]>
Commit | Line | Data |
---|---|---|
10274989 AV |
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 "builtin.h" | |
16 | ||
17 | #include "util/util.h" | |
18 | ||
19 | #include "util/color.h" | |
20 | #include <linux/list.h> | |
21 | #include "util/cache.h" | |
22 | #include <linux/rbtree.h> | |
23 | #include "util/symbol.h" | |
24 | #include "util/string.h" | |
25 | #include "util/callchain.h" | |
26 | #include "util/strlist.h" | |
27 | ||
28 | #include "perf.h" | |
29 | #include "util/header.h" | |
30 | #include "util/parse-options.h" | |
31 | #include "util/parse-events.h" | |
32 | #include "util/svghelper.h" | |
33 | ||
34 | static char const *input_name = "perf.data"; | |
35 | static char const *output_name = "output.svg"; | |
36 | ||
37 | ||
38 | static unsigned long page_size; | |
39 | static unsigned long mmap_window = 32; | |
40 | static u64 sample_type; | |
41 | ||
42 | static unsigned int numcpus; | |
43 | static u64 min_freq; /* Lowest CPU frequency seen */ | |
44 | static u64 max_freq; /* Highest CPU frequency seen */ | |
45 | static u64 turbo_frequency; | |
46 | ||
47 | static u64 first_time, last_time; | |
48 | ||
39a90a8e AV |
49 | static int power_only; |
50 | ||
10274989 AV |
51 | |
52 | static struct perf_header *header; | |
53 | ||
54 | struct per_pid; | |
55 | struct per_pidcomm; | |
56 | ||
57 | struct cpu_sample; | |
58 | struct power_event; | |
59 | struct wake_event; | |
60 | ||
61 | struct sample_wrapper; | |
62 | ||
63 | /* | |
64 | * Datastructure layout: | |
65 | * We keep an list of "pid"s, matching the kernels notion of a task struct. | |
66 | * Each "pid" entry, has a list of "comm"s. | |
67 | * this is because we want to track different programs different, while | |
68 | * exec will reuse the original pid (by design). | |
69 | * Each comm has a list of samples that will be used to draw | |
70 | * final graph. | |
71 | */ | |
72 | ||
73 | struct per_pid { | |
74 | struct per_pid *next; | |
75 | ||
76 | int pid; | |
77 | int ppid; | |
78 | ||
79 | u64 start_time; | |
80 | u64 end_time; | |
81 | u64 total_time; | |
82 | int display; | |
83 | ||
84 | struct per_pidcomm *all; | |
85 | struct per_pidcomm *current; | |
86 | ||
87 | int painted; | |
88 | }; | |
89 | ||
90 | ||
91 | struct per_pidcomm { | |
92 | struct per_pidcomm *next; | |
93 | ||
94 | u64 start_time; | |
95 | u64 end_time; | |
96 | u64 total_time; | |
97 | ||
98 | int Y; | |
99 | int display; | |
100 | ||
101 | long state; | |
102 | u64 state_since; | |
103 | ||
104 | char *comm; | |
105 | ||
106 | struct cpu_sample *samples; | |
107 | }; | |
108 | ||
109 | struct sample_wrapper { | |
110 | struct sample_wrapper *next; | |
111 | ||
112 | u64 timestamp; | |
113 | unsigned char data[0]; | |
114 | }; | |
115 | ||
116 | #define TYPE_NONE 0 | |
117 | #define TYPE_RUNNING 1 | |
118 | #define TYPE_WAITING 2 | |
119 | #define TYPE_BLOCKED 3 | |
120 | ||
121 | struct cpu_sample { | |
122 | struct cpu_sample *next; | |
123 | ||
124 | u64 start_time; | |
125 | u64 end_time; | |
126 | int type; | |
127 | int cpu; | |
128 | }; | |
129 | ||
130 | static struct per_pid *all_data; | |
131 | ||
132 | #define CSTATE 1 | |
133 | #define PSTATE 2 | |
134 | ||
135 | struct power_event { | |
136 | struct power_event *next; | |
137 | int type; | |
138 | int state; | |
139 | u64 start_time; | |
140 | u64 end_time; | |
141 | int cpu; | |
142 | }; | |
143 | ||
144 | struct wake_event { | |
145 | struct wake_event *next; | |
146 | int waker; | |
147 | int wakee; | |
148 | u64 time; | |
149 | }; | |
150 | ||
151 | static struct power_event *power_events; | |
152 | static struct wake_event *wake_events; | |
153 | ||
154 | struct sample_wrapper *all_samples; | |
155 | ||
156 | static struct per_pid *find_create_pid(int pid) | |
157 | { | |
158 | struct per_pid *cursor = all_data; | |
159 | ||
160 | while (cursor) { | |
161 | if (cursor->pid == pid) | |
162 | return cursor; | |
163 | cursor = cursor->next; | |
164 | } | |
165 | cursor = malloc(sizeof(struct per_pid)); | |
166 | assert(cursor != NULL); | |
167 | memset(cursor, 0, sizeof(struct per_pid)); | |
168 | cursor->pid = pid; | |
169 | cursor->next = all_data; | |
170 | all_data = cursor; | |
171 | return cursor; | |
172 | } | |
173 | ||
