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'")
17 #include "util/record.h"
18 #include <traceevent/event-parse.h>
19 #include <api/fs/tracing_path.h>
21 #include "util/bpf_map.h"
22 #include "util/rlimit.h"
24 #include "util/cgroup.h"
25 #include "util/color.h"
26 #include "util/config.h"
27 #include "util/debug.h"
30 #include "util/event.h"
31 #include "util/evsel.h"
32 #include "util/evsel_fprintf.h"
33 #include "util/synthetic-events.h"
34 #include "util/evlist.h"
35 #include "util/evswitch.h"
36 #include "util/mmap.h"
37 #include <subcmd/pager.h>
38 #include <subcmd/exec-cmd.h>
39 #include "util/machine.h"
41 #include "util/symbol.h"
42 #include "util/path.h"
43 #include "util/session.h"
44 #include "util/thread.h"
45 #include <subcmd/parse-options.h>
46 #include "util/strlist.h"
47 #include "util/intlist.h"
48 #include "util/thread_map.h"
49 #include "util/stat.h"
50 #include "util/tool.h"
51 #include "util/util.h"
52 #include "trace/beauty/beauty.h"
53 #include "trace-event.h"
54 #include "util/parse-events.h"
55 #include "util/bpf-loader.h"
56 #include "callchain.h"
57 #include "print_binary.h"
59 #include "syscalltbl.h"
60 #include "rb_resort.h"
69 #include <linux/err.h>
70 #include <linux/filter.h>
71 #include <linux/kernel.h>
72 #include <linux/random.h>
73 #include <linux/stringify.h>
74 #include <linux/time64.h>
75 #include <linux/zalloc.h>
77 #include <sys/sysmacros.h>
79 #include <linux/ctype.h>
80 #include <perf/mmap.h>
83 # define O_CLOEXEC 02000000
86 #ifndef F_LINUX_SPECIFIC_BASE
87 # define F_LINUX_SPECIFIC_BASE 1024
91 * strtoul: Go from a string to a value, i.e. for msr: MSR_FS_BASE to 0xc0000100
93 struct syscall_arg_fmt {
94 size_t (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
95 bool (*strtoul)(char *bf, size_t size, struct syscall_arg *arg, u64 *val);
96 unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
99 u16 nr_entries; // for arrays
107 const char *sys_enter,
110 struct syscall_arg_fmt arg[6];
118 struct perf_tool tool;
119 struct syscalltbl *sctbl;
121 struct syscall *table;
123 struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
124 struct bpf_map *sys_enter,
128 struct evsel *sys_enter,
132 struct bpf_program *unaugmented_prog;
137 struct record_opts opts;
138 struct evlist *evlist;
139 struct machine *host;
140 struct thread *current;
141 struct bpf_object *bpf_obj;
142 struct cgroup *cgroup;
145 unsigned long nr_events;
146 unsigned long nr_events_printed;
147 unsigned long max_events;
148 struct evswitch evswitch;
149 struct strlist *ev_qualifier;
159 double duration_filter;
165 unsigned int max_stack;
166 unsigned int min_stack;
167 int raw_augmented_syscalls_args_size;
168 bool raw_augmented_syscalls;
169 bool fd_path_disabled;
171 bool not_ev_qualifier;
175 bool multiple_threads;
182 bool show_tool_stats;
184 bool libtraceevent_print;
185 bool kernel_syscallchains;
191 bool show_string_prefix;
195 char *perfconfig_events;
197 struct ordered_events data;
205 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
206 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
210 #define TP_UINT_FIELD(bits) \
211 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
214 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
223 #define TP_UINT_FIELD__SWAPPED(bits) \
224 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
227 memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
228 return bswap_##bits(value);\
231 TP_UINT_FIELD__SWAPPED(16);
232 TP_UINT_FIELD__SWAPPED(32);
233 TP_UINT_FIELD__SWAPPED(64);
235 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
237 field->offset = offset;
241 field->integer = tp_field__u8;
244 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
247 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
250 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
259 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
261 return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
264 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
266 return sample->raw_data + field->offset;
269 static int __tp_field__init_ptr(struct tp_field *field, int offset)
271 field->offset = offset;
272 field->pointer = tp_field__ptr;
276 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
278 return __tp_field__init_ptr(field, format_field->offset);
284 struct tp_field args, ret;
289 * The evsel->priv as used by 'perf trace'
290 * sc: for raw_syscalls:sys_{enter,exit} and syscalls:sys_{enter,exit}_SYSCALLNAME
291 * fmt: for all the other tracepoints
294 struct syscall_tp sc;
295 struct syscall_arg_fmt *fmt;
298 static struct evsel_trace *evsel_trace__new(void)
300 return zalloc(sizeof(struct evsel_trace));
303 static void evsel_trace__delete(struct evsel_trace *et)
313 * Used with raw_syscalls:sys_{enter,exit} and with the
314 * syscalls:sys_{enter,exit}_SYSCALL tracepoints
316 static inline struct syscall_tp *__evsel__syscall_tp(struct evsel *evsel)
318 struct evsel_trace *et = evsel->priv;
323 static struct syscall_tp *evsel__syscall_tp(struct evsel *evsel)
325 if (evsel->priv == NULL) {
326 evsel->priv = evsel_trace__new();
327 if (evsel->priv == NULL)
331 return __evsel__syscall_tp(evsel);
335 * Used with all the other tracepoints.
337 static inline struct syscall_arg_fmt *__evsel__syscall_arg_fmt(struct evsel *evsel)
339 struct evsel_trace *et = evsel->priv;
344 static struct syscall_arg_fmt *evsel__syscall_arg_fmt(struct evsel *evsel)
346 struct evsel_trace *et = evsel->priv;
348 if (evsel->priv == NULL) {
349 et = evsel->priv = evsel_trace__new();
355 if (et->fmt == NULL) {
356 et->fmt = calloc(evsel->tp_format->format.nr_fields, sizeof(struct syscall_arg_fmt));
361 return __evsel__syscall_arg_fmt(evsel);
364 evsel_trace__delete(evsel->priv);
369 static int evsel__init_tp_uint_field(struct evsel *evsel, struct tp_field *field, const char *name)
371 struct tep_format_field *format_field = evsel__field(evsel, name);
373 if (format_field == NULL)
376 return tp_field__init_uint(field, format_field, evsel->needs_swap);
379 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
380 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
381 evsel__init_tp_uint_field(evsel, &sc->name, #name); })
383 static int evsel__init_tp_ptr_field(struct evsel *evsel, struct tp_field *field, const char *name)
385 struct tep_format_field *format_field = evsel__field(evsel, name);
387 if (format_field == NULL)
390 return tp_field__init_ptr(field, format_field);
393 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
394 ({ struct syscall_tp *sc = __evsel__syscall_tp(evsel);\
395 evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
397 static void evsel__delete_priv(struct evsel *evsel)
400 evsel__delete(evsel);
403 static int evsel__init_syscall_tp(struct evsel *evsel)
405 struct syscall_tp *sc = evsel__syscall_tp(evsel);
408 if (evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
409 evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
417 static int evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
419 struct syscall_tp *sc = evsel__syscall_tp(evsel);
422 struct tep_format_field *syscall_id = evsel__field(tp, "id");
423 if (syscall_id == NULL)
424 syscall_id = evsel__field(tp, "__syscall_nr");
425 if (syscall_id == NULL ||
426 __tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
435 static int evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
437 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
439 return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
442 static int evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
444 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
446 return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
449 static int evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
451 if (evsel__syscall_tp(evsel) != NULL) {
452 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
455 evsel->handler = handler;
462 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
464 struct evsel *evsel = evsel__newtp("raw_syscalls", direction);
466 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
468 evsel = evsel__newtp("syscalls", direction);
473 if (evsel__init_raw_syscall_tp(evsel, handler))
479 evsel__delete_priv(evsel);
483 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
484 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
485 fields->name.integer(&fields->name, sample); })
487 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
488 ({ struct syscall_tp *fields = __evsel__syscall_tp(evsel); \
489 fields->name.pointer(&fields->name, sample); })
491 size_t strarray__scnprintf_suffix(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_suffix, int val)
493 int idx = val - sa->offset;
495 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
496 size_t printed = scnprintf(bf, size, intfmt, val);
498 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
502 return scnprintf(bf, size, "%s%s", sa->entries[idx], show_suffix ? sa->prefix : "");
505 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
507 int idx = val - sa->offset;
509 if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
510 size_t printed = scnprintf(bf, size, intfmt, val);
512 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
516 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
519 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
521 struct syscall_arg *arg)
523 return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
526 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
527 struct syscall_arg *arg)
529 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
532 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
534 bool syscall_arg__strtoul_strarray(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
536 return strarray__strtoul(arg->parm, bf, size, ret);
539 bool syscall_arg__strtoul_strarray_flags(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
541 return strarray__strtoul_flags(arg->parm, bf, size, ret);
544 bool syscall_arg__strtoul_strarrays(char *bf, size_t size, struct syscall_arg *arg, u64 *ret)
546 return strarrays__strtoul(arg->parm, bf, size, ret);
549 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
551 return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
554 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
559 for (i = 0; i < sas->nr_entries; ++i) {
560 struct strarray *sa = sas->entries[i];
561 int idx = val - sa->offset;
563 if (idx >= 0 && idx < sa->nr_entries) {
564 if (sa->entries[idx] == NULL)
566 return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
570 printed = scnprintf(bf, size, intfmt, val);
572 printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
576 bool strarray__strtoul(struct strarray *sa, char *bf, size_t size, u64 *ret)
580 for (i = 0; i < sa->nr_entries; ++i) {
581 if (sa->entries[i] && strncmp(sa->entries[i], bf, size) == 0 && sa->entries[i][size] == '\0') {
582 *ret = sa->offset + i;
590 bool strarray__strtoul_flags(struct strarray *sa, char *bf, size_t size, u64 *ret)
593 char *tok = bf, *sep, *end;
600 sep = memchr(tok, '|', size);
602 size -= sep - tok + 1;
605 while (end > tok && isspace(*end))
608 toklen = end - tok + 1;
611 while (isspace(*tok))
614 if (isalpha(*tok) || *tok == '_') {
615 if (!strarray__strtoul(sa, tok, toklen, &val))
618 bool is_hexa = tok[0] == 0 && (tok[1] = 'x' || tok[1] == 'X');
620 val = strtoul(tok, NULL, is_hexa ? 16 : 0);
623 *ret |= (1 << (val - 1));
633 bool strarrays__strtoul(struct strarrays *sas, char *bf, size_t size, u64 *ret)
637 for (i = 0; i < sas->nr_entries; ++i) {
638 struct strarray *sa = sas->entries[i];
640 if (strarray__strtoul(sa, bf, size, ret))
647 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
648 struct syscall_arg *arg)
650 return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
654 #define AT_FDCWD -100
657 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
658 struct syscall_arg *arg)
661 const char *prefix = "AT_FD";
664 return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
666 return syscall_arg__scnprintf_fd(bf, size, arg);
669 #define SCA_FDAT syscall_arg__scnprintf_fd_at
671 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
672 struct syscall_arg *arg);
674 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
676 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
678 return scnprintf(bf, size, "%#lx", arg->val);
681 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
684 return scnprintf(bf, size, "NULL");
685 return syscall_arg__scnprintf_hex(bf, size, arg);
688 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
690 return scnprintf(bf, size, "%d", arg->val);
693 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
695 return scnprintf(bf, size, "%ld", arg->val);
698 static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
700 // XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
701 // fill missing comms using thread__set_comm()...
702 // here or in a special syscall_arg__scnprintf_pid_sched_tp...
