4 * Builtin 'trace' command:
6 * Display a continuously updated trace of any workload, CPU, specific PID,
7 * system wide, etc. Default format is loosely strace like, but any other
8 * event may be specified using --event.
10 * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
12 * Initially based on the 'trace' prototype by Thomas Gleixner:
14 * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
16 * Released under the GPL v2. (and only v2, not any later version)
19 #include <traceevent/event-parse.h>
20 #include <api/fs/tracing_path.h>
23 #include "util/cgroup.h"
24 #include "util/color.h"
25 #include "util/config.h"
26 #include "util/debug.h"
28 #include "util/event.h"
29 #include "util/evlist.h"
30 #include <subcmd/exec-cmd.h>
31 #include "util/machine.h"
32 #include "util/path.h"
33 #include "util/session.h"
34 #include "util/thread.h"
35 #include <subcmd/parse-options.h>
36 #include "util/strlist.h"
37 #include "util/intlist.h"
38 #include "util/thread_map.h"
39 #include "util/stat.h"
40 #include "trace/beauty/beauty.h"
41 #include "trace-event.h"
42 #include "util/parse-events.h"
43 #include "util/bpf-loader.h"
44 #include "callchain.h"
45 #include "print_binary.h"
47 #include "syscalltbl.h"
48 #include "rb_resort.h"
56 #include <linux/err.h>
57 #include <linux/filter.h>
58 #include <linux/kernel.h>
59 #include <linux/random.h>
60 #include <linux/stringify.h>
61 #include <linux/time64.h>
63 #include <sys/sysmacros.h>
65 #include "sane_ctype.h"
68 # define O_CLOEXEC 02000000
71 #ifndef F_LINUX_SPECIFIC_BASE
72 # define F_LINUX_SPECIFIC_BASE 1024
76 struct perf_tool tool;
77 struct syscalltbl *sctbl;
80 struct syscall *table;
83 struct perf_evsel *sys_enter,
88 struct record_opts opts;
89 struct perf_evlist *evlist;
91 struct thread *current;
92 struct cgroup *cgroup;
95 unsigned long nr_events;
96 unsigned long nr_events_printed;
97 unsigned long max_events;
98 struct strlist *ev_qualifier;
108 double duration_filter;
114 unsigned int max_stack;
115 unsigned int min_stack;
116 int raw_augmented_syscalls_args_size;
117 bool raw_augmented_syscalls;
119 bool not_ev_qualifier;
123 bool multiple_threads;
129 bool show_tool_stats;
131 bool kernel_syscallchains;
137 bool show_string_prefix;
142 struct ordered_events data;
150 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
151 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
155 #define TP_UINT_FIELD(bits) \
156 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
159 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
168 #define TP_UINT_FIELD__SWAPPED(bits) \
169 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
172 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
173 return bswap_##bits(value);\
176 TP_UINT_FIELD__SWAPPED(16);
177 TP_UINT_FIELD__SWAPPED(32);
178 TP_UINT_FIELD__SWAPPED(64);
180 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
182 field->offset = offset;
186 field->integer = tp_field__u8;
189 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
192 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
195 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
204 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
206 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
209 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
211 return sample->raw_data + field->offset;
214 static int __tp_field__init_ptr(struct tp_field *field, int offset)
216 field->offset = offset;
217 field->pointer = tp_field__ptr;
221 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
223 return __tp_field__init_ptr(field, format_field->offset);
229 struct tp_field args, ret;
233 static int perf_evsel__init_tp_uint_field(struct perf_evsel *evsel,
234 struct tp_field *field,
237 struct tep_format_field *format_field = perf_evsel__field(evsel, name);
239 if (format_field == NULL)
242 return tp_field__init_uint(field, format_field, evsel->needs_swap);
245 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
246 ({ struct syscall_tp *sc = evsel->priv;\
247 perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
249 static int perf_evsel__init_tp_ptr_field(struct perf_evsel *evsel,
250 struct tp_field *field,
253 struct tep_format_field *format_field = perf_evsel__field(evsel, name);
255 if (format_field == NULL)
258 return tp_field__init_ptr(field, format_field);
261 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
262 ({ struct syscall_tp *sc = evsel->priv;\
263 perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
265 static void perf_evsel__delete_priv(struct perf_evsel *evsel)
268 perf_evsel__delete(evsel);
271 static int perf_evsel__init_syscall_tp(struct perf_evsel *evsel)
273 struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
275 if (evsel->priv != NULL) {
276 if (perf_evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
277 perf_evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
288 static int perf_evsel__init_augmented_syscall_tp(struct perf_evsel *evsel, struct perf_evsel *tp)
290 struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
292 if (evsel->priv != NULL) {
293 struct tep_format_field *syscall_id = perf_evsel__field(tp, "id");
294 if (syscall_id == NULL)
295 syscall_id = perf_evsel__field(tp, "__syscall_nr");
296 if (syscall_id == NULL)
298 if (__tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
310 static int perf_evsel__init_augmented_syscall_tp_args(struct perf_evsel *evsel)
312 struct syscall_tp *sc = evsel->priv;
314 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
317 static int perf_evsel__init_augmented_syscall_tp_ret(struct perf_evsel *evsel)
319 struct syscall_tp *sc = evsel->priv;
321 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
324 static int perf_evsel__init_raw_syscall_tp(struct perf_evsel *evsel, void *handler)
326 evsel->priv = malloc(sizeof(struct syscall_tp));
327 if (evsel->priv != NULL) {
328 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
331 evsel->handler = handler;
342 static struct perf_evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
344 struct perf_evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
346 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
348 evsel = perf_evsel__newtp("syscalls", direction);
353 if (perf_evsel__init_raw_syscall_tp(evsel, handler))
359 perf_evsel__delete_priv(evsel);
363 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
364 ({ struct syscall_tp *fields = evsel->priv; \
365 fields->name.integer(&fields->name, sample); })
367 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
368 ({ struct syscall_tp *fields = evsel->priv; \
369 fields->name.pointer(&fields->name, sample); })
371 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
373 int idx = val - sa->offset;
375 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
376 size_t printed = scnprintf(bf, size, intfmt, val);
378 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
382 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
385 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
387 struct syscall_arg *arg)
389 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
392 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
393 struct syscall_arg *arg)
395 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
398 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
400 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
405 for (i = 0; i < sas->nr_entries; ++i) {
406 struct strarray *sa = sas->entries[i];
407 int idx = val - sa->offset;
409 if (idx >= 0 && idx < sa->nr_entries) {
410 if (sa->entries[idx] == NULL)
412 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
416 printed = scnprintf(bf, size, intfmt, val);
418 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
422 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
423 struct syscall_arg *arg)
425 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
429 #define AT_FDCWD -100
432 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
433 struct syscall_arg *arg)
436 const char *prefix = "AT_FD";
439 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
441 return syscall_arg__scnprintf_fd(bf, size, arg);
444 #define SCA_FDAT syscall_arg__scnprintf_fd_at
446 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
447 struct syscall_arg *arg);
449 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
451 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
453 return scnprintf(bf, size, "%#lx", arg->val);
456 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
459 return scnprintf(bf, size, "NULL");
460 return syscall_arg__scnprintf_hex(bf, size, arg);
463 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
465 return scnprintf(bf, size, "%d", arg->val);
468 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
470 return scnprintf(bf, size, "%ld", arg->val);
473 static const char *bpf_cmd[] = {
474 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
475 "MAP_GET_NEXT_KEY", "PROG_LOAD",
477 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
479 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
480 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
482 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
483 static DEFINE_STRARRAY(itimers, "ITIMER_");
485 static const char *keyctl_options[] = {
486 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
487 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
488 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
489 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
490 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
492 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
494 static const char *whences[] = { "SET", "CUR", "END",
502 static DEFINE_STRARRAY(whences, "SEEK_");
504 static const char *fcntl_cmds[] = {
505 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
506 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
507 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
510 static DEFINE_STRARRAY(fcntl_cmds, "F_");
512 static const char *fcntl_linux_specific_cmds[] = {
513 "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
514 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
515 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
518 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
520 static struct strarray *fcntl_cmds_arrays[] = {
521 &strarray__fcntl_cmds,
522 &strarray__fcntl_linux_specific_cmds,
525 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
527 static const char *rlimit_resources[] = {
528 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
529 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
532 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
534 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
535 static DEFINE_STRARRAY(sighow, "SIG_");
537 static const char *clockid[] = {
538 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
539 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
540 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
542 static DEFINE_STRARRAY(clockid, "CLOCK_");
544 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
545 struct syscall_arg *arg)
547 bool show_prefix = arg->show_string_prefix;
548 const char *suffix = "_OK";
552 if (mode == F_OK) /* 0 */
553 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
555 if (mode & n##_OK) { \
556 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
566 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
571 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
573 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
574 struct syscall_arg *arg);
576 #define SCA_FILENAME syscall_arg__scnprintf_filename
578 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
579 struct syscall_arg *arg)
581 bool show_prefix = arg->show_string_prefix;
582 const char *prefix = "O_";
583 int printed = 0, flags = arg->val;
586 if (flags & O_##n) { \
587 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
596 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
601 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
603 #ifndef GRND_NONBLOCK
604 #define GRND_NONBLOCK 0x0001
607 #define GRND_RANDOM 0x0002
610 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
611 struct syscall_arg *arg)
613 bool show_prefix = arg->show_string_prefix;
614 const char *prefix = "GRND_";
615 int printed = 0, flags = arg->val;
618 if (flags & GRND_##n) { \
619 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
620 flags &= ~GRND_##n; \
628 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
633 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
635 #define STRARRAY(name, array) \
636 { .scnprintf = SCA_STRARRAY, \
637 .parm = &strarray__##array, }
639 #include "trace/beauty/arch_errno_names.c"
640 #include "trace/beauty/eventfd.c"
641 #include "trace/beauty/futex_op.c"
642 #include "trace/beauty/futex_val3.c"
643 #include "trace/beauty/mmap.c"
644 #include "trace/beauty/mode_t.c"
645 #include "trace/beauty/msg_flags.c"
646 #include "trace/beauty/open_flags.c"
647 #include "trace/beauty/perf_event_open.c"
648 #include "trace/beauty/pid.c"
649 #include "trace/beauty/sched_policy.c"
650 #include "trace/beauty/seccomp.c"
651 #include "trace/beauty/signum.c"
652 #include "trace/beauty/socket_type.c"
653 #include "trace/beauty/waitid_options.c"
655 struct syscall_arg_fmt {
656 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
657 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
663 static struct syscall_fmt {
666 struct syscall_arg_fmt arg[6];
673 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
674 { .name = "arch_prctl",
675 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
676 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
678 .arg = { [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ }, }, },
680 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
681 { .name = "brk", .hexret = true,
682 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
683 { .name = "clock_gettime",
684 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
685 { .name = "clone", .errpid = true, .nr_args = 5,
686 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
687 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
688 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
689 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
690 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
692 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
694 .arg = { [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ }, }, },
695 { .name = "epoll_ctl",
696 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
697 { .name = "eventfd2",
698 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
699 { .name = "fchmodat",
700 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
701 { .name = "fchownat",
702 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
704 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
705 .parm = &strarrays__fcntl_cmds_arrays,
706 .show_zero = true, },
707 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
709 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
710 { .name = "fstat", .alias = "newfstat", },
711 { .name = "fstatat", .alias = "newfstatat", },
713 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
714 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
715 { .name = "futimesat",
716 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
717 { .name = "getitimer",
718 .arg = { [0] = STRARRAY(which, itimers), }, },
719 { .name = "getpid", .errpid = true, },
720 { .name = "getpgid", .errpid = true, },
721 { .name = "getppid", .errpid = true, },
722 { .name = "getrandom",
723 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
724 { .name = "getrlimit",
725 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
726 { .name = "gettid", .errpid = true, },
729 #if defined(__i386__) || defined(__x86_64__)
731 * FIXME: Make this available to all arches.
