tools/perf/builtin-trace.c

   1 #include <traceevent/event-parse.h>
   2 #include "builtin.h"
   3 #include "util/color.h"
   4 #include "util/debug.h"
   5 #include "util/evlist.h"
   6 #include "util/machine.h"
   7 #include "util/session.h"
   8 #include "util/thread.h"
   9 #include "util/parse-options.h"
  10 #include "util/strlist.h"
  11 #include "util/intlist.h"
  12 #include "util/thread_map.h"
  13
  14 #include <libaudit.h>
  15 #include <stdlib.h>
  16 #include <sys/mman.h>
  17 #include <linux/futex.h>
  18
  19 /* For older distros: */
  20 #ifndef MAP_STACK
  21 # define MAP_STACK              0x20000
  22 #endif
  23
  24 #ifndef MADV_HWPOISON
  25 # define MADV_HWPOISON          100
  26 #endif
  27
  28 #ifndef MADV_MERGEABLE
  29 # define MADV_MERGEABLE         12
  30 #endif
  31
  32 #ifndef MADV_UNMERGEABLE
  33 # define MADV_UNMERGEABLE       13
  34 #endif
  35
  36 struct syscall_arg {
  37         unsigned long val;
  38         void          *parm;
  39         u8            idx;
  40         u8            mask;
  41 };
  42
  43 struct strarray {
  44         int         nr_entries;
  45         const char **entries;
  46 };
  47
  48 #define DEFINE_STRARRAY(array) struct strarray strarray__##array = { \
  49         .nr_entries = ARRAY_SIZE(array), \
  50         .entries = array, \
  51 }
  52
  53 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
  54                                               struct syscall_arg *arg)
  55 {
  56         int idx = arg->val;
  57         struct strarray *sa = arg->parm;
  58
  59         if (idx < 0 || idx >= sa->nr_entries)
  60                 return scnprintf(bf, size, "%d", idx);
  61
  62         return scnprintf(bf, size, "%s", sa->entries[idx]);
  63 }
  64
  65 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
  66
  67 static size_t syscall_arg__scnprintf_hex(char *bf, size_t size,
  68                                          struct syscall_arg *arg)
  69 {
  70         return scnprintf(bf, size, "%#lx", arg->val);
  71 }
  72
  73 #define SCA_HEX syscall_arg__scnprintf_hex
  74
  75 static size_t syscall_arg__scnprintf_mmap_prot(char *bf, size_t size,
  76                                                struct syscall_arg *arg)
  77 {
  78         int printed = 0, prot = arg->val;
  79
  80         if (prot == PROT_NONE)
  81                 return scnprintf(bf, size, "NONE");
  82 #define P_MMAP_PROT(n) \
  83         if (prot & PROT_##n) { \
  84                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
  85                 prot &= ~PROT_##n; \
  86         }
  87
  88         P_MMAP_PROT(EXEC);
  89         P_MMAP_PROT(READ);
  90         P_MMAP_PROT(WRITE);
  91 #ifdef PROT_SEM
  92         P_MMAP_PROT(SEM);
  93 #endif
  94         P_MMAP_PROT(GROWSDOWN);
  95         P_MMAP_PROT(GROWSUP);
  96 #undef P_MMAP_PROT
  97
  98         if (prot)
  99                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", prot);
 100
 101         return printed;
 102 }
 103
 104 #define SCA_MMAP_PROT syscall_arg__scnprintf_mmap_prot
 105
 106 static size_t syscall_arg__scnprintf_mmap_flags(char *bf, size_t size,
 107                                                 struct syscall_arg *arg)
 108 {
 109         int printed = 0, flags = arg->val;
 110
 111 #define P_MMAP_FLAG(n) \
 112         if (flags & MAP_##n) { \
 113                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 114                 flags &= ~MAP_##n; \
 115         }
 116
 117         P_MMAP_FLAG(SHARED);
 118         P_MMAP_FLAG(PRIVATE);
 119 #ifdef MAP_32BIT
 120         P_MMAP_FLAG(32BIT);
 121 #endif
 122         P_MMAP_FLAG(ANONYMOUS);
 123         P_MMAP_FLAG(DENYWRITE);
 124         P_MMAP_FLAG(EXECUTABLE);
 125         P_MMAP_FLAG(FILE);
 126         P_MMAP_FLAG(FIXED);
 127         P_MMAP_FLAG(GROWSDOWN);
 128 #ifdef MAP_HUGETLB
 129         P_MMAP_FLAG(HUGETLB);
 130 #endif
 131         P_MMAP_FLAG(LOCKED);
 132         P_MMAP_FLAG(NONBLOCK);
 133         P_MMAP_FLAG(NORESERVE);
 134         P_MMAP_FLAG(POPULATE);
 135         P_MMAP_FLAG(STACK);
 136 #ifdef MAP_UNINITIALIZED
 137         P_MMAP_FLAG(UNINITIALIZED);
 138 #endif
 139 #undef P_MMAP_FLAG
 140
 141         if (flags)
 142                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 143
 144         return printed;
 145 }
 146
 147 #define SCA_MMAP_FLAGS syscall_arg__scnprintf_mmap_flags
 148
 149 static size_t syscall_arg__scnprintf_madvise_behavior(char *bf, size_t size,
 150                                                       struct syscall_arg *arg)
 151 {
 152         int behavior = arg->val;
 153
 154         switch (behavior) {
 155 #define P_MADV_BHV(n) case MADV_##n: return scnprintf(bf, size, #n)
 156         P_MADV_BHV(NORMAL);
 157         P_MADV_BHV(RANDOM);
 158         P_MADV_BHV(SEQUENTIAL);
 159         P_MADV_BHV(WILLNEED);
 160         P_MADV_BHV(DONTNEED);
 161         P_MADV_BHV(REMOVE);
 162         P_MADV_BHV(DONTFORK);
 163         P_MADV_BHV(DOFORK);
 164         P_MADV_BHV(HWPOISON);
 165 #ifdef MADV_SOFT_OFFLINE
 166         P_MADV_BHV(SOFT_OFFLINE);
 167 #endif
 168         P_MADV_BHV(MERGEABLE);
 169         P_MADV_BHV(UNMERGEABLE);
 170 #ifdef MADV_HUGEPAGE
 171         P_MADV_BHV(HUGEPAGE);
 172 #endif
 173 #ifdef MADV_NOHUGEPAGE
 174         P_MADV_BHV(NOHUGEPAGE);
 175 #endif
 176 #ifdef MADV_DONTDUMP
 177         P_MADV_BHV(DONTDUMP);
 178 #endif
 179 #ifdef MADV_DODUMP
 180         P_MADV_BHV(DODUMP);
 181 #endif
 182 #undef P_MADV_PHV
 183         default: break;
 184         }
 185
 186         return scnprintf(bf, size, "%#x", behavior);
 187 }
 188
 189 #define SCA_MADV_BHV syscall_arg__scnprintf_madvise_behavior
 190
 191 static size_t syscall_arg__scnprintf_futex_op(char *bf, size_t size, struct syscall_arg *arg)
