xref: /linux/tools/perf/builtin-trace.c (revision f11b2803bb88655d90b88c787710b53100913bff)
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/dso.h"
29 #include "util/env.h"
30 #include "util/event.h"
31 #include "util/evsel.h"
32 #include "util/evsel_fprintf.h"
33 #include "util/synthetic-events.h"
34 #include "util/evlist.h"
35 #include "util/evswitch.h"
36 #include "util/mmap.h"
37 #include <subcmd/pager.h>
38 #include <subcmd/exec-cmd.h>
39 #include "util/machine.h"
40 #include "util/map.h"
41 #include "util/symbol.h"
42 #include "util/path.h"
43 #include "util/session.h"
44 #include "util/thread.h"
45 #include <subcmd/parse-options.h>
46 #include "util/strlist.h"
47 #include "util/intlist.h"
48 #include "util/thread_map.h"
49 #include "util/stat.h"
50 #include "util/tool.h"
51 #include "util/util.h"
52 #include "trace/beauty/beauty.h"
53 #include "trace-event.h"
54 #include "util/parse-events.h"
55 #include "util/bpf-loader.h"
56 #include "callchain.h"
57 #include "print_binary.h"
58 #include "string2.h"
59 #include "syscalltbl.h"
60 #include "rb_resort.h"
61 #include "../perf.h"
62 
63 #include <errno.h>
64 #include <inttypes.h>
65 #include <poll.h>
66 #include <signal.h>
67 #include <stdlib.h>
68 #include <string.h>
69 #include <linux/err.h>
70 #include <linux/filter.h>
71 #include <linux/kernel.h>
72 #include <linux/random.h>
73 #include <linux/stringify.h>
74 #include <linux/time64.h>
75 #include <linux/zalloc.h>
76 #include <fcntl.h>
77 #include <sys/sysmacros.h>
78 
79 #include <linux/ctype.h>
80 
81 #ifndef O_CLOEXEC
82 # define O_CLOEXEC		02000000
83 #endif
84 
85 #ifndef F_LINUX_SPECIFIC_BASE
86 # define F_LINUX_SPECIFIC_BASE	1024
87 #endif
88 
89 struct syscall_arg_fmt {
90 	size_t	   (*scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
91 	unsigned long (*mask_val)(struct syscall_arg *arg, unsigned long val);
92 	void	   *parm;
93 	const char *name;
94 	u16	   nr_entries; // for arrays
95 	bool	   show_zero;
96 };
97 
98 struct syscall_fmt {
99 	const char *name;
100 	const char *alias;
101 	struct {
102 		const char *sys_enter,
103 			   *sys_exit;
104 	}	   bpf_prog_name;
105 	struct syscall_arg_fmt arg[6];
106 	u8	   nr_args;
107 	bool	   errpid;
108 	bool	   timeout;
109 	bool	   hexret;
110 };
111 
112 struct trace {
113 	struct perf_tool	tool;
114 	struct syscalltbl	*sctbl;
115 	struct {
116 		struct syscall  *table;
117 		struct bpf_map  *map;
118 		struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
119 			struct bpf_map  *sys_enter,
120 					*sys_exit;
121 		}		prog_array;
122 		struct {
123 			struct evsel *sys_enter,
124 					  *sys_exit,
125 					  *augmented;
126 		}		events;
127 		struct bpf_program *unaugmented_prog;
128 	} syscalls;
129 	struct {
130 		struct bpf_map *map;
131 	} dump;
132 	struct record_opts	opts;
133 	struct evlist	*evlist;
134 	struct machine		*host;
135 	struct thread		*current;
136 	struct bpf_object	*bpf_obj;
137 	struct cgroup		*cgroup;
138 	u64			base_time;
139 	FILE			*output;
140 	unsigned long		nr_events;
141 	unsigned long		nr_events_printed;
142 	unsigned long		max_events;
143 	struct evswitch		evswitch;
144 	struct strlist		*ev_qualifier;
145 	struct {
146 		size_t		nr;
147 		int		*entries;
148 	}			ev_qualifier_ids;
149 	struct {
150 		size_t		nr;
151 		pid_t		*entries;
152 		struct bpf_map  *map;
153 	}			filter_pids;
154 	double			duration_filter;
155 	double			runtime_ms;
156 	struct {
157 		u64		vfs_getname,
158 				proc_getname;
159 	} stats;
160 	unsigned int		max_stack;
161 	unsigned int		min_stack;
162 	int			raw_augmented_syscalls_args_size;
163 	bool			raw_augmented_syscalls;
164 	bool			fd_path_disabled;
165 	bool			sort_events;
166 	bool			not_ev_qualifier;
167 	bool			live;
168 	bool			full_time;
169 	bool			sched;
170 	bool			multiple_threads;
171 	bool			summary;
172 	bool			summary_only;
173 	bool			failure_only;
174 	bool			show_comm;
175 	bool			print_sample;
176 	bool			show_tool_stats;
177 	bool			trace_syscalls;
178 	bool			libtraceevent_print;
179 	bool			kernel_syscallchains;
180 	s16			args_alignment;
181 	bool			show_tstamp;
182 	bool			show_duration;
183 	bool			show_zeros;
184 	bool			show_arg_names;
185 	bool			show_string_prefix;
186 	bool			force;
187 	bool			vfs_getname;
188 	int			trace_pgfaults;
189 	char			*perfconfig_events;
190 	struct {
191 		struct ordered_events	data;
192 		u64			last;
193 	} oe;
194 };
195 
196 struct tp_field {
197 	int offset;
198 	union {
199 		u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
200 		void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
201 	};
202 };
203 
204 #define TP_UINT_FIELD(bits) \
205 static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
206 { \
207 	u##bits value; \
208 	memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
209 	return value;  \
210 }
211 
212 TP_UINT_FIELD(8);
213 TP_UINT_FIELD(16);
214 TP_UINT_FIELD(32);
215 TP_UINT_FIELD(64);
216 
217 #define TP_UINT_FIELD__SWAPPED(bits) \
218 static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
219 { \
220 	u##bits value; \
221 	memcpy(&value, sample->raw_data + field->offset, sizeof(value)); \
222 	return bswap_##bits(value);\
223 }
224 
225 TP_UINT_FIELD__SWAPPED(16);
226 TP_UINT_FIELD__SWAPPED(32);
227 TP_UINT_FIELD__SWAPPED(64);
228 
229 static int __tp_field__init_uint(struct tp_field *field, int size, int offset, bool needs_swap)
230 {
231 	field->offset = offset;
232 
233 	switch (size) {
234 	case 1:
235 		field->integer = tp_field__u8;
236 		break;
237 	case 2:
238 		field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
239 		break;
240 	case 4:
241 		field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
242 		break;
243 	case 8:
244 		field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
245 		break;
246 	default:
247 		return -1;
248 	}
249 
250 	return 0;
251 }
252 
253 static int tp_field__init_uint(struct tp_field *field, struct tep_format_field *format_field, bool needs_swap)
254 {
255 	return __tp_field__init_uint(field, format_field->size, format_field->offset, needs_swap);
256 }
257 
258 static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
259 {
260 	return sample->raw_data + field->offset;
261 }
262 
263 static int __tp_field__init_ptr(struct tp_field *field, int offset)
264 {
265 	field->offset = offset;
266 	field->pointer = tp_field__ptr;
267 	return 0;
268 }
269 
270 static int tp_field__init_ptr(struct tp_field *field, struct tep_format_field *format_field)
271 {
272 	return __tp_field__init_ptr(field, format_field->offset);
273 }
274 
275 struct syscall_tp {
276 	struct tp_field id;
277 	union {
278 		struct tp_field args, ret;
279 	};
280 };
281 
282 static int perf_evsel__init_tp_uint_field(struct evsel *evsel,
283 					  struct tp_field *field,
284 					  const char *name)
285 {
286 	struct tep_format_field *format_field = perf_evsel__field(evsel, name);
287 
288 	if (format_field == NULL)
289 		return -1;
290 
291 	return tp_field__init_uint(field, format_field, evsel->needs_swap);
292 }
293 
294 #define perf_evsel__init_sc_tp_uint_field(evsel, name) \
295 	({ struct syscall_tp *sc = evsel->priv;\
296 	   perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
297 
298 static int perf_evsel__init_tp_ptr_field(struct evsel *evsel,