733 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
734 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
736 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
738 { .name = "kcmp", .nr_args = 5,
739 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
740 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
741 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
742 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
743 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
745 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
747 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
749 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
751 .arg = { [2] = STRARRAY(whence, whences), }, },
752 { .name = "lstat", .alias = "newlstat", },
754 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
755 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
757 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
759 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
760 { .name = "mmap", .hexret = true,
761 /* The standard mmap maps to old_mmap on s390x */
762 #if defined(__s390x__)
765 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
766 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */ },
767 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
769 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
770 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
771 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
772 { .name = "mprotect",
773 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
774 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, },
775 { .name = "mq_unlink",
776 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
777 { .name = "mremap", .hexret = true,
778 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
779 { .name = "name_to_handle_at",
780 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
781 { .name = "newfstatat",
782 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
784 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
785 { .name = "open_by_handle_at",
786 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
787 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
789 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
790 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
791 { .name = "perf_event_open",
792 .arg = { [2] = { .scnprintf = SCA_INT, /* cpu */ },
793 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
794 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
796 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
797 { .name = "pkey_alloc",
798 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
799 { .name = "pkey_free",
800 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
801 { .name = "pkey_mprotect",
802 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
803 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
804 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
805 { .name = "poll", .timeout = true, },
806 { .name = "ppoll", .timeout = true, },
808 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */ },
809 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
810 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
811 { .name = "pread", .alias = "pread64", },
812 { .name = "preadv", .alias = "pread", },
813 { .name = "prlimit64",
814 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
815 { .name = "pwrite", .alias = "pwrite64", },
816 { .name = "readlinkat",
817 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
818 { .name = "recvfrom",
819 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
820 { .name = "recvmmsg",
821 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
823 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
824 { .name = "renameat",
825 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
826 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
827 { .name = "renameat2",
828 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
829 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
830 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
831 { .name = "rt_sigaction",
832 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
833 { .name = "rt_sigprocmask",
834 .arg = { [0] = STRARRAY(how, sighow), }, },
835 { .name = "rt_sigqueueinfo",
836 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
837 { .name = "rt_tgsigqueueinfo",
838 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
839 { .name = "sched_setscheduler",
840 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
842 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
843 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
844 { .name = "select", .timeout = true, },
845 { .name = "sendmmsg",
846 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
848 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
850 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
851 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
852 { .name = "set_tid_address", .errpid = true, },
853 { .name = "setitimer",
854 .arg = { [0] = STRARRAY(which, itimers), }, },
855 { .name = "setrlimit",
856 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
858 .arg = { [0] = STRARRAY(family, socket_families),
859 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
860 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
861 { .name = "socketpair",
862 .arg = { [0] = STRARRAY(family, socket_families),
863 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
864 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
865 { .name = "stat", .alias = "newstat", },
867 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
868 [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
869 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
871 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
873 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
874 { .name = "symlinkat",
875 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
877 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
879 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
880 { .name = "umount2", .alias = "umount",
881 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
882 { .name = "uname", .alias = "newuname", },
883 { .name = "unlinkat",
884 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
885 { .name = "utimensat",
886 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
887 { .name = "wait4", .errpid = true,
888 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
889 { .name = "waitid", .errpid = true,
890 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
893 static int syscall_fmt__cmp(const void *name, const void *fmtp)
895 const struct syscall_fmt *fmt = fmtp;
896 return strcmp(name, fmt->name);
899 static struct syscall_fmt *syscall_fmt__find(const char *name)
901 const int nmemb = ARRAY_SIZE(syscall_fmts);
902 return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
905 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
907 int i, nmemb = ARRAY_SIZE(syscall_fmts);
909 for (i = 0; i < nmemb; ++i) {
910 if (syscall_fmts[i].alias && strcmp(syscall_fmts[i].alias, alias) == 0)
911 return &syscall_fmts[i];
918 * is_exit: is this "exit" or "exit_group"?
919 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
920 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
923 struct tep_event *tp_format;
928 struct tep_format_field *args;
930 struct syscall_fmt *fmt;
931 struct syscall_arg_fmt *arg_fmt;
934 struct bpf_map_syscall_entry {
939 * We need to have this 'calculated' boolean because in some cases we really
940 * don't know what is the duration of a syscall, for instance, when we start
941 * a session and some threads are waiting for a syscall to finish, say 'poll',
942 * in which case all we can do is to print "( ? ) for duration and for the
945 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
947 double duration = (double)t / NSEC_PER_MSEC;
948 size_t printed = fprintf(fp, "(");
951 printed += fprintf(fp, " ");
952 else if (duration >= 1.0)
953 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
954 else if (duration >= 0.01)
955 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
957 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
958 return printed + fprintf(fp, "): ");
962 * filename.ptr: The filename char pointer that will be vfs_getname'd
963 * filename.entry_str_pos: Where to insert the string translated from
964 * filename.ptr by the vfs_getname tracepoint/kprobe.
965 * ret_scnprintf: syscall args may set this to a different syscall return
966 * formatter, for instance, fcntl may return fds, file flags, etc.
968 struct thread_trace {
971 unsigned long nr_events;
972 unsigned long pfmaj, pfmin;
975 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
978 short int entry_str_pos;
980 unsigned int namelen;
988 struct intlist *syscall_stats;
991 static struct thread_trace *thread_trace__new(void)
993 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
996 ttrace->files.max = -1;
998 ttrace->syscall_stats = intlist__new(NULL);
1003 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1005 struct thread_trace *ttrace;
1010 if (thread__priv(thread) == NULL)
1011 thread__set_priv(thread, thread_trace__new());
1013 if (thread__priv(thread) == NULL)
1016 ttrace = thread__priv(thread);
1017 ++ttrace->nr_events;
1021 color_fprintf(fp, PERF_COLOR_RED,
1022 "WARNING: not enough memory, dropping samples!\n");
1027 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1028 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1030 struct thread_trace *ttrace = thread__priv(arg->thread);
1032 ttrace->ret_scnprintf = ret_scnprintf;
1035 #define TRACE_PFMAJ (1 << 0)
1036 #define TRACE_PFMIN (1 << 1)
1038 static const size_t trace__entry_str_size = 2048;
1040 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1042 if (fd > ttrace->files.max) {
1043 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1048 if (ttrace->files.max != -1) {
1049 memset(nfiles + ttrace->files.max + 1, 0,
1050 (fd - ttrace->files.max) * sizeof(struct file));
1052 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1055 ttrace->files.table = nfiles;
1056 ttrace->files.max = fd;
1059 return ttrace->files.table + fd;
1062 struct file *thread__files_entry(struct thread *thread, int fd)
1064 return thread_trace__files_entry(thread__priv(thread), fd);
1067 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1069 struct thread_trace *ttrace = thread__priv(thread);
1070 struct file *file = thread_trace__files_entry(ttrace, fd);
1074 if (stat(pathname, &st) == 0)
1075 file->dev_maj = major(st.st_rdev);
1076 file->pathname = strdup(pathname);
1084 static int thread__read_fd_path(struct thread *thread, int fd)
1086 char linkname[PATH_MAX], pathname[PATH_MAX];
1090 if (thread->pid_ == thread->tid) {
1091 scnprintf(linkname, sizeof(linkname),
1092 "/proc/%d/fd/%d", thread->pid_, fd);
1094 scnprintf(linkname, sizeof(linkname),
1095 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1098 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1101 ret = readlink(linkname, pathname, sizeof(pathname));
1103 if (ret < 0 || ret > st.st_size)
1106 pathname[ret] = '\0';
1107 return trace__set_fd_pathname(thread, fd, pathname);
1110 static const char *thread__fd_path(struct thread *thread, int fd,
1111 struct trace *trace)
1113 struct thread_trace *ttrace = thread__priv(thread);
1121 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1124 ++trace->stats.proc_getname;
1125 if (thread__read_fd_path(thread, fd))
1129 return ttrace->files.table[fd].pathname;
1132 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1135 size_t printed = scnprintf(bf, size, "%d", fd);
1136 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1139 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1144 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1146 size_t printed = scnprintf(bf, size, "%d", fd);
1147 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1150 const char *path = thread__fd_path(thread, fd, trace);
1153 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1155 thread__put(thread);
1161 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1162 struct syscall_arg *arg)
1165 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1166 struct thread_trace *ttrace = thread__priv(arg->thread);
1168 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1169 zfree(&ttrace->files.table[fd].pathname);
1174 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1177 struct thread_trace *ttrace = thread__priv(thread);
1179 ttrace->filename.ptr = ptr;
1180 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1183 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1185 struct augmented_arg *augmented_arg = arg->augmented.args;
1187 return scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1190 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1191 struct syscall_arg *arg)
1193 unsigned long ptr = arg->val;
1195 if (arg->augmented.args)
1196 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1198 if (!arg->trace->vfs_getname)
1199 return scnprintf(bf, size, "%#x", ptr);
1201 thread__set_filename_pos(arg->thread, bf, ptr);
1205 static bool trace__filter_duration(struct trace *trace, double t)
1207 return t < (trace->duration_filter * NSEC_PER_MSEC);
1210 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1212 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1214 return fprintf(fp, "%10.3f ", ts);
1218 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1219 * using ttrace->entry_time for a thread that receives a sys_exit without
1220 * first having received a sys_enter ("poll" issued before tracing session
1221 * starts, lost sys_enter exit due to ring buffer overflow).