703 return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries ?: arg->len, arg->val);
706 #define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
708 static const char *bpf_cmd[] = {
709 "MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
710 "MAP_GET_NEXT_KEY", "PROG_LOAD", "OBJ_PIN", "OBJ_GET", "PROG_ATTACH",
711 "PROG_DETACH", "PROG_TEST_RUN", "PROG_GET_NEXT_ID", "MAP_GET_NEXT_ID",
712 "PROG_GET_FD_BY_ID", "MAP_GET_FD_BY_ID", "OBJ_GET_INFO_BY_FD",
713 "PROG_QUERY", "RAW_TRACEPOINT_OPEN", "BTF_LOAD", "BTF_GET_FD_BY_ID",
714 "TASK_FD_QUERY", "MAP_LOOKUP_AND_DELETE_ELEM", "MAP_FREEZE",
715 "BTF_GET_NEXT_ID", "MAP_LOOKUP_BATCH", "MAP_LOOKUP_AND_DELETE_BATCH",
716 "MAP_UPDATE_BATCH", "MAP_DELETE_BATCH", "LINK_CREATE", "LINK_UPDATE",
717 "LINK_GET_FD_BY_ID", "LINK_GET_NEXT_ID", "ENABLE_STATS", "ITER_CREATE",
718 "LINK_DETACH", "PROG_BIND_MAP",
720 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
722 static const char *fsmount_flags[] = {
725 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
727 #include "trace/beauty/generated/fsconfig_arrays.c"
729 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
731 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
732 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
734 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
735 static DEFINE_STRARRAY(itimers, "ITIMER_");
737 static const char *keyctl_options[] = {
738 "GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
739 "SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
740 "INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
741 "ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
742 "INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
744 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
746 static const char *whences[] = { "SET", "CUR", "END",
754 static DEFINE_STRARRAY(whences, "SEEK_");
756 static const char *fcntl_cmds[] = {
757 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
758 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
759 "SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
762 static DEFINE_STRARRAY(fcntl_cmds, "F_");
764 static const char *fcntl_linux_specific_cmds[] = {
765 "SETLEASE", "GETLEASE", "NOTIFY", [5] = "CANCELLK", "DUPFD_CLOEXEC",
766 "SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
767 "GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
770 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
772 static struct strarray *fcntl_cmds_arrays[] = {
773 &strarray__fcntl_cmds,
774 &strarray__fcntl_linux_specific_cmds,
777 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
779 static const char *rlimit_resources[] = {
780 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
781 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
784 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
786 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
787 static DEFINE_STRARRAY(sighow, "SIG_");
789 static const char *clockid[] = {
790 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
791 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
792 "REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
794 static DEFINE_STRARRAY(clockid, "CLOCK_");
796 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
797 struct syscall_arg *arg)
799 bool show_prefix = arg->show_string_prefix;
800 const char *suffix = "_OK";
804 if (mode == F_OK) /* 0 */
805 return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
807 if (mode & n##_OK) { \
808 printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
818 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
823 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
825 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
826 struct syscall_arg *arg);
828 #define SCA_FILENAME syscall_arg__scnprintf_filename
830 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
831 struct syscall_arg *arg)
833 bool show_prefix = arg->show_string_prefix;
834 const char *prefix = "O_";
835 int printed = 0, flags = arg->val;
838 if (flags & O_##n) { \
839 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
848 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
853 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
855 #ifndef GRND_NONBLOCK
856 #define GRND_NONBLOCK 0x0001
859 #define GRND_RANDOM 0x0002
862 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
863 struct syscall_arg *arg)
865 bool show_prefix = arg->show_string_prefix;
866 const char *prefix = "GRND_";
867 int printed = 0, flags = arg->val;
870 if (flags & GRND_##n) { \
871 printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
872 flags &= ~GRND_##n; \
880 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
885 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
887 #define STRARRAY(name, array) \
888 { .scnprintf = SCA_STRARRAY, \
889 .strtoul = STUL_STRARRAY, \
890 .parm = &strarray__##array, }
892 #define STRARRAY_FLAGS(name, array) \
893 { .scnprintf = SCA_STRARRAY_FLAGS, \
894 .strtoul = STUL_STRARRAY_FLAGS, \
895 .parm = &strarray__##array, }
897 #include "trace/beauty/arch_errno_names.c"
898 #include "trace/beauty/eventfd.c"
899 #include "trace/beauty/futex_op.c"
900 #include "trace/beauty/futex_val3.c"
901 #include "trace/beauty/mmap.c"
902 #include "trace/beauty/mode_t.c"
903 #include "trace/beauty/msg_flags.c"
904 #include "trace/beauty/open_flags.c"
905 #include "trace/beauty/perf_event_open.c"
906 #include "trace/beauty/pid.c"
907 #include "trace/beauty/sched_policy.c"
908 #include "trace/beauty/seccomp.c"
909 #include "trace/beauty/signum.c"
910 #include "trace/beauty/socket_type.c"
911 #include "trace/beauty/waitid_options.c"
913 static struct syscall_fmt syscall_fmts[] = {
915 .arg = { [1] = { .scnprintf = SCA_ACCMODE, /* mode */ }, }, },
916 { .name = "arch_prctl",
917 .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
918 [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
920 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
921 [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
922 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
924 .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
925 { .name = "brk", .hexret = true,
926 .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
927 { .name = "clock_gettime",
928 .arg = { [0] = STRARRAY(clk_id, clockid), }, },
929 { .name = "clone", .errpid = true, .nr_args = 5,
930 .arg = { [0] = { .name = "flags", .scnprintf = SCA_CLONE_FLAGS, },
931 [1] = { .name = "child_stack", .scnprintf = SCA_HEX, },
932 [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
933 [3] = { .name = "child_tidptr", .scnprintf = SCA_HEX, },
934 [4] = { .name = "tls", .scnprintf = SCA_HEX, }, }, },
936 .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
938 .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
939 [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
940 [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
941 { .name = "epoll_ctl",
942 .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
943 { .name = "eventfd2",
944 .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
945 { .name = "fchmodat",
946 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
947 { .name = "fchownat",
948 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
950 .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
951 .strtoul = STUL_STRARRAYS,
952 .parm = &strarrays__fcntl_cmds_arrays,
953 .show_zero = true, },
954 [2] = { .scnprintf = SCA_FCNTL_ARG, /* arg */ }, }, },
956 .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
957 { .name = "fsconfig",
958 .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
960 .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
961 [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
963 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
964 [1] = { .scnprintf = SCA_FILENAME, /* path */ },
965 [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
966 { .name = "fstat", .alias = "newfstat", },
967 { .name = "fstatat", .alias = "newfstatat", },
969 .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
970 [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
971 { .name = "futimesat",
972 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
973 { .name = "getitimer",
974 .arg = { [0] = STRARRAY(which, itimers), }, },
975 { .name = "getpid", .errpid = true, },
976 { .name = "getpgid", .errpid = true, },
977 { .name = "getppid", .errpid = true, },
978 { .name = "getrandom",
979 .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
980 { .name = "getrlimit",
981 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
982 { .name = "getsockopt",
983 .arg = { [1] = STRARRAY(level, socket_level), }, },
984 { .name = "gettid", .errpid = true, },
987 #if defined(__i386__) || defined(__x86_64__)
989 * FIXME: Make this available to all arches.
991 [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
992 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
994 [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
996 { .name = "kcmp", .nr_args = 5,
997 .arg = { [0] = { .name = "pid1", .scnprintf = SCA_PID, },
998 [1] = { .name = "pid2", .scnprintf = SCA_PID, },
999 [2] = { .name = "type", .scnprintf = SCA_KCMP_TYPE, },
1000 [3] = { .name = "idx1", .scnprintf = SCA_KCMP_IDX, },
1001 [4] = { .name = "idx2", .scnprintf = SCA_KCMP_IDX, }, }, },
1003 .arg = { [0] = STRARRAY(option, keyctl_options), }, },
1005 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1007 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1009 .arg = { [2] = STRARRAY(whence, whences), }, },
1010 { .name = "lstat", .alias = "newlstat", },
1011 { .name = "madvise",
1012 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1013 [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
1014 { .name = "mkdirat",
1015 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1016 { .name = "mknodat",
1017 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
1018 { .name = "mmap", .hexret = true,
1019 /* The standard mmap maps to old_mmap on s390x */
1020 #if defined(__s390x__)
1021 .alias = "old_mmap",
1023 .arg = { [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
1024 [3] = { .scnprintf = SCA_MMAP_FLAGS, /* flags */
1025 .strtoul = STUL_STRARRAY_FLAGS,
1026 .parm = &strarray__mmap_flags, },
1027 [5] = { .scnprintf = SCA_HEX, /* offset */ }, }, },
1029 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
1030 [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
1031 .mask_val = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
1032 { .name = "move_mount",
1033 .arg = { [0] = { .scnprintf = SCA_FDAT, /* from_dfd */ },
1034 [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
1035 [2] = { .scnprintf = SCA_FDAT, /* to_dfd */ },
1036 [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
1037 [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
1038 { .name = "mprotect",
1039 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1040 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ }, }, },
1041 { .name = "mq_unlink",
1042 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
1043 { .name = "mremap", .hexret = true,
1044 .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
1045 { .name = "name_to_handle_at",
1046 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1047 { .name = "newfstatat",
1048 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1050 .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1051 { .name = "open_by_handle_at",
1052 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1053 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1055 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ },
1056 [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
1057 { .name = "perf_event_open",
1058 .arg = { [2] = { .scnprintf = SCA_INT, /* cpu */ },
1059 [3] = { .scnprintf = SCA_FD, /* group_fd */ },
1060 [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
1062 .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
1063 { .name = "pkey_alloc",
1064 .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS, /* access_rights */ }, }, },
1065 { .name = "pkey_free",
1066 .arg = { [0] = { .scnprintf = SCA_INT, /* key */ }, }, },
1067 { .name = "pkey_mprotect",
1068 .arg = { [0] = { .scnprintf = SCA_HEX, /* start */ },
1069 [2] = { .scnprintf = SCA_MMAP_PROT, /* prot */ },
1070 [3] = { .scnprintf = SCA_INT, /* pkey */ }, }, },
1071 { .name = "poll", .timeout = true, },
1072 { .name = "ppoll", .timeout = true, },
1074 .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */
1075 .strtoul = STUL_STRARRAY,
1076 .parm = &strarray__prctl_options, },
1077 [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
1078 [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
1079 { .name = "pread", .alias = "pread64", },
1080 { .name = "preadv", .alias = "pread", },
1081 { .name = "prlimit64",
1082 .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
1083 { .name = "pwrite", .alias = "pwrite64", },
1084 { .name = "readlinkat",
1085 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1086 { .name = "recvfrom",
1087 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1088 { .name = "recvmmsg",
1089 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1090 { .name = "recvmsg",
1091 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1092 { .name = "renameat",
1093 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1094 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
1095 { .name = "renameat2",
1096 .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
1097 [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
1098 [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
1099 { .name = "rt_sigaction",
1100 .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1101 { .name = "rt_sigprocmask",
1102 .arg = { [0] = STRARRAY(how, sighow), }, },
1103 { .name = "rt_sigqueueinfo",
1104 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1105 { .name = "rt_tgsigqueueinfo",
1106 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1107 { .name = "sched_setscheduler",
1108 .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
1109 { .name = "seccomp",
1110 .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP, /* op */ },
1111 [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
1112 { .name = "select", .timeout = true, },
1113 { .name = "sendfile", .alias = "sendfile64", },
1114 { .name = "sendmmsg",
1115 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1116 { .name = "sendmsg",
1117 .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
1119 .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
1120 [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
1121 { .name = "set_tid_address", .errpid = true, },
1122 { .name = "setitimer",
1123 .arg = { [0] = STRARRAY(which, itimers), }, },
1124 { .name = "setrlimit",
1125 .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
1126 { .name = "setsockopt",
1127 .arg = { [1] = STRARRAY(level, socket_level), }, },
1129 .arg = { [0] = STRARRAY(family, socket_families),
1130 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1131 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1132 { .name = "socketpair",
1133 .arg = { [0] = STRARRAY(family, socket_families),
1134 [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
1135 [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
1136 { .name = "stat", .alias = "newstat", },
1138 .arg = { [0] = { .scnprintf = SCA_FDAT, /* fdat */ },
1139 [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
1140 [3] = { .scnprintf = SCA_STATX_MASK, /* mask */ }, }, },
1141 { .name = "swapoff",
1142 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1144 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
1145 { .name = "symlinkat",
1146 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1147 { .name = "sync_file_range",
1148 .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
1150 .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1152 .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
1153 { .name = "umount2", .alias = "umount",
1154 .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
1155 { .name = "uname", .alias = "newuname", },
1156 { .name = "unlinkat",
1157 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
1158 { .name = "utimensat",
1159 .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
1160 { .name = "wait4", .errpid = true,
1161 .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1162 { .name = "waitid", .errpid = true,
1163 .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
1166 static int syscall_fmt__cmp(const void *name, const void *fmtp)
1168 const struct syscall_fmt *fmt = fmtp;
1169 return strcmp(name, fmt->name);
1172 static struct syscall_fmt *__syscall_fmt__find(struct syscall_fmt *fmts, const int nmemb, const char *name)
1174 return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1177 static struct syscall_fmt *syscall_fmt__find(const char *name)
1179 const int nmemb = ARRAY_SIZE(syscall_fmts);
1180 return __syscall_fmt__find(syscall_fmts, nmemb, name);
1183 static struct syscall_fmt *__syscall_fmt__find_by_alias(struct syscall_fmt *fmts, const int nmemb, const char *alias)
1187 for (i = 0; i < nmemb; ++i) {
1188 if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
1195 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1197 const int nmemb = ARRAY_SIZE(syscall_fmts);
1198 return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1202 * is_exit: is this "exit" or "exit_group"?
1203 * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
1204 * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
1205 * nonexistent: Just a hole in the syscall table, syscall id not allocated
1208 struct tep_event *tp_format;
1212 struct bpf_program *sys_enter,
1218 struct tep_format_field *args;
1220 struct syscall_fmt *fmt;
1221 struct syscall_arg_fmt *arg_fmt;
1225 * Must match what is in the BPF program:
1227 * tools/perf/examples/bpf/augmented_raw_syscalls.c
1229 struct bpf_map_syscall_entry {
1231 u16 string_args_len[6];
1235 * We need to have this 'calculated' boolean because in some cases we really
1236 * don't know what is the duration of a syscall, for instance, when we start
1237 * a session and some threads are waiting for a syscall to finish, say 'poll',
1238 * in which case all we can do is to print "( ? ) for duration and for the
1241 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1243 double duration = (double)t / NSEC_PER_MSEC;
1244 size_t printed = fprintf(fp, "(");
1247 printed += fprintf(fp, " ");
1248 else if (duration >= 1.0)
1249 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1250 else if (duration >= 0.01)
1251 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1253 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1254 return printed + fprintf(fp, "): ");
1258 * filename.ptr: The filename char pointer that will be vfs_getname'd
1259 * filename.entry_str_pos: Where to insert the string translated from
1260 * filename.ptr by the vfs_getname tracepoint/kprobe.