174 | static void pid_set_comm(int pid, char *comm) | |
175 | { | |
176 | struct per_pid *p; | |
177 | struct per_pidcomm *c; | |
178 | p = find_create_pid(pid); | |
179 | c = p->all; | |
180 | while (c) { | |
181 | if (c->comm && strcmp(c->comm, comm) == 0) { | |
182 | p->current = c; | |
183 | return; | |
184 | } | |
185 | if (!c->comm) { | |
186 | c->comm = strdup(comm); | |
187 | p->current = c; | |
188 | return; | |
189 | } | |
190 | c = c->next; | |
191 | } | |
192 | c = malloc(sizeof(struct per_pidcomm)); | |
193 | assert(c != NULL); | |
194 | memset(c, 0, sizeof(struct per_pidcomm)); | |
195 | c->comm = strdup(comm); | |
196 | p->current = c; | |
197 | c->next = p->all; | |
198 | p->all = c; | |
199 | } | |
200 | ||
201 | static void pid_fork(int pid, int ppid, u64 timestamp) | |
202 | { | |
203 | struct per_pid *p, *pp; | |
204 | p = find_create_pid(pid); | |
205 | pp = find_create_pid(ppid); | |
206 | p->ppid = ppid; | |
207 | if (pp->current && pp->current->comm && !p->current) | |
208 | pid_set_comm(pid, pp->current->comm); | |
209 | ||
210 | p->start_time = timestamp; | |
211 | if (p->current) { | |
212 | p->current->start_time = timestamp; | |
213 | p->current->state_since = timestamp; | |
214 | } | |
215 | } | |
216 | ||
217 | static void pid_exit(int pid, u64 timestamp) | |
218 | { | |
219 | struct per_pid *p; | |
220 | p = find_create_pid(pid); | |
221 | p->end_time = timestamp; | |
222 | if (p->current) | |
223 | p->current->end_time = timestamp; | |
224 | } | |
225 | ||
226 | static void | |
227 | pid_put_sample(int pid, int type, unsigned int cpu, u64 start, u64 end) | |
228 | { | |
229 | struct per_pid *p; | |
230 | struct per_pidcomm *c; | |
231 | struct cpu_sample *sample; | |
232 | ||
233 | p = find_create_pid(pid); | |
234 | c = p->current; | |
235 | if (!c) { | |
236 | c = malloc(sizeof(struct per_pidcomm)); | |
237 | assert(c != NULL); | |
238 | memset(c, 0, sizeof(struct per_pidcomm)); | |
239 | p->current = c; | |
240 | c->next = p->all; | |
241 | p->all = c; | |
242 | } | |
243 | ||
244 | sample = malloc(sizeof(struct cpu_sample)); | |
245 | assert(sample != NULL); | |
246 | memset(sample, 0, sizeof(struct cpu_sample)); | |
247 | sample->start_time = start; | |
248 | sample->end_time = end; | |
249 | sample->type = type; | |
250 | sample->next = c->samples; | |
251 | sample->cpu = cpu; | |
252 | c->samples = sample; | |
253 | ||
254 | if (sample->type == TYPE_RUNNING && end > start && start > 0) { | |
255 | c->total_time += (end-start); | |
256 | p->total_time += (end-start); | |
257 | } | |
258 | ||
259 | if (c->start_time == 0 || c->start_time > start) | |
260 | c->start_time = start; | |
261 | if (p->start_time == 0 || p->start_time > start) | |
262 | p->start_time = start; | |
263 | ||
264 | if (cpu > numcpus) | |
265 | numcpus = cpu; | |
266 | } | |
267 | ||
268 | #define MAX_CPUS 4096 | |
269 | ||
270 | static u64 cpus_cstate_start_times[MAX_CPUS]; | |
271 | static int cpus_cstate_state[MAX_CPUS]; | |
272 | static u64 cpus_pstate_start_times[MAX_CPUS]; | |
273 | static u64 cpus_pstate_state[MAX_CPUS]; | |
274 | ||
275 | static int | |
276 | process_comm_event(event_t *event) | |
277 | { | |
278 | pid_set_comm(event->comm.pid, event->comm.comm); | |
279 | return 0; | |
280 | } | |
281 | static int | |
282 | process_fork_event(event_t *event) | |
283 | { | |
284 | pid_fork(event->fork.pid, event->fork.ppid, event->fork.time); | |
285 | return 0; | |
286 | } | |
287 | ||
288 | static int | |
289 | process_exit_event(event_t *event) | |
290 | { | |
291 | pid_exit(event->fork.pid, event->fork.time); | |
292 | return 0; | |
293 | } | |
294 | ||
295 | struct trace_entry { | |
296 | u32 size; | |
297 | unsigned short type; | |
298 | unsigned char flags; | |
299 | unsigned char preempt_count; | |
300 | int pid; | |
301 | int tgid; | |
302 | }; | |
303 | ||
304 | struct power_entry { | |
305 | struct trace_entry te; | |
306 | s64 type; | |
307 | s64 value; | |
308 | }; | |
309 | ||
310 | #define TASK_COMM_LEN 16 | |
311 | struct wakeup_entry { | |
312 | struct trace_entry te; | |
313 | char comm[TASK_COMM_LEN]; | |
314 | int pid; | |
315 | int prio; | |
316 | int success; | |
317 | }; | |
318 | ||
319 | /* | |
320 | * trace_flag_type is an enumeration that holds different | |