1223 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1226 return __trace__fprintf_tstamp(trace, tstamp, fp);
1228 return fprintf(fp, " ? ");
1231 static bool done = false;
1232 static bool interrupted = false;
1234 static void sig_handler(int sig)
1237 interrupted = sig == SIGINT;
1240 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1244 if (trace->multiple_threads) {
1245 if (trace->show_comm)
1246 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1247 printed += fprintf(fp, "%d ", thread->tid);
1253 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1254 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1258 if (trace->show_tstamp)
1259 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1260 if (trace->show_duration)
1261 printed += fprintf_duration(duration, duration_calculated, fp);
1262 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1265 static int trace__process_event(struct trace *trace, struct machine *machine,
1266 union perf_event *event, struct perf_sample *sample)
1270 switch (event->header.type) {
1271 case PERF_RECORD_LOST:
1272 color_fprintf(trace->output, PERF_COLOR_RED,
1273 "LOST %" PRIu64 " events!\n", event->lost.lost);
1274 ret = machine__process_lost_event(machine, event, sample);
1277 ret = machine__process_event(machine, event, sample);
1284 static int trace__tool_process(struct perf_tool *tool,
1285 union perf_event *event,
1286 struct perf_sample *sample,
1287 struct machine *machine)
1289 struct trace *trace = container_of(tool, struct trace, tool);
1290 return trace__process_event(trace, machine, event, sample);
1293 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1295 struct machine *machine = vmachine;
1297 if (machine->kptr_restrict_warned)
1300 if (symbol_conf.kptr_restrict) {
1301 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1302 "Check /proc/sys/kernel/kptr_restrict.\n\n"
1303 "Kernel samples will not be resolved.\n");
1304 machine->kptr_restrict_warned = true;
1308 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1311 static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
1313 int err = symbol__init(NULL);
1318 trace->host = machine__new_host();
1319 if (trace->host == NULL)
1322 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1326 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1327 evlist->threads, trace__tool_process, false,
1336 static void trace__symbols__exit(struct trace *trace)
1338 machine__exit(trace->host);
1344 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1348 if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1349 nr_args = sc->fmt->nr_args;
1351 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1352 if (sc->arg_fmt == NULL)
1355 for (idx = 0; idx < nr_args; ++idx) {
1357 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1360 sc->nr_args = nr_args;
1364 static int syscall__set_arg_fmts(struct syscall *sc)
1366 struct tep_format_field *field, *last_field = NULL;
1369 for (field = sc->args; field; field = field->next, ++idx) {
1372 if (sc->fmt && sc->fmt->arg[idx].scnprintf)
1375 if (strcmp(field->type, "const char *") == 0 &&
1376 (strcmp(field->name, "filename") == 0 ||
1377 strcmp(field->name, "path") == 0 ||
1378 strcmp(field->name, "pathname") == 0))
1379 sc->arg_fmt[idx].scnprintf = SCA_FILENAME;
1380 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1381 sc->arg_fmt[idx].scnprintf = SCA_PTR;
1382 else if (strcmp(field->type, "pid_t") == 0)
1383 sc->arg_fmt[idx].scnprintf = SCA_PID;
1384 else if (strcmp(field->type, "umode_t") == 0)
1385 sc->arg_fmt[idx].scnprintf = SCA_MODE_T;
1386 else if ((strcmp(field->type, "int") == 0 ||
1387 strcmp(field->type, "unsigned int") == 0 ||
1388 strcmp(field->type, "long") == 0) &&
1389 (len = strlen(field->name)) >= 2 &&
1390 strcmp(field->name + len - 2, "fd") == 0) {
1392 * /sys/kernel/tracing/events/syscalls/sys_enter*
1393 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1398 sc->arg_fmt[idx].scnprintf = SCA_FD;
1403 sc->args_size = last_field->offset + last_field->size;
1408 static int trace__read_syscall_info(struct trace *trace, int id)
1412 const char *name = syscalltbl__name(trace->sctbl, id);
1417 if (id > trace->syscalls.max) {
1418 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1420 if (nsyscalls == NULL)
1423 if (trace->syscalls.max != -1) {
1424 memset(nsyscalls + trace->syscalls.max + 1, 0,
1425 (id - trace->syscalls.max) * sizeof(*sc));
1427 memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
1430 trace->syscalls.table = nsyscalls;
1431 trace->syscalls.max = id;
1434 sc = trace->syscalls.table + id;
1437 sc->fmt = syscall_fmt__find(sc->name);
1439 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1440 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1442 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1443 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1444 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1447 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1450 if (IS_ERR(sc->tp_format))
1453 sc->args = sc->tp_format->format.fields;
1455 * We need to check and discard the first variable '__syscall_nr'
1456 * or 'nr' that mean the syscall number. It is needless here.
1457 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1459 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1460 sc->args = sc->args->next;
1464 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1465 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1467 return syscall__set_arg_fmts(sc);
1470 static int trace__validate_ev_qualifier(struct trace *trace)
1473 size_t nr_allocated;
1474 struct str_node *pos;
1476 trace->ev_qualifier_ids.nr = strlist__nr_entries(trace->ev_qualifier);
1477 trace->ev_qualifier_ids.entries = malloc(trace->ev_qualifier_ids.nr *
1478 sizeof(trace->ev_qualifier_ids.entries[0]));
1480 if (trace->ev_qualifier_ids.entries == NULL) {
1481 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1487 nr_allocated = trace->ev_qualifier_ids.nr;
1490 strlist__for_each_entry(pos, trace->ev_qualifier) {
1491 const char *sc = pos->s;
1492 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1495 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1500 fputs("Error:\tInvalid syscall ", trace->output);
1503 fputs(", ", trace->output);
1506 fputs(sc, trace->output);
1509 trace->ev_qualifier_ids.entries[i++] = id;
1510 if (match_next == -1)
1514 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1517 if (nr_allocated == trace->ev_qualifier_ids.nr) {
1521 entries = realloc(trace->ev_qualifier_ids.entries,
1522 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1523 if (entries == NULL) {
1525 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1528 trace->ev_qualifier_ids.entries = entries;
1530 trace->ev_qualifier_ids.nr++;
1531 trace->ev_qualifier_ids.entries[i++] = id;
1536 fputs("\nHint:\ttry 'perf list syscalls:sys_enter_*'"
1537 "\nHint:\tand: 'man syscalls'\n", trace->output);
1539 zfree(&trace->ev_qualifier_ids.entries);
1540 trace->ev_qualifier_ids.nr = 0;
1547 * args is to be interpreted as a series of longs but we need to handle
1548 * 8-byte unaligned accesses. args points to raw_data within the event
1549 * and raw_data is guaranteed to be 8-byte unaligned because it is
1550 * preceded by raw_size which is a u32. So we need to copy args to a temp
1551 * variable to read it. Most notably this avoids extended load instructions
1552 * on unaligned addresses
1554 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1557 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1559 memcpy(&val, p, sizeof(val));
1563 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1564 struct syscall_arg *arg)
1566 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1567 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1569 return scnprintf(bf, size, "arg%d: ", arg->idx);
1573 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1574 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1575 * in tools/perf/trace/beauty/mount_flags.c
1577 static unsigned long syscall__mask_val(struct syscall *sc, struct syscall_arg *arg, unsigned long val)
1579 if (sc->arg_fmt && sc->arg_fmt[arg->idx].mask_val)
1580 return sc->arg_fmt[arg->idx].mask_val(arg, val);
1585 static size_t syscall__scnprintf_val(struct syscall *sc, char *bf, size_t size,
1586 struct syscall_arg *arg, unsigned long val)
1588 if (sc->arg_fmt && sc->arg_fmt[arg->idx].scnprintf) {
1590 if (sc->arg_fmt[arg->idx].parm)
1591 arg->parm = sc->arg_fmt[arg->idx].parm;
1592 return sc->arg_fmt[arg->idx].scnprintf(bf, size, arg);
1594 return scnprintf(bf, size, "%ld", val);
1597 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1598 unsigned char *args, void *augmented_args, int augmented_args_size,
1599 struct trace *trace, struct thread *thread)
1604 struct syscall_arg arg = {
1607 .size = augmented_args_size,
1608 .args = augmented_args,
1614 .show_string_prefix = trace->show_string_prefix,
1616 struct thread_trace *ttrace = thread__priv(thread);
1619 * Things like fcntl will set this in its 'cmd' formatter to pick the
1620 * right formatter for the return value (an fd? file flags?), which is
1621 * not needed for syscalls that always return a given type, say an fd.
1623 ttrace->ret_scnprintf = NULL;
1625 if (sc->args != NULL) {
1626 struct tep_format_field *field;
1628 for (field = sc->args; field;
1629 field = field->next, ++arg.idx, bit <<= 1) {
1633 val = syscall_arg__val(&arg, arg.idx);
1635 * Some syscall args need some mask, most don't and
1636 * return val untouched.
1638 val = syscall__mask_val(sc, &arg, val);
1641 * Suppress this argument if its value is zero and
1642 * and we don't have a string associated in an
1646 !trace->show_zeros &&
1648 (sc->arg_fmt[arg.idx].show_zero ||
1649 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
1650 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
1651 sc->arg_fmt[arg.idx].parm))
1654 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
1656 if (trace->show_arg_names)
1657 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
1659 printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1661 } else if (IS_ERR(sc->tp_format)) {
1663 * If we managed to read the tracepoint /format file, then we
1664 * may end up not having any args, like with gettid(), so only
1665 * print the raw args when we didn't manage to read it.
