,
900				 sample->time, ret);
901}
902
903static int
904process_enter_tx(struct timechart *tchart,
905		 struct perf_evsel *evsel,
906		 struct perf_sample *sample)
907{
908	long fd = perf_evsel__intval(evsel, sample, "fd");
909	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_TX,
910				   sample->time, fd);
911}
912
913static int
914process_exit_tx(struct timechart *tchart,
915		struct perf_evsel *evsel,
916		struct perf_sample *sample)
917{
918	long ret = perf_evsel__intval(evsel, sample, "ret");
919	return pid_end_io_sample(tchart, sample->tid, IOTYPE_TX,
920				 sample->time, ret);
921}
922
923static int
924process_enter_rx(struct timechart *tchart,
925		 struct perf_evsel *evsel,
926		 struct perf_sample *sample)
927{
928	long fd = perf_evsel__intval(evsel, sample, "fd");
929	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_RX,
930				   sample->time, fd);
931}
932
933static int
934process_exit_rx(struct timechart *tchart,
935		struct perf_evsel *evsel,
936		struct perf_sample *sample)
937{
938	long ret = perf_evsel__intval(evsel, sample, "ret");
939	return pid_end_io_sample(tchart, sample->tid, IOTYPE_RX,
940				 sample->time, ret);
941}
942
943static int
944process_enter_poll(struct timechart *tchart,
945		   struct perf_evsel *evsel,
946		   struct perf_sample *sample)
947{
948	long fd = perf_evsel__intval(evsel, sample, "fd");
949	return pid_begin_io_sample(tchart, sample->tid, IOTYPE_POLL,
950				   sample->time, fd);
951}
952
953static int
954process_exit_poll(struct timechart *tchart,
955		  struct perf_evsel *evsel,
956		  struct perf_sample *sample)
957{
958	long ret = perf_evsel__intval(evsel, sample, "ret");
959	return pid_end_io_sample(tchart, sample->tid, IOTYPE_POLL,
960				 sample->time, ret);
961}
962
963/*
964 * Sort the pid datastructure
965 */
966static void sort_pids(struct timechart *tchart)
967{
968	struct per_pid *new_list, *p, *cursor, *prev;
969	/* sort by ppid first, then by pid, lowest to highest */
970
971	new_list = NULL;
972
973	while (tchart->all_data) {
974		p = tchart->all_data;
975		tchart->all_data = p->next;
976		p->next = NULL;
977
978		if (new_list == NULL) {
979			new_list = p;
980			p->next = NULL;
981			continue;
982		}
983		prev = NULL;
984		cursor = new_list;
985		while (cursor) {
986			if (cursor->ppid > p->ppid ||
987				(cursor->ppid == p->ppid && cursor->pid > p->pid)) {
988				/* must insert before */
989				if (prev) {
990					p->next = prev->next;
991					prev->next = p;
992					cursor = NULL;
993					continue;
994				} else {
995					p->next = new_list;
996					new_list = p;
997					cursor = NULL;
998					continue;
999				}
1000			}
1001
1002			prev = cursor;
1003			cursor = cursor->next;
1004			if (!cursor)
1005				prev->next = p;
1006		}
1007	}
1008	tchart->all_data = new_list;
1009}
1010
1011
1012static void draw_c_p_states(struct timechart *tchart)
1013{
1014	struct power_event *pwr;
1015	pwr = tchart->power_events;
1016
1017	/*
1018	 * two pass drawing so that the P state bars are on top of the C state blocks
1019	 */
1020	while (pwr) {
1021		if (pwr->type == CSTATE)
1022			svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1023		pwr = pwr->next;
1024	}
1025
1026	pwr = tchart->power_events;
1027	while (pwr) {
1028		if (pwr->type == PSTATE) {
1029			if (!pwr->state)
1030				pwr->state = tchart->min_freq;
1031			svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state);
1032		}
1033		pwr = pwr->next;
1034	}
1035}
1036
1037static void draw_wakeups(struct timechart *tchart)
1038{
1039	struct wake_event *we;
1040	struct per_pid *p;
1041	struct per_pidcomm *c;
1042
1043	we = tchart->wake_events;
1044	while (we) {
1045		int from = 0, to = 0;
1046		char *task_from = NULL, *task_to = NULL;
1047
1048		/* locate the column of the waker and wakee */
1049		p = tchart->all_data;
1050		while (p) {
1051			if (p->pid == we->waker || p->pid == we->wakee) {
1052				c = p->all;
