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 pid_t workload_pid = -1;
1540 static bool done = false;
1541 static bool interrupted = false;
1543 static void sighandler_interrupt(int sig __maybe_unused)
1545 done = interrupted = true;
1548 static void sighandler_chld(int sig __maybe_unused, siginfo_t *info,
1549 void *context __maybe_unused)
1551 if (info->si_pid == workload_pid)
1555 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1559 if (trace->multiple_threads) {
1560 if (trace->show_comm)
1561 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1562 printed += fprintf(fp, "%d ", thread->tid);
1568 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1569 u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1573 if (trace->show_tstamp)
1574 printed = trace__fprintf_tstamp(trace, tstamp, fp);
1575 if (trace->show_duration)
1576 printed += fprintf_duration(duration, duration_calculated, fp);
1577 return printed + trace__fprintf_comm_tid(trace, thread, fp);
1580 static int trace__process_event(struct trace *trace, struct machine *machine,
1581 union perf_event *event, struct perf_sample *sample)
1585 switch (event->header.type) {
1586 case PERF_RECORD_LOST:
1587 color_fprintf(trace->output, PERF_COLOR_RED,
1588 "LOST %" PRIu64 " events!\n", event->lost.lost);
1589 ret = machine__process_lost_event(machine, event, sample);
1592 ret = machine__process_event(machine, event, sample);
1599 static int trace__tool_process(struct perf_tool *tool,
1600 union perf_event *event,
1601 struct perf_sample *sample,
1602 struct machine *machine)
1604 struct trace *trace = container_of(tool, struct trace, tool);
1605 return trace__process_event(trace, machine, event, sample);
1608 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1610 struct machine *machine = vmachine;
1612 if (machine->kptr_restrict_warned)
1615 if (symbol_conf.kptr_restrict) {
1616 pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1617 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1618 "Kernel samples will not be resolved.\n");
1619 machine->kptr_restrict_warned = true;
1623 return machine__resolve_kernel_addr(vmachine, addrp, modp);
1626 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1628 int err = symbol__init(NULL);
1633 trace->host = machine__new_host();
1634 if (trace->host == NULL)
1637 err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1641 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1642 evlist->core.threads, trace__tool_process,
1651 static void trace__symbols__exit(struct trace *trace)
1653 machine__exit(trace->host);
1659 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1663 if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1664 nr_args = sc->fmt->nr_args;
1666 sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1667 if (sc->arg_fmt == NULL)
1670 for (idx = 0; idx < nr_args; ++idx) {
1672 sc->arg_fmt[idx] = sc->fmt->arg[idx];
1675 sc->nr_args = nr_args;
1679 static struct syscall_arg_fmt syscall_arg_fmts__by_name[] = {
1680 { .name = "msr", .scnprintf = SCA_X86_MSR, .strtoul = STUL_X86_MSR, },
1681 { .name = "vector", .scnprintf = SCA_X86_IRQ_VECTORS, .strtoul = STUL_X86_IRQ_VECTORS, },
1684 static int syscall_arg_fmt__cmp(const void *name, const void *fmtp)
1686 const struct syscall_arg_fmt *fmt = fmtp;
1687 return strcmp(name, fmt->name);
1690 static struct syscall_arg_fmt *
1691 __syscall_arg_fmt__find_by_name(struct syscall_arg_fmt *fmts, const int nmemb, const char *name)
1693 return bsearch(name, fmts, nmemb, sizeof(struct syscall_arg_fmt), syscall_arg_fmt__cmp);
1696 static struct syscall_arg_fmt *syscall_arg_fmt__find_by_name(const char *name)
1698 const int nmemb = ARRAY_SIZE(syscall_arg_fmts__by_name);
1699 return __syscall_arg_fmt__find_by_name(syscall_arg_fmts__by_name, nmemb, name);
1702 static struct tep_format_field *
1703 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1705 struct tep_format_field *last_field = NULL;
1708 for (; field; field = field->next, ++arg) {
1714 len = strlen(field->name);
1716 if (strcmp(field->type, "const char *") == 0 &&
1717 ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1718 strstr(field->name, "path") != NULL))
1719 arg->scnprintf = SCA_FILENAME;
1720 else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1721 arg->scnprintf = SCA_PTR;
1722 else if (strcmp(field->type, "pid_t") == 0)
1723 arg->scnprintf = SCA_PID;
1724 else if (strcmp(field->type, "umode_t") == 0)
1725 arg->scnprintf = SCA_MODE_T;
1726 else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstr(field->type, "char")) {
1727 arg->scnprintf = SCA_CHAR_ARRAY;
1728 arg->nr_entries = field->arraylen;
1729 } else if ((strcmp(field->type, "int") == 0 ||
1730 strcmp(field->type, "unsigned int") == 0 ||
1731 strcmp(field->type, "long") == 0) &&
1732 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1734 * /sys/kernel/tracing/events/syscalls/sys_enter*
1735 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1740 arg->scnprintf = SCA_FD;
1742 struct syscall_arg_fmt *fmt = syscall_arg_fmt__find_by_name(field->name);
1745 arg->scnprintf = fmt->scnprintf;
1746 arg->strtoul = fmt->strtoul;
1754 static int syscall__set_arg_fmts(struct syscall *sc)
1756 struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1759 sc->args_size = last_field->offset + last_field->size;
1764 static int trace__read_syscall_info(struct trace *trace, int id)
1768 const char *name = syscalltbl__name(trace->sctbl, id);
1770 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
1771 if (trace->syscalls.table == NULL) {
1772 trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1773 if (trace->syscalls.table == NULL)
1777 if (id > trace->sctbl->syscalls.max_id || (id == 0 && trace->syscalls.table == NULL)) {
1778 // When using libaudit we don't know beforehand what is the max syscall id
1779 struct syscall *table = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1784 // Need to memset from offset 0 and +1 members if brand new
1785 if (trace->syscalls.table == NULL)
1786 memset(table, 0, (id + 1) * sizeof(*sc));
1788 memset(table + trace->sctbl->syscalls.max_id + 1, 0, (id - trace->sctbl->syscalls.max_id) * sizeof(*sc));
1790 trace->syscalls.table = table;
1791 trace->sctbl->syscalls.max_id = id;
1794 sc = trace->syscalls.table + id;
1795 if (sc->nonexistent)
1799 sc->nonexistent = true;
1804 sc->fmt = syscall_fmt__find(sc->name);
1806 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1807 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1809 if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1810 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1811 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1814 if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1817 if (IS_ERR(sc->tp_format))
1818 return PTR_ERR(sc->tp_format);
1820 sc->args = sc->tp_format->format.fields;
1822 * We need to check and discard the first variable '__syscall_nr'
1823 * or 'nr' that mean the syscall number. It is needless here.
1824 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1826 if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1827 sc->args = sc->args->next;
1831 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1832 sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1834 return syscall__set_arg_fmts(sc);
1837 static int evsel__init_tp_arg_scnprintf(struct evsel *evsel)
1839 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
1842 syscall_arg_fmt__init_array(fmt, evsel->tp_format->format.fields);
1849 static int intcmp(const void *a, const void *b)
1851 const int *one = a, *another = b;
1853 return *one - *another;
1856 static int trace__validate_ev_qualifier(struct trace *trace)
1859 bool printed_invalid_prefix = false;
1860 struct str_node *pos;
1861 size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1863 trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1864 sizeof(trace->ev_qualifier_ids.entries[0]));
1866 if (trace->ev_qualifier_ids.entries == NULL) {
1867 fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1873 strlist__for_each_entry(pos, trace->ev_qualifier) {
1874 const char *sc = pos->s;
1875 int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1878 id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1882 if (!printed_invalid_prefix) {
1883 pr_debug("Skipping unknown syscalls: ");
1884 printed_invalid_prefix = true;
1893 trace->ev_qualifier_ids.entries[nr_used++] = id;
1894 if (match_next == -1)
1898 id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1901 if (nr_allocated == nr_used) {
1905 entries = realloc(trace->ev_qualifier_ids.entries,
1906 nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1907 if (entries == NULL) {
1909 fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1912 trace->ev_qualifier_ids.entries = entries;
1914 trace->ev_qualifier_ids.entries[nr_used++] = id;
1918 trace->ev_qualifier_ids.nr = nr_used;
1919 qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1921 if (printed_invalid_prefix)
1925 zfree(&trace->ev_qualifier_ids.entries);
1926 trace->ev_qualifier_ids.nr = 0;
1930 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1932 bool in_ev_qualifier;
1934 if (trace->ev_qualifier_ids.nr == 0)
1937 in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1938 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1940 if (in_ev_qualifier)
1941 return !trace->not_ev_qualifier;
1943 return trace->not_ev_qualifier;
1947 * args is to be interpreted as a series of longs but we need to handle
1948 * 8-byte unaligned accesses. args points to raw_data within the event
1949 * and raw_data is guaranteed to be 8-byte unaligned because it is
1950 * preceded by raw_size which is a u32. So we need to copy args to a temp
1951 * variable to read it. Most notably this avoids extended load instructions
1952 * on unaligned addresses
1954 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1957 unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1959 memcpy(&val, p, sizeof(val));
1963 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1964 struct syscall_arg *arg)
1966 if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1967 return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1969 return scnprintf(bf, size, "arg%d: ", arg->idx);
1973 * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1974 * as mount 'flags' argument that needs ignoring some magic flag, see comment
1975 * in tools/perf/trace/beauty/mount_flags.c
1977 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
1979 if (fmt && fmt->mask_val)
1980 return fmt->mask_val(arg, val);
1985 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1986 struct syscall_arg *arg, unsigned long val)
1988 if (fmt && fmt->scnprintf) {
1991 arg->parm = fmt->parm;
1992 return fmt->scnprintf(bf, size, arg);
1994 return scnprintf(bf, size, "%ld", val);
1997 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1998 unsigned char *args, void *augmented_args, int augmented_args_size,
1999 struct trace *trace, struct thread *thread)
2004 struct syscall_arg arg = {
2007 .size = augmented_args_size,
2008 .args = augmented_args,
2014 .show_string_prefix = trace->show_string_prefix,
2016 struct thread_trace *ttrace = thread__priv(thread);
2019 * Things like fcntl will set this in its 'cmd' formatter to pick the
2020 * right formatter for the return value (an fd? file flags?), which is
2021 * not needed for syscalls that always return a given type, say an fd.