1667 while (arg.idx < sc->nr_args) {
1670 val = syscall_arg__val(&arg, arg.idx);
1672 printed += scnprintf(bf + printed, size - printed, ", ");
1673 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
1674 printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1684 typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
1685 union perf_event *event,
1686 struct perf_sample *sample);
1688 static struct syscall *trace__syscall_info(struct trace *trace,
1689 struct perf_evsel *evsel, int id)
1695 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1696 * before that, leaving at a higher verbosity level till that is
1697 * explained. Reproduced with plain ftrace with:
1699 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1700 * grep "NR -1 " /t/trace_pipe
1702 * After generating some load on the machine.
1706 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1707 id, perf_evsel__name(evsel), ++n);
1712 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
1713 trace__read_syscall_info(trace, id))
1716 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
1719 return &trace->syscalls.table[id];
1723 fprintf(trace->output, "Problems reading syscall %d", id);
1724 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
1725 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1726 fputs(" information\n", trace->output);
1731 static void thread__update_stats(struct thread_trace *ttrace,
1732 int id, struct perf_sample *sample)
1734 struct int_node *inode;
1735 struct stats *stats;
1738 inode = intlist__findnew(ttrace->syscall_stats, id);
1742 stats = inode->priv;
1743 if (stats == NULL) {
1744 stats = malloc(sizeof(struct stats));
1748 inode->priv = stats;
1751 if (ttrace->entry_time && sample->time > ttrace->entry_time)
1752 duration = sample->time - ttrace->entry_time;
1754 update_stats(stats, duration);
1757 static int trace__printf_interrupted_entry(struct trace *trace)
1759 struct thread_trace *ttrace;
1763 if (trace->failure_only || trace->current == NULL)
1766 ttrace = thread__priv(trace->current);
1768 if (!ttrace->entry_pending)
1771 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
1772 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
1774 if (len < trace->args_alignment - 4)
1775 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
1777 printed += fprintf(trace->output, " ...\n");
1779 ttrace->entry_pending = false;
1780 ++trace->nr_events_printed;
1785 static int trace__fprintf_sample(struct trace *trace, struct perf_evsel *evsel,
1786 struct perf_sample *sample, struct thread *thread)
1790 if (trace->print_sample) {
1791 double ts = (double)sample->time / NSEC_PER_MSEC;
1793 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
1794 perf_evsel__name(evsel), ts,
1795 thread__comm_str(thread),
1796 sample->pid, sample->tid, sample->cpu);
1802 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
1804 void *augmented_args = NULL;
1806 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
1807 * and there we get all 6 syscall args plus the tracepoint common fields
1808 * that gets calculated at the start and the syscall_nr (another long).
1809 * So we check if that is the case and if so don't look after the
1810 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
1813 * We'll revisit this later to pass s->args_size to the BPF augmenter
1814 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
1815 * copies only what we need for each syscall, like what happens when we
1816 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
1817 * traffic to just what is needed for each syscall.
1819 int args_size = raw_augmented_args_size ?: sc->args_size;
1821 *augmented_args_size = sample->raw_size - args_size;
1822 if (*augmented_args_size > 0)
1823 augmented_args = sample->raw_data + args_size;
1825 return augmented_args;
1828 static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
1829 union perf_event *event __maybe_unused,
1830 struct perf_sample *sample)
1835 struct thread *thread;
1836 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
1837 int augmented_args_size = 0;
1838 void *augmented_args = NULL;
1839 struct syscall *sc = trace__syscall_info(trace, evsel, id);
1840 struct thread_trace *ttrace;
1845 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1846 ttrace = thread__trace(thread, trace->output);
1850 trace__fprintf_sample(trace, evsel, sample, thread);
1852 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
1854 if (ttrace->entry_str == NULL) {
1855 ttrace->entry_str = malloc(trace__entry_str_size);
1856 if (!ttrace->entry_str)
1860 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
1861 trace__printf_interrupted_entry(trace);
1863 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
1864 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
1865 * this breaks syscall__augmented_args() check for augmented args, as we calculate
1866 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
1867 * so when handling, say the openat syscall, we end up getting 6 args for the
1868 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
1869 * thinking that the extra 2 u64 args are the augmented filename, so just check
1870 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
1872 if (evsel != trace->syscalls.events.sys_enter)
1873 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
1874 ttrace->entry_time = sample->time;
1875 msg = ttrace->entry_str;
1876 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
1878 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
1879 args, augmented_args, augmented_args_size, trace, thread);
1882 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
1885 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
1886 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
1887 if (trace->args_alignment > printed)
1888 alignment = trace->args_alignment - printed;
1889 fprintf(trace->output, "%*s= ?\n", alignment, " ");
1892 ttrace->entry_pending = true;
1893 /* See trace__vfs_getname & trace__sys_exit */
1894 ttrace->filename.pending_open = false;
1897 if (trace->current != thread) {
1898 thread__put(trace->current);
1899 trace->current = thread__get(thread);
1903 thread__put(thread);
1907 static int trace__fprintf_sys_enter(struct trace *trace, struct perf_evsel *evsel,
1908 struct perf_sample *sample)
1910 struct thread_trace *ttrace;
1911 struct thread *thread;
1912 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
1913 struct syscall *sc = trace__syscall_info(trace, evsel, id);
1915 void *args, *augmented_args = NULL;
1916 int augmented_args_size;
1921 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1922 ttrace = thread__trace(thread, trace->output);
1924 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
1925 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
1930 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
1931 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
1932 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
1933 fprintf(trace->output, "%s", msg);
1936 thread__put(thread);
1940 static int trace__resolve_callchain(struct trace *trace, struct perf_evsel *evsel,
1941 struct perf_sample *sample,
1942 struct callchain_cursor *cursor)
1944 struct addr_location al;
1945 int max_stack = evsel->attr.sample_max_stack ?
1946 evsel->attr.sample_max_stack :
1950 if (machine__resolve(trace->host, &al, sample) < 0)
1953 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
1954 addr_location__put(&al);
1958 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
1960 /* TODO: user-configurable print_opts */
1961 const unsigned int print_opts = EVSEL__PRINT_SYM |
1963 EVSEL__PRINT_UNKNOWN_AS_ADDR;
1965 return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, trace->output);
1968 static const char *errno_to_name(struct perf_evsel *evsel, int err)
1970 struct perf_env *env = perf_evsel__env(evsel);
1971 const char *arch_name = perf_env__arch(env);
1973 return arch_syscalls__strerrno(arch_name, err);
1976 static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
1977 union perf_event *event __maybe_unused,
1978 struct perf_sample *sample)
1982 bool duration_calculated = false;
1983 struct thread *thread;
1984 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
1985 int alignment = trace->args_alignment;
1986 struct syscall *sc = trace__syscall_info(trace, evsel, id);
1987 struct thread_trace *ttrace;
1992 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1993 ttrace = thread__trace(thread, trace->output);
1997 trace__fprintf_sample(trace, evsel, sample, thread);
2000 thread__update_stats(ttrace, id, sample);
2002 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2004 if (sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2005 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2006 ttrace->filename.pending_open = false;
2007 ++trace->stats.vfs_getname;
2010 if (ttrace->entry_time) {
2011 duration = sample->time - ttrace->entry_time;
2012 if (trace__filter_duration(trace, duration))
2014 duration_calculated = true;
2015 } else if (trace->duration_filter)
2018 if (sample->callchain) {
2019 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2020 if (callchain_ret == 0) {
2021 if (callchain_cursor.nr < trace->min_stack)
2027 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2030 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2032 if (ttrace->entry_pending) {
2033 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2035 printed += fprintf(trace->output, " ... [");
2036 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2038 printed += fprintf(trace->output, "]: %s()", sc->name);
2041 printed++; /* the closing ')' */
2043 if (alignment > printed)
2044 alignment -= printed;
2048 fprintf(trace->output, ")%*s= ", alignment, " ");
2050 if (sc->fmt == NULL) {
2054 fprintf(trace->output, "%ld", ret);
2055 } else if (ret < 0) {
2057 char bf[STRERR_BUFSIZE];
2058 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2059 *e = errno_to_name(evsel, -ret);
2061 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2063 } else if (ret == 0 && sc->fmt->timeout)
2064 fprintf(trace->output, "0 (Timeout)");
2065 else if (ttrace->ret_scnprintf) {
2067 struct syscall_arg arg = {
2072 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2073 ttrace->ret_scnprintf = NULL;
2074 fprintf(trace->output, "%s", bf);
2075 } else if (sc->fmt->hexret)
2076 fprintf(trace->output, "%#lx", ret);
2077 else if (sc->fmt->errpid) {
2078 struct thread *child = machine__find_thread(trace->host, ret, ret);
2080 if (child != NULL) {
2081 fprintf(trace->output, "%ld", ret);
2082 if (child->comm_set)
2083 fprintf(trace->output, " (%s)", thread__comm_str(child));
2089 fputc('\n', trace->output);
2092 * We only consider an 'event' for the sake of --max-events a non-filtered
2093 * sys_enter + sys_exit and other tracepoint events.