 192 {
 193         enum syscall_futex_args {
 194                 SCF_UADDR   = (1 << 0),
 195                 SCF_OP      = (1 << 1),
 196                 SCF_VAL     = (1 << 2),
 197                 SCF_TIMEOUT = (1 << 3),
 198                 SCF_UADDR2  = (1 << 4),
 199                 SCF_VAL3    = (1 << 5),
 200         };
 201         int op = arg->val;
 202         int cmd = op & FUTEX_CMD_MASK;
 203         size_t printed = 0;
 204
 205         switch (cmd) {
 206 #define P_FUTEX_OP(n) case FUTEX_##n: printed = scnprintf(bf, size, #n);
 207         P_FUTEX_OP(WAIT);           arg->mask |= SCF_VAL3|SCF_UADDR2;             break;
 208         P_FUTEX_OP(WAKE);           arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 209         P_FUTEX_OP(FD);             arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 210         P_FUTEX_OP(REQUEUE);        arg->mask |= SCF_VAL3|SCF_TIMEOUT;            break;
 211         P_FUTEX_OP(CMP_REQUEUE);    arg->mask |= SCF_TIMEOUT;                     break;
 212         P_FUTEX_OP(CMP_REQUEUE_PI); arg->mask |= SCF_TIMEOUT;                     break;
 213         P_FUTEX_OP(WAKE_OP);                                                      break;
 214         P_FUTEX_OP(LOCK_PI);        arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 215         P_FUTEX_OP(UNLOCK_PI);      arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 216         P_FUTEX_OP(TRYLOCK_PI);     arg->mask |= SCF_VAL3|SCF_UADDR2;             break;
 217         P_FUTEX_OP(WAIT_BITSET);    arg->mask |= SCF_UADDR2;                      break;
 218         P_FUTEX_OP(WAKE_BITSET);    arg->mask |= SCF_UADDR2;                      break;
 219         P_FUTEX_OP(WAIT_REQUEUE_PI);                                              break;
 220         default: printed = scnprintf(bf, size, "%#x", cmd);                       break;
 221         }
 222
 223         if (op & FUTEX_PRIVATE_FLAG)
 224                 printed += scnprintf(bf + printed, size - printed, "|PRIV");
 225
 226         if (op & FUTEX_CLOCK_REALTIME)
 227                 printed += scnprintf(bf + printed, size - printed, "|CLKRT");
 228
 229         return printed;
 230 }
 231
 232 #define SCA_FUTEX_OP  syscall_arg__scnprintf_futex_op
 233
 234 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
 235 static DEFINE_STRARRAY(itimers);
 236
 237 static const char *whences[] = { "SET", "CUR", "END",
 238 #ifdef SEEK_DATA
 239 "DATA",
 240 #endif
 241 #ifdef SEEK_HOLE
 242 "HOLE",
 243 #endif
 244 };
 245 static DEFINE_STRARRAY(whences);
 246
 247 static const char *fcntl_cmds[] = {
 248         "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
 249         "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "F_GETLK64",
 250         "F_SETLK64", "F_SETLKW64", "F_SETOWN_EX", "F_GETOWN_EX",
 251         "F_GETOWNER_UIDS",
 252 };
 253 static DEFINE_STRARRAY(fcntl_cmds);
 254
 255 static size_t syscall_arg__scnprintf_open_flags(char *bf, size_t size,
 256                                                struct syscall_arg *arg)
 257 {
 258         int printed = 0, flags = arg->val;
 259
 260         if (!(flags & O_CREAT))
 261                 arg->mask |= 1 << (arg->idx + 1); /* Mask the mode parm */
 262
 263         if (flags == 0)
 264                 return scnprintf(bf, size, "RDONLY");
 265 #define P_FLAG(n) \
 266         if (flags & O_##n) { \
 267                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 268                 flags &= ~O_##n; \
 269         }
 270
 271         P_FLAG(APPEND);
 272         P_FLAG(ASYNC);
 273         P_FLAG(CLOEXEC);
 274         P_FLAG(CREAT);
 275         P_FLAG(DIRECT);
 276         P_FLAG(DIRECTORY);
 277         P_FLAG(EXCL);
 278         P_FLAG(LARGEFILE);
 279         P_FLAG(NOATIME);
 280         P_FLAG(NOCTTY);
 281 #ifdef O_NONBLOCK
 282         P_FLAG(NONBLOCK);
 283 #elif O_NDELAY
 284         P_FLAG(NDELAY);
 285 #endif
 286 #ifdef O_PATH
 287         P_FLAG(PATH);
 288 #endif
 289         P_FLAG(RDWR);
 290 #ifdef O_DSYNC
 291         if ((flags & O_SYNC) == O_SYNC)
 292                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", "SYNC");
 293         else {
 294                 P_FLAG(DSYNC);
 295         }
 296 #else
 297         P_FLAG(SYNC);
 298 #endif
 299         P_FLAG(TRUNC);
 300         P_FLAG(WRONLY);
 301 #undef P_FLAG
 302
 303         if (flags)
 304                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 305
 306         return printed;
 307 }
 308
 309 #define SCA_OPEN_FLAGS syscall_arg__scnprintf_open_flags
 310
 311 static struct syscall_fmt {
 312         const char *name;
 313         const char *alias;
 314         size_t     (*arg_scnprintf[6])(char *bf, size_t size, struct syscall_arg *arg);
 315         void       *arg_parm[6];
 316         bool       errmsg;
 317         bool       timeout;
 318         bool       hexret;
 319 } syscall_fmts[] = {
 320         { .name     = "access",     .errmsg = true, },
 321         { .name     = "arch_prctl", .errmsg = true, .alias = "prctl", },
 322         { .name     = "brk",        .hexret = true,
 323           .arg_scnprintf = { [0] = SCA_HEX, /* brk */ }, },
 324         { .name     = "mmap",       .hexret = true, },
 325         { .name     = "connect",    .errmsg = true, },
 326         { .name     = "fcntl",      .errmsg = true,
 327           .arg_scnprintf = { [1] = SCA_STRARRAY, /* cmd */ },
 328           .arg_parm      = { [1] = &strarray__fcntl_cmds, /* cmd */ }, },
 329         { .name     = "fstat",      .errmsg = true, .alias = "newfstat", },
 330         { .name     = "fstatat",    .errmsg = true, .alias = "newfstatat", },
 331         { .name     = "futex",      .errmsg = true,