299 					 struct tp_field *field,
300 					 const char *name)
301 {
302 	struct tep_format_field *format_field = perf_evsel__field(evsel, name);
303 
304 	if (format_field == NULL)
305 		return -1;
306 
307 	return tp_field__init_ptr(field, format_field);
308 }
309 
310 #define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
311 	({ struct syscall_tp *sc = evsel->priv;\
312 	   perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
313 
314 static void evsel__delete_priv(struct evsel *evsel)
315 {
316 	zfree(&evsel->priv);
317 	evsel__delete(evsel);
318 }
319 
320 static int perf_evsel__init_syscall_tp(struct evsel *evsel)
321 {
322 	struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
323 
324 	if (evsel->priv != NULL) {
325 		if (perf_evsel__init_tp_uint_field(evsel, &sc->id, "__syscall_nr") &&
326 		    perf_evsel__init_tp_uint_field(evsel, &sc->id, "nr"))
327 			goto out_delete;
328 		return 0;
329 	}
330 
331 	return -ENOMEM;
332 out_delete:
333 	zfree(&evsel->priv);
334 	return -ENOENT;
335 }
336 
337 static int perf_evsel__init_augmented_syscall_tp(struct evsel *evsel, struct evsel *tp)
338 {
339 	struct syscall_tp *sc = evsel->priv = malloc(sizeof(struct syscall_tp));
340 
341 	if (evsel->priv != NULL) {
342 		struct tep_format_field *syscall_id = perf_evsel__field(tp, "id");
343 		if (syscall_id == NULL)
344 			syscall_id = perf_evsel__field(tp, "__syscall_nr");
345 		if (syscall_id == NULL)
346 			goto out_delete;
347 		if (__tp_field__init_uint(&sc->id, syscall_id->size, syscall_id->offset, evsel->needs_swap))
348 			goto out_delete;
349 
350 		return 0;
351 	}
352 
353 	return -ENOMEM;
354 out_delete:
355 	zfree(&evsel->priv);
356 	return -EINVAL;
357 }
358 
359 static int perf_evsel__init_augmented_syscall_tp_args(struct evsel *evsel)
360 {
361 	struct syscall_tp *sc = evsel->priv;
362 
363 	return __tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64));
364 }
365 
366 static int perf_evsel__init_augmented_syscall_tp_ret(struct evsel *evsel)
367 {
368 	struct syscall_tp *sc = evsel->priv;
369 
370 	return __tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap);
371 }
372 
373 static int perf_evsel__init_raw_syscall_tp(struct evsel *evsel, void *handler)
374 {
375 	evsel->priv = malloc(sizeof(struct syscall_tp));
376 	if (evsel->priv != NULL) {
377 		if (perf_evsel__init_sc_tp_uint_field(evsel, id))
378 			goto out_delete;
379 
380 		evsel->handler = handler;
381 		return 0;
382 	}
383 
384 	return -ENOMEM;
385 
386 out_delete:
387 	zfree(&evsel->priv);
388 	return -ENOENT;
389 }
390 
391 static struct evsel *perf_evsel__raw_syscall_newtp(const char *direction, void *handler)
392 {
393 	struct evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
394 
395 	/* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
396 	if (IS_ERR(evsel))
397 		evsel = perf_evsel__newtp("syscalls", direction);
398 
399 	if (IS_ERR(evsel))
400 		return NULL;
401 
402 	if (perf_evsel__init_raw_syscall_tp(evsel, handler))
403 		goto out_delete;
404 
405 	return evsel;
406 
407 out_delete:
408 	evsel__delete_priv(evsel);
409 	return NULL;
410 }
411 
412 #define perf_evsel__sc_tp_uint(evsel, name, sample) \
413 	({ struct syscall_tp *fields = evsel->priv; \
414 	   fields->name.integer(&fields->name, sample); })
415 
416 #define perf_evsel__sc_tp_ptr(evsel, name, sample) \
417 	({ struct syscall_tp *fields = evsel->priv; \
418 	   fields->name.pointer(&fields->name, sample); })
419 
420 size_t strarray__scnprintf(struct strarray *sa, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
421 {
422 	int idx = val - sa->offset;
423 
424 	if (idx < 0 || idx >= sa->nr_entries || sa->entries[idx] == NULL) {
425 		size_t printed = scnprintf(bf, size, intfmt, val);
426 		if (show_prefix)
427 			printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sa->prefix);
428 		return printed;
429 	}
430 
431 	return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
432 }
433 
434 static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
435 						const char *intfmt,
436 					        struct syscall_arg *arg)
437 {
438 	return strarray__scnprintf(arg->parm, bf, size, intfmt, arg->show_string_prefix, arg->val);
439 }
440 
441 static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
442 					      struct syscall_arg *arg)
443 {
444 	return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
445 }
446 
447 #define SCA_STRARRAY syscall_arg__scnprintf_strarray
448 
449 size_t syscall_arg__scnprintf_strarray_flags(char *bf, size_t size, struct syscall_arg *arg)
450 {
451 	return strarray__scnprintf_flags(arg->parm, bf, size, arg->show_string_prefix, arg->val);
452 }
453 
454 size_t strarrays__scnprintf(struct strarrays *sas, char *bf, size_t size, const char *intfmt, bool show_prefix, int val)
455 {
456 	size_t printed;
457 	int i;
458 
459 	for (i = 0; i < sas->nr_entries; ++i) {
460 		struct strarray *sa = sas->entries[i];
461 		int idx = val - sa->offset;
462 
463 		if (idx >= 0 && idx < sa->nr_entries) {
464 			if (sa->entries[idx] == NULL)
465 				break;
466 			return scnprintf(bf, size, "%s%s", show_prefix ? sa->prefix : "", sa->entries[idx]);
467 		}
468 	}
469 
470 	printed = scnprintf(bf, size, intfmt, val);
471 	if (show_prefix)
472 		printed += scnprintf(bf + printed, size - printed, " /* %s??? */", sas->entries[0]->prefix);
473 	return printed;
474 }
475 
476 size_t syscall_arg__scnprintf_strarrays(char *bf, size_t size,
477 					struct syscall_arg *arg)
478 {
479 	return strarrays__scnprintf(arg->parm, bf, size, "%d", arg->show_string_prefix, arg->val);
480 }
481 
482 #ifndef AT_FDCWD
483 #define AT_FDCWD	-100
484 #endif
485 
486 static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
487 					   struct syscall_arg *arg)
488 {
489 	int fd = arg->val;
490 	const char *prefix = "AT_FD";
491 
492 	if (fd == AT_FDCWD)
493 		return scnprintf(bf, size, "%s%s", arg->show_string_prefix ? prefix : "", "CWD");
494 
495 	return syscall_arg__scnprintf_fd(bf, size, arg);
496 }
497 
498 #define SCA_FDAT syscall_arg__scnprintf_fd_at
499 
500 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
501 					      struct syscall_arg *arg);
502 
503 #define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
504 
505 size_t syscall_arg__scnprintf_hex(char *bf, size_t size, struct syscall_arg *arg)
506 {
507 	return scnprintf(bf, size, "%#lx", arg->val);
508 }
509 
510 size_t syscall_arg__scnprintf_ptr(char *bf, size_t size, struct syscall_arg *arg)
511 {
512 	if (arg->val == 0)
513 		return scnprintf(bf, size, "NULL");
514 	return syscall_arg__scnprintf_hex(bf, size, arg);
515 }
516 
517 size_t syscall_arg__scnprintf_int(char *bf, size_t size, struct syscall_arg *arg)
518 {
519 	return scnprintf(bf, size, "%d", arg->val);
520 }
521 
522 size_t syscall_arg__scnprintf_long(char *bf, size_t size, struct syscall_arg *arg)
523 {
524 	return scnprintf(bf, size, "%ld", arg->val);
525 }
526 
527 static size_t syscall_arg__scnprintf_char_array(char *bf, size_t size, struct syscall_arg *arg)
528 {
529 	// XXX Hey, maybe for sched:sched_switch prev/next comm fields we can
530 	//     fill missing comms using thread__set_comm()...
531 	//     here or in a special syscall_arg__scnprintf_pid_sched_tp...