1053				while (c) {
1054					if (c->Y && c->start_time <= we->time && c->end_time >= we->time) {
1055						if (p->pid == we->waker && !from) {
1056							from = c->Y;
1057							task_from = strdup(c->comm);
1058						}
1059						if (p->pid == we->wakee && !to) {
1060							to = c->Y;
1061							task_to = strdup(c->comm);
1062						}
1063					}
1064					c = c->next;
1065				}
1066				c = p->all;
1067				while (c) {
1068					if (p->pid == we->waker && !from) {
1069						from = c->Y;
1070						task_from = strdup(c->comm);
1071					}
1072					if (p->pid == we->wakee && !to) {
1073						to = c->Y;
1074						task_to = strdup(c->comm);
1075					}
1076					c = c->next;
1077				}
1078			}
1079			p = p->next;
1080		}
1081
1082		if (!task_from) {
1083			task_from = malloc(40);
1084			sprintf(task_from, "[%i]", we->waker);
1085		}
1086		if (!task_to) {
1087			task_to = malloc(40);
1088			sprintf(task_to, "[%i]", we->wakee);
1089		}
1090
1091		if (we->waker == -1)
1092			svg_interrupt(we->time, to, we->backtrace);
1093		else if (from && to && abs(from - to) == 1)
1094			svg_wakeline(we->time, from, to, we->backtrace);
1095		else
1096			svg_partial_wakeline(we->time, from, task_from, to,
1097					     task_to, we->backtrace);
1098		we = we->next;
1099
1100		free(task_from);
1101		free(task_to);
1102	}
1103}
1104
1105static void draw_cpu_usage(struct timechart *tchart)
1106{
1107	struct per_pid *p;
1108	struct per_pidcomm *c;
1109	struct cpu_sample *sample;
1110	p = tchart->all_data;
1111	while (p) {
1112		c = p->all;
1113		while (c) {
1114			sample = c->samples;
1115			while (sample) {
1116				if (sample->type == TYPE_RUNNING) {
1117					svg_process(sample->cpu,
1118						    sample->start_time,
1119						    sample->end_time,
1120						    p->pid,
1121						    c->comm,
1122						    sample->backtrace);
1123				}
1124
1125				sample = sample->next;
1126			}
1127			c = c->next;
1128		}
1129		p = p->next;
1130	}
1131}
1132
1133static void draw_io_bars(struct timechart *tchart)
1134{
1135	const char *suf;
1136	double bytes;
1137	char comm[256];
1138	struct per_pid *p;
1139	struct per_pidcomm *c;
1140	struct io_sample *sample;
1141	int Y = 1;
1142
1143	p = tchart->all_data;
1144	while (p) {
1145		c = p->all;
1146		while (c) {
1147			if (!c->display) {
1148				c->Y = 0;
1149				c = c->next;
1150				continue;
1151			}
1152
1153			svg_box(Y, c->start_time, c->end_time, "process3");
1154			sample = c->io_samples;
1155			for (sample = c->io_samples; sample; sample = sample->next) {
1156				double h = (double)sample->bytes / c->max_bytes;
1157
1158				if (tchart->skip_eagain &&
1159				    sample->err == -EAGAIN)
1160					continue;
1161
1162				if (sample->err)
1163					h = 1;
1164
1165				if (sample->type == IOTYPE_SYNC)
1166					svg_fbox(Y,
1167						sample->start_time,
1168						sample->end_time,
1169						1,
1170						sample->err ? "error" : "sync",
1171						sample->fd,
1172						sample->err,
1173						sample->merges);
1174				else if (sample->type == IOTYPE_POLL)
1175					svg_fbox(Y,
1176						sample->start_time,
1177						sample->end_time,
1178						1,
1179						sample->err ? "error" : "poll",
1180						sample->fd,
1181						sample->err,
1182						sample->merges);
1183				else if (sample->type == IOTYPE_READ)
1184					svg_ubox(Y,
1185						sample->start_time,
1186						sample->end_time,
1187						h,
1188						sample->err ? "error" : "disk",
1189						sample->fd,
1190						sample->err,
1191						sample->merges);
1192				else if (sample->type == IOTYPE_WRITE)
1193					svg_lbox(Y,
1194						sample->start_time,
1195						sample->end_time,
1196						h,
1197						sample->err ? "error" : "disk",
1198						sample->fd,
1199						sample->err,
1200						sample->merges);
1201				else if (sample->type == IOTYPE_RX)
1202					svg_ubox(Y,
1203						sample->start_time,
1204						sample->end_time,
1205						h,
1206						sample->err ? "error" : "net",
1207						sample->fd,