2095 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2098 if (callchain_ret > 0)
2099 trace__fprintf_callchain(trace, sample);
2100 else if (callchain_ret < 0)
2101 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2103 ttrace->entry_pending = false;
2106 thread__put(thread);
2110 static int trace__vfs_getname(struct trace *trace, struct perf_evsel *evsel,
2111 union perf_event *event __maybe_unused,
2112 struct perf_sample *sample)
2114 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2115 struct thread_trace *ttrace;
2116 size_t filename_len, entry_str_len, to_move;
2117 ssize_t remaining_space;
2119 const char *filename = perf_evsel__rawptr(evsel, sample, "pathname");
2124 ttrace = thread__priv(thread);
2128 filename_len = strlen(filename);
2129 if (filename_len == 0)
2132 if (ttrace->filename.namelen < filename_len) {
2133 char *f = realloc(ttrace->filename.name, filename_len + 1);
2138 ttrace->filename.namelen = filename_len;
2139 ttrace->filename.name = f;
2142 strcpy(ttrace->filename.name, filename);
2143 ttrace->filename.pending_open = true;
2145 if (!ttrace->filename.ptr)
2148 entry_str_len = strlen(ttrace->entry_str);
2149 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2150 if (remaining_space <= 0)
2153 if (filename_len > (size_t)remaining_space) {
2154 filename += filename_len - remaining_space;
2155 filename_len = remaining_space;
2158 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2159 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2160 memmove(pos + filename_len, pos, to_move);
2161 memcpy(pos, filename, filename_len);
2163 ttrace->filename.ptr = 0;
2164 ttrace->filename.entry_str_pos = 0;
2166 thread__put(thread);
2171 static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
2172 union perf_event *event __maybe_unused,
2173 struct perf_sample *sample)
2175 u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
2176 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2177 struct thread *thread = machine__findnew_thread(trace->host,
2180 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2185 ttrace->runtime_ms += runtime_ms;
2186 trace->runtime_ms += runtime_ms;
2188 thread__put(thread);
2192 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2194 perf_evsel__strval(evsel, sample, "comm"),
2195 (pid_t)perf_evsel__intval(evsel, sample, "pid"),
2197 perf_evsel__intval(evsel, sample, "vruntime"));
2201 static int bpf_output__printer(enum binary_printer_ops op,
2202 unsigned int val, void *extra __maybe_unused, FILE *fp)
2204 unsigned char ch = (unsigned char)val;
2207 case BINARY_PRINT_CHAR_DATA:
2208 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2209 case BINARY_PRINT_DATA_BEGIN:
2210 case BINARY_PRINT_LINE_BEGIN:
2211 case BINARY_PRINT_ADDR:
2212 case BINARY_PRINT_NUM_DATA:
2213 case BINARY_PRINT_NUM_PAD:
2214 case BINARY_PRINT_SEP:
2215 case BINARY_PRINT_CHAR_PAD:
2216 case BINARY_PRINT_LINE_END:
2217 case BINARY_PRINT_DATA_END:
2225 static void bpf_output__fprintf(struct trace *trace,
2226 struct perf_sample *sample)
2228 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2229 bpf_output__printer, NULL, trace->output);
2230 ++trace->nr_events_printed;
2233 static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
2234 union perf_event *event __maybe_unused,
2235 struct perf_sample *sample)
2237 struct thread *thread;
2238 int callchain_ret = 0;
2240 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2241 * this event's max_events having been hit and this is an entry coming
2242 * from the ring buffer that we should discard, since the max events
2243 * have already been considered/printed.
2245 if (evsel->disabled)
2248 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2250 if (sample->callchain) {
2251 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2252 if (callchain_ret == 0) {
2253 if (callchain_cursor.nr < trace->min_stack)
2259 trace__printf_interrupted_entry(trace);
2260 trace__fprintf_tstamp(trace, sample->time, trace->output);
2262 if (trace->trace_syscalls && trace->show_duration)
2263 fprintf(trace->output, "( ): ");
2266 trace__fprintf_comm_tid(trace, thread, trace->output);
2268 if (evsel == trace->syscalls.events.augmented) {
2269 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2270 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2273 fprintf(trace->output, "%s(", sc->name);
2274 trace__fprintf_sys_enter(trace, evsel, sample);
2275 fputc(')', trace->output);
2280 * XXX: Not having the associated syscall info or not finding/adding
2281 * the thread should never happen, but if it does...
2282 * fall thru and print it as a bpf_output event.
2286 fprintf(trace->output, "%s:", evsel->name);
2288 if (perf_evsel__is_bpf_output(evsel)) {
2289 bpf_output__fprintf(trace, sample);
2290 } else if (evsel->tp_format) {
2291 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2292 trace__fprintf_sys_enter(trace, evsel, sample)) {
2293 event_format__fprintf(evsel->tp_format, sample->cpu,
2294 sample->raw_data, sample->raw_size,
2296 ++trace->nr_events_printed;
2298 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2299 perf_evsel__disable(evsel);
2300 perf_evsel__close(evsel);
2306 fprintf(trace->output, "\n");
2308 if (callchain_ret > 0)
2309 trace__fprintf_callchain(trace, sample);
2310 else if (callchain_ret < 0)
2311 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2313 thread__put(thread);
2317 static void print_location(FILE *f, struct perf_sample *sample,
2318 struct addr_location *al,
2319 bool print_dso, bool print_sym)
2322 if ((verbose > 0 || print_dso) && al->map)
2323 fprintf(f, "%s@", al->map->dso->long_name);
2325 if ((verbose > 0 || print_sym) && al->sym)
2326 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2327 al->addr - al->sym->start);
2329 fprintf(f, "0x%" PRIx64, al->addr);
2331 fprintf(f, "0x%" PRIx64, sample->addr);
2334 static int trace__pgfault(struct trace *trace,
2335 struct perf_evsel *evsel,
2336 union perf_event *event __maybe_unused,
2337 struct perf_sample *sample)
2339 struct thread *thread;
2340 struct addr_location al;
2341 char map_type = 'd';
2342 struct thread_trace *ttrace;
2344 int callchain_ret = 0;
2346 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2348 if (sample->callchain) {
2349 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2350 if (callchain_ret == 0) {
2351 if (callchain_cursor.nr < trace->min_stack)
2357 ttrace = thread__trace(thread, trace->output);
2361 if (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2366 if (trace->summary_only)
2369 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2371 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2373 fprintf(trace->output, "%sfault [",
2374 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2377 print_location(trace->output, sample, &al, false, true);
2379 fprintf(trace->output, "] => ");
2381 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2384 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2392 print_location(trace->output, sample, &al, true, false);
2394 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2396 if (callchain_ret > 0)
2397 trace__fprintf_callchain(trace, sample);
2398 else if (callchain_ret < 0)
2399 pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2401 ++trace->nr_events_printed;
2405 thread__put(thread);
2409 static void trace__set_base_time(struct trace *trace,
2410 struct perf_evsel *evsel,
2411 struct perf_sample *sample)
2414 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2415 * and don't use sample->time unconditionally, we may end up having
2416 * some other event in the future without PERF_SAMPLE_TIME for good
2417 * reason, i.e. we may not be interested in its timestamps, just in
2418 * it taking place, picking some piece of information when it
2419 * appears in our event stream (vfs_getname comes to mind).
2421 if (trace->base_time == 0 && !trace->full_time &&
2422 (evsel->attr.sample_type & PERF_SAMPLE_TIME))
2423 trace->base_time = sample->time;
2426 static int trace__process_sample(struct perf_tool *tool,
2427 union perf_event *event,
2428 struct perf_sample *sample,
2429 struct perf_evsel *evsel,
2430 struct machine *machine __maybe_unused)
2432 struct trace *trace = container_of(tool, struct trace, tool);
2433 struct thread *thread;
2436 tracepoint_handler handler = evsel->handler;
2438 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2439 if (thread && thread__is_filtered(thread))
2442 trace__set_base_time(trace, evsel, sample);
2446 handler(trace, evsel, event, sample);
2449 thread__put(thread);
2453 static int trace__record(struct trace *trace, int argc, const char **argv)
2455 unsigned int rec_argc, i, j;
2456 const char **rec_argv;
2457 const char * const record_args[] = {
2464 const char * const sc_args[] = { "-e", };
2465 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2466 const char * const majpf_args[] = { "-e", "major-faults" };
2467 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2468 const char * const minpf_args[] = { "-e", "minor-faults" };
2469 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2471 /* +1 is for the event string below */
2472 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2473 majpf_args_nr + minpf_args_nr + argc;
2474 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2476 if (rec_argv == NULL)
2480 for (i = 0; i < ARRAY_SIZE(record_args); i++)
2481 rec_argv[j++] = record_args[i];
2483 if (trace->trace_syscalls) {
2484 for (i = 0; i < sc_args_nr; i++)
2485 rec_argv[j++] = sc_args[i];
2487 /* event string may be different for older kernels - e.g., RHEL6 */
2488 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2489 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2490 else if (is_valid_tracepoint("syscalls:sys_enter"))
2491 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2493 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2499 if (trace->trace_pgfaults & TRACE_PFMAJ)
2500 for (i = 0; i < majpf_args_nr; i++)
2501 rec_argv[j++] = majpf_args[i];
2503 if (trace->trace_pgfaults & TRACE_PFMIN)
2504 for (i = 0; i < minpf_args_nr; i++)
2505 rec_argv[j++] = minpf_args[i];
2507 for (i = 0; i < (unsigned int)argc; i++)
2508 rec_argv[j++] = argv[i];
2510 return cmd_record(j, rec_argv);
2513 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2515 static bool perf_evlist__add_vfs_getname(struct perf_evlist *evlist)
2518 struct perf_evsel *evsel, *tmp;
2519 struct parse_events_error err = { .idx = 0, };
2520 int ret = parse_events(evlist, "probe:vfs_getname*", &err);
2525 evlist__for_each_entry_safe(evlist, evsel, tmp) {
2526 if (!strstarts(perf_evsel__name(evsel), "probe:vfs_getname"))
2529 if (perf_evsel__field(evsel, "pathname")) {
2530 evsel->handler = trace__vfs_getname;
2535 list_del_init(&evsel->node);
2536 evsel->evlist = NULL;
2537 perf_evsel__delete(evsel);
2543 static struct perf_evsel *perf_evsel__new_pgfault(u64 config)
2545 struct perf_evsel *evsel;
2546 struct perf_event_attr attr = {