321 | * states when a trace occurs. These are: | |
322 | * IRQS_OFF - interrupts were disabled | |
323 | * IRQS_NOSUPPORT - arch does not support irqs_disabled_flags | |
324 | * NEED_RESCED - reschedule is requested | |
325 | * HARDIRQ - inside an interrupt handler | |
326 | * SOFTIRQ - inside a softirq handler | |
327 | */ | |
328 | enum trace_flag_type { | |
329 | TRACE_FLAG_IRQS_OFF = 0x01, | |
330 | TRACE_FLAG_IRQS_NOSUPPORT = 0x02, | |
331 | TRACE_FLAG_NEED_RESCHED = 0x04, | |
332 | TRACE_FLAG_HARDIRQ = 0x08, | |
333 | TRACE_FLAG_SOFTIRQ = 0x10, | |
334 | }; | |
335 | ||
336 | ||
337 | ||
338 | struct sched_switch { | |
339 | struct trace_entry te; | |
340 | char prev_comm[TASK_COMM_LEN]; | |
341 | int prev_pid; | |
342 | int prev_prio; | |
343 | long prev_state; /* Arjan weeps. */ | |
344 | char next_comm[TASK_COMM_LEN]; | |
345 | int next_pid; | |
346 | int next_prio; | |
347 | }; | |
348 | ||
349 | static void c_state_start(int cpu, u64 timestamp, int state) | |
350 | { | |
351 | cpus_cstate_start_times[cpu] = timestamp; | |
352 | cpus_cstate_state[cpu] = state; | |
353 | } | |
354 | ||
355 | static void c_state_end(int cpu, u64 timestamp) | |
356 | { | |
357 | struct power_event *pwr; | |
358 | pwr = malloc(sizeof(struct power_event)); | |
359 | if (!pwr) | |
360 | return; | |
361 | memset(pwr, 0, sizeof(struct power_event)); | |
362 | ||
363 | pwr->state = cpus_cstate_state[cpu]; | |
364 | pwr->start_time = cpus_cstate_start_times[cpu]; | |
365 | pwr->end_time = timestamp; | |
366 | pwr->cpu = cpu; | |
367 | pwr->type = CSTATE; | |
368 | pwr->next = power_events; | |
369 | ||
370 | power_events = pwr; | |
371 | } | |
372 | ||
373 | static void p_state_change(int cpu, u64 timestamp, u64 new_freq) | |
374 | { | |
375 | struct power_event *pwr; | |
376 | pwr = malloc(sizeof(struct power_event)); | |
377 | ||
378 | if (new_freq > 8000000) /* detect invalid data */ | |
379 | return; | |
380 | ||
381 | if (!pwr) | |
382 | return; | |
383 | memset(pwr, 0, sizeof(struct power_event)); | |
384 | ||
385 | pwr->state = cpus_pstate_state[cpu]; | |
386 | pwr->start_time = cpus_pstate_start_times[cpu]; | |
387 | pwr->end_time = timestamp; | |
388 | pwr->cpu = cpu; | |
389 | pwr->type = PSTATE; | |
390 | pwr->next = power_events; | |
391 | ||
392 | if (!pwr->start_time) | |
393 | pwr->start_time = first_time; | |
394 | ||
395 | power_events = pwr; | |
396 | ||
397 | cpus_pstate_state[cpu] = new_freq; | |
398 | cpus_pstate_start_times[cpu] = timestamp; | |
399 | ||
400 | if ((u64)new_freq > max_freq) | |
401 | max_freq = new_freq; | |
402 | ||
403 | if (new_freq < min_freq || min_freq == 0) | |
404 | min_freq = new_freq; | |
405 | ||
406 | if (new_freq == max_freq - 1000) | |
407 | turbo_frequency = max_freq; | |
408 | } | |
409 | ||
410 | static void | |
411 | sched_wakeup(int cpu, u64 timestamp, int pid, struct trace_entry *te) | |
412 | { | |
413 | struct wake_event *we; | |
414 | struct per_pid *p; | |
415 | struct wakeup_entry *wake = (void *)te; | |
416 | ||
417 | we = malloc(sizeof(struct wake_event)); | |
418 | if (!we) | |
419 | return; | |
420 | ||
421 | memset(we, 0, sizeof(struct wake_event)); | |
422 | we->time = timestamp; | |
423 | we->waker = pid; | |
424 | ||
425 | if ((te->flags & TRACE_FLAG_HARDIRQ) || (te->flags & TRACE_FLAG_SOFTIRQ)) | |
426 | we->waker = -1; | |
427 | ||
428 | we->wakee = wake->pid; | |
429 | we->next = wake_events; | |
430 | wake_events = we; | |
431 | p = find_create_pid(we->wakee); | |
432 | ||
433 | if (p && p->current && p->current->state == TYPE_NONE) { | |
434 | p->current->state_since = timestamp; | |
435 | p->current->state = TYPE_WAITING; | |
436 | } | |
437 | if (p && p->current && p->current->state == TYPE_BLOCKED) { | |
438 | pid_put_sample(p->pid, p->current->state, cpu, p->current->state_since, timestamp); | |
439 | p->current->state_since = timestamp; | |
440 | p->current->state = TYPE_WAITING; | |
441 | } | |
442 | } | |
443 | ||
444 | static void sched_switch(int cpu, u64 timestamp, struct trace_entry *te) | |
445 | { | |
446 | struct per_pid *p = NULL, *prev_p; | |
447 | struct sched_switch *sw = (void *)te; | |
448 | ||
449 | ||
450 | prev_p = find_create_pid(sw->prev_pid); | |
451 | ||
452 | p = find_create_pid(sw->next_pid); | |
453 | ||
