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 #include "util/stat.h"
  14 #include "trace-event.h"
  15 #include "util/parse-events.h"
  16
  17 #include <libaudit.h>
  18 #include <stdlib.h>
  19 #include <sys/eventfd.h>
  20 #include <sys/mman.h>
  21 #include <linux/futex.h>
  22
  23 /* For older distros: */
  24 #ifndef MAP_STACK
  25 # define MAP_STACK              0x20000
  26 #endif
  27
  28 #ifndef MADV_HWPOISON
  29 # define MADV_HWPOISON          100
  30 #endif
  31
  32 #ifndef MADV_MERGEABLE
  33 # define MADV_MERGEABLE         12
  34 #endif
  35
  36 #ifndef MADV_UNMERGEABLE
  37 # define MADV_UNMERGEABLE       13
  38 #endif
  39
  40 #ifndef EFD_SEMAPHORE
  41 # define EFD_SEMAPHORE          1
  42 #endif
  43
  44 struct tp_field {
  45         int offset;
  46         union {
  47                 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
  48                 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
  49         };
  50 };
  51
  52 #define TP_UINT_FIELD(bits) \
  53 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
  54 { \
  55         return *(u##bits *)(sample->raw_data + field->offset); \
  56 }
  57
  58 TP_UINT_FIELD(8);
  59 TP_UINT_FIELD(16);
  60 TP_UINT_FIELD(32);
  61 TP_UINT_FIELD(64);
  62
  63 #define TP_UINT_FIELD__SWAPPED(bits) \
  64 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
  65 { \
  66         u##bits value = *(u##bits *)(sample->raw_data + field->offset); \
  67         return bswap_##bits(value);\
  68 }
  69
  70 TP_UINT_FIELD__SWAPPED(16);
  71 TP_UINT_FIELD__SWAPPED(32);
  72 TP_UINT_FIELD__SWAPPED(64);
  73
  74 static int tp_field__init_uint(struct tp_field *field,
  75                                struct format_field *format_field,
  76                                bool needs_swap)
  77 {
  78         field->offset = format_field->offset;
  79
  80         switch (format_field->size) {
  81         case 1:
  82                 field->integer = tp_field__u8;
  83                 break;
  84         case 2:
  85                 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
  86                 break;
  87         case 4:
  88                 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
  89                 break;
  90         case 8:
  91                 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
  92                 break;
  93         default:
  94                 return -1;
  95         }
  96
  97         return 0;
  98 }
  99
 100 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
 101 {
 102         return sample->raw_data + field->offset;
 103 }
 104
 105 static int tp_field__init_ptr(struct tp_field *field, struct format_field *format_field)
 106 {
 107         field->offset = format_field->offset;
 108         field->pointer = tp_field__ptr;
 109         return 0;
 110 }
 111
 112 struct syscall_tp {
 113         struct tp_field id;
 114         union {
 115                 struct tp_field args, ret;
 116         };
 117 };
 118
 119 static int perf_evsel__init_tp_uint_field(struct perf_evsel *evsel,
 120                                           struct tp_field *field,
 121                                           const char *name)
 122 {
 123         struct format_field *format_field = perf_evsel__field(evsel, name);
 124
 125         if (format_field == NULL)
 126                 return -1;
 127
 128         return tp_field__init_uint(field, format_field, evsel->needs_swap);
 129 }
 130
 131 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
 132         ({ struct syscall_tp *sc = evsel->priv;\
 133            perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
 134
 135 static int perf_evsel__init_tp_ptr_field(struct perf_evsel *evsel,
 136                                          struct tp_field *field,
 137                                          const char *name)
 138 {
 139         struct format_field *format_field = perf_evsel__field(evsel, name);
 140
 141         if (format_field == NULL)
 142                 return -1;
 143
 144         return tp_field__init_ptr(field, format_field);
 145 }
 146
 147 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
 148         ({ struct syscall_tp *sc = evsel->priv;\
 149            perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
 150
 151 static void perf_evsel__delete_priv(struct perf_evsel *evsel)
 152 {
 153         zfree(&evsel->priv);
 154         perf_evsel__delete(evsel);
 155 }
 156
 157 static int perf_evsel__init_syscall_tp(struct perf_evsel *evsel, void *handler)
 158 {
 159         evsel->priv = malloc(sizeof(struct syscall_tp));
 160         if (evsel->priv != NULL) {
 161                 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
 162                         goto out_delete;
 163
 164                 evsel->handler = handler;
 165                 return 0;
 166         }
 167
 168         return -ENOMEM;
 169
 170 out_delete:
 171         zfree(&evsel->priv);
 172         return -ENOENT;
 173 }
 174
 175 static struct perf_evsel *perf_evsel__syscall_newtp(const char *direction, void *handler)
 176 {
 177         struct perf_evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
 178
 179         /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
 180         if (evsel == NULL)
 181                 evsel = perf_evsel__newtp("syscalls", direction);
 182
 183         if (evsel) {
 184                 if (perf_evsel__init_syscall_tp(evsel, handler))
 185                         goto out_delete;
 186         }
 187
 188         return evsel;
 189
 190 out_delete:
 191         perf_evsel__delete_priv(evsel);
 192         return NULL;
 193 }
 194
 195 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
 196         ({ struct syscall_tp *fields = evsel->priv; \
 197            fields->name.integer(&fields->name, sample); })
 198
 199 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
 200         ({ struct syscall_tp *fields = evsel->priv; \
 201            fields->name.pointer(&fields->name, sample); })
 202
 203 static int perf_evlist__add_syscall_newtp(struct perf_evlist *evlist,
 204                                           void *sys_enter_handler,
 205                                           void *sys_exit_handler)
 206 {
 207         int ret = -1;
 208         struct perf_evsel *sys_enter, *sys_exit;
 209
 210         sys_enter = perf_evsel__syscall_newtp("sys_enter", sys_enter_handler);
 211         if (sys_enter == NULL)
 212                 goto out;
 213
 214         if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
 215                 goto out_delete_sys_enter;
 216
 217         sys_exit = perf_evsel__syscall_newtp("sys_exit", sys_exit_handler);
 218         if (sys_exit == NULL)
 219                 goto out_delete_sys_enter;
 220
 221         if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
 222                 goto out_delete_sys_exit;
 223
 224         perf_evlist__add(evlist, sys_enter);
 225         perf_evlist__add(evlist, sys_exit);
 226
 227         ret = 0;
 228 out:
 229         return ret;
 230
 231 out_delete_sys_exit:
 232         perf_evsel__delete_priv(sys_exit);
 233 out_delete_sys_enter:
 234         perf_evsel__delete_priv(sys_enter);
 235         goto out;
 236 }
 237
 238
 239 struct syscall_arg {
 240         unsigned long val;
 241         struct thread *thread;
 242         struct trace  *trace;
 243         void          *parm;
 244         u8            idx;
 245         u8            mask;
 246 };
 247
 248 struct strarray {
 249         int         offset;
 250         int         nr_entries;
 251         const char **entries;
 252 };
 253
 254 #define DEFINE_STRARRAY(array) struct strarray strarray__##array = { \
 255         .nr_entries = ARRAY_SIZE(array), \
 256         .entries = array, \
 257 }
 258
 259 #define DEFINE_STRARRAY_OFFSET(array, off) struct strarray strarray__##array = { \
 260         .offset     = off, \
 261         .nr_entries = ARRAY_SIZE(array), \
 262         .entries = array, \
 263 }
 264
 265 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
 266                                                 const char *intfmt,
 267                                                 struct syscall_arg *arg)
 268 {
 269         struct strarray *sa = arg->parm;
 270         int idx = arg->val - sa->offset;
 271
 272         if (idx < 0 || idx >= sa->nr_entries)
 273                 return scnprintf(bf, size, intfmt, arg->val);
 274
 275         return scnprintf(bf, size, "%s", sa->entries[idx]);
 276 }
 277
 278 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
 279                                               struct syscall_arg *arg)
 280 {
 281         return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
 282 }
 283
 284 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
 285
 286 #if defined(__i386__) || defined(__x86_64__)
 287 /*
 288  * FIXME: Make this available to all arches as soon as the ioctl beautifier
 289  *        gets rewritten to support all arches.
 290  */
 291 static size_t syscall_arg__scnprintf_strhexarray(char *bf, size_t size,
 292                                                  struct syscall_arg *arg)
 293 {
 294         return __syscall_arg__scnprintf_strarray(bf, size, "%#x", arg);
 295 }
 296
 297 #define SCA_STRHEXARRAY syscall_arg__scnprintf_strhexarray
 298 #endif /* defined(__i386__) || defined(__x86_64__) */
 299
 300 static size_t syscall_arg__scnprintf_fd(char *bf, size_t size,
 301                                         struct syscall_arg *arg);
 302
 303 #define SCA_FD syscall_arg__scnprintf_fd
 304
 305 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
 306                                            struct syscall_arg *arg)
 307 {
 308         int fd = arg->val;
 309
 310         if (fd == AT_FDCWD)
 311                 return scnprintf(bf, size, "CWD");
 312
 313         return syscall_arg__scnprintf_fd(bf, size, arg);
 314 }
 315
 316 #define SCA_FDAT syscall_arg__scnprintf_fd_at
 317
 318 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
 319                                               struct syscall_arg *arg);
 320
 321 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
 322
 323 static size_t syscall_arg__scnprintf_hex(char *bf, size_t size,