1261 * ret_scnprintf: syscall args may set this to a different syscall return
1262 * formatter, for instance, fcntl may return fds, file flags, etc.
1264 struct thread_trace {
1267 unsigned long nr_events;
1268 unsigned long pfmaj, pfmin;
1271 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1274 short int entry_str_pos;
1276 unsigned int namelen;
1284 struct intlist *syscall_stats;
1287 static struct thread_trace *thread_trace__new(void)
1289 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1292 ttrace->files.max = -1;
1293 ttrace->syscall_stats = intlist__new(NULL);
1299 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1301 struct thread_trace *ttrace;
1306 if (thread__priv(thread) == NULL)
1307 thread__set_priv(thread, thread_trace__new());
1309 if (thread__priv(thread) == NULL)
1312 ttrace = thread__priv(thread);
1313 ++ttrace->nr_events;
1317 color_fprintf(fp, PERF_COLOR_RED,
1318 "WARNING: not enough memory, dropping samples!\n");
1323 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1324 size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1326 struct thread_trace *ttrace = thread__priv(arg->thread);
1328 ttrace->ret_scnprintf = ret_scnprintf;
1331 #define TRACE_PFMAJ (1 << 0)
1332 #define TRACE_PFMIN (1 << 1)
1334 static const size_t trace__entry_str_size = 2048;
1336 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1341 if (fd > ttrace->files.max) {
1342 struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1347 if (ttrace->files.max != -1) {
1348 memset(nfiles + ttrace->files.max + 1, 0,
1349 (fd - ttrace->files.max) * sizeof(struct file));
1351 memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1354 ttrace->files.table = nfiles;
1355 ttrace->files.max = fd;
1358 return ttrace->files.table + fd;
1361 struct file *thread__files_entry(struct thread *thread, int fd)
1363 return thread_trace__files_entry(thread__priv(thread), fd);
1366 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1368 struct thread_trace *ttrace = thread__priv(thread);
1369 struct file *file = thread_trace__files_entry(ttrace, fd);
1373 if (stat(pathname, &st) == 0)
1374 file->dev_maj = major(st.st_rdev);
1375 file->pathname = strdup(pathname);
1383 static int thread__read_fd_path(struct thread *thread, int fd)
1385 char linkname[PATH_MAX], pathname[PATH_MAX];
1389 if (thread->pid_ == thread->tid) {
1390 scnprintf(linkname, sizeof(linkname),
1391 "/proc/%d/fd/%d", thread->pid_, fd);
1393 scnprintf(linkname, sizeof(linkname),
1394 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1397 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1400 ret = readlink(linkname, pathname, sizeof(pathname));
1402 if (ret < 0 || ret > st.st_size)
1405 pathname[ret] = '\0';
1406 return trace__set_fd_pathname(thread, fd, pathname);
1409 static const char *thread__fd_path(struct thread *thread, int fd,
1410 struct trace *trace)
1412 struct thread_trace *ttrace = thread__priv(thread);
1414 if (ttrace == NULL || trace->fd_path_disabled)
1420 if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1423 ++trace->stats.proc_getname;
1424 if (thread__read_fd_path(thread, fd))
1428 return ttrace->files.table[fd].pathname;
1431 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1434 size_t printed = scnprintf(bf, size, "%d", fd);
1435 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1438 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1443 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1445 size_t printed = scnprintf(bf, size, "%d", fd);
1446 struct thread *thread = machine__find_thread(trace->host, pid, pid);
1449 const char *path = thread__fd_path(thread, fd, trace);
1452 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1454 thread__put(thread);
1460 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1461 struct syscall_arg *arg)
1464 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1465 struct thread_trace *ttrace = thread__priv(arg->thread);
1467 if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1468 zfree(&ttrace->files.table[fd].pathname);
1473 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1476 struct thread_trace *ttrace = thread__priv(thread);
1478 ttrace->filename.ptr = ptr;
1479 ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1482 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1484 struct augmented_arg *augmented_arg = arg->augmented.args;
1485 size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1487 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1488 * we would have two strings, each prefixed by its size.
1490 int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1492 arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1493 arg->augmented.size -= consumed;
1498 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1499 struct syscall_arg *arg)
1501 unsigned long ptr = arg->val;
1503 if (arg->augmented.args)
1504 return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1506 if (!arg->trace->vfs_getname)
1507 return scnprintf(bf, size, "%#x", ptr);
1509 thread__set_filename_pos(arg->thread, bf, ptr);
1513 static bool trace__filter_duration(struct trace *trace, double t)
1515 return t < (trace->duration_filter * NSEC_PER_MSEC);
1518 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1520 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1522 return fprintf(fp, "%10.3f ", ts);
1526 * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1527 * using ttrace->entry_time for a thread that receives a sys_exit without
1528 * first having received a sys_enter ("poll" issued before tracing session
1529 * starts, lost sys_enter exit due to ring buffer overflow).
1531 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1534 return __trace__fprintf_tstamp(trace, tstamp, fp);
1536 return fprintf(fp, " ? ");
1539 static bool done = false;
1540 static bool interrupted = false;
1542 static void sig_handler(int sig)
1545 interrupted = sig == SIGINT;
1548 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1552 if (trace->multiple_threads) {
1553 if (trace->show_comm)
1554 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1555 printed += fprintf(fp, "%d ", thread->tid);
1561 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1562 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1566 if (trace->show_tstamp)
1567 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1568 if (trace->show_duration)
1569 printed += fprintf_duration(duration, duration_calculated, fp);
1570 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1573 static int trace__process_event(struct trace *trace, struct machine *machine,
1574 union perf_event *event, struct perf_sample *sample)
1578 switch (event->header.type) {
1579 case PERF_RECORD_LOST:
1580 color_fprintf(trace->output, PERF_COLOR_RED,
1581 "LOST %" PRIu64 " events!\n", event->lost.lost);
1582 ret = machine__process_lost_event(machine, event, sample);
1585 ret = machine__process_event(machine, event, sample);
1592 static int trace__tool_process(struct perf_tool *tool,
1593 union perf_event *event,
1594 struct perf_sample *sample,
1595 struct machine *machine)
1597 struct trace *trace = container_of(tool, struct trace, tool);
1598 return trace__process_event(trace, machine, event, sample);
1601 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1603 struct machine *machine = vmachine;
1605 if (machine->kptr_restrict_warned)
1608 if (symbol_conf.kptr_restrict) {
1609 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1610 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1611 "Kernel samples will not be resolved.\n");
1612 machine->kptr_restrict_warned = true;
1616 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1619 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1621 int err = symbol__init(NULL);
1626 trace->host = machine__new_host();
1627 if (trace->host == NULL)
1630 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1634 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1635 evlist->core.threads, trace__tool_process,
1644 static void trace__symbols__exit(struct trace *trace)
1646 machine__exit(trace->host);
1652 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1656 if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1657 nr_args = sc->fmt->nr_args;
1659 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1660 if (sc->arg_fmt == NULL)
1663 for (idx = 0; idx < nr_args; ++idx) {
1665 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1668 sc->nr_args = nr_args;
1672 static struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1673 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1674 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
1677 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1679 const struct syscall_arg_fmt *fmt = fmtp;
1680 return strcmp(name, fmt->name);
1683 static struct syscall_arg_fmt *
1684 __syscall_arg_fmt__find_by_name(struct syscall_arg_fmt *fmts, const int nmemb, const char *name)
1686 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1689 static struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1691 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1692 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1695 static struct tep_format_field *
1696 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1698 struct tep_format_field *last_field = NULL;
1701 for (; field; field = field->next, ++arg) {
1707 len = strlen(field->name);
1709 if (strcmp(field->type, "const char *") == 0 &&
1710 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1711 strstr(field->name, "path") != NULL))
1712 arg->scnprintf = SCA_FILENAME;
1713 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1714 arg->scnprintf = SCA_PTR;
1715 else if (strcmp(field->type, "pid_t") == 0)
1716 arg->scnprintf = SCA_PID;
1717 else if (strcmp(field->type, "umode_t") == 0)
1718 arg->scnprintf = SCA_MODE_T;
1719 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
1720 arg->scnprintf = SCA_CHAR_ARRAY;
1721 arg->nr_entries = field->arraylen;
1722 } else if ((strcmp(field->type, "int") == 0 ||
1723 strcmp(field->type, "unsigned int") == 0 ||
1724 strcmp(field->type, "long") == 0) &&
1725 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1727 * /sys/kernel/tracing/events/syscalls/sys_enter*
1728 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1733 arg->scnprintf = SCA_FD;
1735 struct syscall_arg_fmt *fmt = syscall_arg_fmt__find_by_name(field->name);
1738 arg->scnprintf = fmt->scnprintf;
1739 arg->strtoul = fmt->strtoul;
1747 static int syscall__set_arg_fmts(struct syscall *sc)
1749 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1752 sc->args_size = last_field->offset + last_field->size;
1757 static int trace__read_syscall_info(struct trace *trace, int id)
1761 const char *name = syscalltbl__name(trace->sctbl, id);
1763 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
1764 if (trace->syscalls.table == NULL) {
1765 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1766 if (trace->syscalls.table == NULL)
1770 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1771 // When using libaudit we don't know beforehand what is the max syscall id
1772 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1777 // Need to memset from offset 0 and +1 members if brand new
1778 if (trace->syscalls.table == NULL)
1779 memset(table, 0, (id + 1) * sizeof(*sc));
1781 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
1783 trace->syscalls.table = table;
1784 trace->sctbl->syscalls.max_id = id;
1787 sc = trace->syscalls.table + id;
1788 if (sc->nonexistent)
1792 sc->nonexistent = true;
1797 sc->fmt = syscall_fmt__find(sc->name);
1799 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1800 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1802 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1803 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1804 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1807 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1810 if (IS_ERR(sc->tp_format))
1811 return PTR_ERR(sc->tp_format);
1813 sc->args = sc->tp_format->format.fields;
1815 * We need to check and discard the first variable '__syscall_nr'
1816 * or 'nr' that mean the syscall number. It is needless here.
1817 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1819 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1820 sc->args = sc->args->next;
1824 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1825 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1827 return syscall__set_arg_fmts(sc);
1830 static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
1832 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
1835 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
1842 static int intcmp(const void *a, const void *b)
1844 const int *one = a, *another = b;
1846 return *one - *another;
1849 static int trace__validate_ev_qualifier(struct trace *trace)
1852 bool printed_invalid_prefix = false;
1853 struct str_node *pos;
1854 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1856 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1857 sizeof(trace->ev_qualifier_ids.entries[0]));
1859 if (trace->ev_qualifier_ids.entries == NULL) {
1860 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1866 strlist__for_each_entry(pos, trace->ev_qualifier) {
1867 const char *sc = pos->s;
1868 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1871 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1875 if (!printed_invalid_prefix) {
1876 pr_debug("Skipping unknown syscalls: ");
1877 printed_invalid_prefix = true;
1886 trace->ev_qualifier_ids.entries[nr_used++] = id;
1887 if (match_next == -1)
1891 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1894 if (nr_allocated == nr_used) {
1898 entries = realloc(trace->ev_qualifier_ids.entries,
1899 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1900 if (entries == NULL) {
1902 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1905 trace->ev_qualifier_ids.entries = entries;
1907 trace->ev_qualifier_ids.entries[nr_used++] = id;
1911 trace->ev_qualifier_ids.nr = nr_used;
1912 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1914 if (printed_invalid_prefix)
1918 zfree(&trace->ev_qualifier_ids.entries);
1919 trace->ev_qualifier_ids.nr = 0;
1923 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1925 bool in_ev_qualifier;
1927 if (trace->ev_qualifier_ids.nr == 0)
1930 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1931 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1933 if (in_ev_qualifier)
1934 return !trace->not_ev_qualifier;
1936 return trace->not_ev_qualifier;
1940 * args is to be interpreted as a series of longs but we need to handle
1941 * 8-byte unaligned accesses. args points to raw_data within the event
1942 * and raw_data is guaranteed to be 8-byte unaligned because it is
1943 * preceded by raw_size which is a u32. So we need to copy args to a temp
1944 * variable to read it. Most notably this avoids extended load instructions
1945 * on unaligned addresses
1947 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1950 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1952 memcpy(&val, p, sizeof(val));
1956 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1957 struct syscall_arg *arg)
1959 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1960 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1962 return scnprintf(bf, size, "arg%d: ", arg->idx);
1966 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1967 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1968 * in tools/perf/trace/beauty/mount_flags.c
1970 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
1972 if (fmt && fmt->mask_val)
1973 return fmt->mask_val(arg, val);
1978 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1979 struct syscall_arg *arg, unsigned long val)
1981 if (fmt && fmt->scnprintf) {
1984 arg->parm = fmt->parm;
1985 return fmt->scnprintf(bf, size, arg);
1987 return scnprintf(bf, size, "%ld", val);
1990 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1991 unsigned char *args, void *augmented_args, int augmented_args_size,
1992 struct trace *trace, struct thread *thread)
1997 struct syscall_arg arg = {
2000 .size = augmented_args_size,
2001 .args = augmented_args,
2007 .show_string_prefix = trace->show_string_prefix,
2009 struct thread_trace *ttrace = thread__priv(thread);
2012 * Things like fcntl will set this in its 'cmd' formatter to pick the
2013 * right formatter for the return value (an fd? file flags?), which is
2014 * not needed for syscalls that always return a given type, say an fd.