532 	return scnprintf(bf, size, "\"%-.*s\"", arg->fmt->nr_entries, arg->val);
533 }
534 
535 #define SCA_CHAR_ARRAY syscall_arg__scnprintf_char_array
536 
537 static const char *bpf_cmd[] = {
538 	"MAP_CREATE", "MAP_LOOKUP_ELEM", "MAP_UPDATE_ELEM", "MAP_DELETE_ELEM",
539 	"MAP_GET_NEXT_KEY", "PROG_LOAD",
540 };
541 static DEFINE_STRARRAY(bpf_cmd, "BPF_");
542 
543 static const char *fsmount_flags[] = {
544 	[1] = "CLOEXEC",
545 };
546 static DEFINE_STRARRAY(fsmount_flags, "FSMOUNT_");
547 
548 #include "trace/beauty/generated/fsconfig_arrays.c"
549 
550 static DEFINE_STRARRAY(fsconfig_cmds, "FSCONFIG_");
551 
552 static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
553 static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, "EPOLL_CTL_", 1);
554 
555 static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
556 static DEFINE_STRARRAY(itimers, "ITIMER_");
557 
558 static const char *keyctl_options[] = {
559 	"GET_KEYRING_ID", "JOIN_SESSION_KEYRING", "UPDATE", "REVOKE", "CHOWN",
560 	"SETPERM", "DESCRIBE", "CLEAR", "LINK", "UNLINK", "SEARCH", "READ",
561 	"INSTANTIATE", "NEGATE", "SET_REQKEY_KEYRING", "SET_TIMEOUT",
562 	"ASSUME_AUTHORITY", "GET_SECURITY", "SESSION_TO_PARENT", "REJECT",
563 	"INSTANTIATE_IOV", "INVALIDATE", "GET_PERSISTENT",
564 };
565 static DEFINE_STRARRAY(keyctl_options, "KEYCTL_");
566 
567 static const char *whences[] = { "SET", "CUR", "END",
568 #ifdef SEEK_DATA
569 "DATA",
570 #endif
571 #ifdef SEEK_HOLE
572 "HOLE",
573 #endif
574 };
575 static DEFINE_STRARRAY(whences, "SEEK_");
576 
577 static const char *fcntl_cmds[] = {
578 	"DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
579 	"SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "GETLK64",
580 	"SETLK64", "SETLKW64", "SETOWN_EX", "GETOWN_EX",
581 	"GETOWNER_UIDS",
582 };
583 static DEFINE_STRARRAY(fcntl_cmds, "F_");
584 
585 static const char *fcntl_linux_specific_cmds[] = {
586 	"SETLEASE", "GETLEASE", "NOTIFY", [5] =	"CANCELLK", "DUPFD_CLOEXEC",
587 	"SETPIPE_SZ", "GETPIPE_SZ", "ADD_SEALS", "GET_SEALS",
588 	"GET_RW_HINT", "SET_RW_HINT", "GET_FILE_RW_HINT", "SET_FILE_RW_HINT",
589 };
590 
591 static DEFINE_STRARRAY_OFFSET(fcntl_linux_specific_cmds, "F_", F_LINUX_SPECIFIC_BASE);
592 
593 static struct strarray *fcntl_cmds_arrays[] = {
594 	&strarray__fcntl_cmds,
595 	&strarray__fcntl_linux_specific_cmds,
596 };
597 
598 static DEFINE_STRARRAYS(fcntl_cmds_arrays);
599 
600 static const char *rlimit_resources[] = {
601 	"CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
602 	"MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
603 	"RTTIME",
604 };
605 static DEFINE_STRARRAY(rlimit_resources, "RLIMIT_");
606 
607 static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
608 static DEFINE_STRARRAY(sighow, "SIG_");
609 
610 static const char *clockid[] = {
611 	"REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
612 	"MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE", "BOOTTIME",
613 	"REALTIME_ALARM", "BOOTTIME_ALARM", "SGI_CYCLE", "TAI"
614 };
615 static DEFINE_STRARRAY(clockid, "CLOCK_");
616 
617 static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
618 						 struct syscall_arg *arg)
619 {
620 	bool show_prefix = arg->show_string_prefix;
621 	const char *suffix = "_OK";
622 	size_t printed = 0;
623 	int mode = arg->val;
624 
625 	if (mode == F_OK) /* 0 */
626 		return scnprintf(bf, size, "F%s", show_prefix ? suffix : "");
627 #define	P_MODE(n) \
628 	if (mode & n##_OK) { \
629 		printed += scnprintf(bf + printed, size - printed, "%s%s", #n, show_prefix ? suffix : ""); \
630 		mode &= ~n##_OK; \
631 	}
632 
633 	P_MODE(R);
634 	P_MODE(W);
635 	P_MODE(X);
636 #undef P_MODE
637 
638 	if (mode)
639 		printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
640 
641 	return printed;
642 }
643 
644 #define SCA_ACCMODE syscall_arg__scnprintf_access_mode
645 
646 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
647 					      struct syscall_arg *arg);
648 
649 #define SCA_FILENAME syscall_arg__scnprintf_filename
650 
651 static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
652 						struct syscall_arg *arg)
653 {
654 	bool show_prefix = arg->show_string_prefix;
655 	const char *prefix = "O_";
656 	int printed = 0, flags = arg->val;
657 
658 #define	P_FLAG(n) \
659 	if (flags & O_##n) { \
660 		printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
661 		flags &= ~O_##n; \
662 	}
663 
664 	P_FLAG(CLOEXEC);
665 	P_FLAG(NONBLOCK);
666 #undef P_FLAG
667 
668 	if (flags)
669 		printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
670 
671 	return printed;
672 }
673 
674 #define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
675 
676 #ifndef GRND_NONBLOCK
677 #define GRND_NONBLOCK	0x0001
678 #endif
679 #ifndef GRND_RANDOM
680 #define GRND_RANDOM	0x0002
681 #endif
682 
683 static size_t syscall_arg__scnprintf_getrandom_flags(char *bf, size_t size,
684 						   struct syscall_arg *arg)
685 {
686 	bool show_prefix = arg->show_string_prefix;
687 	const char *prefix = "GRND_";
688 	int printed = 0, flags = arg->val;
689 
690 #define	P_FLAG(n) \
691 	if (flags & GRND_##n) { \
692 		printed += scnprintf(bf + printed, size - printed, "%s%s%s", printed ? "|" : "", show_prefix ? prefix : "", #n); \
693 		flags &= ~GRND_##n; \
694 	}
695 
696 	P_FLAG(RANDOM);
697 	P_FLAG(NONBLOCK);
698 #undef P_FLAG
699 
700 	if (flags)
701 		printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
702 
703 	return printed;
704 }
705 
706 #define SCA_GETRANDOM_FLAGS syscall_arg__scnprintf_getrandom_flags
707 
708 #define STRARRAY(name, array) \
709 	  { .scnprintf	= SCA_STRARRAY, \
710 	    .parm	= &strarray__##array, }
711 
712 #define STRARRAY_FLAGS(name, array) \
713 	  { .scnprintf	= SCA_STRARRAY_FLAGS, \
714 	    .parm	= &strarray__##array, }
715 
716 #include "trace/beauty/arch_errno_names.c"
717 #include "trace/beauty/eventfd.c"
718 #include "trace/beauty/futex_op.c"
719 #include "trace/beauty/futex_val3.c"
720 #include "trace/beauty/mmap.c"
721 #include "trace/beauty/mode_t.c"
722 #include "trace/beauty/msg_flags.c"
723 #include "trace/beauty/open_flags.c"
724 #include "trace/beauty/perf_event_open.c"
725 #include "trace/beauty/pid.c"
726 #include "trace/beauty/sched_policy.c"
727 #include "trace/beauty/seccomp.c"
728 #include "trace/beauty/signum.c"
729 #include "trace/beauty/socket_type.c"
730 #include "trace/beauty/waitid_options.c"
731 
732 static struct syscall_fmt syscall_fmts[] = {
733 	{ .name	    = "access",
734 	  .arg = { [1] = { .scnprintf = SCA_ACCMODE,  /* mode */ }, }, },
735 	{ .name	    = "arch_prctl",
736 	  .arg = { [0] = { .scnprintf = SCA_X86_ARCH_PRCTL_CODE, /* code */ },
737 		   [1] = { .scnprintf = SCA_PTR, /* arg2 */ }, }, },
738 	{ .name	    = "bind",
739 	  .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
740 		   [1] = { .scnprintf = SCA_SOCKADDR, /* umyaddr */ },
741 		   [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
742 	{ .name	    = "bpf",
743 	  .arg = { [0] = STRARRAY(cmd, bpf_cmd), }, },
744 	{ .name	    = "brk",	    .hexret = true,
745 	  .arg = { [0] = { .scnprintf = SCA_PTR, /* brk */ }, }, },
746 	{ .name     = "clock_gettime",
747 	  .arg = { [0] = STRARRAY(clk_id, clockid), }, },
748 	{ .name	    = "clone",	    .errpid = true, .nr_args = 5,
749 	  .arg = { [0] = { .name = "flags",	    .scnprintf = SCA_CLONE_FLAGS, },
750 		   [1] = { .name = "child_stack",   .scnprintf = SCA_HEX, },