 324                                          struct syscall_arg *arg)
 325 {
 326         return scnprintf(bf, size, "%#lx", arg->val);
 327 }
 328
 329 #define SCA_HEX syscall_arg__scnprintf_hex
 330
 331 static size_t syscall_arg__scnprintf_mmap_prot(char *bf, size_t size,
 332                                                struct syscall_arg *arg)
 333 {
 334         int printed = 0, prot = arg->val;
 335
 336         if (prot == PROT_NONE)
 337                 return scnprintf(bf, size, "NONE");
 338 #define P_MMAP_PROT(n) \
 339         if (prot & PROT_##n) { \
 340                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 341                 prot &= ~PROT_##n; \
 342         }
 343
 344         P_MMAP_PROT(EXEC);
 345         P_MMAP_PROT(READ);
 346         P_MMAP_PROT(WRITE);
 347 #ifdef PROT_SEM
 348         P_MMAP_PROT(SEM);
 349 #endif
 350         P_MMAP_PROT(GROWSDOWN);
 351         P_MMAP_PROT(GROWSUP);
 352 #undef P_MMAP_PROT
 353
 354         if (prot)
 355                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", prot);
 356
 357         return printed;
 358 }
 359
 360 #define SCA_MMAP_PROT syscall_arg__scnprintf_mmap_prot
 361
 362 static size_t syscall_arg__scnprintf_mmap_flags(char *bf, size_t size,
 363                                                 struct syscall_arg *arg)
 364 {
 365         int printed = 0, flags = arg->val;
 366
 367 #define P_MMAP_FLAG(n) \
 368         if (flags & MAP_##n) { \
 369                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 370                 flags &= ~MAP_##n; \
 371         }
 372
 373         P_MMAP_FLAG(SHARED);
 374         P_MMAP_FLAG(PRIVATE);
 375 #ifdef MAP_32BIT
 376         P_MMAP_FLAG(32BIT);
 377 #endif
 378         P_MMAP_FLAG(ANONYMOUS);
 379         P_MMAP_FLAG(DENYWRITE);
 380         P_MMAP_FLAG(EXECUTABLE);
 381         P_MMAP_FLAG(FILE);
 382         P_MMAP_FLAG(FIXED);
 383         P_MMAP_FLAG(GROWSDOWN);
 384 #ifdef MAP_HUGETLB
 385         P_MMAP_FLAG(HUGETLB);
 386 #endif
 387         P_MMAP_FLAG(LOCKED);
 388         P_MMAP_FLAG(NONBLOCK);
 389         P_MMAP_FLAG(NORESERVE);
 390         P_MMAP_FLAG(POPULATE);
 391         P_MMAP_FLAG(STACK);
 392 #ifdef MAP_UNINITIALIZED
 393         P_MMAP_FLAG(UNINITIALIZED);
 394 #endif
 395 #undef P_MMAP_FLAG
 396
 397         if (flags)
 398                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 399
 400         return printed;
 401 }
 402
 403 #define SCA_MMAP_FLAGS syscall_arg__scnprintf_mmap_flags
 404
 405 static size_t syscall_arg__scnprintf_mremap_flags(char *bf, size_t size,
 406                                                   struct syscall_arg *arg)
 407 {
 408         int printed = 0, flags = arg->val;
 409
 410 #define P_MREMAP_FLAG(n) \
 411         if (flags & MREMAP_##n) { \
 412                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 413                 flags &= ~MREMAP_##n; \
 414         }
 415
 416         P_MREMAP_FLAG(MAYMOVE);
 417 #ifdef MREMAP_FIXED
 418         P_MREMAP_FLAG(FIXED);
 419 #endif
 420 #undef P_MREMAP_FLAG
 421
 422         if (flags)
 423                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 424
 425         return printed;
 426 }
 427
 428 #define SCA_MREMAP_FLAGS syscall_arg__scnprintf_mremap_flags
 429
 430 static size_t syscall_arg__scnprintf_madvise_behavior(char *bf, size_t size,
 431                                                       struct syscall_arg *arg)
 432 {
 433         int behavior = arg->val;
 434
 435         switch (behavior) {
 436 #define P_MADV_BHV(n) case MADV_##n: return scnprintf(bf, size, #n)
 437         P_MADV_BHV(NORMAL);
 438         P_MADV_BHV(RANDOM);
 439         P_MADV_BHV(SEQUENTIAL);
 440         P_MADV_BHV(WILLNEED);
 441         P_MADV_BHV(DONTNEED);
 442         P_MADV_BHV(REMOVE);
 443         P_MADV_BHV(DONTFORK);
 444         P_MADV_BHV(DOFORK);
 445         P_MADV_BHV(HWPOISON);
 446 #ifdef MADV_SOFT_OFFLINE
 447         P_MADV_BHV(SOFT_OFFLINE);
 448 #endif
 449         P_MADV_BHV(MERGEABLE);
 450         P_MADV_BHV(UNMERGEABLE);
 451 #ifdef MADV_HUGEPAGE
 452         P_MADV_BHV(HUGEPAGE);
 453 #endif
 454 #ifdef MADV_NOHUGEPAGE
 455         P_MADV_BHV(NOHUGEPAGE);
 456 #endif
 457 #ifdef MADV_DONTDUMP
 458         P_MADV_BHV(DONTDUMP);
 459 #endif
 460 #ifdef MADV_DODUMP
 461         P_MADV_BHV(DODUMP);
 462 #endif
 463 #undef P_MADV_PHV
 464         default: break;
 465         }
 466
 467         return scnprintf(bf, size, "%#x", behavior);
 468 }
 469
 470 #define SCA_MADV_BHV syscall_arg__scnprintf_madvise_behavior
 471
 472 static size_t syscall_arg__scnprintf_flock(char *bf, size_t size,
 473                                            struct syscall_arg *arg)
 474 {
 475         int printed = 0, op = arg->val;
 476
 477         if (op == 0)
 478                 return scnprintf(bf, size, "NONE");
 479 #define P_CMD(cmd) \
 480         if ((op & LOCK_##cmd) == LOCK_##cmd) { \
 481                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #cmd); \
 482                 op &= ~LOCK_##cmd; \
 483         }
 484
 485         P_CMD(SH);
 486         P_CMD(EX);
 487         P_CMD(NB);
 488         P_CMD(UN);
 489         P_CMD(MAND);
 490         P_CMD(RW);
 491         P_CMD(READ);
 492         P_CMD(WRITE);
 493 #undef P_OP
 494
 495         if (op)
 496                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", op);
 497
 498         return printed;
 499 }
 500
 501 #define SCA_FLOCK syscall_arg__scnprintf_flock
 502
 503 static size_t syscall_arg__scnprintf_futex_op(char *bf, size_t size, struct syscall_arg *arg)
 504 {
 505         enum syscall_futex_args {
 506                 SCF_UADDR   = (1 << 0),
 507                 SCF_OP      = (1 << 1),
 508                 SCF_VAL     = (1 << 2),
 509                 SCF_TIMEOUT = (1 << 3),
 510                 SCF_UADDR2  = (1 << 4),
 511                 SCF_VAL3    = (1 << 5),
 512         };
 513         int op = arg->val;
 514         int cmd = op & FUTEX_CMD_MASK;
 515         size_t printed = 0;
 516
 517         switch (cmd) {
 518 #define P_FUTEX_OP(n) case FUTEX_##n: printed = scnprintf(bf, size, #n);
 519         P_FUTEX_OP(WAIT);           arg->mask |= SCF_VAL3|SCF_UADDR2;             break;
 520         P_FUTEX_OP(WAKE);           arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 521         P_FUTEX_OP(FD);             arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 522         P_FUTEX_OP(REQUEUE);        arg->mask |= SCF_VAL3|SCF_TIMEOUT;            break;
 523         P_FUTEX_OP(CMP_REQUEUE);    arg->mask |= SCF_TIMEOUT;                     break;
 524         P_FUTEX_OP(CMP_REQUEUE_PI); arg->mask |= SCF_TIMEOUT;                     break;
 525         P_FUTEX_OP(WAKE_OP);                                                      break;
 526         P_FUTEX_OP(LOCK_PI);        arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 527         P_FUTEX_OP(UNLOCK_PI);      arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
 528         P_FUTEX_OP(TRYLOCK_PI);     arg->mask |= SCF_VAL3|SCF_UADDR2;             break;
 529         P_FUTEX_OP(WAIT_BITSET);    arg->mask |= SCF_UADDR2;                      break;
 530         P_FUTEX_OP(WAKE_BITSET);    arg->mask |= SCF_UADDR2;                      break;
 531         P_FUTEX_OP(WAIT_REQUEUE_PI);                                              break;
 532         default: printed = scnprintf(bf, size, "%#x", cmd);                       break;
 533         }
 534
 535         if (op & FUTEX_PRIVATE_FLAG)
 536                 printed += scnprintf(bf + printed, size - printed, "|PRIV");
 537
 538         if (op & FUTEX_CLOCK_REALTIME)
 539                 printed += scnprintf(bf + printed, size - printed, "|CLKRT");
 540
 541         return printed;
 542 }
 543
 544 #define SCA_FUTEX_OP  syscall_arg__scnprintf_futex_op
 545
 546 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
 547 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, 1);
 548
 549 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
 550 static DEFINE_STRARRAY(itimers);
 551
 552 static const char *whences[] = { "SET", "CUR", "END",
 553 #ifdef SEEK_DATA
 554 "DATA",
 555 #endif
 556 #ifdef SEEK_HOLE
 557 "HOLE",
 558 #endif
 559 };
 560 static DEFINE_STRARRAY(whences);
 561
 562 static const char *fcntl_cmds[] = {
 563         "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
 564         "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "F_GETLK64",
 565         "F_SETLK64", "F_SETLKW64", "F_SETOWN_EX", "F_GETOWN_EX",
 566         "F_GETOWNER_UIDS",
 567 };
 568 static DEFINE_STRARRAY(fcntl_cmds);
 569
 570 static const char *rlimit_resources[] = {
 571         "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
 572         "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
 573         "RTTIME",
 574 };
 575 static DEFINE_STRARRAY(rlimit_resources);
 576
 577 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
 578 static DEFINE_STRARRAY(sighow);
 579
 580 static const char *clockid[] = {
 581         "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
 582         "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE",
 583 };
 584 static DEFINE_STRARRAY(clockid);
 585
 586 static const char *socket_families[] = {
 587         "UNSPEC", "LOCAL", "INET", "AX25", "IPX", "APPLETALK", "NETROM",
 588         "BRIDGE", "ATMPVC", "X25", "INET6", "ROSE", "DECnet", "NETBEUI",
 589         "SECURITY", "KEY", "NETLINK", "PACKET", "ASH", "ECONET", "ATMSVC",
 590         "RDS", "SNA", "IRDA", "PPPOX", "WANPIPE", "LLC", "IB", "CAN", "TIPC",
 591         "BLUETOOTH", "IUCV", "RXRPC", "ISDN", "PHONET", "IEEE802154", "CAIF",
 592         "ALG", "NFC", "VSOCK",
 593 };
 594 static DEFINE_STRARRAY(socket_families);
 595
 596 #ifndef SOCK_TYPE_MASK
 597 #define SOCK_TYPE_MASK 0xf
 598 #endif
 599
 600 static size_t syscall_arg__scnprintf_socket_type(char *bf, size_t size,
 601                                                       struct syscall_arg *arg)
 602 {
 603         size_t printed;
 604         int type = arg->val,
 605             flags = type & ~SOCK_TYPE_MASK;
 606
 607         type &= SOCK_TYPE_MASK;
 608         /*
 609          * Can't use a strarray, MIPS may override for ABI reasons.
