xref: /linux/tools/perf/builtin-trace.c (revision c1a604dff486399ae0be95e6396e0158df95ad5d)
1 /*
2  * builtin-trace.c
3  *
4  * Builtin 'trace' command:
5  *
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.
9  *
10  * Copyright (C) 2012, 2013, 2014, 2015 Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
11  *
12  * Initially based on the 'trace' prototype by Thomas Gleixner:
13  *
14  * http://lwn.net/Articles/415728/ ("Announcing a new utility: 'trace'")
15  */
16 
17 #include "util/record.h"
18 #include <traceevent/event-parse.h>
19 #include <api/fs/tracing_path.h>
20 #include <bpf/bpf.h>
21 #include "util/bpf_map.h"
22 #include "util/rlimit.h"
23 #include "builtin.h"
24 #include "util/cgroup.h"
25 #include "util/color.h"
26 #include "util/config.h"
27 #include "util/debug.h"
28 #include "util/env.h"
29 #include "util/event.h"
30 #include "util/evlist.h"
31 #include "util/evswitch.h"
32 #include <subcmd/exec-cmd.h>
33 #include "util/machine.h"
34 #include "util/map.h"
35 #include "util/symbol.h"
36 #include "util/path.h"
37 #include "util/session.h"
38 #include "util/thread.h"
39 #include <subcmd/parse-options.h>
40 #include "util/strlist.h"
41 #include "util/intlist.h"
42 #include "util/thread_map.h"
43 #include "util/stat.h"
44 #include "util/util.h"
45 #include "trace/beauty/beauty.h"
46 #include "trace-event.h"
47 #include "util/parse-events.h"
48 #include "util/bpf-loader.h"
49 #include "callchain.h"
50 #include "print_binary.h"
51 #include "string2.h"
52 #include "syscalltbl.h"
53 #include "rb_resort.h"
54 #include "../perf.h"
55 
56 #include <errno.h>
57 #include <inttypes.h>
58 #include <poll.h>
59 #include <signal.h>
60 #include <stdlib.h>
61 #include <string.h>
62 #include <linux/err.h>
63 #include <linux/filter.h>
64 #include <linux/kernel.h>
65 #include <linux/random.h>
66 #include <linux/stringify.h>
67 #include <linux/time64.h>
68 #include <linux/zalloc.h>
69 #include <fcntl.h>
70 #include <sys/sysmacros.h>
71 
72 #include <linux/ctype.h>
73 
74 #ifndef O_CLOEXEC
75 # define O_CLOEXEC		02000000
76 #endif
77 
78 #ifndef F_LINUX_SPECIFIC_BASE
79 # define F_LINUX_SPECIFIC_BASE	1024
80 #endif
81 
82 struct trace {
83 	struct perf_tool	tool;
84 	struct syscalltbl	*sctbl;
85 	struct {
86 		struct syscall  *table;
87 		struct bpf_map  *map;
88 		struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
89 			struct bpf_map  *sys_enter,
90 					*sys_exit;
91 		}		prog_array;
92 		struct {
93 			struct evsel *sys_enter,
94 					  *sys_exit,
95 					  *augmented;
96 		}		events;
97 		struct bpf_program *unaugmented_prog;
98 	} syscalls;
99 	struct {
100 		struct bpf_map *map;
101 	} dump;
102 	struct record_opts	opts;
103 	struct evlist	*evlist;
104 	struct machine		*host;
105 	struct thread		*current;
106 	struct bpf_object	*bpf_obj;
107 	struct cgroup		*cgroup;
108 	u64			base_time;
109 	FILE			*output;
110 	unsigned long		nr_events;
111 	unsigned long		nr_events_printed;
112 	unsigned long		max_events;
113 	struct evswitch		evswitch;
114 	struct strlist		*ev_qualifier;
115 	struct {
116 		size_t		nr;
117 		int		*entries;
118 	}			ev_qualifier_ids;
119 	struct {
120 		size_t		nr;
121 		pid_t		*entries;
122 		struct bpf_map  *map;
123 	}			filter_pids;
124 	double			duration_filter;
125 	double			runtime_ms;
126 	struct {
127 		u64		vfs_getname,
128 				proc_getname;
129 	} stats;
130 	unsigned int		max_stack;
131 	unsigned int		min_stack;
132 	int			raw_augmented_syscalls_args_size;
133 	bool			raw_augmented_syscalls;
134 	bool			fd_path_disabled;
135 	bool			sort_events;
136 	bool			not_ev_qualifier;
137 	bool			live;
138 	bool			full_time;
139 	bool			sched;
140 	bool			multiple_threads;
141 	bool			summary;
142 	bool			summary_only;
143 	bool			failure_only;
144 	bool			show_comm;
145 	bool			print_sample;
146 	bool			show_tool_stats;
147 	bool			trace_syscalls;
148 	bool			kernel_syscallchains;
149 	s16			args_alignment;
150 	bool			show_tstamp;
151 	bool			show_duration;
152 	bool			show_zeros;
153 	bool			show_arg_names;
154 	bool			show_string_prefix;
155 	bool			force;
156 	bool			vfs_getname;
157 	int			trace_pgfaults;
158 	struct {
159 		struct ordered_events	data;
160 		u64			last;
161 	} oe;
162 };
163 
164 struct tp_field {
165 	int offset;
166 	union {
167 		u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
168 		void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
169 	};
170 };
171 
172 #define TP_UINT_FIELD(bits) \
173 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
174 { \
175 	u##bits value; \
176 	memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
177 	return value;  \
178 }
179 
180 TP_UINT_FIELD(8);
181 TP_UINT_FIELD(16);
182 TP_UINT_FIELD(32);
183 TP_UINT_FIELD(64);
184 
185 #define TP_UINT_FIELD__SWAPPED(bits) \
186 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
187 { \
188 	u##bits value; \
189 	memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
190 	return bswap_##bits(value);\
191 }
192 
193 TP_UINT_FIELD__SWAPPED(16);
194 TP_UINT_FIELD__SWAPPED(32);
195 TP_UINT_FIELD__SWAPPED(64);
196 
197 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
198 {
199 	field->offset = offset;
200 
201 	switch (size) {
202 	case 1:
203 		field->integer = tp_field__u8;
204 		break;
205 	case 2:
206 		field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
207 		break;
208 	case 4:
209 		field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
210 		break;
211 	case 8:
212 		field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
213 		break;
214 	default:
215 		return -1;
216 	}
217 
218 	return 0;
219 }
220 
221 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
222 {
223 	return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
224 }
225 
226 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
227 {
228 	return sample->raw_data + field->offset;
229 }
230 
231 static int __tp_field__init_ptr(struct tp_field *field, int offset)
232 {
233 	field->offset = offset;
234 	field->pointer = tp_field__ptr;
235 	return 0;
236 }
237 
238 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
239 {
240 	return __tp_field__init_ptr(field, format_field->offset);
241 }
242 
243 struct syscall_tp {
244 	struct tp_field id;
245 	union {
246 		struct tp_field args, ret;
247 	};
248 };
249 
250 static int perf_evsel__init_tp_uint_field(struct evsel *evsel,
251 					  struct tp_field *field,
252 					  const char *name)
253 {
254 	struct tep_format_field *format_field = perf_evsel__field(evsel, name);
255 
256 	if (format_field == NULL)
257 		return -1;
258 
259 	return tp_field__init_uint(field, format_field, evsel->needs_swap);
260 }
261 
262 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
263 	({ struct syscall_tp *sc = evsel->priv;\
264 	   perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
265 
266 static int perf_evsel__init_tp_ptr_field(struct evsel *evsel,
267 					 struct tp_field *field,
268 					 const char *name)
269 {
270 	struct tep_format_field *format_field = perf_evsel__field(evsel, name);
271 
272 	if (format_field == NULL)
273 		return -1;
274 
275 	return tp_field__init_ptr(field, format_field);
276 }
277 
278 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
279 	({ struct syscall_tp *sc = evsel->priv;\
280 	   perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
281 
282 static void evsel__delete_priv(struct evsel *evsel)
283 {
284 	zfree(&evsel->priv);
285 	evsel__delete(evsel);
286 }
287 
288 static int perf_evsel__init_syscall_tp(struct evsel *evsel)
289 {
290 	struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
291 
292 	if (evsel->priv != NULL) {
293 		if (perf_evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
294 		    perf_evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
295 			goto out_delete;
296 		return 0;
297 	}
298 
299 	return -ENOMEM;
300 out_delete:
301 	zfree(&evsel->priv);
302 	return -ENOENT;
303 }
304 
305 static int perf_evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
306 {
307 	struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
308 
309 	if (evsel->priv != NULL) {
310 		struct tep_format_field *syscall_id = perf_evsel__field(tp, "id");
311 		if (syscall_id == NULL)
312 			syscall_id = perf_evsel__field(tp, "__syscall_nr");
313 		if (syscall_id == NULL)
314 			goto out_delete;
315 		if (__tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
316 			goto out_delete;
317 
318 		return 0;
319 	}
320 
321 	return -ENOMEM;
322 out_delete:
323 	zfree(&evsel->priv);
324 	return -EINVAL;
325 }
326 
327 static int perf_evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
328 {
329 	struct syscall_tp *sc = evsel->priv;
330 
331 	return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
332 }
333 
334 static int perf_evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
335 {
336 	struct syscall_tp *sc = evsel->priv;
337 
338 	return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
339 }
340 
341 static int perf_evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
342 {
343 	evsel->priv = malloc(sizeof(struct syscall_tp));
344 	if (evsel->priv != NULL) {
345 		if (perf_evsel__init_sc_tp_uint_field(evsel, id))
346 			goto out_delete;
347 
348 		evsel->handler = handler;
349 		return 0;
350 	}
351 
352 	return -ENOMEM;
353 
354 out_delete:
355 	zfree(&evsel->priv);
356 	return -ENOENT;
357 }
358 
359 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
360 {
361 	struct evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
362 
363 	/* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
364 	if (IS_ERR(evsel))
365 		evsel = perf_evsel__newtp("syscalls", direction);
366 
367 	if (IS_ERR(evsel))
368 		return NULL;
369 
370 	if (perf_evsel__init_raw_syscall_tp(evsel, handler))
371 		goto out_delete;
372 
373 	return evsel;
374 
375 out_delete:
376 	evsel__delete_priv(evsel);
377 	return NULL;
378 }
379 
380 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
381 	({ struct syscall_tp *fields = evsel->priv; \
382 	   fields->name.integer(&fields->name, sample); })
383 
384 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
385 	({ struct syscall_tp *fields = evsel->priv; \
386 	   fields->name.pointer(&fields->name, sample); })
387 
388 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
389 {
390 	int idx = val - sa->offset;
391 
392 	if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
393 		size_t printed = scnprintf(bf, size, intfmt, val);
394 		if (show_prefix)
395 			printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
396 		return printed;
397 	}
398 
399 	return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
400 }
401 
402 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
403 						const char *intfmt,
404 					        struct syscall_arg *arg)
405 {
406 	return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
407 }
408 
409 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
410 					      struct syscall_arg *arg)
411 {
412 	return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
413 }
414 
415 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
416 
417 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
418 {
419 	return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
420 }
421 
422 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
423 {
424 	size_t printed;
425 	int i;
426 
427 	for (i = 0; i < sas->nr_entries; ++i) {
428 		struct strarray *sa = sas->entries[i];
429 		int idx = val - sa->offset;
430 
431 		if (idx >= 0 && idx < sa->nr_entries) {
432 			if (sa->entries[idx] == NULL)
433 				break;
434 			return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
435 		}
436 	}
437 
438 	printed = scnprintf(bf, size, intfmt, val);
439 	if (show_prefix)
440 		printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
441 	return printed;
442 }
443 
444 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
445 					struct syscall_arg *arg)
446 {
447 	return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
448 }
449 
450 #ifndef AT_FDCWD
451 #define AT_FDCWD	-100
452 #endif
453 
454 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
455 					   struct syscall_arg *arg)
456 {
457 	int fd = arg->val;
458 	const char *prefix = "AT_FD";
459 
460 	if (fd == AT_FDCWD)
461 		return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
462 
463 	return syscall_arg__scnprintf_fd(bf, size, arg);
464 }
465 
466 #define SCA_FDAT syscall_arg__scnprintf_fd_at
467 
468 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
469 					      struct syscall_arg *arg);
470 
471 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
472 
473 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
474 {
475 	return scnprintf(bf, size, "%#lx", arg->val);
476 }
477 
478 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
479 {
480 	if (arg->val == 0)
481 		return scnprintf(bf, size, "NULL");
482 	return syscall_arg__scnprintf_hex(bf, size, arg);
483 }
484 
485 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
486 {
487 	return scnprintf(bf, size, "%d", arg->val);
488 }
489 
490 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
491 {
492 	return scnprintf(bf, size, "%ld", arg->val);
493 }
494 
495 static const char *bpf_cmd[] = {
496 	"MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
497 	"MAP_GET_NEXT_KEY", "PROG_LOAD",
498 };
499 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
500 
501 static const char *fsmount_flags[] = {
502 	[1] = "CLOEXEC",
503 };
504 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
505 
506 #include "trace/beauty/generated/fsconfig_arrays.c"
507 
508 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
509 
510 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
511 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
512 
513 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
514 static DEFINE_STRARRAY(itimers, "ITIMER_");
515 
516 static const char *keyctl_options[] = {
517 	"GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
518 	"SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
519 	"INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
520 	"ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
521 	"INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
522 };
523 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
524 
525 static const char *whences[] = { "SET", "CUR", "END",
526 #ifdef SEEK_DATA
527 "DATA",
528 #endif
529 #ifdef SEEK_HOLE
530 "HOLE",
531 #endif
532 };
533 static DEFINE_STRARRAY(whences, "SEEK_");
534 
535 static const char *fcntl_cmds[] = {
536 	"DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
537 	"SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
538 	"SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
539 	"GETOWNER_UIDS",
540 };
541 static DEFINE_STRARRAY(fcntl_cmds, "F_");
542 
543 static const char *fcntl_linux_specific_cmds[] = {
544 	"SETLEASE", "GETLEASE", "NOTIFY", [5] =	"CANCELLK", "DUPFD_CLOEXEC",
545 	"SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
546 	"GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
547 };
548 
549 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
550 
551 static struct strarray *fcntl_cmds_arrays[] = {
552 	&strarray__fcntl_cmds,
553 	&strarray__fcntl_linux_specific_cmds,
554 };
555 
556 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
557 
558 static const char *rlimit_resources[] = {
559 	"CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
560 	"MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
561 	"RTTIME",
562 };
563 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
564 
565 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
566 static DEFINE_STRARRAY(sighow, "SIG_");
567 
568 static const char *clockid[] = {
569 	"REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
570 	"MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
571 	"REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
572 };
573 static DEFINE_STRARRAY(clockid, "CLOCK_");
574 
575 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
576 						 struct syscall_arg *arg)
577 {
578 	bool show_prefix = arg->show_string_prefix;
579 	const char *suffix = "_OK";
580 	size_t printed = 0;
581 	int mode = arg->val;
582 
583 	if (mode == F_OK) /* 0 */
584 		return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
585 #define	P_MODE(n) \
586 	if (mode & n##_OK) { \
587 		printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
588 		mode &= ~n##_OK; \
589 	}
590 
591 	P_MODE(R);
592 	P_MODE(W);
593 	P_MODE(X);
594 #undef P_MODE
595 
596 	if (mode)
597 		printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
598 
599 	return printed;
600 }
601 
602 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
603 
604 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
605 					      struct syscall_arg *arg);
606 
607 #define SCA_FILENAME syscall_arg__scnprintf_filename
608 
609 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
610 						struct syscall_arg *arg)
611 {
612 	bool show_prefix = arg->show_string_prefix;
613 	const char *prefix = "O_";
614 	int printed = 0, flags = arg->val;
615 
616 #define	P_FLAG(n) \
617 	if (flags & O_##n) { \
618 		printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
619 		flags &= ~O_##n; \
620 	}
621 
622 	P_FLAG(CLOEXEC);
623 	P_FLAG(NONBLOCK);
624 #undef P_FLAG
625 
626 	if (flags)
627 		printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
628 
629 	return printed;
630 }
631 
632 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
633 
634 #ifndef GRND_NONBLOCK
635 #define GRND_NONBLOCK	0x0001
636 #endif
637 #ifndef GRND_RANDOM
638 #define GRND_RANDOM	0x0002
639 #endif
640 
641 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
642 						   struct syscall_arg *arg)
643 {
644 	bool show_prefix = arg->show_string_prefix;
645 	const char *prefix = "GRND_";
646 	int printed = 0, flags = arg->val;
647 
648 #define	P_FLAG(n) \
649 	if (flags & GRND_##n) { \
650 		printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
651 		flags &= ~GRND_##n; \
652 	}
653 
654 	P_FLAG(RANDOM);
655 	P_FLAG(NONBLOCK);
656 #undef P_FLAG
657 
658 	if (flags)
659 		printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
660 
661 	return printed;
662 }
663 
664 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
665 
666 #define STRARRAY(name, array) \
667 	  { .scnprintf	= SCA_STRARRAY, \
668 	    .parm	= &strarray__##array, }
669 
670 #define STRARRAY_FLAGS(name, array) \
671 	  { .scnprintf	= SCA_STRARRAY_FLAGS, \
672 	    .parm	= &strarray__##array, }
673 
674 #include "trace/beauty/arch_errno_names.c"
675 #include "trace/beauty/eventfd.c"
676 #include "trace/beauty/futex_op.c"
677 #include "trace/beauty/futex_val3.c"
678 #include "trace/beauty/mmap.c"
679 #include "trace/beauty/mode_t.c"
680 #include "trace/beauty/msg_flags.c"
681 #include "trace/beauty/open_flags.c"
682 #include "trace/beauty/perf_event_open.c"
683 #include "trace/beauty/pid.c"
684 #include "trace/beauty/sched_policy.c"
685 #include "trace/beauty/seccomp.c"
686 #include "trace/beauty/signum.c"
687 #include "trace/beauty/socket_type.c"
688 #include "trace/beauty/waitid_options.c"
689 
690 struct syscall_arg_fmt {
691 	size_t	   (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
692 	unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
693 	void	   *parm;
694 	const char *name;
695 	bool	   show_zero;
696 };
697 
698 static struct syscall_fmt {
699 	const char *name;
700 	const char *alias;
701 	struct {
702 		const char *sys_enter,
703 			   *sys_exit;
704 	}	   bpf_prog_name;
705 	struct syscall_arg_fmt arg[6];
706 	u8	   nr_args;
707 	bool	   errpid;
708 	bool	   timeout;
709 	bool	   hexret;
710 } syscall_fmts[] = {
711 	{ .name	    = "access",
712 	  .arg = { [1] = { .scnprintf = SCA_ACCMODE,  /* mode */ }, }, },
713 	{ .name	    = "arch_prctl",
714 	  .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
715 		   [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
716 	{ .name	    = "bind",
717 	  .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
718 		   [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
719 		   [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
720 	{ .name	    = "bpf",
721 	  .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
722 	{ .name	    = "brk",	    .hexret = true,
723 	  .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
724 	{ .name     = "clock_gettime",
725 	  .arg = { [0] = STRARRAY(clk_id, clockid), }, },
726 	{ .name	    = "clone",	    .errpid = true, .nr_args = 5,
727 	  .arg = { [0] = { .name = "flags",	    .scnprintf = SCA_CLONE_FLAGS, },
728 		   [1] = { .name = "child_stack",   .scnprintf = SCA_HEX, },
729 		   [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
730 		   [3] = { .name = "child_tidptr",  .scnprintf = SCA_HEX, },
731 		   [4] = { .name = "tls",	    .scnprintf = SCA_HEX, }, }, },
732 	{ .name	    = "close",
733 	  .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
734 	{ .name	    = "connect",
735 	  .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
736 		   [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
737 		   [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
738 	{ .name	    = "epoll_ctl",
739 	  .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
740 	{ .name	    = "eventfd2",
741 	  .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
742 	{ .name	    = "fchmodat",
743 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
744 	{ .name	    = "fchownat",
745 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
746 	{ .name	    = "fcntl",
747 	  .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
748 			   .parm      = &strarrays__fcntl_cmds_arrays,
749 			   .show_zero = true, },
750 		   [2] = { .scnprintf =  SCA_FCNTL_ARG, /* arg */ }, }, },
751 	{ .name	    = "flock",
752 	  .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
753 	{ .name     = "fsconfig",
754 	  .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
755 	{ .name     = "fsmount",
756 	  .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
757 		   [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
758 	{ .name     = "fspick",
759 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dfd */ },
760 		   [1] = { .scnprintf = SCA_FILENAME,	  /* path */ },
761 		   [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
762 	{ .name	    = "fstat", .alias = "newfstat", },
763 	{ .name	    = "fstatat", .alias = "newfstatat", },
764 	{ .name	    = "futex",
765 	  .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
766 		   [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
767 	{ .name	    = "futimesat",
768 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
769 	{ .name	    = "getitimer",
770 	  .arg = { [0] = STRARRAY(which, itimers), }, },
771 	{ .name	    = "getpid",	    .errpid = true, },
772 	{ .name	    = "getpgid",    .errpid = true, },
773 	{ .name	    = "getppid",    .errpid = true, },
774 	{ .name	    = "getrandom",
775 	  .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
776 	{ .name	    = "getrlimit",
777 	  .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
778 	{ .name	    = "gettid",	    .errpid = true, },
779 	{ .name	    = "ioctl",
780 	  .arg = {
781 #if defined(__i386__) || defined(__x86_64__)
782 /*
783  * FIXME: Make this available to all arches.