2023 ttrace->ret_scnprintf = NULL;
2025 if (sc->args != NULL) {
2026 struct tep_format_field *field;
2028 for (field = sc->args; field;
2029 field = field->next, ++arg.idx, bit <<= 1) {
2033 arg.fmt = &sc->arg_fmt[arg.idx];
2034 val = syscall_arg__val(&arg, arg.idx);
2036 * Some syscall args need some mask, most don't and
2037 * return val untouched.
2039 val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
2042 * Suppress this argument if its value is zero and
2043 * and we don't have a string associated in an
2047 !trace->show_zeros &&
2049 (sc->arg_fmt[arg.idx].show_zero ||
2050 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
2051 sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
2052 sc->arg_fmt[arg.idx].parm))
2055 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2057 if (trace->show_arg_names)
2058 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2060 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
2061 bf + printed, size - printed, &arg, val);
2063 } else if (IS_ERR(sc->tp_format)) {
2065 * If we managed to read the tracepoint /format file, then we
2066 * may end up not having any args, like with gettid(), so only
2067 * print the raw args when we didn't manage to read it.
2069 while (arg.idx < sc->nr_args) {
2072 val = syscall_arg__val(&arg, arg.idx);
2074 printed += scnprintf(bf + printed, size - printed, ", ");
2075 printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
2076 printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
2086 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
2087 union perf_event *event,
2088 struct perf_sample *sample);
2090 static struct syscall *trace__syscall_info(struct trace *trace,
2091 struct evsel *evsel, int id)
2098 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
2099 * before that, leaving at a higher verbosity level till that is
2100 * explained. Reproduced with plain ftrace with:
2102 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
2103 * grep "NR -1 " /t/trace_pipe
2105 * After generating some load on the machine.
2109 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
2110 id, evsel__name(evsel), ++n);
2117 #ifdef HAVE_SYSCALL_TABLE_SUPPORT
2118 if (id > trace->sctbl->syscalls.max_id) {
2120 if (id >= trace->sctbl->syscalls.max_id) {
2122 * With libaudit we don't know beforehand what is the max_id,
2123 * so we let trace__read_syscall_info() figure that out as we
2124 * go on reading syscalls.
2126 err = trace__read_syscall_info(trace, id);
2132 if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
2133 (err = trace__read_syscall_info(trace, id)) != 0)
2136 if (trace->syscalls.table[id].name == NULL) {
2137 if (trace->syscalls.table[id].nonexistent)
2142 return &trace->syscalls.table[id];
2146 char sbuf[STRERR_BUFSIZE];
2147 fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
2148 if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
2149 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
2150 fputs(" information\n", trace->output);
2155 struct syscall_stats {
2162 static void thread__update_stats(struct thread *thread, struct thread_trace *ttrace,
2163 int id, struct perf_sample *sample, long err, bool errno_summary)
2165 struct int_node *inode;
2166 struct syscall_stats *stats;
2169 inode = intlist__findnew(ttrace->syscall_stats, id);
2173 stats = inode->priv;
2174 if (stats == NULL) {
2175 stats = malloc(sizeof(*stats));
2179 stats->nr_failures = 0;
2180 stats->max_errno = 0;
2181 stats->errnos = NULL;
2182 init_stats(&stats->stats);
2183 inode->priv = stats;
2186 if (ttrace->entry_time && sample->time > ttrace->entry_time)
2187 duration = sample->time - ttrace->entry_time;
2189 update_stats(&stats->stats, duration);
2192 ++stats->nr_failures;
2198 if (err > stats->max_errno) {
2199 u32 *new_errnos = realloc(stats->errnos, err * sizeof(u32));
2202 memset(new_errnos + stats->max_errno, 0, (err - stats->max_errno) * sizeof(u32));
2204 pr_debug("Not enough memory for errno stats for thread \"%s\"(%d/%d), results will be incomplete\n",
2205 thread__comm_str(thread), thread->pid_, thread->tid);
2209 stats->errnos = new_errnos;
2210 stats->max_errno = err;
2213 ++stats->errnos[err - 1];
2217 static int trace__printf_interrupted_entry(struct trace *trace)
2219 struct thread_trace *ttrace;
2223 if (trace->failure_only || trace->current == NULL)
2226 ttrace = thread__priv(trace->current);
2228 if (!ttrace->entry_pending)
2231 printed = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
2232 printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
2234 if (len < trace->args_alignment - 4)
2235 printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
2237 printed += fprintf(trace->output, " ...\n");
2239 ttrace->entry_pending = false;
2240 ++trace->nr_events_printed;
2245 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
2246 struct perf_sample *sample, struct thread *thread)
2250 if (trace->print_sample) {
2251 double ts = (double)sample->time / NSEC_PER_MSEC;
2253 printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
2254 evsel__name(evsel), ts,
2255 thread__comm_str(thread),
2256 sample->pid, sample->tid, sample->cpu);
2262 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
2264 void *augmented_args = NULL;
2266 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
2267 * and there we get all 6 syscall args plus the tracepoint common fields
2268 * that gets calculated at the start and the syscall_nr (another long).
2269 * So we check if that is the case and if so don't look after the
2270 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
2273 * We'll revisit this later to pass s->args_size to the BPF augmenter
2274 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
2275 * copies only what we need for each syscall, like what happens when we
2276 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
2277 * traffic to just what is needed for each syscall.
2279 int args_size = raw_augmented_args_size ?: sc->args_size;
2281 *augmented_args_size = sample->raw_size - args_size;
2282 if (*augmented_args_size > 0)
2283 augmented_args = sample->raw_data + args_size;
2285 return augmented_args;
2288 static void syscall__exit(struct syscall *sc)
2296 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
2297 union perf_event *event __maybe_unused,
2298 struct perf_sample *sample)
2303 struct thread *thread;
2304 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2305 int augmented_args_size = 0;
2306 void *augmented_args = NULL;
2307 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2308 struct thread_trace *ttrace;
2313 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2314 ttrace = thread__trace(thread, trace->output);
2318 trace__fprintf_sample(trace, evsel, sample, thread);
2320 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2322 if (ttrace->entry_str == NULL) {
2323 ttrace->entry_str = malloc(trace__entry_str_size);
2324 if (!ttrace->entry_str)
2328 if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2329 trace__printf_interrupted_entry(trace);
2331 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2332 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2333 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2334 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2335 * so when handling, say the openat syscall, we end up getting 6 args for the
2336 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2337 * thinking that the extra 2 u64 args are the augmented filename, so just check
2338 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2340 if (evsel != trace->syscalls.events.sys_enter)
2341 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2342 ttrace->entry_time = sample->time;
2343 msg = ttrace->entry_str;
2344 printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2346 printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2347 args, augmented_args, augmented_args_size, trace, thread);
2350 if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2353 trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2354 printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2355 if (trace->args_alignment > printed)
2356 alignment = trace->args_alignment - printed;
2357 fprintf(trace->output, "%*s= ?\n", alignment, " ");
2360 ttrace->entry_pending = true;
2361 /* See trace__vfs_getname & trace__sys_exit */
2362 ttrace->filename.pending_open = false;
2365 if (trace->current != thread) {
2366 thread__put(trace->current);
2367 trace->current = thread__get(thread);
2371 thread__put(thread);
2375 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2376 struct perf_sample *sample)
2378 struct thread_trace *ttrace;
2379 struct thread *thread;
2380 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2381 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2383 void *args, *augmented_args = NULL;
2384 int augmented_args_size;
2389 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2390 ttrace = thread__trace(thread, trace->output);
2392 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2393 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2398 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2399 augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2400 syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2401 fprintf(trace->output, "%s", msg);
2404 thread__put(thread);
2408 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2409 struct perf_sample *sample,
2410 struct callchain_cursor *cursor)
2412 struct addr_location al;
2413 int max_stack = evsel->core.attr.sample_max_stack ?
2414 evsel->core.attr.sample_max_stack :
2418 if (machine__resolve(trace->host, &al, sample) < 0)
2421 err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2422 addr_location__put(&al);
2426 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2428 /* TODO: user-configurable print_opts */
2429 const unsigned int print_opts = EVSEL__PRINT_SYM |
2431 EVSEL__PRINT_UNKNOWN_AS_ADDR;
2433 return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
2436 static const char *errno_to_name(struct evsel *evsel, int err)
2438 struct perf_env *env = evsel__env(evsel);
2439 const char *arch_name = perf_env__arch(env);
2441 return arch_syscalls__strerrno(arch_name, err);
2444 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2445 union perf_event *event __maybe_unused,
2446 struct perf_sample *sample)
2450 bool duration_calculated = false;
2451 struct thread *thread;
2452 int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2453 int alignment = trace->args_alignment;
2454 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2455 struct thread_trace *ttrace;
2460 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2461 ttrace = thread__trace(thread, trace->output);
2465 trace__fprintf_sample(trace, evsel, sample, thread);
2467 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2470 thread__update_stats(thread, ttrace, id, sample, ret, trace->errno_summary);
2472 if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2473 trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2474 ttrace->filename.pending_open = false;
2475 ++trace->stats.vfs_getname;
2478 if (ttrace->entry_time) {
2479 duration = sample->time - ttrace->entry_time;
2480 if (trace__filter_duration(trace, duration))
2482 duration_calculated = true;
2483 } else if (trace->duration_filter)
2486 if (sample->callchain) {
2487 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2488 if (callchain_ret == 0) {
2489 if (callchain_cursor.nr < trace->min_stack)
2495 if (trace->summary_only || (ret >= 0 && trace->failure_only))
2498 trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2500 if (ttrace->entry_pending) {
2501 printed = fprintf(trace->output, "%s", ttrace->entry_str);
2503 printed += fprintf(trace->output, " ... [");
2504 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2506 printed += fprintf(trace->output, "]: %s()", sc->name);
2509 printed++; /* the closing ')' */
2511 if (alignment > printed)
2512 alignment -= printed;
2516 fprintf(trace->output, ")%*s= ", alignment, " ");
2518 if (sc->fmt == NULL) {
2522 fprintf(trace->output, "%ld", ret);
2523 } else if (ret < 0) {
2525 char bf[STRERR_BUFSIZE];
2526 const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2527 *e = errno_to_name(evsel, -ret);
2529 fprintf(trace->output, "-1 %s (%s)", e, emsg);
2531 } else if (ret == 0 && sc->fmt->timeout)
2532 fprintf(trace->output, "0 (Timeout)");
2533 else if (ttrace->ret_scnprintf) {
2535 struct syscall_arg arg = {
2540 ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2541 ttrace->ret_scnprintf = NULL;
2542 fprintf(trace->output, "%s", bf);
2543 } else if (sc->fmt->hexret)
2544 fprintf(trace->output, "%#lx", ret);
2545 else if (sc->fmt->errpid) {
2546 struct thread *child = machine__find_thread(trace->host, ret, ret);
2548 if (child != NULL) {
2549 fprintf(trace->output, "%ld", ret);
2550 if (child->comm_set)
2551 fprintf(trace->output, " (%s)", thread__comm_str(child));
2557 fputc('\n', trace->output);
2560 * We only consider an 'event' for the sake of --max-events a non-filtered
2561 * sys_enter + sys_exit and other tracepoint events.