2016 ttrace->ret_scnprintf = NULL;
2018 if (sc->args != NULL) {
2019 struct tep_format_field *field;
2021 for (field = sc->args; field;
2022 field = field->next, ++arg.idx, bit <<= 1) {
2026 arg.fmt = &sc->arg_fmt[arg.idx];
2027 val = syscall_arg__val(&arg, arg.idx);
2029 * Some syscall args need some mask, most don't and
2030 * return val untouched.
2032 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
2035 * Suppress this argument if its value is zero and
2036 * and we don't have a string associated in an
2040 !trace->show_zeros &&
2042 (sc->arg_fmt[arg.idx].show_zero ||
2043 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
2044 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2045 sc->arg_fmt[arg.idx].parm))
2048 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2050 if (trace->show_arg_names)
2051 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2053 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2054 bf + printed, size - printed, &arg, val);
2056 } else if (IS_ERR(sc->tp_format)) {
2058 * If we managed to read the tracepoint /format file, then we
2059 * may end up not having any args, like with gettid(), so only
2060 * print the raw args when we didn't manage to read it.
2062 while (arg.idx < sc->nr_args) {
2065 val = syscall_arg__val(&arg, arg.idx);
2067 printed += scnprintf(bf + printed, size - printed, ", ");
2068 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
2069 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
2079 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
2080 union perf_event *event,
2081 struct perf_sample *sample);
2083 static struct syscall *trace__syscall_info(struct trace *trace,
2084 struct evsel *evsel, int id)
2091 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2092 * before that, leaving at a higher verbosity level till that is
2093 * explained. Reproduced with plain ftrace with:
2095 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2096 * grep "NR -1 " /t/trace_pipe
2098 * After generating some load on the machine.
2102 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
2103 id, evsel__name(evsel), ++n);
2110 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2111 if (id > trace->sctbl->syscalls.max_id) {
2113 if (id >= trace->sctbl->syscalls.max_id) {
2115 * With libaudit we don't know beforehand what is the max_id,
2116 * so we let trace__read_syscall_info() figure that out as we
2117 * go on reading syscalls.
2119 err = trace__read_syscall_info(trace, id);
2125 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2126 (err = trace__read_syscall_info(trace, id)) != 0)
2129 if (trace->syscalls.table[id].name == NULL) {
2130 if (trace->syscalls.table[id].nonexistent)
2135 return &trace->syscalls.table[id];
2139 char sbuf[STRERR_BUFSIZE];
2140 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
2141 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
2142 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2143 fputs(" information\n", trace->output);
2148 struct syscall_stats {
2155 static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2156 int id, struct perf_sample *sample, long err, bool errno_summary)
2158 struct int_node *inode;
2159 struct syscall_stats *stats;
2162 inode = intlist__findnew(ttrace->syscall_stats, id);
2166 stats = inode->priv;
2167 if (stats == NULL) {
2168 stats = malloc(sizeof(*stats));
2172 stats->nr_failures = 0;
2173 stats->max_errno = 0;
2174 stats->errnos = NULL;
2175 init_stats(&stats->stats);
2176 inode->priv = stats;
2179 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2180 duration = sample->time - ttrace->entry_time;
2182 update_stats(&stats->stats, duration);
2185 ++stats->nr_failures;
2191 if (err > stats->max_errno) {
2192 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2195 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2197 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2198 thread__comm_str(thread), thread->pid_, thread->tid);
2202 stats->errnos = new_errnos;
2203 stats->max_errno = err;
2206 ++stats->errnos[err - 1];
2210 static int trace__printf_interrupted_entry(struct trace *trace)
2212 struct thread_trace *ttrace;
2216 if (trace->failure_only || trace->current == NULL)
2219 ttrace = thread__priv(trace->current);
2221 if (!ttrace->entry_pending)
2224 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
2225 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2227 if (len < trace->args_alignment - 4)
2228 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
2230 printed += fprintf(trace->output, " ...\n");
2232 ttrace->entry_pending = false;
2233 ++trace->nr_events_printed;
2238 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
2239 struct perf_sample *sample, struct thread *thread)
2243 if (trace->print_sample) {
2244 double ts = (double)sample->time / NSEC_PER_MSEC;
2246 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
2247 evsel__name(evsel), ts,
2248 thread__comm_str(thread),
2249 sample->pid, sample->tid, sample->cpu);
2255 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
2257 void *augmented_args = NULL;
2259 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
2260 * and there we get all 6 syscall args plus the tracepoint common fields
2261 * that gets calculated at the start and the syscall_nr (another long).
2262 * So we check if that is the case and if so don't look after the
2263 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2266 * We'll revisit this later to pass s->args_size to the BPF augmenter
2267 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2268 * copies only what we need for each syscall, like what happens when we
2269 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2270 * traffic to just what is needed for each syscall.
2272 int args_size = raw_augmented_args_size ?: sc->args_size;
2274 *augmented_args_size = sample->raw_size - args_size;
2275 if (*augmented_args_size > 0)
2276 augmented_args = sample->raw_data + args_size;
2278 return augmented_args;
2281 static void syscall__exit(struct syscall *sc)
2289 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
2290 union perf_event *event __maybe_unused,
2291 struct perf_sample *sample)
2296 struct thread *thread;
2297 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2298 int augmented_args_size = 0;
2299 void *augmented_args = NULL;
2300 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2301 struct thread_trace *ttrace;
2306 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2307 ttrace = thread__trace(thread, trace->output);
2311 trace__fprintf_sample(trace, evsel, sample, thread);
2313 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2315 if (ttrace->entry_str == NULL) {
2316 ttrace->entry_str = malloc(trace__entry_str_size);
2317 if (!ttrace->entry_str)
2321 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2322 trace__printf_interrupted_entry(trace);
2324 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2325 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2326 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2327 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2328 * so when handling, say the openat syscall, we end up getting 6 args for the
2329 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2330 * thinking that the extra 2 u64 args are the augmented filename, so just check
2331 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2333 if (evsel != trace->syscalls.events.sys_enter)
2334 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2335 ttrace->entry_time = sample->time;
2336 msg = ttrace->entry_str;
2337 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2339 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2340 args, augmented_args, augmented_args_size, trace, thread);
2343 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2346 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2347 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2348 if (trace->args_alignment > printed)
2349 alignment = trace->args_alignment - printed;
2350 fprintf(trace->output, "%*s= ?\n", alignment, " ");
2353 ttrace->entry_pending = true;
2354 /* See trace__vfs_getname & trace__sys_exit */
2355 ttrace->filename.pending_open = false;
2358 if (trace->current != thread) {
2359 thread__put(trace->current);
2360 trace->current = thread__get(thread);
2364 thread__put(thread);
2368 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2369 struct perf_sample *sample)
2371 struct thread_trace *ttrace;
2372 struct thread *thread;
2373 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2374 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2376 void *args, *augmented_args = NULL;
2377 int augmented_args_size;
2382 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2383 ttrace = thread__trace(thread, trace->output);
2385 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2386 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2391 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2392 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2393 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2394 fprintf(trace->output, "%s", msg);
2397 thread__put(thread);
2401 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2402 struct perf_sample *sample,
2403 struct callchain_cursor *cursor)
2405 struct addr_location al;
2406 int max_stack = evsel->core.attr.sample_max_stack ?
2407 evsel->core.attr.sample_max_stack :
2411 if (machine__resolve(trace->host, &al, sample) < 0)
2414 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2415 addr_location__put(&al);
2419 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2421 /* TODO: user-configurable print_opts */
2422 const unsigned int print_opts = EVSEL__PRINT_SYM |
2424 EVSEL__PRINT_UNKNOWN_AS_ADDR;
2426 return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
2429 static const char *errno_to_name(struct evsel *evsel, int err)
2431 struct perf_env *env = evsel__env(evsel);
2432 const char *arch_name = perf_env__arch(env);
2434 return arch_syscalls__strerrno(arch_name, err);
2437 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2438 union perf_event *event __maybe_unused,
2439 struct perf_sample *sample)
2443 bool duration_calculated = false;
2444 struct thread *thread;
2445 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2446 int alignment = trace->args_alignment;
2447 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2448 struct thread_trace *ttrace;
2453 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2454 ttrace = thread__trace(thread, trace->output);
2458 trace__fprintf_sample(trace, evsel, sample, thread);
2460 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2463 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
2465 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2466 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2467 ttrace->filename.pending_open = false;
2468 ++trace->stats.vfs_getname;
2471 if (ttrace->entry_time) {
2472 duration = sample->time - ttrace->entry_time;
2473 if (trace__filter_duration(trace, duration))
2475 duration_calculated = true;
2476 } else if (trace->duration_filter)
2479 if (sample->callchain) {
2480 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2481 if (callchain_ret == 0) {
2482 if (callchain_cursor.nr < trace->min_stack)
2488 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2491 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2493 if (ttrace->entry_pending) {
2494 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2496 printed += fprintf(trace->output, " ... [");
2497 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2499 printed += fprintf(trace->output, "]: %s()", sc->name);
2502 printed++; /* the closing ')' */
2504 if (alignment > printed)
2505 alignment -= printed;
2509 fprintf(trace->output, ")%*s= ", alignment, " ");
2511 if (sc->fmt == NULL) {
2515 fprintf(trace->output, "%ld", ret);
2516 } else if (ret < 0) {
2518 char bf[STRERR_BUFSIZE];
2519 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2520 *e = errno_to_name(evsel, -ret);
2522 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2524 } else if (ret == 0 && sc->fmt->timeout)
2525 fprintf(trace->output, "0 (Timeout)");
2526 else if (ttrace->ret_scnprintf) {
2528 struct syscall_arg arg = {
2533 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2534 ttrace->ret_scnprintf = NULL;
2535 fprintf(trace->output, "%s", bf);
2536 } else if (sc->fmt->hexret)
2537 fprintf(trace->output, "%#lx", ret);
2538 else if (sc->fmt->errpid) {
2539 struct thread *child = machine__find_thread(trace->host, ret, ret);
2541 if (child != NULL) {
2542 fprintf(trace->output, "%ld", ret);
2543 if (child->comm_set)
2544 fprintf(trace->output, " (%s)", thread__comm_str(child));
2550 fputc('\n', trace->output);
2553 * We only consider an 'event' for the sake of --max-events a non-filtered
2554 * sys_enter + sys_exit and other tracepoint events.
2556 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2559 if (callchain_ret > 0)
2560 trace__fprintf_callchain(trace, sample);
2561 else if (callchain_ret < 0)
2562 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2564 ttrace->entry_pending = false;
2567 thread__put(thread);
2571 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2572 union perf_event *event __maybe_unused,
2573 struct perf_sample *sample)
2575 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2576 struct thread_trace *ttrace;
2577 size_t filename_len, entry_str_len, to_move;
2578 ssize_t remaining_space;
2580 const char *filename = evsel__rawptr(evsel, sample, "pathname");
2585 ttrace = thread__priv(thread);
2589 filename_len = strlen(filename);
2590 if (filename_len == 0)
2593 if (ttrace->filename.namelen < filename_len) {
2594 char *f = realloc(ttrace->filename.name, filename_len + 1);
2599 ttrace->filename.namelen = filename_len;
2600 ttrace->filename.name = f;
2603 strcpy(ttrace->filename.name, filename);
2604 ttrace->filename.pending_open = true;
2606 if (!ttrace->filename.ptr)
2609 entry_str_len = strlen(ttrace->entry_str);
2610 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2611 if (remaining_space <= 0)
2614 if (filename_len > (size_t)remaining_space) {
2615 filename += filename_len - remaining_space;
2616 filename_len = remaining_space;
2619 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2620 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2621 memmove(pos + filename_len, pos, to_move);
2622 memcpy(pos, filename, filename_len);
2624 ttrace->filename.ptr = 0;
2625 ttrace->filename.entry_str_pos = 0;
2627 thread__put(thread);
2632 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2633 union perf_event *event __maybe_unused,
2634 struct perf_sample *sample)
2636 u64 runtime = evsel__intval(evsel, sample, "runtime");
2637 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2638 struct thread *thread = machine__findnew_thread(trace->host,
2641 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2646 ttrace->runtime_ms += runtime_ms;
2647 trace->runtime_ms += runtime_ms;
2649 thread__put(thread);
2653 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2655 evsel__strval(evsel, sample, "comm"),
2656 (pid_t)evsel__intval(evsel, sample, "pid"),
2658 evsel__intval(evsel, sample, "vruntime"));
2662 static int bpf_output__printer(enum binary_printer_ops op,
2663 unsigned int val, void *extra __maybe_unused, FILE *fp)
2665 unsigned char ch = (unsigned char)val;
2668 case BINARY_PRINT_CHAR_DATA:
2669 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2670 case BINARY_PRINT_DATA_BEGIN:
2671 case BINARY_PRINT_LINE_BEGIN:
2672 case BINARY_PRINT_ADDR:
2673 case BINARY_PRINT_NUM_DATA:
2674 case BINARY_PRINT_NUM_PAD:
2675 case BINARY_PRINT_SEP:
2676 case BINARY_PRINT_CHAR_PAD:
2677 case BINARY_PRINT_LINE_END:
2678 case BINARY_PRINT_DATA_END:
2686 static void bpf_output__fprintf(struct trace *trace,
2687 struct perf_sample *sample)
2689 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2690 bpf_output__printer, NULL, trace->output);
2691 ++trace->nr_events_printed;
2694 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2695 struct thread *thread, void *augmented_args, int augmented_args_size)
2698 size_t size = sizeof(bf);
2699 struct tep_format_field *field = evsel->tp_format->format.fields;
2700 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
2704 struct syscall_arg syscall_arg = {
2706 .size = augmented_args_size,
2707 .args = augmented_args,
2713 .show_string_prefix = trace->show_string_prefix,
2716 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2717 if (syscall_arg.mask & bit)
2720 syscall_arg.len = 0;
2721 syscall_arg.fmt = arg;
2722 if (field->flags & TEP_FIELD_IS_ARRAY) {
2723 int offset = field->offset;
2725 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2726 offset = format_field__intval(field, sample, evsel->needs_swap);
2727 syscall_arg.len = offset >> 16;
2729 if (field->flags & TEP_FIELD_IS_RELATIVE)
2730 offset += field->offset + field->size;
2733 val = (uintptr_t)(sample->raw_data + offset);
2735 val = format_field__intval(field, sample, evsel->needs_swap);
2737 * Some syscall args need some mask, most don't and
2738 * return val untouched.