751 		   [2] = { .name = "parent_tidptr", .scnprintf = SCA_HEX, },
752 		   [3] = { .name = "child_tidptr",  .scnprintf = SCA_HEX, },
753 		   [4] = { .name = "tls",	    .scnprintf = SCA_HEX, }, }, },
754 	{ .name	    = "close",
755 	  .arg = { [0] = { .scnprintf = SCA_CLOSE_FD, /* fd */ }, }, },
756 	{ .name	    = "connect",
757 	  .arg = { [0] = { .scnprintf = SCA_INT, /* fd */ },
758 		   [1] = { .scnprintf = SCA_SOCKADDR, /* servaddr */ },
759 		   [2] = { .scnprintf = SCA_INT, /* addrlen */ }, }, },
760 	{ .name	    = "epoll_ctl",
761 	  .arg = { [1] = STRARRAY(op, epoll_ctl_ops), }, },
762 	{ .name	    = "eventfd2",
763 	  .arg = { [1] = { .scnprintf = SCA_EFD_FLAGS, /* flags */ }, }, },
764 	{ .name	    = "fchmodat",
765 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
766 	{ .name	    = "fchownat",
767 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
768 	{ .name	    = "fcntl",
769 	  .arg = { [1] = { .scnprintf = SCA_FCNTL_CMD, /* cmd */
770 			   .parm      = &strarrays__fcntl_cmds_arrays,
771 			   .show_zero = true, },
772 		   [2] = { .scnprintf =  SCA_FCNTL_ARG, /* arg */ }, }, },
773 	{ .name	    = "flock",
774 	  .arg = { [1] = { .scnprintf = SCA_FLOCK, /* cmd */ }, }, },
775 	{ .name     = "fsconfig",
776 	  .arg = { [1] = STRARRAY(cmd, fsconfig_cmds), }, },
777 	{ .name     = "fsmount",
778 	  .arg = { [1] = STRARRAY_FLAGS(flags, fsmount_flags),
779 		   [2] = { .scnprintf = SCA_FSMOUNT_ATTR_FLAGS, /* attr_flags */ }, }, },
780 	{ .name     = "fspick",
781 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	  /* dfd */ },
782 		   [1] = { .scnprintf = SCA_FILENAME,	  /* path */ },
783 		   [2] = { .scnprintf = SCA_FSPICK_FLAGS, /* flags */ }, }, },
784 	{ .name	    = "fstat", .alias = "newfstat", },
785 	{ .name	    = "fstatat", .alias = "newfstatat", },
786 	{ .name	    = "futex",
787 	  .arg = { [1] = { .scnprintf = SCA_FUTEX_OP, /* op */ },
788 		   [5] = { .scnprintf = SCA_FUTEX_VAL3, /* val3 */ }, }, },
789 	{ .name	    = "futimesat",
790 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
791 	{ .name	    = "getitimer",
792 	  .arg = { [0] = STRARRAY(which, itimers), }, },
793 	{ .name	    = "getpid",	    .errpid = true, },
794 	{ .name	    = "getpgid",    .errpid = true, },
795 	{ .name	    = "getppid",    .errpid = true, },
796 	{ .name	    = "getrandom",
797 	  .arg = { [2] = { .scnprintf = SCA_GETRANDOM_FLAGS, /* flags */ }, }, },
798 	{ .name	    = "getrlimit",
799 	  .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
800 	{ .name	    = "gettid",	    .errpid = true, },
801 	{ .name	    = "ioctl",
802 	  .arg = {
803 #if defined(__i386__) || defined(__x86_64__)
804 /*
805  * FIXME: Make this available to all arches.
806  */
807 		   [1] = { .scnprintf = SCA_IOCTL_CMD, /* cmd */ },
808 		   [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
809 #else
810 		   [2] = { .scnprintf = SCA_HEX, /* arg */ }, }, },
811 #endif
812 	{ .name	    = "kcmp",	    .nr_args = 5,
813 	  .arg = { [0] = { .name = "pid1",	.scnprintf = SCA_PID, },
814 		   [1] = { .name = "pid2",	.scnprintf = SCA_PID, },
815 		   [2] = { .name = "type",	.scnprintf = SCA_KCMP_TYPE, },
816 		   [3] = { .name = "idx1",	.scnprintf = SCA_KCMP_IDX, },
817 		   [4] = { .name = "idx2",	.scnprintf = SCA_KCMP_IDX, }, }, },
818 	{ .name	    = "keyctl",
819 	  .arg = { [0] = STRARRAY(option, keyctl_options), }, },
820 	{ .name	    = "kill",
821 	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
822 	{ .name	    = "linkat",
823 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
824 	{ .name	    = "lseek",
825 	  .arg = { [2] = STRARRAY(whence, whences), }, },
826 	{ .name	    = "lstat", .alias = "newlstat", },
827 	{ .name     = "madvise",
828 	  .arg = { [0] = { .scnprintf = SCA_HEX,      /* start */ },
829 		   [2] = { .scnprintf = SCA_MADV_BHV, /* behavior */ }, }, },
830 	{ .name	    = "mkdirat",
831 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
832 	{ .name	    = "mknodat",
833 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* fd */ }, }, },
834 	{ .name	    = "mmap",	    .hexret = true,
835 /* The standard mmap maps to old_mmap on s390x */
836 #if defined(__s390x__)
837 	.alias = "old_mmap",
838 #endif
839 	  .arg = { [2] = { .scnprintf = SCA_MMAP_PROT,	/* prot */ },
840 		   [3] = { .scnprintf = SCA_MMAP_FLAGS,	/* flags */ },
841 		   [5] = { .scnprintf = SCA_HEX,	/* offset */ }, }, },
842 	{ .name	    = "mount",
843 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* dev_name */ },
844 		   [3] = { .scnprintf = SCA_MOUNT_FLAGS, /* flags */
845 			   .mask_val  = SCAMV_MOUNT_FLAGS, /* flags */ }, }, },
846 	{ .name	    = "move_mount",
847 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* from_dfd */ },
848 		   [1] = { .scnprintf = SCA_FILENAME, /* from_pathname */ },
849 		   [2] = { .scnprintf = SCA_FDAT,	/* to_dfd */ },
850 		   [3] = { .scnprintf = SCA_FILENAME, /* to_pathname */ },
851 		   [4] = { .scnprintf = SCA_MOVE_MOUNT_FLAGS, /* flags */ }, }, },
852 	{ .name	    = "mprotect",
853 	  .arg = { [0] = { .scnprintf = SCA_HEX,	/* start */ },
854 		   [2] = { .scnprintf = SCA_MMAP_PROT,	/* prot */ }, }, },
855 	{ .name	    = "mq_unlink",
856 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* u_name */ }, }, },
857 	{ .name	    = "mremap",	    .hexret = true,
858 	  .arg = { [3] = { .scnprintf = SCA_MREMAP_FLAGS, /* flags */ }, }, },
859 	{ .name	    = "name_to_handle_at",
860 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
861 	{ .name	    = "newfstatat",
862 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
863 	{ .name	    = "open",
864 	  .arg = { [1] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
865 	{ .name	    = "open_by_handle_at",
866 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },
867 		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
868 	{ .name	    = "openat",
869 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },
870 		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
871 	{ .name	    = "perf_event_open",
872 	  .arg = { [2] = { .scnprintf = SCA_INT,	/* cpu */ },
873 		   [3] = { .scnprintf = SCA_FD,		/* group_fd */ },
874 		   [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
875 	{ .name	    = "pipe2",
876 	  .arg = { [1] = { .scnprintf = SCA_PIPE_FLAGS, /* flags */ }, }, },
877 	{ .name	    = "pkey_alloc",
878 	  .arg = { [1] = { .scnprintf = SCA_PKEY_ALLOC_ACCESS_RIGHTS,	/* access_rights */ }, }, },
879 	{ .name	    = "pkey_free",
880 	  .arg = { [0] = { .scnprintf = SCA_INT,	/* key */ }, }, },
881 	{ .name	    = "pkey_mprotect",
882 	  .arg = { [0] = { .scnprintf = SCA_HEX,	/* start */ },
883 		   [2] = { .scnprintf = SCA_MMAP_PROT,	/* prot */ },
884 		   [3] = { .scnprintf = SCA_INT,	/* pkey */ }, }, },
885 	{ .name	    = "poll", .timeout = true, },
886 	{ .name	    = "ppoll", .timeout = true, },
887 	{ .name	    = "prctl",
888 	  .arg = { [0] = { .scnprintf = SCA_PRCTL_OPTION, /* option */ },
889 		   [1] = { .scnprintf = SCA_PRCTL_ARG2, /* arg2 */ },
890 		   [2] = { .scnprintf = SCA_PRCTL_ARG3, /* arg3 */ }, }, },
891 	{ .name	    = "pread", .alias = "pread64", },
892 	{ .name	    = "preadv", .alias = "pread", },
893 	{ .name	    = "prlimit64",
894 	  .arg = { [1] = STRARRAY(resource, rlimit_resources), }, },
895 	{ .name	    = "pwrite", .alias = "pwrite64", },
896 	{ .name	    = "readlinkat",
897 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