2563 if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2566 if (callchain_ret > 0)
2567 trace__fprintf_callchain(trace, sample);
2568 else if (callchain_ret < 0)
2569 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2571 ttrace->entry_pending = false;
2574 thread__put(thread);
2578 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2579 union perf_event *event __maybe_unused,
2580 struct perf_sample *sample)
2582 struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2583 struct thread_trace *ttrace;
2584 size_t filename_len, entry_str_len, to_move;
2585 ssize_t remaining_space;
2587 const char *filename = evsel__rawptr(evsel, sample, "pathname");
2592 ttrace = thread__priv(thread);
2596 filename_len = strlen(filename);
2597 if (filename_len == 0)
2600 if (ttrace->filename.namelen < filename_len) {
2601 char *f = realloc(ttrace->filename.name, filename_len + 1);
2606 ttrace->filename.namelen = filename_len;
2607 ttrace->filename.name = f;
2610 strcpy(ttrace->filename.name, filename);
2611 ttrace->filename.pending_open = true;
2613 if (!ttrace->filename.ptr)
2616 entry_str_len = strlen(ttrace->entry_str);
2617 remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2618 if (remaining_space <= 0)
2621 if (filename_len > (size_t)remaining_space) {
2622 filename += filename_len - remaining_space;
2623 filename_len = remaining_space;
2626 to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2627 pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2628 memmove(pos + filename_len, pos, to_move);
2629 memcpy(pos, filename, filename_len);
2631 ttrace->filename.ptr = 0;
2632 ttrace->filename.entry_str_pos = 0;
2634 thread__put(thread);
2639 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2640 union perf_event *event __maybe_unused,
2641 struct perf_sample *sample)
2643 u64 runtime = evsel__intval(evsel, sample, "runtime");
2644 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2645 struct thread *thread = machine__findnew_thread(trace->host,
2648 struct thread_trace *ttrace = thread__trace(thread, trace->output);
2653 ttrace->runtime_ms += runtime_ms;
2654 trace->runtime_ms += runtime_ms;
2656 thread__put(thread);
2660 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2662 evsel__strval(evsel, sample, "comm"),
2663 (pid_t)evsel__intval(evsel, sample, "pid"),
2665 evsel__intval(evsel, sample, "vruntime"));
2669 static int bpf_output__printer(enum binary_printer_ops op,
2670 unsigned int val, void *extra __maybe_unused, FILE *fp)
2672 unsigned char ch = (unsigned char)val;
2675 case BINARY_PRINT_CHAR_DATA:
2676 return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2677 case BINARY_PRINT_DATA_BEGIN:
2678 case BINARY_PRINT_LINE_BEGIN:
2679 case BINARY_PRINT_ADDR:
2680 case BINARY_PRINT_NUM_DATA:
2681 case BINARY_PRINT_NUM_PAD:
2682 case BINARY_PRINT_SEP:
2683 case BINARY_PRINT_CHAR_PAD:
2684 case BINARY_PRINT_LINE_END:
2685 case BINARY_PRINT_DATA_END:
2693 static void bpf_output__fprintf(struct trace *trace,
2694 struct perf_sample *sample)
2696 binary__fprintf(sample->raw_data, sample->raw_size, 8,
2697 bpf_output__printer, NULL, trace->output);
2698 ++trace->nr_events_printed;
2701 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2702 struct thread *thread, void *augmented_args, int augmented_args_size)
2705 size_t size = sizeof(bf);
2706 struct tep_format_field *field = evsel->tp_format->format.fields;
2707 struct syscall_arg_fmt *arg = __evsel__syscall_arg_fmt(evsel);
2711 struct syscall_arg syscall_arg = {
2713 .size = augmented_args_size,
2714 .args = augmented_args,
2720 .show_string_prefix = trace->show_string_prefix,
2723 for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2724 if (syscall_arg.mask & bit)
2727 syscall_arg.len = 0;
2728 syscall_arg.fmt = arg;
2729 if (field->flags & TEP_FIELD_IS_ARRAY) {
2730 int offset = field->offset;
2732 if (field->flags & TEP_FIELD_IS_DYNAMIC) {
2733 offset = format_field__intval(field, sample, evsel->needs_swap);
2734 syscall_arg.len = offset >> 16;
2736 if (field->flags & TEP_FIELD_IS_RELATIVE)
2737 offset += field->offset + field->size;
2740 val = (uintptr_t)(sample->raw_data + offset);
2742 val = format_field__intval(field, sample, evsel->needs_swap);
2744 * Some syscall args need some mask, most don't and
2745 * return val untouched.
2747 val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2750 * Suppress this argument if its value is zero and
2751 * and we don't have a string associated in an
2755 !trace->show_zeros &&
2756 !((arg->show_zero ||
2757 arg->scnprintf == SCA_STRARRAY ||
2758 arg->scnprintf == SCA_STRARRAYS) &&
2762 printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2765 * XXX Perhaps we should have a show_tp_arg_names,
2766 * leaving show_arg_names just for syscalls?
2768 if (1 || trace->show_arg_names)
2769 printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2771 printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2774 return printed + fprintf(trace->output, "%s", bf);
2777 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2778 union perf_event *event __maybe_unused,
2779 struct perf_sample *sample)
2781 struct thread *thread;
2782 int callchain_ret = 0;
2784 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2785 * this event's max_events having been hit and this is an entry coming
2786 * from the ring buffer that we should discard, since the max events
2787 * have already been considered/printed.
2789 if (evsel->disabled)
2792 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2794 if (sample->callchain) {
2795 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2796 if (callchain_ret == 0) {
2797 if (callchain_cursor.nr < trace->min_stack)
2803 trace__printf_interrupted_entry(trace);
2804 trace__fprintf_tstamp(trace, sample->time, trace->output);
2806 if (trace->trace_syscalls && trace->show_duration)
2807 fprintf(trace->output, "( ): ");
2810 trace__fprintf_comm_tid(trace, thread, trace->output);
2812 if (evsel == trace->syscalls.events.augmented) {
2813 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2814 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2817 fprintf(trace->output, "%s(", sc->name);
2818 trace__fprintf_sys_enter(trace, evsel, sample);
2819 fputc(')', trace->output);
2824 * XXX: Not having the associated syscall info or not finding/adding
2825 * the thread should never happen, but if it does...
2826 * fall thru and print it as a bpf_output event.
2830 fprintf(trace->output, "%s(", evsel->name);
2832 if (evsel__is_bpf_output(evsel)) {
2833 bpf_output__fprintf(trace, sample);
2834 } else if (evsel->tp_format) {
2835 if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2836 trace__fprintf_sys_enter(trace, evsel, sample)) {
2837 if (trace->libtraceevent_print) {
2838 event_format__fprintf(evsel->tp_format, sample->cpu,
2839 sample->raw_data, sample->raw_size,
2842 trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2848 fprintf(trace->output, ")\n");
2850 if (callchain_ret > 0)
2851 trace__fprintf_callchain(trace, sample);
2852 else if (callchain_ret < 0)
2853 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2855 ++trace->nr_events_printed;
2857 if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2858 evsel__disable(evsel);
2859 evsel__close(evsel);
2862 thread__put(thread);
2866 static void print_location(FILE *f, struct perf_sample *sample,
2867 struct addr_location *al,
2868 bool print_dso, bool print_sym)
2871 if ((verbose > 0 || print_dso) && al->map)
2872 fprintf(f, "%s@", al->map->dso->long_name);
2874 if ((verbose > 0 || print_sym) && al->sym)
2875 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2876 al->addr - al->sym->start);
2878 fprintf(f, "0x%" PRIx64, al->addr);
2880 fprintf(f, "0x%" PRIx64, sample->addr);
2883 static int trace__pgfault(struct trace *trace,
2884 struct evsel *evsel,
2885 union perf_event *event __maybe_unused,
2886 struct perf_sample *sample)
2888 struct thread *thread;
2889 struct addr_location al;
2890 char map_type = 'd';
2891 struct thread_trace *ttrace;
2893 int callchain_ret = 0;
2895 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2897 if (sample->callchain) {
2898 callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2899 if (callchain_ret == 0) {
2900 if (callchain_cursor.nr < trace->min_stack)
2906 ttrace = thread__trace(thread, trace->output);
2910 if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2915 if (trace->summary_only)
2918 thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2920 trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2922 fprintf(trace->output, "%sfault [",
2923 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2926 print_location(trace->output, sample, &al, false, true);
2928 fprintf(trace->output, "] => ");
2930 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2933 thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2941 print_location(trace->output, sample, &al, true, false);
2943 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2945 if (callchain_ret > 0)
2946 trace__fprintf_callchain(trace, sample);
2947 else if (callchain_ret < 0)
2948 pr_err("Problem processing %s callchain, skipping...\n", evsel__name(evsel));
2950 ++trace->nr_events_printed;
2954 thread__put(thread);
2958 static void trace__set_base_time(struct trace *trace,
2959 struct evsel *evsel,
2960 struct perf_sample *sample)
2963 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2964 * and don't use sample->time unconditionally, we may end up having
2965 * some other event in the future without PERF_SAMPLE_TIME for good
2966 * reason, i.e. we may not be interested in its timestamps, just in
2967 * it taking place, picking some piece of information when it
2968 * appears in our event stream (vfs_getname comes to mind).