 610          */
 611         switch (type) {
 612 #define P_SK_TYPE(n) case SOCK_##n: printed = scnprintf(bf, size, #n); break;
 613         P_SK_TYPE(STREAM);
 614         P_SK_TYPE(DGRAM);
 615         P_SK_TYPE(RAW);
 616         P_SK_TYPE(RDM);
 617         P_SK_TYPE(SEQPACKET);
 618         P_SK_TYPE(DCCP);
 619         P_SK_TYPE(PACKET);
 620 #undef P_SK_TYPE
 621         default:
 622                 printed = scnprintf(bf, size, "%#x", type);
 623         }
 624
 625 #define P_SK_FLAG(n) \
 626         if (flags & SOCK_##n) { \
 627                 printed += scnprintf(bf + printed, size - printed, "|%s", #n); \
 628                 flags &= ~SOCK_##n; \
 629         }
 630
 631         P_SK_FLAG(CLOEXEC);
 632         P_SK_FLAG(NONBLOCK);
 633 #undef P_SK_FLAG
 634
 635         if (flags)
 636                 printed += scnprintf(bf + printed, size - printed, "|%#x", flags);
 637
 638         return printed;
 639 }
 640
 641 #define SCA_SK_TYPE syscall_arg__scnprintf_socket_type
 642
 643 #ifndef MSG_PROBE
 644 #define MSG_PROBE            0x10
 645 #endif
 646 #ifndef MSG_WAITFORONE
 647 #define MSG_WAITFORONE  0x10000
 648 #endif
 649 #ifndef MSG_SENDPAGE_NOTLAST
 650 #define MSG_SENDPAGE_NOTLAST 0x20000
 651 #endif
 652 #ifndef MSG_FASTOPEN
 653 #define MSG_FASTOPEN         0x20000000
 654 #endif
 655
 656 static size_t syscall_arg__scnprintf_msg_flags(char *bf, size_t size,
 657                                                struct syscall_arg *arg)
 658 {
 659         int printed = 0, flags = arg->val;
 660
 661         if (flags == 0)
 662                 return scnprintf(bf, size, "NONE");
 663 #define P_MSG_FLAG(n) \
 664         if (flags & MSG_##n) { \
 665                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 666                 flags &= ~MSG_##n; \
 667         }
 668
 669         P_MSG_FLAG(OOB);
 670         P_MSG_FLAG(PEEK);
 671         P_MSG_FLAG(DONTROUTE);
 672         P_MSG_FLAG(TRYHARD);
 673         P_MSG_FLAG(CTRUNC);
 674         P_MSG_FLAG(PROBE);
 675         P_MSG_FLAG(TRUNC);
 676         P_MSG_FLAG(DONTWAIT);
 677         P_MSG_FLAG(EOR);
 678         P_MSG_FLAG(WAITALL);
 679         P_MSG_FLAG(FIN);
 680         P_MSG_FLAG(SYN);
 681         P_MSG_FLAG(CONFIRM);
 682         P_MSG_FLAG(RST);
 683         P_MSG_FLAG(ERRQUEUE);
 684         P_MSG_FLAG(NOSIGNAL);
 685         P_MSG_FLAG(MORE);
 686         P_MSG_FLAG(WAITFORONE);
 687         P_MSG_FLAG(SENDPAGE_NOTLAST);
 688         P_MSG_FLAG(FASTOPEN);
 689         P_MSG_FLAG(CMSG_CLOEXEC);
 690 #undef P_MSG_FLAG
 691
 692         if (flags)
 693                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 694
 695         return printed;
 696 }
 697
 698 #define SCA_MSG_FLAGS syscall_arg__scnprintf_msg_flags
 699
 700 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
 701                                                  struct syscall_arg *arg)
 702 {
 703         size_t printed = 0;
 704         int mode = arg->val;
 705
 706         if (mode == F_OK) /* 0 */
 707                 return scnprintf(bf, size, "F");
 708 #define P_MODE(n) \
 709         if (mode & n##_OK) { \
 710                 printed += scnprintf(bf + printed, size - printed, "%s", #n); \
 711                 mode &= ~n##_OK; \
 712         }
 713
 714         P_MODE(R);
 715         P_MODE(W);
 716         P_MODE(X);
 717 #undef P_MODE
 718
 719         if (mode)
 720                 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
 721
 722         return printed;
 723 }
 724
 725 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
 726
 727 static size_t syscall_arg__scnprintf_open_flags(char *bf, size_t size,
 728                                                struct syscall_arg *arg)
 729 {
 730         int printed = 0, flags = arg->val;
 731
 732         if (!(flags & O_CREAT))
 733                 arg->mask |= 1 << (arg->idx + 1); /* Mask the mode parm */
 734
 735         if (flags == 0)
 736                 return scnprintf(bf, size, "RDONLY");
 737 #define P_FLAG(n) \
 738         if (flags & O_##n) { \
 739                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 740                 flags &= ~O_##n; \
 741         }
 742
 743         P_FLAG(APPEND);
 744         P_FLAG(ASYNC);
 745         P_FLAG(CLOEXEC);
 746         P_FLAG(CREAT);
 747         P_FLAG(DIRECT);
 748         P_FLAG(DIRECTORY);
 749         P_FLAG(EXCL);
 750         P_FLAG(LARGEFILE);
 751         P_FLAG(NOATIME);
 752         P_FLAG(NOCTTY);
 753 #ifdef O_NONBLOCK
 754         P_FLAG(NONBLOCK);
 755 #elif O_NDELAY
 756         P_FLAG(NDELAY);
 757 #endif
 758 #ifdef O_PATH
 759         P_FLAG(PATH);
 760 #endif
 761         P_FLAG(RDWR);
 762 #ifdef O_DSYNC
 763         if ((flags & O_SYNC) == O_SYNC)
 764                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", "SYNC");
 765         else {
 766                 P_FLAG(DSYNC);
 767         }
 768 #else
 769         P_FLAG(SYNC);
 770 #endif
 771         P_FLAG(TRUNC);
 772         P_FLAG(WRONLY);
 773 #undef P_FLAG
 774
 775         if (flags)
 776                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 777
 778         return printed;
 779 }
 780
 781 #define SCA_OPEN_FLAGS syscall_arg__scnprintf_open_flags
 782
 783 static size_t syscall_arg__scnprintf_eventfd_flags(char *bf, size_t size,
 784                                                    struct syscall_arg *arg)
 785 {
 786         int printed = 0, flags = arg->val;
 787
 788         if (flags == 0)
 789                 return scnprintf(bf, size, "NONE");
 790 #define P_FLAG(n) \
 791         if (flags & EFD_##n) { \
 792                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 793                 flags &= ~EFD_##n; \
 794         }
 795
 796         P_FLAG(SEMAPHORE);
 797         P_FLAG(CLOEXEC);
 798         P_FLAG(NONBLOCK);
 799 #undef P_FLAG
 800
 801         if (flags)
 802                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 803
 804         return printed;
 805 }
 806
 807 #define SCA_EFD_FLAGS syscall_arg__scnprintf_eventfd_flags
 808
 809 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
 810                                                 struct syscall_arg *arg)
 811 {
 812         int printed = 0, flags = arg->val;
 813
 814 #define P_FLAG(n) \
 815         if (flags & O_##n) { \
 816                 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
 817                 flags &= ~O_##n; \
 818         }
 819
 820         P_FLAG(CLOEXEC);
 821         P_FLAG(NONBLOCK);
 822 #undef P_FLAG
 823
 824         if (flags)
 825                 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
 826
 827         return printed;
 828 }
 829
 830 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
 831
 832 static size_t syscall_arg__scnprintf_signum(char *bf, size_t size, struct syscall_arg *arg)
 833 {
 834         int sig = arg->val;
 835
 836         switch (sig) {
 837 #define P_SIGNUM(n) case SIG##n: return scnprintf(bf, size, #n)
 838         P_SIGNUM(HUP);
 839         P_SIGNUM(INT);
 840         P_SIGNUM(QUIT);
 841         P_SIGNUM(ILL);
 842         P_SIGNUM(TRAP);
 843         P_SIGNUM(ABRT);
 844         P_SIGNUM(BUS);
 845         P_SIGNUM(FPE);
 846         P_SIGNUM(KILL);
 847         P_SIGNUM(USR1);
 848         P_SIGNUM(SEGV);
 849         P_SIGNUM(USR2);
 850         P_SIGNUM(PIPE);
 851         P_SIGNUM(ALRM);
 852         P_SIGNUM(TERM);
 853         P_SIGNUM(CHLD);
 854         P_SIGNUM(CONT);
 855         P_SIGNUM(STOP);
 856         P_SIGNUM(TSTP);
 857         P_SIGNUM(TTIN);
 858         P_SIGNUM(TTOU);
 859         P_SIGNUM(URG);
 860         P_SIGNUM(XCPU);
 861         P_SIGNUM(XFSZ);
 862         P_SIGNUM(VTALRM);
 863         P_SIGNUM(PROF);
 864         P_SIGNUM(WINCH);
 865         P_SIGNUM(IO);
 866         P_SIGNUM(PWR);
 867         P_SIGNUM(SYS);
 868 #ifdef SIGEMT
 869         P_SIGNUM(EMT);
 870 #endif
 871 #ifdef SIGSTKFLT
 872         P_SIGNUM(STKFLT);
 873 #endif
 874 #ifdef SIGSWI
 875         P_SIGNUM(SWI);
 876 #endif
 877         default: break;
 878         }
 879
 880         return scnprintf(bf, size, "%#x", sig);
 881 }
 882
 883 #define SCA_SIGNUM syscall_arg__scnprintf_signum
 884
 885 #if defined(__i386__) || defined(__x86_64__)
 886 /*
 887  * FIXME: Make this available to all arches.
 888  */
 889 #define TCGETS          0x5401
 890
 891 static const char *tioctls[] = {
 892         "TCGETS", "TCSETS", "TCSETSW", "TCSETSF", "TCGETA", "TCSETA", "TCSETAW",
 893         "TCSETAF", "TCSBRK", "TCXONC", "TCFLSH", "TIOCEXCL", "TIOCNXCL",
 894         "TIOCSCTTY", "TIOCGPGRP", "TIOCSPGRP", "TIOCOUTQ", "TIOCSTI",
 895         "TIOCGWINSZ", "TIOCSWINSZ", "TIOCMGET", "TIOCMBIS", "TIOCMBIC",
 896         "TIOCMSET", "TIOCGSOFTCAR", "TIOCSSOFTCAR", "FIONREAD", "TIOCLINUX",
 897         "TIOCCONS", "TIOCGSERIAL", "TIOCSSERIAL", "TIOCPKT", "FIONBIO",
 898         "TIOCNOTTY", "TIOCSETD", "TIOCGETD", "TCSBRKP", [0x27] = "TIOCSBRK",
 899         "TIOCCBRK", "TIOCGSID", "TCGETS2", "TCSETS2", "TCSETSW2", "TCSETSF2",
 900         "TIOCGRS485", "TIOCSRS485", "TIOCGPTN", "TIOCSPTLCK",
 901         "TIOCGDEV||TCGETX", "TCSETX", "TCSETXF", "TCSETXW", "TIOCSIG",
 902         "TIOCVHANGUP", "TIOCGPKT", "TIOCGPTLCK", "TIOCGEXCL",
 903         [0x50] = "FIONCLEX", "FIOCLEX", "FIOASYNC", "TIOCSERCONFIG",
 904         "TIOCSERGWILD", "TIOCSERSWILD", "TIOCGLCKTRMIOS", "TIOCSLCKTRMIOS",
 905         "TIOCSERGSTRUCT", "TIOCSERGETLSR", "TIOCSERGETMULTI", "TIOCSERSETMULTI",
 906         "TIOCMIWAIT", "TIOCGICOUNT", [0x60] = "FIOQSIZE",
 907 };
 908
 909 static DEFINE_STRARRAY_OFFSET(tioctls, 0x5401);
 910 #endif /* defined(__i386__) || defined(__x86_64__) */
 911
 912 #define STRARRAY(arg, name, array) \
 913           .arg_scnprintf = { [arg] = SCA_STRARRAY, }, \
 914           .arg_parm      = { [arg] = &strarray__##array, }
 915
 916 static struct syscall_fmt {
 917         const char *name;
 918         const char *alias;
 919         size_t     (*arg_scnprintf[6])(char *bf, size_t size, struct syscall_arg *arg);
 920         void       *arg_parm[6];
 921         bool       errmsg;
 922         bool       timeout;
 923         bool       hexret;
 924 } syscall_fmts[] = {
 925         { .name     = "access",     .errmsg = true,
 926           .arg_scnprintf = { [1] = SCA_ACCMODE, /* mode */ }, },
 927         { .name     = "arch_prctl", .errmsg = true, .alias = "prctl", },
 928         { .name     = "brk",        .hexret = true,
 929           .arg_scnprintf = { [0] = SCA_HEX, /* brk */ }, },
 930         { .name     = "clock_gettime",  .errmsg = true, STRARRAY(0, clk_id, clockid), },
 931         { .name     = "close",      .errmsg = true,
 932           .arg_scnprintf = { [0] = SCA_CLOSE_FD, /* fd */ }, },
 933         { .name     = "connect",    .errmsg = true, },
 934         { .name     = "dup",        .errmsg = true,
 935           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 936         { .name     = "dup2",       .errmsg = true,
 937           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 938         { .name     = "dup3",       .errmsg = true,
 939           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 940         { .name     = "epoll_ctl",  .errmsg = true, STRARRAY(1, op, epoll_ctl_ops), },
 941         { .name     = "eventfd2",   .errmsg = true,
 942           .arg_scnprintf = { [1] = SCA_EFD_FLAGS, /* flags */ }, },
 943         { .name     = "faccessat",  .errmsg = true,
 944           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
 945         { .name     = "fadvise64",  .errmsg = true,
 946           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 947         { .name     = "fallocate",  .errmsg = true,
 948           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 949         { .name     = "fchdir",     .errmsg = true,
 950           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 951         { .name     = "fchmod",     .errmsg = true,
 952           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 953         { .name     = "fchmodat",   .errmsg = true,
 954           .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
 955         { .name     = "fchown",     .errmsg = true,
 956           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 957         { .name     = "fchownat",   .errmsg = true,
 958           .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
 959         { .name     = "fcntl",      .errmsg = true,
 960           .arg_scnprintf = { [0] = SCA_FD, /* fd */
 961                              [1] = SCA_STRARRAY, /* cmd */ },
 962           .arg_parm      = { [1] = &strarray__fcntl_cmds, /* cmd */ }, },
 963         { .name     = "fdatasync",  .errmsg = true,
 964           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 965         { .name     = "flock",      .errmsg = true,
 966           .arg_scnprintf = { [0] = SCA_FD, /* fd */
 967                              [1] = SCA_FLOCK, /* cmd */ }, },
 968         { .name     = "fsetxattr",  .errmsg = true,
 969           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 970         { .name     = "fstat",      .errmsg = true, .alias = "newfstat",
 971           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 972         { .name     = "fstatat",    .errmsg = true, .alias = "newfstatat",
 973           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
 974         { .name     = "fstatfs",    .errmsg = true,
 975           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 976         { .name     = "fsync",    .errmsg = true,
 977           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 978         { .name     = "ftruncate", .errmsg = true,
 979           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 980         { .name     = "futex",      .errmsg = true,
 981           .arg_scnprintf = { [1] = SCA_FUTEX_OP, /* op */ }, },
 982         { .name     = "futimesat", .errmsg = true,
 983           .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
 984         { .name     = "getdents",   .errmsg = true,
 985           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 986         { .name     = "getdents64", .errmsg = true,
 987           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
 988         { .name     = "getitimer",  .errmsg = true, STRARRAY(0, which, itimers), },
 989         { .name     = "getrlimit",  .errmsg = true, STRARRAY(0, resource, rlimit_resources), },
 990         { .name     = "ioctl",      .errmsg = true,
 991           .arg_scnprintf = { [0] = SCA_FD, /* fd */
 992 #if defined(__i386__) || defined(__x86_64__)
 993 /*
 994  * FIXME: Make this available to all arches.