2740 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2743 * Suppress this argument if its value is zero and
2744 * and we don't have a string associated in an
2748 !trace->show_zeros &&
2749 !((arg->show_zero ||
2750 arg->scnprintf == SCA_STRARRAY ||
2751 arg->scnprintf == SCA_STRARRAYS) &&
2755 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2758 * XXX Perhaps we should have a show_tp_arg_names,
2759 * leaving show_arg_names just for syscalls?
2761 if (1 || trace->show_arg_names)
2762 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2764 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2767 return printed + fprintf(trace->output, "%s", bf);
2770 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2771 union perf_event *event __maybe_unused,
2772 struct perf_sample *sample)
2774 struct thread *thread;
2775 int callchain_ret = 0;
2777 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2778 * this event's max_events having been hit and this is an entry coming
2779 * from the ring buffer that we should discard, since the max events
2780 * have already been considered/printed.
2782 if (evsel->disabled)
2785 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2787 if (sample->callchain) {
2788 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2789 if (callchain_ret == 0) {
2790 if (callchain_cursor.nr < trace->min_stack)
2796 trace__printf_interrupted_entry(trace);
2797 trace__fprintf_tstamp(trace, sample->time, trace->output);
2799 if (trace->trace_syscalls && trace->show_duration)
2800 fprintf(trace->output, "( ): ");
2803 trace__fprintf_comm_tid(trace, thread, trace->output);
2805 if (evsel == trace->syscalls.events.augmented) {
2806 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2807 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2810 fprintf(trace->output, "%s(", sc->name);
2811 trace__fprintf_sys_enter(trace, evsel, sample);
2812 fputc(')', trace->output);
2817 * XXX: Not having the associated syscall info or not finding/adding
2818 * the thread should never happen, but if it does...
2819 * fall thru and print it as a bpf_output event.
2823 fprintf(trace->output, "%s(", evsel->name);
2825 if (evsel__is_bpf_output(evsel)) {
2826 bpf_output__fprintf(trace, sample);
2827 } else if (evsel->tp_format) {
2828 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2829 trace__fprintf_sys_enter(trace, evsel, sample)) {
2830 if (trace->libtraceevent_print) {
2831 event_format__fprintf(evsel->tp_format, sample->cpu,
2832 sample->raw_data, sample->raw_size,
2835 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2841 fprintf(trace->output, ")\n");
2843 if (callchain_ret > 0)
2844 trace__fprintf_callchain(trace, sample);
2845 else if (callchain_ret < 0)
2846 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2848 ++trace->nr_events_printed;
2850 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2851 evsel__disable(evsel);
2852 evsel__close(evsel);
2855 thread__put(thread);
2859 static void print_location(FILE *f, struct perf_sample *sample,
2860 struct addr_location *al,
2861 bool print_dso, bool print_sym)
2864 if ((verbose > 0 || print_dso) && al->map)
2865 fprintf(f, "%s@", al->map->dso->long_name);
2867 if ((verbose > 0 || print_sym) && al->sym)
2868 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2869 al->addr - al->sym->start);
2871 fprintf(f, "0x%" PRIx64, al->addr);
2873 fprintf(f, "0x%" PRIx64, sample->addr);
2876 static int trace__pgfault(struct trace *trace,
2877 struct evsel *evsel,
2878 union perf_event *event __maybe_unused,
2879 struct perf_sample *sample)
2881 struct thread *thread;
2882 struct addr_location al;
2883 char map_type = 'd';
2884 struct thread_trace *ttrace;
2886 int callchain_ret = 0;
2888 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2890 if (sample->callchain) {
2891 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2892 if (callchain_ret == 0) {
2893 if (callchain_cursor.nr < trace->min_stack)
2899 ttrace = thread__trace(thread, trace->output);
2903 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2908 if (trace->summary_only)
2911 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2913 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2915 fprintf(trace->output, "%sfault [",
2916 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2919 print_location(trace->output, sample, &al, false, true);
2921 fprintf(trace->output, "] => ");
2923 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2926 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2934 print_location(trace->output, sample, &al, true, false);
2936 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2938 if (callchain_ret > 0)
2939 trace__fprintf_callchain(trace, sample);
2940 else if (callchain_ret < 0)
2941 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2943 ++trace->nr_events_printed;
2947 thread__put(thread);
2951 static void trace__set_base_time(struct trace *trace,
2952 struct evsel *evsel,
2953 struct perf_sample *sample)
2956 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2957 * and don't use sample->time unconditionally, we may end up having
2958 * some other event in the future without PERF_SAMPLE_TIME for good
2959 * reason, i.e. we may not be interested in its timestamps, just in
2960 * it taking place, picking some piece of information when it
2961 * appears in our event stream (vfs_getname comes to mind).
2963 if (trace->base_time == 0 && !trace->full_time &&
2964 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2965 trace->base_time = sample->time;
2968 static int trace__process_sample(struct perf_tool *tool,
2969 union perf_event *event,
2970 struct perf_sample *sample,
2971 struct evsel *evsel,
2972 struct machine *machine __maybe_unused)
2974 struct trace *trace = container_of(tool, struct trace, tool);
2975 struct thread *thread;
2978 tracepoint_handler handler = evsel->handler;
2980 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2981 if (thread && thread__is_filtered(thread))
2984 trace__set_base_time(trace, evsel, sample);
2988 handler(trace, evsel, event, sample);
2991 thread__put(thread);
2995 static int trace__record(struct trace *trace, int argc, const char **argv)
2997 unsigned int rec_argc, i, j;
2998 const char **rec_argv;
2999 const char * const record_args[] = {
3005 pid_t pid = getpid();
3006 char *filter = asprintf__tp_filter_pids(1, &pid);
3007 const char * const sc_args[] = { "-e", };
3008 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3009 const char * const majpf_args[] = { "-e", "major-faults" };
3010 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3011 const char * const minpf_args[] = { "-e", "minor-faults" };
3012 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
3015 /* +3 is for the event string below and the pid filter */
3016 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
3017 majpf_args_nr + minpf_args_nr + argc;
3018 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3020 if (rec_argv == NULL || filter == NULL)
3024 for (i = 0; i < ARRAY_SIZE(record_args); i++)
3025 rec_argv[j++] = record_args[i];
3027 if (trace->trace_syscalls) {
3028 for (i = 0; i < sc_args_nr; i++)
3029 rec_argv[j++] = sc_args[i];
3031 /* event string may be different for older kernels - e.g., RHEL6 */
3032 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3033 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3034 else if (is_valid_tracepoint("syscalls:sys_enter"))
3035 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3037 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
3042 rec_argv[j++] = "--filter";
3043 rec_argv[j++] = filter;
3045 if (trace->trace_pgfaults & TRACE_PFMAJ)
3046 for (i = 0; i < majpf_args_nr; i++)
3047 rec_argv[j++] = majpf_args[i];
3049 if (trace->trace_pgfaults & TRACE_PFMIN)
3050 for (i = 0; i < minpf_args_nr; i++)
3051 rec_argv[j++] = minpf_args[i];
3053 for (i = 0; i < (unsigned int)argc; i++)
3054 rec_argv[j++] = argv[i];
3056 err = cmd_record(j, rec_argv);
3063 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3065 static bool evlist__add_vfs_getname(struct evlist *evlist)
3068 struct evsel *evsel, *tmp;
3069 struct parse_events_error err;
3072 parse_events_error__init(&err);
3073 ret = parse_events(evlist, "probe:vfs_getname*", &err);
3074 parse_events_error__exit(&err);
3078 evlist__for_each_entry_safe(evlist, evsel, tmp) {
3079 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
3082 if (evsel__field(evsel, "pathname")) {
3083 evsel->handler = trace__vfs_getname;
3088 list_del_init(&evsel->core.node);
3089 evsel->evlist = NULL;
3090 evsel__delete(evsel);
3096 static struct evsel *evsel__new_pgfault(u64 config)
3098 struct evsel *evsel;
3099 struct perf_event_attr attr = {
3100 .type = PERF_TYPE_SOFTWARE,
3104 attr.config = config;
3105 attr.sample_period = 1;
3107 event_attr_init(&attr);
3109 evsel = evsel__new(&attr);
3111 evsel->handler = trace__pgfault;
3116 static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3118 struct evsel *evsel;
3120 evlist__for_each_entry(evlist, evsel) {
3121 struct evsel_trace *et = evsel->priv;
3123 if (!et || !evsel->tp_format || strcmp(evsel->tp_format->system, "syscalls"))
3131 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3133 const u32 type = event->header.type;
3134 struct evsel *evsel;
3136 if (type != PERF_RECORD_SAMPLE) {
3137 trace__process_event(trace, trace->host, event, sample);
3141 evsel = evlist__id2evsel(trace->evlist, sample->id);
3142 if (evsel == NULL) {
3143 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3147 if (evswitch__discard(&trace->evswitch, evsel))
3150 trace__set_base_time(trace, evsel, sample);
3152 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
3153 sample->raw_data == NULL) {
3154 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
3155 evsel__name(evsel), sample->tid,
3156 sample->cpu, sample->raw_size);
3158 tracepoint_handler handler = evsel->handler;
3159 handler(trace, evsel, event, sample);
3162 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3166 static int trace__add_syscall_newtp(struct trace *trace)
3169 struct evlist *evlist = trace->evlist;
3170 struct evsel *sys_enter, *sys_exit;
3172 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
3173 if (sys_enter == NULL)
3176 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3177 goto out_delete_sys_enter;
3179 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
3180 if (sys_exit == NULL)
3181 goto out_delete_sys_enter;
3183 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3184 goto out_delete_sys_exit;
3186 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3187 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
3189 evlist__add(evlist, sys_enter);
3190 evlist__add(evlist, sys_exit);
3192 if (callchain_param.enabled && !trace->kernel_syscallchains) {
3194 * We're interested only in the user space callchain
3195 * leading to the syscall, allow overriding that for
3196 * debugging reasons using --kernel_syscall_callchains
3198 sys_exit->core.attr.exclude_callchain_kernel = 1;
3201 trace->syscalls.events.sys_enter = sys_enter;
3202 trace->syscalls.events.sys_exit = sys_exit;
3208 out_delete_sys_exit:
3209 evsel__delete_priv(sys_exit);
3210 out_delete_sys_enter:
3211 evsel__delete_priv(sys_enter);
3215 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
3218 struct evsel *sys_exit;
3219 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3220 trace->ev_qualifier_ids.nr,
3221 trace->ev_qualifier_ids.entries);
3226 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
3227 sys_exit = trace->syscalls.events.sys_exit;
3228 err = evsel__append_tp_filter(sys_exit, filter);
3239 #ifdef HAVE_LIBBPF_SUPPORT
3240 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
3242 if (trace->bpf_obj == NULL)
3245 return bpf_object__find_map_by_name(trace->bpf_obj, name);
3248 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3250 trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
3253 static void trace__set_bpf_map_syscalls(struct trace *trace)
3255 trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
3256 trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
3257 trace->syscalls.prog_array.sys_exit = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
3260 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3262 struct bpf_program *pos, *prog = NULL;
3263 const char *sec_name;
3265 if (trace->bpf_obj == NULL)
3268 bpf_object__for_each_program(pos, trace->bpf_obj) {
3269 sec_name = bpf_program__section_name(pos);
3270 if (sec_name && !strcmp(sec_name, name)) {
3279 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3280 const char *prog_name, const char *type)
3282 struct bpf_program *prog;
3284 if (prog_name == NULL) {
3285 char default_prog_name[256];
3286 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
3287 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3290 if (sc->fmt && sc->fmt->alias) {
3291 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
3292 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3296 goto out_unaugmented;
3299 prog = trace__find_bpf_program_by_title(trace, prog_name);
3306 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3307 prog_name, type, sc->name);
3309 return trace->syscalls.unaugmented_prog;
3312 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3314 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3319 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3320 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
3323 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3325 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3326 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3329 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3331 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3332 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3335 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
3337 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3343 for (; arg < sc->nr_args; ++arg) {
3344 entry->string_args_len[arg] = 0;
3345 if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
3346 /* Should be set like strace -s strsize */
3347 entry->string_args_len[arg] = PATH_MAX;
3351 for (; arg < 6; ++arg)
3352 entry->string_args_len[arg] = 0;
3354 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
3356 int fd = bpf_map__fd(trace->syscalls.map);
3357 struct bpf_map_syscall_entry value = {
3358 .enabled = !trace->not_ev_qualifier,
3363 for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
3364 int key = trace->ev_qualifier_ids.entries[i];
3366 if (value.enabled) {
3367 trace__init_bpf_map_syscall_args(trace, key, &value);
3368 trace__init_syscall_bpf_progs(trace, key);
3371 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
3379 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
3381 int fd = bpf_map__fd(trace->syscalls.map);
3382 struct bpf_map_syscall_entry value = {
3387 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3389 trace__init_bpf_map_syscall_args(trace, key, &value);
3391 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
3399 static int trace__init_syscalls_bpf_map(struct trace *trace)
3401 bool enabled = true;
3403 if (trace->ev_qualifier_ids.nr)
3404 enabled = trace->not_ev_qualifier;
3406 return __trace__init_syscalls_bpf_map(trace, enabled);
3409 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3411 struct tep_format_field *field, *candidate_field;
3415 * We're only interested in syscalls that have a pointer:
3417 for (field = sc->args; field; field = field->next) {
3418 if (field->flags & TEP_FIELD_IS_POINTER)
3419 goto try_to_find_pair;
3425 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3426 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3427 struct bpf_program *pair_prog;
3428 bool is_candidate = false;
3430 if (pair == NULL || pair == sc ||
3431 pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3434 for (field = sc->args, candidate_field = pair->args;
3435 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3436 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3437 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3440 if (!candidate_is_pointer) {
3441 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3445 if (candidate_is_pointer) {
3446 // The candidate might copy a pointer we don't have, skip it.