2970 if (trace->base_time == 0 && !trace->full_time &&
2971 (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2972 trace->base_time = sample->time;
2975 static int trace__process_sample(struct perf_tool *tool,
2976 union perf_event *event,
2977 struct perf_sample *sample,
2978 struct evsel *evsel,
2979 struct machine *machine __maybe_unused)
2981 struct trace *trace = container_of(tool, struct trace, tool);
2982 struct thread *thread;
2985 tracepoint_handler handler = evsel->handler;
2987 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2988 if (thread && thread__is_filtered(thread))
2991 trace__set_base_time(trace, evsel, sample);
2995 handler(trace, evsel, event, sample);
2998 thread__put(thread);
3002 static int trace__record(struct trace *trace, int argc, const char **argv)
3004 unsigned int rec_argc, i, j;
3005 const char **rec_argv;
3006 const char * const record_args[] = {
3012 pid_t pid = getpid();
3013 char *filter = asprintf__tp_filter_pids(1, &pid);
3014 const char * const sc_args[] = { "-e", };
3015 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
3016 const char * const majpf_args[] = { "-e", "major-faults" };
3017 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
3018 const char * const minpf_args[] = { "-e", "minor-faults" };
3019 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
3022 /* +3 is for the event string below and the pid filter */
3023 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 3 +
3024 majpf_args_nr + minpf_args_nr + argc;
3025 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3027 if (rec_argv == NULL || filter == NULL)
3031 for (i = 0; i < ARRAY_SIZE(record_args); i++)
3032 rec_argv[j++] = record_args[i];
3034 if (trace->trace_syscalls) {
3035 for (i = 0; i < sc_args_nr; i++)
3036 rec_argv[j++] = sc_args[i];
3038 /* event string may be different for older kernels - e.g., RHEL6 */
3039 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
3040 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
3041 else if (is_valid_tracepoint("syscalls:sys_enter"))
3042 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
3044 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
3049 rec_argv[j++] = "--filter";
3050 rec_argv[j++] = filter;
3052 if (trace->trace_pgfaults & TRACE_PFMAJ)
3053 for (i = 0; i < majpf_args_nr; i++)
3054 rec_argv[j++] = majpf_args[i];
3056 if (trace->trace_pgfaults & TRACE_PFMIN)
3057 for (i = 0; i < minpf_args_nr; i++)
3058 rec_argv[j++] = minpf_args[i];
3060 for (i = 0; i < (unsigned int)argc; i++)
3061 rec_argv[j++] = argv[i];
3063 err = cmd_record(j, rec_argv);
3070 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
3072 static bool evlist__add_vfs_getname(struct evlist *evlist)
3075 struct evsel *evsel, *tmp;
3076 struct parse_events_error err;
3079 parse_events_error__init(&err);
3080 ret = parse_events(evlist, "probe:vfs_getname*", &err);
3081 parse_events_error__exit(&err);
3085 evlist__for_each_entry_safe(evlist, evsel, tmp) {
3086 if (!strstarts(evsel__name(evsel), "probe:vfs_getname"))
3089 if (evsel__field(evsel, "pathname")) {
3090 evsel->handler = trace__vfs_getname;
3095 list_del_init(&evsel->core.node);
3096 evsel->evlist = NULL;
3097 evsel__delete(evsel);
3103 static struct evsel *evsel__new_pgfault(u64 config)
3105 struct evsel *evsel;
3106 struct perf_event_attr attr = {
3107 .type = PERF_TYPE_SOFTWARE,
3111 attr.config = config;
3112 attr.sample_period = 1;
3114 event_attr_init(&attr);
3116 evsel = evsel__new(&attr);
3118 evsel->handler = trace__pgfault;
3123 static void evlist__free_syscall_tp_fields(struct evlist *evlist)
3125 struct evsel *evsel;
3127 evlist__for_each_entry(evlist, evsel) {
3128 struct evsel_trace *et = evsel->priv;
3130 if (!et || !evsel->tp_format || strcmp(evsel->tp_format->system, "syscalls"))
3138 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
3140 const u32 type = event->header.type;
3141 struct evsel *evsel;
3143 if (type != PERF_RECORD_SAMPLE) {
3144 trace__process_event(trace, trace->host, event, sample);
3148 evsel = evlist__id2evsel(trace->evlist, sample->id);
3149 if (evsel == NULL) {
3150 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
3154 if (evswitch__discard(&trace->evswitch, evsel))
3157 trace__set_base_time(trace, evsel, sample);
3159 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
3160 sample->raw_data == NULL) {
3161 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
3162 evsel__name(evsel), sample->tid,
3163 sample->cpu, sample->raw_size);
3165 tracepoint_handler handler = evsel->handler;
3166 handler(trace, evsel, event, sample);
3169 if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
3173 static int trace__add_syscall_newtp(struct trace *trace)
3176 struct evlist *evlist = trace->evlist;
3177 struct evsel *sys_enter, *sys_exit;
3179 sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
3180 if (sys_enter == NULL)
3183 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
3184 goto out_delete_sys_enter;
3186 sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
3187 if (sys_exit == NULL)
3188 goto out_delete_sys_enter;
3190 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
3191 goto out_delete_sys_exit;
3193 evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
3194 evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
3196 evlist__add(evlist, sys_enter);
3197 evlist__add(evlist, sys_exit);
3199 if (callchain_param.enabled && !trace->kernel_syscallchains) {
3201 * We're interested only in the user space callchain
3202 * leading to the syscall, allow overriding that for
3203 * debugging reasons using --kernel_syscall_callchains
3205 sys_exit->core.attr.exclude_callchain_kernel = 1;
3208 trace->syscalls.events.sys_enter = sys_enter;
3209 trace->syscalls.events.sys_exit = sys_exit;
3215 out_delete_sys_exit:
3216 evsel__delete_priv(sys_exit);
3217 out_delete_sys_enter:
3218 evsel__delete_priv(sys_enter);
3222 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
3225 struct evsel *sys_exit;
3226 char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
3227 trace->ev_qualifier_ids.nr,
3228 trace->ev_qualifier_ids.entries);
3233 if (!evsel__append_tp_filter(trace->syscalls.events.sys_enter, filter)) {
3234 sys_exit = trace->syscalls.events.sys_exit;
3235 err = evsel__append_tp_filter(sys_exit, filter);
3246 #ifdef HAVE_LIBBPF_SUPPORT
3247 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
3249 if (trace->bpf_obj == NULL)
3252 return bpf_object__find_map_by_name(trace->bpf_obj, name);
3255 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
3257 trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
3260 static void trace__set_bpf_map_syscalls(struct trace *trace)
3262 trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
3263 trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
3264 trace->syscalls.prog_array.sys_exit = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
3267 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
3269 struct bpf_program *pos, *prog = NULL;
3270 const char *sec_name;
3272 if (trace->bpf_obj == NULL)
3275 bpf_object__for_each_program(pos, trace->bpf_obj) {
3276 sec_name = bpf_program__section_name(pos);
3277 if (sec_name && !strcmp(sec_name, name)) {
3286 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
3287 const char *prog_name, const char *type)
3289 struct bpf_program *prog;
3291 if (prog_name == NULL) {
3292 char default_prog_name[256];
3293 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
3294 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3297 if (sc->fmt && sc->fmt->alias) {
3298 scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
3299 prog = trace__find_bpf_program_by_title(trace, default_prog_name);
3303 goto out_unaugmented;
3306 prog = trace__find_bpf_program_by_title(trace, prog_name);
3313 pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
3314 prog_name, type, sc->name);
3316 return trace->syscalls.unaugmented_prog;
3319 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
3321 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3326 sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3327 sc->bpf_prog.sys_exit = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit : NULL, "exit");
3330 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
3332 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3333 return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3336 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
3338 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3339 return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
3342 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
3344 struct syscall *sc = trace__syscall_info(trace, NULL, id);
3350 for (; arg < sc->nr_args; ++arg) {
3351 entry->string_args_len[arg] = 0;
3352 if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
3353 /* Should be set like strace -s strsize */
3354 entry->string_args_len[arg] = PATH_MAX;
3358 for (; arg < 6; ++arg)
3359 entry->string_args_len[arg] = 0;
3361 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
3363 int fd = bpf_map__fd(trace->syscalls.map);
3364 struct bpf_map_syscall_entry value = {
3365 .enabled = !trace->not_ev_qualifier,
3370 for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
3371 int key = trace->ev_qualifier_ids.entries[i];
3373 if (value.enabled) {
3374 trace__init_bpf_map_syscall_args(trace, key, &value);
3375 trace__init_syscall_bpf_progs(trace, key);
3378 err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
3386 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
3388 int fd = bpf_map__fd(trace->syscalls.map);
3389 struct bpf_map_syscall_entry value = {
3394 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3396 trace__init_bpf_map_syscall_args(trace, key, &value);
3398 err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
3406 static int trace__init_syscalls_bpf_map(struct trace *trace)
3408 bool enabled = true;
3410 if (trace->ev_qualifier_ids.nr)
3411 enabled = trace->not_ev_qualifier;
3413 return __trace__init_syscalls_bpf_map(trace, enabled);
3416 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3418 struct tep_format_field *field, *candidate_field;
3422 * We're only interested in syscalls that have a pointer:
3424 for (field = sc->args; field; field = field->next) {
3425 if (field->flags & TEP_FIELD_IS_POINTER)
3426 goto try_to_find_pair;
3432 for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3433 struct syscall *pair = trace__syscall_info(trace, NULL, id);
3434 struct bpf_program *pair_prog;
3435 bool is_candidate = false;
3437 if (pair == NULL || pair == sc ||
3438 pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3441 for (field = sc->args, candidate_field = pair->args;
3442 field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3443 bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3444 candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3447 if (!candidate_is_pointer) {
3448 // The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3452 if (candidate_is_pointer) {
3453 // The candidate might copy a pointer we don't have, skip it.
3454 goto next_candidate;
3459 if (strcmp(field->type, candidate_field->type))
3460 goto next_candidate;
3462 is_candidate = true;
3466 goto next_candidate;
3469 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3470 * then it may be collecting that and we then can't use it, as it would collect
3471 * more than what is common to the two syscalls.
3473 if (candidate_field) {
3474 for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3475 if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3476 goto next_candidate;
3479 pair_prog = pair->bpf_prog.sys_enter;
3481 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3482 * have been searched for, so search it here and if it returns the
3483 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3484 * program for a filtered syscall on a non-filtered one.
3486 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3487 * useful for "renameat2".
3489 if (pair_prog == NULL) {
3490 pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3491 if (pair_prog == trace->syscalls.unaugmented_prog)
3492 goto next_candidate;
3495 pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3504 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3506 int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3507 map_exit_fd = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3510 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3513 if (!trace__syscall_enabled(trace, key))
3516 trace__init_syscall_bpf_progs(trace, key);
3518 // It'll get at least the "!raw_syscalls:unaugmented"
3519 prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3520 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3523 prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3524 err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3530 * Now lets do a second pass looking for enabled syscalls without
3531 * an augmenter that have a signature that is a superset of another
3532 * syscall with an augmenter so that we can auto-reuse it.