 995  */
 996                              [1] = SCA_STRHEXARRAY, /* cmd */
 997                              [2] = SCA_HEX, /* arg */ },
 998           .arg_parm      = { [1] = &strarray__tioctls, /* cmd */ }, },
 999 #else
1000                              [2] = SCA_HEX, /* arg */ }, },
1001 #endif
1002         { .name     = "kill",       .errmsg = true,
1003           .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
1004         { .name     = "linkat",     .errmsg = true,
1005           .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
1006         { .name     = "lseek",      .errmsg = true,
1007           .arg_scnprintf = { [0] = SCA_FD, /* fd */
1008                              [2] = SCA_STRARRAY, /* whence */ },
1009           .arg_parm      = { [2] = &strarray__whences, /* whence */ }, },
1010         { .name     = "lstat",      .errmsg = true, .alias = "newlstat", },
1011         { .name     = "madvise",    .errmsg = true,
1012           .arg_scnprintf = { [0] = SCA_HEX,      /* start */
1013                              [2] = SCA_MADV_BHV, /* behavior */ }, },
1014         { .name     = "mkdirat",    .errmsg = true,
1015           .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
1016         { .name     = "mknodat",    .errmsg = true,
1017           .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
1018         { .name     = "mlock",      .errmsg = true,
1019           .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
1020         { .name     = "mlockall",   .errmsg = true,
1021           .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
1022         { .name     = "mmap",       .hexret = true,
1023           .arg_scnprintf = { [0] = SCA_HEX,       /* addr */
1024                              [2] = SCA_MMAP_PROT, /* prot */
1025                              [3] = SCA_MMAP_FLAGS, /* flags */
1026                              [4] = SCA_FD,        /* fd */ }, },
1027         { .name     = "mprotect",   .errmsg = true,
1028           .arg_scnprintf = { [0] = SCA_HEX, /* start */
1029                              [2] = SCA_MMAP_PROT, /* prot */ }, },
1030         { .name     = "mremap",     .hexret = true,
1031           .arg_scnprintf = { [0] = SCA_HEX, /* addr */
1032                              [3] = SCA_MREMAP_FLAGS, /* flags */
1033                              [4] = SCA_HEX, /* new_addr */ }, },
1034         { .name     = "munlock",    .errmsg = true,
1035           .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
1036         { .name     = "munmap",     .errmsg = true,
1037           .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
1038         { .name     = "name_to_handle_at", .errmsg = true,
1039           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1040         { .name     = "newfstatat", .errmsg = true,
1041           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1042         { .name     = "open",       .errmsg = true,
1043           .arg_scnprintf = { [1] = SCA_OPEN_FLAGS, /* flags */ }, },
1044         { .name     = "open_by_handle_at", .errmsg = true,
1045           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1046                              [2] = SCA_OPEN_FLAGS, /* flags */ }, },
1047         { .name     = "openat",     .errmsg = true,
1048           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1049                              [2] = SCA_OPEN_FLAGS, /* flags */ }, },
1050         { .name     = "pipe2",      .errmsg = true,
1051           .arg_scnprintf = { [1] = SCA_PIPE_FLAGS, /* flags */ }, },
1052         { .name     = "poll",       .errmsg = true, .timeout = true, },
1053         { .name     = "ppoll",      .errmsg = true, .timeout = true, },
1054         { .name     = "pread",      .errmsg = true, .alias = "pread64",
1055           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1056         { .name     = "preadv",     .errmsg = true, .alias = "pread",
1057           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1058         { .name     = "prlimit64",  .errmsg = true, STRARRAY(1, resource, rlimit_resources), },
1059         { .name     = "pwrite",     .errmsg = true, .alias = "pwrite64",
1060           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1061         { .name     = "pwritev",    .errmsg = true,
1062           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1063         { .name     = "read",       .errmsg = true,
1064           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1065         { .name     = "readlinkat", .errmsg = true,
1066           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1067         { .name     = "readv",      .errmsg = true,
1068           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1069         { .name     = "recvfrom",   .errmsg = true,
1070           .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1071         { .name     = "recvmmsg",   .errmsg = true,
1072           .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1073         { .name     = "recvmsg",    .errmsg = true,
1074           .arg_scnprintf = { [2] = SCA_MSG_FLAGS, /* flags */ }, },
1075         { .name     = "renameat",   .errmsg = true,
1076           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1077         { .name     = "rt_sigaction", .errmsg = true,
1078           .arg_scnprintf = { [0] = SCA_SIGNUM, /* sig */ }, },
1079         { .name     = "rt_sigprocmask",  .errmsg = true, STRARRAY(0, how, sighow), },
1080         { .name     = "rt_sigqueueinfo", .errmsg = true,
1081           .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
1082         { .name     = "rt_tgsigqueueinfo", .errmsg = true,
1083           .arg_scnprintf = { [2] = SCA_SIGNUM, /* sig */ }, },
1084         { .name     = "select",     .errmsg = true, .timeout = true, },
1085         { .name     = "sendmmsg",    .errmsg = true,
1086           .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1087         { .name     = "sendmsg",    .errmsg = true,
1088           .arg_scnprintf = { [2] = SCA_MSG_FLAGS, /* flags */ }, },
1089         { .name     = "sendto",     .errmsg = true,
1090           .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1091         { .name     = "setitimer",  .errmsg = true, STRARRAY(0, which, itimers), },
1092         { .name     = "setrlimit",  .errmsg = true, STRARRAY(0, resource, rlimit_resources), },
1093         { .name     = "shutdown",   .errmsg = true,
1094           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1095         { .name     = "socket",     .errmsg = true,
1096           .arg_scnprintf = { [0] = SCA_STRARRAY, /* family */
1097                              [1] = SCA_SK_TYPE, /* type */ },
1098           .arg_parm      = { [0] = &strarray__socket_families, /* family */ }, },
1099         { .name     = "socketpair", .errmsg = true,
1100           .arg_scnprintf = { [0] = SCA_STRARRAY, /* family */
1101                              [1] = SCA_SK_TYPE, /* type */ },
1102           .arg_parm      = { [0] = &strarray__socket_families, /* family */ }, },
1103         { .name     = "stat",       .errmsg = true, .alias = "newstat", },
1104         { .name     = "symlinkat",  .errmsg = true,
1105           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1106         { .name     = "tgkill",     .errmsg = true,
1107           .arg_scnprintf = { [2] = SCA_SIGNUM, /* sig */ }, },
1108         { .name     = "tkill",      .errmsg = true,
1109           .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
1110         { .name     = "uname",      .errmsg = true, .alias = "newuname", },
1111         { .name     = "unlinkat",   .errmsg = true,
1112           .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1113         { .name     = "utimensat",  .errmsg = true,
1114           .arg_scnprintf = { [0] = SCA_FDAT, /* dirfd */ }, },
1115         { .name     = "write",      .errmsg = true,
1116           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1117         { .name     = "writev",     .errmsg = true,
1118           .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
1119 };
1120
1121 static int syscall_fmt__cmp(const void *name, const void *fmtp)
1122 {
1123         const struct syscall_fmt *fmt = fmtp;
1124         return strcmp(name, fmt->name);
1125 }
1126
1127 static struct syscall_fmt *syscall_fmt__find(const char *name)
1128 {
1129         const int nmemb = ARRAY_SIZE(syscall_fmts);
1130         return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1131 }
1132
1133 struct syscall {
1134         struct event_format *tp_format;
1135         const char          *name;
1136         bool                filtered;
1137         bool                is_exit;
1138         struct syscall_fmt  *fmt;
1139         size_t              (**arg_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1140         void                **arg_parm;
1141 };
1142
1143 static size_t fprintf_duration(unsigned long t, FILE *fp)
1144 {
1145         double duration = (double)t / NSEC_PER_MSEC;
1146         size_t printed = fprintf(fp, "(");
1147
1148         if (duration >= 1.0)
1149                 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1150         else if (duration >= 0.01)
1151                 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1152         else
1153                 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1154         return printed + fprintf(fp, "): ");
1155 }
1156
1157 struct thread_trace {
1158         u64               entry_time;
1159         u64               exit_time;
1160         bool              entry_pending;
1161         unsigned long     nr_events;
1162         unsigned long     pfmaj, pfmin;
1163         char              *entry_str;
1164         double            runtime_ms;
1165         struct {
1166                 int       max;
1167                 char      **table;
1168         } paths;
1169
1170         struct intlist *syscall_stats;
1171 };
1172
1173 static struct thread_trace *thread_trace__new(void)
1174 {
1175         struct thread_trace *ttrace =  zalloc(sizeof(struct thread_trace));
1176
1177         if (ttrace)
1178                 ttrace->paths.max = -1;
1179
1180         ttrace->syscall_stats = intlist__new(NULL);
1181
1182         return ttrace;
1183 }
1184
1185 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1186 {
1187         struct thread_trace *ttrace;
1188
1189         if (thread == NULL)
1190                 goto fail;
1191
1192         if (thread__priv(thread) == NULL)
1193                 thread__set_priv(thread, thread_trace__new());
1194
1195         if (thread__priv(thread) == NULL)
1196                 goto fail;
1197
1198         ttrace = thread__priv(thread);
1199         ++ttrace->nr_events;
1200
1201         return ttrace;
1202 fail:
1203         color_fprintf(fp, PERF_COLOR_RED,
1204                       "WARNING: not enough memory, dropping samples!\n");
1205         return NULL;
1206 }
1207
1208 #define TRACE_PFMAJ             (1 << 0)
1209 #define TRACE_PFMIN             (1 << 1)
1210
1211 struct trace {
1212         struct perf_tool        tool;
1213         struct {
1214                 int             machine;
1215                 int             open_id;
1216         }                       audit;
1217         struct {
1218                 int             max;
1219                 struct syscall  *table;
1220         } syscalls;
1221         struct record_opts      opts;
1222         struct perf_evlist      *evlist;
1223         struct machine          *host;
1224         struct thread           *current;
1225         u64                     base_time;
1226         FILE                    *output;
1227         unsigned long           nr_events;
1228         struct strlist          *ev_qualifier;
1229         const char              *last_vfs_getname;
1230         struct intlist          *tid_list;
1231         struct intlist          *pid_list;
1232         double                  duration_filter;
1233         double                  runtime_ms;
1234         struct {
1235                 u64             vfs_getname,
1236                                 proc_getname;
1237         } stats;
1238         bool                    not_ev_qualifier;
1239         bool                    live;
1240         bool                    full_time;
1241         bool                    sched;
1242         bool                    multiple_threads;
1243         bool                    summary;
1244         bool                    summary_only;
1245         bool                    show_comm;
1246         bool                    show_tool_stats;
1247         bool                    trace_syscalls;
1248         int                     trace_pgfaults;
1249 };
1250
1251 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1252 {
1253         struct thread_trace *ttrace = thread__priv(thread);
1254
1255         if (fd > ttrace->paths.max) {