3447 goto next_candidate;
3452 if (strcmp(field->type, candidate_field->type))
3453 goto next_candidate;
3455 is_candidate = true;
3459 goto next_candidate;
3462 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3463 * then it may be collecting that and we then can't use it, as it would collect
3464 * more than what is common to the two syscalls.
3466 if (candidate_field) {
3467 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3468 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3469 goto next_candidate;
3472 pair_prog = pair->bpf_prog.sys_enter;
3474 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3475 * have been searched for, so search it here and if it returns the
3476 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3477 * program for a filtered syscall on a non-filtered one.
3479 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3480 * useful for "renameat2".
3482 if (pair_prog == NULL) {
3483 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3484 if (pair_prog == trace->syscalls.unaugmented_prog)
3485 goto next_candidate;
3488 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3497 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3499 int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3500 map_exit_fd = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3503 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3506 if (!trace__syscall_enabled(trace, key))
3509 trace__init_syscall_bpf_progs(trace, key);
3511 // It'll get at least the "!raw_syscalls:unaugmented"
3512 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3513 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3516 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3517 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3523 * Now lets do a second pass looking for enabled syscalls without
3524 * an augmenter that have a signature that is a superset of another
3525 * syscall with an augmenter so that we can auto-reuse it.
3527 * I.e. if we have an augmenter for the "open" syscall that has
3530 * int open(const char *pathname, int flags, mode_t mode);
3532 * I.e. that will collect just the first string argument, then we
3533 * can reuse it for the 'creat' syscall, that has this signature:
3535 * int creat(const char *pathname, mode_t mode);
3539 * int stat(const char *pathname, struct stat *statbuf);
3540 * int lstat(const char *pathname, struct stat *statbuf);
3542 * Because the 'open' augmenter will collect the first arg as a string,
3543 * and leave alone all the other args, which already helps with
3544 * beautifying 'stat' and 'lstat''s pathname arg.
3546 * Then, in time, when 'stat' gets an augmenter that collects both
3547 * first and second arg (this one on the raw_syscalls:sys_exit prog
3548 * array tail call, then that one will be used.
3550 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3551 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3552 struct bpf_program *pair_prog;
3555 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3559 * For now we're just reusing the sys_enter prog, and if it
3560 * already has an augmenter, we don't need to find one.
3562 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3566 * Look at all the other syscalls for one that has a signature
3567 * that is close enough that we can share:
3569 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3570 if (pair_prog == NULL)
3573 sc->bpf_prog.sys_enter = pair_prog;
3576 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3577 * with the fd for the program we're reusing:
3579 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3580 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3589 static void trace__delete_augmented_syscalls(struct trace *trace)
3591 struct evsel *evsel, *tmp;
3593 evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3594 evsel__delete(trace->syscalls.events.augmented);
3595 trace->syscalls.events.augmented = NULL;
3597 evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3598 if (evsel->bpf_obj == trace->bpf_obj) {
3599 evlist__remove(trace->evlist, evsel);
3600 evsel__delete(evsel);
3605 bpf_object__close(trace->bpf_obj);
3606 trace->bpf_obj = NULL;
3608 #else // HAVE_LIBBPF_SUPPORT
3609 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace __maybe_unused,
3610 const char *name __maybe_unused)
3615 static void trace__set_bpf_map_filtered_pids(struct trace *trace __maybe_unused)
3619 static void trace__set_bpf_map_syscalls(struct trace *trace __maybe_unused)
3623 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3628 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3633 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3634 const char *name __maybe_unused)
3639 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3644 static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3647 #endif // HAVE_LIBBPF_SUPPORT
3649 static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3651 struct evsel *evsel;
3653 evlist__for_each_entry(trace->evlist, evsel) {
3654 if (evsel == trace->syscalls.events.augmented ||
3655 evsel->bpf_obj == trace->bpf_obj)
3664 static int trace__set_ev_qualifier_filter(struct trace *trace)
3666 if (trace->syscalls.map)
3667 return trace__set_ev_qualifier_bpf_filter(trace);
3668 if (trace->syscalls.events.sys_enter)
3669 return trace__set_ev_qualifier_tp_filter(trace);
3673 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3674 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3677 #ifdef HAVE_LIBBPF_SUPPORT
3679 int map_fd = bpf_map__fd(map);
3682 for (i = 0; i < npids; ++i) {
3683 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3691 static int trace__set_filter_loop_pids(struct trace *trace)
3693 unsigned int nr = 1, err;
3697 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3699 while (thread && nr < ARRAY_SIZE(pids)) {
3700 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3705 if (!strcmp(thread__comm_str(parent), "sshd") ||
3706 strstarts(thread__comm_str(parent), "gnome-terminal")) {
3707 pids[nr++] = parent->tid;
3713 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
3714 if (!err && trace->filter_pids.map)
3715 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3720 static int trace__set_filter_pids(struct trace *trace)
3724 * Better not use !target__has_task() here because we need to cover the
3725 * case where no threads were specified in the command line, but a
3726 * workload was, and in that case we will fill in the thread_map when
3727 * we fork the workload in evlist__prepare_workload.
3729 if (trace->filter_pids.nr > 0) {
3730 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3731 trace->filter_pids.entries);
3732 if (!err && trace->filter_pids.map) {
3733 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3734 trace->filter_pids.entries);
3736 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3737 err = trace__set_filter_loop_pids(trace);
3743 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3745 struct evlist *evlist = trace->evlist;
3746 struct perf_sample sample;
3747 int err = evlist__parse_sample(evlist, event, &sample);
3750 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3752 trace__handle_event(trace, event, &sample);
3757 static int __trace__flush_events(struct trace *trace)
3759 u64 first = ordered_events__first_time(&trace->oe.data);
3760 u64 flush = trace->oe.last - NSEC_PER_SEC;
3762 /* Is there some thing to flush.. */
3763 if (first && first < flush)
3764 return ordered_events__flush_time(&trace->oe.data, flush);
3769 static int trace__flush_events(struct trace *trace)
3771 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3774 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3778 if (!trace->sort_events)
3779 return __trace__deliver_event(trace, event);
3781 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3782 if (err && err != -1)
3785 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0);
3789 return trace__flush_events(trace);
3792 static int ordered_events__deliver_event(struct ordered_events *oe,
3793 struct ordered_event *event)
3795 struct trace *trace = container_of(oe, struct trace, oe.data);
3797 return __trace__deliver_event(trace, event->event);
3800 static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
3802 struct tep_format_field *field;
3803 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
3805 if (evsel->tp_format == NULL || fmt == NULL)
3808 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3809 if (strcmp(field->name, arg) == 0)
3815 static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3817 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3819 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3820 char *right = tok + 1, *right_end;
3825 while (isspace(*right))
3831 while (!isalpha(*left))
3832 if (++left == tok) {
3834 * Bail out, can't find the name of the argument that is being
3835 * used in the filter, let it try to set this filter, will fail later.
3840 right_end = right + 1;
3841 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
3844 if (isalpha(*right)) {
3845 struct syscall_arg_fmt *fmt;
3846 int left_size = tok - left,
3847 right_size = right_end - right;
3850 while (isspace(left[left_size - 1]))
3853 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3855 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
3857 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3858 arg, evsel->name, evsel->filter);
3862 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3863 arg, (int)(right - tok), tok, right_size, right);
3867 struct syscall_arg syscall_arg = {
3871 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
3872 char *n, expansion[19];
3873 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3874 int expansion_offset = right - new_filter;
3876 pr_debug("%s", expansion);
3878 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3879 pr_debug(" out of memory!\n");
3883 if (new_filter != evsel->filter)
3885 left = n + expansion_offset + expansion_lenght;
3888 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3889 right_size, right, arg, evsel->name, evsel->filter);
3893 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3894 arg, evsel->name, evsel->filter);
3904 if (new_filter != evsel->filter) {
3905 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
3906 evsel__set_filter(evsel, new_filter);
3913 static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3915 struct evlist *evlist = trace->evlist;
3916 struct evsel *evsel;
3918 evlist__for_each_entry(evlist, evsel) {
3919 if (evsel->filter == NULL)
3922 if (trace__expand_filter(trace, evsel)) {
3931 static int trace__run(struct trace *trace, int argc, const char **argv)
3933 struct evlist *evlist = trace->evlist;
3934 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3936 unsigned long before;
3937 const bool forks = argc > 0;
3938 bool draining = false;
3941 signal(SIGCHLD, sig_handler);
3943 if (!trace->raw_augmented_syscalls) {
3944 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3945 goto out_error_raw_syscalls;
3947 if (trace->trace_syscalls)
3948 trace->vfs_getname = evlist__add_vfs_getname(evlist);
3951 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3952 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3953 if (pgfault_maj == NULL)
3955 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3956 evlist__add(evlist, pgfault_maj);
3959 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3960 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3961 if (pgfault_min == NULL)
3963 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3964 evlist__add(evlist, pgfault_min);
3967 /* Enable ignoring missing threads when -u/-p option is defined. */
3968 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3971 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
3972 goto out_error_sched_stat_runtime;
3974 * If a global cgroup was set, apply it to all the events without an
3975 * explicit cgroup. I.e.:
3977 * trace -G A -e sched:*switch
3979 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3980 * _and_ sched:sched_switch to the 'A' cgroup, while:
3982 * trace -e sched:*switch -G A
3984 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3985 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3986 * a cgroup (on the root cgroup, sys wide, etc).
3990 * trace -G A -e sched:*switch -G B
3992 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3993 * to the 'B' cgroup.