3534 * I.e. if we have an augmenter for the "open" syscall that has
3537 * int open(const char *pathname, int flags, mode_t mode);
3539 * I.e. that will collect just the first string argument, then we
3540 * can reuse it for the 'creat' syscall, that has this signature:
3542 * int creat(const char *pathname, mode_t mode);
3546 * int stat(const char *pathname, struct stat *statbuf);
3547 * int lstat(const char *pathname, struct stat *statbuf);
3549 * Because the 'open' augmenter will collect the first arg as a string,
3550 * and leave alone all the other args, which already helps with
3551 * beautifying 'stat' and 'lstat''s pathname arg.
3553 * Then, in time, when 'stat' gets an augmenter that collects both
3554 * first and second arg (this one on the raw_syscalls:sys_exit prog
3555 * array tail call, then that one will be used.
3557 for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3558 struct syscall *sc = trace__syscall_info(trace, NULL, key);
3559 struct bpf_program *pair_prog;
3562 if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3566 * For now we're just reusing the sys_enter prog, and if it
3567 * already has an augmenter, we don't need to find one.
3569 if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3573 * Look at all the other syscalls for one that has a signature
3574 * that is close enough that we can share:
3576 pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3577 if (pair_prog == NULL)
3580 sc->bpf_prog.sys_enter = pair_prog;
3583 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3584 * with the fd for the program we're reusing:
3586 prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3587 err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3596 static void trace__delete_augmented_syscalls(struct trace *trace)
3598 struct evsel *evsel, *tmp;
3600 evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3601 evsel__delete(trace->syscalls.events.augmented);
3602 trace->syscalls.events.augmented = NULL;
3604 evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3605 if (evsel->bpf_obj == trace->bpf_obj) {
3606 evlist__remove(trace->evlist, evsel);
3607 evsel__delete(evsel);
3612 bpf_object__close(trace->bpf_obj);
3613 trace->bpf_obj = NULL;
3615 #else // HAVE_LIBBPF_SUPPORT
3616 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace __maybe_unused,
3617 const char *name __maybe_unused)
3622 static void trace__set_bpf_map_filtered_pids(struct trace *trace __maybe_unused)
3626 static void trace__set_bpf_map_syscalls(struct trace *trace __maybe_unused)
3630 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3635 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3640 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3641 const char *name __maybe_unused)
3646 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3651 static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3654 #endif // HAVE_LIBBPF_SUPPORT
3656 static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3658 struct evsel *evsel;
3660 evlist__for_each_entry(trace->evlist, evsel) {
3661 if (evsel == trace->syscalls.events.augmented ||
3662 evsel->bpf_obj == trace->bpf_obj)
3671 static int trace__set_ev_qualifier_filter(struct trace *trace)
3673 if (trace->syscalls.map)
3674 return trace__set_ev_qualifier_bpf_filter(trace);
3675 if (trace->syscalls.events.sys_enter)
3676 return trace__set_ev_qualifier_tp_filter(trace);
3680 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3681 size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3684 #ifdef HAVE_LIBBPF_SUPPORT
3686 int map_fd = bpf_map__fd(map);
3689 for (i = 0; i < npids; ++i) {
3690 err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3698 static int trace__set_filter_loop_pids(struct trace *trace)
3700 unsigned int nr = 1, err;
3704 struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3706 while (thread && nr < ARRAY_SIZE(pids)) {
3707 struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3712 if (!strcmp(thread__comm_str(parent), "sshd") ||
3713 strstarts(thread__comm_str(parent), "gnome-terminal")) {
3714 pids[nr++] = parent->tid;
3720 err = evlist__append_tp_filter_pids(trace->evlist, nr, pids);
3721 if (!err && trace->filter_pids.map)
3722 err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3727 static int trace__set_filter_pids(struct trace *trace)
3731 * Better not use !target__has_task() here because we need to cover the
3732 * case where no threads were specified in the command line, but a
3733 * workload was, and in that case we will fill in the thread_map when
3734 * we fork the workload in evlist__prepare_workload.
3736 if (trace->filter_pids.nr > 0) {
3737 err = evlist__append_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3738 trace->filter_pids.entries);
3739 if (!err && trace->filter_pids.map) {
3740 err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3741 trace->filter_pids.entries);
3743 } else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3744 err = trace__set_filter_loop_pids(trace);
3750 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3752 struct evlist *evlist = trace->evlist;
3753 struct perf_sample sample;
3754 int err = evlist__parse_sample(evlist, event, &sample);
3757 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3759 trace__handle_event(trace, event, &sample);
3764 static int __trace__flush_events(struct trace *trace)
3766 u64 first = ordered_events__first_time(&trace->oe.data);
3767 u64 flush = trace->oe.last - NSEC_PER_SEC;
3769 /* Is there some thing to flush.. */
3770 if (first && first < flush)
3771 return ordered_events__flush_time(&trace->oe.data, flush);
3776 static int trace__flush_events(struct trace *trace)
3778 return !trace->sort_events ? 0 : __trace__flush_events(trace);
3781 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3785 if (!trace->sort_events)
3786 return __trace__deliver_event(trace, event);
3788 err = evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3789 if (err && err != -1)
3792 err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0, NULL);
3796 return trace__flush_events(trace);
3799 static int ordered_events__deliver_event(struct ordered_events *oe,
3800 struct ordered_event *event)
3802 struct trace *trace = container_of(oe, struct trace, oe.data);
3804 return __trace__deliver_event(trace, event->event);
3807 static struct syscall_arg_fmt *evsel__find_syscall_arg_fmt_by_name(struct evsel *evsel, char *arg)
3809 struct tep_format_field *field;
3810 struct syscall_arg_fmt *fmt = __evsel__syscall_arg_fmt(evsel);
3812 if (evsel->tp_format == NULL || fmt == NULL)
3815 for (field = evsel->tp_format->format.fields; field; field = field->next, ++fmt)
3816 if (strcmp(field->name, arg) == 0)
3822 static int trace__expand_filter(struct trace *trace __maybe_unused, struct evsel *evsel)
3824 char *tok, *left = evsel->filter, *new_filter = evsel->filter;
3826 while ((tok = strpbrk(left, "=<>!")) != NULL) {
3827 char *right = tok + 1, *right_end;
3832 while (isspace(*right))
3838 while (!isalpha(*left))
3839 if (++left == tok) {
3841 * Bail out, can't find the name of the argument that is being
3842 * used in the filter, let it try to set this filter, will fail later.
3847 right_end = right + 1;
3848 while (isalnum(*right_end) || *right_end == '_' || *right_end == '|')
3851 if (isalpha(*right)) {
3852 struct syscall_arg_fmt *fmt;
3853 int left_size = tok - left,
3854 right_size = right_end - right;
3857 while (isspace(left[left_size - 1]))
3860 scnprintf(arg, sizeof(arg), "%.*s", left_size, left);
3862 fmt = evsel__find_syscall_arg_fmt_by_name(evsel, arg);
3864 pr_err("\"%s\" not found in \"%s\", can't set filter \"%s\"\n",
3865 arg, evsel->name, evsel->filter);
3869 pr_debug2("trying to expand \"%s\" \"%.*s\" \"%.*s\" -> ",
3870 arg, (int)(right - tok), tok, right_size, right);
3874 struct syscall_arg syscall_arg = {
3878 if (fmt->strtoul(right, right_size, &syscall_arg, &val)) {
3879 char *n, expansion[19];
3880 int expansion_lenght = scnprintf(expansion, sizeof(expansion), "%#" PRIx64, val);
3881 int expansion_offset = right - new_filter;
3883 pr_debug("%s", expansion);
3885 if (asprintf(&n, "%.*s%s%s", expansion_offset, new_filter, expansion, right_end) < 0) {
3886 pr_debug(" out of memory!\n");
3890 if (new_filter != evsel->filter)
3892 left = n + expansion_offset + expansion_lenght;
3895 pr_err("\"%.*s\" not found for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3896 right_size, right, arg, evsel->name, evsel->filter);
3900 pr_err("No resolver (strtoul) for \"%s\" in \"%s\", can't set filter \"%s\"\n",
3901 arg, evsel->name, evsel->filter);
3911 if (new_filter != evsel->filter) {
3912 pr_debug("New filter for %s: %s\n", evsel->name, new_filter);
3913 evsel__set_filter(evsel, new_filter);
3920 static int trace__expand_filters(struct trace *trace, struct evsel **err_evsel)
3922 struct evlist *evlist = trace->evlist;
3923 struct evsel *evsel;
3925 evlist__for_each_entry(evlist, evsel) {
3926 if (evsel->filter == NULL)
3929 if (trace__expand_filter(trace, evsel)) {
3938 static int trace__run(struct trace *trace, int argc, const char **argv)
3940 struct evlist *evlist = trace->evlist;
3941 struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3943 unsigned long before;
3944 const bool forks = argc > 0;
3945 bool draining = false;
3949 if (!trace->raw_augmented_syscalls) {
3950 if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3951 goto out_error_raw_syscalls;
3953 if (trace->trace_syscalls)
3954 trace->vfs_getname = evlist__add_vfs_getname(evlist);
3957 if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3958 pgfault_maj = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3959 if (pgfault_maj == NULL)
3961 evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3962 evlist__add(evlist, pgfault_maj);
3965 if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3966 pgfault_min = evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3967 if (pgfault_min == NULL)
3969 evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3970 evlist__add(evlist, pgfault_min);
3973 /* Enable ignoring missing threads when -u/-p option is defined. */
3974 trace->opts.ignore_missing_thread = trace->opts.target.uid != UINT_MAX || trace->opts.target.pid;
3977 evlist__add_newtp(evlist, "sched", "sched_stat_runtime", trace__sched_stat_runtime))
3978 goto out_error_sched_stat_runtime;
3980 * If a global cgroup was set, apply it to all the events without an
3981 * explicit cgroup. I.e.:
3983 * trace -G A -e sched:*switch
3985 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3986 * _and_ sched:sched_switch to the 'A' cgroup, while:
3988 * trace -e sched:*switch -G A
3990 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3991 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3992 * a cgroup (on the root cgroup, sys wide, etc).
3996 * trace -G A -e sched:*switch -G B
3998 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3999 * to the 'B' cgroup.