1256                 char **npath = realloc(ttrace->paths.table, (fd + 1) * sizeof(char *));
1257
1258                 if (npath == NULL)
1259                         return -1;
1260
1261                 if (ttrace->paths.max != -1) {
1262                         memset(npath + ttrace->paths.max + 1, 0,
1263                                (fd - ttrace->paths.max) * sizeof(char *));
1264                 } else {
1265                         memset(npath, 0, (fd + 1) * sizeof(char *));
1266                 }
1267
1268                 ttrace->paths.table = npath;
1269                 ttrace->paths.max   = fd;
1270         }
1271
1272         ttrace->paths.table[fd] = strdup(pathname);
1273
1274         return ttrace->paths.table[fd] != NULL ? 0 : -1;
1275 }
1276
1277 static int thread__read_fd_path(struct thread *thread, int fd)
1278 {
1279         char linkname[PATH_MAX], pathname[PATH_MAX];
1280         struct stat st;
1281         int ret;
1282
1283         if (thread->pid_ == thread->tid) {
1284                 scnprintf(linkname, sizeof(linkname),
1285                           "/proc/%d/fd/%d", thread->pid_, fd);
1286         } else {
1287                 scnprintf(linkname, sizeof(linkname),
1288                           "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1289         }
1290
1291         if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1292                 return -1;
1293
1294         ret = readlink(linkname, pathname, sizeof(pathname));
1295
1296         if (ret < 0 || ret > st.st_size)
1297                 return -1;
1298
1299         pathname[ret] = '\0';
1300         return trace__set_fd_pathname(thread, fd, pathname);
1301 }
1302
1303 static const char *thread__fd_path(struct thread *thread, int fd,
1304                                    struct trace *trace)
1305 {
1306         struct thread_trace *ttrace = thread__priv(thread);
1307
1308         if (ttrace == NULL)
1309                 return NULL;
1310
1311         if (fd < 0)
1312                 return NULL;
1313
1314         if ((fd > ttrace->paths.max || ttrace->paths.table[fd] == NULL)) {
1315                 if (!trace->live)
1316                         return NULL;
1317                 ++trace->stats.proc_getname;
1318                 if (thread__read_fd_path(thread, fd))
1319                         return NULL;
1320         }
1321
1322         return ttrace->paths.table[fd];
1323 }
1324
1325 static size_t syscall_arg__scnprintf_fd(char *bf, size_t size,
1326                                         struct syscall_arg *arg)
1327 {
1328         int fd = arg->val;
1329         size_t printed = scnprintf(bf, size, "%d", fd);
1330         const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1331
1332         if (path)
1333                 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1334
1335         return printed;
1336 }
1337
1338 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1339                                               struct syscall_arg *arg)
1340 {
1341         int fd = arg->val;
1342         size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1343         struct thread_trace *ttrace = thread__priv(arg->thread);
1344
1345         if (ttrace && fd >= 0 && fd <= ttrace->paths.max)
1346                 zfree(&ttrace->paths.table[fd]);
1347
1348         return printed;
1349 }
1350
1351 static bool trace__filter_duration(struct trace *trace, double t)
1352 {
1353         return t < (trace->duration_filter * NSEC_PER_MSEC);
1354 }
1355
1356 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1357 {
1358         double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1359
1360         return fprintf(fp, "%10.3f ", ts);
1361 }
1362
1363 static bool done = false;
1364 static bool interrupted = false;
1365
1366 static void sig_handler(int sig)
1367 {
1368         done = true;
1369         interrupted = sig == SIGINT;
1370 }
1371
1372 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1373                                         u64 duration, u64 tstamp, FILE *fp)
1374 {
1375         size_t printed = trace__fprintf_tstamp(trace, tstamp, fp);
1376         printed += fprintf_duration(duration, fp);
1377
1378         if (trace->multiple_threads) {
1379                 if (trace->show_comm)
1380                         printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1381                 printed += fprintf(fp, "%d ", thread->tid);
1382         }
1383
1384         return printed;
1385 }
1386
1387 static int trace__process_event(struct trace *trace, struct machine *machine,
1388                                 union perf_event *event, struct perf_sample *sample)
1389 {
1390         int ret = 0;
1391
1392         switch (event->header.type) {
1393         case PERF_RECORD_LOST:
1394                 color_fprintf(trace->output, PERF_COLOR_RED,
1395                               "LOST %" PRIu64 " events!\n", event->lost.lost);
1396                 ret = machine__process_lost_event(machine, event, sample);
1397         default:
1398                 ret = machine__process_event(machine, event, sample);
1399                 break;
1400         }
1401
1402         return ret;
1403 }
1404
1405 static int trace__tool_process(struct perf_tool *tool,
1406                                union perf_event *event,
1407                                struct perf_sample *sample,
1408                                struct machine *machine)
1409 {
1410         struct trace *trace = container_of(tool, struct trace, tool);
1411         return trace__process_event(trace, machine, event, sample);
1412 }
1413
1414 static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
1415 {
1416         int err = symbol__init(NULL);
1417
1418         if (err)
1419                 return err;
1420
1421         trace->host = machine__new_host();
1422         if (trace->host == NULL)
1423                 return -ENOMEM;
1424
1425         err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1426                                             evlist->threads, trace__tool_process, false);
1427         if (err)
1428                 symbol__exit();
1429
1430         return err;
1431 }
1432
1433 static int syscall__set_arg_fmts(struct syscall *sc)
1434 {
1435         struct format_field *field;
1436         int idx = 0;
1437
1438         sc->arg_scnprintf = calloc(sc->tp_format->format.nr_fields - 1, sizeof(void *));
1439         if (sc->arg_scnprintf == NULL)
1440                 return -1;
1441
1442         if (sc->fmt)
1443                 sc->arg_parm = sc->fmt->arg_parm;
1444
1445         for (field = sc->tp_format->format.fields->next; field; field = field->next) {
1446                 if (sc->fmt && sc->fmt->arg_scnprintf[idx])
1447                         sc->arg_scnprintf[idx] = sc->fmt->arg_scnprintf[idx];
1448                 else if (field->flags & FIELD_IS_POINTER)
1449                         sc->arg_scnprintf[idx] = syscall_arg__scnprintf_hex;
1450                 ++idx;
1451         }
1452
1453         return 0;
1454 }
1455
1456 static int trace__read_syscall_info(struct trace *trace, int id)
1457 {
1458         char tp_name[128];
1459         struct syscall *sc;
1460         const char *name = audit_syscall_to_name(id, trace->audit.machine);
1461
1462         if (name == NULL)
1463                 return -1;
1464
1465         if (id > trace->syscalls.max) {
1466                 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1467
1468                 if (nsyscalls == NULL)
1469                         return -1;
1470
1471                 if (trace->syscalls.max != -1) {
1472                         memset(nsyscalls + trace->syscalls.max + 1, 0,
1473                                (id - trace->syscalls.max) * sizeof(*sc));
1474                 } else {
1475                         memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
1476                 }
1477
1478                 trace->syscalls.table = nsyscalls;
1479                 trace->syscalls.max   = id;
1480         }
1481
1482         sc = trace->syscalls.table + id;
1483         sc->name = name;
1484
1485         if (trace->ev_qualifier) {
1486                 bool in = strlist__find(trace->ev_qualifier, name) != NULL;
1487
1488                 if (!(in ^ trace->not_ev_qualifier)) {
1489                         sc->filtered = true;
1490                         /*
1491                          * No need to do read tracepoint information since this will be
1492                          * filtered out.
1493                          */
1494                         return 0;
1495                 }
1496         }
1497
1498         sc->fmt  = syscall_fmt__find(sc->name);
1499
1500         snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1501         sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1502
1503         if (sc->tp_format == NULL && sc->fmt && sc->fmt->alias) {
1504                 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1505                 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1506         }
1507
1508         if (sc->tp_format == NULL)
1509                 return -1;
1510
1511         sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1512
1513         return syscall__set_arg_fmts(sc);
1514 }
1515
1516 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1517                                       unsigned long *args, struct trace *trace,
1518                                       struct thread *thread)
1519 {
1520         size_t printed = 0;
1521
1522         if (sc->tp_format != NULL) {
1523                 struct format_field *field;
1524                 u8 bit = 1;
1525                 struct syscall_arg arg = {
1526                         .idx    = 0,
1527                         .mask   = 0,
1528                         .trace  = trace,
1529                         .thread = thread,
1530                 };
1531
1532                 for (field = sc->tp_format->format.fields->next; field;
1533                      field = field->next, ++arg.idx, bit <<= 1) {
1534                         if (arg.mask & bit)
1535                                 continue;
1536                         /*
1537                          * Suppress this argument if its value is zero and
1538                          * and we don't have a string associated in an
1539                          * strarray for it.
1540                          */
1541                         if (args[arg.idx] == 0 &&
1542                             !(sc->arg_scnprintf &&
1543                               sc->arg_scnprintf[arg.idx] == SCA_STRARRAY &&
1544                               sc->arg_parm[arg.idx]))
1545                                 continue;
1546
1547                         printed += scnprintf(bf + printed, size - printed,
1548                                              "%s%s: ", printed ? ", " : "", field->name);
1549                         if (sc->arg_scnprintf && sc->arg_scnprintf[arg.idx]) {
1550                                 arg.val = args[arg.idx];
1551                                 if (sc->arg_parm)
1552                                         arg.parm = sc->arg_parm[arg.idx];
1553                                 printed += sc->arg_scnprintf[arg.idx](bf + printed,
1554                                                                       size - printed, &arg);
1555                         } else {
1556                                 printed += scnprintf(bf + printed, size - printed,
1557                                                      "%ld", args[arg.idx]);
1558                         }
1559                 }
1560         } else {
1561                 int i = 0;
1562
1563                 while (i < 6) {
1564                         printed += scnprintf(bf + printed, size - printed,
1565                                              "%sarg%d: %ld",
1566                                              printed ? ", " : "", i, args[i]);
1567                         ++i;
1568                 }
1569         }
1570
1571         return printed;
1572 }
1573
1574 typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
1575                                   union perf_event *event,
1576                                   struct perf_sample *sample);
1577
1578 static struct syscall *trace__syscall_info(struct trace *trace,
1579                                            struct perf_evsel *evsel, int id)
1580 {
1581
1582         if (id < 0) {
1583
1584                 /*
1585                  * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1586                  * before that, leaving at a higher verbosity level till that is
1587                  * explained. Reproduced with plain ftrace with:
1588                  *
1589                  * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1590                  * grep "NR -1 " /t/trace_pipe
1591                  *
1592                  * After generating some load on the machine.