1208						sample->err,
1209						sample->merges);
1210				else if (sample->type == IOTYPE_TX)
1211					svg_lbox(Y,
1212						sample->start_time,
1213						sample->end_time,
1214						h,
1215						sample->err ? "error" : "net",
1216						sample->fd,
1217						sample->err,
1218						sample->merges);
1219			}
1220
1221			suf = "";
1222			bytes = c->total_bytes;
1223			if (bytes > 1024) {
1224				bytes = bytes / 1024;
1225				suf = "K";
1226			}
1227			if (bytes > 1024) {
1228				bytes = bytes / 1024;
1229				suf = "M";
1230			}
1231			if (bytes > 1024) {
1232				bytes = bytes / 1024;
1233				suf = "G";
1234			}
1235
1236
1237			sprintf(comm, "%s:%i (%3.1f %sbytes)", c->comm ?: "", p->pid, bytes, suf);
1238			svg_text(Y, c->start_time, comm);
1239
1240			c->Y = Y;
1241			Y++;
1242			c = c->next;
1243		}
1244		p = p->next;
1245	}
1246}
1247
1248static void draw_process_bars(struct timechart *tchart)
1249{
1250	struct per_pid *p;
1251	struct per_pidcomm *c;
1252	struct cpu_sample *sample;
1253	int Y = 0;
1254
1255	Y = 2 * tchart->numcpus + 2;
1256
1257	p = tchart->all_data;
1258	while (p) {
1259		c = p->all;
1260		while (c) {
1261			if (!c->display) {
1262				c->Y = 0;
1263				c = c->next;
1264				continue;
1265			}
1266
1267			svg_box(Y, c->start_time, c->end_time, "process");
1268			sample = c->samples;
1269			while (sample) {
1270				if (sample->type == TYPE_RUNNING)
1271					svg_running(Y, sample->cpu,
1272						    sample->start_time,
1273						    sample->end_time,
1274						    sample->backtrace);
1275				if (sample->type == TYPE_BLOCKED)
1276					svg_blocked(Y, sample->cpu,
1277						    sample->start_time,
1278						    sample->end_time,
1279						    sample->backtrace);
1280				if (sample->type == TYPE_WAITING)
1281					svg_waiting(Y, sample->cpu,
1282						    sample->start_time,
1283						    sample->end_time,
1284						    sample->backtrace);
1285				sample = sample->next;
1286			}
1287
1288			if (c->comm) {
1289				char comm[256];
1290				if (c->total_time > 5000000000) /* 5 seconds */
1291					sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / 1000000000.0);
1292				else
1293					sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / 1000000.0);
1294
1295				svg_text(Y, c->start_time, comm);
1296			}
1297			c->Y = Y;
1298			Y++;
1299			c = c->next;
1300		}
1301		p = p->next;
1302	}
1303}
1304
1305static void add_process_filter(const char *string)
1306{
1307	int pid = strtoull(string, NULL, 10);
1308	struct process_filter *filt = malloc(sizeof(*filt));
1309
1310	if (!filt)
1311		return;
1312
1313	filt->name = strdup(string);
1314	filt->pid  = pid;
1315	filt->next = process_filter;
1316
1317	process_filter = filt;
1318}
1319
1320static int passes_filter(struct per_pid *p, struct per_pidcomm *c)
1321{
1322	struct process_filter *filt;
1323	if (!process_filter)
1324		return 1;
1325
1326	filt = process_filter;
1327	while (filt) {
1328		if (filt->pid && p->pid == filt->pid)
1329			return 1;
1330		if (strcmp(filt->name, c->comm) == 0)
1331			return 1;
1332		filt = filt->next;
1333	}
1334	return 0;
1335}
1336
1337static int determine_display_tasks_filtered(struct timechart *tchart)
1338{
1339	struct per_pid *p;
1340	struct per_pidcomm *c;
1341	int count = 0;
1342
1343	p = tchart->all_data;
1344	while (p) {
1345		p->display = 0;
1346		if (p->start_time == 1)
1347			p->start_time = tchart->first_time;
1348
1349		/* no exit marker, task kept running to the end */
1350		if (p->end_time == 0)
1351			p->end_time = tchart->last_time;
1352
1353		c = p->all;
1354
1355		while (c) {
1356			c->display = 0;
1357
1358			if (c->start_time == 1)
1359				c->start_time = tchart->first_time;
1360
1361			if (passes_filter(p, c)) {
1362				c->display = 1;
1363				p->display = 1;
1364				count++;
1365			}
1366
1367			if (c->end_time == 0)
1368				c->end_time = tchart->last_time;
1369
1370			c = c->next;
1371		}
1372		p = p->next;
1373	}
1374	return count;