 332           .arg_scnprintf = { [1] = SCA_FUTEX_OP, /* op */ }, },
 333         { .name     = "getitimer",  .errmsg = true,
 334           .arg_scnprintf = { [0] = SCA_STRARRAY, /* which */ },
 335           .arg_parm      = { [0] = &strarray__itimers, /* which */ }, },
 336         { .name     = "ioctl",      .errmsg = true,
 337           .arg_scnprintf = { [2] = SCA_HEX, /* arg */ }, },
 338         { .name     = "lseek",      .errmsg = true,
 339           .arg_scnprintf = { [2] = SCA_STRARRAY, /* whence */ },
 340           .arg_parm      = { [2] = &strarray__whences, /* whence */ }, },
 341         { .name     = "lstat",      .errmsg = true, .alias = "newlstat", },
 342         { .name     = "madvise",    .errmsg = true,
 343           .arg_scnprintf = { [0] = SCA_HEX,      /* start */
 344                              [2] = SCA_MADV_BHV, /* behavior */ }, },
 345         { .name     = "mmap",       .hexret = true,
 346           .arg_scnprintf = { [0] = SCA_HEX,       /* addr */
 347                              [2] = SCA_MMAP_PROT, /* prot */
 348                              [3] = SCA_MMAP_FLAGS, /* flags */ }, },
 349         { .name     = "mprotect",   .errmsg = true,
 350           .arg_scnprintf = { [0] = SCA_HEX, /* start */
 351                              [2] = SCA_MMAP_PROT, /* prot */ }, },
 352         { .name     = "mremap",     .hexret = true,
 353           .arg_scnprintf = { [0] = SCA_HEX, /* addr */
 354                              [4] = SCA_HEX, /* new_addr */ }, },
 355         { .name     = "munmap",     .errmsg = true,
 356           .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
 357         { .name     = "open",       .errmsg = true,
 358           .arg_scnprintf = { [1] = SCA_OPEN_FLAGS, /* flags */ }, },
 359         { .name     = "open_by_handle_at", .errmsg = true,
 360           .arg_scnprintf = { [2] = SCA_OPEN_FLAGS, /* flags */ }, },
 361         { .name     = "openat",     .errmsg = true,
 362           .arg_scnprintf = { [2] = SCA_OPEN_FLAGS, /* flags */ }, },
 363         { .name     = "poll",       .errmsg = true, .timeout = true, },
 364         { .name     = "ppoll",      .errmsg = true, .timeout = true, },
 365         { .name     = "pread",      .errmsg = true, .alias = "pread64", },
 366         { .name     = "pwrite",     .errmsg = true, .alias = "pwrite64", },
 367         { .name     = "read",       .errmsg = true, },
 368         { .name     = "recvfrom",   .errmsg = true, },
 369         { .name     = "select",     .errmsg = true, .timeout = true, },
 370         { .name     = "setitimer",  .errmsg = true,
 371           .arg_scnprintf = { [0] = SCA_STRARRAY, /* which */ },
 372           .arg_parm      = { [0] = &strarray__itimers, /* which */ }, },
 373         { .name     = "socket",     .errmsg = true, },
 374         { .name     = "stat",       .errmsg = true, .alias = "newstat", },
 375         { .name     = "uname",      .errmsg = true, .alias = "newuname", },
 376 };
 377
 378 static int syscall_fmt__cmp(const void *name, const void *fmtp)
 379 {
 380         const struct syscall_fmt *fmt = fmtp;
 381         return strcmp(name, fmt->name);
 382 }
 383
 384 static struct syscall_fmt *syscall_fmt__find(const char *name)
 385 {
 386         const int nmemb = ARRAY_SIZE(syscall_fmts);
 387         return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
 388 }
 389
 390 struct syscall {
 391         struct event_format *tp_format;
 392         const char          *name;
 393         bool                filtered;
 394         struct syscall_fmt  *fmt;
 395         size_t              (**arg_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
 396         void                **arg_parm;
 397 };
 398
 399 static size_t fprintf_duration(unsigned long t, FILE *fp)
 400 {
 401         double duration = (double)t / NSEC_PER_MSEC;
 402         size_t printed = fprintf(fp, "(");
 403
 404         if (duration >= 1.0)
 405                 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
 406         else if (duration >= 0.01)
 407                 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
 408         else
 409                 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
 410         return printed + fprintf(fp, "): ");
 411 }
 412
 413 struct thread_trace {
 414         u64               entry_time;
 415         u64               exit_time;
 416         bool              entry_pending;
 417         unsigned long     nr_events;
 418         char              *entry_str;
 419         double            runtime_ms;
 420 };
 421
 422 static struct thread_trace *thread_trace__new(void)
 423 {
 424         return zalloc(sizeof(struct thread_trace));
 425 }
 426
 427 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
 428 {
 429         struct thread_trace *ttrace;
 430
 431         if (thread == NULL)
 432                 goto fail;
 433
 434         if (thread->priv == NULL)
 435                 thread->priv = thread_trace__new();
 436
 437         if (thread->priv == NULL)
 438                 goto fail;
 439
 440         ttrace = thread->priv;
 441         ++ttrace->nr_events;
 442
 443         return ttrace;
 444 fail:
 445         color_fprintf(fp, PERF_COLOR_RED,
 446                       "WARNING: not enough memory, dropping samples!\n");
 447         return NULL;
 448 }
 449
 450 struct trace {
 451         struct perf_tool        tool;
 452         int                     audit_machine;
 453         struct {
 454                 int             max;
 455                 struct syscall  *table;
 456         } syscalls;
 457         struct perf_record_opts opts;
 458         struct machine          host;
 459         u64                     base_time;