4001 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
4002 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
4005 evlist__set_default_cgroup(trace->evlist, trace->cgroup);
4007 err = evlist__create_maps(evlist, &trace->opts.target);
4009 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
4010 goto out_delete_evlist;
4013 err = trace__symbols_init(trace, evlist);
4015 fprintf(trace->output, "Problems initializing symbol libraries!\n");
4016 goto out_delete_evlist;
4019 evlist__config(evlist, &trace->opts, &callchain_param);
4022 err = evlist__prepare_workload(evlist, &trace->opts.target, argv, false, NULL);
4024 fprintf(trace->output, "Couldn't run the workload!\n");
4025 goto out_delete_evlist;
4027 workload_pid = evlist->workload.pid;
4030 err = evlist__open(evlist);
4032 goto out_error_open;
4034 err = bpf__apply_obj_config();
4036 char errbuf[BUFSIZ];
4038 bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
4039 pr_err("ERROR: Apply config to BPF failed: %s\n",
4041 goto out_error_open;
4044 err = trace__set_filter_pids(trace);
4048 if (trace->syscalls.map)
4049 trace__init_syscalls_bpf_map(trace);
4051 if (trace->syscalls.prog_array.sys_enter)
4052 trace__init_syscalls_bpf_prog_array_maps(trace);
4054 if (trace->ev_qualifier_ids.nr > 0) {
4055 err = trace__set_ev_qualifier_filter(trace);
4059 if (trace->syscalls.events.sys_exit) {
4060 pr_debug("event qualifier tracepoint filter: %s\n",
4061 trace->syscalls.events.sys_exit->filter);
4066 * If the "close" syscall is not traced, then we will not have the
4067 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
4068 * fd->pathname table and were ending up showing the last value set by
4069 * syscalls opening a pathname and associating it with a descriptor or
4070 * reading it from /proc/pid/fd/ in cases where that doesn't make
4073 * So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
4076 trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
4078 err = trace__expand_filters(trace, &evsel);
4080 goto out_delete_evlist;
4081 err = evlist__apply_filters(evlist, &evsel);
4083 goto out_error_apply_filters;
4085 if (trace->dump.map)
4086 bpf_map__fprintf(trace->dump.map, trace->output);
4088 err = evlist__mmap(evlist, trace->opts.mmap_pages);
4090 goto out_error_mmap;
4092 if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
4093 evlist__enable(evlist);
4096 evlist__start_workload(evlist);
4098 if (trace->opts.initial_delay) {
4099 usleep(trace->opts.initial_delay * 1000);
4100 evlist__enable(evlist);
4103 trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
4104 evlist->core.threads->nr > 1 ||
4105 evlist__first(evlist)->core.attr.inherit;
4108 * Now that we already used evsel->core.attr to ask the kernel to setup the
4109 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
4110 * trace__resolve_callchain(), allowing per-event max-stack settings
4111 * to override an explicitly set --max-stack global setting.
4113 evlist__for_each_entry(evlist, evsel) {
4114 if (evsel__has_callchain(evsel) &&
4115 evsel->core.attr.sample_max_stack == 0)
4116 evsel->core.attr.sample_max_stack = trace->max_stack;
4119 before = trace->nr_events;
4121 for (i = 0; i < evlist->core.nr_mmaps; i++) {
4122 union perf_event *event;
4125 md = &evlist->mmap[i];
4126 if (perf_mmap__read_init(&md->core) < 0)
4129 while ((event = perf_mmap__read_event(&md->core)) != NULL) {
4132 err = trace__deliver_event(trace, event);
4136 perf_mmap__consume(&md->core);
4141 if (done && !draining) {
4142 evlist__disable(evlist);
4146 perf_mmap__read_done(&md->core);
4149 if (trace->nr_events == before) {
4150 int timeout = done ? 100 : -1;
4152 if (!draining && evlist__poll(evlist, timeout) > 0) {
4153 if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
4158 if (trace__flush_events(trace))
4166 thread__zput(trace->current);
4168 evlist__disable(evlist);
4170 if (trace->sort_events)
4171 ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
4175 trace__fprintf_thread_summary(trace, trace->output);
4177 if (trace->show_tool_stats) {
4178 fprintf(trace->output, "Stats:\n "
4179 " vfs_getname : %" PRIu64 "\n"
4180 " proc_getname: %" PRIu64 "\n",
4181 trace->stats.vfs_getname,
4182 trace->stats.proc_getname);
4187 trace__symbols__exit(trace);
4188 evlist__free_syscall_tp_fields(evlist);
4189 evlist__delete(evlist);
4190 cgroup__put(trace->cgroup);
4191 trace->evlist = NULL;
4192 trace->live = false;
4195 char errbuf[BUFSIZ];
4197 out_error_sched_stat_runtime:
4198 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
4201 out_error_raw_syscalls:
4202 tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
4206 evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
4210 evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
4213 fprintf(trace->output, "%s\n", errbuf);
4214 goto out_delete_evlist;
4216 out_error_apply_filters:
4217 fprintf(trace->output,
4218 "Failed to set filter \"%s\" on event %s with %d (%s)\n",
4219 evsel->filter, evsel__name(evsel), errno,
4220 str_error_r(errno, errbuf, sizeof(errbuf)));
4221 goto out_delete_evlist;
4224 fprintf(trace->output, "Not enough memory to run!\n");
4225 goto out_delete_evlist;
4228 fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
4229 goto out_delete_evlist;
4232 static int trace__replay(struct trace *trace)
4234 const struct evsel_str_handler handlers[] = {
4235 { "probe:vfs_getname", trace__vfs_getname, },
4237 struct perf_data data = {
4239 .mode = PERF_DATA_MODE_READ,
4240 .force = trace->force,
4242 struct perf_session *session;
4243 struct evsel *evsel;
4246 trace->tool.sample = trace__process_sample;
4247 trace->tool.mmap = perf_event__process_mmap;
4248 trace->tool.mmap2 = perf_event__process_mmap2;
4249 trace->tool.comm = perf_event__process_comm;
4250 trace->tool.exit = perf_event__process_exit;
4251 trace->tool.fork = perf_event__process_fork;
4252 trace->tool.attr = perf_event__process_attr;
4253 trace->tool.tracing_data = perf_event__process_tracing_data;
4254 trace->tool.build_id = perf_event__process_build_id;
4255 trace->tool.namespaces = perf_event__process_namespaces;
4257 trace->tool.ordered_events = true;
4258 trace->tool.ordering_requires_timestamps = true;
4260 /* add tid to output */
4261 trace->multiple_threads = true;
4263 session = perf_session__new(&data, &trace->tool);
4264 if (IS_ERR(session))
4265 return PTR_ERR(session);
4267 if (trace->opts.target.pid)
4268 symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
4270 if (trace->opts.target.tid)
4271 symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
4273 if (symbol__init(&session->header.env) < 0)
4276 trace->host = &session->machines.host;
4278 err = perf_session__set_tracepoints_handlers(session, handlers);
4282 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_enter");
4283 /* older kernels have syscalls tp versus raw_syscalls */
4285 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_enter");
4288 (evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
4289 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
4290 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
4294 evsel = evlist__find_tracepoint_by_name(session->evlist, "raw_syscalls:sys_exit");
4296 evsel = evlist__find_tracepoint_by_name(session->evlist, "syscalls:sys_exit");
4298 (evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
4299 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
4300 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
4304 evlist__for_each_entry(session->evlist, evsel) {
4305 if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
4306 (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
4307 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
4308 evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
4309 evsel->handler = trace__pgfault;
4314 err = perf_session__process_events(session);
4316 pr_err("Failed to process events, error %d", err);
4318 else if (trace->summary)
4319 trace__fprintf_thread_summary(trace, trace->output);
4322 perf_session__delete(session);
4327 static size_t trace__fprintf_threads_header(FILE *fp)
4331 printed = fprintf(fp, "\n Summary of events:\n\n");
4336 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
4337 struct syscall_stats *stats;
4342 struct int_node *source = rb_entry(nd, struct int_node, rb_node);
4343 struct syscall_stats *stats = source->priv;
4345 entry->syscall = source->i;
4346 entry->stats = stats;
4347 entry->msecs = stats ? (u64)stats->stats.n * (avg_stats(&stats->stats) / NSEC_PER_MSEC) : 0;
4350 static size_t thread__dump_stats(struct thread_trace *ttrace,
4351 struct trace *trace, FILE *fp)
4356 DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
4358 if (syscall_stats == NULL)
4361 printed += fprintf(fp, "\n");
4363 printed += fprintf(fp, " syscall calls errors total min avg max stddev\n");
4364 printed += fprintf(fp, " (msec) (msec) (msec) (msec) (%%)\n");
4365 printed += fprintf(fp, " --------------- -------- ------ -------- --------- --------- --------- ------\n");
4367 resort_rb__for_each_entry(nd, syscall_stats) {
4368 struct syscall_stats *stats = syscall_stats_entry->stats;
4370 double min = (double)(stats->stats.min) / NSEC_PER_MSEC;
4371 double max = (double)(stats->stats.max) / NSEC_PER_MSEC;
4372 double avg = avg_stats(&stats->stats);
4374 u64 n = (u64)stats->stats.n;
4376 pct = avg ? 100.0 * stddev_stats(&stats->stats) / avg : 0.0;
4377 avg /= NSEC_PER_MSEC;
4379 sc = &trace->syscalls.table[syscall_stats_entry->syscall];
4380 printed += fprintf(fp, " %-15s", sc->name);
4381 printed += fprintf(fp, " %8" PRIu64 " %6" PRIu64 " %9.3f %9.3f %9.3f",
4382 n, stats->nr_failures, syscall_stats_entry->msecs, min, avg);
4383 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
4385 if (trace->errno_summary && stats->nr_failures) {
4386 const char *arch_name = perf_env__arch(trace->host->env);
4389 for (e = 0; e < stats->max_errno; ++e) {
4390 if (stats->errnos[e] != 0)
4391 fprintf(fp, "\t\t\t\t%s: %d\n", arch_syscalls__strerrno(arch_name, e + 1), stats->errnos[e]);
4397 resort_rb__delete(syscall_stats);
4398 printed += fprintf(fp, "\n\n");
4403 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
4406 struct thread_trace *ttrace = thread__priv(thread);
4412 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
4414 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
4415 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
4416 printed += fprintf(fp, "%.1f%%", ratio);
4418 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
4420 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
4422 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
4423 else if (fputc('\n', fp) != EOF)
4426 printed += thread__dump_stats(ttrace, trace, fp);
4431 static unsigned long thread__nr_events(struct thread_trace *ttrace)
4433 return ttrace ? ttrace->nr_events : 0;
4436 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
4437 struct thread *thread;
4440 entry->thread = rb_entry(nd, struct thread, rb_node);
4443 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
4445 size_t printed = trace__fprintf_threads_header(fp);
4449 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
4450 DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
4452 if (threads == NULL) {
4453 fprintf(fp, "%s", "Error sorting output by nr_events!\n");
4457 resort_rb__for_each_entry(nd, threads)
4458 printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
4460 resort_rb__delete(threads);
4465 static int trace__set_duration(const struct option *opt, const char *str,
4466 int unset __maybe_unused)
4468 struct trace *trace = opt->value;
4470 trace->duration_filter = atof(str);
4474 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
4475 int unset __maybe_unused)
4479 struct trace *trace = opt->value;
4481 * FIXME: introduce a intarray class, plain parse csv and create a
4482 * { int nr, int entries[] } struct...