1375}
1376
1377static int determine_display_tasks(struct timechart *tchart, u64 threshold)
1378{
1379	struct per_pid *p;
1380	struct per_pidcomm *c;
1381	int count = 0;
1382
1383	p = tchart->all_data;
1384	while (p) {
1385		p->display = 0;
1386		if (p->start_time == 1)
1387			p->start_time = tchart->first_time;
1388
1389		/* no exit marker, task kept running to the end */
1390		if (p->end_time == 0)
1391			p->end_time = tchart->last_time;
1392		if (p->total_time >= threshold)
1393			p->display = 1;
1394
1395		c = p->all;
1396
1397		while (c) {
1398			c->display = 0;
1399
1400			if (c->start_time == 1)
1401				c->start_time = tchart->first_time;
1402
1403			if (c->total_time >= threshold) {
1404				c->display = 1;
1405				count++;
1406			}
1407
1408			if (c->end_time == 0)
1409				c->end_time = tchart->last_time;
1410
1411			c = c->next;
1412		}
1413		p = p->next;
1414	}
1415	return count;
1416}
1417
1418static int determine_display_io_tasks(struct timechart *timechart, u64 threshold)
1419{
1420	struct per_pid *p;
1421	struct per_pidcomm *c;
1422	int count = 0;
1423
1424	p = timechart->all_data;
1425	while (p) {
1426		/* no exit marker, task kept running to the end */
1427		if (p->end_time == 0)
1428			p->end_time = timechart->last_time;
1429
1430		c = p->all;
1431
1432		while (c) {
1433			c->display = 0;
1434
1435			if (c->total_bytes >= threshold) {
1436				c->display = 1;
1437				count++;
1438			}
1439
1440			if (c->end_time == 0)
1441				c->end_time = timechart->last_time;
1442
1443			c = c->next;
1444		}
1445		p = p->next;
1446	}
1447	return count;
1448}
1449
1450#define BYTES_THRESH (1 * 1024 * 1024)
1451#define TIME_THRESH 10000000
1452
1453static void write_svg_file(struct timechart *tchart, const char *filename)
1454{
1455	u64 i;
1456	int count;
1457	int thresh = tchart->io_events ? BYTES_THRESH : TIME_THRESH;
1458
1459	if (tchart->power_only)
1460		tchart->proc_num = 0;
1461
1462	/* We'd like to show at least proc_num tasks;
1463	 * be less picky if we have fewer */
1464	do {
1465		if (process_filter)
1466			count = determine_display_tasks_filtered(tchart);
1467		else if (tchart->io_events)
1468			count = determine_display_io_tasks(tchart, thresh);
1469		else
1470			count = determine_display_tasks(tchart, thresh);
1471		thresh /= 10;
1472	} while (!process_filter && thresh && count < tchart->proc_num);
1473
1474	if (!tchart->proc_num)
1475		count = 0;
1476
1477	if (tchart->io_events) {
1478		open_svg(filename, 0, count, tchart->first_time, tchart->last_time);
1479
1480		svg_time_grid(0.5);
1481		svg_io_legenda();
1482
1483		draw_io_bars(tchart);
1484	} else {
1485		open_svg(filename, tchart->numcpus, count, tchart->first_time, tchart->last_time);
1486
1487		svg_time_grid(0);
1488
1489		svg_legenda();
1490
1491		for (i = 0; i < tchart->numcpus; i++)
1492			svg_cpu_box(i, tchart->max_freq, tchart->turbo_frequency);
1493
1494		draw_cpu_usage(tchart);
1495		if (tchart->proc_num)
1496			draw_process_bars(tchart);
1497		if (!tchart->tasks_only)
1498			draw_c_p_states(tchart);
1499		if (tchart->proc_num)
1500			draw_wakeups(tchart);
1501	}
1502
1503	svg_close();
1504}
1505
1506static int process_header(struct perf_file_section *section __maybe_unused,
1507			  struct perf_header *ph,
1508			  int feat,
1509			  int fd __maybe_unused,
1510			  void *data)
1511{
1512	struct timechart *tchart = data;
1513
1514	switch (feat) {
1515	case HEADER_NRCPUS:
1516		tchart->numcpus = ph->env.nr_cpus_avail;
1517		break;
1518
1519	case HEADER_CPU_TOPOLOGY:
1520		if (!tchart->topology)
1521			break;
1522
1523		if (svg_build_topology_map(ph->env.sibling_cores,
1524					   ph->env.nr_sibling_cores,
1525					   ph->env.sibling_threads,
1526					   ph->env.nr_sibling_threads))
1527			fprintf(stderr, "problem building topology\n");
1528		break;
1529
1530	default:
1531		break;
1532	}
1533
1534	return 0;
1535}
1536
1537static int __cmd_timechart(struct timechart *tchart, const char *output_name)