898 	{ .name	    = "recvfrom",
899 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
900 	{ .name	    = "recvmmsg",
901 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
902 	{ .name	    = "recvmsg",
903 	  .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
904 	{ .name	    = "renameat",
905 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
906 		   [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ }, }, },
907 	{ .name	    = "renameat2",
908 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* olddirfd */ },
909 		   [2] = { .scnprintf = SCA_FDAT, /* newdirfd */ },
910 		   [4] = { .scnprintf = SCA_RENAMEAT2_FLAGS, /* flags */ }, }, },
911 	{ .name	    = "rt_sigaction",
912 	  .arg = { [0] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
913 	{ .name	    = "rt_sigprocmask",
914 	  .arg = { [0] = STRARRAY(how, sighow), }, },
915 	{ .name	    = "rt_sigqueueinfo",
916 	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
917 	{ .name	    = "rt_tgsigqueueinfo",
918 	  .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
919 	{ .name	    = "sched_setscheduler",
920 	  .arg = { [1] = { .scnprintf = SCA_SCHED_POLICY, /* policy */ }, }, },
921 	{ .name	    = "seccomp",
922 	  .arg = { [0] = { .scnprintf = SCA_SECCOMP_OP,	   /* op */ },
923 		   [1] = { .scnprintf = SCA_SECCOMP_FLAGS, /* flags */ }, }, },
924 	{ .name	    = "select", .timeout = true, },
925 	{ .name	    = "sendfile", .alias = "sendfile64", },
926 	{ .name	    = "sendmmsg",
927 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
928 	{ .name	    = "sendmsg",
929 	  .arg = { [2] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ }, }, },
930 	{ .name	    = "sendto",
931 	  .arg = { [3] = { .scnprintf = SCA_MSG_FLAGS, /* flags */ },
932 		   [4] = { .scnprintf = SCA_SOCKADDR, /* addr */ }, }, },
933 	{ .name	    = "set_tid_address", .errpid = true, },
934 	{ .name	    = "setitimer",
935 	  .arg = { [0] = STRARRAY(which, itimers), }, },
936 	{ .name	    = "setrlimit",
937 	  .arg = { [0] = STRARRAY(resource, rlimit_resources), }, },
938 	{ .name	    = "socket",
939 	  .arg = { [0] = STRARRAY(family, socket_families),
940 		   [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
941 		   [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
942 	{ .name	    = "socketpair",
943 	  .arg = { [0] = STRARRAY(family, socket_families),
944 		   [1] = { .scnprintf = SCA_SK_TYPE, /* type */ },
945 		   [2] = { .scnprintf = SCA_SK_PROTO, /* protocol */ }, }, },
946 	{ .name	    = "stat", .alias = "newstat", },
947 	{ .name	    = "statx",
948 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	 /* fdat */ },
949 		   [2] = { .scnprintf = SCA_STATX_FLAGS, /* flags */ } ,
950 		   [3] = { .scnprintf = SCA_STATX_MASK,	 /* mask */ }, }, },
951 	{ .name	    = "swapoff",
952 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
953 	{ .name	    = "swapon",
954 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* specialfile */ }, }, },
955 	{ .name	    = "symlinkat",
956 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
957 	{ .name	    = "sync_file_range",
958 	  .arg = { [3] = { .scnprintf = SCA_SYNC_FILE_RANGE_FLAGS, /* flags */ }, }, },
959 	{ .name	    = "tgkill",
960 	  .arg = { [2] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
961 	{ .name	    = "tkill",
962 	  .arg = { [1] = { .scnprintf = SCA_SIGNUM, /* sig */ }, }, },
963 	{ .name     = "umount2", .alias = "umount",
964 	  .arg = { [0] = { .scnprintf = SCA_FILENAME, /* name */ }, }, },
965 	{ .name	    = "uname", .alias = "newuname", },
966 	{ .name	    = "unlinkat",
967 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dfd */ }, }, },
968 	{ .name	    = "utimensat",
969 	  .arg = { [0] = { .scnprintf = SCA_FDAT, /* dirfd */ }, }, },
970 	{ .name	    = "wait4",	    .errpid = true,
971 	  .arg = { [2] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
972 	{ .name	    = "waitid",	    .errpid = true,
973 	  .arg = { [3] = { .scnprintf = SCA_WAITID_OPTIONS, /* options */ }, }, },
974 };
975 
976 static int syscall_fmt__cmp(const void *name, const void *fmtp)
977 {
978 	const struct syscall_fmt *fmt = fmtp;
979 	return strcmp(name, fmt->name);
980 }
981 
982 static struct syscall_fmt *__syscall_fmt__find(struct syscall_fmt *fmts, const int nmemb, const char *name)
983 {
984 	return bsearch(name, fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
985 }
986 
987 static struct syscall_fmt *syscall_fmt__find(const char *name)
988 {
989 	const int nmemb = ARRAY_SIZE(syscall_fmts);
990 	return __syscall_fmt__find(syscall_fmts, nmemb, name);
991 }
992 
993 static struct syscall_fmt *__syscall_fmt__find_by_alias(struct syscall_fmt *fmts, const int nmemb, const char *alias)
994 {
995 	int i;
996 
997 	for (i = 0; i < nmemb; ++i) {
998 		if (fmts[i].alias && strcmp(fmts[i].alias, alias) == 0)
999 			return &fmts[i];
1000 	}
1001 
1002 	return NULL;
1003 }
1004 
1005 static struct syscall_fmt *syscall_fmt__find_by_alias(const char *alias)
1006 {
1007 	const int nmemb = ARRAY_SIZE(syscall_fmts);
1008 	return __syscall_fmt__find_by_alias(syscall_fmts, nmemb, alias);
1009 }
1010 
1011 /*
1012  * is_exit: is this "exit" or "exit_group"?
1013  * is_open: is this "open" or "openat"? To associate the fd returned in sys_exit with the pathname in sys_enter.
1014  * args_size: sum of the sizes of the syscall arguments, anything after that is augmented stuff: pathname for openat, etc.
1015  * nonexistent: Just a hole in the syscall table, syscall id not allocated
1016  */
1017 struct syscall {
1018 	struct tep_event    *tp_format;
1019 	int		    nr_args;
1020 	int		    args_size;
1021 	struct {
1022 		struct bpf_program *sys_enter,
1023 				   *sys_exit;
1024 	}		    bpf_prog;
1025 	bool		    is_exit;
1026 	bool		    is_open;
1027 	bool		    nonexistent;
1028 	struct tep_format_field *args;
1029 	const char	    *name;
1030 	struct syscall_fmt  *fmt;
1031 	struct syscall_arg_fmt *arg_fmt;
1032 };
1033 
1034 /*
1035  * Must match what is in the BPF program:
1036  *
1037  * tools/perf/examples/bpf/augmented_raw_syscalls.c
1038  */
1039 struct bpf_map_syscall_entry {
1040 	bool	enabled;
1041 	u16	string_args_len[6];
1042 };
1043 
1044 /*
1045  * We need to have this 'calculated' boolean because in some cases we really
1046  * don't know what is the duration of a syscall, for instance, when we start
1047  * a session and some threads are waiting for a syscall to finish, say 'poll',
1048  * in which case all we can do is to print "( ? ) for duration and for the
1049  * start timestamp.
1050  */
1051 static size_t fprintf_duration(unsigned long t, bool calculated, FILE *fp)
1052 {
1053 	double duration = (double)t / NSEC_PER_MSEC;
1054 	size_t printed = fprintf(fp, "(");
1055 
1056 	if (!calculated)
1057 		printed += fprintf(fp, "         ");
1058 	else if (duration >= 1.0)
1059 		printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1060 	else if (duration >= 0.01)
1061 		printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1062 	else
1063 		printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
1064 	return printed + fprintf(fp, "): ");
1065 }
1066 
1067 /**
1068  * filename.ptr: The filename char pointer that will be vfs_getname'd
1069  * filename.entry_str_pos: Where to insert the string translated from
1070  *                         filename.ptr by the vfs_getname tracepoint/kprobe.