3995 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3996 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3999 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
4001 err = evlist__create_maps(evlist, &trace->opts.target);
4003 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
4004 goto out_delete_evlist;
4007 err = trace__symbols_init(trace, evlist);
4009 fprintf(trace->output, "Problems initializing symbol libraries!\n");
4010 goto out_delete_evlist;
4013 evlist__config(evlist, &trace->opts, &callchain_param);
4016 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
4018 fprintf(trace->output, "Couldn't run the workload!\n");
4019 goto out_delete_evlist;
4023 err = evlist__open(evlist);
4025 goto out_error_open;
4027 err = bpf__apply_obj_config();
4029 char errbuf[BUFSIZ];
4031 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
4032 pr_err("ERROR: Apply config to BPF failed: %s\n",
4034 goto out_error_open;
4037 err = trace__set_filter_pids(trace);
4041 if (trace->syscalls.map)
4042 trace__init_syscalls_bpf_map(trace);
4044 if (trace->syscalls.prog_array.sys_enter)
4045 trace__init_syscalls_bpf_prog_array_maps(trace);
4047 if (trace->ev_qualifier_ids.nr > 0) {
4048 err = trace__set_ev_qualifier_filter(trace);
4052 if (trace->syscalls.events.sys_exit) {
4053 pr_debug("event qualifier tracepoint filter: %s\n",
4054 trace->syscalls.events.sys_exit->filter);
4059 * If the "close" syscall is not traced, then we will not have the
4060 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
4061 * fd->pathname table and were ending up showing the last value set by
4062 * syscalls opening a pathname and associating it with a descriptor or
4063 * reading it from /proc/pid/fd/ in cases where that doesn't make
4066 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
4069 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
4071 err = trace__expand_filters(trace, &evsel);
4073 goto out_delete_evlist;
4074 err = evlist__apply_filters(evlist, &evsel);
4076 goto out_error_apply_filters;
4078 if (trace->dump.map)
4079 bpf_map__fprintf(trace->dump.map, trace->output);
4081 err = evlist__mmap(evlist, trace->opts.mmap_pages);
4083 goto out_error_mmap;
4085 if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
4086 evlist__enable(evlist);
4089 evlist__start_workload(evlist);
4091 if (trace->opts.initial_delay) {
4092 usleep(trace->opts.initial_delay * 1000);
4093 evlist__enable(evlist);
4096 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
4097 evlist->core.threads->nr > 1 ||
4098 evlist__first(evlist)->core.attr.inherit;
4101 * Now that we already used evsel->core.attr to ask the kernel to setup the
4102 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
4103 * trace__resolve_callchain(), allowing per-event max-stack settings
4104 * to override an explicitly set --max-stack global setting.
4106 evlist__for_each_entry(evlist, evsel) {
4107 if (evsel__has_callchain(evsel) &&
4108 evsel->core.attr.sample_max_stack == 0)
4109 evsel->core.attr.sample_max_stack = trace->max_stack;
4112 before = trace->nr_events;
4114 for (i = 0; i < evlist->core.nr_mmaps; i++) {
4115 union perf_event *event;
4118 md = &evlist->mmap[i];
4119 if (perf_mmap__read_init(&md->core) < 0)
4122 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
4125 err = trace__deliver_event(trace, event);
4129 perf_mmap__consume(&md->core);
4134 if (done && !draining) {
4135 evlist__disable(evlist);
4139 perf_mmap__read_done(&md->core);
4142 if (trace->nr_events == before) {
4143 int timeout = done ? 100 : -1;
4145 if (!draining && evlist__poll(evlist, timeout) > 0) {
4146 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
4151 if (trace__flush_events(trace))
4159 thread__zput(trace->current);
4161 evlist__disable(evlist);
4163 if (trace->sort_events)
4164 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
4168 trace__fprintf_thread_summary(trace, trace->output);
4170 if (trace->show_tool_stats) {
4171 fprintf(trace->output, "Stats:\n "
4172 " vfs_getname : %" PRIu64 "\n"
4173 " proc_getname: %" PRIu64 "\n",
4174 trace->stats.vfs_getname,
4175 trace->stats.proc_getname);
4180 trace__symbols__exit(trace);
4181 evlist__free_syscall_tp_fields(evlist);
4182 evlist__delete(evlist);
4183 cgroup__put(trace->cgroup);
4184 trace->evlist = NULL;
4185 trace->live = false;
4188 char errbuf[BUFSIZ];
4190 out_error_sched_stat_runtime:
4191 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
4194 out_error_raw_syscalls:
4195 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
4199 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
4203 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
4206 fprintf(trace->output, "%s\n", errbuf);
4207 goto out_delete_evlist;
4209 out_error_apply_filters:
4210 fprintf(trace->output,
4211 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
4212 evsel->filter, evsel__name(evsel), errno,
4213 str_error_r(errno, errbuf, sizeof(errbuf)));
4214 goto out_delete_evlist;
4217 fprintf(trace->output, "Not enough memory to run!\n");
4218 goto out_delete_evlist;
4221 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4222 goto out_delete_evlist;
4225 static int trace__replay(struct trace *trace)
4227 const struct evsel_str_handler handlers[] = {
4228 { "probe:vfs_getname", trace__vfs_getname, },
4230 struct perf_data data = {
4232 .mode = PERF_DATA_MODE_READ,
4233 .force = trace->force,
4235 struct perf_session *session;
4236 struct evsel *evsel;
4239 trace->tool.sample = trace__process_sample;
4240 trace->tool.mmap = perf_event__process_mmap;
4241 trace->tool.mmap2 = perf_event__process_mmap2;
4242 trace->tool.comm = perf_event__process_comm;
4243 trace->tool.exit = perf_event__process_exit;
4244 trace->tool.fork = perf_event__process_fork;
4245 trace->tool.attr = perf_event__process_attr;
4246 trace->tool.tracing_data = perf_event__process_tracing_data;
4247 trace->tool.build_id = perf_event__process_build_id;
4248 trace->tool.namespaces = perf_event__process_namespaces;
4250 trace->tool.ordered_events = true;
4251 trace->tool.ordering_requires_timestamps = true;
4253 /* add tid to output */
4254 trace->multiple_threads = true;
4256 session = perf_session__new(&data, &trace->tool);
4257 if (IS_ERR(session))
4258 return PTR_ERR(session);
4260 if (trace->opts.target.pid)
4261 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4263 if (trace->opts.target.tid)
4264 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4266 if (symbol__init(&session->header.env) < 0)
4269 trace->host = &session->machines.host;
4271 err = perf_session__set_tracepoints_handlers(session, handlers);
4275 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4276 /* older kernels have syscalls tp versus raw_syscalls */
4278 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
4281 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
4282 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
4283 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4287 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4289 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
4291 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
4292 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
4293 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
4297 evlist__for_each_entry(session->evlist, evsel) {
4298 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4299 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4300 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4301 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
4302 evsel->handler = trace__pgfault;
4307 err = perf_session__process_events(session);
4309 pr_err("Failed to process events, error %d", err);
4311 else if (trace->summary)
4312 trace__fprintf_thread_summary(trace, trace->output);
4315 perf_session__delete(session);
4320 static size_t trace__fprintf_threads_header(FILE *fp)
4324 printed = fprintf(fp, "\n Summary of events:\n\n");
4329 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
4330 struct syscall_stats *stats;
4335 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
4336 struct syscall_stats *stats = source->priv;
4338 entry->syscall = source->i;
4339 entry->stats = stats;
4340 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
4343 static size_t thread__dump_stats(struct thread_trace *ttrace,
4344 struct trace *trace, FILE *fp)
4349 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
4351 if (syscall_stats == NULL)
4354 printed += fprintf(fp, "\n");
4356 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4357 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4358 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
4360 resort_rb__for_each_entry(nd, syscall_stats) {
4361 struct syscall_stats *stats = syscall_stats_entry->stats;
4363 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4364 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4365 double avg = avg_stats(&stats->stats);
4367 u64 n = (u64)stats->stats.n;
4369 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
4370 avg /= NSEC_PER_MSEC;
4372 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
4373 printed += fprintf(fp, " %-15s", sc->name);
4374 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4375 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
4376 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
4378 if (trace->errno_summary && stats->nr_failures) {
4379 const char *arch_name = perf_env__arch(trace->host->env);
4382 for (e = 0; e < stats->max_errno; ++e) {
4383 if (stats->errnos[e] != 0)
4384 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4390 resort_rb__delete(syscall_stats);
4391 printed += fprintf(fp, "\n\n");
4396 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
4399 struct thread_trace *ttrace = thread__priv(thread);
4405 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4407 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
4408 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
4409 printed += fprintf(fp, "%.1f%%", ratio);
4411 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4413 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
4415 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4416 else if (fputc('\n', fp) != EOF)
4419 printed += thread__dump_stats(ttrace, trace, fp);
4424 static unsigned long thread__nr_events(struct thread_trace *ttrace)
4426 return ttrace ? ttrace->nr_events : 0;
4429 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
4430 struct thread *thread;
4433 entry->thread = rb_entry(nd, struct thread, rb_node);
4436 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4438 size_t printed = trace__fprintf_threads_header(fp);
4442 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4443 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
4445 if (threads == NULL) {
4446 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4450 resort_rb__for_each_entry(nd, threads)
4451 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
4453 resort_rb__delete(threads);
4458 static int trace__set_duration(const struct option *opt, const char *str,
4459 int unset __maybe_unused)
4461 struct trace *trace = opt->value;
4463 trace->duration_filter = atof(str);
4467 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4468 int unset __maybe_unused)
4472 struct trace *trace = opt->value;
4474 * FIXME: introduce a intarray class, plain parse csv and create a
4475 * { int nr, int entries[] } struct...
4477 struct intlist *list = intlist__new(str);
4482 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4483 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4485 if (trace->filter_pids.entries == NULL)
4488 trace->filter_pids.entries[0] = getpid();
4490 for (i = 1; i < trace->filter_pids.nr; ++i)
4491 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4493 intlist__delete(list);
4499 static int trace__open_output(struct trace *trace, const char *filename)
4503 if (!stat(filename, &st) && st.st_size) {
4504 char oldname[PATH_MAX];
4506 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4508 rename(filename, oldname);
4511 trace->output = fopen(filename, "w");
4513 return trace->output == NULL ? -errno : 0;
4516 static int parse_pagefaults(const struct option *opt, const char *str,
4517 int unset __maybe_unused)
4519 int *trace_pgfaults = opt->value;
4521 if (strcmp(str, "all") == 0)
4522 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4523 else if (strcmp(str, "maj") == 0)
4524 *trace_pgfaults |= TRACE_PFMAJ;
4525 else if (strcmp(str, "min") == 0)
4526 *trace_pgfaults |= TRACE_PFMIN;
4533 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4535 struct evsel *evsel;
4537 evlist__for_each_entry(evlist, evsel) {
4538 if (evsel->handler == NULL)
4539 evsel->handler = handler;
4543 static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4545 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4548 struct syscall_fmt *scfmt = syscall_fmt__find(name);
4553 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4554 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4557 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4562 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
4564 struct evsel *evsel;
4566 evlist__for_each_entry(evlist, evsel) {
4567 if (evsel->priv || !evsel->tp_format)
4570 if (strcmp(evsel->tp_format->system, "syscalls")) {
4571 evsel__init_tp_arg_scnprintf(evsel);
4575 if (evsel__init_syscall_tp(evsel))
4578 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4579 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4581 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4584 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
4585 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4586 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4588 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4591 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
4599 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4600 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4601 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4603 * It'd be better to introduce a parse_options() variant that would return a
4604 * list with the terms it didn't match to an event...