454 | if (prev_p->current && prev_p->current->state != TYPE_NONE) | |
455 | pid_put_sample(sw->prev_pid, TYPE_RUNNING, cpu, prev_p->current->state_since, timestamp); | |
456 | if (p && p->current) { | |
457 | if (p->current->state != TYPE_NONE) | |
458 | pid_put_sample(sw->next_pid, p->current->state, cpu, p->current->state_since, timestamp); | |
459 | ||
460 | p->current->state_since = timestamp; | |
461 | p->current->state = TYPE_RUNNING; | |
462 | } | |
463 | ||
464 | if (prev_p->current) { | |
465 | prev_p->current->state = TYPE_NONE; | |
466 | prev_p->current->state_since = timestamp; | |
467 | if (sw->prev_state & 2) | |
468 | prev_p->current->state = TYPE_BLOCKED; | |
469 | if (sw->prev_state == 0) | |
470 | prev_p->current->state = TYPE_WAITING; | |
471 | } | |
472 | } | |
473 | ||
474 | ||
475 | static int | |
476 | process_sample_event(event_t *event) | |
477 | { | |
478 | int cursor = 0; | |
479 | u64 addr = 0; | |
480 | u64 stamp = 0; | |
481 | u32 cpu = 0; | |
482 | u32 pid = 0; | |
483 | struct trace_entry *te; | |
484 | ||
485 | if (sample_type & PERF_SAMPLE_IP) | |
486 | cursor++; | |
487 | ||
488 | if (sample_type & PERF_SAMPLE_TID) { | |
489 | pid = event->sample.array[cursor]>>32; | |
490 | cursor++; | |
491 | } | |
492 | if (sample_type & PERF_SAMPLE_TIME) { | |
493 | stamp = event->sample.array[cursor++]; | |
494 | ||
495 | if (!first_time || first_time > stamp) | |
496 | first_time = stamp; | |
497 | if (last_time < stamp) | |
498 | last_time = stamp; | |
499 | ||
500 | } | |
501 | if (sample_type & PERF_SAMPLE_ADDR) | |
502 | addr = event->sample.array[cursor++]; | |
503 | if (sample_type & PERF_SAMPLE_ID) | |
504 | cursor++; | |
505 | if (sample_type & PERF_SAMPLE_STREAM_ID) | |
506 | cursor++; | |
507 | if (sample_type & PERF_SAMPLE_CPU) | |
508 | cpu = event->sample.array[cursor++] & 0xFFFFFFFF; | |
509 | if (sample_type & PERF_SAMPLE_PERIOD) | |
510 | cursor++; | |
511 | ||
512 | te = (void *)&event->sample.array[cursor]; | |
513 | ||
514 | if (sample_type & PERF_SAMPLE_RAW && te->size > 0) { | |
515 | char *event_str; | |
516 | struct power_entry *pe; | |
517 | ||
518 | pe = (void *)te; | |
519 | ||
520 | event_str = perf_header__find_event(te->type); | |
521 | ||
522 | if (!event_str) | |
523 | return 0; | |
524 | ||
525 | if (strcmp(event_str, "power:power_start") == 0) | |
526 | c_state_start(cpu, stamp, pe->value); | |
527 | ||
528 | if (strcmp(event_str, "power:power_end") == 0) | |
529 | c_state_end(cpu, stamp); | |
530 | ||
531 | if (strcmp(event_str, "power:power_frequency") == 0) | |
532 | p_state_change(cpu, stamp, pe->value); | |
533 | ||
534 | if (strcmp(event_str, "sched:sched_wakeup") == 0) | |
535 | sched_wakeup(cpu, stamp, pid, te); | |
536 | ||
537 | if (strcmp(event_str, "sched:sched_switch") == 0) | |
538 | sched_switch(cpu, stamp, te); | |
539 | } | |
540 | return 0; | |
541 | } | |
542 | ||
543 | /* | |
544 | * After the last sample we need to wrap up the current C/P state | |
545 | * and close out each CPU for these. | |
546 | */ | |
547 | static void end_sample_processing(void) | |
548 | { | |
549 | u64 cpu; | |
550 | struct power_event *pwr; | |
551 | ||
39a90a8e | 552 | for (cpu = 0; cpu <= numcpus; cpu++) { |
10274989 AV |
553 | pwr = malloc(sizeof(struct power_event)); |
554 | if (!pwr) | |
555 | return; | |
556 | memset(pwr, 0, sizeof(struct power_event)); | |
557 | ||
558 | /* C state */ | |
559 | #if 0 | |
560 | pwr->state = cpus_cstate_state[cpu]; | |
561 | pwr->start_time = cpus_cstate_start_times[cpu]; | |
562 | pwr->end_time = last_time; | |
563 | pwr->cpu = cpu; | |
564 | pwr->type = CSTATE; | |
565 | pwr->next = power_events; | |
566 | ||
567 | power_events = pwr; | |
568 | #endif | |
569 | /* P state */ | |
570 | ||
571 | pwr = malloc(sizeof(struct power_event)); | |
572 | if (!pwr) | |
573 | return; | |
574 | memset(pwr, 0, sizeof(struct power_event)); | |
575 | ||
576 | pwr->state = cpus_pstate_state[cpu]; | |
577 | pwr->start_time = cpus_pstate_start_times[cpu]; | |
578 | pwr->end_time = last_time; | |
579 | pwr->cpu = cpu; | |
580 | pwr->type = PSTATE; | |
581 | pwr->next = power_events; | |
582 | ||
583 | if (!pwr->start_time) | |
584 | pwr->start_time = first_time; | |
585 | if (!pwr->state) | |
586 | pwr->state = min_freq; | |
