tools/perf/builtin-timechart.c

   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
  49 static int              power_only;
  50
  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
 552         for (cpu = 0; cpu <= numcpus; cpu++) {
 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(&copy->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;
 757                 char *task_from = NULL, *task_to = NULL;
 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) {
 766                                                 if (p->pid == we->waker) {
 767                                                         from = c->Y;
 768                                                         task_from = c->comm;
 769                                                 }
 770                                                 if (p->pid == we->wakee) {
 771                                                         to = c->Y;
 772                                                         task_to = c->comm;
 773                                                 }
 774                                         }
 775                                         c = c->next;
 776                                 }
 777                         }
 778                         p = p->next;
 779                 }
 780
 781                 if (we->waker == -1)
 782                         svg_interrupt(we->time, to);
 783                 else if (from && to && abs(from - to) == 1)
 784                         svg_wakeline(we->time, from, to);
 785                 else
 786                         svg_partial_wakeline(we->time, from, task_from, to, task_to);
 787                 we = we->next;
 788         }
 789 }
 790
 791 static void draw_cpu_usage(void)
 792 {
 793         struct per_pid *p;
 794         struct per_pidcomm *c;
 795         struct cpu_sample *sample;
 796         p = all_data;
 797         while (p) {
 798                 c = p->all;
 799                 while (c) {
 800                         sample = c->samples;
 801                         while (sample) {
 802                                 if (sample->type == TYPE_RUNNING)
 803                                         svg_process(sample->cpu, sample->start_time, sample->end_time, "sample", c->comm);
 804
 805                                 sample = sample->next;
 806                         }
 807                         c = c->next;
 808                 }
 809                 p = p->next;
 810         }
 811 }
 812
 813 static void draw_process_bars(void)
 814 {
 815         struct per_pid *p;
 816         struct per_pidcomm *c;
 817         struct cpu_sample *sample;
 818         int Y = 0;
 819
 820         Y = 2 * numcpus + 2;
 821
 822         p = all_data;
 823         while (p) {
 824                 c = p->all;
 825                 while (c) {
 826                         if (!c->display) {
 827                                 c->Y = 0;
 828                                 c = c->next;
 829                                 continue;
 830                         }
 831
 832                         svg_box(Y, c->start_time, c->end_time, "process");
 833                         sample = c->samples;
 834                         while (sample) {
 835                                 if (sample->type == TYPE_RUNNING)
 836                                         svg_sample(Y, sample->cpu, sample->start_time, sample->end_time);
 837                                 if (sample->type == TYPE_BLOCKED)
 838                                         svg_box(Y, sample->start_time, sample->end_time, "blocked");
 839                                 if (sample->type == TYPE_WAITING)
 840                                         svg_waiting(Y, sample->start_time, sample->end_time);
 841                                 sample = sample->next;
 842                         }
 843
 844                         if (c->comm) {
 845                                 char comm[256];
 846                                 if (c->total_time > 5000000000) /* 5 seconds */
 847                                         sprintf(comm, "%s:%i (%2.2fs)", c->comm, p->pid, c->total_time / 1000000000.0);
 848                                 else
 849                                         sprintf(comm, "%s:%i (%3.1fms)", c->comm, p->pid, c->total_time / 1000000.0);
 850
 851                                 svg_text(Y, c->start_time, comm);
 852                         }
 853                         c->Y = Y;
 854                         Y++;
 855                         c = c->next;
 856                 }
 857                 p = p->next;
 858         }
 859 }
 860
 861 static int determine_display_tasks(u64 threshold)
 862 {
 863         struct per_pid *p;
 864         struct per_pidcomm *c;
 865         int count = 0;
 866
 867         p = all_data;
 868         while (p) {
 869                 p->display = 0;
 870                 if (p->start_time == 1)
 871                         p->start_time = first_time;
 872
 873                 /* no exit marker, task kept running to the end */
 874                 if (p->end_time == 0)
 875                         p->end_time = last_time;
 876                 if (p->total_time >= threshold && !power_only)
 877                         p->display = 1;
 878
 879                 c = p->all;
 880
 881                 while (c) {
 882                         c->display = 0;
 883