784  */
785 		   [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
786 		   [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
787 #else
788 		   [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
789 #endif
790 	{ .name	    = "kcmp",	    .nr_args = 5,
791 	  .arg = { [0] = { .name = "pid1",	.scnprintf = SCA_PID, },
792 		   [1] = { .name = "pid2",	.scnprintf = SCA_PID, },
793 		   [2] = { .name = "type",	.scnprintf = SCA_KCMP_TYPE, },
794 		   [3] = { .name = "idx1",	.scnprintf = SCA_KCMP_IDX, },
795 		   [4] = { .name = "idx2",	.scnprintf = SCA_KCMP_IDX, }, }, },
796 	{ .name	    = "keyctl",
797 	  .arg = { [0] = STRARRAY(option, keyctl_options), }, },
798 	{ .name	    = "kill",
799 	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
800 	{ .name	    = "linkat",
801 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
802 	{ .name	    = "lseek",
803 	  .arg = { [2] = STRARRAY(whence, whences), }, },
804 	{ .name	    = "lstat", .alias = "newlstat", },
805 	{ .name     = "madvise",
806 	  .arg = { [0] = { .scnprintf = SCA_HEX,      /* start */ },
807 		   [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
808 	{ .name	    = "mkdirat",
809 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
810 	{ .name	    = "mknodat",
811 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
812 	{ .name	    = "mmap",	    .hexret = true,
813 /* The standard mmap maps to old_mmap on s390x */
814 #if defined(__s390x__)
815 	.alias = "old_mmap",
816 #endif
817 	  .arg = { [2] = { .scnprintf = SCA_MMAP_PROT,	/* prot */ },
818 		   [3] = { .scnprintf = SCA_MMAP_FLAGS,	/* flags */ },
819 		   [5] = { .scnprintf = SCA_HEX,	/* offset */ }, }, },
820 	{ .name	    = "mount",
821 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
822 		   [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
823 			   .mask_val  = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
824 	{ .name	    = "move_mount",
825 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* from_dfd */ },
826 		   [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
827 		   [2] = { .scnprintf = SCA_FDAT,	/* to_dfd */ },
828 		   [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
829 		   [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
830 	{ .name	    = "mprotect",
831 	  .arg = { [0] = { .scnprintf = SCA_HEX,	/* start */ },
832 		   [2] = { .scnprintf = SCA_MMAP_PROT,	/* prot */ }, }, },
833 	{ .name	    = "mq_unlink",
834 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
835 	{ .name	    = "mremap",	    .hexret = true,
836 	  .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
837 	{ .name	    = "name_to_handle_at",
838 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
839 	{ .name	    = "newfstatat",
840 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
841 	{ .name	    = "open",
842 	  .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
843 	{ .name	    = "open_by_handle_at",
844 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },
845 		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
846 	{ .name	    = "openat",
847 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },
848 		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
849 	{ .name	    = "perf_event_open",
850 	  .arg = { [2] = { .scnprintf = SCA_INT,	/* cpu */ },
851 		   [3] = { .scnprintf = SCA_FD,		/* group_fd */ },
852 		   [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
853 	{ .name	    = "pipe2",
854 	  .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
855 	{ .name	    = "pkey_alloc",
856 	  .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS,	/* access_rights */ }, }, },
857 	{ .name	    = "pkey_free",
858 	  .arg = { [0] = { .scnprintf = SCA_INT,	/* key */ }, }, },
859 	{ .name	    = "pkey_mprotect",
860 	  .arg = { [0] = { .scnprintf = SCA_HEX,	/* start */ },
861 		   [2] = { .scnprintf = SCA_MMAP_PROT,	/* prot */ },
862 		   [3] = { .scnprintf = SCA_INT,	/* pkey */ }, }, },
863 	{ .name	    = "poll", .timeout = true, },
864 	{ .name	    = "ppoll", .timeout = true, },
865 	{ .name	    = "prctl",
866 	  .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */ },
867 		   [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
868 		   [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
869 	{ .name	    = "pread", .alias = "pread64", },
870 	{ .name	    = "preadv", .alias = "pread", },
871 	{ .name	    = "prlimit64",
872 	  .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
873 	{ .name	    = "pwrite", .alias = "pwrite64", },
874 	{ .name	    = "readlinkat",
875 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
876 	{ .name	    = "recvfrom",
877 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
878 	{ .name	    = "recvmmsg",
879 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
880 	{ .name	    = "recvmsg",
881 	  .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
882 	{ .name	    = "renameat",
883 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
884 		   [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
885 	{ .name	    = "renameat2",
886 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
887 		   [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
888 		   [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
889 	{ .name	    = "rt_sigaction",
890 	  .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
891 	{ .name	    = "rt_sigprocmask",
892 	  .arg = { [0] = STRARRAY(how, sighow), }, },
893 	{ .name	    = "rt_sigqueueinfo",
894 	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
895 	{ .name	    = "rt_tgsigqueueinfo",
896 	  .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
897 	{ .name	    = "sched_setscheduler",
898 	  .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
899 	{ .name	    = "seccomp",
900 	  .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP,	   /* op */ },
901 		   [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
902 	{ .name	    = "select", .timeout = true, },
903 	{ .name	    = "sendfile", .alias = "sendfile64", },
904 	{ .name	    = "sendmmsg",
905 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
906 	{ .name	    = "sendmsg",
907 	  .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
908 	{ .name	    = "sendto",
909 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
910 		   [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
911 	{ .name	    = "set_tid_address", .errpid = true, },
912 	{ .name	    = "setitimer",
913 	  .arg = { [0] = STRARRAY(which, itimers), }, },
914 	{ .name	    = "setrlimit",
915 	  .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
916 	{ .name	    = "socket",
917 	  .arg = { [0] = STRARRAY(family, socket_families),
918 		   [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
919 		   [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
920 	{ .name	    = "socketpair",
921 	  .arg = { [0] = STRARRAY(family, socket_families),
922 		   [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
923 		   [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
924 	{ .name	    = "stat", .alias = "newstat", },
925 	{ .name	    = "statx",
926 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	 /* fdat */ },
927 		   [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
928 		   [3] = { .scnprintf = SCA_STATX_MASK,	 /* mask */ }, }, },
929 	{ .name	    = "swapoff",
930 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
931 	{ .name	    = "swapon",
932 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
933 	{ .name	    = "symlinkat",
934 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
935 	{ .name	    = "sync_file_range",
936 	  .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
937 	{ .name	    = "tgkill",
938 	  .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
939 	{ .name	    = "tkill",
940 	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
941 	{ .name     = "umount2", .alias = "umount",
942 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
943 	{ .name	    = "uname", .alias = "newuname", },
944 	{ .name	    = "unlinkat",
945 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
946 	{ .name	    = "utimensat",
947 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
948 	{ .name	    = "wait4",	    .errpid = true,
949 	  .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
950 	{ .name	    = "waitid",	    .errpid = true,
951 	  .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
952 };
953 
954 static int syscall_fmt__cmp(const void *name, const void *fmtp)
955 {
956 	const struct syscall_fmt *fmt = fmtp;
957 	return strcmp(name, fmt->name);
958 }
959 
960 static struct syscall_fmt *syscall_fmt__find(const char *name)
961 {
962 	const int nmemb = ARRAY_SIZE(syscall_fmts);
963 	return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
964 }
965 
966 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
967 {
968 	int i, nmemb = ARRAY_SIZE(syscall_fmts);
969 
970 	for (i = 0; i < nmemb; ++i) {
971 		if (syscall_fmts[i].alias && strcmp(syscall_fmts[i].alias, alias) == 0)
972 			return &syscall_fmts[i];
973 	}
974 
975 	return NULL;
976 }
977 
978 /*
979  * is_exit: is this "exit" or "exit_group"?
980  * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
981  * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
982  * nonexistent: Just a hole in the syscall table, syscall id not allocated
983  */
984 struct syscall {
985 	struct tep_event    *tp_format;
986 	int		    nr_args;
987 	int		    args_size;
988 	struct {
989 		struct bpf_program *sys_enter,
990 				   *sys_exit;
991 	}		    bpf_prog;
992 	bool		    is_exit;
993 	bool		    is_open;
994 	bool		    nonexistent;
995 	struct tep_format_field *args;
996 	const char	    *name;
997 	struct syscall_fmt  *fmt;
998 	struct syscall_arg_fmt *arg_fmt;
999 };
1000 
1001 /*
1002  * Must match what is in the BPF program:
1003  *
1004  * tools/perf/examples/bpf/augmented_raw_syscalls.c
1005  */
1006 struct bpf_map_syscall_entry {
1007 	bool	enabled;
1008 	u16	string_args_len[6];
1009 };
1010 
1011 /*
1012  * We need to have this 'calculated' boolean because in some cases we really
1013  * don't know what is the duration of a syscall, for instance, when we start
1014  * a session and some threads are waiting for a syscall to finish, say 'poll',
1015  * in which case all we can do is to print "( ? ) for duration and for the
1016  * start timestamp.
1017  */
1018 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1019 {
1020 	double duration = (double)t / NSEC_PER_MSEC;
1021 	size_t printed = fprintf(fp, "(");
1022 
1023 	if (!calculated)
1024 		printed += fprintf(fp, "         ");
1025 	else if (duration >= 1.0)
1026 		printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1027 	else if (duration >= 0.01)
1028 		printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1029 	else
1030 		printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1031 	return printed + fprintf(fp, "): ");
1032 }
1033 
1034 /**
1035  * filename.ptr: The filename char pointer that will be vfs_getname'd
1036  * filename.entry_str_pos: Where to insert the string translated from
1037  *                         filename.ptr by the vfs_getname tracepoint/kprobe.
1038  * ret_scnprintf: syscall args may set this to a different syscall return
1039  *                formatter, for instance, fcntl may return fds, file flags, etc.