587 | power_events = pwr; | |
588 | } | |
589 | } | |
590 | ||
591 | static u64 sample_time(event_t *event) | |
592 | { | |
593 | int cursor; | |
594 | ||
595 | cursor = 0; | |
596 | if (sample_type & PERF_SAMPLE_IP) | |
597 | cursor++; | |
598 | if (sample_type & PERF_SAMPLE_TID) | |
599 | cursor++; | |
600 | if (sample_type & PERF_SAMPLE_TIME) | |
601 | return event->sample.array[cursor]; | |
602 | return 0; | |
603 | } | |
604 | ||
605 | ||
606 | /* | |
607 | * We first queue all events, sorted backwards by insertion. | |
608 | * The order will get flipped later. | |
609 | */ | |
610 | static int | |
611 | queue_sample_event(event_t *event) | |
612 | { | |
613 | struct sample_wrapper *copy, *prev; | |
614 | int size; | |
615 | ||
616 | size = event->sample.header.size + sizeof(struct sample_wrapper) + 8; | |
617 | ||
618 | copy = malloc(size); | |
619 | if (!copy) | |
620 | return 1; | |
621 | ||
622 | memset(copy, 0, size); | |
623 | ||
624 | copy->next = NULL; | |
625 | copy->timestamp = sample_time(event); | |
626 | ||
627 | memcpy(©->data, event, event->sample.header.size); | |
628 | ||
629 | /* insert in the right place in the list */ | |
630 | ||
631 | if (!all_samples) { | |
632 | /* first sample ever */ | |
633 | all_samples = copy; | |
634 | return 0; | |
635 | } | |
636 | ||
637 | if (all_samples->timestamp < copy->timestamp) { | |
638 | /* insert at the head of the list */ | |
639 | copy->next = all_samples; | |
640 | all_samples = copy; | |
641 | return 0; | |
642 | } | |
643 | ||
644 | prev = all_samples; | |
645 | while (prev->next) { | |
646 | if (prev->next->timestamp < copy->timestamp) { | |
647 | copy->next = prev->next; | |
648 | prev->next = copy; | |
649 | return 0; | |
650 | } | |
651 | prev = prev->next; | |
652 | } | |
653 | /* insert at the end of the list */ | |
654 | prev->next = copy; | |
655 | ||
656 | return 0; | |
657 | } | |
658 | ||
659 | static void sort_queued_samples(void) | |
660 | { | |
661 | struct sample_wrapper *cursor, *next; | |
662 | ||
663 | cursor = all_samples; | |
664 | all_samples = NULL; | |
665 | ||
666 | while (cursor) { | |
667 | next = cursor->next; | |
668 | cursor->next = all_samples; | |
669 | all_samples = cursor; | |
670 | cursor = next; | |
671 | } | |
672 | } | |
673 | ||
674 | /* | |
675 | * Sort the pid datastructure | |
676 | */ | |
677 | static void sort_pids(void) | |
678 | { | |
679 | struct per_pid *new_list, *p, *cursor, *prev; | |
680 | /* sort by ppid first, then by pid, lowest to highest */ | |
681 | ||
682 | new_list = NULL; | |
683 | ||
684 | while (all_data) { | |
685 | p = all_data; | |
686 | all_data = p->next; | |
687 | p->next = NULL; | |
688 | ||
689 | if (new_list == NULL) { | |
690 | new_list = p; | |
691 | p->next = NULL; | |
692 | continue; | |
693 | } | |
694 | prev = NULL; | |
695 | cursor = new_list; | |
696 | while (cursor) { | |
697 | if (cursor->ppid > p->ppid || | |
698 | (cursor->ppid == p->ppid && cursor->pid > p->pid)) { | |
699 | /* must insert before */ | |
700 | if (prev) { | |
701 | p->next = prev->next; | |
702 | prev->next = p; | |
703 | cursor = NULL; | |
704 | continue; | |
705 | } else { | |
706 | p->next = new_list; | |
707 | new_list = p; | |
708 | cursor = NULL; | |
709 | continue; | |
710 | } | |
711 | } | |
712 | ||
713 | prev = cursor; | |
714 | cursor = cursor->next; | |
715 | if (!cursor) | |
716 | prev->next = p; | |
717 | } | |
718 | } | |
719 | all_data = new_list; | |
720 | } | |
721 | ||
722 | ||
723 | static void draw_c_p_states(void) | |
724 | { | |
725 | struct power_event *pwr; | |
726 | pwr = power_events; | |
727 | ||
728 | /* | |
729 | * two pass drawing so that the P state bars are on top of the C state blocks | |
730 | */ | |
731 | while (pwr) { | |
732 | if (pwr->type == CSTATE) | |
733 | svg_cstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state); | |
734 | pwr = pwr->next; | |
735 | } | |
736 | ||
737 | pwr = power_events; | |
738 | while (pwr) { | |
739 | if (pwr->type == PSTATE) { | |
740 | if (!pwr->state) | |
741 | pwr->state = min_freq; | |
742 | svg_pstate(pwr->cpu, pwr->start_time, pwr->end_time, pwr->state); | |
743 | } | |
744 | pwr = pwr->next; | |
745 | } | |
746 | } | |
747 | ||
748 | static void draw_wakeups(void) | |
749 | { | |
750 | struct wake_event *we; | |