1538{
1539	const struct perf_evsel_str_handler power_tracepoints[] = {
1540		{ "power:cpu_idle",		process_sample_cpu_idle },
1541		{ "power:cpu_frequency",	process_sample_cpu_frequency },
1542		{ "sched:sched_wakeup",		process_sample_sched_wakeup },
1543		{ "sched:sched_switch",		process_sample_sched_switch },
1544#ifdef SUPPORT_OLD_POWER_EVENTS
1545		{ "power:power_start",		process_sample_power_start },
1546		{ "power:power_end",		process_sample_power_end },
1547		{ "power:power_frequency",	process_sample_power_frequency },
1548#endif
1549
1550		{ "syscalls:sys_enter_read",		process_enter_read },
1551		{ "syscalls:sys_enter_pread64",		process_enter_read },
1552		{ "syscalls:sys_enter_readv",		process_enter_read },
1553		{ "syscalls:sys_enter_preadv",		process_enter_read },
1554		{ "syscalls:sys_enter_write",		process_enter_write },
1555		{ "syscalls:sys_enter_pwrite64",	process_enter_write },
1556		{ "syscalls:sys_enter_writev",		process_enter_write },
1557		{ "syscalls:sys_enter_pwritev",		process_enter_write },
1558		{ "syscalls:sys_enter_sync",		process_enter_sync },
1559		{ "syscalls:sys_enter_sync_file_range",	process_enter_sync },
1560		{ "syscalls:sys_enter_fsync",		process_enter_sync },
1561		{ "syscalls:sys_enter_msync",		process_enter_sync },
1562		{ "syscalls:sys_enter_recvfrom",	process_enter_rx },
1563		{ "syscalls:sys_enter_recvmmsg",	process_enter_rx },
1564		{ "syscalls:sys_enter_recvmsg",		process_enter_rx },
1565		{ "syscalls:sys_enter_sendto",		process_enter_tx },
1566		{ "syscalls:sys_enter_sendmsg",		process_enter_tx },
1567		{ "syscalls:sys_enter_sendmmsg",	process_enter_tx },
1568		{ "syscalls:sys_enter_epoll_pwait",	process_enter_poll },
1569		{ "syscalls:sys_enter_epoll_wait",	process_enter_poll },
1570		{ "syscalls:sys_enter_poll",		process_enter_poll },
1571		{ "syscalls:sys_enter_ppoll",		process_enter_poll },
1572		{ "syscalls:sys_enter_pselect6",	process_enter_poll },
1573		{ "syscalls:sys_enter_select",		process_enter_poll },
1574
1575		{ "syscalls:sys_exit_read",		process_exit_read },
1576		{ "syscalls:sys_exit_pread64",		process_exit_read },
1577		{ "syscalls:sys_exit_readv",		process_exit_read },
1578		{ "syscalls:sys_exit_preadv",		process_exit_read },
1579		{ "syscalls:sys_exit_write",		process_exit_write },
1580		{ "syscalls:sys_exit_pwrite64",		process_exit_write },
1581		{ "syscalls:sys_exit_writev",		process_exit_write },
1582		{ "syscalls:sys_exit_pwritev",		process_exit_write },
1583		{ "syscalls:sys_exit_sync",		process_exit_sync },
1584		{ "syscalls:sys_exit_sync_file_range",	process_exit_sync },
1585		{ "syscalls:sys_exit_fsync",		process_exit_sync },
1586		{ "syscalls:sys_exit_msync",		process_exit_sync },
1587		{ "syscalls:sys_exit_recvfrom",		process_exit_rx },
1588		{ "syscalls:sys_exit_recvmmsg",		process_exit_rx },
1589		{ "syscalls:sys_exit_recvmsg",		process_exit_rx },
1590		{ "syscalls:sys_exit_sendto",		process_exit_tx },
1591		{ "syscalls:sys_exit_sendmsg",		process_exit_tx },
1592		{ "syscalls:sys_exit_sendmmsg",		process_exit_tx },
1593		{ "syscalls:sys_exit_epoll_pwait",	process_exit_poll },
1594		{ "syscalls:sys_exit_epoll_wait",	process_exit_poll },
1595		{ "syscalls:sys_exit_poll",		process_exit_poll },
1596		{ "syscalls:sys_exit_ppoll",		process_exit_poll },
1597		{ "syscalls:sys_exit_pselect6",		process_exit_poll },
1598		{ "syscalls:sys_exit_select",		process_exit_poll },
1599	};
1600	struct perf_data_file file = {
1601		.path = input_name,
1602		.mode = PERF_DATA_MODE_READ,
1603		.force = tchart->force,
1604	};
1605
1606	struct perf_session *session = perf_session__new(&file, false,
1607							 &tchart->tool);
1608	int ret = -EINVAL;
1609
1610	if (session == NULL)
1611		return -1;
1612
1613	symbol__init(&session->header.env);
1614
1615	(void)perf_header__process_sections(&session->header,
1616					    perf_data_file__fd(session->file),