1071  * ret_scnprintf: syscall args may set this to a different syscall return
1072  *                formatter, for instance, fcntl may return fds, file flags, etc.
1073  */
1074 struct thread_trace {
1075 	u64		  entry_time;
1076 	bool		  entry_pending;
1077 	unsigned long	  nr_events;
1078 	unsigned long	  pfmaj, pfmin;
1079 	char		  *entry_str;
1080 	double		  runtime_ms;
1081 	size_t		  (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1082         struct {
1083 		unsigned long ptr;
1084 		short int     entry_str_pos;
1085 		bool	      pending_open;
1086 		unsigned int  namelen;
1087 		char	      *name;
1088 	} filename;
1089 	struct {
1090 		int	      max;
1091 		struct file   *table;
1092 	} files;
1093 
1094 	struct intlist *syscall_stats;
1095 };
1096 
1097 static struct thread_trace *thread_trace__new(void)
1098 {
1099 	struct thread_trace *ttrace =  zalloc(sizeof(struct thread_trace));
1100 
1101 	if (ttrace) {
1102 		ttrace->files.max = -1;
1103 		ttrace->syscall_stats = intlist__new(NULL);
1104 	}
1105 
1106 	return ttrace;
1107 }
1108 
1109 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
1110 {
1111 	struct thread_trace *ttrace;
1112 
1113 	if (thread == NULL)
1114 		goto fail;
1115 
1116 	if (thread__priv(thread) == NULL)
1117 		thread__set_priv(thread, thread_trace__new());
1118 
1119 	if (thread__priv(thread) == NULL)
1120 		goto fail;
1121 
1122 	ttrace = thread__priv(thread);
1123 	++ttrace->nr_events;
1124 
1125 	return ttrace;
1126 fail:
1127 	color_fprintf(fp, PERF_COLOR_RED,
1128 		      "WARNING: not enough memory, dropping samples!\n");
1129 	return NULL;
1130 }
1131 
1132 
1133 void syscall_arg__set_ret_scnprintf(struct syscall_arg *arg,
1134 				    size_t (*ret_scnprintf)(char *bf, size_t size, struct syscall_arg *arg))
1135 {
1136 	struct thread_trace *ttrace = thread__priv(arg->thread);
1137 
1138 	ttrace->ret_scnprintf = ret_scnprintf;
1139 }
1140 
1141 #define TRACE_PFMAJ		(1 << 0)
1142 #define TRACE_PFMIN		(1 << 1)
1143 
1144 static const size_t trace__entry_str_size = 2048;
1145 
1146 static struct file *thread_trace__files_entry(struct thread_trace *ttrace, int fd)
1147 {
1148 	if (fd < 0)
1149 		return NULL;
1150 
1151 	if (fd > ttrace->files.max) {
1152 		struct file *nfiles = realloc(ttrace->files.table, (fd + 1) * sizeof(struct file));
1153 
1154 		if (nfiles == NULL)
1155 			return NULL;
1156 
1157 		if (ttrace->files.max != -1) {
1158 			memset(nfiles + ttrace->files.max + 1, 0,
1159 			       (fd - ttrace->files.max) * sizeof(struct file));
1160 		} else {
1161 			memset(nfiles, 0, (fd + 1) * sizeof(struct file));
1162 		}
1163 
1164 		ttrace->files.table = nfiles;
1165 		ttrace->files.max   = fd;
1166 	}
1167 
1168 	return ttrace->files.table + fd;
1169 }
1170 
1171 struct file *thread__files_entry(struct thread *thread, int fd)
1172 {
1173 	return thread_trace__files_entry(thread__priv(thread), fd);
1174 }
1175 
1176 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
1177 {
1178 	struct thread_trace *ttrace = thread__priv(thread);
1179 	struct file *file = thread_trace__files_entry(ttrace, fd);
1180 
1181 	if (file != NULL) {
1182 		struct stat st;
1183 		if (stat(pathname, &st) == 0)
1184 			file->dev_maj = major(st.st_rdev);
1185 		file->pathname = strdup(pathname);
1186 		if (file->pathname)
1187 			return 0;
1188 	}
1189 
1190 	return -1;
1191 }
1192 
1193 static int thread__read_fd_path(struct thread *thread, int fd)
1194 {
1195 	char linkname[PATH_MAX], pathname[PATH_MAX];
1196 	struct stat st;
1197 	int ret;
1198 
1199 	if (thread->pid_ == thread->tid) {
1200 		scnprintf(linkname, sizeof(linkname),
1201 			  "/proc/%d/fd/%d", thread->pid_, fd);
1202 	} else {
1203 		scnprintf(linkname, sizeof(linkname),
1204 			  "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1205 	}
1206 
1207 	if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1208 		return -1;
1209 
1210 	ret = readlink(linkname, pathname, sizeof(pathname));
1211 
1212 	if (ret < 0 || ret > st.st_size)
1213 		return -1;
1214 
1215 	pathname[ret] = '\0';
1216 	return trace__set_fd_pathname(thread, fd, pathname);
1217 }
1218 
1219 static const char *thread__fd_path(struct thread *thread, int fd,
1220 				   struct trace *trace)
1221 {
1222 	struct thread_trace *ttrace = thread__priv(thread);
1223 
1224 	if (ttrace == NULL || trace->fd_path_disabled)
1225 		return NULL;
1226 
1227 	if (fd < 0)
1228 		return NULL;
1229 
1230 	if ((fd > ttrace->files.max || ttrace->files.table[fd].pathname == NULL)) {
1231 		if (!trace->live)
1232 			return NULL;
1233 		++trace->stats.proc_getname;
1234 		if (thread__read_fd_path(thread, fd))
1235 			return NULL;
1236 	}
1237 
1238 	return ttrace->files.table[fd].pathname;
1239 }
1240 
1241 size_t syscall_arg__scnprintf_fd(char *bf, size_t size, struct syscall_arg *arg)
1242 {
1243 	int fd = arg->val;
1244 	size_t printed = scnprintf(bf, size, "%d", fd);
1245 	const char *path = thread__fd_path(arg->thread, fd, arg->trace);
1246 
1247 	if (path)
1248 		printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1249 
1250 	return printed;
1251 }
1252 
1253 size_t pid__scnprintf_fd(struct trace *trace, pid_t pid, int fd, char *bf, size_t size)
1254 {
1255         size_t printed = scnprintf(bf, size, "%d", fd);
1256 	struct thread *thread = machine__find_thread(trace->host, pid, pid);
1257 
1258 	if (thread) {
1259 		const char *path = thread__fd_path(thread, fd, trace);
1260 
1261 		if (path)
1262 			printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1263 
1264 		thread__put(thread);
1265 	}
1266 
1267         return printed;
1268 }
1269 
1270 static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1271 					      struct syscall_arg *arg)
1272 {
1273 	int fd = arg->val;
1274 	size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
1275 	struct thread_trace *ttrace = thread__priv(arg->thread);
1276 
1277 	if (ttrace && fd >= 0 && fd <= ttrace->files.max)
1278 		zfree(&ttrace->files.table[fd].pathname);
1279 
1280 	return printed;
1281 }
1282 
1283 static void thread__set_filename_pos(struct thread *thread, const char *bf,
1284 				     unsigned long ptr)
1285 {
1286 	struct thread_trace *ttrace = thread__priv(thread);
1287 
1288 	ttrace->filename.ptr = ptr;
1289 	ttrace->filename.entry_str_pos = bf - ttrace->entry_str;
1290 }
1291 
1292 static size_t syscall_arg__scnprintf_augmented_string(struct syscall_arg *arg, char *bf, size_t size)
1293 {
1294 	struct augmented_arg *augmented_arg = arg->augmented.args;
1295 	size_t printed = scnprintf(bf, size, "\"%.*s\"", augmented_arg->size, augmented_arg->value);
1296 	/*
1297 	 * So that the next arg with a payload can consume its augmented arg, i.e. for rename* syscalls
1298 	 * we would have two strings, each prefixed by its size.