 460         FILE                    *output;
 461         unsigned long           nr_events;
 462         struct strlist          *ev_qualifier;
 463         bool                    not_ev_qualifier;
 464         struct intlist          *tid_list;
 465         struct intlist          *pid_list;
 466         bool                    sched;
 467         bool                    multiple_threads;
 468         double                  duration_filter;
 469         double                  runtime_ms;
 470 };
 471
 472 static bool trace__filter_duration(struct trace *trace, double t)
 473 {
 474         return t < (trace->duration_filter * NSEC_PER_MSEC);
 475 }
 476
 477 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
 478 {
 479         double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
 480
 481         return fprintf(fp, "%10.3f ", ts);
 482 }
 483
 484 static bool done = false;
 485
 486 static void sig_handler(int sig __maybe_unused)
 487 {
 488         done = true;
 489 }
 490
 491 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
 492                                         u64 duration, u64 tstamp, FILE *fp)
 493 {
 494         size_t printed = trace__fprintf_tstamp(trace, tstamp, fp);
 495         printed += fprintf_duration(duration, fp);
 496
 497         if (trace->multiple_threads)
 498                 printed += fprintf(fp, "%d ", thread->tid);
 499
 500         return printed;
 501 }
 502
 503 static int trace__process_event(struct trace *trace, struct machine *machine,
 504                                 union perf_event *event)
 505 {
 506         int ret = 0;
 507
 508         switch (event->header.type) {
 509         case PERF_RECORD_LOST:
 510                 color_fprintf(trace->output, PERF_COLOR_RED,
 511                               "LOST %" PRIu64 " events!\n", event->lost.lost);
 512                 ret = machine__process_lost_event(machine, event);
 513         default:
 514                 ret = machine__process_event(machine, event);
 515                 break;
 516         }
 517
 518         return ret;
 519 }
 520
 521 static int trace__tool_process(struct perf_tool *tool,
 522                                union perf_event *event,
 523                                struct perf_sample *sample __maybe_unused,
 524                                struct machine *machine)
 525 {
 526         struct trace *trace = container_of(tool, struct trace, tool);
 527         return trace__process_event(trace, machine, event);
 528 }
 529
 530 static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
 531 {
 532         int err = symbol__init();
 533
 534         if (err)
 535                 return err;
 536
 537         machine__init(&trace->host, "", HOST_KERNEL_ID);
 538         machine__create_kernel_maps(&trace->host);
 539
 540         if (perf_target__has_task(&trace->opts.target)) {
 541                 err = perf_event__synthesize_thread_map(&trace->tool, evlist->threads,
 542                                                         trace__tool_process,
 543                                                         &trace->host);
 544         } else {
 545                 err = perf_event__synthesize_threads(&trace->tool, trace__tool_process,
 546                                                      &trace->host);
 547         }
 548
 549         if (err)
 550                 symbol__exit();
 551
 552         return err;
 553 }
 554
 555 static int syscall__set_arg_fmts(struct syscall *sc)
 556 {
 557         struct format_field *field;
 558         int idx = 0;
 559
 560         sc->arg_scnprintf = calloc(sc->tp_format->format.nr_fields - 1, sizeof(void *));
 561         if (sc->arg_scnprintf == NULL)
 562                 return -1;
 563
 564         if (sc->fmt)
 565                 sc->arg_parm = sc->fmt->arg_parm;
 566
 567         for (field = sc->tp_format->format.fields->next; field; field = field->next) {
 568                 if (sc->fmt && sc->fmt->arg_scnprintf[idx])
 569                         sc->arg_scnprintf[idx] = sc->fmt->arg_scnprintf[idx];
 570                 else if (field->flags & FIELD_IS_POINTER)
 571                         sc->arg_scnprintf[idx] = syscall_arg__scnprintf_hex;
 572                 ++idx;
 573         }
 574
 575         return 0;
 576 }
 577
 578 static int trace__read_syscall_info(struct trace *trace, int id)
 579 {
 580         char tp_name[128];
 581         struct syscall *sc;
 582         const char *name = audit_syscall_to_name(id, trace->audit_machine);
 583
 584         if (name == NULL)
 585                 return -1;
 586
 587         if (id > trace->syscalls.max) {
 588                 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
 589
 590                 if (nsyscalls == NULL)
 591                         return -1;
 592
 593                 if (trace->syscalls.max != -1) {
 594                         memset(nsyscalls + trace->syscalls.max + 1, 0,
 595                                (id - trace->syscalls.max) * sizeof(*sc));
 596                 } else {
 597                         memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
 598                 }
 599
 600                 trace->syscalls.table = nsyscalls;
 601                 trace->syscalls.max   = id;
 602         }
 603
 604         sc = trace->syscalls.table + id;
 605         sc->name = name;
 606
 607         if (trace->ev_qualifier) {
 608                 bool in = strlist__find(trace->ev_qualifier, name) != NULL;
 609
 610                 if (!(in ^ trace->not_ev_qualifier)) {
 611                         sc->filtered = true;
 612                         /*
 613                          * No need to do read tracepoint information since this will be
 614                          * filtered out.