751 | struct per_pid *p; | |
752 | struct per_pidcomm *c; | |
753 | ||
754 | we = wake_events; | |
755 | while (we) { | |
756 | int from = 0, to = 0; | |
4f1202c8 | 757 | char *task_from = NULL, *task_to = NULL; |
10274989 AV |
758 | |
759 | /* locate the column of the waker and wakee */ | |
760 | p = all_data; | |
761 | while (p) { | |
762 | if (p->pid == we->waker || p->pid == we->wakee) { | |
763 | c = p->all; | |
764 | while (c) { | |
765 | if (c->Y && c->start_time <= we->time && c->end_time >= we->time) { | |
4f1202c8 | 766 | if (p->pid == we->waker) { |
10274989 | 767 | from = c->Y; |
3bc2a39c | 768 | task_from = strdup(c->comm); |
4f1202c8 AV |
769 | } |
770 | if (p->pid == we->wakee) { | |
10274989 | 771 | to = c->Y; |
3bc2a39c | 772 | task_to = strdup(c->comm); |
4f1202c8 | 773 | } |
10274989 AV |
774 | } |
775 | c = c->next; | |
776 | } | |
3bc2a39c AV |
777 | c = p->all; |
778 | while (c) { | |
779 | if (p->pid == we->waker && !from) { | |
780 | from = c->Y; | |
781 | task_from = strdup(c->comm); | |
782 | } | |
783 | if (p->pid == we->wakee && !to) { | |
784 | to = c->Y; | |
785 | task_to = strdup(c->comm); | |
786 | } | |
787 | c = c->next; | |
788 | } | |
10274989 AV |
789 | } |
790 | p = p->next; | |
791 | } | |
792 | ||
3bc2a39c AV |
793 | if (!task_from) { |
794 | task_from = malloc(40); | |
795 | sprintf(task_from, "[%i]", we->waker); | |
796 | } | |
797 | if (!task_to) { | |
798 | task_to = malloc(40); | |
799 | sprintf(task_to, "[%i]", we->wakee); | |
800 | } | |
801 | ||
10274989 AV |
802 | if (we->waker == -1) |
803 | svg_interrupt(we->time, to); | |
804 | else if (from && to && abs(from - to) == 1) | |
805 | svg_wakeline(we->time, from, to); | |
806 | else | |
4f1202c8 | 807 | svg_partial_wakeline(we->time, from, task_from, to, task_to); |
10274989 | 808 | we = we->next; |
3bc2a39c AV |
809 | |
810 | free(task_from); | |
811 | free(task_to); | |
10274989 AV |
812 | } |
813 | } | |
814 | ||
815 | static void draw_cpu_usage(void) | |
816 | { | |
817 | struct per_pid *p; | |
818 | struct per_pidcomm *c; | |
819 | struct cpu_sample *sample; | |
820 | p = all_data; | |
821 | while (p) { | |
822 | c = p->all; | |
823 | while (c) { | |
824 | sample = c->samples; | |
825 | while (sample) { | |
826 | if (sample->type == TYPE_RUNNING) | |
827 | svg_process(sample->cpu, sample->start_time, sample->end_time, "sample", c->comm); | |
828 | ||
829 | sample = sample->next; | |
830 | } | |
831 | c = c->next; | |
832 | } | |
833 | p = p->next; | |
834 | } | |
835 | } | |
836 | ||
837 | static void draw_process_bars(void) | |
838 | { | |
839 | struct per_pid *p; | |
840 | struct per_pidcomm *c; | |
841 | struct cpu_sample *sample; | |
842 | int Y = 0; | |
843 | ||
844 | Y = 2 * numcpus + 2; | |
845 | ||
846 | p = all_data; | |
847 | while (p) { | |
848 | c = p->all; | |
849 | while (c) { | |
850 | if (!c->display) { | |
851 | c->Y = 0; | |
852 | c = c->next; | |
853 | continue; | |
854 | } | |
855 | ||
a92fe7b3 | 856 | svg_box(Y, c->start_time, c->end_time, "process"); |
10274989 AV |
857 | sample = c->samples; |
858 | while (sample) { | |
859 | if (sample->type == TYPE_RUNNING) | |
a92fe7b3 | 860 | svg_sample(Y, sample->cpu, sample->start_time, sample->end_time); |
10274989 AV |
861 | if (sample->type == TYPE_BLOCKED) |
862 | svg_box(Y, sample->start_time, sample->end_time, "blocked"); | |
863 | if (sample->type == TYPE_WAITING) | |
a92fe7b3 | 864 | svg_waiting(Y, sample->start_time, sample->end_time); |
10274989 AV |
865 | sample = sample->next; |
866 | } | |
867 | ||
868 | if (c->comm) { | |
869 | char comm[256]; | |
870 | if (c->total_time > 5000000000) /* 5 seconds */ | |
871 | sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / 1000000000.0); | |
872 | else | |
873 | sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / 1000000.0); | |
874 | ||
875 | svg_text(Y, c->start_time, comm); | |
876 | } | |
877 | c->Y = Y; | |
878 | Y++; | |
879 | c = c->next; | |
880 | } | |
881 | p = p->next; | |
882 | } | |
883 | } | |
884 | ||
885 | static int determine_display_tasks(u64 threshold) | |
886 | { | |
887 | struct per_pid *p; | |
888 | struct per_pidcomm *c; | |
889 | int count = 0; | |
890 | ||
891 | p = all_data; | |
892 | while (p) { | |
893 | p->display = 0; | |