1617					    tchart,
1618					    process_header);
1619
1620	if (!perf_session__has_traces(session, "timechart record"))
1621		goto out_delete;
1622
1623	if (perf_session__set_tracepoints_handlers(session,
1624						   power_tracepoints)) {
1625		pr_err("Initializing session tracepoint handlers failed\n");
1626		goto out_delete;
1627	}
1628
1629	ret = perf_session__process_events(session);
1630	if (ret)
1631		goto out_delete;
1632
1633	end_sample_processing(tchart);
1634
1635	sort_pids(tchart);
1636
1637	write_svg_file(tchart, output_name);
1638
1639	pr_info("Written %2.1f seconds of trace to %s.\n",
1640		(tchart->last_time - tchart->first_time) / 1000000000.0, output_name);
1641out_delete:
1642	perf_session__delete(session);
1643	return ret;
1644}
1645
1646static int timechart__io_record(int argc, const char **argv)
1647{
1648	unsigned int rec_argc, i;
1649	const char **rec_argv;
1650	const char **p;
1651	char *filter = NULL;
1652
1653	const char * const common_args[] = {
1654		"record", "-a", "-R", "-c", "1",
1655	};
1656	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1657
1658	const char * const disk_events[] = {
1659		"syscalls:sys_enter_read",
1660		"syscalls:sys_enter_pread64",
1661		"syscalls:sys_enter_readv",
1662		"syscalls:sys_enter_preadv",
1663		"syscalls:sys_enter_write",
1664		"syscalls:sys_enter_pwrite64",
1665		"syscalls:sys_enter_writev",
1666		"syscalls:sys_enter_pwritev",
1667		"syscalls:sys_enter_sync",
1668		"syscalls:sys_enter_sync_file_range",
1669		"syscalls:sys_enter_fsync",
1670		"syscalls:sys_enter_msync",
1671
1672		"syscalls:sys_exit_read",
1673		"syscalls:sys_exit_pread64",
1674		"syscalls:sys_exit_readv",
1675		"syscalls:sys_exit_preadv",
1676		"syscalls:sys_exit_write",
1677		"syscalls:sys_exit_pwrite64",
1678		"syscalls:sys_exit_writev",
1679		"syscalls:sys_exit_pwritev",
1680		"syscalls:sys_exit_sync",
1681		"syscalls:sys_exit_sync_file_range",
1682		"syscalls:sys_exit_fsync",
1683		"syscalls:sys_exit_msync",
1684	};
1685	unsigned int disk_events_nr = ARRAY_SIZE(disk_events);
1686
1687	const char * const net_events[] = {
1688		"syscalls:sys_enter_recvfrom",
1689		"syscalls:sys_enter_recvmmsg",
1690		"syscalls:sys_enter_recvmsg",
1691		"syscalls:sys_enter_sendto",
1692		"syscalls:sys_enter_sendmsg",
1693		"syscalls:sys_enter_sendmmsg",
1694
1695		"syscalls:sys_exit_recvfrom",
1696		"syscalls:sys_exit_recvmmsg",
1697		"syscalls:sys_exit_recvmsg",
1698		"syscalls:sys_exit_sendto",
1699		"syscalls:sys_exit_sendmsg",
1700		"syscalls:sys_exit_sendmmsg",
1701	};
1702	unsigned int net_events_nr = ARRAY_SIZE(net_events);
1703
1704	const char * const poll_events[] = {
1705		"syscalls:sys_enter_epoll_pwait",
1706		"syscalls:sys_enter_epoll_wait",
1707		"syscalls:sys_enter_poll",
1708		"syscalls:sys_enter_ppoll",
1709		"syscalls:sys_enter_pselect6",
1710		"syscalls:sys_enter_select",
1711
1712		"syscalls:sys_exit_epoll_pwait",
1713		"syscalls:sys_exit_epoll_wait",
1714		"syscalls:sys_exit_poll",
1715		"syscalls:sys_exit_ppoll",
1716		"syscalls:sys_exit_pselect6",
1717		"syscalls:sys_exit_select",
1718	};
1719	unsigned int poll_events_nr = ARRAY_SIZE(poll_events);
1720
1721	rec_argc = common_args_nr +
1722		disk_events_nr * 4 +
1723		net_events_nr * 4 +
1724		poll_events_nr * 4 +
1725		argc;
1726	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1727
1728	if (rec_argv == NULL)
1729		return -ENOMEM;
1730
1731	if (asprintf(&filter, "common_pid != %d", getpid()) < 0)
1732		return -ENOMEM;
1733
1734	p = rec_argv;
1735	for (i = 0; i < common_args_nr; i++)
1736		*p++ = strdup(common_args[i]);
1737
1738	for (i = 0; i < disk_events_nr; i++) {
1739		if (!is_valid_tracepoint(disk_events[i])) {
1740			rec_argc -= 4;
1741			continue;
1742		}
1743
1744		*p++ = "-e";
1745		*p++ = strdup(disk_events[i]);
1746		*p++ = "--filter";
1747		*p++ = filter;
1748	}
1749	for (i = 0; i < net_events_nr; i++) {