 615                          */
 616                         return 0;
 617                 }
 618         }
 619
 620         sc->fmt  = syscall_fmt__find(sc->name);
 621
 622         snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
 623         sc->tp_format = event_format__new("syscalls", tp_name);
 624
 625         if (sc->tp_format == NULL && sc->fmt && sc->fmt->alias) {
 626                 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
 627                 sc->tp_format = event_format__new("syscalls", tp_name);
 628         }
 629
 630         if (sc->tp_format == NULL)
 631                 return -1;
 632
 633         return syscall__set_arg_fmts(sc);
 634 }
 635
 636 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
 637                                       unsigned long *args)
 638 {
 639         size_t printed = 0;
 640
 641         if (sc->tp_format != NULL) {
 642                 struct format_field *field;
 643                 u8 bit = 1;
 644                 struct syscall_arg arg = {
 645                         .idx  = 0,
 646                         .mask = 0,
 647                 };
 648
 649                 for (field = sc->tp_format->format.fields->next; field;
 650                      field = field->next, ++arg.idx, bit <<= 1) {
 651                         if (arg.mask & bit)
 652                                 continue;
 653
 654                         printed += scnprintf(bf + printed, size - printed,
 655                                              "%s%s: ", printed ? ", " : "", field->name);
 656                         if (sc->arg_scnprintf && sc->arg_scnprintf[arg.idx]) {
 657                                 arg.val = args[arg.idx];
 658                                 if (sc->arg_parm)
 659                                         arg.parm = sc->arg_parm[arg.idx];
 660                                 printed += sc->arg_scnprintf[arg.idx](bf + printed,
 661                                                                       size - printed, &arg);
 662                         } else {
 663                                 printed += scnprintf(bf + printed, size - printed,
 664                                                      "%ld", args[arg.idx]);
 665                         }
 666                 }
 667         } else {
 668                 int i = 0;
 669
 670                 while (i < 6) {
 671                         printed += scnprintf(bf + printed, size - printed,
 672                                              "%sarg%d: %ld",
 673                                              printed ? ", " : "", i, args[i]);
 674                         ++i;
 675                 }
 676         }
 677
 678         return printed;
 679 }
 680
 681 typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
 682                                   struct perf_sample *sample);
 683
 684 static struct syscall *trace__syscall_info(struct trace *trace,
 685                                            struct perf_evsel *evsel,
 686                                            struct perf_sample *sample)
 687 {
 688         int id = perf_evsel__intval(evsel, sample, "id");
 689
 690         if (id < 0) {
 691
 692                 /*
 693                  * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
 694                  * before that, leaving at a higher verbosity level till that is
 695                  * explained. Reproduced with plain ftrace with:
 696                  *
 697                  * echo 1 > /t/events/raw_syscalls/sys_exit/enable
 698                  * grep "NR -1 " /t/trace_pipe
 699                  *
 700                  * After generating some load on the machine.
 701                  */
 702                 if (verbose > 1) {
 703                         static u64 n;
 704                         fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
 705                                 id, perf_evsel__name(evsel), ++n);
 706                 }
 707                 return NULL;
 708         }
 709
 710         if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
 711             trace__read_syscall_info(trace, id))
 712                 goto out_cant_read;
 713
 714         if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
 715                 goto out_cant_read;
 716
 717         return &trace->syscalls.table[id];
 718
 719 out_cant_read:
 720         if (verbose) {
 721                 fprintf(trace->output, "Problems reading syscall %d", id);
 722                 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
 723                         fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
 724                 fputs(" information\n", trace->output);
 725         }
 726         return NULL;
 727 }
 728
 729 static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
 730                             struct perf_sample *sample)
 731 {
 732         char *msg;
 733         void *args;
 734         size_t printed = 0;
 735         struct thread *thread;
 736         struct syscall *sc = trace__syscall_info(trace, evsel, sample);
 737         struct thread_trace *ttrace;
 738
 739         if (sc == NULL)
 740                 return -1;
 741
 742         if (sc->filtered)
 743                 return 0;
 744
 745         thread = machine__findnew_thread(&trace->host, sample->pid,
 746                                          sample->tid);
 747         ttrace = thread__trace(thread, trace->output);
 748         if (ttrace == NULL)
 749                 return -1;
 750
 751         args = perf_evsel__rawptr(evsel, sample, "args");
 752         if (args == NULL) {
 753                 fprintf(trace->output, "Problems reading syscall arguments\n");
 754                 return -1;
 755         }
 756
 757         ttrace = thread->priv;
 758
 759         if (ttrace->entry_str == NULL) {
 760                 ttrace->entry_str = malloc(1024);
 761                 if (!ttrace->entry_str)
 762                         return -1;
 763         }
 764
 765         ttrace->entry_time = sample->time;
 766         msg = ttrace->entry_str;
 767         printed += scnprintf(msg + printed, 1024 - printed, "%s(", sc->name);
 768
 769         printed += syscall__scnprintf_args(sc, msg + printed, 1024 - printed,  args);
 770
 771         if (!strcmp(sc->name, "exit_group") || !strcmp(sc->name, "exit")) {
 772                 if (!trace->duration_filter) {
 773                         trace__fprintf_entry_head(trace, thread, 1, sample->time, trace->output);
 774                         fprintf(trace->output, "%-70s\n", ttrace->entry_str);
 775                 }
 776         } else
 777                 ttrace->entry_pending = true;
 778
 779         return 0;
 780 }
 781
 782 static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
 783                            struct perf_sample *sample)
 784 {
 785         int ret;
 786         u64 duration = 0;
 787         struct thread *thread;
 788         struct syscall *sc = trace__syscall_info(trace, evsel, sample);
 789         struct thread_trace *ttrace;