4484 struct intlist *list = intlist__new(str);
4489 i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
4490 trace->filter_pids.entries = calloc(i, sizeof(pid_t));
4492 if (trace->filter_pids.entries == NULL)
4495 trace->filter_pids.entries[0] = getpid();
4497 for (i = 1; i < trace->filter_pids.nr; ++i)
4498 trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
4500 intlist__delete(list);
4506 static int trace__open_output(struct trace *trace, const char *filename)
4510 if (!stat(filename, &st) && st.st_size) {
4511 char oldname[PATH_MAX];
4513 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
4515 rename(filename, oldname);
4518 trace->output = fopen(filename, "w");
4520 return trace->output == NULL ? -errno : 0;
4523 static int parse_pagefaults(const struct option *opt, const char *str,
4524 int unset __maybe_unused)
4526 int *trace_pgfaults = opt->value;
4528 if (strcmp(str, "all") == 0)
4529 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4530 else if (strcmp(str, "maj") == 0)
4531 *trace_pgfaults |= TRACE_PFMAJ;
4532 else if (strcmp(str, "min") == 0)
4533 *trace_pgfaults |= TRACE_PFMIN;
4540 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4542 struct evsel *evsel;
4544 evlist__for_each_entry(evlist, evsel) {
4545 if (evsel->handler == NULL)
4546 evsel->handler = handler;
4550 static void evsel__set_syscall_arg_fmt(struct evsel *evsel, const char *name)
4552 struct syscall_arg_fmt *fmt = evsel__syscall_arg_fmt(evsel);
4555 struct syscall_fmt *scfmt = syscall_fmt__find(name);
4560 if (strcmp(evsel->tp_format->format.fields->name, "__syscall_nr") == 0 ||
4561 strcmp(evsel->tp_format->format.fields->name, "nr") == 0)
4564 memcpy(fmt + skip, scfmt->arg, (evsel->tp_format->format.nr_fields - skip) * sizeof(*fmt));
4569 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
4571 struct evsel *evsel;
4573 evlist__for_each_entry(evlist, evsel) {
4574 if (evsel->priv || !evsel->tp_format)
4577 if (strcmp(evsel->tp_format->system, "syscalls")) {
4578 evsel__init_tp_arg_scnprintf(evsel);
4582 if (evsel__init_syscall_tp(evsel))
4585 if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4586 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4588 if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4591 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_enter_") - 1);
4592 } else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4593 struct syscall_tp *sc = __evsel__syscall_tp(evsel);
4595 if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4598 evsel__set_syscall_arg_fmt(evsel, evsel->tp_format->name + sizeof("sys_exit_") - 1);
4606 * XXX: Hackish, just splitting the combined -e+--event (syscalls
4607 * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4608 * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4610 * It'd be better to introduce a parse_options() variant that would return a
4611 * list with the terms it didn't match to an event...
4613 static int trace__parse_events_option(const struct option *opt, const char *str,
4614 int unset __maybe_unused)
4616 struct trace *trace = (struct trace *)opt->value;
4617 const char *s = str;
4618 char *sep = NULL, *lists[2] = { NULL, NULL, };
4619 int len = strlen(str) + 1, err = -1, list, idx;
4620 char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4621 char group_name[PATH_MAX];
4622 struct syscall_fmt *fmt;
4624 if (strace_groups_dir == NULL)
4629 trace->not_ev_qualifier = true;
4633 if ((sep = strchr(s, ',')) != NULL)
4637 if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4638 syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4643 fmt = syscall_fmt__find_by_alias(s);
4648 path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4649 if (access(group_name, R_OK) == 0)
4654 sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4656 lists[list] = malloc(len);
4657 if (lists[list] == NULL)
4659 strcpy(lists[list], s);
4669 if (lists[1] != NULL) {
4670 struct strlist_config slist_config = {
4671 .dirname = strace_groups_dir,
4674 trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4675 if (trace->ev_qualifier == NULL) {
4676 fputs("Not enough memory to parse event qualifier", trace->output);
4680 if (trace__validate_ev_qualifier(trace))
4682 trace->trace_syscalls = true;
4689 .value = &trace->evlist,
4691 err = parse_events_option(&o, lists[0], 0);
4694 free(strace_groups_dir);
4703 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4705 struct trace *trace = opt->value;
4707 if (!list_empty(&trace->evlist->core.entries)) {
4709 .value = &trace->evlist,
4711 return parse_cgroups(&o, str, unset);
4713 trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4718 static int trace__config(const char *var, const char *value, void *arg)
4720 struct trace *trace = arg;
4723 if (!strcmp(var, "trace.add_events")) {
4724 trace->perfconfig_events = strdup(value);
4725 if (trace->perfconfig_events == NULL) {
4726 pr_err("Not enough memory for %s\n", "trace.add_events");
4729 } else if (!strcmp(var, "trace.show_timestamp")) {
4730 trace->show_tstamp = perf_config_bool(var, value);
4731 } else if (!strcmp(var, "trace.show_duration")) {
4732 trace->show_duration = perf_config_bool(var, value);
4733 } else if (!strcmp(var, "trace.show_arg_names")) {
4734 trace->show_arg_names = perf_config_bool(var, value);
4735 if (!trace->show_arg_names)
4736 trace->show_zeros = true;
4737 } else if (!strcmp(var, "trace.show_zeros")) {
4738 bool new_show_zeros = perf_config_bool(var, value);
4739 if (!trace->show_arg_names && !new_show_zeros) {
4740 pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4743 trace->show_zeros = new_show_zeros;
4744 } else if (!strcmp(var, "trace.show_prefix")) {
4745 trace->show_string_prefix = perf_config_bool(var, value);
4746 } else if (!strcmp(var, "trace.no_inherit")) {
4747 trace->opts.no_inherit = perf_config_bool(var, value);
4748 } else if (!strcmp(var, "trace.args_alignment")) {
4749 int args_alignment = 0;
4750 if (perf_config_int(&args_alignment, var, value) == 0)
4751 trace->args_alignment = args_alignment;
4752 } else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4753 if (strcasecmp(value, "libtraceevent") == 0)
4754 trace->libtraceevent_print = true;
4755 else if (strcasecmp(value, "libbeauty") == 0)
4756 trace->libtraceevent_print = false;
4762 static void trace__exit(struct trace *trace)
4766 strlist__delete(trace->ev_qualifier);
4767 free(trace->ev_qualifier_ids.entries);
4768 if (trace->syscalls.table) {
4769 for (i = 0; i <= trace->sctbl->syscalls.max_id; i++)
4770 syscall__exit(&trace->syscalls.table[i]);
4771 free(trace->syscalls.table);
4773 syscalltbl__delete(trace->sctbl);
4774 zfree(&trace->perfconfig_events);
4777 int cmd_trace(int argc, const char **argv)
4779 const char *trace_usage[] = {
4780 "perf trace [<options>] [<command>]",
4781 "perf trace [<options>] -- <command> [<options>]",
4782 "perf trace record [<options>] [<command>]",
4783 "perf trace record [<options>] -- <command> [<options>]",
4786 struct trace trace = {
4792 .user_freq = UINT_MAX,
4793 .user_interval = ULLONG_MAX,
4794 .no_buffering = true,
4795 .mmap_pages = UINT_MAX,
4799 .show_tstamp = true,
4800 .show_duration = true,
4801 .show_arg_names = true,
4802 .args_alignment = 70,
4803 .trace_syscalls = false,
4804 .kernel_syscallchains = false,
4805 .max_stack = UINT_MAX,
4806 .max_events = ULONG_MAX,
4808 const char *map_dump_str = NULL;
4809 const char *output_name = NULL;
4810 const struct option trace_options[] = {
4811 OPT_CALLBACK('e', "event", &trace, "event",
4812 "event/syscall selector. use 'perf list' to list available events",
4813 trace__parse_events_option),
4814 OPT_CALLBACK(0, "filter", &trace.evlist, "filter",
4815 "event filter", parse_filter),
4816 OPT_BOOLEAN(0, "comm", &trace.show_comm,
4817 "show the thread COMM next to its id"),
4818 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4819 OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4820 trace__parse_events_option),
4821 OPT_STRING('o', "output", &output_name, "file", "output file name"),
4822 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4823 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4824 "trace events on existing process id"),
4825 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4826 "trace events on existing thread id"),
4827 OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4828 "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4829 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4830 "system-wide collection from all CPUs"),
4831 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4832 "list of cpus to monitor"),
4833 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4834 "child tasks do not inherit counters"),
4835 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4836 "number of mmap data pages", evlist__parse_mmap_pages),
4837 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4839 OPT_CALLBACK(0, "duration", &trace, "float",
4840 "show only events with duration > N.M ms",
4841 trace__set_duration),
4842 #ifdef HAVE_LIBBPF_SUPPORT
4843 OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4845 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4846 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4847 OPT_BOOLEAN('T', "time", &trace.full_time,
4848 "Show full timestamp, not time relative to first start"),
4849 OPT_BOOLEAN(0, "failure", &trace.failure_only,
4850 "Show only syscalls that failed"),
4851 OPT_BOOLEAN('s', "summary", &trace.summary_only,
4852 "Show only syscall summary with statistics"),
4853 OPT_BOOLEAN('S', "with-summary", &trace.summary,
4854 "Show all syscalls and summary with statistics"),
4855 OPT_BOOLEAN(0, "errno-summary", &trace.errno_summary,
4856 "Show errno stats per syscall, use with -s or -S"),
4857 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4858 "Trace pagefaults", parse_pagefaults, "maj"),
4859 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4860 OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4861 OPT_CALLBACK(0, "call-graph", &trace.opts,
4862 "record_mode[,record_size]", record_callchain_help,
4863 &record_parse_callchain_opt),
4864 OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4865 "Use libtraceevent to print the tracepoint arguments."),
4866 OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4867 "Show the kernel callchains on the syscall exit path"),
4868 OPT_ULONG(0, "max-events", &trace.max_events,
4869 "Set the maximum number of events to print, exit after that is reached. "),
4870 OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4871 "Set the minimum stack depth when parsing the callchain, "
4872 "anything below the specified depth will be ignored."),
4873 OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4874 "Set the maximum stack depth when parsing the callchain, "
4875 "anything beyond the specified depth will be ignored. "
4876 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4877 OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4878 "Sort batch of events before processing, use if getting out of order events"),
4879 OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4880 "print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4881 OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4882 "per thread proc mmap processing timeout in ms"),
4883 OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4884 trace__parse_cgroups),
4885 OPT_INTEGER('D', "delay", &trace.opts.initial_delay,
4886 "ms to wait before starting measurement after program "
4888 OPTS_EVSWITCH(&trace.evswitch),
4891 bool __maybe_unused max_stack_user_set = true;
4892 bool mmap_pages_user_set = true;
4893 struct evsel *evsel;
4894 const char * const trace_subcommands[] = { "record", NULL };
4897 struct sigaction sigchld_act;
4899 signal(SIGSEGV, sighandler_dump_stack);
4900 signal(SIGFPE, sighandler_dump_stack);
4901 signal(SIGINT, sighandler_interrupt);
4903 memset(&sigchld_act, 0, sizeof(sigchld_act));
4904 sigchld_act.sa_flags = SA_SIGINFO;
4905 sigchld_act.sa_sigaction = sighandler_chld;
4906 sigaction(SIGCHLD, &sigchld_act, NULL);
4908 trace.evlist = evlist__new();
4909 trace.sctbl = syscalltbl__new();
4911 if (trace.evlist == NULL || trace.sctbl == NULL) {
4912 pr_err("Not enough memory to run!\n");
4918 * Parsing .perfconfig may entail creating a BPF event, that may need
4919 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4920 * is too small. This affects just this process, not touching the
4921 * global setting. If it fails we'll get something in 'perf trace -v'
4922 * to help diagnose the problem.