1750		if (!is_valid_tracepoint(net_events[i])) {
1751			rec_argc -= 4;
1752			continue;
1753		}
1754
1755		*p++ = "-e";
1756		*p++ = strdup(net_events[i]);
1757		*p++ = "--filter";
1758		*p++ = filter;
1759	}
1760	for (i = 0; i < poll_events_nr; i++) {
1761		if (!is_valid_tracepoint(poll_events[i])) {
1762			rec_argc -= 4;
1763			continue;
1764		}
1765
1766		*p++ = "-e";
1767		*p++ = strdup(poll_events[i]);
1768		*p++ = "--filter";
1769		*p++ = filter;
1770	}
1771
1772	for (i = 0; i < (unsigned int)argc; i++)
1773		*p++ = argv[i];
1774
1775	return cmd_record(rec_argc, rec_argv, NULL);
1776}
1777
1778
1779static int timechart__record(struct timechart *tchart, int argc, const char **argv)
1780{
1781	unsigned int rec_argc, i, j;
1782	const char **rec_argv;
1783	const char **p;
1784	unsigned int record_elems;
1785
1786	const char * const common_args[] = {
1787		"record", "-a", "-R", "-c", "1",
1788	};
1789	unsigned int common_args_nr = ARRAY_SIZE(common_args);
1790
1791	const char * const backtrace_args[] = {
1792		"-g",
1793	};
1794	unsigned int backtrace_args_no = ARRAY_SIZE(backtrace_args);
1795
1796	const char * const power_args[] = {
1797		"-e", "power:cpu_frequency",
1798		"-e", "power:cpu_idle",
1799	};
1800	unsigned int power_args_nr = ARRAY_SIZE(power_args);
1801
1802	const char * const old_power_args[] = {
1803#ifdef SUPPORT_OLD_POWER_EVENTS
1804		"-e", "power:power_start",
1805		"-e", "power:power_end",
1806		"-e", "power:power_frequency",
1807#endif
1808	};
1809	unsigned int old_power_args_nr = ARRAY_SIZE(old_power_args);
1810
1811	const char * const tasks_args[] = {
1812		"-e", "sched:sched_wakeup",
1813		"-e", "sched:sched_switch",
1814	};
1815	unsigned int tasks_args_nr = ARRAY_SIZE(tasks_args);
1816
1817#ifdef SUPPORT_OLD_POWER_EVENTS
1818	if (!is_valid_tracepoint("power:cpu_idle") &&
1819	    is_valid_tracepoint("power:power_start")) {
1820		use_old_power_events = 1;
1821		power_args_nr = 0;
1822	} else {
1823		old_power_args_nr = 0;
1824	}
1825#endif
1826
1827	if (tchart->power_only)
1828		tasks_args_nr = 0;
1829
1830	if (tchart->tasks_only) {
1831		power_args_nr = 0;
1832		old_power_args_nr = 0;
1833	}
1834
1835	if (!tchart->with_backtrace)
1836		backtrace_args_no = 0;
1837
1838	record_elems = common_args_nr + tasks_args_nr +
1839		power_args_nr + old_power_args_nr + backtrace_args_no;
1840
1841	rec_argc = record_elems + argc;
1842	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1843
1844	if (rec_argv == NULL)
1845		return -ENOMEM;
1846
1847	p = rec_argv;
1848	for (i = 0; i < common_args_nr; i++)
1849		*p++ = strdup(common_args[i]);
1850
1851	for (i = 0; i < backtrace_args_no; i++)
1852		*p++ = strdup(backtrace_args[i]);
1853
1854	for (i = 0; i < tasks_args_nr; i++)
1855		*p++ = strdup(tasks_args[i]);
1856
1857	for (i = 0; i < power_args_nr; i++)
1858		*p++ = strdup(power_args[i]);
1859
1860	for (i = 0; i < old_power_args_nr; i++)
1861		*p++ = strdup(old_power_args[i]);
1862
1863	for (j = 0; j < (unsigned int)argc; j++)
1864		*p++ = argv[j];
1865
1866	return cmd_record(rec_argc, rec_argv, NULL);
1867}
1868
1869static int
1870parse_process(const struct option *opt __maybe_unused, const char *arg,
1871	      int __maybe_unused unset)
1872{
1873	if (arg)
1874		add_process_filter(arg);
1875	return 0;
1876}
1877
1878static int
1879parse_highlight(const struct option *opt __maybe_unused, const char *arg,
1880		int __maybe_unused unset)
1881{
1882	unsigned long duration = strtoul(arg, NULL, 0);
1883
1884	if (svg_highlight || svg_highlight_name)
1885		return -1;
1886
1887	if (duration)
1888		svg_highlight = duration;
1889	else
1890		svg_highlight_name = strdup(arg);
1891
1892	return 0;
1893}
1894
1895static int
1896parse_time(const struct option *opt, const char *arg, int __maybe_unused unset)
1897{
1898	char unit = 'n';
1899	u64 *value = opt->value;
1900
1901	if (sscanf(arg, "%" PRIu64 "%cs", value, &unit) > 0) {
1902		switch (unit) {
1903		case 'm':