 790
 791         if (sc == NULL)
 792                 return -1;
 793
 794         if (sc->filtered)
 795                 return 0;
 796
 797         thread = machine__findnew_thread(&trace->host, sample->pid,
 798                                          sample->tid);
 799         ttrace = thread__trace(thread, trace->output);
 800         if (ttrace == NULL)
 801                 return -1;
 802
 803         ret = perf_evsel__intval(evsel, sample, "ret");
 804
 805         ttrace = thread->priv;
 806
 807         ttrace->exit_time = sample->time;
 808
 809         if (ttrace->entry_time) {
 810                 duration = sample->time - ttrace->entry_time;
 811                 if (trace__filter_duration(trace, duration))
 812                         goto out;
 813         } else if (trace->duration_filter)
 814                 goto out;
 815
 816         trace__fprintf_entry_head(trace, thread, duration, sample->time, trace->output);
 817
 818         if (ttrace->entry_pending) {
 819                 fprintf(trace->output, "%-70s", ttrace->entry_str);
 820         } else {
 821                 fprintf(trace->output, " ... [");
 822                 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
 823                 fprintf(trace->output, "]: %s()", sc->name);
 824         }
 825
 826         if (sc->fmt == NULL) {
 827 signed_print:
 828                 fprintf(trace->output, ") = %d", ret);
 829         } else if (ret < 0 && sc->fmt->errmsg) {
 830                 char bf[256];
 831                 const char *emsg = strerror_r(-ret, bf, sizeof(bf)),
 832                            *e = audit_errno_to_name(-ret);
 833
 834                 fprintf(trace->output, ") = -1 %s %s", e, emsg);
 835         } else if (ret == 0 && sc->fmt->timeout)
 836                 fprintf(trace->output, ") = 0 Timeout");
 837         else if (sc->fmt->hexret)
 838                 fprintf(trace->output, ") = %#x", ret);
 839         else
 840                 goto signed_print;
 841
 842         fputc('\n', trace->output);
 843 out:
 844         ttrace->entry_pending = false;
 845
 846         return 0;
 847 }
 848
 849 static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
 850                                      struct perf_sample *sample)
 851 {
 852         u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
 853         double runtime_ms = (double)runtime / NSEC_PER_MSEC;
 854         struct thread *thread = machine__findnew_thread(&trace->host,
 855                                                         sample->pid,
 856                                                         sample->tid);
 857         struct thread_trace *ttrace = thread__trace(thread, trace->output);
 858
 859         if (ttrace == NULL)
 860                 goto out_dump;
 861
 862         ttrace->runtime_ms += runtime_ms;
 863         trace->runtime_ms += runtime_ms;
 864         return 0;
 865
 866 out_dump:
 867         fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
 868                evsel->name,
 869                perf_evsel__strval(evsel, sample, "comm"),
 870                (pid_t)perf_evsel__intval(evsel, sample, "pid"),
 871                runtime,
 872                perf_evsel__intval(evsel, sample, "vruntime"));
 873         return 0;
 874 }
 875
 876 static bool skip_sample(struct trace *trace, struct perf_sample *sample)
 877 {
 878         if ((trace->pid_list && intlist__find(trace->pid_list, sample->pid)) ||
 879             (trace->tid_list && intlist__find(trace->tid_list, sample->tid)))
 880                 return false;
 881
 882         if (trace->pid_list || trace->tid_list)
 883                 return true;
 884
 885         return false;
 886 }
 887
 888 static int trace__process_sample(struct perf_tool *tool,
 889                                  union perf_event *event __maybe_unused,
 890                                  struct perf_sample *sample,
 891                                  struct perf_evsel *evsel,
 892                                  struct machine *machine __maybe_unused)
 893 {
 894         struct trace *trace = container_of(tool, struct trace, tool);
 895         int err = 0;
 896
 897         tracepoint_handler handler = evsel->handler.func;
 898
 899         if (skip_sample(trace, sample))
 900                 return 0;
 901
 902         if (trace->base_time == 0)
 903                 trace->base_time = sample->time;
 904
 905         if (handler)
 906                 handler(trace, evsel, sample);
 907
 908         return err;
 909 }
 910
 911 static bool
 912 perf_session__has_tp(struct perf_session *session, const char *name)
 913 {
 914         struct perf_evsel *evsel;
 915
 916         evsel = perf_evlist__find_tracepoint_by_name(session->evlist, name);
 917
 918         return evsel != NULL;
 919 }
 920
 921 static int parse_target_str(struct trace *trace)
 922 {
 923         if (trace->opts.target.pid) {
 924                 trace->pid_list = intlist__new(trace->opts.target.pid);
 925                 if (trace->pid_list == NULL) {
 926                         pr_err("Error parsing process id string\n");
 927                         return -EINVAL;
 928                 }
 929         }
 930
 931         if (trace->opts.target.tid) {
 932                 trace->tid_list = intlist__new(trace->opts.target.tid);
 933                 if (trace->tid_list == NULL) {
 934                         pr_err("Error parsing thread id string\n");
 935                         return -EINVAL;
 936                 }
 937         }
 938
 939         return 0;
 940 }
 941
 942 static int trace__run(struct trace *trace, int argc, const char **argv)
 943 {
 944         struct perf_evlist *evlist = perf_evlist__new();
 945         struct perf_evsel *evsel;
 946         int err = -1, i;
 947         unsigned long before;
 948         const bool forks = argc > 0;
 949
 950         if (evlist == NULL) {
 951                 fprintf(trace->output, "Not enough memory to run!\n");
 952                 goto out;
 953         }
 954
 955         if (perf_evlist__add_newtp(evlist, "raw_syscalls", "sys_enter", trace__sys_enter) ||
 956             perf_evlist__add_newtp(evlist, "raw_syscalls", "sys_exit", trace__sys_exit)) {
 957                 fprintf(trace->output, "Couldn't read the raw_syscalls tracepoints information!\n");
 958                 goto out_delete_evlist;
 959         }
 960
 961         if (trace->sched &&
 962             perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
 963                                    trace__sched_stat_runtime)) {
 964                 fprintf(trace->output, "Couldn't read the sched_stat_runtime tracepoint information!\n");
 965                 goto out_delete_evlist;
 966         }
 967
 968         err = perf_evlist__create_maps(evlist, &trace->opts.target);
 969         if (err < 0) {
 970                 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
 971                 goto out_delete_evlist;
 972         }
 973
 974         err = trace__symbols_init(trace, evlist);
 975         if (err < 0) {