4606 static int trace__parse_events_option(const struct option *opt, const char *str,
4607 int unset __maybe_unused)
4609 struct trace *trace = (struct trace *)opt->value;
4610 const char *s = str;
4611 char *sep = NULL, *lists[2] = { NULL, NULL, };
4612 int len = strlen(str) + 1, err = -1, list, idx;
4613 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4614 char group_name[PATH_MAX];
4615 struct syscall_fmt *fmt;
4617 if (strace_groups_dir == NULL)
4622 trace->not_ev_qualifier = true;
4626 if ((sep = strchr(s, ',')) != NULL)
4630 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4631 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4636 fmt = syscall_fmt__find_by_alias(s);
4641 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4642 if (access(group_name, R_OK) == 0)
4647 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4649 lists[list] = malloc(len);
4650 if (lists[list] == NULL)
4652 strcpy(lists[list], s);
4662 if (lists[1] != NULL) {
4663 struct strlist_config slist_config = {
4664 .dirname = strace_groups_dir,
4667 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4668 if (trace->ev_qualifier == NULL) {
4669 fputs("Not enough memory to parse event qualifier", trace->output);
4673 if (trace__validate_ev_qualifier(trace))
4675 trace->trace_syscalls = true;
4682 .value = &trace->evlist,
4684 err = parse_events_option(&o, lists[0], 0);
4687 free(strace_groups_dir);
4696 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4698 struct trace *trace = opt->value;
4700 if (!list_empty(&trace->evlist->core.entries)) {
4702 .value = &trace->evlist,
4704 return parse_cgroups(&o, str, unset);
4706 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4711 static int trace__config(const char *var, const char *value, void *arg)
4713 struct trace *trace = arg;
4716 if (!strcmp(var, "trace.add_events")) {
4717 trace->perfconfig_events = strdup(value);
4718 if (trace->perfconfig_events == NULL) {
4719 pr_err("Not enough memory for %s\n", "trace.add_events");
4722 } else if (!strcmp(var, "trace.show_timestamp")) {
4723 trace->show_tstamp = perf_config_bool(var, value);
4724 } else if (!strcmp(var, "trace.show_duration")) {
4725 trace->show_duration = perf_config_bool(var, value);
4726 } else if (!strcmp(var, "trace.show_arg_names")) {
4727 trace->show_arg_names = perf_config_bool(var, value);
4728 if (!trace->show_arg_names)
4729 trace->show_zeros = true;
4730 } else if (!strcmp(var, "trace.show_zeros")) {
4731 bool new_show_zeros = perf_config_bool(var, value);
4732 if (!trace->show_arg_names && !new_show_zeros) {
4733 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4736 trace->show_zeros = new_show_zeros;
4737 } else if (!strcmp(var, "trace.show_prefix")) {
4738 trace->show_string_prefix = perf_config_bool(var, value);
4739 } else if (!strcmp(var, "trace.no_inherit")) {
4740 trace->opts.no_inherit = perf_config_bool(var, value);
4741 } else if (!strcmp(var, "trace.args_alignment")) {
4742 int args_alignment = 0;
4743 if (perf_config_int(&args_alignment, var, value) == 0)
4744 trace->args_alignment = args_alignment;
4745 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4746 if (strcasecmp(value, "libtraceevent") == 0)
4747 trace->libtraceevent_print = true;
4748 else if (strcasecmp(value, "libbeauty") == 0)
4749 trace->libtraceevent_print = false;
4755 static void trace__exit(struct trace *trace)
4759 strlist__delete(trace->ev_qualifier);
4760 free(trace->ev_qualifier_ids.entries);
4761 if (trace->syscalls.table) {
4762 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4763 syscall__exit(&trace->syscalls.table[i]);
4764 free(trace->syscalls.table);
4766 syscalltbl__delete(trace->sctbl);
4767 zfree(&trace->perfconfig_events);
4770 int cmd_trace(int argc, const char **argv)
4772 const char *trace_usage[] = {
4773 "perf trace [<options>] [<command>]",
4774 "perf trace [<options>] -- <command> [<options>]",
4775 "perf trace record [<options>] [<command>]",
4776 "perf trace record [<options>] -- <command> [<options>]",
4779 struct trace trace = {
4785 .user_freq = UINT_MAX,
4786 .user_interval = ULLONG_MAX,
4787 .no_buffering = true,
4788 .mmap_pages = UINT_MAX,
4792 .show_tstamp = true,
4793 .show_duration = true,
4794 .show_arg_names = true,
4795 .args_alignment = 70,
4796 .trace_syscalls = false,
4797 .kernel_syscallchains = false,
4798 .max_stack = UINT_MAX,
4799 .max_events = ULONG_MAX,
4801 const char *map_dump_str = NULL;
4802 const char *output_name = NULL;
4803 const struct option trace_options[] = {
4804 OPT_CALLBACK('e', "event", &trace, "event",
4805 "event/syscall selector. use 'perf list' to list available events",
4806 trace__parse_events_option),
4807 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4808 "event filter", parse_filter),
4809 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4810 "show the thread COMM next to its id"),
4811 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4812 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4813 trace__parse_events_option),
4814 OPT_STRING('o', "output", &output_name, "file", "output file name"),
4815 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4816 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4817 "trace events on existing process id"),
4818 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4819 "trace events on existing thread id"),
4820 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4821 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4822 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4823 "system-wide collection from all CPUs"),
4824 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4825 "list of cpus to monitor"),
4826 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4827 "child tasks do not inherit counters"),
4828 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4829 "number of mmap data pages", evlist__parse_mmap_pages),
4830 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4832 OPT_CALLBACK(0, "duration", &trace, "float",
4833 "show only events with duration > N.M ms",
4834 trace__set_duration),
4835 #ifdef HAVE_LIBBPF_SUPPORT
4836 OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4838 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4839 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4840 OPT_BOOLEAN('T', "time", &trace.full_time,
4841 "Show full timestamp, not time relative to first start"),
4842 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4843 "Show only syscalls that failed"),
4844 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4845 "Show only syscall summary with statistics"),
4846 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4847 "Show all syscalls and summary with statistics"),
4848 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4849 "Show errno stats per syscall, use with -s or -S"),
4850 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4851 "Trace pagefaults", parse_pagefaults, "maj"),
4852 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4853 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4854 OPT_CALLBACK(0, "call-graph", &trace.opts,
4855 "record_mode[,record_size]", record_callchain_help,
4856 &record_parse_callchain_opt),
4857 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4858 "Use libtraceevent to print the tracepoint arguments."),
4859 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4860 "Show the kernel callchains on the syscall exit path"),
4861 OPT_ULONG(0, "max-events", &trace.max_events,
4862 "Set the maximum number of events to print, exit after that is reached. "),
4863 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4864 "Set the minimum stack depth when parsing the callchain, "
4865 "anything below the specified depth will be ignored."),
4866 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4867 "Set the maximum stack depth when parsing the callchain, "
4868 "anything beyond the specified depth will be ignored. "
4869 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4870 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4871 "Sort batch of events before processing, use if getting out of order events"),
4872 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4873 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4874 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4875 "per thread proc mmap processing timeout in ms"),
4876 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4877 trace__parse_cgroups),
4878 OPT_INTEGER('D', "delay", &trace.opts.initial_delay,
4879 "ms to wait before starting measurement after program "
4881 OPTS_EVSWITCH(&trace.evswitch),
4884 bool __maybe_unused max_stack_user_set = true;
4885 bool mmap_pages_user_set = true;
4886 struct evsel *evsel;
4887 const char * const trace_subcommands[] = { "record", NULL };
4891 signal(SIGSEGV, sighandler_dump_stack);
4892 signal(SIGFPE, sighandler_dump_stack);
4893 signal(SIGINT, sig_handler);
4895 trace.evlist = evlist__new();
4896 trace.sctbl = syscalltbl__new();
4898 if (trace.evlist == NULL || trace.sctbl == NULL) {
4899 pr_err("Not enough memory to run!\n");
4905 * Parsing .perfconfig may entail creating a BPF event, that may need
4906 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4907 * is too small. This affects just this process, not touching the
4908 * global setting. If it fails we'll get something in 'perf trace -v'
4909 * to help diagnose the problem.
4911 rlimit__bump_memlock();
4913 err = perf_config(trace__config, &trace);
4917 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4918 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4921 * Here we already passed thru trace__parse_events_option() and it has
4922 * already figured out if -e syscall_name, if not but if --event
4923 * foo:bar was used, the user is interested _just_ in those, say,
4924 * tracepoint events, not in the strace-like syscall-name-based mode.
4926 * This is important because we need to check if strace-like mode is
4927 * needed to decided if we should filter out the eBPF
4928 * __augmented_syscalls__ code, if it is in the mix, say, via
4929 * .perfconfig trace.add_events, and filter those out.
4931 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4932 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4933 trace.trace_syscalls = true;
4936 * Now that we have --verbose figured out, lets see if we need to parse
4937 * events from .perfconfig, so that if those events fail parsing, say some
4938 * BPF program fails, then we'll be able to use --verbose to see what went
4939 * wrong in more detail.
4941 if (trace.perfconfig_events != NULL) {
4942 struct parse_events_error parse_err;
4944 parse_events_error__init(&parse_err);
4945 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
4947 parse_events_error__print(&parse_err, trace.perfconfig_events);
4948 parse_events_error__exit(&parse_err);
4953 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4954 usage_with_options_msg(trace_usage, trace_options,
4955 "cgroup monitoring only available in system-wide mode");
4958 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4959 if (IS_ERR(evsel)) {
4960 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4961 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4966 trace.syscalls.events.augmented = evsel;
4968 evsel = evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4969 if (evsel == NULL) {
4970 pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4974 if (evsel->bpf_obj == NULL) {
4975 pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4979 trace.bpf_obj = evsel->bpf_obj;
4982 * If we have _just_ the augmenter event but don't have a
4983 * explicit --syscalls, then assume we want all strace-like
4986 if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
4987 trace.trace_syscalls = true;
4989 * So, if we have a syscall augmenter, but trace_syscalls, aka
4990 * strace-like syscall tracing is not set, then we need to trow
4991 * away the augmenter, i.e. all the events that were created
4992 * from that BPF object file.
4994 * This is more to fix the current .perfconfig trace.add_events
4995 * style of setting up the strace-like eBPF based syscall point
4996 * payload augmenter.
4998 * All this complexity will be avoided by adding an alternative
4999 * to trace.add_events in the form of
5000 * trace.bpf_augmented_syscalls, that will be only parsed if we
5003 * .perfconfig trace.add_events is still useful if we want, for
5004 * instance, have msr_write.msr in some .perfconfig profile based
5005 * 'perf trace --config determinism.profile' mode, where for some
5006 * particular goal/workload type we want a set of events and
5007 * output mode (with timings, etc) instead of having to add
5008 * all via the command line.
5010 * Also --config to specify an alternate .perfconfig file needs
5011 * to be implemented.
5013 if (!trace.trace_syscalls) {
5014 trace__delete_augmented_syscalls(&trace);
5016 trace__set_bpf_map_filtered_pids(&trace);
5017 trace__set_bpf_map_syscalls(&trace);
5018 trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
5022 err = bpf__setup_stdout(trace.evlist);
5024 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
5025 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
5032 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
5033 if (trace.dump.map == NULL) {
5034 pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
5039 if (trace.trace_pgfaults) {
5040 trace.opts.sample_address = true;
5041 trace.opts.sample_time = true;
5044 if (trace.opts.mmap_pages == UINT_MAX)
5045 mmap_pages_user_set = false;
5047 if (trace.max_stack == UINT_MAX) {
5048 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
5049 max_stack_user_set = false;
5052 #ifdef HAVE_DWARF_UNWIND_SUPPORT
5053 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
5054 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
5058 if (callchain_param.enabled) {
5059 if (!mmap_pages_user_set && geteuid() == 0)
5060 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
5062 symbol_conf.use_callchain = true;
5065 if (trace.evlist->core.nr_entries > 0) {
5066 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
5067 if (evlist__set_syscall_tp_fields(trace.evlist)) {
5068 perror("failed to set syscalls:* tracepoint fields");
5073 if (trace.sort_events) {
5074 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
5075 ordered_events__set_copy_on_queue(&trace.oe.data, true);
5079 * If we are augmenting syscalls, then combine what we put in the
5080 * __augmented_syscalls__ BPF map with what is in the
5081 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5082 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5084 * We'll switch to look at two BPF maps, one for sys_enter and the
5085 * other for sys_exit when we start augmenting the sys_exit paths with
5086 * buffers that are being copied from kernel to userspace, think 'read'
5089 if (trace.syscalls.events.augmented) {
5090 evlist__for_each_entry(trace.evlist, evsel) {
5091 bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
5093 if (raw_syscalls_sys_exit) {
5094 trace.raw_augmented_syscalls = true;
5095 goto init_augmented_syscall_tp;
5098 if (trace.syscalls.events.augmented->priv == NULL &&
5099 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
5100 struct evsel *augmented = trace.syscalls.events.augmented;
5101 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5102 evsel__init_augmented_syscall_tp_args(augmented))
5105 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5106 * Above we made sure we can get from the payload the tp fields
5107 * that we get from syscalls:sys_enter tracefs format file.
5109 augmented->handler = trace__sys_enter;
5111 * Now we do the same for the *syscalls:sys_enter event so that
5112 * if we handle it directly, i.e. if the BPF prog returns 0 so
5113 * as not to filter it, then we'll handle it just like we would
5114 * for the BPF_OUTPUT one:
5116 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5117 evsel__init_augmented_syscall_tp_args(evsel))
5119 evsel->handler = trace__sys_enter;
5122 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
5123 struct syscall_tp *sc;
5124 init_augmented_syscall_tp:
5125 if (evsel__init_augmented_syscall_tp(evsel, evsel))
5127 sc = __evsel__syscall_tp(evsel);
5129 * For now with BPF raw_augmented we hook into
5130 * raw_syscalls:sys_enter and there we get all
5131 * 6 syscall args plus the tracepoint common
5132 * fields and the syscall_nr (another long).
5133 * So we check if that is the case and if so
5134 * don't look after the sc->args_size but
5135 * always after the full raw_syscalls:sys_enter
5136 * payload, which is fixed.
5138 * We'll revisit this later to pass
5139 * s->args_size to the BPF augmenter (now
5140 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5141 * so that it copies only what we need for each
5142 * syscall, like what happens when we use
5143 * syscalls:sys_enter_NAME, so that we reduce
5144 * the kernel/userspace traffic to just what is
5145 * needed for each syscall.
5147 if (trace.raw_augmented_syscalls)
5148 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
5149 evsel__init_augmented_syscall_tp_ret(evsel);
5150 evsel->handler = trace__sys_exit;
5155 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5156 return trace__record(&trace, argc-1, &argv[1]);
5158 /* Using just --errno-summary will trigger --summary */
5159 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5160 trace.summary_only = true;
5162 /* summary_only implies summary option, but don't overwrite summary if set */
5163 if (trace.summary_only)
5164 trace.summary = trace.summary_only;
5166 if (output_name != NULL) {
5167 err = trace__open_output(&trace, output_name);
5169 perror("failed to create output file");
5174 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5178 err = target__validate(&trace.opts.target);
5180 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5181 fprintf(trace.output, "%s", bf);
5185 err = target__parse_uid(&trace.opts.target);
5187 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5188 fprintf(trace.output, "%s", bf);
5192 if (!argc && target__none(&trace.opts.target))
5193 trace.opts.target.system_wide = true;
5196 err = trace__replay(&trace);
5198 err = trace__run(&trace, argc, argv);
5201 if (output_name != NULL)
5202 fclose(trace.output);
5204 trace__exit(&trace);