1299 	 */
1300 	int consumed = sizeof(*augmented_arg) + augmented_arg->size;
1301 
1302 	arg->augmented.args = ((void *)arg->augmented.args) + consumed;
1303 	arg->augmented.size -= consumed;
1304 
1305 	return printed;
1306 }
1307 
1308 static size_t syscall_arg__scnprintf_filename(char *bf, size_t size,
1309 					      struct syscall_arg *arg)
1310 {
1311 	unsigned long ptr = arg->val;
1312 
1313 	if (arg->augmented.args)
1314 		return syscall_arg__scnprintf_augmented_string(arg, bf, size);
1315 
1316 	if (!arg->trace->vfs_getname)
1317 		return scnprintf(bf, size, "%#x", ptr);
1318 
1319 	thread__set_filename_pos(arg->thread, bf, ptr);
1320 	return 0;
1321 }
1322 
1323 static bool trace__filter_duration(struct trace *trace, double t)
1324 {
1325 	return t < (trace->duration_filter * NSEC_PER_MSEC);
1326 }
1327 
1328 static size_t __trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1329 {
1330 	double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1331 
1332 	return fprintf(fp, "%10.3f ", ts);
1333 }
1334 
1335 /*
1336  * We're handling tstamp=0 as an undefined tstamp, i.e. like when we are
1337  * using ttrace->entry_time for a thread that receives a sys_exit without
1338  * first having received a sys_enter ("poll" issued before tracing session
1339  * starts, lost sys_enter exit due to ring buffer overflow).
1340  */
1341 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1342 {
1343 	if (tstamp > 0)
1344 		return __trace__fprintf_tstamp(trace, tstamp, fp);
1345 
1346 	return fprintf(fp, "         ? ");
1347 }
1348 
1349 static bool done = false;
1350 static bool interrupted = false;
1351 
1352 static void sig_handler(int sig)
1353 {
1354 	done = true;
1355 	interrupted = sig == SIGINT;
1356 }
1357 
1358 static size_t trace__fprintf_comm_tid(struct trace *trace, struct thread *thread, FILE *fp)
1359 {
1360 	size_t printed = 0;
1361 
1362 	if (trace->multiple_threads) {
1363 		if (trace->show_comm)
1364 			printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
1365 		printed += fprintf(fp, "%d ", thread->tid);
1366 	}
1367 
1368 	return printed;
1369 }
1370 
1371 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
1372 					u64 duration, bool duration_calculated, u64 tstamp, FILE *fp)
1373 {
1374 	size_t printed = 0;
1375 
1376 	if (trace->show_tstamp)
1377 		printed = trace__fprintf_tstamp(trace, tstamp, fp);
1378 	if (trace->show_duration)
1379 		printed += fprintf_duration(duration, duration_calculated, fp);
1380 	return printed + trace__fprintf_comm_tid(trace, thread, fp);
1381 }
1382 
1383 static int trace__process_event(struct trace *trace, struct machine *machine,
1384 				union perf_event *event, struct perf_sample *sample)
1385 {
1386 	int ret = 0;
1387 
1388 	switch (event->header.type) {
1389 	case PERF_RECORD_LOST:
1390 		color_fprintf(trace->output, PERF_COLOR_RED,
1391 			      "LOST %" PRIu64 " events!\n", event->lost.lost);
1392 		ret = machine__process_lost_event(machine, event, sample);
1393 		break;
1394 	default:
1395 		ret = machine__process_event(machine, event, sample);
1396 		break;
1397 	}
1398 
1399 	return ret;
1400 }
1401 
1402 static int trace__tool_process(struct perf_tool *tool,
1403 			       union perf_event *event,
1404 			       struct perf_sample *sample,
1405 			       struct machine *machine)
1406 {
1407 	struct trace *trace = container_of(tool, struct trace, tool);
1408 	return trace__process_event(trace, machine, event, sample);
1409 }
1410 
1411 static char *trace__machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
1412 {
1413 	struct machine *machine = vmachine;
1414 
1415 	if (machine->kptr_restrict_warned)
1416 		return NULL;
1417 
1418 	if (symbol_conf.kptr_restrict) {
1419 		pr_warning("Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
1420 			   "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
1421 			   "Kernel samples will not be resolved.\n");
1422 		machine->kptr_restrict_warned = true;
1423 		return NULL;
1424 	}
1425 
1426 	return machine__resolve_kernel_addr(vmachine, addrp, modp);
1427 }
1428 
1429 static int trace__symbols_init(struct trace *trace, struct evlist *evlist)
1430 {
1431 	int err = symbol__init(NULL);
1432 
1433 	if (err)
1434 		return err;
1435 
1436 	trace->host = machine__new_host();
1437 	if (trace->host == NULL)
1438 		return -ENOMEM;
1439 
1440 	err = trace_event__register_resolver(trace->host, trace__machine__resolve_kernel_addr);
1441 	if (err < 0)
1442 		goto out;
1443 
1444 	err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1445 					    evlist->core.threads, trace__tool_process, false,
1446 					    1);
1447 out:
1448 	if (err)
1449 		symbol__exit();
1450 
1451 	return err;
1452 }
1453 
1454 static void trace__symbols__exit(struct trace *trace)
1455 {
1456 	machine__exit(trace->host);
1457 	trace->host = NULL;
1458 
1459 	symbol__exit();
1460 }
1461 
1462 static int syscall__alloc_arg_fmts(struct syscall *sc, int nr_args)
1463 {
1464 	int idx;
1465 
1466 	if (nr_args == 6 && sc->fmt && sc->fmt->nr_args != 0)
1467 		nr_args = sc->fmt->nr_args;
1468 
1469 	sc->arg_fmt = calloc(nr_args, sizeof(*sc->arg_fmt));
1470 	if (sc->arg_fmt == NULL)
1471 		return -1;
1472 
1473 	for (idx = 0; idx < nr_args; ++idx) {
1474 		if (sc->fmt)
1475 			sc->arg_fmt[idx] = sc->fmt->arg[idx];
1476 	}
1477 
1478 	sc->nr_args = nr_args;
1479 	return 0;
1480 }
1481 
1482 static struct tep_format_field *
1483 syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field *field)
1484 {
1485 	struct tep_format_field *last_field = NULL;
1486 	int len;
1487 
1488 	for (; field; field = field->next, ++arg) {
1489 		last_field = field;
1490 
1491 		if (arg->scnprintf)
1492 			continue;
1493 
1494 		len = strlen(field->name);
1495 
1496 		if (strcmp(field->type, "const char *") == 0 &&
1497 		    ((len >= 4 && strcmp(field->name + len - 4, "name") == 0) ||
1498 		     strstr(field->name, "path") != NULL))
1499 			arg->scnprintf = SCA_FILENAME;
1500 		else if ((field->flags & TEP_FIELD_IS_POINTER) || strstr(field->name, "addr"))
1501 			arg->scnprintf = SCA_PTR;
1502 		else if (strcmp(field->type, "pid_t") == 0)
1503 			arg->scnprintf = SCA_PID;
1504 		else if (strcmp(field->type, "umode_t") == 0)
1505 			arg->scnprintf = SCA_MODE_T;
1506 		else if ((field->flags & TEP_FIELD_IS_ARRAY) && strstarts(field->type, "char")) {
1507 			arg->scnprintf = SCA_CHAR_ARRAY;
1508 			arg->nr_entries = field->arraylen;
1509 		} else if ((strcmp(field->type, "int") == 0 ||
1510 			  strcmp(field->type, "unsigned int") == 0 ||
1511 			  strcmp(field->type, "long") == 0) &&
1512 			 len >= 2 && strcmp(field->name + len - 2, "fd") == 0) {
1513 			/*
1514 			 * /sys/kernel/tracing/events/syscalls/sys_enter*
1515 			 * egrep 'field:.*fd;' .../format|sed -r 's/.*field:([a-z ]+) [a-z_]*fd.+/\1/g'|sort|uniq -c
1516 			 * 65 int
1517 			 * 23 unsigned int
1518 			 * 7 unsigned long
1519 			 */
1520 			arg->scnprintf = SCA_FD;
1521 		}
1522 	}
1523 
1524 	return last_field;
1525 }
1526 
1527 static int syscall__set_arg_fmts(struct syscall *sc)
1528 {
1529 	struct tep_format_field *last_field = syscall_arg_fmt__init_array(sc->arg_fmt, sc->args);
1530 
1531 	if (last_field)
1532 		sc->args_size = last_field->offset + last_field->size;
1533 
1534 	return 0;
1535 }
1536 
1537 static int trace__read_syscall_info(struct trace *trace, int id)
1538 {
1539 	char tp_name[128];
1540 	struct syscall *sc;
1541 	const char *name = syscalltbl__name(trace->sctbl, id);
1542 
1543 	if (trace->syscalls.table == NULL) {
1544 		trace->syscalls.table = calloc(trace->sctbl->syscalls.max_id + 1, sizeof(*sc));
1545 		if (trace->syscalls.table == NULL)
1546 			return -ENOMEM;
1547 	}
1548 
1549 	sc = trace->syscalls.table + id;
1550 	if (sc->nonexistent)
1551 		return 0;
1552 
1553 	if (name == NULL) {
1554 		sc->nonexistent = true;
1555 		return 0;
1556 	}
1557 
1558 	sc->name = name;
1559 	sc->fmt  = syscall_fmt__find(sc->name);
1560 
1561 	snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
1562 	sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1563 
1564 	if (IS_ERR(sc->tp_format) && sc->fmt && sc->fmt->alias) {
1565 		snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
1566 		sc->tp_format = trace_event__tp_format("syscalls", tp_name);
1567 	}
1568 
1569 	if (syscall__alloc_arg_fmts(sc, IS_ERR(sc->tp_format) ? 6 : sc->tp_format->format.nr_fields))
1570 		return -ENOMEM;
1571 
1572 	if (IS_ERR(sc->tp_format))
1573 		return PTR_ERR(sc->tp_format);
1574 
1575 	sc->args = sc->tp_format->format.fields;
1576 	/*
1577 	 * We need to check and discard the first variable '__syscall_nr'
1578 	 * or 'nr' that mean the syscall number. It is needless here.