894 | if (p->start_time == 1) | |
895 | p->start_time = first_time; | |
896 | ||
897 | /* no exit marker, task kept running to the end */ | |
898 | if (p->end_time == 0) | |
899 | p->end_time = last_time; | |
39a90a8e | 900 | if (p->total_time >= threshold && !power_only) |
10274989 AV |
901 | p->display = 1; |
902 | ||
903 | c = p->all; | |
904 | ||
905 | while (c) { | |
906 | c->display = 0; | |
907 | ||
908 | if (c->start_time == 1) | |
909 | c->start_time = first_time; | |
910 | ||
39a90a8e | 911 | if (c->total_time >= threshold && !power_only) { |
10274989 AV |
912 | c->display = 1; |
913 | count++; | |
914 | } | |
915 | ||
916 | if (c->end_time == 0) | |
917 | c->end_time = last_time; | |
918 | ||
919 | c = c->next; | |
920 | } | |
921 | p = p->next; | |
922 | } | |
923 | return count; | |
924 | } | |
925 | ||
926 | ||
927 | ||
928 | #define TIME_THRESH 10000000 | |
929 | ||
930 | static void write_svg_file(const char *filename) | |
931 | { | |
932 | u64 i; | |
933 | int count; | |
934 | ||
935 | numcpus++; | |
936 | ||
937 | ||
938 | count = determine_display_tasks(TIME_THRESH); | |
939 | ||
940 | /* We'd like to show at least 15 tasks; be less picky if we have fewer */ | |
941 | if (count < 15) | |
942 | count = determine_display_tasks(TIME_THRESH / 10); | |
943 | ||
5094b655 | 944 | open_svg(filename, numcpus, count, first_time, last_time); |
10274989 | 945 | |
5094b655 | 946 | svg_time_grid(); |
10274989 AV |
947 | svg_legenda(); |
948 | ||
949 | for (i = 0; i < numcpus; i++) | |
950 | svg_cpu_box(i, max_freq, turbo_frequency); | |
951 | ||
952 | draw_cpu_usage(); | |
953 | draw_process_bars(); | |
954 | draw_c_p_states(); | |
955 | draw_wakeups(); | |
956 | ||
957 | svg_close(); | |
958 | } | |
959 | ||
960 | static int | |
961 | process_event(event_t *event) | |
962 | { | |
963 | ||
964 | switch (event->header.type) { | |
965 | ||
cdd6c482 | 966 | case PERF_RECORD_COMM: |
10274989 | 967 | return process_comm_event(event); |
cdd6c482 | 968 | case PERF_RECORD_FORK: |
10274989 | 969 | return process_fork_event(event); |
cdd6c482 | 970 | case PERF_RECORD_EXIT: |
10274989 | 971 | return process_exit_event(event); |
cdd6c482 | 972 | case PERF_RECORD_SAMPLE: |
10274989 AV |
973 | return queue_sample_event(event); |
974 | ||
975 | /* | |
976 | * We dont process them right now but they are fine: | |
977 | */ | |
cdd6c482 IM |
978 | case PERF_RECORD_MMAP: |
979 | case PERF_RECORD_THROTTLE: | |
980 | case PERF_RECORD_UNTHROTTLE: | |
10274989 AV |
981 | return 0; |
982 | ||
983 | default: | |
984 | return -1; | |
985 | } | |
986 | ||
987 | return 0; | |
988 | } | |
989 | ||
990 | static void process_samples(void) | |
991 | { | |
992 | struct sample_wrapper *cursor; | |
993 | event_t *event; | |
994 | ||
995 | sort_queued_samples(); | |
996 | ||
997 | cursor = all_samples; | |
998 | while (cursor) { | |
999 | event = (void *)&cursor->data; | |
1000 | cursor = cursor->next; | |
1001 | process_sample_event(event); | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | ||
1006 | static int __cmd_timechart(void) | |
1007 | { | |
1008 | int ret, rc = EXIT_FAILURE; | |
1009 | unsigned long offset = 0; | |
1010 | unsigned long head, shift; | |
1011 | struct stat statbuf; | |
1012 | event_t *event; | |
1013 | uint32_t size; | |
1014 | char *buf; | |
1015 | int input; | |
1016 | ||
1017 | input = open(input_name, O_RDONLY); | |
1018 | if (input < 0) { | |
1019 | fprintf(stderr, " failed to open file: %s", input_name); | |
1020 | if (!strcmp(input_name, "perf.data")) | |
1021 | fprintf(stderr, " (try 'perf record' first)"); | |
1022 | fprintf(stderr, "\n"); | |
1023 | exit(-1); | |
1024 | } | |
1025 | ||
1026 | ret = fstat(input, &statbuf); | |
1027 | if (ret < 0) { | |
1028 | perror("failed to stat file"); | |
1029 | exit(-1); | |
1030 | } | |
1031 | ||
1032 | if (!statbuf.st_size) { | |
1033 | fprintf(stderr, "zero-sized file, nothing to do!\n"); | |
1034 | exit(0); | |
1035 | } | |
1036 | ||
1037 | header = perf_header__read(input); | |
1038 | head = header->data_offset; | |
1039 | ||
1040 | sample_type = perf_header__sample_type(header); | |
1041 | ||
1042 | shift = page_size * (head / page_size); | |
1043 | offset += shift; | |
1044 | head -= shift; | |
1045 | ||
1046 | remap: | |