1040  */
1041 struct thread_trace {
1042 	u64		  entry_time;
1043 	bool		  entry_pending;
1044 	unsigned long	  nr_events;
1045 	unsigned long	  pfmaj, pfmin;
1046 	char		  *entry_str;
1047 	double		  runtime_ms;
1048 	size_t		  (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1049         struct {
1050 		unsigned long ptr;
1051 		short int     entry_str_pos;
1052 		bool	      pending_open;
1053 		unsigned int  namelen;
1054 		char	      *name;
1055 	} filename;
1056 	struct {
1057 		int	      max;
1058 		struct file   *table;
1059 	} files;
1060 
1061 	struct intlist *syscall_stats;
1062 };
1063 
1064 static struct thread_trace *thread_trace__new(void)
1065 {
1066 	struct thread_trace *ttrace =  zalloc(sizeof(struct thread_trace));
1067 
1068 	if (ttrace) {
1069 		ttrace->files.max = -1;
1070 		ttrace->syscall_stats = intlist__new(NULL);
1071 	}
1072 
1073 	return ttrace;
1074 }
1075 
1076 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1077 {
1078 	struct thread_trace *ttrace;
1079 
1080 	if (thread == NULL)
1081 		goto fail;
1082 
1083 	if (thread__priv(thread) == NULL)
1084 		thread__set_priv(thread, thread_trace__new());
1085 
1086 	if (thread__priv(thread) == NULL)
1087 		goto fail;
1088 
1089 	ttrace = thread__priv(thread);
1090 	++ttrace->nr_events;
1091 
1092 	return ttrace;
1093 fail:
1094 	color_fprintf(fp, PERF_COLOR_RED,
1095 		      "WARNING: not enough memory, dropping samples!\n");
1096 	return NULL;
1097 }
1098 
1099 
1100 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1101 				    size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1102 {
1103 	struct thread_trace *ttrace = thread__priv(arg->thread);
1104 
1105 	ttrace->ret_scnprintf = ret_scnprintf;
1106 }
1107 
1108 #define TRACE_PFMAJ		(1 << 0)
1109 #define TRACE_PFMIN		(1 << 1)
1110 
1111 static const size_t trace__entry_str_size = 2048;
1112 
1113 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1114 {
1115 	if (fd < 0)
1116 		return NULL;
1117 
1118 	if (fd > ttrace->files.max) {
1119 		struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1120 
1121 		if (nfiles == NULL)
1122 			return NULL;
1123 
1124 		if (ttrace->files.max != -1) {
1125 			memset(nfiles + ttrace->files.max + 1, 0,
1126 			       (fd - ttrace->files.max) * sizeof(struct file));
1127 		} else {
1128 			memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1129 		}
1130 
1131 		ttrace->files.table = nfiles;
1132 		ttrace->files.max   = fd;
1133 	}
1134 
1135 	return ttrace->files.table + fd;
1136 }
1137 
1138 struct file *thread__files_entry(struct thread *thread, int fd)
1139 {
1140 	return thread_trace__files_entry(thread__priv(thread), fd);
1141 }
1142 
1143 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1144 {
1145 	struct thread_trace *ttrace = thread__priv(thread);
1146 	struct file *file = thread_trace__files_entry(ttrace, fd);
1147 
1148 	if (file != NULL) {
1149 		struct stat st;
1150 		if (stat(pathname, &st) == 0)
1151 			file->dev_maj = major(st.st_rdev);
1152 		file->pathname = strdup(pathname);
1153 		if (file->pathname)
1154 			return 0;
1155 	}
1156 
1157 	return -1;
1158 }
1159 
1160 static int thread__read_fd_path(struct thread *thread, int fd)
1161 {
1162 	char linkname[PATH_MAX], pathname[PATH_MAX];
1163 	struct stat st;
1164 	int ret;
1165 
1166 	if (thread->pid_ == thread->tid) {
1167 		scnprintf(linkname, sizeof(linkname),
1168 			  "/proc/%d/fd/%d", thread->pid_, fd);
1169 	} else {
1170 		scnprintf(linkname, sizeof(linkname),
1171 			  "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1172 	}
1173 
1174 	if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1175 		return -1;
1176 
1177 	ret = readlink(linkname, pathname, sizeof(pathname));
1178 
1179 	if (ret < 0 || ret > st.st_size)
1180 		return -1;
1181 
1182 	pathname[ret] = '\0';
1183 	return trace__set_fd_pathname(thread, fd, pathname);
1184 }
1185 
1186 static const char *thread__fd_path(struct thread *thread, int fd,
1187 				   struct trace *trace)
1188 {
1189 	struct thread_trace *ttrace = thread__priv(thread);
1190 
1191 	if (ttrace == NULL || trace->fd_path_disabled)
1192 		return NULL;
1193 
1194 	if (fd < 0)
1195 		return NULL;
1196 
1197 	if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1198 		if (!trace->live)
1199 			return NULL;
1200 		++trace->stats.proc_getname;
1201 		if (thread__read_fd_path(thread, fd))
1202 			return NULL;
1203 	}
1204 
1205 	return ttrace->files.table[fd].pathname;
1206 }
1207 
1208 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1209 {
1210 	int fd = arg->val;
1211 	size_t printed = scnprintf(bf, size, "%d", fd);
1212 	const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1213 
1214 	if (path)
1215 		printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1216 
1217 	return printed;
1218 }
1219 
1220 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1221 {
1222         size_t printed = scnprintf(bf, size, "%d", fd);
1223 	struct thread *thread = machine__find_thread(trace->host, pid, pid);
1224 
1225 	if (thread) {
1226 		const char *path = thread__fd_path(thread, fd, trace);
1227 
1228 		if (path)
1229 			printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1230 
1231 		thread__put(thread);
1232 	}
1233 
1234         return printed;
1235 }
1236 
1237 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1238 					      struct syscall_arg *arg)
1239 {
1240 	int fd = arg->val;
1241 	size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1242 	struct thread_trace *ttrace = thread__priv(arg->thread);
1243 
1244 	if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1245 		zfree(&ttrace->files.table[fd].pathname);
1246 
1247 	return printed;
1248 }
1249 
1250 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1251 				     unsigned long ptr)
1252 {
1253 	struct thread_trace *ttrace = thread__priv(thread);
1254 
1255 	ttrace->filename.ptr = ptr;
1256 	ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1257 }
1258 
1259 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1260 {
1261 	struct augmented_arg *augmented_arg = arg->augmented.args;
1262 	size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1263 	/*
1264 	 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1265 	 * we would have two strings, each prefixed by its size.
1266 	 */
1267 	int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1268 
1269 	arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1270 	arg->augmented.size -= consumed;
1271 
1272 	return printed;
1273 }
1274 
1275 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1276 					      struct syscall_arg *arg)
1277 {
1278 	unsigned long ptr = arg->val;
1279 
1280 	if (arg->augmented.args)
1281 		return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1282 
1283 	if (!arg->trace->vfs_getname)
1284 		return scnprintf(bf, size, "%#x", ptr);
1285 
1286 	thread__set_filename_pos(arg->thread, bf, ptr);
1287 	return 0;
1288 }
1289 
1290 static bool trace__filter_duration(struct trace *trace, double t)
1291 {
1292 	return t < (trace->duration_filter * NSEC_PER_MSEC);
1293 }
1294 
1295 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1296 {
1297 	double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1298 
1299 	return fprintf(fp, "%10.3f ", ts);
1300 }
1301 
1302 /*
1303  * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1304  * using ttrace->entry_time for a thread that receives a sys_exit without
1305  * first having received a sys_enter ("poll" issued before tracing session
1306  * starts, lost sys_enter exit due to ring buffer overflow).
1307  */
1308 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1309 {
1310 	if (tstamp > 0)
1311 		return __trace__fprintf_tstamp(trace, tstamp, fp);
1312 
1313 	return fprintf(fp, "         ? ");
1314 }
1315 
1316 static bool done = false;
1317 static bool interrupted = false;
1318 
1319 static void sig_handler(int sig)
1320 {
1321 	done = true;
1322 	interrupted = sig == SIGINT;
1323 }
1324 
1325 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1326 {
1327 	size_t printed = 0;
1328 
1329 	if (trace->multiple_threads) {
1330 		if (trace->show_comm)
1331 			printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1332 		printed += fprintf(fp, "%d ", thread->tid);
1333 	}
1334 
1335 	return printed;
1336 }
1337 
1338 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1339 					u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1340 {
1341 	size_t printed = 0;
1342 
1343 	if (trace->show_tstamp)
1344 		printed = trace__fprintf_tstamp(trace, tstamp, fp);
1345 	if (trace->show_duration)
1346 		printed += fprintf_duration(duration, duration_calculated, fp);
1347 	return printed + trace__fprintf_comm_tid(trace, thread, fp);
1348 }
1349 
1350 static int trace__process_event(struct trace *trace, struct machine *machine,
1351 				union perf_event *event, struct perf_sample *sample)
1352 {
1353 	int ret = 0;
1354 
1355 	switch (event->header.type) {
1356 	case PERF_RECORD_LOST:
1357 		color_fprintf(trace->output, PERF_COLOR_RED,
1358 			      "LOST %" PRIu64 " events!\n", event->lost.lost);
1359 		ret = machine__process_lost_event(machine, event, sample);
1360 		break;
1361 	default:
1362 		ret = machine__process_event(machine, event, sample);
1363 		break;
1364 	}
1365 
1366 	return ret;
1367 }
1368 
1369 static int trace__tool_process(struct perf_tool *tool,
1370 			       union perf_event *event,
1371 			       struct perf_sample *sample,
1372 			       struct machine *machine)
1373 {
1374 	struct trace *trace = container_of(tool, struct trace, tool);
1375 	return trace__process_event(trace, machine, event, sample);
1376 }
1377 
1378 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1379 {
1380 	struct machine *machine = vmachine;
1381 
1382 	if (machine->kptr_restrict_warned)
1383 		return NULL;
1384 
1385 	if (symbol_conf.kptr_restrict) {
1386 		pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1387 			   "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1388 			   "Kernel samples will not be resolved.\n");
1389 		machine->kptr_restrict_warned = true;
1390 		return NULL;
1391 	}
1392 
1393 	return machine__resolve_kernel_addr(vmachine, addrp, modp);
1394 }
1395 
1396 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1397 {
1398 	int err = symbol__init(NULL);
1399 
1400 	if (err)
1401 		return err;
1402 
1403 	trace->host = machine__new_host();
1404 	if (trace->host == NULL)
1405 		return -ENOMEM;
1406 
1407 	err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1408 	if (err < 0)
1409 		goto out;
1410 
1411 	err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1412 					    evlist->core.threads, trace__tool_process, false,
1413 					    1);
1414 out:
1415 	if (err)
1416 		symbol__exit();
1417 
1418 	return err;
1419 }
1420 
1421 static void trace__symbols__exit(struct trace *trace)
1422 {
1423 	machine__exit(trace->host);
1424 	trace->host = NULL;
1425 
1426 	symbol__exit();
1427 }
1428 
1429 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1430 {
1431 	int idx;
1432 
1433 	if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1434 		nr_args = sc->fmt->nr_args;
1435 
1436 	sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1437 	if (sc->arg_fmt == NULL)
1438 		return -1;
1439 
1440 	for (idx = 0; idx < nr_args; ++idx) {
1441 		if (sc->fmt)
1442 			sc->arg_fmt[idx] = sc->fmt->arg[idx];
1443 	}
1444 
1445 	sc->nr_args = nr_args;
1446 	return 0;
1447 }
1448 
1449 static int syscall__set_arg_fmts(struct syscall *sc)
1450 {
1451 	struct tep_format_field *field, *last_field = NULL;
1452 	int idx = 0, len;
1453 
1454 	for (field = sc->args; field; field = field->next, ++idx) {
1455 		last_field = field;
1456 
1457 		if (sc->fmt && sc->fmt->arg[idx].scnprintf)
1458 			continue;
1459 
1460 		len = strlen(field->name);
1461 
1462 		if (strcmp(field->type, "const char *") == 0 &&
1463 		    ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1464 		     strstr(field->name, "path") != NULL))
1465 			sc->arg_fmt[idx].scnprintf = SCA_FILENAME;
1466 		else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1467 			sc->arg_fmt[idx].scnprintf = SCA_PTR;
1468 		else if (strcmp(field->type, "pid_t") == 0)
1469 			sc->arg_fmt[idx].scnprintf = SCA_PID;
1470 		else if (strcmp(field->type, "umode_t") == 0)
1471 			sc->arg_fmt[idx].scnprintf = SCA_MODE_T;
1472 		else if ((strcmp(field->type, "int") == 0 ||
1473 			  strcmp(field->type, "unsigned int") == 0 ||
1474 			  strcmp(field->type, "long") == 0) &&
1475 			 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1476 			/*
1477 			 * /sys/kernel/tracing/events/syscalls/sys_enter*
1478 			 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1479 			 * 65 int
1480 			 * 23 unsigned int
1481 			 * 7 unsigned long
1482 			 */
1483 			sc->arg_fmt[idx].scnprintf = SCA_FD;
1484 		}
1485 	}
1486 
1487 	if (last_field)
1488 		sc->args_size = last_field->offset + last_field->size;
1489 
1490 	return 0;
1491 }
1492 
1493 static int trace__read_syscall_info(struct trace *trace, int id)
1494 {
1495 	char tp_name[128];
1496 	struct syscall *sc;
1497 	const char *name = syscalltbl__name(trace->sctbl, id);
1498 
1499 	if (trace->syscalls.table == NULL) {
1500 		trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1501 		if (trace->syscalls.table == NULL)
1502 			return -ENOMEM;
1503 	}
1504 
1505 	sc = trace->syscalls.table + id;
1506 	if (sc->nonexistent)
1507 		return 0;
1508 
1509 	if (name == NULL) {
1510 		sc->nonexistent = true;
1511 		return 0;
1512 	}
1513 
1514 	sc->name = name;
1515 	sc->fmt  = syscall_fmt__find(sc->name);
1516 
1517 	snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1518 	sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1519 
1520 	if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1521 		snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1522 		sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1523 	}
1524 
1525 	if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1526 		return -ENOMEM;
1527 
1528 	if (IS_ERR(sc->tp_format))
1529 		return PTR_ERR(sc->tp_format);
1530 
1531 	sc->args = sc->tp_format->format.fields;
1532 	/*
1533 	 * We need to check and discard the first variable '__syscall_nr'
1534 	 * or 'nr' that mean the syscall number. It is needless here.
1535 	 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1536 	 */
1537 	if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1538 		sc->args = sc->args->next;
1539 		--sc->nr_args;
1540 	}
1541 
1542 	sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1543 	sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1544 
1545 	return syscall__set_arg_fmts(sc);
1546 }
1547 
1548 static int intcmp(const void *a, const void *b)
1549 {
1550 	const int *one = a, *another = b;
1551 
1552 	return *one - *another;
1553 }
1554 
1555 static int trace__validate_ev_qualifier(struct trace *trace)
1556 {
1557 	int err = 0;
1558 	bool printed_invalid_prefix = false;
1559 	struct str_node *pos;
1560 	size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1561 
1562 	trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1563 						 sizeof(trace->ev_qualifier_ids.entries[0]));
1564 
1565 	if (trace->ev_qualifier_ids.entries == NULL) {
1566 		fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1567 		       trace->output);
1568 		err = -EINVAL;
1569 		goto out;
1570 	}
1571 
1572 	strlist__for_each_entry(pos, trace->ev_qualifier) {
1573 		const char *sc = pos->s;
1574 		int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1575 
1576 		if (id < 0) {
1577 			id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1578 			if (id >= 0)
1579 				goto matches;
1580 
1581 			if (!printed_invalid_prefix) {
1582 				pr_debug("Skipping unknown syscalls: ");
1583 				printed_invalid_prefix = true;
1584 			} else {
1585 				pr_debug(", ");
1586 			}
1587 
1588 			pr_debug("%s", sc);
1589 			continue;
1590 		}
1591 matches:
1592 		trace->ev_qualifier_ids.entries[nr_used++] = id;
1593 		if (match_next == -1)
1594 			continue;
1595 
1596 		while (1) {
1597 			id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1598 			if (id < 0)
1599 				break;
1600 			if (nr_allocated == nr_used) {
1601 				void *entries;
1602 
1603 				nr_allocated += 8;
1604 				entries = realloc(trace->ev_qualifier_ids.entries,
1605 						  nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1606 				if (entries == NULL) {
1607 					err = -ENOMEM;
1608 					fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1609 					goto out_free;
1610 				}
1611 				trace->ev_qualifier_ids.entries = entries;
1612 			}
1613 			trace->ev_qualifier_ids.entries[nr_used++] = id;
1614 		}
1615 	}
1616 
1617 	trace->ev_qualifier_ids.nr = nr_used;
1618 	qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1619 out:
1620 	if (printed_invalid_prefix)
1621 		pr_debug("\n");
1622 	return err;
1623 out_free:
1624 	zfree(&trace->ev_qualifier_ids.entries);
1625 	trace->ev_qualifier_ids.nr = 0;
1626 	goto out;
1627 }
1628 
1629 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1630 {
1631 	bool in_ev_qualifier;
1632 
1633 	if (trace->ev_qualifier_ids.nr == 0)
1634 		return true;
1635 
1636 	in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1637 				  trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1638 
1639 	if (in_ev_qualifier)
1640 	       return !trace->not_ev_qualifier;
1641 
1642 	return trace->not_ev_qualifier;
1643 }
1644 
1645 /*
1646  * args is to be interpreted as a series of longs but we need to handle
1647  * 8-byte unaligned accesses. args points to raw_data within the event
1648  * and raw_data is guaranteed to be 8-byte unaligned because it is
1649  * preceded by raw_size which is a u32. So we need to copy args to a temp
1650  * variable to read it. Most notably this avoids extended load instructions
1651  * on unaligned addresses
1652  */
1653 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1654 {
1655 	unsigned long val;
1656 	unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1657 
1658 	memcpy(&val, p, sizeof(val));
1659 	return val;
1660 }
1661 
1662 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1663 				      struct syscall_arg *arg)
1664 {
1665 	if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1666 		return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1667 
1668 	return scnprintf(bf, size, "arg%d: ", arg->idx);
1669 }
1670 
1671 /*
1672  * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1673  * as mount 'flags' argument that needs ignoring some magic flag, see comment
1674  * in tools/perf/trace/beauty/mount_flags.c
1675  */
1676 static unsigned long syscall__mask_val(struct syscall *sc, struct syscall_arg *arg, unsigned long val)
1677 {
1678 	if (sc->arg_fmt && sc->arg_fmt[arg->idx].mask_val)
1679 		return sc->arg_fmt[arg->idx].mask_val(arg, val);
1680 
1681 	return val;
1682 }
1683 
1684 static size_t syscall__scnprintf_val(struct syscall *sc, char *bf, size_t size,
1685 				     struct syscall_arg *arg, unsigned long val)
1686 {
1687 	if (sc->arg_fmt && sc->arg_fmt[arg->idx].scnprintf) {
1688 		arg->val = val;
1689 		if (sc->arg_fmt[arg->idx].parm)
1690 			arg->parm = sc->arg_fmt[arg->idx].parm;
1691 		return sc->arg_fmt[arg->idx].scnprintf(bf, size, arg);
1692 	}
1693 	return scnprintf(bf, size, "%ld", val);
1694 }
1695 
1696 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1697 				      unsigned char *args, void *augmented_args, int augmented_args_size,
1698 				      struct trace *trace, struct thread *thread)
1699 {
1700 	size_t printed = 0;
1701 	unsigned long val;
1702 	u8 bit = 1;
1703 	struct syscall_arg arg = {
1704 		.args	= args,
1705 		.augmented = {
1706 			.size = augmented_args_size,
1707 			.args = augmented_args,
1708 		},
1709 		.idx	= 0,
1710 		.mask	= 0,
1711 		.trace  = trace,
1712 		.thread = thread,
1713 		.show_string_prefix = trace->show_string_prefix,
1714 	};
1715 	struct thread_trace *ttrace = thread__priv(thread);
1716 
1717 	/*
1718 	 * Things like fcntl will set this in its 'cmd' formatter to pick the
1719 	 * right formatter for the return value (an fd? file flags?), which is
1720 	 * not needed for syscalls that always return a given type, say an fd.