2547 .type = PERF_TYPE_SOFTWARE,
2551 attr.config = config;
2552 attr.sample_period = 1;
2554 event_attr_init(&attr);
2556 evsel = perf_evsel__new(&attr);
2558 evsel->handler = trace__pgfault;
2563 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
2565 const u32 type = event->header.type;
2566 struct perf_evsel *evsel;
2568 if (type != PERF_RECORD_SAMPLE) {
2569 trace__process_event(trace, trace->host, event, sample);
2573 evsel = perf_evlist__id2evsel(trace->evlist, sample->id);
2574 if (evsel == NULL) {
2575 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
2579 trace__set_base_time(trace, evsel, sample);
2581 if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
2582 sample->raw_data == NULL) {
2583 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2584 perf_evsel__name(evsel), sample->tid,
2585 sample->cpu, sample->raw_size);
2587 tracepoint_handler handler = evsel->handler;
2588 handler(trace, evsel, event, sample);
2591 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
2595 static int trace__add_syscall_newtp(struct trace *trace)
2598 struct perf_evlist *evlist = trace->evlist;
2599 struct perf_evsel *sys_enter, *sys_exit;
2601 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
2602 if (sys_enter == NULL)
2605 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
2606 goto out_delete_sys_enter;
2608 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
2609 if (sys_exit == NULL)
2610 goto out_delete_sys_enter;
2612 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
2613 goto out_delete_sys_exit;
2615 perf_evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
2616 perf_evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
2618 perf_evlist__add(evlist, sys_enter);
2619 perf_evlist__add(evlist, sys_exit);
2621 if (callchain_param.enabled && !trace->kernel_syscallchains) {
2623 * We're interested only in the user space callchain
2624 * leading to the syscall, allow overriding that for
2625 * debugging reasons using --kernel_syscall_callchains
2627 sys_exit->attr.exclude_callchain_kernel = 1;
2630 trace->syscalls.events.sys_enter = sys_enter;
2631 trace->syscalls.events.sys_exit = sys_exit;
2637 out_delete_sys_exit:
2638 perf_evsel__delete_priv(sys_exit);
2639 out_delete_sys_enter:
2640 perf_evsel__delete_priv(sys_enter);
2644 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
2647 struct perf_evsel *sys_exit;
2648 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
2649 trace->ev_qualifier_ids.nr,
2650 trace->ev_qualifier_ids.entries);
2655 if (!perf_evsel__append_tp_filter(trace->syscalls.events.sys_enter,
2657 sys_exit = trace->syscalls.events.sys_exit;
2658 err = perf_evsel__append_tp_filter(sys_exit, filter);
2669 #ifdef HAVE_LIBBPF_SUPPORT
2670 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
2672 int fd = bpf_map__fd(trace->syscalls.map);
2673 struct bpf_map_syscall_entry value = {
2674 .enabled = !trace->not_ev_qualifier,
2679 for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
2680 int key = trace->ev_qualifier_ids.entries[i];
2682 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
2690 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
2692 int fd = bpf_map__fd(trace->syscalls.map);
2693 struct bpf_map_syscall_entry value = {
2698 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
2699 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
2707 static int trace__init_syscalls_bpf_map(struct trace *trace)
2709 bool enabled = true;
2711 if (trace->ev_qualifier_ids.nr)
2712 enabled = trace->not_ev_qualifier;
2714 return __trace__init_syscalls_bpf_map(trace, enabled);
2717 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
2722 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
2726 #endif // HAVE_LIBBPF_SUPPORT
2728 static int trace__set_ev_qualifier_filter(struct trace *trace)
2730 if (trace->syscalls.map)
2731 return trace__set_ev_qualifier_bpf_filter(trace);
2732 if (trace->syscalls.events.sys_enter)
2733 return trace__set_ev_qualifier_tp_filter(trace);
2737 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
2738 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
2741 #ifdef HAVE_LIBBPF_SUPPORT
2743 int map_fd = bpf_map__fd(map);
2746 for (i = 0; i < npids; ++i) {
2747 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
2755 static int trace__set_filter_loop_pids(struct trace *trace)
2757 unsigned int nr = 1, err;
2761 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
2763 while (thread && nr < ARRAY_SIZE(pids)) {
2764 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
2769 if (!strcmp(thread__comm_str(parent), "sshd")) {
2770 pids[nr++] = parent->tid;
2776 err = perf_evlist__set_tp_filter_pids(trace->evlist, nr, pids);
2777 if (!err && trace->filter_pids.map)
2778 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
2783 static int trace__set_filter_pids(struct trace *trace)
2787 * Better not use !target__has_task() here because we need to cover the
2788 * case where no threads were specified in the command line, but a
2789 * workload was, and in that case we will fill in the thread_map when
2790 * we fork the workload in perf_evlist__prepare_workload.
2792 if (trace->filter_pids.nr > 0) {
2793 err = perf_evlist__set_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
2794 trace->filter_pids.entries);
2795 if (!err && trace->filter_pids.map) {
2796 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
2797 trace->filter_pids.entries);
2799 } else if (thread_map__pid(trace->evlist->threads, 0) == -1) {
2800 err = trace__set_filter_loop_pids(trace);
2806 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
2808 struct perf_evlist *evlist = trace->evlist;
2809 struct perf_sample sample;
2812 err = perf_evlist__parse_sample(evlist, event, &sample);
2814 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
2816 trace__handle_event(trace, event, &sample);
2821 static int __trace__flush_events(struct trace *trace)
2823 u64 first = ordered_events__first_time(&trace->oe.data);
2824 u64 flush = trace->oe.last - NSEC_PER_SEC;
2826 /* Is there some thing to flush.. */
2827 if (first && first < flush)
2828 return ordered_events__flush_time(&trace->oe.data, flush);
2833 static int trace__flush_events(struct trace *trace)
2835 return !trace->sort_events ? 0 : __trace__flush_events(trace);
2838 static int trace__deliver_event(struct trace *trace, union perf_event *event)
2842 if (!trace->sort_events)
2843 return __trace__deliver_event(trace, event);
2845 err = perf_evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
2846 if (err && err != -1)
2849 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0);
2853 return trace__flush_events(trace);
2856 static int ordered_events__deliver_event(struct ordered_events *oe,
2857 struct ordered_event *event)
2859 struct trace *trace = container_of(oe, struct trace, oe.data);
2861 return __trace__deliver_event(trace, event->event);
2864 static int trace__run(struct trace *trace, int argc, const char **argv)
2866 struct perf_evlist *evlist = trace->evlist;
2867 struct perf_evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
2869 unsigned long before;
2870 const bool forks = argc > 0;
2871 bool draining = false;
2875 if (!trace->raw_augmented_syscalls) {
2876 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
2877 goto out_error_raw_syscalls;
2879 if (trace->trace_syscalls)
2880 trace->vfs_getname = perf_evlist__add_vfs_getname(evlist);
2883 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
2884 pgfault_maj = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
2885 if (pgfault_maj == NULL)
2887 perf_evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
2888 perf_evlist__add(evlist, pgfault_maj);
2891 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
2892 pgfault_min = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
2893 if (pgfault_min == NULL)
2895 perf_evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
2896 perf_evlist__add(evlist, pgfault_min);
2900 perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
2901 trace__sched_stat_runtime))
2902 goto out_error_sched_stat_runtime;
2905 * If a global cgroup was set, apply it to all the events without an
2906 * explicit cgroup. I.e.:
2908 * trace -G A -e sched:*switch
2910 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
2911 * _and_ sched:sched_switch to the 'A' cgroup, while:
2913 * trace -e sched:*switch -G A
2915 * will only set the sched:sched_switch event to the 'A' cgroup, all the
2916 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
2917 * a cgroup (on the root cgroup, sys wide, etc).
2921 * trace -G A -e sched:*switch -G B
2923 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
2924 * to the 'B' cgroup.
2926 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
2927 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
2930 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
2932 err = perf_evlist__create_maps(evlist, &trace->opts.target);
2934 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
2935 goto out_delete_evlist;
2938 err = trace__symbols_init(trace, evlist);
2940 fprintf(trace->output, "Problems initializing symbol libraries!\n");
2941 goto out_delete_evlist;
2944 perf_evlist__config(evlist, &trace->opts, &callchain_param);
2946 signal(SIGCHLD, sig_handler);
2947 signal(SIGINT, sig_handler);
2950 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
2953 fprintf(trace->output, "Couldn't run the workload!\n");
2954 goto out_delete_evlist;
2958 err = perf_evlist__open(evlist);
2960 goto out_error_open;
2962 err = bpf__apply_obj_config();
2964 char errbuf[BUFSIZ];
2966 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
2967 pr_err("ERROR: Apply config to BPF failed: %s\n",
2969 goto out_error_open;
2972 err = trace__set_filter_pids(trace);
2976 if (trace->syscalls.map)
2977 trace__init_syscalls_bpf_map(trace);
2979 if (trace->ev_qualifier_ids.nr > 0) {
2980 err = trace__set_ev_qualifier_filter(trace);
2984 if (trace->syscalls.events.sys_exit) {
2985 pr_debug("event qualifier tracepoint filter: %s\n",
2986 trace->syscalls.events.sys_exit->filter);
2990 err = perf_evlist__apply_filters(evlist, &evsel);
2992 goto out_error_apply_filters;
2994 err = perf_evlist__mmap(evlist, trace->opts.mmap_pages);
2996 goto out_error_mmap;
2998 if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
2999 perf_evlist__enable(evlist);
3002 perf_evlist__start_workload(evlist);
3004 if (trace->opts.initial_delay) {
3005 usleep(trace->opts.initial_delay * 1000);
3006 perf_evlist__enable(evlist);
3009 trace->multiple_threads = thread_map__pid(evlist->threads, 0) == -1 ||
3010 evlist->threads->nr > 1 ||
3011 perf_evlist__first(evlist)->attr.inherit;
3014 * Now that we already used evsel->attr to ask the kernel to setup the
3015 * events, lets reuse evsel->attr.sample_max_stack as the limit in
3016 * trace__resolve_callchain(), allowing per-event max-stack settings
3017 * to override an explicitly set --max-stack global setting.