1593                  */
1594                 if (verbose > 1) {
1595                         static u64 n;
1596                         fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1597                                 id, perf_evsel__name(evsel), ++n);
1598                 }
1599                 return NULL;
1600         }
1601
1602         if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
1603             trace__read_syscall_info(trace, id))
1604                 goto out_cant_read;
1605
1606         if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
1607                 goto out_cant_read;
1608
1609         return &trace->syscalls.table[id];
1610
1611 out_cant_read:
1612         if (verbose) {
1613                 fprintf(trace->output, "Problems reading syscall %d", id);
1614                 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
1615                         fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1616                 fputs(" information\n", trace->output);
1617         }
1618         return NULL;
1619 }
1620
1621 static void thread__update_stats(struct thread_trace *ttrace,
1622                                  int id, struct perf_sample *sample)
1623 {
1624         struct int_node *inode;
1625         struct stats *stats;
1626         u64 duration = 0;
1627
1628         inode = intlist__findnew(ttrace->syscall_stats, id);
1629         if (inode == NULL)
1630                 return;
1631
1632         stats = inode->priv;
1633         if (stats == NULL) {
1634                 stats = malloc(sizeof(struct stats));
1635                 if (stats == NULL)
1636                         return;
1637                 init_stats(stats);
1638                 inode->priv = stats;
1639         }
1640
1641         if (ttrace->entry_time && sample->time > ttrace->entry_time)
1642                 duration = sample->time - ttrace->entry_time;
1643
1644         update_stats(stats, duration);
1645 }
1646
1647 static int trace__printf_interrupted_entry(struct trace *trace, struct perf_sample *sample)
1648 {
1649         struct thread_trace *ttrace;
1650         u64 duration;
1651         size_t printed;
1652
1653         if (trace->current == NULL)
1654                 return 0;
1655
1656         ttrace = thread__priv(trace->current);
1657
1658         if (!ttrace->entry_pending)
1659                 return 0;
1660
1661         duration = sample->time - ttrace->entry_time;
1662
1663         printed  = trace__fprintf_entry_head(trace, trace->current, duration, sample->time, trace->output);
1664         printed += fprintf(trace->output, "%-70s) ...\n", ttrace->entry_str);
1665         ttrace->entry_pending = false;
1666
1667         return printed;
1668 }
1669
1670 static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
1671                             union perf_event *event __maybe_unused,
1672                             struct perf_sample *sample)
1673 {
1674         char *msg;
1675         void *args;
1676         size_t printed = 0;
1677         struct thread *thread;
1678         int id = perf_evsel__sc_tp_uint(evsel, id, sample);
1679         struct syscall *sc = trace__syscall_info(trace, evsel, id);
1680         struct thread_trace *ttrace;
1681
1682         if (sc == NULL)
1683                 return -1;
1684
1685         if (sc->filtered)
1686                 return 0;
1687
1688         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1689         ttrace = thread__trace(thread, trace->output);
1690         if (ttrace == NULL)
1691                 return -1;
1692
1693         args = perf_evsel__sc_tp_ptr(evsel, args, sample);
1694
1695         if (ttrace->entry_str == NULL) {
1696                 ttrace->entry_str = malloc(1024);
1697                 if (!ttrace->entry_str)
1698                         return -1;
1699         }
1700
1701         printed += trace__printf_interrupted_entry(trace, sample);
1702
1703         ttrace->entry_time = sample->time;
1704         msg = ttrace->entry_str;
1705         printed += scnprintf(msg + printed, 1024 - printed, "%s(", sc->name);
1706
1707         printed += syscall__scnprintf_args(sc, msg + printed, 1024 - printed,
1708                                            args, trace, thread);
1709
1710         if (sc->is_exit) {
1711                 if (!trace->duration_filter && !trace->summary_only) {
1712                         trace__fprintf_entry_head(trace, thread, 1, sample->time, trace->output);
1713                         fprintf(trace->output, "%-70s\n", ttrace->entry_str);
1714                 }
1715         } else
1716                 ttrace->entry_pending = true;
1717
1718         trace->current = thread;
1719
1720         return 0;
1721 }
1722
1723 static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
1724                            union perf_event *event __maybe_unused,
1725                            struct perf_sample *sample)
1726 {
1727         long ret;
1728         u64 duration = 0;
1729         struct thread *thread;
1730         int id = perf_evsel__sc_tp_uint(evsel, id, sample);
1731         struct syscall *sc = trace__syscall_info(trace, evsel, id);
1732         struct thread_trace *ttrace;
1733
1734         if (sc == NULL)
1735                 return -1;
1736
1737         if (sc->filtered)
1738                 return 0;
1739
1740         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1741         ttrace = thread__trace(thread, trace->output);
1742         if (ttrace == NULL)
1743                 return -1;
1744
1745         if (trace->summary)
1746                 thread__update_stats(ttrace, id, sample);
1747
1748         ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
1749
1750         if (id == trace->audit.open_id && ret >= 0 && trace->last_vfs_getname) {
1751                 trace__set_fd_pathname(thread, ret, trace->last_vfs_getname);
1752                 trace->last_vfs_getname = NULL;
1753                 ++trace->stats.vfs_getname;
1754         }
1755
1756         ttrace->exit_time = sample->time;
1757
1758         if (ttrace->entry_time) {
1759                 duration = sample->time - ttrace->entry_time;
1760                 if (trace__filter_duration(trace, duration))
1761                         goto out;
1762         } else if (trace->duration_filter)
1763                 goto out;
1764
1765         if (trace->summary_only)
1766                 goto out;
1767
1768         trace__fprintf_entry_head(trace, thread, duration, sample->time, trace->output);
1769
1770         if (ttrace->entry_pending) {
1771                 fprintf(trace->output, "%-70s", ttrace->entry_str);
1772         } else {
1773                 fprintf(trace->output, " ... [");
1774                 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
1775                 fprintf(trace->output, "]: %s()", sc->name);
1776         }
1777
1778         if (sc->fmt == NULL) {
1779 signed_print:
1780                 fprintf(trace->output, ") = %ld", ret);
1781         } else if (ret < 0 && sc->fmt->errmsg) {
1782                 char bf[STRERR_BUFSIZE];
1783                 const char *emsg = strerror_r(-ret, bf, sizeof(bf)),
1784                            *e = audit_errno_to_name(-ret);
1785
1786                 fprintf(trace->output, ") = -1 %s %s", e, emsg);
1787         } else if (ret == 0 && sc->fmt->timeout)
1788                 fprintf(trace->output, ") = 0 Timeout");
1789         else if (sc->fmt->hexret)
1790                 fprintf(trace->output, ") = %#lx", ret);
1791         else
1792                 goto signed_print;
1793
1794         fputc('\n', trace->output);
1795 out:
1796         ttrace->entry_pending = false;
1797
1798         return 0;
1799 }
1800
1801 static int trace__vfs_getname(struct trace *trace, struct perf_evsel *evsel,
1802                               union perf_event *event __maybe_unused,
1803                               struct perf_sample *sample)
1804 {
1805         trace->last_vfs_getname = perf_evsel__rawptr(evsel, sample, "pathname");
1806         return 0;
1807 }
1808
1809 static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
1810                                      union perf_event *event __maybe_unused,
1811                                      struct perf_sample *sample)
1812 {
1813         u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
1814         double runtime_ms = (double)runtime / NSEC_PER_MSEC;
1815         struct thread *thread = machine__findnew_thread(trace->host,
1816                                                         sample->pid,
1817                                                         sample->tid);
1818         struct thread_trace *ttrace = thread__trace(thread, trace->output);
1819
1820         if (ttrace == NULL)
1821                 goto out_dump;
1822
1823         ttrace->runtime_ms += runtime_ms;
1824         trace->runtime_ms += runtime_ms;
1825         return 0;
1826
1827 out_dump:
1828         fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
1829                evsel->name,
1830                perf_evsel__strval(evsel, sample, "comm"),
1831                (pid_t)perf_evsel__intval(evsel, sample, "pid"),
1832                runtime,
1833                perf_evsel__intval(evsel, sample, "vruntime"));
1834         return 0;
1835 }
1836
1837 static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
1838                                 union perf_event *event __maybe_unused,
1839                                 struct perf_sample *sample)
1840 {
1841         trace__printf_interrupted_entry(trace, sample);
1842         trace__fprintf_tstamp(trace, sample->time, trace->output);
1843         fprintf(trace->output, "(%9.9s): %s:", " ", evsel->name);
1844
1845         if (evsel->tp_format) {
1846                 event_format__fprintf(evsel->tp_format, sample->cpu,
1847                                       sample->raw_data, sample->raw_size,
1848                                       trace->output);
1849         }
1850
1851         fprintf(trace->output, ")\n");
1852         return 0;
1853 }
1854
1855 static void print_location(FILE *f, struct perf_sample *sample,
1856                            struct addr_location *al,
1857                            bool print_dso, bool print_sym)
1858 {
1859
1860         if ((verbose || print_dso) && al->map)
1861                 fprintf(f, "%s@", al->map->dso->long_name);
1862
1863         if ((verbose || print_sym) && al->sym)
1864                 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
1865                         al->addr - al->sym->start);
1866         else if (al->map)
1867                 fprintf(f, "0x%" PRIx64, al->addr);
1868         else
1869                 fprintf(f, "0x%" PRIx64, sample->addr);
1870 }
1871
1872 static int trace__pgfault(struct trace *trace,
1873                           struct perf_evsel *evsel,
1874                           union perf_event *event,
1875                           struct perf_sample *sample)
1876 {
1877         struct thread *thread;
1878         u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1879         struct addr_location al;
1880         char map_type = 'd';
1881         struct thread_trace *ttrace;
1882
1883         thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1884         ttrace = thread__trace(thread, trace->output);
1885         if (ttrace == NULL)
1886                 return -1;
1887
1888         if (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
1889                 ttrace->pfmaj++;
1890         else
1891                 ttrace->pfmin++;
1892
1893         if (trace->summary_only)
1894                 return 0;
1895
1896         thread__find_addr_location(thread, cpumode, MAP__FUNCTION,
1897                               sample->ip, &al);
1898
1899         trace__fprintf_entry_head(trace, thread, 0, sample->time, trace->output);
1900
1901         fprintf(trace->output, "%sfault [",
1902                 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
1903                 "maj" : "min");
1904
1905         print_location(trace->output, sample, &al, false, true);
1906
1907         fprintf(trace->output, "] => ");
1908
1909         thread__find_addr_location(thread, cpumode, MAP__VARIABLE,
1910                                    sample->addr, &al);
1911
1912         if (!al.map) {
1913                 thread__find_addr_location(thread, cpumode,
1914                                            MAP__FUNCTION, sample->addr, &al);
1915
1916                 if (al.map)
1917                         map_type = 'x';
1918                 else
1919                         map_type = '?';
1920         }
1921
1922         print_location(trace->output, sample, &al, true, false);
1923
1924         fprintf(trace->output, " (%c%c)\n", map_type, al.level);
1925
1926         return 0;
1927 }
1928
1929 static bool skip_sample(struct trace *trace, struct perf_sample *sample)
1930 {
1931         if ((trace->pid_list && intlist__find(trace->pid_list, sample->pid)) ||
1932             (trace->tid_list && intlist__find(trace->tid_list, sample->tid)))
1933                 return false;
1934
1935         if (trace->pid_list || trace->tid_list)
1936                 return true;
1937
1938         return false;
1939 }
1940
1941 static int trace__process_sample(struct perf_tool *tool,
1942                                  union perf_event *event,
1943                                  struct perf_sample *sample,
1944                                  struct perf_evsel *evsel,
1945                                  struct machine *machine __maybe_unused)
1946 {
1947         struct trace *trace = container_of(tool, struct trace, tool);
1948         int err = 0;
1949
1950         tracepoint_handler handler = evsel->handler;
1951
1952         if (skip_sample(trace, sample))
1953                 return 0;
1954
1955         if (!trace->full_time && trace->base_time == 0)
1956                 trace->base_time = sample->time;
1957
1958         if (handler) {
1959                 ++trace->nr_events;
1960                 handler(trace, evsel, event, sample);
1961         }
1962
1963         return err;
1964 }
1965
1966 static int parse_target_str(struct trace *trace)
1967 {
1968         if (trace->opts.target.pid) {
1969                 trace->pid_list = intlist__new(trace->opts.target.pid);