1721 	 */
1722 	ttrace->ret_scnprintf = NULL;
1723 
1724 	if (sc->args != NULL) {
1725 		struct tep_format_field *field;
1726 
1727 		for (field = sc->args; field;
1728 		     field = field->next, ++arg.idx, bit <<= 1) {
1729 			if (arg.mask & bit)
1730 				continue;
1731 
1732 			val = syscall_arg__val(&arg, arg.idx);
1733 			/*
1734 			 * Some syscall args need some mask, most don't and
1735 			 * return val untouched.
1736 			 */
1737 			val = syscall__mask_val(sc, &arg, val);
1738 
1739 			/*
1740  			 * Suppress this argument if its value is zero and
1741  			 * and we don't have a string associated in an
1742  			 * strarray for it.
1743  			 */
1744 			if (val == 0 &&
1745 			    !trace->show_zeros &&
1746 			    !(sc->arg_fmt &&
1747 			      (sc->arg_fmt[arg.idx].show_zero ||
1748 			       sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
1749 			       sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
1750 			      sc->arg_fmt[arg.idx].parm))
1751 				continue;
1752 
1753 			printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
1754 
1755 			if (trace->show_arg_names)
1756 				printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
1757 
1758 			printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1759 		}
1760 	} else if (IS_ERR(sc->tp_format)) {
1761 		/*
1762 		 * If we managed to read the tracepoint /format file, then we
1763 		 * may end up not having any args, like with gettid(), so only
1764 		 * print the raw args when we didn't manage to read it.
1765 		 */
1766 		while (arg.idx < sc->nr_args) {
1767 			if (arg.mask & bit)
1768 				goto next_arg;
1769 			val = syscall_arg__val(&arg, arg.idx);
1770 			if (printed)
1771 				printed += scnprintf(bf + printed, size - printed, ", ");
1772 			printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
1773 			printed += syscall__scnprintf_val(sc, bf + printed, size - printed, &arg, val);
1774 next_arg:
1775 			++arg.idx;
1776 			bit <<= 1;
1777 		}
1778 	}
1779 
1780 	return printed;
1781 }
1782 
1783 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
1784 				  union perf_event *event,
1785 				  struct perf_sample *sample);
1786 
1787 static struct syscall *trace__syscall_info(struct trace *trace,
1788 					   struct evsel *evsel, int id)
1789 {
1790 	int err = 0;
1791 
1792 	if (id < 0) {
1793 
1794 		/*
1795 		 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1796 		 * before that, leaving at a higher verbosity level till that is
1797 		 * explained. Reproduced with plain ftrace with:
1798 		 *
1799 		 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1800 		 * grep "NR -1 " /t/trace_pipe
1801 		 *
1802 		 * After generating some load on the machine.
1803  		 */
1804 		if (verbose > 1) {
1805 			static u64 n;
1806 			fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1807 				id, perf_evsel__name(evsel), ++n);
1808 		}
1809 		return NULL;
1810 	}
1811 
1812 	err = -EINVAL;
1813 
1814 	if (id > trace->sctbl->syscalls.max_id)
1815 		goto out_cant_read;
1816 
1817 	if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
1818 	    (err = trace__read_syscall_info(trace, id)) != 0)
1819 		goto out_cant_read;
1820 
1821 	if (trace->syscalls.table[id].name == NULL) {
1822 		if (trace->syscalls.table[id].nonexistent)
1823 			return NULL;
1824 		goto out_cant_read;
1825 	}
1826 
1827 	return &trace->syscalls.table[id];
1828 
1829 out_cant_read:
1830 	if (verbose > 0) {
1831 		char sbuf[STRERR_BUFSIZE];
1832 		fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
1833 		if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
1834 			fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1835 		fputs(" information\n", trace->output);
1836 	}
1837 	return NULL;
1838 }
1839 
1840 static void thread__update_stats(struct thread_trace *ttrace,
1841 				 int id, struct perf_sample *sample)
1842 {
1843 	struct int_node *inode;
1844 	struct stats *stats;
1845 	u64 duration = 0;
1846 
1847 	inode = intlist__findnew(ttrace->syscall_stats, id);
1848 	if (inode == NULL)
1849 		return;
1850 
1851 	stats = inode->priv;
1852 	if (stats == NULL) {
1853 		stats = malloc(sizeof(struct stats));
1854 		if (stats == NULL)
1855 			return;
1856 		init_stats(stats);
1857 		inode->priv = stats;
1858 	}
1859 
1860 	if (ttrace->entry_time && sample->time > ttrace->entry_time)
1861 		duration = sample->time - ttrace->entry_time;
1862 
1863 	update_stats(stats, duration);
1864 }
1865 
1866 static int trace__printf_interrupted_entry(struct trace *trace)
1867 {
1868 	struct thread_trace *ttrace;
1869 	size_t printed;
1870 	int len;
1871 
1872 	if (trace->failure_only || trace->current == NULL)
1873 		return 0;
1874 
1875 	ttrace = thread__priv(trace->current);
1876 
1877 	if (!ttrace->entry_pending)
1878 		return 0;
1879 
1880 	printed  = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
1881 	printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
1882 
1883 	if (len < trace->args_alignment - 4)
1884 		printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
1885 
1886 	printed += fprintf(trace->output, " ...\n");
1887 
1888 	ttrace->entry_pending = false;
1889 	++trace->nr_events_printed;
1890 
1891 	return printed;
1892 }
1893 
1894 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
1895 				 struct perf_sample *sample, struct thread *thread)
1896 {
1897 	int printed = 0;
1898 
1899 	if (trace->print_sample) {
1900 		double ts = (double)sample->time / NSEC_PER_MSEC;
1901 
1902 		printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
1903 				   perf_evsel__name(evsel), ts,
1904 				   thread__comm_str(thread),
1905 				   sample->pid, sample->tid, sample->cpu);
1906 	}
1907 
1908 	return printed;
1909 }
1910 
1911 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
1912 {
1913 	void *augmented_args = NULL;
1914 	/*
1915 	 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
1916 	 * and there we get all 6 syscall args plus the tracepoint common fields
1917 	 * that gets calculated at the start and the syscall_nr (another long).
1918 	 * So we check if that is the case and if so don't look after the
1919 	 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
1920 	 * which is fixed.
1921 	 *
1922 	 * We'll revisit this later to pass s->args_size to the BPF augmenter
1923 	 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
1924 	 * copies only what we need for each syscall, like what happens when we
1925 	 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
1926 	 * traffic to just what is needed for each syscall.
1927 	 */
1928 	int args_size = raw_augmented_args_size ?: sc->args_size;
1929 
1930 	*augmented_args_size = sample->raw_size - args_size;
1931 	if (*augmented_args_size > 0)
1932 		augmented_args = sample->raw_data + args_size;
1933 
1934 	return augmented_args;
1935 }
1936 
1937 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
1938 			    union perf_event *event __maybe_unused,
1939 			    struct perf_sample *sample)
1940 {
1941 	char *msg;
1942 	void *args;
1943 	int printed = 0;
1944 	struct thread *thread;
1945 	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
1946 	int augmented_args_size = 0;
1947 	void *augmented_args = NULL;
1948 	struct syscall *sc = trace__syscall_info(trace, evsel, id);
1949 	struct thread_trace *ttrace;
1950 
1951 	if (sc == NULL)
1952 		return -1;
1953 
1954 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
1955 	ttrace = thread__trace(thread, trace->output);
1956 	if (ttrace == NULL)
1957 		goto out_put;
1958 
1959 	trace__fprintf_sample(trace, evsel, sample, thread);
1960 
1961 	args = perf_evsel__sc_tp_ptr(evsel, args, sample);
1962 
1963 	if (ttrace->entry_str == NULL) {
1964 		ttrace->entry_str = malloc(trace__entry_str_size);
1965 		if (!ttrace->entry_str)
1966 			goto out_put;
1967 	}
1968 
1969 	if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
1970 		trace__printf_interrupted_entry(trace);
1971 	/*
1972 	 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
1973 	 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
1974 	 * this breaks syscall__augmented_args() check for augmented args, as we calculate
1975 	 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
1976 	 * so when handling, say the openat syscall, we end up getting 6 args for the
1977 	 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
1978 	 * thinking that the extra 2 u64 args are the augmented filename, so just check
1979 	 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
1980 	 */
1981 	if (evsel != trace->syscalls.events.sys_enter)
1982 		augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
1983 	ttrace->entry_time = sample->time;
1984 	msg = ttrace->entry_str;
1985 	printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
1986 
1987 	printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
1988 					   args, augmented_args, augmented_args_size, trace, thread);
1989 
1990 	if (sc->is_exit) {
1991 		if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
1992 			int alignment = 0;
1993 
1994 			trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
1995 			printed = fprintf(trace->output, "%s)", ttrace->entry_str);
1996 			if (trace->args_alignment > printed)
1997 				alignment = trace->args_alignment - printed;
1998 			fprintf(trace->output, "%*s= ?\n", alignment, " ");
1999 		}
2000 	} else {
2001 		ttrace->entry_pending = true;
2002 		/* See trace__vfs_getname & trace__sys_exit */
2003 		ttrace->filename.pending_open = false;
2004 	}
2005 
2006 	if (trace->current != thread) {
2007 		thread__put(trace->current);
2008 		trace->current = thread__get(thread);
2009 	}
2010 	err = 0;
2011 out_put:
2012 	thread__put(thread);
2013 	return err;
2014 }
2015 
2016 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2017 				    struct perf_sample *sample)
2018 {
2019 	struct thread_trace *ttrace;
2020 	struct thread *thread;
2021 	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2022 	struct syscall *sc = trace__syscall_info(trace, evsel, id);
2023 	char msg[1024];
2024 	void *args, *augmented_args = NULL;
2025 	int augmented_args_size;
2026 
2027 	if (sc == NULL)
2028 		return -1;
2029 
2030 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2031 	ttrace = thread__trace(thread, trace->output);
2032 	/*
2033 	 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2034 	 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2035 	 */
2036 	if (ttrace == NULL)
2037 		goto out_put;
2038 
2039 	args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2040 	augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2041 	syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2042 	fprintf(trace->output, "%s", msg);
2043 	err = 0;
2044 out_put:
2045 	thread__put(thread);
2046 	return err;
2047 }
2048 
2049 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2050 				    struct perf_sample *sample,
2051 				    struct callchain_cursor *cursor)
2052 {
2053 	struct addr_location al;
2054 	int max_stack = evsel->core.attr.sample_max_stack ?
2055 			evsel->core.attr.sample_max_stack :
2056 			trace->max_stack;
2057 	int err;
2058 
2059 	if (machine__resolve(trace->host, &al, sample) < 0)
2060 		return -1;
2061 
2062 	err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2063 	addr_location__put(&al);
2064 	return err;
2065 }
2066 
2067 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2068 {
2069 	/* TODO: user-configurable print_opts */
2070 	const unsigned int print_opts = EVSEL__PRINT_SYM |
2071 				        EVSEL__PRINT_DSO |
2072 				        EVSEL__PRINT_UNKNOWN_AS_ADDR;
2073 
2074 	return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, trace->output);
2075 }
2076 
2077 static const char *errno_to_name(struct evsel *evsel, int err)
2078 {
2079 	struct perf_env *env = perf_evsel__env(evsel);
2080 	const char *arch_name = perf_env__arch(env);
2081 
2082 	return arch_syscalls__strerrno(arch_name, err);
2083 }
2084 
2085 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2086 			   union perf_event *event __maybe_unused,
2087 			   struct perf_sample *sample)
2088 {
2089 	long ret;
2090 	u64 duration = 0;
2091 	bool duration_calculated = false;
2092 	struct thread *thread;
2093 	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2094 	int alignment = trace->args_alignment;
2095 	struct syscall *sc = trace__syscall_info(trace, evsel, id);
2096 	struct thread_trace *ttrace;
2097 
2098 	if (sc == NULL)
2099 		return -1;
2100 
2101 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2102 	ttrace = thread__trace(thread, trace->output);
2103 	if (ttrace == NULL)
2104 		goto out_put;
2105 
2106 	trace__fprintf_sample(trace, evsel, sample, thread);
2107 
2108 	if (trace->summary)
2109 		thread__update_stats(ttrace, id, sample);
2110 
2111 	ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2112 
2113 	if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2114 		trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2115 		ttrace->filename.pending_open = false;
2116 		++trace->stats.vfs_getname;
2117 	}
2118 
2119 	if (ttrace->entry_time) {
2120 		duration = sample->time - ttrace->entry_time;
2121 		if (trace__filter_duration(trace, duration))
2122 			goto out;
2123 		duration_calculated = true;
2124 	} else if (trace->duration_filter)
2125 		goto out;
2126 
2127 	if (sample->callchain) {
2128 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2129 		if (callchain_ret == 0) {
2130 			if (callchain_cursor.nr < trace->min_stack)
2131 				goto out;
2132 			callchain_ret = 1;
2133 		}
2134 	}
2135 
2136 	if (trace->summary_only || (ret >= 0 && trace->failure_only))
2137 		goto out;
2138 
2139 	trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2140 
2141 	if (ttrace->entry_pending) {
2142 		printed = fprintf(trace->output, "%s", ttrace->entry_str);
2143 	} else {
2144 		printed += fprintf(trace->output, " ... [");
2145 		color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2146 		printed += 9;
2147 		printed += fprintf(trace->output, "]: %s()", sc->name);
2148 	}
2149 
2150 	printed++; /* the closing ')' */
2151 
2152 	if (alignment > printed)
2153 		alignment -= printed;
2154 	else
2155 		alignment = 0;
2156 
2157 	fprintf(trace->output, ")%*s= ", alignment, " ");
2158 
2159 	if (sc->fmt == NULL) {
2160 		if (ret < 0)
2161 			goto errno_print;
2162 signed_print:
2163 		fprintf(trace->output, "%ld", ret);
2164 	} else if (ret < 0) {
2165 errno_print: {
2166 		char bf[STRERR_BUFSIZE];
2167 		const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2168 			   *e = errno_to_name(evsel, -ret);
2169 
2170 		fprintf(trace->output, "-1 %s (%s)", e, emsg);
2171 	}
2172 	} else if (ret == 0 && sc->fmt->timeout)
2173 		fprintf(trace->output, "0 (Timeout)");
2174 	else if (ttrace->ret_scnprintf) {
2175 		char bf[1024];
2176 		struct syscall_arg arg = {
2177 			.val	= ret,
2178 			.thread	= thread,
2179 			.trace	= trace,
2180 		};
2181 		ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2182 		ttrace->ret_scnprintf = NULL;
2183 		fprintf(trace->output, "%s", bf);
2184 	} else if (sc->fmt->hexret)
2185 		fprintf(trace->output, "%#lx", ret);
2186 	else if (sc->fmt->errpid) {
2187 		struct thread *child = machine__find_thread(trace->host, ret, ret);
2188 
2189 		if (child != NULL) {
2190 			fprintf(trace->output, "%ld", ret);
2191 			if (child->comm_set)
2192 				fprintf(trace->output, " (%s)", thread__comm_str(child));
2193 			thread__put(child);
2194 		}
2195 	} else
2196 		goto signed_print;
2197 
2198 	fputc('\n', trace->output);
2199 
2200 	/*
2201 	 * We only consider an 'event' for the sake of --max-events a non-filtered
2202 	 * sys_enter + sys_exit and other tracepoint events.