 976                 fprintf(trace->output, "Problems initializing symbol libraries!\n");
 977                 goto out_delete_maps;
 978         }
 979
 980         perf_evlist__config(evlist, &trace->opts);
 981
 982         signal(SIGCHLD, sig_handler);
 983         signal(SIGINT, sig_handler);
 984
 985         if (forks) {
 986                 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
 987                                                     argv, false, false);
 988                 if (err < 0) {
 989                         fprintf(trace->output, "Couldn't run the workload!\n");
 990                         goto out_delete_maps;
 991                 }
 992         }
 993
 994         err = perf_evlist__open(evlist);
 995         if (err < 0) {
 996                 fprintf(trace->output, "Couldn't create the events: %s\n", strerror(errno));
 997                 goto out_delete_maps;
 998         }
 999
1000         err = perf_evlist__mmap(evlist, UINT_MAX, false);
1001         if (err < 0) {
1002                 fprintf(trace->output, "Couldn't mmap the events: %s\n", strerror(errno));
1003                 goto out_close_evlist;
1004         }
1005
1006         perf_evlist__enable(evlist);
1007
1008         if (forks)
1009                 perf_evlist__start_workload(evlist);
1010
1011         trace->multiple_threads = evlist->threads->map[0] == -1 || evlist->threads->nr > 1;
1012 again:
1013         before = trace->nr_events;
1014
1015         for (i = 0; i < evlist->nr_mmaps; i++) {
1016                 union perf_event *event;
1017
1018                 while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
1019                         const u32 type = event->header.type;
1020                         tracepoint_handler handler;
1021                         struct perf_sample sample;
1022
1023                         ++trace->nr_events;
1024
1025                         err = perf_evlist__parse_sample(evlist, event, &sample);
1026                         if (err) {
1027                                 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
1028                                 continue;
1029                         }
1030
1031                         if (trace->base_time == 0)
1032                                 trace->base_time = sample.time;
1033
1034                         if (type != PERF_RECORD_SAMPLE) {
1035                                 trace__process_event(trace, &trace->host, event);
1036                                 continue;
1037                         }
1038
1039                         evsel = perf_evlist__id2evsel(evlist, sample.id);
1040                         if (evsel == NULL) {
1041                                 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample.id);
1042                                 continue;
1043                         }
1044
1045                         if (sample.raw_data == NULL) {
1046                                 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
1047                                        perf_evsel__name(evsel), sample.tid,
1048                                        sample.cpu, sample.raw_size);
1049                                 continue;
1050                         }
1051
1052                         handler = evsel->handler.func;
1053                         handler(trace, evsel, &sample);
1054
1055                         if (done)
1056                                 goto out_unmap_evlist;
1057                 }
1058         }
1059
1060         if (trace->nr_events == before) {
1061                 if (done)
1062                         goto out_unmap_evlist;
1063
1064                 poll(evlist->pollfd, evlist->nr_fds, -1);
1065         }
1066
1067         if (done)
1068                 perf_evlist__disable(evlist);
1069
1070         goto again;
1071
1072 out_unmap_evlist:
1073         perf_evlist__munmap(evlist);
1074 out_close_evlist:
1075         perf_evlist__close(evlist);
1076 out_delete_maps:
1077         perf_evlist__delete_maps(evlist);
1078 out_delete_evlist:
1079         perf_evlist__delete(evlist);
1080 out:
1081         return err;
1082 }
1083
1084 static int trace__replay(struct trace *trace)
1085 {
1086         const struct perf_evsel_str_handler handlers[] = {
1087                 { "raw_syscalls:sys_enter",  trace__sys_enter, },
1088                 { "raw_syscalls:sys_exit",   trace__sys_exit, },
1089         };
1090
1091         struct perf_session *session;
1092         int err = -1;
1093
1094         trace->tool.sample        = trace__process_sample;
1095         trace->tool.mmap          = perf_event__process_mmap;
1096         trace->tool.mmap2         = perf_event__process_mmap2;
1097         trace->tool.comm          = perf_event__process_comm;
1098         trace->tool.exit          = perf_event__process_exit;
1099         trace->tool.fork          = perf_event__process_fork;
1100         trace->tool.attr          = perf_event__process_attr;
1101         trace->tool.tracing_data = perf_event__process_tracing_data;
1102         trace->tool.build_id      = perf_event__process_build_id;
1103
1104         trace->tool.ordered_samples = true;
1105         trace->tool.ordering_requires_timestamps = true;
1106
1107         /* add tid to output */
1108         trace->multiple_threads = true;
1109
1110         if (symbol__init() < 0)
1111                 return -1;
1112
1113         session = perf_session__new(input_name, O_RDONLY, 0, false,
1114                                     &trace->tool);
1115         if (session == NULL)
1116                 return -ENOMEM;
1117
1118         err = perf_session__set_tracepoints_handlers(session, handlers);
1119         if (err)
1120                 goto out;
1121
1122         if (!perf_session__has_tp(session, "raw_syscalls:sys_enter")) {
1123                 pr_err("Data file does not have raw_syscalls:sys_enter events\n");
1124                 goto out;
1125         }
1126
1127         if (!perf_session__has_tp(session, "raw_syscalls:sys_exit")) {
1128                 pr_err("Data file does not have raw_syscalls:sys_exit events\n");
1129                 goto out;
1130         }
1131
1132         err = parse_target_str(trace);
1133         if (err != 0)
1134                 goto out;
1135
1136         setup_pager();
1137
1138         err = perf_session__process_events(session, &trace->tool);
1139         if (err)
1140                 pr_err("Failed to process events, error %d", err);
1141
1142 out:
1143         perf_session__delete(session);
1144
1145         return err;
1146 }
1147
1148 static size_t trace__fprintf_threads_header(FILE *fp)
1149 {
1150         size_t printed;
1151
1152         printed  = fprintf(fp, "\n _____________________________________________________________________\n");
1153         printed += fprintf(fp," __)    Summary of events    (__\n\n");
1154         printed += fprintf(fp,"              [ task - pid ]     [ events ] [ ratio ]  [ runtime ]\n");
1155         printed += fprintf(fp," _____________________________________________________________________\n\n");