1904			*value *= 1000000;
1905			break;
1906		case 'u':
1907			*value *= 1000;
1908			break;
1909		case 'n':
1910			break;
1911		default:
1912			return -1;
1913		}
1914	}
1915
1916	return 0;
1917}
1918
1919int cmd_timechart(int argc, const char **argv,
1920		  const char *prefix __maybe_unused)
1921{
1922	struct timechart tchart = {
1923		.tool = {
1924			.comm		 = process_comm_event,
1925			.fork		 = process_fork_event,
1926			.exit		 = process_exit_event,
1927			.sample		 = process_sample_event,
1928			.ordered_events	 = true,
1929		},
1930		.proc_num = 15,
1931		.min_time = 1000000,
1932		.merge_dist = 1000,
1933	};
1934	const char *output_name = "output.svg";
1935	const struct option timechart_options[] = {
1936	OPT_STRING('i', "input", &input_name, "file", "input file name"),
1937	OPT_STRING('o', "output", &output_name, "file", "output file name"),
1938	OPT_INTEGER('w', "width", &svg_page_width, "page width"),
1939	OPT_CALLBACK(0, "highlight", NULL, "duration or task name",
1940		      "highlight tasks. Pass duration in ns or process name.",
1941		       parse_highlight),
1942	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1943	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only,
1944		    "output processes data only"),
1945	OPT_CALLBACK('p', "process", NULL, "process",
1946		      "process selector. Pass a pid or process name.",
1947		       parse_process),
1948	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
1949		    "Look for files with symbols relative to this directory"),
1950	OPT_INTEGER('n', "proc-num", &tchart.proc_num,
1951		    "min. number of tasks to print"),
1952	OPT_BOOLEAN('t', "topology", &tchart.topology,
1953		    "sort CPUs according to topology"),
1954	OPT_BOOLEAN(0, "io-skip-eagain", &tchart.skip_eagain,
1955		    "skip EAGAIN errors"),
1956	OPT_CALLBACK(0, "io-min-time", &tchart.min_time, "time",
1957		     "all IO faster than min-time will visually appear longer",
1958		     parse_time),
1959	OPT_CALLBACK(0, "io-merge-dist", &tchart.merge_dist, "time",
1960		     "merge events that are merge-dist us apart",
1961		     parse_time),
1962	OPT_BOOLEAN('f', "force", &tchart.force, "don't complain, do it"),
1963	OPT_END()
1964	};
1965	const char * const timechart_subcommands[] = { "record", NULL };
1966	const char *timechart_usage[] = {
1967		"perf timechart [<options>] {record}",
1968		NULL
1969	};
1970
1971	const struct option timechart_record_options[] = {
1972	OPT_BOOLEAN('P', "power-only", &tchart.power_only, "output power data only"),
1973	OPT_BOOLEAN('T', "tasks-only", &tchart.tasks_only,
1974		    "output processes data only"),
1975	OPT_BOOLEAN('I', "io-only", &tchart.io_only,
1976		    "record only IO data"),
1977	OPT_BOOLEAN('g', "callchain", &tchart.with_backtrace, "record callchain"),
1978	OPT_END()
1979	};
1980	const char * const timechart_record_usage[] = {
1981		"perf timechart record [<options>]",
1982		NULL
1983	};
1984	argc = parse_options_subcommand(argc, argv, timechart_options, timechart_subcommands,
1985			timechart_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1986
1987	if (tchart.power_only && tchart.tasks_only) {
1988		pr_err("-P and -T options cannot be used at the same time.\n");
1989		return -1;
1990	}
1991
1992	if (argc && !strncmp(argv[0], "rec", 3)) {
1993		argc = parse_options(argc, argv, timechart_record_options,
1994				     timechart_record_usage,
1995				     PARSE_OPT_STOP_AT_NON_OPTION);
1996
1997		if (tchart.power_only && tchart.tasks_only) {
1998			pr_err("-P and -T options cannot be used at the same time.\n");
1999			return -1;
2000		}
2001
2002		if (tchart.io_only)
2003			return timechart__io_record(argc, argv);
2004		else
2005			return timechart__record(&tchart, argc, argv);
2006	} else if (argc)
2007		usage_with_options(timechart_usage, timechart_options);
2008
2009	setup_pager();
2010
2011	return __cmd_timechart(&tchart, output_name);
2012}
2013