1047 | buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ, | |
1048 | MAP_SHARED, input, offset); | |
1049 | if (buf == MAP_FAILED) { | |
1050 | perror("failed to mmap file"); | |
1051 | exit(-1); | |
1052 | } | |
1053 | ||
1054 | more: | |
1055 | event = (event_t *)(buf + head); | |
1056 | ||
1057 | size = event->header.size; | |
1058 | if (!size) | |
1059 | size = 8; | |
1060 | ||
1061 | if (head + event->header.size >= page_size * mmap_window) { | |
1062 | int ret2; | |
1063 | ||
1064 | shift = page_size * (head / page_size); | |
1065 | ||
1066 | ret2 = munmap(buf, page_size * mmap_window); | |
1067 | assert(ret2 == 0); | |
1068 | ||
1069 | offset += shift; | |
1070 | head -= shift; | |
1071 | goto remap; | |
1072 | } | |
1073 | ||
1074 | size = event->header.size; | |
1075 | ||
1076 | if (!size || process_event(event) < 0) { | |
1077 | ||
1078 | printf("%p [%p]: skipping unknown header type: %d\n", | |
1079 | (void *)(offset + head), | |
1080 | (void *)(long)(event->header.size), | |
1081 | event->header.type); | |
1082 | ||
1083 | /* | |
1084 | * assume we lost track of the stream, check alignment, and | |
1085 | * increment a single u64 in the hope to catch on again 'soon'. | |
1086 | */ | |
1087 | ||
1088 | if (unlikely(head & 7)) | |
1089 | head &= ~7ULL; | |
1090 | ||
1091 | size = 8; | |
1092 | } | |
1093 | ||
1094 | head += size; | |
1095 | ||
1096 | if (offset + head >= header->data_offset + header->data_size) | |
1097 | goto done; | |
1098 | ||
1099 | if (offset + head < (unsigned long)statbuf.st_size) | |
1100 | goto more; | |
1101 | ||
1102 | done: | |
1103 | rc = EXIT_SUCCESS; | |
1104 | close(input); | |
1105 | ||
1106 | ||
1107 | process_samples(); | |
1108 | ||
1109 | end_sample_processing(); | |
1110 | ||
1111 | sort_pids(); | |
1112 | ||
1113 | write_svg_file(output_name); | |
1114 | ||
1115 | printf("Written %2.1f seconds of trace to %s.\n", (last_time - first_time) / 1000000000.0, output_name); | |
1116 | ||
1117 | return rc; | |
1118 | } | |
1119 | ||
3c09eebd AV |
1120 | static const char * const timechart_usage[] = { |
1121 | "perf timechart [<options>] {record}", | |
10274989 AV |
1122 | NULL |
1123 | }; | |
1124 | ||
3c09eebd AV |
1125 | static const char *record_args[] = { |
1126 | "record", | |
1127 | "-a", | |
1128 | "-R", | |
1129 | "-M", | |
1130 | "-f", | |
1131 | "-c", "1", | |
1132 | "-e", "power:power_start", | |
1133 | "-e", "power:power_end", | |
1134 | "-e", "power:power_frequency", | |
1135 | "-e", "sched:sched_wakeup", | |
1136 | "-e", "sched:sched_switch", | |
1137 | }; | |
1138 | ||
1139 | static int __cmd_record(int argc, const char **argv) | |
1140 | { | |
1141 | unsigned int rec_argc, i, j; | |
1142 | const char **rec_argv; | |
1143 | ||
1144 | rec_argc = ARRAY_SIZE(record_args) + argc - 1; | |
1145 | rec_argv = calloc(rec_argc + 1, sizeof(char *)); | |
1146 | ||
1147 | for (i = 0; i < ARRAY_SIZE(record_args); i++) | |
1148 | rec_argv[i] = strdup(record_args[i]); | |
1149 | ||
1150 | for (j = 1; j < (unsigned int)argc; j++, i++) | |
1151 | rec_argv[i] = argv[j]; | |
1152 | ||
1153 | return cmd_record(i, rec_argv, NULL); | |
1154 | } | |
1155 | ||
10274989 AV |
1156 | static const struct option options[] = { |
1157 | OPT_STRING('i', "input", &input_name, "file", | |
1158 | "input file name"), | |
1159 | OPT_STRING('o', "output", &output_name, "file", | |
1160 | "output file name"), | |
5094b655 AV |
1161 | OPT_INTEGER('w', "width", &svg_page_width, |
1162 | "page width"), | |
39a90a8e AV |
1163 | OPT_BOOLEAN('p', "power-only", &power_only, |
1164 | "output power data only"), | |
10274989 AV |
1165 | OPT_END() |
1166 | }; | |
1167 | ||
1168 | ||
1169 | int cmd_timechart(int argc, const char **argv, const char *prefix __used) | |
1170 | { | |
1171 | symbol__init(); | |
1172 | ||
1173 | page_size = getpagesize(); | |
1174 | ||
3c09eebd AV |
1175 | argc = parse_options(argc, argv, options, timechart_usage, |
1176 | PARSE_OPT_STOP_AT_NON_OPTION); | |
10274989 | 1177 | |
3c09eebd AV |
1178 | if (argc && !strncmp(argv[0], "rec", 3)) |
1179 | return __cmd_record(argc, argv); | |
1180 | else if (argc) | |
1181 | usage_with_options(timechart_usage, options); | |
10274989 AV |
1182 | |
1183 | setup_pager(); | |
1184 | ||
1185 | return __cmd_timechart(); | |
1186 | } |