2203 	 */
2204 	if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2205 		interrupted = true;
2206 
2207 	if (callchain_ret > 0)
2208 		trace__fprintf_callchain(trace, sample);
2209 	else if (callchain_ret < 0)
2210 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2211 out:
2212 	ttrace->entry_pending = false;
2213 	err = 0;
2214 out_put:
2215 	thread__put(thread);
2216 	return err;
2217 }
2218 
2219 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2220 			      union perf_event *event __maybe_unused,
2221 			      struct perf_sample *sample)
2222 {
2223 	struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2224 	struct thread_trace *ttrace;
2225 	size_t filename_len, entry_str_len, to_move;
2226 	ssize_t remaining_space;
2227 	char *pos;
2228 	const char *filename = perf_evsel__rawptr(evsel, sample, "pathname");
2229 
2230 	if (!thread)
2231 		goto out;
2232 
2233 	ttrace = thread__priv(thread);
2234 	if (!ttrace)
2235 		goto out_put;
2236 
2237 	filename_len = strlen(filename);
2238 	if (filename_len == 0)
2239 		goto out_put;
2240 
2241 	if (ttrace->filename.namelen < filename_len) {
2242 		char *f = realloc(ttrace->filename.name, filename_len + 1);
2243 
2244 		if (f == NULL)
2245 			goto out_put;
2246 
2247 		ttrace->filename.namelen = filename_len;
2248 		ttrace->filename.name = f;
2249 	}
2250 
2251 	strcpy(ttrace->filename.name, filename);
2252 	ttrace->filename.pending_open = true;
2253 
2254 	if (!ttrace->filename.ptr)
2255 		goto out_put;
2256 
2257 	entry_str_len = strlen(ttrace->entry_str);
2258 	remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2259 	if (remaining_space <= 0)
2260 		goto out_put;
2261 
2262 	if (filename_len > (size_t)remaining_space) {
2263 		filename += filename_len - remaining_space;
2264 		filename_len = remaining_space;
2265 	}
2266 
2267 	to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2268 	pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2269 	memmove(pos + filename_len, pos, to_move);
2270 	memcpy(pos, filename, filename_len);
2271 
2272 	ttrace->filename.ptr = 0;
2273 	ttrace->filename.entry_str_pos = 0;
2274 out_put:
2275 	thread__put(thread);
2276 out:
2277 	return 0;
2278 }
2279 
2280 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2281 				     union perf_event *event __maybe_unused,
2282 				     struct perf_sample *sample)
2283 {
2284         u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
2285 	double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2286 	struct thread *thread = machine__findnew_thread(trace->host,
2287 							sample->pid,
2288 							sample->tid);
2289 	struct thread_trace *ttrace = thread__trace(thread, trace->output);
2290 
2291 	if (ttrace == NULL)
2292 		goto out_dump;
2293 
2294 	ttrace->runtime_ms += runtime_ms;
2295 	trace->runtime_ms += runtime_ms;
2296 out_put:
2297 	thread__put(thread);
2298 	return 0;
2299 
2300 out_dump:
2301 	fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2302 	       evsel->name,
2303 	       perf_evsel__strval(evsel, sample, "comm"),
2304 	       (pid_t)perf_evsel__intval(evsel, sample, "pid"),
2305 	       runtime,
2306 	       perf_evsel__intval(evsel, sample, "vruntime"));
2307 	goto out_put;
2308 }
2309 
2310 static int bpf_output__printer(enum binary_printer_ops op,
2311 			       unsigned int val, void *extra __maybe_unused, FILE *fp)
2312 {
2313 	unsigned char ch = (unsigned char)val;
2314 
2315 	switch (op) {
2316 	case BINARY_PRINT_CHAR_DATA:
2317 		return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2318 	case BINARY_PRINT_DATA_BEGIN:
2319 	case BINARY_PRINT_LINE_BEGIN:
2320 	case BINARY_PRINT_ADDR:
2321 	case BINARY_PRINT_NUM_DATA:
2322 	case BINARY_PRINT_NUM_PAD:
2323 	case BINARY_PRINT_SEP:
2324 	case BINARY_PRINT_CHAR_PAD:
2325 	case BINARY_PRINT_LINE_END:
2326 	case BINARY_PRINT_DATA_END:
2327 	default:
2328 		break;
2329 	}
2330 
2331 	return 0;
2332 }
2333 
2334 static void bpf_output__fprintf(struct trace *trace,
2335 				struct perf_sample *sample)
2336 {
2337 	binary__fprintf(sample->raw_data, sample->raw_size, 8,
2338 			bpf_output__printer, NULL, trace->output);
2339 	++trace->nr_events_printed;
2340 }
2341 
2342 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2343 				union perf_event *event __maybe_unused,
2344 				struct perf_sample *sample)
2345 {
2346 	struct thread *thread;
2347 	int callchain_ret = 0;
2348 	/*
2349 	 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2350 	 * this event's max_events having been hit and this is an entry coming
2351 	 * from the ring buffer that we should discard, since the max events
2352 	 * have already been considered/printed.
2353 	 */
2354 	if (evsel->disabled)
2355 		return 0;
2356 
2357 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2358 
2359 	if (sample->callchain) {
2360 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2361 		if (callchain_ret == 0) {
2362 			if (callchain_cursor.nr < trace->min_stack)
2363 				goto out;
2364 			callchain_ret = 1;
2365 		}
2366 	}
2367 
2368 	trace__printf_interrupted_entry(trace);
2369 	trace__fprintf_tstamp(trace, sample->time, trace->output);
2370 
2371 	if (trace->trace_syscalls && trace->show_duration)
2372 		fprintf(trace->output, "(         ): ");
2373 
2374 	if (thread)
2375 		trace__fprintf_comm_tid(trace, thread, trace->output);
2376 
2377 	if (evsel == trace->syscalls.events.augmented) {
2378 		int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2379 		struct syscall *sc = trace__syscall_info(trace, evsel, id);
2380 
2381 		if (sc) {
2382 			fprintf(trace->output, "%s(", sc->name);
2383 			trace__fprintf_sys_enter(trace, evsel, sample);
2384 			fputc(')', trace->output);
2385 			goto newline;
2386 		}
2387 
2388 		/*
2389 		 * XXX: Not having the associated syscall info or not finding/adding
2390 		 * 	the thread should never happen, but if it does...
2391 		 * 	fall thru and print it as a bpf_output event.
2392 		 */
2393 	}
2394 
2395 	fprintf(trace->output, "%s:", evsel->name);
2396 
2397 	if (perf_evsel__is_bpf_output(evsel)) {
2398 		bpf_output__fprintf(trace, sample);
2399 	} else if (evsel->tp_format) {
2400 		if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2401 		    trace__fprintf_sys_enter(trace, evsel, sample)) {
2402 			event_format__fprintf(evsel->tp_format, sample->cpu,
2403 					      sample->raw_data, sample->raw_size,
2404 					      trace->output);
2405 			++trace->nr_events_printed;
2406 
2407 			if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2408 				evsel__disable(evsel);
2409 				evsel__close(evsel);
2410 			}
2411 		}
2412 	}
2413 
2414 newline:
2415 	fprintf(trace->output, "\n");
2416 
2417 	if (callchain_ret > 0)
2418 		trace__fprintf_callchain(trace, sample);
2419 	else if (callchain_ret < 0)
2420 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2421 out:
2422 	thread__put(thread);
2423 	return 0;
2424 }
2425 
2426 static void print_location(FILE *f, struct perf_sample *sample,
2427 			   struct addr_location *al,
2428 			   bool print_dso, bool print_sym)
2429 {
2430 
2431 	if ((verbose > 0 || print_dso) && al->map)
2432 		fprintf(f, "%s@", al->map->dso->long_name);
2433 
2434 	if ((verbose > 0 || print_sym) && al->sym)
2435 		fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2436 			al->addr - al->sym->start);
2437 	else if (al->map)
2438 		fprintf(f, "0x%" PRIx64, al->addr);
2439 	else
2440 		fprintf(f, "0x%" PRIx64, sample->addr);
2441 }
2442 
2443 static int trace__pgfault(struct trace *trace,
2444 			  struct evsel *evsel,
2445 			  union perf_event *event __maybe_unused,
2446 			  struct perf_sample *sample)
2447 {
2448 	struct thread *thread;
2449 	struct addr_location al;
2450 	char map_type = 'd';
2451 	struct thread_trace *ttrace;
2452 	int err = -1;
2453 	int callchain_ret = 0;
2454 
2455 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2456 
2457 	if (sample->callchain) {
2458 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2459 		if (callchain_ret == 0) {
2460 			if (callchain_cursor.nr < trace->min_stack)
2461 				goto out_put;
2462 			callchain_ret = 1;
2463 		}
2464 	}
2465 
2466 	ttrace = thread__trace(thread, trace->output);
2467 	if (ttrace == NULL)
2468 		goto out_put;
2469 
2470 	if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2471 		ttrace->pfmaj++;
2472 	else
2473 		ttrace->pfmin++;
2474 
2475 	if (trace->summary_only)
2476 		goto out;
2477 
2478 	thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2479 
2480 	trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2481 
2482 	fprintf(trace->output, "%sfault [",
2483 		evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2484 		"maj" : "min");
2485 
2486 	print_location(trace->output, sample, &al, false, true);
2487 
2488 	fprintf(trace->output, "] => ");
2489 
2490 	thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2491 
2492 	if (!al.map) {
2493 		thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2494 
2495 		if (al.map)
2496 			map_type = 'x';
2497 		else
2498 			map_type = '?';
2499 	}
2500 
2501 	print_location(trace->output, sample, &al, true, false);
2502 
2503 	fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2504 
2505 	if (callchain_ret > 0)
2506 		trace__fprintf_callchain(trace, sample);
2507 	else if (callchain_ret < 0)
2508 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2509 
2510 	++trace->nr_events_printed;
2511 out:
2512 	err = 0;
2513 out_put:
2514 	thread__put(thread);
2515 	return err;
2516 }
2517 
2518 static void trace__set_base_time(struct trace *trace,
2519 				 struct evsel *evsel,
2520 				 struct perf_sample *sample)
2521 {
2522 	/*
2523 	 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2524 	 * and don't use sample->time unconditionally, we may end up having
2525 	 * some other event in the future without PERF_SAMPLE_TIME for good
2526 	 * reason, i.e. we may not be interested in its timestamps, just in
2527 	 * it taking place, picking some piece of information when it
2528 	 * appears in our event stream (vfs_getname comes to mind).
2529 	 */
2530 	if (trace->base_time == 0 && !trace->full_time &&
2531 	    (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2532 		trace->base_time = sample->time;
2533 }
2534 
2535 static int trace__process_sample(struct perf_tool *tool,
2536 				 union perf_event *event,
2537 				 struct perf_sample *sample,
2538 				 struct evsel *evsel,
2539 				 struct machine *machine __maybe_unused)
2540 {
2541 	struct trace *trace = container_of(tool, struct trace, tool);
2542 	struct thread *thread;
2543 	int err = 0;
2544 
2545 	tracepoint_handler handler = evsel->handler;
2546 
2547 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2548 	if (thread && thread__is_filtered(thread))
2549 		goto out;
2550 
2551 	trace__set_base_time(trace, evsel, sample);
2552 
2553 	if (handler) {
2554 		++trace->nr_events;
2555 		handler(trace, evsel, event, sample);
2556 	}
2557 out:
2558 	thread__put(thread);
2559 	return err;
2560 }
2561 
2562 static int trace__record(struct trace *trace, int argc, const char **argv)
2563 {
2564 	unsigned int rec_argc, i, j;
2565 	const char **rec_argv;
2566 	const char * const record_args[] = {
2567 		"record",
2568 		"-R",
2569 		"-m", "1024",
2570 		"-c", "1",
2571 	};
2572 
2573 	const char * const sc_args[] = { "-e", };
2574 	unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2575 	const char * const majpf_args[] = { "-e", "major-faults" };
2576 	unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2577 	const char * const minpf_args[] = { "-e", "minor-faults" };
2578 	unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2579 
2580 	/* +1 is for the event string below */
2581 	rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2582 		majpf_args_nr + minpf_args_nr + argc;
2583 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
2584 
2585 	if (rec_argv == NULL)
2586 		return -ENOMEM;
2587 
2588 	j = 0;
2589 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
2590 		rec_argv[j++] = record_args[i];
2591 
2592 	if (trace->trace_syscalls) {
2593 		for (i = 0; i < sc_args_nr; i++)
2594 			rec_argv[j++] = sc_args[i];
2595 
2596 		/* event string may be different for older kernels - e.g., RHEL6 */
2597 		if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2598 			rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2599 		else if (is_valid_tracepoint("syscalls:sys_enter"))
2600 			rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2601 		else {
2602 			pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2603 			free(rec_argv);
2604 			return -1;
2605 		}
2606 	}
2607 
2608 	if (trace->trace_pgfaults & TRACE_PFMAJ)
2609 		for (i = 0; i < majpf_args_nr; i++)
2610 			rec_argv[j++] = majpf_args[i];
2611 
2612 	if (trace->trace_pgfaults & TRACE_PFMIN)
2613 		for (i = 0; i < minpf_args_nr; i++)
2614 			rec_argv[j++] = minpf_args[i];
2615 
2616 	for (i = 0; i < (unsigned int)argc; i++)
2617 		rec_argv[j++] = argv[i];
2618 
2619 	return cmd_record(j, rec_argv);
2620 }
2621 
2622 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2623 
2624 static bool evlist__add_vfs_getname(struct evlist *evlist)
2625 {
2626 	bool found = false;
2627 	struct evsel *evsel, *tmp;
2628 	struct parse_events_error err = { .idx = 0, };
2629 	int ret = parse_events(evlist, "probe:vfs_getname*", &err);
2630 
2631 	if (ret)
2632 		return false;
2633 
2634 	evlist__for_each_entry_safe(evlist, evsel, tmp) {
2635 		if (!strstarts(perf_evsel__name(evsel), "probe:vfs_getname"))
2636 			continue;
2637 
2638 		if (perf_evsel__field(evsel, "pathname")) {
2639 			evsel->handler = trace__vfs_getname;
2640 			found = true;
2641 			continue;
2642 		}
2643 
2644 		list_del_init(&evsel->core.node);
2645 		evsel->evlist = NULL;
2646 		evsel__delete(evsel);
2647 	}
2648 
2649 	return found;
2650 }
2651 
2652 static struct evsel *perf_evsel__new_pgfault(u64 config)
2653 {
2654 	struct evsel *evsel;
2655 	struct perf_event_attr attr = {
2656 		.type = PERF_TYPE_SOFTWARE,
2657 		.mmap_data = 1,
2658 	};
2659 
2660 	attr.config = config;
2661 	attr.sample_period = 1;
2662 
2663 	event_attr_init(&attr);
2664 
2665 	evsel = evsel__new(&attr);
2666 	if (evsel)
2667 		evsel->handler = trace__pgfault;
2668 
2669 	return evsel;
2670 }
2671 
2672 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
2673 {
2674 	const u32 type = event->header.type;
2675 	struct evsel *evsel;
2676 
2677 	if (type != PERF_RECORD_SAMPLE) {
2678 		trace__process_event(trace, trace->host, event, sample);
2679 		return;
2680 	}
2681 
2682 	evsel = perf_evlist__id2evsel(trace->evlist, sample->id);
2683 	if (evsel == NULL) {
2684 		fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
2685 		return;
2686 	}
2687 
2688 	if (evswitch__discard(&trace->evswitch, evsel))
2689 		return;
2690 
2691 	trace__set_base_time(trace, evsel, sample);
2692 
2693 	if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
2694 	    sample->raw_data == NULL) {
2695 		fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2696 		       perf_evsel__name(evsel), sample->tid,
2697 		       sample->cpu, sample->raw_size);
2698 	} else {
2699 		tracepoint_handler handler = evsel->handler;
2700 		handler(trace, evsel, event, sample);
2701 	}
2702 
2703 	if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
2704 		interrupted = true;
2705 }
2706 
2707 static int trace__add_syscall_newtp(struct trace *trace)
2708 {
2709 	int ret = -1;
2710 	struct evlist *evlist = trace->evlist;
2711 	struct evsel *sys_enter, *sys_exit;
2712 
2713 	sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
2714 	if (sys_enter == NULL)
2715 		goto out;
2716 
2717 	if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
2718 		goto out_delete_sys_enter;
2719 
2720 	sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
2721 	if (sys_exit == NULL)
2722 		goto out_delete_sys_enter;
2723 
2724 	if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
2725 		goto out_delete_sys_exit;
2726 
2727 	perf_evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
2728 	perf_evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
2729 
2730 	evlist__add(evlist, sys_enter);
2731 	evlist__add(evlist, sys_exit);
2732 
2733 	if (callchain_param.enabled && !trace->kernel_syscallchains) {
2734 		/*
2735 		 * We're interested only in the user space callchain
2736 		 * leading to the syscall, allow overriding that for
2737 		 * debugging reasons using --kernel_syscall_callchains
2738 		 */
2739 		sys_exit->core.attr.exclude_callchain_kernel = 1;
2740 	}
2741 
2742 	trace->syscalls.events.sys_enter = sys_enter;
2743 	trace->syscalls.events.sys_exit  = sys_exit;
2744 
2745 	ret = 0;
2746 out:
2747 	return ret;
2748 
2749 out_delete_sys_exit:
2750 	evsel__delete_priv(sys_exit);
2751 out_delete_sys_enter:
2752 	evsel__delete_priv(sys_enter);
2753 	goto out;
2754 }
2755 
2756 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
2757 {
2758 	int err = -1;
2759 	struct evsel *sys_exit;
2760 	char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
2761 						trace->ev_qualifier_ids.nr,
2762 						trace->ev_qualifier_ids.entries);
2763 
2764 	if (filter == NULL)
2765 		goto out_enomem;
2766 
2767 	if (!perf_evsel__append_tp_filter(trace->syscalls.events.sys_enter,
2768 					  filter)) {
2769 		sys_exit = trace->syscalls.events.sys_exit;
2770 		err = perf_evsel__append_tp_filter(sys_exit, filter);
2771 	}
2772 
2773 	free(filter);
2774 out:
2775 	return err;
2776 out_enomem:
2777 	errno = ENOMEM;
2778 	goto out;
2779 }
2780 
2781 #ifdef HAVE_LIBBPF_SUPPORT
2782 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
2783 {
2784 	if (trace->bpf_obj == NULL)
2785 		return NULL;
2786 
2787 	return bpf_object__find_program_by_title(trace->bpf_obj, name);
2788 }
2789 
2790 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
2791 							const char *prog_name, const char *type)
2792 {
2793 	struct bpf_program *prog;
2794 
2795 	if (prog_name == NULL) {
2796 		char default_prog_name[256];
2797 		scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
2798 		prog = trace__find_bpf_program_by_title(trace, default_prog_name);
2799 		if (prog != NULL)
2800 			goto out_found;
2801 		if (sc->fmt && sc->fmt->alias) {
2802 			scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
2803 			prog = trace__find_bpf_program_by_title(trace, default_prog_name);
2804 			if (prog != NULL)
2805 				goto out_found;
2806 		}
2807 		goto out_unaugmented;
2808 	}
2809 
2810 	prog = trace__find_bpf_program_by_title(trace, prog_name);
2811 
2812 	if (prog != NULL) {
2813 out_found:
2814 		return prog;
2815 	}
2816 
2817 	pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
2818 		 prog_name, type, sc->name);
2819 out_unaugmented:
2820 	return trace->syscalls.unaugmented_prog;
2821 }
2822 
2823 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
2824 {
2825 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2826 
2827 	if (sc == NULL)
2828 		return;
2829 
2830 	sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
2831 	sc->bpf_prog.sys_exit  = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit  : NULL,  "exit");
2832 }
2833 
2834 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
2835 {
2836 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2837 	return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
2838 }
2839 
2840 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
2841 {
2842 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2843 	return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
2844 }
2845 
2846 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
2847 {
2848 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2849 	int arg = 0;
2850 
2851 	if (sc == NULL)
2852 		goto out;
2853 
2854 	for (; arg < sc->nr_args; ++arg) {
2855 		entry->string_args_len[arg] = 0;
2856 		if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
2857 			/* Should be set like strace -s strsize */
2858 			entry->string_args_len[arg] = PATH_MAX;
2859 		}
2860 	}
2861 out:
2862 	for (; arg < 6; ++arg)
2863 		entry->string_args_len[arg] = 0;
2864 }
2865 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
2866 {
2867 	int fd = bpf_map__fd(trace->syscalls.map);
2868 	struct bpf_map_syscall_entry value = {
2869 		.enabled = !trace->not_ev_qualifier,
2870 	};
2871 	int err = 0;
2872 	size_t i;
2873 
2874 	for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
2875 		int key = trace->ev_qualifier_ids.entries[i];
2876 
2877 		if (value.enabled) {
2878 			trace__init_bpf_map_syscall_args(trace, key, &value);
2879 			trace__init_syscall_bpf_progs(trace, key);
2880 		}
2881 
2882 		err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
2883 		if (err)
2884 			break;
2885 	}
2886 
2887 	return err;
2888 }
2889 
2890 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
2891 {
2892 	int fd = bpf_map__fd(trace->syscalls.map);
2893 	struct bpf_map_syscall_entry value = {
2894 		.enabled = enabled,
2895 	};
2896 	int err = 0, key;
2897 
2898 	for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
2899 		if (enabled)
2900 			trace__init_bpf_map_syscall_args(trace, key, &value);
2901 
2902 		err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
2903 		if (err)
2904 			break;
2905 	}
2906 
2907 	return err;
2908 }
2909 
2910 static int trace__init_syscalls_bpf_map(struct trace *trace)
2911 {
2912 	bool enabled = true;
2913 
2914 	if (trace->ev_qualifier_ids.nr)
2915 		enabled = trace->not_ev_qualifier;
2916 
2917 	return __trace__init_syscalls_bpf_map(trace, enabled);
2918 }
2919 
2920 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
2921 {
2922 	struct tep_format_field *field, *candidate_field;
2923 	int id;
2924 
2925 	/*
2926 	 * We're only interested in syscalls that have a pointer:
2927 	 */
2928 	for (field = sc->args; field; field = field->next) {
2929 		if (field->flags & TEP_FIELD_IS_POINTER)
2930 			goto try_to_find_pair;
2931 	}
2932 
2933 	return NULL;
2934 
2935 try_to_find_pair:
2936 	for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
2937 		struct syscall *pair = trace__syscall_info(trace, NULL, id);
2938 		struct bpf_program *pair_prog;
2939 		bool is_candidate = false;
2940 
2941 		if (pair == NULL || pair == sc ||
2942 		    pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
2943 			continue;
2944 
2945 		for (field = sc->args, candidate_field = pair->args;
2946 		     field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
2947 			bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
2948 			     candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
2949 
2950 			if (is_pointer) {
2951 			       if (!candidate_is_pointer) {
2952 					// The candidate just doesn't copies our pointer arg, might copy other pointers we want.