4924 rlimit__bump_memlock();
4926 err = perf_config(trace__config, &trace);
4930 argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4931 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4934 * Here we already passed thru trace__parse_events_option() and it has
4935 * already figured out if -e syscall_name, if not but if --event
4936 * foo:bar was used, the user is interested _just_ in those, say,
4937 * tracepoint events, not in the strace-like syscall-name-based mode.
4939 * This is important because we need to check if strace-like mode is
4940 * needed to decided if we should filter out the eBPF
4941 * __augmented_syscalls__ code, if it is in the mix, say, via
4942 * .perfconfig trace.add_events, and filter those out.
4944 if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4945 trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4946 trace.trace_syscalls = true;
4949 * Now that we have --verbose figured out, lets see if we need to parse
4950 * events from .perfconfig, so that if those events fail parsing, say some
4951 * BPF program fails, then we'll be able to use --verbose to see what went
4952 * wrong in more detail.
4954 if (trace.perfconfig_events != NULL) {
4955 struct parse_events_error parse_err;
4957 parse_events_error__init(&parse_err);
4958 err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
4960 parse_events_error__print(&parse_err, trace.perfconfig_events);
4961 parse_events_error__exit(&parse_err);
4966 if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4967 usage_with_options_msg(trace_usage, trace_options,
4968 "cgroup monitoring only available in system-wide mode");
4971 evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4972 if (IS_ERR(evsel)) {
4973 bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4974 pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4979 trace.syscalls.events.augmented = evsel;
4981 evsel = evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4982 if (evsel == NULL) {
4983 pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4987 if (evsel->bpf_obj == NULL) {
4988 pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4992 trace.bpf_obj = evsel->bpf_obj;
4995 * If we have _just_ the augmenter event but don't have a
4996 * explicit --syscalls, then assume we want all strace-like
4999 if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
5000 trace.trace_syscalls = true;
5002 * So, if we have a syscall augmenter, but trace_syscalls, aka
5003 * strace-like syscall tracing is not set, then we need to trow
5004 * away the augmenter, i.e. all the events that were created
5005 * from that BPF object file.
5007 * This is more to fix the current .perfconfig trace.add_events
5008 * style of setting up the strace-like eBPF based syscall point
5009 * payload augmenter.
5011 * All this complexity will be avoided by adding an alternative
5012 * to trace.add_events in the form of
5013 * trace.bpf_augmented_syscalls, that will be only parsed if we
5016 * .perfconfig trace.add_events is still useful if we want, for
5017 * instance, have msr_write.msr in some .perfconfig profile based
5018 * 'perf trace --config determinism.profile' mode, where for some
5019 * particular goal/workload type we want a set of events and
5020 * output mode (with timings, etc) instead of having to add
5021 * all via the command line.
5023 * Also --config to specify an alternate .perfconfig file needs
5024 * to be implemented.
5026 if (!trace.trace_syscalls) {
5027 trace__delete_augmented_syscalls(&trace);
5029 trace__set_bpf_map_filtered_pids(&trace);
5030 trace__set_bpf_map_syscalls(&trace);
5031 trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
5035 err = bpf__setup_stdout(trace.evlist);
5037 bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
5038 pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
5045 trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
5046 if (trace.dump.map == NULL) {
5047 pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
5052 if (trace.trace_pgfaults) {
5053 trace.opts.sample_address = true;
5054 trace.opts.sample_time = true;
5057 if (trace.opts.mmap_pages == UINT_MAX)
5058 mmap_pages_user_set = false;
5060 if (trace.max_stack == UINT_MAX) {
5061 trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
5062 max_stack_user_set = false;
5065 #ifdef HAVE_DWARF_UNWIND_SUPPORT
5066 if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
5067 record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
5071 if (callchain_param.enabled) {
5072 if (!mmap_pages_user_set && geteuid() == 0)
5073 trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
5075 symbol_conf.use_callchain = true;
5078 if (trace.evlist->core.nr_entries > 0) {
5079 evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
5080 if (evlist__set_syscall_tp_fields(trace.evlist)) {
5081 perror("failed to set syscalls:* tracepoint fields");
5086 if (trace.sort_events) {
5087 ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
5088 ordered_events__set_copy_on_queue(&trace.oe.data, true);
5092 * If we are augmenting syscalls, then combine what we put in the
5093 * __augmented_syscalls__ BPF map with what is in the
5094 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
5095 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
5097 * We'll switch to look at two BPF maps, one for sys_enter and the
5098 * other for sys_exit when we start augmenting the sys_exit paths with
5099 * buffers that are being copied from kernel to userspace, think 'read'
5102 if (trace.syscalls.events.augmented) {
5103 evlist__for_each_entry(trace.evlist, evsel) {
5104 bool raw_syscalls_sys_exit = strcmp(evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
5106 if (raw_syscalls_sys_exit) {
5107 trace.raw_augmented_syscalls = true;
5108 goto init_augmented_syscall_tp;
5111 if (trace.syscalls.events.augmented->priv == NULL &&
5112 strstr(evsel__name(evsel), "syscalls:sys_enter")) {
5113 struct evsel *augmented = trace.syscalls.events.augmented;
5114 if (evsel__init_augmented_syscall_tp(augmented, evsel) ||
5115 evsel__init_augmented_syscall_tp_args(augmented))
5118 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
5119 * Above we made sure we can get from the payload the tp fields
5120 * that we get from syscalls:sys_enter tracefs format file.
5122 augmented->handler = trace__sys_enter;
5124 * Now we do the same for the *syscalls:sys_enter event so that
5125 * if we handle it directly, i.e. if the BPF prog returns 0 so
5126 * as not to filter it, then we'll handle it just like we would
5127 * for the BPF_OUTPUT one:
5129 if (evsel__init_augmented_syscall_tp(evsel, evsel) ||
5130 evsel__init_augmented_syscall_tp_args(evsel))
5132 evsel->handler = trace__sys_enter;
5135 if (strstarts(evsel__name(evsel), "syscalls:sys_exit_")) {
5136 struct syscall_tp *sc;
5137 init_augmented_syscall_tp:
5138 if (evsel__init_augmented_syscall_tp(evsel, evsel))
5140 sc = __evsel__syscall_tp(evsel);
5142 * For now with BPF raw_augmented we hook into
5143 * raw_syscalls:sys_enter and there we get all
5144 * 6 syscall args plus the tracepoint common
5145 * fields and the syscall_nr (another long).
5146 * So we check if that is the case and if so
5147 * don't look after the sc->args_size but
5148 * always after the full raw_syscalls:sys_enter
5149 * payload, which is fixed.
5151 * We'll revisit this later to pass
5152 * s->args_size to the BPF augmenter (now
5153 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
5154 * so that it copies only what we need for each
5155 * syscall, like what happens when we use
5156 * syscalls:sys_enter_NAME, so that we reduce
5157 * the kernel/userspace traffic to just what is
5158 * needed for each syscall.
5160 if (trace.raw_augmented_syscalls)
5161 trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
5162 evsel__init_augmented_syscall_tp_ret(evsel);
5163 evsel->handler = trace__sys_exit;
5168 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
5169 return trace__record(&trace, argc-1, &argv[1]);
5171 /* Using just --errno-summary will trigger --summary */
5172 if (trace.errno_summary && !trace.summary && !trace.summary_only)
5173 trace.summary_only = true;
5175 /* summary_only implies summary option, but don't overwrite summary if set */
5176 if (trace.summary_only)
5177 trace.summary = trace.summary_only;
5179 if (output_name != NULL) {
5180 err = trace__open_output(&trace, output_name);
5182 perror("failed to create output file");
5187 err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
5191 err = target__validate(&trace.opts.target);
5193 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5194 fprintf(trace.output, "%s", bf);
5198 err = target__parse_uid(&trace.opts.target);
5200 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
5201 fprintf(trace.output, "%s", bf);
5205 if (!argc && target__none(&trace.opts.target))
5206 trace.opts.target.system_wide = true;
5209 err = trace__replay(&trace);
5211 err = trace__run(&trace, argc, argv);
5214 if (output_name != NULL)
5215 fclose(trace.output);
5217 trace__exit(&trace);