1156
1157         return printed;
1158 }
1159
1160 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
1161 {
1162         size_t printed = trace__fprintf_threads_header(fp);
1163         struct rb_node *nd;
1164
1165         for (nd = rb_first(&trace->host.threads); nd; nd = rb_next(nd)) {
1166                 struct thread *thread = rb_entry(nd, struct thread, rb_node);
1167                 struct thread_trace *ttrace = thread->priv;
1168                 const char *color;
1169                 double ratio;
1170
1171                 if (ttrace == NULL)
1172                         continue;
1173
1174                 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
1175
1176                 color = PERF_COLOR_NORMAL;
1177                 if (ratio > 50.0)
1178                         color = PERF_COLOR_RED;
1179                 else if (ratio > 25.0)
1180                         color = PERF_COLOR_GREEN;
1181                 else if (ratio > 5.0)
1182                         color = PERF_COLOR_YELLOW;
1183
1184                 printed += color_fprintf(fp, color, "%20s", thread->comm);
1185                 printed += fprintf(fp, " - %-5d :%11lu   [", thread->tid, ttrace->nr_events);
1186                 printed += color_fprintf(fp, color, "%5.1f%%", ratio);
1187                 printed += fprintf(fp, " ] %10.3f ms\n", ttrace->runtime_ms);
1188         }
1189
1190         return printed;
1191 }
1192
1193 static int trace__set_duration(const struct option *opt, const char *str,
1194                                int unset __maybe_unused)
1195 {
1196         struct trace *trace = opt->value;
1197
1198         trace->duration_filter = atof(str);
1199         return 0;
1200 }
1201
1202 static int trace__open_output(struct trace *trace, const char *filename)
1203 {
1204         struct stat st;
1205
1206         if (!stat(filename, &st) && st.st_size) {
1207                 char oldname[PATH_MAX];
1208
1209                 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
1210                 unlink(oldname);
1211                 rename(filename, oldname);
1212         }
1213
1214         trace->output = fopen(filename, "w");
1215
1216         return trace->output == NULL ? -errno : 0;
1217 }
1218
1219 int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
1220 {
1221         const char * const trace_usage[] = {
1222                 "perf trace [<options>] [<command>]",
1223                 "perf trace [<options>] -- <command> [<options>]",
1224                 NULL
1225         };
1226         struct trace trace = {
1227                 .audit_machine = audit_detect_machine(),
1228                 .syscalls = {
1229                         . max = -1,
1230                 },
1231                 .opts = {
1232                         .target = {
1233                                 .uid       = UINT_MAX,
1234                                 .uses_mmap = true,
1235                         },
1236                         .user_freq     = UINT_MAX,
1237                         .user_interval = ULLONG_MAX,
1238                         .no_delay      = true,
1239                         .mmap_pages    = 1024,
1240                 },
1241                 .output = stdout,
1242         };
1243         const char *output_name = NULL;
1244         const char *ev_qualifier_str = NULL;
1245         const struct option trace_options[] = {
1246         OPT_STRING('e', "expr", &ev_qualifier_str, "expr",
1247                     "list of events to trace"),
1248         OPT_STRING('o', "output", &output_name, "file", "output file name"),
1249         OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
1250         OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
1251                     "trace events on existing process id"),
1252         OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
1253                     "trace events on existing thread id"),
1254         OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
1255                     "system-wide collection from all CPUs"),
1256         OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
1257                     "list of cpus to monitor"),
1258         OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
1259                     "child tasks do not inherit counters"),
1260         OPT_UINTEGER('m', "mmap-pages", &trace.opts.mmap_pages,
1261                      "number of mmap data pages"),
1262         OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
1263                    "user to profile"),
1264         OPT_CALLBACK(0, "duration", &trace, "float",
1265                      "show only events with duration > N.M ms",
1266                      trace__set_duration),
1267         OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
1268         OPT_INCR('v', "verbose", &verbose, "be more verbose"),
1269         OPT_END()
1270         };
1271         int err;
1272         char bf[BUFSIZ];
1273
1274         argc = parse_options(argc, argv, trace_options, trace_usage, 0);
1275
1276         if (output_name != NULL) {
1277                 err = trace__open_output(&trace, output_name);
1278                 if (err < 0) {
1279                         perror("failed to create output file");
1280                         goto out;
1281                 }
1282         }
1283
1284         if (ev_qualifier_str != NULL) {
1285                 const char *s = ev_qualifier_str;
1286
1287                 trace.not_ev_qualifier = *s == '!';
1288                 if (trace.not_ev_qualifier)
1289                         ++s;
1290                 trace.ev_qualifier = strlist__new(true, s);
1291                 if (trace.ev_qualifier == NULL) {
1292                         fputs("Not enough memory to parse event qualifier",
1293                               trace.output);
1294                         err = -ENOMEM;
1295                         goto out_close;
1296                 }
1297         }
1298
1299         err = perf_target__validate(&trace.opts.target);
1300         if (err) {
1301                 perf_target__strerror(&trace.opts.target, err, bf, sizeof(bf));
1302                 fprintf(trace.output, "%s", bf);
1303                 goto out_close;
1304         }
1305
1306         err = perf_target__parse_uid(&trace.opts.target);
1307         if (err) {
1308                 perf_target__strerror(&trace.opts.target, err, bf, sizeof(bf));
1309                 fprintf(trace.output, "%s", bf);
1310                 goto out_close;
1311         }
1312
1313         if (!argc && perf_target__none(&trace.opts.target))
1314                 trace.opts.target.system_wide = true;
1315
1316         if (input_name)
1317                 err = trace__replay(&trace);
1318         else
1319                 err = trace__run(&trace, argc, argv);
1320
1321         if (trace.sched && !err)
1322                 trace__fprintf_thread_summary(&trace, trace.output);
1323
1324 out_close:
1325         if (output_name != NULL)
1326                 fclose(trace.output);
1327 out:
1328         return err;
1329 }