 884                         if (c->start_time == 1)
 885                                 c->start_time = first_time;
 886
 887                         if (c->total_time >= threshold && !power_only) {
 888                                 c->display = 1;
 889                                 count++;
 890                         }
 891
 892                         if (c->end_time == 0)
 893                                 c->end_time = last_time;
 894
 895                         c = c->next;
 896                 }
 897                 p = p->next;
 898         }
 899         return count;
 900 }
 901
 902
 903
 904 #define TIME_THRESH 10000000
 905
 906 static void write_svg_file(const char *filename)
 907 {
 908         u64 i;
 909         int count;
 910
 911         numcpus++;
 912
 913
 914         count = determine_display_tasks(TIME_THRESH);
 915
 916         /* We'd like to show at least 15 tasks; be less picky if we have fewer */
 917         if (count < 15)
 918                 count = determine_display_tasks(TIME_THRESH / 10);
 919
 920         open_svg(filename, numcpus, count, first_time, last_time);
 921
 922         svg_time_grid();
 923         svg_legenda();
 924
 925         for (i = 0; i < numcpus; i++)
 926                 svg_cpu_box(i, max_freq, turbo_frequency);
 927
 928         draw_cpu_usage();
 929         draw_process_bars();
 930         draw_c_p_states();
 931         draw_wakeups();
 932
 933         svg_close();
 934 }
 935
 936 static int
 937 process_event(event_t *event)
 938 {
 939
 940         switch (event->header.type) {
 941
 942         case PERF_RECORD_COMM:
 943                 return process_comm_event(event);
 944         case PERF_RECORD_FORK:
 945                 return process_fork_event(event);
 946         case PERF_RECORD_EXIT:
 947                 return process_exit_event(event);
 948         case PERF_RECORD_SAMPLE:
 949                 return queue_sample_event(event);
 950
 951         /*
 952          * We dont process them right now but they are fine:
 953          */
 954         case PERF_RECORD_MMAP:
 955         case PERF_RECORD_THROTTLE:
 956         case PERF_RECORD_UNTHROTTLE:
 957                 return 0;
 958
 959         default:
 960                 return -1;
 961         }
 962
 963         return 0;
 964 }
 965
 966 static void process_samples(void)
 967 {
 968         struct sample_wrapper *cursor;
 969         event_t *event;
 970
 971         sort_queued_samples();
 972
 973         cursor = all_samples;
 974         while (cursor) {
 975                 event = (void *)&cursor->data;
 976                 cursor = cursor->next;
 977                 process_sample_event(event);
 978         }
 979 }
 980
 981
 982 static int __cmd_timechart(void)
 983 {
 984         int ret, rc = EXIT_FAILURE;
 985         unsigned long offset = 0;
 986         unsigned long head, shift;
 987         struct stat statbuf;
 988         event_t *event;
 989         uint32_t size;
 990         char *buf;
 991         int input;
 992
 993         input = open(input_name, O_RDONLY);
 994         if (input < 0) {
 995                 fprintf(stderr, " failed to open file: %s", input_name);
 996                 if (!strcmp(input_name, "perf.data"))
 997                         fprintf(stderr, "  (try 'perf record' first)");
 998                 fprintf(stderr, "\n");
 999                 exit(-1);
1000         }
1001
1002         ret = fstat(input, &statbuf);
1003         if (ret < 0) {
1004                 perror("failed to stat file");
1005                 exit(-1);
1006         }
1007
1008         if (!statbuf.st_size) {
1009                 fprintf(stderr, "zero-sized file, nothing to do!\n");
1010                 exit(0);
1011         }
1012
1013         header = perf_header__read(input);
1014         head = header->data_offset;
1015
1016         sample_type = perf_header__sample_type(header);
1017
1018         shift = page_size * (head / page_size);
1019         offset += shift;
1020         head -= shift;
1021
1022 remap:
1023         buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1024                            MAP_SHARED, input, offset);
1025         if (buf == MAP_FAILED) {
1026                 perror("failed to mmap file");
1027                 exit(-1);
1028         }
1029
1030 more:
1031         event = (event_t *)(buf + head);
1032
1033         size = event->header.size;
1034         if (!size)
1035                 size = 8;
1036
1037         if (head + event->header.size >= page_size * mmap_window) {
1038                 int ret2;
1039
1040                 shift = page_size * (head / page_size);
1041
1042                 ret2 = munmap(buf, page_size * mmap_window);
1043                 assert(ret2 == 0);
1044
1045                 offset += shift;
1046                 head -= shift;
1047                 goto remap;
1048         }
1049
1050         size = event->header.size;
1051
1052         if (!size || process_event(event) < 0) {
1053
1054                 printf("%p [%p]: skipping unknown header type: %d\n",
1055                         (void *)(offset + head),
1056                         (void *)(long)(event->header.size),
1057                         event->header.type);
1058
1059                 /*
1060                  * assume we lost track of the stream, check alignment, and
1061                  * increment a single u64 in the hope to catch on again 'soon'.