3019 evlist__for_each_entry(evlist, evsel) {
3020 if (evsel__has_callchain(evsel) &&
3021 evsel->attr.sample_max_stack == 0)
3022 evsel->attr.sample_max_stack = trace->max_stack;
3025 before = trace->nr_events;
3027 for (i = 0; i < evlist->nr_mmaps; i++) {
3028 union perf_event *event;
3029 struct perf_mmap *md;
3031 md = &evlist->mmap[i];
3032 if (perf_mmap__read_init(md) < 0)
3035 while ((event = perf_mmap__read_event(md)) != NULL) {
3038 err = trace__deliver_event(trace, event);
3042 perf_mmap__consume(md);
3047 if (done && !draining) {
3048 perf_evlist__disable(evlist);
3052 perf_mmap__read_done(md);
3055 if (trace->nr_events == before) {
3056 int timeout = done ? 100 : -1;
3058 if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
3059 if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
3064 if (trace__flush_events(trace))
3072 thread__zput(trace->current);
3074 perf_evlist__disable(evlist);
3076 if (trace->sort_events)
3077 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
3081 trace__fprintf_thread_summary(trace, trace->output);
3083 if (trace->show_tool_stats) {
3084 fprintf(trace->output, "Stats:\n "
3085 " vfs_getname : %" PRIu64 "\n"
3086 " proc_getname: %" PRIu64 "\n",
3087 trace->stats.vfs_getname,
3088 trace->stats.proc_getname);
3093 trace__symbols__exit(trace);
3095 perf_evlist__delete(evlist);
3096 cgroup__put(trace->cgroup);
3097 trace->evlist = NULL;
3098 trace->live = false;
3101 char errbuf[BUFSIZ];
3103 out_error_sched_stat_runtime:
3104 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
3107 out_error_raw_syscalls:
3108 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
3112 perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
3116 perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
3119 fprintf(trace->output, "%s\n", errbuf);
3120 goto out_delete_evlist;
3122 out_error_apply_filters:
3123 fprintf(trace->output,
3124 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
3125 evsel->filter, perf_evsel__name(evsel), errno,
3126 str_error_r(errno, errbuf, sizeof(errbuf)));
3127 goto out_delete_evlist;
3130 fprintf(trace->output, "Not enough memory to run!\n");
3131 goto out_delete_evlist;
3134 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
3135 goto out_delete_evlist;
3138 static int trace__replay(struct trace *trace)
3140 const struct perf_evsel_str_handler handlers[] = {
3141 { "probe:vfs_getname", trace__vfs_getname, },
3143 struct perf_data data = {
3147 .mode = PERF_DATA_MODE_READ,
3148 .force = trace->force,
3150 struct perf_session *session;
3151 struct perf_evsel *evsel;
3154 trace->tool.sample = trace__process_sample;
3155 trace->tool.mmap = perf_event__process_mmap;
3156 trace->tool.mmap2 = perf_event__process_mmap2;
3157 trace->tool.comm = perf_event__process_comm;
3158 trace->tool.exit = perf_event__process_exit;
3159 trace->tool.fork = perf_event__process_fork;
3160 trace->tool.attr = perf_event__process_attr;
3161 trace->tool.tracing_data = perf_event__process_tracing_data;
3162 trace->tool.build_id = perf_event__process_build_id;
3163 trace->tool.namespaces = perf_event__process_namespaces;
3165 trace->tool.ordered_events = true;
3166 trace->tool.ordering_requires_timestamps = true;
3168 /* add tid to output */
3169 trace->multiple_threads = true;
3171 session = perf_session__new(&data, false, &trace->tool);
3172 if (session == NULL)
3175 if (trace->opts.target.pid)
3176 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
3178 if (trace->opts.target.tid)
3179 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
3181 if (symbol__init(&session->header.env) < 0)
3184 trace->host = &session->machines.host;
3186 err = perf_session__set_tracepoints_handlers(session, handlers);
3190 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3191 "raw_syscalls:sys_enter");
3192 /* older kernels have syscalls tp versus raw_syscalls */
3194 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3195 "syscalls:sys_enter");
3198 (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
3199 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
3200 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
3204 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3205 "raw_syscalls:sys_exit");
3207 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3208 "syscalls:sys_exit");
3210 (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
3211 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
3212 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
3216 evlist__for_each_entry(session->evlist, evsel) {
3217 if (evsel->attr.type == PERF_TYPE_SOFTWARE &&
3218 (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
3219 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
3220 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS))
3221 evsel->handler = trace__pgfault;
3226 err = perf_session__process_events(session);
3228 pr_err("Failed to process events, error %d", err);
3230 else if (trace->summary)
3231 trace__fprintf_thread_summary(trace, trace->output);
3234 perf_session__delete(session);
3239 static size_t trace__fprintf_threads_header(FILE *fp)
3243 printed = fprintf(fp, "\n Summary of events:\n\n");
3248 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
3249 struct stats *stats;
3254 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
3255 struct stats *stats = source->priv;
3257 entry->syscall = source->i;
3258 entry->stats = stats;
3259 entry->msecs = stats ? (u64)stats->n * (avg_stats(stats) / NSEC_PER_MSEC) : 0;
3262 static size_t thread__dump_stats(struct thread_trace *ttrace,
3263 struct trace *trace, FILE *fp)
3268 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
3270 if (syscall_stats == NULL)
3273 printed += fprintf(fp, "\n");
3275 printed += fprintf(fp, " syscall calls total min avg max stddev\n");
3276 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
3277 printed += fprintf(fp, " --------------- -------- --------- --------- --------- --------- ------\n");
3279 resort_rb__for_each_entry(nd, syscall_stats) {
3280 struct stats *stats = syscall_stats_entry->stats;
3282 double min = (double)(stats->min) / NSEC_PER_MSEC;
3283 double max = (double)(stats->max) / NSEC_PER_MSEC;
3284 double avg = avg_stats(stats);
3286 u64 n = (u64) stats->n;
3288 pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
3289 avg /= NSEC_PER_MSEC;
3291 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
3292 printed += fprintf(fp, " %-15s", sc->name);
3293 printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f %9.3f",
3294 n, syscall_stats_entry->msecs, min, avg);
3295 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
3299 resort_rb__delete(syscall_stats);
3300 printed += fprintf(fp, "\n\n");
3305 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
3308 struct thread_trace *ttrace = thread__priv(thread);
3314 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
3316 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
3317 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
3318 printed += fprintf(fp, "%.1f%%", ratio);
3320 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
3322 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
3324 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
3325 else if (fputc('\n', fp) != EOF)
3328 printed += thread__dump_stats(ttrace, trace, fp);
3333 static unsigned long thread__nr_events(struct thread_trace *ttrace)
3335 return ttrace ? ttrace->nr_events : 0;
3338 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
3339 struct thread *thread;
3342 entry->thread = rb_entry(nd, struct thread, rb_node);
3345 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
3347 size_t printed = trace__fprintf_threads_header(fp);
3351 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3352 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
3354 if (threads == NULL) {
3355 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
3359 resort_rb__for_each_entry(nd, threads)
3360 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
3362 resort_rb__delete(threads);
3367 static int trace__set_duration(const struct option *opt, const char *str,
3368 int unset __maybe_unused)
3370 struct trace *trace = opt->value;
3372 trace->duration_filter = atof(str);
3376 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
3377 int unset __maybe_unused)
3381 struct trace *trace = opt->value;
3383 * FIXME: introduce a intarray class, plain parse csv and create a
3384 * { int nr, int entries[] } struct...
3386 struct intlist *list = intlist__new(str);
3391 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
3392 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
3394 if (trace->filter_pids.entries == NULL)
3397 trace->filter_pids.entries[0] = getpid();
3399 for (i = 1; i < trace->filter_pids.nr; ++i)
3400 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
3402 intlist__delete(list);
3408 static int trace__open_output(struct trace *trace, const char *filename)
3412 if (!stat(filename, &st) && st.st_size) {
3413 char oldname[PATH_MAX];
3415 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
3417 rename(filename, oldname);
3420 trace->output = fopen(filename, "w");
3422 return trace->output == NULL ? -errno : 0;
3425 static int parse_pagefaults(const struct option *opt, const char *str,
3426 int unset __maybe_unused)
3428 int *trace_pgfaults = opt->value;
3430 if (strcmp(str, "all") == 0)
3431 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
3432 else if (strcmp(str, "maj") == 0)
3433 *trace_pgfaults |= TRACE_PFMAJ;
3434 else if (strcmp(str, "min") == 0)
3435 *trace_pgfaults |= TRACE_PFMIN;
3442 static void evlist__set_evsel_handler(struct perf_evlist *evlist, void *handler)
3444 struct perf_evsel *evsel;
3446 evlist__for_each_entry(evlist, evsel)
3447 evsel->handler = handler;
3450 static int evlist__set_syscall_tp_fields(struct perf_evlist *evlist)
3452 struct perf_evsel *evsel;
3454 evlist__for_each_entry(evlist, evsel) {
3455 if (evsel->priv || !evsel->tp_format)
3458 if (strcmp(evsel->tp_format->system, "syscalls"))
3461 if (perf_evsel__init_syscall_tp(evsel))
3464 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
3465 struct syscall_tp *sc = evsel->priv;
3467 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
3469 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
3470 struct syscall_tp *sc = evsel->priv;
3472 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
3481 * XXX: Hackish, just splitting the combined -e+--event (syscalls
3482 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
3483 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
3485 * It'd be better to introduce a parse_options() variant that would return a
3486 * list with the terms it didn't match to an event...