1970                 if (trace->pid_list == NULL) {
1971                         pr_err("Error parsing process id string\n");
1972                         return -EINVAL;
1973                 }
1974         }
1975
1976         if (trace->opts.target.tid) {
1977                 trace->tid_list = intlist__new(trace->opts.target.tid);
1978                 if (trace->tid_list == NULL) {
1979                         pr_err("Error parsing thread id string\n");
1980                         return -EINVAL;
1981                 }
1982         }
1983
1984         return 0;
1985 }
1986
1987 static int trace__record(struct trace *trace, int argc, const char **argv)
1988 {
1989         unsigned int rec_argc, i, j;
1990         const char **rec_argv;
1991         const char * const record_args[] = {
1992                 "record",
1993                 "-R",
1994                 "-m", "1024",
1995                 "-c", "1",
1996         };
1997
1998         const char * const sc_args[] = { "-e", };
1999         unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2000         const char * const majpf_args[] = { "-e", "major-faults" };
2001         unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2002         const char * const minpf_args[] = { "-e", "minor-faults" };
2003         unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2004
2005         /* +1 is for the event string below */
2006         rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2007                 majpf_args_nr + minpf_args_nr + argc;
2008         rec_argv = calloc(rec_argc + 1, sizeof(char *));
2009
2010         if (rec_argv == NULL)
2011                 return -ENOMEM;
2012
2013         j = 0;
2014         for (i = 0; i < ARRAY_SIZE(record_args); i++)
2015                 rec_argv[j++] = record_args[i];
2016
2017         if (trace->trace_syscalls) {
2018                 for (i = 0; i < sc_args_nr; i++)
2019                         rec_argv[j++] = sc_args[i];
2020
2021                 /* event string may be different for older kernels - e.g., RHEL6 */
2022                 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2023                         rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2024                 else if (is_valid_tracepoint("syscalls:sys_enter"))
2025                         rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2026                 else {
2027                         pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2028                         return -1;
2029                 }
2030         }
2031
2032         if (trace->trace_pgfaults & TRACE_PFMAJ)
2033                 for (i = 0; i < majpf_args_nr; i++)
2034                         rec_argv[j++] = majpf_args[i];
2035
2036         if (trace->trace_pgfaults & TRACE_PFMIN)
2037                 for (i = 0; i < minpf_args_nr; i++)
2038                         rec_argv[j++] = minpf_args[i];
2039
2040         for (i = 0; i < (unsigned int)argc; i++)
2041                 rec_argv[j++] = argv[i];
2042
2043         return cmd_record(j, rec_argv, NULL);
2044 }
2045
2046 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2047
2048 static void perf_evlist__add_vfs_getname(struct perf_evlist *evlist)
2049 {
2050         struct perf_evsel *evsel = perf_evsel__newtp("probe", "vfs_getname");
2051         if (evsel == NULL)
2052                 return;
2053
2054         if (perf_evsel__field(evsel, "pathname") == NULL) {
2055                 perf_evsel__delete(evsel);
2056                 return;
2057         }
2058
2059         evsel->handler = trace__vfs_getname;
2060         perf_evlist__add(evlist, evsel);
2061 }
2062
2063 static int perf_evlist__add_pgfault(struct perf_evlist *evlist,
2064                                     u64 config)
2065 {
2066         struct perf_evsel *evsel;
2067         struct perf_event_attr attr = {
2068                 .type = PERF_TYPE_SOFTWARE,
2069                 .mmap_data = 1,
2070         };
2071
2072         attr.config = config;
2073         attr.sample_period = 1;
2074
2075         event_attr_init(&attr);
2076
2077         evsel = perf_evsel__new(&attr);
2078         if (!evsel)
2079                 return -ENOMEM;
2080
2081         evsel->handler = trace__pgfault;
2082         perf_evlist__add(evlist, evsel);
2083
2084         return 0;
2085 }
2086
2087 static int trace__run(struct trace *trace, int argc, const char **argv)
2088 {
2089         struct perf_evlist *evlist = trace->evlist;
2090         struct perf_evsel *evsel;
2091         int err = -1, i;
2092         unsigned long before;
2093         const bool forks = argc > 0;
2094         bool draining = false;
2095
2096         trace->live = true;
2097
2098         if (trace->trace_syscalls &&
2099             perf_evlist__add_syscall_newtp(evlist, trace__sys_enter,
2100                                            trace__sys_exit))
2101                 goto out_error_raw_syscalls;
2102
2103         if (trace->trace_syscalls)
2104                 perf_evlist__add_vfs_getname(evlist);
2105
2106         if ((trace->trace_pgfaults & TRACE_PFMAJ) &&
2107             perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MAJ)) {
2108                 goto out_error_mem;
2109         }
2110
2111         if ((trace->trace_pgfaults & TRACE_PFMIN) &&
2112             perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MIN))
2113                 goto out_error_mem;
2114
2115         if (trace->sched &&
2116             perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
2117                                    trace__sched_stat_runtime))
2118                 goto out_error_sched_stat_runtime;
2119
2120         err = perf_evlist__create_maps(evlist, &trace->opts.target);
2121         if (err < 0) {
2122                 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
2123                 goto out_delete_evlist;
2124         }
2125
2126         err = trace__symbols_init(trace, evlist);
2127         if (err < 0) {
2128                 fprintf(trace->output, "Problems initializing symbol libraries!\n");
2129                 goto out_delete_evlist;
2130         }
2131
2132         perf_evlist__config(evlist, &trace->opts);
2133
2134         signal(SIGCHLD, sig_handler);
2135         signal(SIGINT, sig_handler);
2136
2137         if (forks) {
2138                 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
2139                                                     argv, false, NULL);
2140                 if (err < 0) {
2141                         fprintf(trace->output, "Couldn't run the workload!\n");
2142                         goto out_delete_evlist;
2143                 }
2144         }
2145
2146         err = perf_evlist__open(evlist);
2147         if (err < 0)
2148                 goto out_error_open;
2149
2150         err = perf_evlist__mmap(evlist, trace->opts.mmap_pages, false);
2151         if (err < 0)
2152                 goto out_error_mmap;
2153
2154         if (forks)
2155                 perf_evlist__start_workload(evlist);
2156         else
2157                 perf_evlist__enable(evlist);
2158
2159         trace->multiple_threads = evlist->threads->map[0] == -1 ||
2160                                   evlist->threads->nr > 1 ||
2161                                   perf_evlist__first(evlist)->attr.inherit;
2162 again:
2163         before = trace->nr_events;
2164
2165         for (i = 0; i < evlist->nr_mmaps; i++) {
2166                 union perf_event *event;
2167
2168                 while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
2169                         const u32 type = event->header.type;
2170                         tracepoint_handler handler;
2171                         struct perf_sample sample;
2172
2173                         ++trace->nr_events;
2174
2175                         err = perf_evlist__parse_sample(evlist, event, &sample);
2176                         if (err) {
2177                                 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
2178                                 goto next_event;
2179                         }
2180
2181                         if (!trace->full_time && trace->base_time == 0)
2182                                 trace->base_time = sample.time;
2183
2184                         if (type != PERF_RECORD_SAMPLE) {
2185                                 trace__process_event(trace, trace->host, event, &sample);
2186                                 continue;
2187                         }
2188
2189                         evsel = perf_evlist__id2evsel(evlist, sample.id);
2190                         if (evsel == NULL) {
2191                                 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample.id);
2192                                 goto next_event;
2193                         }
2194
2195                         if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
2196                             sample.raw_data == NULL) {
2197                                 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2198                                        perf_evsel__name(evsel), sample.tid,
2199                                        sample.cpu, sample.raw_size);
2200                                 goto next_event;
2201                         }
2202
2203                         handler = evsel->handler;
2204                         handler(trace, evsel, event, &sample);
2205 next_event:
2206                         perf_evlist__mmap_consume(evlist, i);
2207
2208                         if (interrupted)
2209                                 goto out_disable;
2210                 }
2211         }
2212
2213         if (trace->nr_events == before) {
2214                 int timeout = done ? 100 : -1;
2215
2216                 if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
2217                         if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP) == 0)
2218                                 draining = true;
2219
2220                         goto again;
2221                 }
2222         } else {
2223                 goto again;
2224         }
2225
2226 out_disable:
2227         perf_evlist__disable(evlist);
2228
2229         if (!err) {
2230                 if (trace->summary)
2231                         trace__fprintf_thread_summary(trace, trace->output);
2232
2233                 if (trace->show_tool_stats) {
2234                         fprintf(trace->output, "Stats:\n "
2235                                                " vfs_getname : %" PRIu64 "\n"
2236                                                " proc_getname: %" PRIu64 "\n",
2237                                 trace->stats.vfs_getname,
2238                                 trace->stats.proc_getname);
2239                 }
2240         }
2241
2242 out_delete_evlist:
2243         perf_evlist__delete(evlist);
2244         trace->evlist = NULL;
2245         trace->live = false;
2246         return err;
2247 {
2248         char errbuf[BUFSIZ];
2249
2250 out_error_sched_stat_runtime:
2251         debugfs__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
2252         goto out_error;
2253
2254 out_error_raw_syscalls:
2255         debugfs__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
2256         goto out_error;
2257
2258 out_error_mmap:
2259         perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
2260         goto out_error;
2261
2262 out_error_open:
2263         perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
2264
2265 out_error:
2266         fprintf(trace->output, "%s\n", errbuf);
2267         goto out_delete_evlist;
2268 }
2269 out_error_mem:
2270         fprintf(trace->output, "Not enough memory to run!\n");
2271         goto out_delete_evlist;
2272 }
2273
2274 static int trace__replay(struct trace *trace)
2275 {
2276         const struct perf_evsel_str_handler handlers[] = {
2277                 { "probe:vfs_getname",       trace__vfs_getname, },
2278         };
2279         struct perf_data_file file = {
2280                 .path  = input_name,
2281                 .mode  = PERF_DATA_MODE_READ,
2282         };
2283         struct perf_session *session;
2284         struct perf_evsel *evsel;
2285         int err = -1;
2286
2287         trace->tool.sample        = trace__process_sample;
2288         trace->tool.mmap          = perf_event__process_mmap;
2289         trace->tool.mmap2         = perf_event__process_mmap2;
2290         trace->tool.comm          = perf_event__process_comm;
2291         trace->tool.exit          = perf_event__process_exit;
2292         trace->tool.fork          = perf_event__process_fork;
2293         trace->tool.attr          = perf_event__process_attr;
2294         trace->tool.tracing_data = perf_event__process_tracing_data;
2295         trace->tool.build_id      = perf_event__process_build_id;
2296
2297         trace->tool.ordered_events = true;
2298         trace->tool.ordering_requires_timestamps = true;
2299
2300         /* add tid to output */
2301         trace->multiple_threads = true;
2302
2303         session = perf_session__new(&file, false, &trace->tool);
2304         if (session == NULL)
2305                 return -1;
2306
2307         if (symbol__init(&session->header.env) < 0)
2308                 goto out;
2309
2310         trace->host = &session->machines.host;
2311
2312         err = perf_session__set_tracepoints_handlers(session, handlers);
2313         if (err)
2314                 goto out;
2315
2316         evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2317                                                      "raw_syscalls:sys_enter");
2318         /* older kernels have syscalls tp versus raw_syscalls */
2319         if (evsel == NULL)
2320                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2321                                                              "syscalls:sys_enter");
2322
2323         if (evsel &&
2324             (perf_evsel__init_syscall_tp(evsel, trace__sys_enter) < 0 ||
2325             perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
2326                 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
2327                 goto out;
2328         }
2329
2330         evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2331                                                      "raw_syscalls:sys_exit");