1062                  */
1063
1064                 if (unlikely(head & 7))
1065                         head &= ~7ULL;
1066
1067                 size = 8;
1068         }
1069
1070         head += size;
1071
1072         if (offset + head >= header->data_offset + header->data_size)
1073                 goto done;
1074
1075         if (offset + head < (unsigned long)statbuf.st_size)
1076                 goto more;
1077
1078 done:
1079         rc = EXIT_SUCCESS;
1080         close(input);
1081
1082
1083         process_samples();
1084
1085         end_sample_processing();
1086
1087         sort_pids();
1088
1089         write_svg_file(output_name);
1090
1091         printf("Written %2.1f seconds of trace to %s.\n", (last_time - first_time) / 1000000000.0, output_name);
1092
1093         return rc;
1094 }
1095
1096 static const char * const timechart_usage[] = {
1097         "perf timechart [<options>] {record}",
1098         NULL
1099 };
1100
1101 static const char *record_args[] = {
1102         "record",
1103         "-a",
1104         "-R",
1105         "-M",
1106         "-f",
1107         "-c", "1",
1108         "-e", "power:power_start",
1109         "-e", "power:power_end",
1110         "-e", "power:power_frequency",
1111         "-e", "sched:sched_wakeup",
1112         "-e", "sched:sched_switch",
1113 };
1114
1115 static int __cmd_record(int argc, const char **argv)
1116 {
1117         unsigned int rec_argc, i, j;
1118         const char **rec_argv;
1119
1120         rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1121         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1122
1123         for (i = 0; i < ARRAY_SIZE(record_args); i++)
1124                 rec_argv[i] = strdup(record_args[i]);
1125
1126         for (j = 1; j < (unsigned int)argc; j++, i++)
1127                 rec_argv[i] = argv[j];
1128
1129         return cmd_record(i, rec_argv, NULL);
1130 }
1131
1132 static const struct option options[] = {
1133         OPT_STRING('i', "input", &input_name, "file",
1134                     "input file name"),
1135         OPT_STRING('o', "output", &output_name, "file",
1136                     "output file name"),
1137         OPT_INTEGER('w', "width", &svg_page_width,
1138                     "page width"),
1139         OPT_BOOLEAN('p', "power-only", &power_only,
1140                     "output power data only"),
1141         OPT_END()
1142 };
1143
1144
1145 int cmd_timechart(int argc, const char **argv, const char *prefix __used)
1146 {
1147         symbol__init();
1148
1149         page_size = getpagesize();
1150
1151         argc = parse_options(argc, argv, options, timechart_usage,
1152                         PARSE_OPT_STOP_AT_NON_OPTION);
1153
1154         if (argc && !strncmp(argv[0], "rec", 3))
1155                 return __cmd_record(argc, argv);
1156         else if (argc)
1157                 usage_with_options(timechart_usage, options);
1158
1159         setup_pager();
1160
1161         return __cmd_timechart();
1162 }