2953 					continue;
2954 			       }
2955 			} else {
2956 				if (candidate_is_pointer) {
2957 					// The candidate might copy a pointer we don't have, skip it.
2958 					goto next_candidate;
2959 				}
2960 				continue;
2961 			}
2962 
2963 			if (strcmp(field->type, candidate_field->type))
2964 				goto next_candidate;
2965 
2966 			is_candidate = true;
2967 		}
2968 
2969 		if (!is_candidate)
2970 			goto next_candidate;
2971 
2972 		/*
2973 		 * Check if the tentative pair syscall augmenter has more pointers, if it has,
2974 		 * then it may be collecting that and we then can't use it, as it would collect
2975 		 * more than what is common to the two syscalls.
2976 		 */
2977 		if (candidate_field) {
2978 			for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
2979 				if (candidate_field->flags & TEP_FIELD_IS_POINTER)
2980 					goto next_candidate;
2981 		}
2982 
2983 		pair_prog = pair->bpf_prog.sys_enter;
2984 		/*
2985 		 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
2986 		 * have been searched for, so search it here and if it returns the
2987 		 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
2988 		 * program for a filtered syscall on a non-filtered one.
2989 		 *
2990 		 * For instance, we have "!syscalls:sys_enter_renameat" and that is
2991 		 * useful for "renameat2".
2992 		 */
2993 		if (pair_prog == NULL) {
2994 			pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
2995 			if (pair_prog == trace->syscalls.unaugmented_prog)
2996 				goto next_candidate;
2997 		}
2998 
2999 		pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3000 		return pair_prog;
3001 	next_candidate:
3002 		continue;
3003 	}
3004 
3005 	return NULL;
3006 }
3007 
3008 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3009 {
3010 	int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3011 	    map_exit_fd  = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3012 	int err = 0, key;
3013 
3014 	for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3015 		int prog_fd;
3016 
3017 		if (!trace__syscall_enabled(trace, key))
3018 			continue;
3019 
3020 		trace__init_syscall_bpf_progs(trace, key);
3021 
3022 		// It'll get at least the "!raw_syscalls:unaugmented"
3023 		prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3024 		err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3025 		if (err)
3026 			break;
3027 		prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3028 		err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3029 		if (err)
3030 			break;
3031 	}
3032 
3033 	/*
3034 	 * Now lets do a second pass looking for enabled syscalls without
3035 	 * an augmenter that have a signature that is a superset of another
3036 	 * syscall with an augmenter so that we can auto-reuse it.
3037 	 *
3038 	 * I.e. if we have an augmenter for the "open" syscall that has
3039 	 * this signature:
3040 	 *
3041 	 *   int open(const char *pathname, int flags, mode_t mode);
3042 	 *
3043 	 * I.e. that will collect just the first string argument, then we
3044 	 * can reuse it for the 'creat' syscall, that has this signature:
3045 	 *
3046 	 *   int creat(const char *pathname, mode_t mode);
3047 	 *
3048 	 * and for:
3049 	 *
3050 	 *   int stat(const char *pathname, struct stat *statbuf);
3051 	 *   int lstat(const char *pathname, struct stat *statbuf);
3052 	 *
3053 	 * Because the 'open' augmenter will collect the first arg as a string,
3054 	 * and leave alone all the other args, which already helps with
3055 	 * beautifying 'stat' and 'lstat''s pathname arg.
3056 	 *
3057 	 * Then, in time, when 'stat' gets an augmenter that collects both
3058 	 * first and second arg (this one on the raw_syscalls:sys_exit prog
3059 	 * array tail call, then that one will be used.
3060 	 */
3061 	for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3062 		struct syscall *sc = trace__syscall_info(trace, NULL, key);
3063 		struct bpf_program *pair_prog;
3064 		int prog_fd;
3065 
3066 		if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3067 			continue;
3068 
3069 		/*
3070 		 * For now we're just reusing the sys_enter prog, and if it
3071 		 * already has an augmenter, we don't need to find one.
3072 		 */
3073 		if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3074 			continue;
3075 
3076 		/*
3077 		 * Look at all the other syscalls for one that has a signature
3078 		 * that is close enough that we can share:
3079 		 */
3080 		pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3081 		if (pair_prog == NULL)
3082 			continue;
3083 
3084 		sc->bpf_prog.sys_enter = pair_prog;
3085 
3086 		/*
3087 		 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3088 		 * with the fd for the program we're reusing:
3089 		 */
3090 		prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3091 		err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3092 		if (err)
3093 			break;
3094 	}
3095 
3096 
3097 	return err;
3098 }
3099 #else
3100 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3101 {
3102 	return 0;
3103 }
3104 
3105 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3106 {
3107 	return 0;
3108 }
3109 
3110 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3111 							    const char *name __maybe_unused)
3112 {
3113 	return NULL;
3114 }
3115 
3116 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3117 {
3118 	return 0;
3119 }
3120 #endif // HAVE_LIBBPF_SUPPORT
3121 
3122 static int trace__set_ev_qualifier_filter(struct trace *trace)
3123 {
3124 	if (trace->syscalls.map)
3125 		return trace__set_ev_qualifier_bpf_filter(trace);
3126 	if (trace->syscalls.events.sys_enter)
3127 		return trace__set_ev_qualifier_tp_filter(trace);
3128 	return 0;
3129 }
3130 
3131 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3132 				    size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3133 {
3134 	int err = 0;
3135 #ifdef HAVE_LIBBPF_SUPPORT
3136 	bool value = true;
3137 	int map_fd = bpf_map__fd(map);
3138 	size_t i;
3139 
3140 	for (i = 0; i < npids; ++i) {
3141 		err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3142 		if (err)
3143 			break;
3144 	}
3145 #endif
3146 	return err;
3147 }
3148 
3149 static int trace__set_filter_loop_pids(struct trace *trace)
3150 {
3151 	unsigned int nr = 1, err;
3152 	pid_t pids[32] = {
3153 		getpid(),
3154 	};
3155 	struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3156 
3157 	while (thread && nr < ARRAY_SIZE(pids)) {
3158 		struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3159 
3160 		if (parent == NULL)
3161 			break;
3162 
3163 		if (!strcmp(thread__comm_str(parent), "sshd") ||
3164 		    strstarts(thread__comm_str(parent), "gnome-terminal")) {
3165 			pids[nr++] = parent->tid;
3166 			break;
3167 		}
3168 		thread = parent;
3169 	}
3170 
3171 	err = perf_evlist__set_tp_filter_pids(trace->evlist, nr, pids);
3172 	if (!err && trace->filter_pids.map)
3173 		err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3174 
3175 	return err;
3176 }
3177 
3178 static int trace__set_filter_pids(struct trace *trace)
3179 {
3180 	int err = 0;
3181 	/*
3182 	 * Better not use !target__has_task() here because we need to cover the
3183 	 * case where no threads were specified in the command line, but a
3184 	 * workload was, and in that case we will fill in the thread_map when
3185 	 * we fork the workload in perf_evlist__prepare_workload.
3186 	 */
3187 	if (trace->filter_pids.nr > 0) {
3188 		err = perf_evlist__set_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3189 						      trace->filter_pids.entries);
3190 		if (!err && trace->filter_pids.map) {
3191 			err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3192 						       trace->filter_pids.entries);
3193 		}
3194 	} else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3195 		err = trace__set_filter_loop_pids(trace);
3196 	}
3197 
3198 	return err;
3199 }
3200 
3201 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3202 {
3203 	struct evlist *evlist = trace->evlist;
3204 	struct perf_sample sample;
3205 	int err;
3206 
3207 	err = perf_evlist__parse_sample(evlist, event, &sample);
3208 	if (err)
3209 		fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3210 	else
3211 		trace__handle_event(trace, event, &sample);
3212 
3213 	return 0;
3214 }
3215 
3216 static int __trace__flush_events(struct trace *trace)
3217 {
3218 	u64 first = ordered_events__first_time(&trace->oe.data);
3219 	u64 flush = trace->oe.last - NSEC_PER_SEC;
3220 
3221 	/* Is there some thing to flush.. */
3222 	if (first && first < flush)
3223 		return ordered_events__flush_time(&trace->oe.data, flush);
3224 
3225 	return 0;
3226 }
3227 
3228 static int trace__flush_events(struct trace *trace)
3229 {
3230 	return !trace->sort_events ? 0 : __trace__flush_events(trace);
3231 }
3232 
3233 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3234 {
3235 	int err;
3236 
3237 	if (!trace->sort_events)
3238 		return __trace__deliver_event(trace, event);
3239 
3240 	err = perf_evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3241 	if (err && err != -1)
3242 		return err;
3243 
3244 	err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0);
3245 	if (err)
3246 		return err;
3247 
3248 	return trace__flush_events(trace);
3249 }
3250 
3251 static int ordered_events__deliver_event(struct ordered_events *oe,
3252 					 struct ordered_event *event)
3253 {
3254 	struct trace *trace = container_of(oe, struct trace, oe.data);
3255 
3256 	return __trace__deliver_event(trace, event->event);
3257 }
3258 
3259 static int trace__run(struct trace *trace, int argc, const char **argv)
3260 {
3261 	struct evlist *evlist = trace->evlist;
3262 	struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3263 	int err = -1, i;
3264 	unsigned long before;
3265 	const bool forks = argc > 0;
3266 	bool draining = false;
3267 
3268 	trace->live = true;
3269 
3270 	if (!trace->raw_augmented_syscalls) {
3271 		if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3272 			goto out_error_raw_syscalls;
3273 
3274 		if (trace->trace_syscalls)
3275 			trace->vfs_getname = evlist__add_vfs_getname(evlist);
3276 	}
3277 
3278 	if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3279 		pgfault_maj = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3280 		if (pgfault_maj == NULL)
3281 			goto out_error_mem;
3282 		perf_evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3283 		evlist__add(evlist, pgfault_maj);
3284 	}
3285 
3286 	if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3287 		pgfault_min = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3288 		if (pgfault_min == NULL)
3289 			goto out_error_mem;
3290 		perf_evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3291 		evlist__add(evlist, pgfault_min);
3292 	}
3293 
3294 	if (trace->sched &&
3295 	    perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
3296 				   trace__sched_stat_runtime))
3297 		goto out_error_sched_stat_runtime;
3298 
3299 	/*
3300 	 * If a global cgroup was set, apply it to all the events without an
3301 	 * explicit cgroup. I.e.:
3302 	 *
3303 	 * 	trace -G A -e sched:*switch
3304 	 *
3305 	 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3306 	 * _and_ sched:sched_switch to the 'A' cgroup, while:
3307 	 *
3308 	 * trace -e sched:*switch -G A
3309 	 *
3310 	 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3311 	 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3312 	 * a cgroup (on the root cgroup, sys wide, etc).
3313 	 *
3314 	 * Multiple cgroups:
3315 	 *
3316 	 * trace -G A -e sched:*switch -G B
3317 	 *
3318 	 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3319 	 * to the 'B' cgroup.
3320 	 *
3321 	 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3322 	 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3323 	 */
3324 	if (trace->cgroup)
3325 		evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3326 
3327 	err = perf_evlist__create_maps(evlist, &trace->opts.target);
3328 	if (err < 0) {
3329 		fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
3330 		goto out_delete_evlist;
3331 	}
3332 
3333 	err = trace__symbols_init(trace, evlist);
3334 	if (err < 0) {
3335 		fprintf(trace->output, "Problems initializing symbol libraries!\n");
3336 		goto out_delete_evlist;
3337 	}
3338 
3339 	perf_evlist__config(evlist, &trace->opts, &callchain_param);
3340 
3341 	signal(SIGCHLD, sig_handler);
3342 	signal(SIGINT, sig_handler);
3343 
3344 	if (forks) {
3345 		err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
3346 						    argv, false, NULL);
3347 		if (err < 0) {
3348 			fprintf(trace->output, "Couldn't run the workload!\n");
3349 			goto out_delete_evlist;
3350 		}
3351 	}
3352 
3353 	err = evlist__open(evlist);
3354 	if (err < 0)
3355 		goto out_error_open;
3356 
3357 	err = bpf__apply_obj_config();
3358 	if (err) {
3359 		char errbuf[BUFSIZ];
3360 
3361 		bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
3362 		pr_err("ERROR: Apply config to BPF failed: %s\n",
3363 			 errbuf);
3364 		goto out_error_open;
3365 	}
3366 
3367 	err = trace__set_filter_pids(trace);
3368 	if (err < 0)
3369 		goto out_error_mem;
3370 
3371 	if (trace->syscalls.map)
3372 		trace__init_syscalls_bpf_map(trace);
3373 
3374 	if (trace->syscalls.prog_array.sys_enter)
3375 		trace__init_syscalls_bpf_prog_array_maps(trace);
3376 
3377 	if (trace->ev_qualifier_ids.nr > 0) {
3378 		err = trace__set_ev_qualifier_filter(trace);
3379 		if (err < 0)
3380 			goto out_errno;
3381 
3382 		if (trace->syscalls.events.sys_exit) {
3383 			pr_debug("event qualifier tracepoint filter: %s\n",
3384 				 trace->syscalls.events.sys_exit->filter);
3385 		}
3386 	}
3387 
3388 	/*
3389 	 * If the "close" syscall is not traced, then we will not have the
3390 	 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3391 	 * fd->pathname table and were ending up showing the last value set by
3392 	 * syscalls opening a pathname and associating it with a descriptor or
3393 	 * reading it from /proc/pid/fd/ in cases where that doesn't make
3394 	 * sense.