3488 static int trace__parse_events_option(const struct option *opt, const char *str,
3489 int unset __maybe_unused)
3491 struct trace *trace = (struct trace *)opt->value;
3492 const char *s = str;
3493 char *sep = NULL, *lists[2] = { NULL, NULL, };
3494 int len = strlen(str) + 1, err = -1, list, idx;
3495 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
3496 char group_name[PATH_MAX];
3497 struct syscall_fmt *fmt;
3499 if (strace_groups_dir == NULL)
3504 trace->not_ev_qualifier = true;
3508 if ((sep = strchr(s, ',')) != NULL)
3512 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
3513 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
3518 fmt = syscall_fmt__find_by_alias(s);
3523 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
3524 if (access(group_name, R_OK) == 0)
3529 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
3531 lists[list] = malloc(len);
3532 if (lists[list] == NULL)
3534 strcpy(lists[list], s);
3544 if (lists[1] != NULL) {
3545 struct strlist_config slist_config = {
3546 .dirname = strace_groups_dir,
3549 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
3550 if (trace->ev_qualifier == NULL) {
3551 fputs("Not enough memory to parse event qualifier", trace->output);
3555 if (trace__validate_ev_qualifier(trace))
3557 trace->trace_syscalls = true;
3563 struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
3564 "event selector. use 'perf list' to list available events",
3565 parse_events_option);
3566 err = parse_events_option(&o, lists[0], 0);
3575 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
3577 struct trace *trace = opt->value;
3579 if (!list_empty(&trace->evlist->entries))
3580 return parse_cgroups(opt, str, unset);
3582 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
3587 static struct bpf_map *bpf__find_map_by_name(const char *name)
3589 struct bpf_object *obj, *tmp;
3591 bpf_object__for_each_safe(obj, tmp) {
3592 struct bpf_map *map = bpf_object__find_map_by_name(obj, name);
3601 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3603 trace->filter_pids.map = bpf__find_map_by_name("pids_filtered");
3606 static void trace__set_bpf_map_syscalls(struct trace *trace)
3608 trace->syscalls.map = bpf__find_map_by_name("syscalls");
3611 static int trace__config(const char *var, const char *value, void *arg)
3613 struct trace *trace = arg;
3616 if (!strcmp(var, "trace.add_events")) {
3617 struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
3618 "event selector. use 'perf list' to list available events",
3619 parse_events_option);
3620 err = parse_events_option(&o, value, 0);
3621 } else if (!strcmp(var, "trace.show_timestamp")) {
3622 trace->show_tstamp = perf_config_bool(var, value);
3623 } else if (!strcmp(var, "trace.show_duration")) {
3624 trace->show_duration = perf_config_bool(var, value);
3625 } else if (!strcmp(var, "trace.show_arg_names")) {
3626 trace->show_arg_names = perf_config_bool(var, value);
3627 if (!trace->show_arg_names)
3628 trace->show_zeros = true;
3629 } else if (!strcmp(var, "trace.show_zeros")) {
3630 bool new_show_zeros = perf_config_bool(var, value);
3631 if (!trace->show_arg_names && !new_show_zeros) {
3632 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
3635 trace->show_zeros = new_show_zeros;
3636 } else if (!strcmp(var, "trace.show_prefix")) {
3637 trace->show_string_prefix = perf_config_bool(var, value);
3638 } else if (!strcmp(var, "trace.no_inherit")) {
3639 trace->opts.no_inherit = perf_config_bool(var, value);
3640 } else if (!strcmp(var, "trace.args_alignment")) {
3641 int args_alignment = 0;
3642 if (perf_config_int(&args_alignment, var, value) == 0)
3643 trace->args_alignment = args_alignment;
3649 int cmd_trace(int argc, const char **argv)
3651 const char *trace_usage[] = {
3652 "perf trace [<options>] [<command>]",
3653 "perf trace [<options>] -- <command> [<options>]",
3654 "perf trace record [<options>] [<command>]",
3655 "perf trace record [<options>] -- <command> [<options>]",
3658 struct trace trace = {
3667 .user_freq = UINT_MAX,
3668 .user_interval = ULLONG_MAX,
3669 .no_buffering = true,
3670 .mmap_pages = UINT_MAX,
3674 .show_tstamp = true,
3675 .show_duration = true,
3676 .show_arg_names = true,
3677 .args_alignment = 70,
3678 .trace_syscalls = false,
3679 .kernel_syscallchains = false,
3680 .max_stack = UINT_MAX,
3681 .max_events = ULONG_MAX,
3683 const char *output_name = NULL;
3684 const struct option trace_options[] = {
3685 OPT_CALLBACK('e', "event", &trace, "event",
3686 "event/syscall selector. use 'perf list' to list available events",
3687 trace__parse_events_option),
3688 OPT_BOOLEAN(0, "comm", &trace.show_comm,
3689 "show the thread COMM next to its id"),
3690 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
3691 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
3692 trace__parse_events_option),
3693 OPT_STRING('o', "output", &output_name, "file", "output file name"),
3694 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
3695 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
3696 "trace events on existing process id"),
3697 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
3698 "trace events on existing thread id"),
3699 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
3700 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
3701 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
3702 "system-wide collection from all CPUs"),
3703 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
3704 "list of cpus to monitor"),
3705 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
3706 "child tasks do not inherit counters"),
3707 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
3708 "number of mmap data pages",
3709 perf_evlist__parse_mmap_pages),
3710 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
3712 OPT_CALLBACK(0, "duration", &trace, "float",
3713 "show only events with duration > N.M ms",
3714 trace__set_duration),
3715 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
3716 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
3717 OPT_BOOLEAN('T', "time", &trace.full_time,
3718 "Show full timestamp, not time relative to first start"),
3719 OPT_BOOLEAN(0, "failure", &trace.failure_only,
3720 "Show only syscalls that failed"),
3721 OPT_BOOLEAN('s', "summary", &trace.summary_only,
3722 "Show only syscall summary with statistics"),
3723 OPT_BOOLEAN('S', "with-summary", &trace.summary,
3724 "Show all syscalls and summary with statistics"),
3725 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
3726 "Trace pagefaults", parse_pagefaults, "maj"),
3727 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
3728 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
3729 OPT_CALLBACK(0, "call-graph", &trace.opts,
3730 "record_mode[,record_size]", record_callchain_help,
3731 &record_parse_callchain_opt),
3732 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
3733 "Show the kernel callchains on the syscall exit path"),
3734 OPT_ULONG(0, "max-events", &trace.max_events,
3735 "Set the maximum number of events to print, exit after that is reached. "),
3736 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
3737 "Set the minimum stack depth when parsing the callchain, "
3738 "anything below the specified depth will be ignored."),
3739 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
3740 "Set the maximum stack depth when parsing the callchain, "
3741 "anything beyond the specified depth will be ignored. "
3742 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
3743 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
3744 "Sort batch of events before processing, use if getting out of order events"),
3745 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
3746 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
3747 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
3748 "per thread proc mmap processing timeout in ms"),
3749 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
3750 trace__parse_cgroups),
3751 OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
3752 "ms to wait before starting measurement after program "
3756 bool __maybe_unused max_stack_user_set = true;
3757 bool mmap_pages_user_set = true;
3758 struct perf_evsel *evsel;
3759 const char * const trace_subcommands[] = { "record", NULL };
3763 signal(SIGSEGV, sighandler_dump_stack);
3764 signal(SIGFPE, sighandler_dump_stack);
3766 trace.evlist = perf_evlist__new();
3767 trace.sctbl = syscalltbl__new();
3769 if (trace.evlist == NULL || trace.sctbl == NULL) {
3770 pr_err("Not enough memory to run!\n");
3775 err = perf_config(trace__config, &trace);
3779 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
3780 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
3782 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
3783 usage_with_options_msg(trace_usage, trace_options,
3784 "cgroup monitoring only available in system-wide mode");
3787 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
3788 if (IS_ERR(evsel)) {
3789 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
3790 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
3795 trace.syscalls.events.augmented = evsel;
3796 trace__set_bpf_map_filtered_pids(&trace);
3797 trace__set_bpf_map_syscalls(&trace);
3800 err = bpf__setup_stdout(trace.evlist);
3802 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
3803 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
3809 if (trace.trace_pgfaults) {
3810 trace.opts.sample_address = true;
3811 trace.opts.sample_time = true;
3814 if (trace.opts.mmap_pages == UINT_MAX)
3815 mmap_pages_user_set = false;
3817 if (trace.max_stack == UINT_MAX) {
3818 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
3819 max_stack_user_set = false;
3822 #ifdef HAVE_DWARF_UNWIND_SUPPORT
3823 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
3824 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
3828 if (callchain_param.enabled) {
3829 if (!mmap_pages_user_set && geteuid() == 0)
3830 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
3832 symbol_conf.use_callchain = true;
3835 if (trace.evlist->nr_entries > 0) {
3836 evlist__set_evsel_handler(trace.evlist, trace__event_handler);
3837 if (evlist__set_syscall_tp_fields(trace.evlist)) {
3838 perror("failed to set syscalls:* tracepoint fields");
3843 if (trace.sort_events) {
3844 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
3845 ordered_events__set_copy_on_queue(&trace.oe.data, true);
3849 * If we are augmenting syscalls, then combine what we put in the
3850 * __augmented_syscalls__ BPF map with what is in the
3851 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
3852 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
3854 * We'll switch to look at two BPF maps, one for sys_enter and the
3855 * other for sys_exit when we start augmenting the sys_exit paths with
3856 * buffers that are being copied from kernel to userspace, think 'read'
3859 if (trace.syscalls.events.augmented) {
3860 evlist__for_each_entry(trace.evlist, evsel) {
3861 bool raw_syscalls_sys_exit = strcmp(perf_evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
3863 if (raw_syscalls_sys_exit) {
3864 trace.raw_augmented_syscalls = true;
3865 goto init_augmented_syscall_tp;
3868 if (strcmp(perf_evsel__name(evsel), "raw_syscalls:sys_enter") == 0) {
3869 struct perf_evsel *augmented = trace.syscalls.events.augmented;
3870 if (perf_evsel__init_augmented_syscall_tp(augmented, evsel) ||
3871 perf_evsel__init_augmented_syscall_tp_args(augmented))
3873 augmented->handler = trace__sys_enter;
3876 if (strstarts(perf_evsel__name(evsel), "syscalls:sys_exit_")) {
3877 struct syscall_tp *sc;
3878 init_augmented_syscall_tp:
3879 if (perf_evsel__init_augmented_syscall_tp(evsel, evsel))
3883 * For now with BPF raw_augmented we hook into
3884 * raw_syscalls:sys_enter and there we get all
3885 * 6 syscall args plus the tracepoint common
3886 * fields and the syscall_nr (another long).
3887 * So we check if that is the case and if so
3888 * don't look after the sc->args_size but
3889 * always after the full raw_syscalls:sys_enter
3890 * payload, which is fixed.
3892 * We'll revisit this later to pass
3893 * s->args_size to the BPF augmenter (now
3894 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
3895 * so that it copies only what we need for each
3896 * syscall, like what happens when we use
3897 * syscalls:sys_enter_NAME, so that we reduce
3898 * the kernel/userspace traffic to just what is
3899 * needed for each syscall.
3901 if (trace.raw_augmented_syscalls)
3902 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
3903 perf_evsel__init_augmented_syscall_tp_ret(evsel);
3904 evsel->handler = trace__sys_exit;
3909 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
3910 return trace__record(&trace, argc-1, &argv[1]);
3912 /* summary_only implies summary option, but don't overwrite summary if set */
3913 if (trace.summary_only)
3914 trace.summary = trace.summary_only;
3916 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
3917 trace.evlist->nr_entries == 0 /* Was --events used? */) {
3918 trace.trace_syscalls = true;
3921 if (output_name != NULL) {
3922 err = trace__open_output(&trace, output_name);
3924 perror("failed to create output file");
3929 err = target__validate(&trace.opts.target);
3931 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
3932 fprintf(trace.output, "%s", bf);
3936 err = target__parse_uid(&trace.opts.target);
3938 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
3939 fprintf(trace.output, "%s", bf);
3943 if (!argc && target__none(&trace.opts.target))
3944 trace.opts.target.system_wide = true;
3947 err = trace__replay(&trace);
3949 err = trace__run(&trace, argc, argv);
3952 if (output_name != NULL)
3953 fclose(trace.output);