2332         if (evsel == NULL)
2333                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2334                                                              "syscalls:sys_exit");
2335         if (evsel &&
2336             (perf_evsel__init_syscall_tp(evsel, trace__sys_exit) < 0 ||
2337             perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
2338                 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
2339                 goto out;
2340         }
2341
2342         evlist__for_each(session->evlist, evsel) {
2343                 if (evsel->attr.type == PERF_TYPE_SOFTWARE &&
2344                     (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
2345                      evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
2346                      evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS))
2347                         evsel->handler = trace__pgfault;
2348         }
2349
2350         err = parse_target_str(trace);
2351         if (err != 0)
2352                 goto out;
2353
2354         setup_pager();
2355
2356         err = perf_session__process_events(session, &trace->tool);
2357         if (err)
2358                 pr_err("Failed to process events, error %d", err);
2359
2360         else if (trace->summary)
2361                 trace__fprintf_thread_summary(trace, trace->output);
2362
2363 out:
2364         perf_session__delete(session);
2365
2366         return err;
2367 }
2368
2369 static size_t trace__fprintf_threads_header(FILE *fp)
2370 {
2371         size_t printed;
2372
2373         printed  = fprintf(fp, "\n Summary of events:\n\n");
2374
2375         return printed;
2376 }
2377
2378 static size_t thread__dump_stats(struct thread_trace *ttrace,
2379                                  struct trace *trace, FILE *fp)
2380 {
2381         struct stats *stats;
2382         size_t printed = 0;
2383         struct syscall *sc;
2384         struct int_node *inode = intlist__first(ttrace->syscall_stats);
2385
2386         if (inode == NULL)
2387                 return 0;
2388
2389         printed += fprintf(fp, "\n");
2390
2391         printed += fprintf(fp, "   syscall            calls      min       avg       max      stddev\n");
2392         printed += fprintf(fp, "                               (msec)    (msec)    (msec)        (%%)\n");
2393         printed += fprintf(fp, "   --------------- -------- --------- --------- ---------     ------\n");
2394
2395         /* each int_node is a syscall */
2396         while (inode) {
2397                 stats = inode->priv;
2398                 if (stats) {
2399                         double min = (double)(stats->min) / NSEC_PER_MSEC;
2400                         double max = (double)(stats->max) / NSEC_PER_MSEC;
2401                         double avg = avg_stats(stats);
2402                         double pct;
2403                         u64 n = (u64) stats->n;
2404
2405                         pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
2406                         avg /= NSEC_PER_MSEC;
2407
2408                         sc = &trace->syscalls.table[inode->i];
2409                         printed += fprintf(fp, "   %-15s", sc->name);
2410                         printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f",
2411                                            n, min, avg);
2412                         printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
2413                 }
2414
2415                 inode = intlist__next(inode);
2416         }
2417
2418         printed += fprintf(fp, "\n\n");
2419
2420         return printed;
2421 }
2422
2423 /* struct used to pass data to per-thread function */
2424 struct summary_data {
2425         FILE *fp;
2426         struct trace *trace;
2427         size_t printed;
2428 };
2429
2430 static int trace__fprintf_one_thread(struct thread *thread, void *priv)
2431 {
2432         struct summary_data *data = priv;
2433         FILE *fp = data->fp;
2434         size_t printed = data->printed;
2435         struct trace *trace = data->trace;
2436         struct thread_trace *ttrace = thread__priv(thread);
2437         double ratio;
2438
2439         if (ttrace == NULL)
2440                 return 0;
2441
2442         ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
2443
2444         printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
2445         printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
2446         printed += fprintf(fp, "%.1f%%", ratio);
2447         if (ttrace->pfmaj)
2448                 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
2449         if (ttrace->pfmin)
2450                 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
2451         printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
2452         printed += thread__dump_stats(ttrace, trace, fp);
2453
2454         data->printed += printed;
2455
2456         return 0;
2457 }
2458
2459 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
2460 {
2461         struct summary_data data = {
2462                 .fp = fp,
2463                 .trace = trace
2464         };
2465         data.printed = trace__fprintf_threads_header(fp);
2466
2467         machine__for_each_thread(trace->host, trace__fprintf_one_thread, &data);
2468
2469         return data.printed;
2470 }
2471
2472 static int trace__set_duration(const struct option *opt, const char *str,
2473                                int unset __maybe_unused)
2474 {
2475         struct trace *trace = opt->value;
2476
2477         trace->duration_filter = atof(str);
2478         return 0;
2479 }
2480
2481 static int trace__open_output(struct trace *trace, const char *filename)
2482 {
2483         struct stat st;
2484
2485         if (!stat(filename, &st) && st.st_size) {
2486                 char oldname[PATH_MAX];
2487
2488                 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
2489                 unlink(oldname);
2490                 rename(filename, oldname);
2491         }
2492
2493         trace->output = fopen(filename, "w");
2494
2495         return trace->output == NULL ? -errno : 0;
2496 }
2497
2498 static int parse_pagefaults(const struct option *opt, const char *str,
2499                             int unset __maybe_unused)
2500 {
2501         int *trace_pgfaults = opt->value;
2502
2503         if (strcmp(str, "all") == 0)
2504                 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
2505         else if (strcmp(str, "maj") == 0)
2506                 *trace_pgfaults |= TRACE_PFMAJ;
2507         else if (strcmp(str, "min") == 0)
2508                 *trace_pgfaults |= TRACE_PFMIN;
2509         else
2510                 return -1;
2511
2512         return 0;
2513 }
2514
2515 static void evlist__set_evsel_handler(struct perf_evlist *evlist, void *handler)
2516 {
2517         struct perf_evsel *evsel;
2518
2519         evlist__for_each(evlist, evsel)
2520                 evsel->handler = handler;
2521 }
2522
2523 int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
2524 {
2525         const char * const trace_usage[] = {
2526                 "perf trace [<options>] [<command>]",
2527                 "perf trace [<options>] -- <command> [<options>]",
2528                 "perf trace record [<options>] [<command>]",
2529                 "perf trace record [<options>] -- <command> [<options>]",
2530                 NULL
2531         };
2532         struct trace trace = {
2533                 .audit = {
2534                         .machine = audit_detect_machine(),
2535                         .open_id = audit_name_to_syscall("open", trace.audit.machine),
2536                 },
2537                 .syscalls = {
2538                         . max = -1,
2539                 },
2540                 .opts = {
2541                         .target = {
2542                                 .uid       = UINT_MAX,
2543                                 .uses_mmap = true,
2544                         },
2545                         .user_freq     = UINT_MAX,
2546                         .user_interval = ULLONG_MAX,
2547                         .no_buffering  = true,
2548                         .mmap_pages    = UINT_MAX,
2549                 },
2550                 .output = stdout,
2551                 .show_comm = true,
2552                 .trace_syscalls = true,
2553         };
2554         const char *output_name = NULL;
2555         const char *ev_qualifier_str = NULL;
2556         const struct option trace_options[] = {
2557         OPT_CALLBACK(0, "event", &trace.evlist, "event",
2558                      "event selector. use 'perf list' to list available events",
2559                      parse_events_option),
2560         OPT_BOOLEAN(0, "comm", &trace.show_comm,
2561                     "show the thread COMM next to its id"),
2562         OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
2563         OPT_STRING('e', "expr", &ev_qualifier_str, "expr",
2564                     "list of events to trace"),
2565         OPT_STRING('o', "output", &output_name, "file", "output file name"),
2566         OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
2567         OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
2568                     "trace events on existing process id"),
2569         OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
2570                     "trace events on existing thread id"),
2571         OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
2572                     "system-wide collection from all CPUs"),
2573         OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
2574                     "list of cpus to monitor"),
2575         OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
2576                     "child tasks do not inherit counters"),
2577         OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
2578                      "number of mmap data pages",
2579                      perf_evlist__parse_mmap_pages),
2580         OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
2581                    "user to profile"),
2582         OPT_CALLBACK(0, "duration", &trace, "float",
2583                      "show only events with duration > N.M ms",
2584                      trace__set_duration),
2585         OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
2586         OPT_INCR('v', "verbose", &verbose, "be more verbose"),
2587         OPT_BOOLEAN('T', "time", &trace.full_time,
2588                     "Show full timestamp, not time relative to first start"),
2589         OPT_BOOLEAN('s', "summary", &trace.summary_only,
2590                     "Show only syscall summary with statistics"),
2591         OPT_BOOLEAN('S', "with-summary", &trace.summary,
2592                     "Show all syscalls and summary with statistics"),
2593         OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
2594                      "Trace pagefaults", parse_pagefaults, "maj"),
2595         OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
2596         OPT_END()
2597         };
2598         int err;
2599         char bf[BUFSIZ];
2600
2601         trace.evlist = perf_evlist__new();
2602         if (trace.evlist == NULL)
2603                 return -ENOMEM;
2604
2605         if (trace.evlist == NULL) {
2606                 pr_err("Not enough memory to run!\n");
2607                 goto out;
2608         }
2609
2610         argc = parse_options(argc, argv, trace_options, trace_usage,
2611                              PARSE_OPT_STOP_AT_NON_OPTION);
2612
2613         if (trace.trace_pgfaults) {
2614                 trace.opts.sample_address = true;
2615                 trace.opts.sample_time = true;
2616         }
2617
2618         if (trace.evlist->nr_entries > 0)
2619                 evlist__set_evsel_handler(trace.evlist, trace__event_handler);
2620
2621         if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
2622                 return trace__record(&trace, argc-1, &argv[1]);
2623
2624         /* summary_only implies summary option, but don't overwrite summary if set */
2625         if (trace.summary_only)
2626                 trace.summary = trace.summary_only;
2627
2628         if (!trace.trace_syscalls && !trace.trace_pgfaults &&
2629             trace.evlist->nr_entries == 0 /* Was --events used? */) {
2630                 pr_err("Please specify something to trace.\n");
2631                 return -1;
2632         }
2633
2634         if (output_name != NULL) {
2635                 err = trace__open_output(&trace, output_name);
2636                 if (err < 0) {
2637                         perror("failed to create output file");
2638                         goto out;
2639                 }
2640         }
2641
2642         if (ev_qualifier_str != NULL) {
2643                 const char *s = ev_qualifier_str;
2644
2645                 trace.not_ev_qualifier = *s == '!';
2646                 if (trace.not_ev_qualifier)
2647                         ++s;
2648                 trace.ev_qualifier = strlist__new(true, s);
2649                 if (trace.ev_qualifier == NULL) {
2650                         fputs("Not enough memory to parse event qualifier",
2651                               trace.output);
2652                         err = -ENOMEM;
2653                         goto out_close;
2654                 }
2655         }
2656
2657         err = target__validate(&trace.opts.target);
2658         if (err) {
2659                 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
2660                 fprintf(trace.output, "%s", bf);
2661                 goto out_close;
2662         }
2663
2664         err = target__parse_uid(&trace.opts.target);
2665         if (err) {
2666                 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
2667                 fprintf(trace.output, "%s", bf);
2668                 goto out_close;
2669         }
2670
2671         if (!argc && target__none(&trace.opts.target))
2672                 trace.opts.target.system_wide = true;
2673
2674         if (input_name)
2675                 err = trace__replay(&trace);
2676         else
2677                 err = trace__run(&trace, argc, argv);
2678
2679 out_close:
2680         if (output_name != NULL)
2681                 fclose(trace.output);
2682 out:
2683         return err;
2684 }