3395 	 *
3396 	 *  So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3397 	 *  not in use.
3398 	 */
3399 	trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
3400 
3401 	err = perf_evlist__apply_filters(evlist, &evsel);
3402 	if (err < 0)
3403 		goto out_error_apply_filters;
3404 
3405 	if (trace->dump.map)
3406 		bpf_map__fprintf(trace->dump.map, trace->output);
3407 
3408 	err = perf_evlist__mmap(evlist, trace->opts.mmap_pages);
3409 	if (err < 0)
3410 		goto out_error_mmap;
3411 
3412 	if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
3413 		evlist__enable(evlist);
3414 
3415 	if (forks)
3416 		perf_evlist__start_workload(evlist);
3417 
3418 	if (trace->opts.initial_delay) {
3419 		usleep(trace->opts.initial_delay * 1000);
3420 		evlist__enable(evlist);
3421 	}
3422 
3423 	trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
3424 				  evlist->core.threads->nr > 1 ||
3425 				  perf_evlist__first(evlist)->core.attr.inherit;
3426 
3427 	/*
3428 	 * Now that we already used evsel->core.attr to ask the kernel to setup the
3429 	 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
3430 	 * trace__resolve_callchain(), allowing per-event max-stack settings
3431 	 * to override an explicitly set --max-stack global setting.
3432 	 */
3433 	evlist__for_each_entry(evlist, evsel) {
3434 		if (evsel__has_callchain(evsel) &&
3435 		    evsel->core.attr.sample_max_stack == 0)
3436 			evsel->core.attr.sample_max_stack = trace->max_stack;
3437 	}
3438 again:
3439 	before = trace->nr_events;
3440 
3441 	for (i = 0; i < evlist->nr_mmaps; i++) {
3442 		union perf_event *event;
3443 		struct perf_mmap *md;
3444 
3445 		md = &evlist->mmap[i];
3446 		if (perf_mmap__read_init(md) < 0)
3447 			continue;
3448 
3449 		while ((event = perf_mmap__read_event(md)) != NULL) {
3450 			++trace->nr_events;
3451 
3452 			err = trace__deliver_event(trace, event);
3453 			if (err)
3454 				goto out_disable;
3455 
3456 			perf_mmap__consume(md);
3457 
3458 			if (interrupted)
3459 				goto out_disable;
3460 
3461 			if (done && !draining) {
3462 				evlist__disable(evlist);
3463 				draining = true;
3464 			}
3465 		}
3466 		perf_mmap__read_done(md);
3467 	}
3468 
3469 	if (trace->nr_events == before) {
3470 		int timeout = done ? 100 : -1;
3471 
3472 		if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
3473 			if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
3474 				draining = true;
3475 
3476 			goto again;
3477 		} else {
3478 			if (trace__flush_events(trace))
3479 				goto out_disable;
3480 		}
3481 	} else {
3482 		goto again;
3483 	}
3484 
3485 out_disable:
3486 	thread__zput(trace->current);
3487 
3488 	evlist__disable(evlist);
3489 
3490 	if (trace->sort_events)
3491 		ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
3492 
3493 	if (!err) {
3494 		if (trace->summary)
3495 			trace__fprintf_thread_summary(trace, trace->output);
3496 
3497 		if (trace->show_tool_stats) {
3498 			fprintf(trace->output, "Stats:\n "
3499 					       " vfs_getname : %" PRIu64 "\n"
3500 					       " proc_getname: %" PRIu64 "\n",
3501 				trace->stats.vfs_getname,
3502 				trace->stats.proc_getname);
3503 		}
3504 	}
3505 
3506 out_delete_evlist:
3507 	trace__symbols__exit(trace);
3508 
3509 	evlist__delete(evlist);
3510 	cgroup__put(trace->cgroup);
3511 	trace->evlist = NULL;
3512 	trace->live = false;
3513 	return err;
3514 {
3515 	char errbuf[BUFSIZ];
3516 
3517 out_error_sched_stat_runtime:
3518 	tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
3519 	goto out_error;
3520 
3521 out_error_raw_syscalls:
3522 	tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
3523 	goto out_error;
3524 
3525 out_error_mmap:
3526 	perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
3527 	goto out_error;
3528 
3529 out_error_open:
3530 	perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
3531 
3532 out_error:
3533 	fprintf(trace->output, "%s\n", errbuf);
3534 	goto out_delete_evlist;
3535 
3536 out_error_apply_filters:
3537 	fprintf(trace->output,
3538 		"Failed to set filter \"%s\" on event %s with %d (%s)\n",
3539 		evsel->filter, perf_evsel__name(evsel), errno,
3540 		str_error_r(errno, errbuf, sizeof(errbuf)));
3541 	goto out_delete_evlist;
3542 }
3543 out_error_mem:
3544 	fprintf(trace->output, "Not enough memory to run!\n");
3545 	goto out_delete_evlist;
3546 
3547 out_errno:
3548 	fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
3549 	goto out_delete_evlist;
3550 }
3551 
3552 static int trace__replay(struct trace *trace)
3553 {
3554 	const struct evsel_str_handler handlers[] = {
3555 		{ "probe:vfs_getname",	     trace__vfs_getname, },
3556 	};
3557 	struct perf_data data = {
3558 		.path  = input_name,
3559 		.mode  = PERF_DATA_MODE_READ,
3560 		.force = trace->force,
3561 	};
3562 	struct perf_session *session;
3563 	struct evsel *evsel;
3564 	int err = -1;
3565 
3566 	trace->tool.sample	  = trace__process_sample;
3567 	trace->tool.mmap	  = perf_event__process_mmap;
3568 	trace->tool.mmap2	  = perf_event__process_mmap2;
3569 	trace->tool.comm	  = perf_event__process_comm;
3570 	trace->tool.exit	  = perf_event__process_exit;
3571 	trace->tool.fork	  = perf_event__process_fork;
3572 	trace->tool.attr	  = perf_event__process_attr;
3573 	trace->tool.tracing_data  = perf_event__process_tracing_data;
3574 	trace->tool.build_id	  = perf_event__process_build_id;
3575 	trace->tool.namespaces	  = perf_event__process_namespaces;
3576 
3577 	trace->tool.ordered_events = true;
3578 	trace->tool.ordering_requires_timestamps = true;
3579 
3580 	/* add tid to output */
3581 	trace->multiple_threads = true;
3582 
3583 	session = perf_session__new(&data, false, &trace->tool);
3584 	if (session == NULL)
3585 		return -1;
3586 
3587 	if (trace->opts.target.pid)
3588 		symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
3589 
3590 	if (trace->opts.target.tid)
3591 		symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
3592 
3593 	if (symbol__init(&session->header.env) < 0)
3594 		goto out;
3595 
3596 	trace->host = &session->machines.host;
3597 
3598 	err = perf_session__set_tracepoints_handlers(session, handlers);
3599 	if (err)
3600 		goto out;
3601 
3602 	evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3603 						     "raw_syscalls:sys_enter");
3604 	/* older kernels have syscalls tp versus raw_syscalls */
3605 	if (evsel == NULL)
3606 		evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3607 							     "syscalls:sys_enter");
3608 
3609 	if (evsel &&
3610 	    (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
3611 	    perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
3612 		pr_err("Error during initialize raw_syscalls:sys_enter event\n");
3613 		goto out;
3614 	}
3615 
3616 	evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3617 						     "raw_syscalls:sys_exit");
3618 	if (evsel == NULL)
3619 		evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3620 							     "syscalls:sys_exit");
3621 	if (evsel &&
3622 	    (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
3623 	    perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
3624 		pr_err("Error during initialize raw_syscalls:sys_exit event\n");
3625 		goto out;
3626 	}
3627 
3628 	evlist__for_each_entry(session->evlist, evsel) {
3629 		if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
3630 		    (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
3631 		     evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
3632 		     evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
3633 			evsel->handler = trace__pgfault;
3634 	}
3635 
3636 	setup_pager();
3637 
3638 	err = perf_session__process_events(session);
3639 	if (err)
3640 		pr_err("Failed to process events, error %d", err);
3641 
3642 	else if (trace->summary)
3643 		trace__fprintf_thread_summary(trace, trace->output);
3644 
3645 out:
3646 	perf_session__delete(session);
3647 
3648 	return err;
3649 }
3650 
3651 static size_t trace__fprintf_threads_header(FILE *fp)
3652 {
3653 	size_t printed;
3654 
3655 	printed  = fprintf(fp, "\n Summary of events:\n\n");
3656 
3657 	return printed;
3658 }
3659 
3660 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
3661 	struct stats 	*stats;
3662 	double		msecs;
3663 	int		syscall;
3664 )
3665 {
3666 	struct int_node *source = rb_entry(nd, struct int_node, rb_node);
3667 	struct stats *stats = source->priv;
3668 
3669 	entry->syscall = source->i;
3670 	entry->stats   = stats;
3671 	entry->msecs   = stats ? (u64)stats->n * (avg_stats(stats) / NSEC_PER_MSEC) : 0;
3672 }
3673 
3674 static size_t thread__dump_stats(struct thread_trace *ttrace,
3675 				 struct trace *trace, FILE *fp)
3676 {
3677 	size_t printed = 0;
3678 	struct syscall *sc;
3679 	struct rb_node *nd;
3680 	DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
3681 
3682 	if (syscall_stats == NULL)
3683 		return 0;
3684 
3685 	printed += fprintf(fp, "\n");
3686 
3687 	printed += fprintf(fp, "   syscall            calls    total       min       avg       max      stddev\n");
3688 	printed += fprintf(fp, "                               (msec)    (msec)    (msec)    (msec)        (%%)\n");
3689 	printed += fprintf(fp, "   --------------- -------- --------- --------- --------- ---------     ------\n");
3690 
3691 	resort_rb__for_each_entry(nd, syscall_stats) {
3692 		struct stats *stats = syscall_stats_entry->stats;
3693 		if (stats) {
3694 			double min = (double)(stats->min) / NSEC_PER_MSEC;
3695 			double max = (double)(stats->max) / NSEC_PER_MSEC;
3696 			double avg = avg_stats(stats);
3697 			double pct;
3698 			u64 n = (u64) stats->n;
3699 
3700 			pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
3701 			avg /= NSEC_PER_MSEC;
3702 
3703 			sc = &trace->syscalls.table[syscall_stats_entry->syscall];
3704 			printed += fprintf(fp, "   %-15s", sc->name);
3705 			printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f %9.3f",
3706 					   n, syscall_stats_entry->msecs, min, avg);
3707 			printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
3708 		}
3709 	}
3710 
3711 	resort_rb__delete(syscall_stats);
3712 	printed += fprintf(fp, "\n\n");
3713 
3714 	return printed;
3715 }
3716 
3717 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
3718 {
3719 	size_t printed = 0;
3720 	struct thread_trace *ttrace = thread__priv(thread);
3721 	double ratio;
3722 
3723 	if (ttrace == NULL)
3724 		return 0;
3725 
3726 	ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
3727 
3728 	printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
3729 	printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
3730 	printed += fprintf(fp, "%.1f%%", ratio);
3731 	if (ttrace->pfmaj)
3732 		printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
3733 	if (ttrace->pfmin)
3734 		printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
3735 	if (trace->sched)
3736 		printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
3737 	else if (fputc('\n', fp) != EOF)
3738 		++printed;
3739 
3740 	printed += thread__dump_stats(ttrace, trace, fp);
3741 
3742 	return printed;
3743 }
3744 
3745 static unsigned long thread__nr_events(struct thread_trace *ttrace)
3746 {
3747 	return ttrace ? ttrace->nr_events : 0;
3748 }
3749 
3750 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
3751 	struct thread *thread;
3752 )
3753 {
3754 	entry->thread = rb_entry(nd, struct thread, rb_node);
3755 }
3756 
3757 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
3758 {
3759 	size_t printed = trace__fprintf_threads_header(fp);
3760 	struct rb_node *nd;
3761 	int i;
3762 
3763 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3764 		DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
3765 
3766 		if (threads == NULL) {
3767 			fprintf(fp, "%s", "Error sorting output by nr_events!\n");
3768 			return 0;
3769 		}
3770 
3771 		resort_rb__for_each_entry(nd, threads)
3772 			printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
3773 
3774 		resort_rb__delete(threads);
3775 	}
3776 	return printed;
3777 }
3778 
3779 static int trace__set_duration(const struct option *opt, const char *str,
3780 			       int unset __maybe_unused)
3781 {
3782 	struct trace *trace = opt->value;
3783 
3784 	trace->duration_filter = atof(str);
3785 	return 0;
3786 }
3787 
3788 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
3789 					      int unset __maybe_unused)
3790 {
3791 	int ret = -1;
3792 	size_t i;
3793 	struct trace *trace = opt->value;
3794 	/*
3795 	 * FIXME: introduce a intarray class, plain parse csv and create a
3796 	 * { int nr, int entries[] } struct...
3797 	 */
3798 	struct intlist *list = intlist__new(str);
3799 
3800 	if (list == NULL)
3801 		return -1;
3802 
3803 	i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
3804 	trace->filter_pids.entries = calloc(i, sizeof(pid_t));
3805 
3806 	if (trace->filter_pids.entries == NULL)
3807 		goto out;
3808 
3809 	trace->filter_pids.entries[0] = getpid();
3810 
3811 	for (i = 1; i < trace->filter_pids.nr; ++i)
3812 		trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
3813 
3814 	intlist__delete(list);
3815 	ret = 0;
3816 out:
3817 	return ret;
3818 }
3819 
3820 static int trace__open_output(struct trace *trace, const char *filename)
3821 {
3822 	struct stat st;
3823 
3824 	if (!stat(filename, &st) && st.st_size) {
3825 		char oldname[PATH_MAX];
3826 
3827 		scnprintf(oldname, sizeof(oldname), "%s.old", filename);
3828 		unlink(oldname);
3829 		rename(filename, oldname);
3830 	}
3831 
3832 	trace->output = fopen(filename, "w");
3833 
3834 	return trace->output == NULL ? -errno : 0;
3835 }
3836 
3837 static int parse_pagefaults(const struct option *opt, const char *str,
3838 			    int unset __maybe_unused)
3839 {
3840 	int *trace_pgfaults = opt->value;
3841 
3842 	if (strcmp(str, "all") == 0)
3843 		*trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
3844 	else if (strcmp(str, "maj") == 0)
3845 		*trace_pgfaults |= TRACE_PFMAJ;
3846 	else if (strcmp(str, "min") == 0)
3847 		*trace_pgfaults |= TRACE_PFMIN;
3848 	else
3849 		return -1;
3850 
3851 	return 0;
3852 }
3853 
3854 static void evlist__set_evsel_handler(struct evlist *evlist, void *handler)
3855 {
3856 	struct evsel *evsel;
3857 
3858 	evlist__for_each_entry(evlist, evsel)
3859 		evsel->handler = handler;
3860 }
3861 
3862 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
3863 {
3864 	struct evsel *evsel;
3865 
3866 	evlist__for_each_entry(evlist, evsel) {
3867 		if (evsel->priv || !evsel->tp_format)
3868 			continue;
3869 
3870 		if (strcmp(evsel->tp_format->system, "syscalls"))
3871 			continue;
3872 
3873 		if (perf_evsel__init_syscall_tp(evsel))
3874 			return -1;
3875 
3876 		if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
3877 			struct syscall_tp *sc = evsel->priv;
3878 
3879 			if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
3880 				return -1;
3881 		} else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
3882 			struct syscall_tp *sc = evsel->priv;
3883 
3884 			if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
3885 				return -1;
3886 		}
3887 	}
3888 
3889 	return 0;
3890 }
3891 
3892 /*
3893  * XXX: Hackish, just splitting the combined -e+--event (syscalls
3894  * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
3895  * existing facilities unchanged (trace->ev_qualifier + parse_options()).
3896  *
3897  * It'd be better to introduce a parse_options() variant that would return a
3898  * list with the terms it didn't match to an event...
3899  */
3900 static int trace__parse_events_option(const struct option *opt, const char *str,
3901 				      int unset __maybe_unused)
3902 {
3903 	struct trace *trace = (struct trace *)opt->value;
3904 	const char *s = str;
3905 	char *sep = NULL, *lists[2] = { NULL, NULL, };
3906 	int len = strlen(str) + 1, err = -1, list, idx;
3907 	char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
3908 	char group_name[PATH_MAX];
3909 	struct syscall_fmt *fmt;
3910 
3911 	if (strace_groups_dir == NULL)
3912 		return -1;
3913 
3914 	if (*s == '!') {
3915 		++s;
3916 		trace->not_ev_qualifier = true;
3917 	}
3918 
3919 	while (1) {
3920 		if ((sep = strchr(s, ',')) != NULL)
3921 			*sep = '\0';
3922 
3923 		list = 0;
3924 		if (syscalltbl__id(trace->sctbl, s) >= 0 ||
3925 		    syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
3926 			list = 1;
3927 			goto do_concat;
3928 		}
3929 
3930 		fmt = syscall_fmt__find_by_alias(s);
3931 		if (fmt != NULL) {
3932 			list = 1;
3933 			s = fmt->name;
3934 		} else {
3935 			path__join(group_name, sizeof(group_name), strace_groups_dir, s);
3936 			if (access(group_name, R_OK) == 0)
3937 				list = 1;
3938 		}
3939 do_concat:
3940 		if (lists[list]) {
3941 			sprintf(lists[list] + strlen(lists[list]), ",%s", s);
3942 		} else {
3943 			lists[list] = malloc(len);
3944 			if (lists[list] == NULL)
3945 				goto out;
3946 			strcpy(lists[list], s);
3947 		}
3948 
3949 		if (!sep)
3950 			break;
3951 
3952 		*sep = ',';
3953 		s = sep + 1;
3954 	}
3955 
3956 	if (lists[1] != NULL) {
3957 		struct strlist_config slist_config = {
3958 			.dirname = strace_groups_dir,
3959 		};
3960 
3961 		trace->ev_qualifier = strlist__new(lists[1], &slist_config);
3962 		if (trace->ev_qualifier == NULL) {
3963 			fputs("Not enough memory to parse event qualifier", trace->output);
3964 			goto out;
3965 		}
3966 
3967 		if (trace__validate_ev_qualifier(trace))
3968 			goto out;
3969 		trace->trace_syscalls = true;
3970 	}
3971 
3972 	err = 0;
3973 
3974 	if (lists[0]) {
3975 		struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
3976 					       "event selector. use 'perf list' to list available events",
3977 					       parse_events_option);
3978 		err = parse_events_option(&o, lists[0], 0);
3979 	}
3980 out:
3981 	if (sep)
3982 		*sep = ',';
3983 
3984 	return err;
3985 }
3986 
3987 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
3988 {
3989 	struct trace *trace = opt->value;
3990 
3991 	if (!list_empty(&trace->evlist->core.entries))
3992 		return parse_cgroups(opt, str, unset);
3993 
3994 	trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
3995 
3996 	return 0;
3997 }
3998 
3999 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
4000 {
4001 	if (trace->bpf_obj == NULL)
4002 		return NULL;
4003 
4004 	return bpf_object__find_map_by_name(trace->bpf_obj, name);
4005 }
4006 
4007 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
4008 {
4009 	trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
4010 }
4011 
4012 static void trace__set_bpf_map_syscalls(struct trace *trace)
4013 {
4014 	trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
4015 	trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
4016 	trace->syscalls.prog_array.sys_exit  = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
4017 }
4018 
4019 static int trace__config(const char *var, const char *value, void *arg)
4020 {
4021 	struct trace *trace = arg;
4022 	int err = 0;
4023 
4024 	if (!strcmp(var, "trace.add_events")) {
4025 		struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
4026 					       "event selector. use 'perf list' to list available events",
4027 					       parse_events_option);
4028 		/*
4029 		 * We can't propagate parse_event_option() return, as it is 1
4030 		 * for failure while perf_config() expects -1.
4031 		 */
4032 		if (parse_events_option(&o, value, 0))
4033 			err = -1;
4034 	} else if (!strcmp(var, "trace.show_timestamp")) {
4035 		trace->show_tstamp = perf_config_bool(var, value);
4036 	} else if (!strcmp(var, "trace.show_duration")) {
4037 		trace->show_duration = perf_config_bool(var, value);
4038 	} else if (!strcmp(var, "trace.show_arg_names")) {
4039 		trace->show_arg_names = perf_config_bool(var, value);
4040 		if (!trace->show_arg_names)
4041 			trace->show_zeros = true;
4042 	} else if (!strcmp(var, "trace.show_zeros")) {
4043 		bool new_show_zeros = perf_config_bool(var, value);
4044 		if (!trace->show_arg_names && !new_show_zeros) {
4045 			pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4046 			goto out;
4047 		}
4048 		trace->show_zeros = new_show_zeros;
4049 	} else if (!strcmp(var, "trace.show_prefix")) {
4050 		trace->show_string_prefix = perf_config_bool(var, value);
4051 	} else if (!strcmp(var, "trace.no_inherit")) {
4052 		trace->opts.no_inherit = perf_config_bool(var, value);
4053 	} else if (!strcmp(var, "trace.args_alignment")) {
4054 		int args_alignment = 0;
4055 		if (perf_config_int(&args_alignment, var, value) == 0)
4056 			trace->args_alignment = args_alignment;
4057 	}
4058 out:
4059 	return err;
4060 }
4061 
4062 int cmd_trace(int argc, const char **argv)
4063 {
4064 	const char *trace_usage[] = {
4065 		"perf trace [<options>] [<command>]",
4066 		"perf trace [<options>] -- <command> [<options>]",
4067 		"perf trace record [<options>] [<command>]",
4068 		"perf trace record [<options>] -- <command> [<options>]",
4069 		NULL
4070 	};
4071 	struct trace trace = {
4072 		.opts = {
4073 			.target = {
4074 				.uid	   = UINT_MAX,
4075 				.uses_mmap = true,
4076 			},
4077 			.user_freq     = UINT_MAX,
4078 			.user_interval = ULLONG_MAX,
4079 			.no_buffering  = true,
4080 			.mmap_pages    = UINT_MAX,
4081 		},
4082 		.output = stderr,
4083 		.show_comm = true,
4084 		.show_tstamp = true,
4085 		.show_duration = true,
4086 		.show_arg_names = true,
4087 		.args_alignment = 70,
4088 		.trace_syscalls = false,
4089 		.kernel_syscallchains = false,
4090 		.max_stack = UINT_MAX,
4091 		.max_events = ULONG_MAX,
4092 	};
4093 	const char *map_dump_str = NULL;
4094 	const char *output_name = NULL;
4095 	const struct option trace_options[] = {
4096 	OPT_CALLBACK('e', "event", &trace, "event",
4097 		     "event/syscall selector. use 'perf list' to list available events",
4098 		     trace__parse_events_option),
4099 	OPT_BOOLEAN(0, "comm", &trace.show_comm,
4100 		    "show the thread COMM next to its id"),
4101 	OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4102 	OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4103 		     trace__parse_events_option),
4104 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
4105 	OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4106 	OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4107 		    "trace events on existing process id"),
4108 	OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4109 		    "trace events on existing thread id"),
4110 	OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4111 		     "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4112 	OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4113 		    "system-wide collection from all CPUs"),
4114 	OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4115 		    "list of cpus to monitor"),
4116 	OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4117 		    "child tasks do not inherit counters"),
4118 	OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4119 		     "number of mmap data pages",
4120 		     perf_evlist__parse_mmap_pages),
4121 	OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4122 		   "user to profile"),
4123 	OPT_CALLBACK(0, "duration", &trace, "float",
4124 		     "show only events with duration > N.M ms",
4125 		     trace__set_duration),
4126 #ifdef HAVE_LIBBPF_SUPPORT
4127 	OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4128 #endif
4129 	OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4130 	OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4131 	OPT_BOOLEAN('T', "time", &trace.full_time,
4132 		    "Show full timestamp, not time relative to first start"),
4133 	OPT_BOOLEAN(0, "failure", &trace.failure_only,
4134 		    "Show only syscalls that failed"),
4135 	OPT_BOOLEAN('s', "summary", &trace.summary_only,
4136 		    "Show only syscall summary with statistics"),
4137 	OPT_BOOLEAN('S', "with-summary", &trace.summary,
4138 		    "Show all syscalls and summary with statistics"),
4139 	OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4140 		     "Trace pagefaults", parse_pagefaults, "maj"),
4141 	OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4142 	OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4143 	OPT_CALLBACK(0, "call-graph", &trace.opts,
4144 		     "record_mode[,record_size]", record_callchain_help,
4145 		     &record_parse_callchain_opt),
4146 	OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4147 		    "Show the kernel callchains on the syscall exit path"),
4148 	OPT_ULONG(0, "max-events", &trace.max_events,
4149 		"Set the maximum number of events to print, exit after that is reached. "),
4150 	OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4151 		     "Set the minimum stack depth when parsing the callchain, "
4152 		     "anything below the specified depth will be ignored."),
4153 	OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4154 		     "Set the maximum stack depth when parsing the callchain, "
4155 		     "anything beyond the specified depth will be ignored. "
4156 		     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4157 	OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4158 			"Sort batch of events before processing, use if getting out of order events"),
4159 	OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4160 			"print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4161 	OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4162 			"per thread proc mmap processing timeout in ms"),
4163 	OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4164 		     trace__parse_cgroups),
4165 	OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
4166 		     "ms to wait before starting measurement after program "
4167 		     "start"),
4168 	OPTS_EVSWITCH(&trace.evswitch),
4169 	OPT_END()
4170 	};
4171 	bool __maybe_unused max_stack_user_set = true;
4172 	bool mmap_pages_user_set = true;
4173 	struct evsel *evsel;
4174 	const char * const trace_subcommands[] = { "record", NULL };
4175 	int err = -1;
4176 	char bf[BUFSIZ];
4177 
4178 	signal(SIGSEGV, sighandler_dump_stack);
4179 	signal(SIGFPE, sighandler_dump_stack);
4180 
4181 	trace.evlist = evlist__new();
4182 	trace.sctbl = syscalltbl__new();
4183 
4184 	if (trace.evlist == NULL || trace.sctbl == NULL) {
4185 		pr_err("Not enough memory to run!\n");
4186 		err = -ENOMEM;
4187 		goto out;
4188 	}
4189 
4190 	/*
4191 	 * Parsing .perfconfig may entail creating a BPF event, that may need
4192 	 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4193 	 * is too small. This affects just this process, not touching the
4194 	 * global setting. If it fails we'll get something in 'perf trace -v'
4195 	 * to help diagnose the problem.
4196 	 */
4197 	rlimit__bump_memlock();
4198 
4199 	err = perf_config(trace__config, &trace);
4200 	if (err)
4201 		goto out;
4202 
4203 	argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4204 				 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4205 
4206 	if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4207 		usage_with_options_msg(trace_usage, trace_options,
4208 				       "cgroup monitoring only available in system-wide mode");
4209 	}
4210 
4211 	evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4212 	if (IS_ERR(evsel)) {
4213 		bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4214 		pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4215 		goto out;
4216 	}
4217 
4218 	if (evsel) {
4219 		trace.syscalls.events.augmented = evsel;
4220 
4221 		evsel = perf_evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4222 		if (evsel == NULL) {
4223 			pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4224 			goto out;
4225 		}
4226 
4227 		if (evsel->bpf_obj == NULL) {
4228 			pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4229 			goto out;
4230 		}
4231 
4232 		trace.bpf_obj = evsel->bpf_obj;
4233 
4234 		trace__set_bpf_map_filtered_pids(&trace);
4235 		trace__set_bpf_map_syscalls(&trace);
4236 		trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
4237 	}
4238 
4239 	err = bpf__setup_stdout(trace.evlist);
4240 	if (err) {
4241 		bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
4242 		pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
4243 		goto out;
4244 	}
4245 
4246 	err = -1;
4247 
4248 	if (map_dump_str) {
4249 		trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
4250 		if (trace.dump.map == NULL) {
4251 			pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
4252 			goto out;
4253 		}
4254 	}
4255 
4256 	if (trace.trace_pgfaults) {
4257 		trace.opts.sample_address = true;
4258 		trace.opts.sample_time = true;
4259 	}
4260 
4261 	if (trace.opts.mmap_pages == UINT_MAX)
4262 		mmap_pages_user_set = false;
4263 
4264 	if (trace.max_stack == UINT_MAX) {
4265 		trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
4266 		max_stack_user_set = false;
4267 	}
4268 
4269 #ifdef HAVE_DWARF_UNWIND_SUPPORT
4270 	if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
4271 		record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
4272 	}
4273 #endif
4274 
4275 	if (callchain_param.enabled) {
4276 		if (!mmap_pages_user_set && geteuid() == 0)
4277 			trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
4278 
4279 		symbol_conf.use_callchain = true;
4280 	}
4281 
4282 	if (trace.evlist->core.nr_entries > 0) {
4283 		evlist__set_evsel_handler(trace.evlist, trace__event_handler);
4284 		if (evlist__set_syscall_tp_fields(trace.evlist)) {
4285 			perror("failed to set syscalls:* tracepoint fields");
4286 			goto out;
4287 		}
4288 	}
4289 
4290 	if (trace.sort_events) {
4291 		ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
4292 		ordered_events__set_copy_on_queue(&trace.oe.data, true);
4293 	}
4294 
4295 	/*
4296 	 * If we are augmenting syscalls, then combine what we put in the
4297 	 * __augmented_syscalls__ BPF map with what is in the
4298 	 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
4299 	 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
4300 	 *
4301 	 * We'll switch to look at two BPF maps, one for sys_enter and the
4302 	 * other for sys_exit when we start augmenting the sys_exit paths with
4303 	 * buffers that are being copied from kernel to userspace, think 'read'
4304 	 * syscall.
4305 	 */
4306 	if (trace.syscalls.events.augmented) {
4307 		evlist__for_each_entry(trace.evlist, evsel) {
4308 			bool raw_syscalls_sys_exit = strcmp(perf_evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
4309 
4310 			if (raw_syscalls_sys_exit) {
4311 				trace.raw_augmented_syscalls = true;
4312 				goto init_augmented_syscall_tp;
4313 			}
4314 
4315 			if (trace.syscalls.events.augmented->priv == NULL &&
4316 			    strstr(perf_evsel__name(evsel), "syscalls:sys_enter")) {
4317 				struct evsel *augmented = trace.syscalls.events.augmented;
4318 				if (perf_evsel__init_augmented_syscall_tp(augmented, evsel) ||
4319 				    perf_evsel__init_augmented_syscall_tp_args(augmented))
4320 					goto out;
4321 				/*
4322 				 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
4323 				 * Above we made sure we can get from the payload the tp fields
4324 				 * that we get from syscalls:sys_enter tracefs format file.
4325 				 */
4326 				augmented->handler = trace__sys_enter;
4327 				/*
4328 				 * Now we do the same for the *syscalls:sys_enter event so that
4329 				 * if we handle it directly, i.e. if the BPF prog returns 0 so
4330 				 * as not to filter it, then we'll handle it just like we would
4331 				 * for the BPF_OUTPUT one:
4332 				 */
4333 				if (perf_evsel__init_augmented_syscall_tp(evsel, evsel) ||
4334 				    perf_evsel__init_augmented_syscall_tp_args(evsel))
4335 					goto out;
4336 				evsel->handler = trace__sys_enter;
4337 			}
4338 
4339 			if (strstarts(perf_evsel__name(evsel), "syscalls:sys_exit_")) {
4340 				struct syscall_tp *sc;
4341 init_augmented_syscall_tp:
4342 				if (perf_evsel__init_augmented_syscall_tp(evsel, evsel))
4343 					goto out;
4344 				sc = evsel->priv;
4345 				/*
4346 				 * For now with BPF raw_augmented we hook into
4347 				 * raw_syscalls:sys_enter and there we get all
4348 				 * 6 syscall args plus the tracepoint common
4349 				 * fields and the syscall_nr (another long).
4350 				 * So we check if that is the case and if so
4351 				 * don't look after the sc->args_size but
4352 				 * always after the full raw_syscalls:sys_enter
4353 				 * payload, which is fixed.
4354 				 *
4355 				 * We'll revisit this later to pass
4356 				 * s->args_size to the BPF augmenter (now
4357 				 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
4358 				 * so that it copies only what we need for each
4359 				 * syscall, like what happens when we use
4360 				 * syscalls:sys_enter_NAME, so that we reduce
4361 				 * the kernel/userspace traffic to just what is
4362 				 * needed for each syscall.
4363 				 */
4364 				if (trace.raw_augmented_syscalls)
4365 					trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
4366 				perf_evsel__init_augmented_syscall_tp_ret(evsel);
4367 				evsel->handler = trace__sys_exit;
4368 			}
4369 		}
4370 	}
4371 
4372 	if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
4373 		return trace__record(&trace, argc-1, &argv[1]);
4374 
4375 	/* summary_only implies summary option, but don't overwrite summary if set */
4376 	if (trace.summary_only)
4377 		trace.summary = trace.summary_only;
4378 
4379 	if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4380 	    trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4381 		trace.trace_syscalls = true;
4382 	}
4383 
4384 	if (output_name != NULL) {
4385 		err = trace__open_output(&trace, output_name);
4386 		if (err < 0) {
4387 			perror("failed to create output file");
4388 			goto out;
4389 		}
4390 	}
4391 
4392 	err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
4393 	if (err)
4394 		goto out_close;
4395 
4396 	err = target__validate(&trace.opts.target);
4397 	if (err) {
4398 		target__strerror(&trace.opts.target, err, bf, sizeof(bf));
4399 		fprintf(trace.output, "%s", bf);
4400 		goto out_close;
4401 	}
4402 
4403 	err = target__parse_uid(&trace.opts.target);
4404 	if (err) {
4405 		target__strerror(&trace.opts.target, err, bf, sizeof(bf));
4406 		fprintf(trace.output, "%s", bf);
4407 		goto out_close;
4408 	}
4409 
4410 	if (!argc && target__none(&trace.opts.target))
4411 		trace.opts.target.system_wide = true;
4412 
4413 	if (input_name)
4414 		err = trace__replay(&trace);
4415 	else
4416 		err = trace__run(&trace, argc, argv);
4417 
4418 out_close:
4419 	if (output_name != NULL)
4420 		fclose(trace.output);
4421 out:
4422 	return err;
4423 }
4424