1579 	 * So drop '__syscall_nr' or 'nr' field but does not exist on older kernels.
1580 	 */
1581 	if (sc->args && (!strcmp(sc->args->name, "__syscall_nr") || !strcmp(sc->args->name, "nr"))) {
1582 		sc->args = sc->args->next;
1583 		--sc->nr_args;
1584 	}
1585 
1586 	sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1587 	sc->is_open = !strcmp(name, "open") || !strcmp(name, "openat");
1588 
1589 	return syscall__set_arg_fmts(sc);
1590 }
1591 
1592 static int perf_evsel__init_tp_arg_scnprintf(struct evsel *evsel)
1593 {
1594 	int nr_args = evsel->tp_format->format.nr_fields;
1595 
1596 	evsel->priv = calloc(nr_args, sizeof(struct syscall_arg_fmt));
1597 	if (evsel->priv != NULL) {
1598 		syscall_arg_fmt__init_array(evsel->priv, evsel->tp_format->format.fields);
1599 		return 0;
1600 	}
1601 
1602 	return -ENOMEM;
1603 }
1604 
1605 static int intcmp(const void *a, const void *b)
1606 {
1607 	const int *one = a, *another = b;
1608 
1609 	return *one - *another;
1610 }
1611 
1612 static int trace__validate_ev_qualifier(struct trace *trace)
1613 {
1614 	int err = 0;
1615 	bool printed_invalid_prefix = false;
1616 	struct str_node *pos;
1617 	size_t nr_used = 0, nr_allocated = strlist__nr_entries(trace->ev_qualifier);
1618 
1619 	trace->ev_qualifier_ids.entries = malloc(nr_allocated *
1620 						 sizeof(trace->ev_qualifier_ids.entries[0]));
1621 
1622 	if (trace->ev_qualifier_ids.entries == NULL) {
1623 		fputs("Error:\tNot enough memory for allocating events qualifier ids\n",
1624 		       trace->output);
1625 		err = -EINVAL;
1626 		goto out;
1627 	}
1628 
1629 	strlist__for_each_entry(pos, trace->ev_qualifier) {
1630 		const char *sc = pos->s;
1631 		int id = syscalltbl__id(trace->sctbl, sc), match_next = -1;
1632 
1633 		if (id < 0) {
1634 			id = syscalltbl__strglobmatch_first(trace->sctbl, sc, &match_next);
1635 			if (id >= 0)
1636 				goto matches;
1637 
1638 			if (!printed_invalid_prefix) {
1639 				pr_debug("Skipping unknown syscalls: ");
1640 				printed_invalid_prefix = true;
1641 			} else {
1642 				pr_debug(", ");
1643 			}
1644 
1645 			pr_debug("%s", sc);
1646 			continue;
1647 		}
1648 matches:
1649 		trace->ev_qualifier_ids.entries[nr_used++] = id;
1650 		if (match_next == -1)
1651 			continue;
1652 
1653 		while (1) {
1654 			id = syscalltbl__strglobmatch_next(trace->sctbl, sc, &match_next);
1655 			if (id < 0)
1656 				break;
1657 			if (nr_allocated == nr_used) {
1658 				void *entries;
1659 
1660 				nr_allocated += 8;
1661 				entries = realloc(trace->ev_qualifier_ids.entries,
1662 						  nr_allocated * sizeof(trace->ev_qualifier_ids.entries[0]));
1663 				if (entries == NULL) {
1664 					err = -ENOMEM;
1665 					fputs("\nError:\t Not enough memory for parsing\n", trace->output);
1666 					goto out_free;
1667 				}
1668 				trace->ev_qualifier_ids.entries = entries;
1669 			}
1670 			trace->ev_qualifier_ids.entries[nr_used++] = id;
1671 		}
1672 	}
1673 
1674 	trace->ev_qualifier_ids.nr = nr_used;
1675 	qsort(trace->ev_qualifier_ids.entries, nr_used, sizeof(int), intcmp);
1676 out:
1677 	if (printed_invalid_prefix)
1678 		pr_debug("\n");
1679 	return err;
1680 out_free:
1681 	zfree(&trace->ev_qualifier_ids.entries);
1682 	trace->ev_qualifier_ids.nr = 0;
1683 	goto out;
1684 }
1685 
1686 static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
1687 {
1688 	bool in_ev_qualifier;
1689 
1690 	if (trace->ev_qualifier_ids.nr == 0)
1691 		return true;
1692 
1693 	in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
1694 				  trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
1695 
1696 	if (in_ev_qualifier)
1697 	       return !trace->not_ev_qualifier;
1698 
1699 	return trace->not_ev_qualifier;
1700 }
1701 
1702 /*
1703  * args is to be interpreted as a series of longs but we need to handle
1704  * 8-byte unaligned accesses. args points to raw_data within the event
1705  * and raw_data is guaranteed to be 8-byte unaligned because it is
1706  * preceded by raw_size which is a u32. So we need to copy args to a temp
1707  * variable to read it. Most notably this avoids extended load instructions
1708  * on unaligned addresses
1709  */
1710 unsigned long syscall_arg__val(struct syscall_arg *arg, u8 idx)
1711 {
1712 	unsigned long val;
1713 	unsigned char *p = arg->args + sizeof(unsigned long) * idx;
1714 
1715 	memcpy(&val, p, sizeof(val));
1716 	return val;
1717 }
1718 
1719 static size_t syscall__scnprintf_name(struct syscall *sc, char *bf, size_t size,
1720 				      struct syscall_arg *arg)
1721 {
1722 	if (sc->arg_fmt && sc->arg_fmt[arg->idx].name)
1723 		return scnprintf(bf, size, "%s: ", sc->arg_fmt[arg->idx].name);
1724 
1725 	return scnprintf(bf, size, "arg%d: ", arg->idx);
1726 }
1727 
1728 /*
1729  * Check if the value is in fact zero, i.e. mask whatever needs masking, such
1730  * as mount 'flags' argument that needs ignoring some magic flag, see comment
1731  * in tools/perf/trace/beauty/mount_flags.c
1732  */
1733 static unsigned long syscall_arg_fmt__mask_val(struct syscall_arg_fmt *fmt, struct syscall_arg *arg, unsigned long val)
1734 {
1735 	if (fmt && fmt->mask_val)
1736 		return fmt->mask_val(arg, val);
1737 
1738 	return val;
1739 }
1740 
1741 static size_t syscall_arg_fmt__scnprintf_val(struct syscall_arg_fmt *fmt, char *bf, size_t size,
1742 					     struct syscall_arg *arg, unsigned long val)
1743 {
1744 	if (fmt && fmt->scnprintf) {
1745 		arg->val = val;
1746 		if (fmt->parm)
1747 			arg->parm = fmt->parm;
1748 		return fmt->scnprintf(bf, size, arg);
1749 	}
1750 	return scnprintf(bf, size, "%ld", val);
1751 }
1752 
1753 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
1754 				      unsigned char *args, void *augmented_args, int augmented_args_size,
1755 				      struct trace *trace, struct thread *thread)
1756 {
1757 	size_t printed = 0;
1758 	unsigned long val;
1759 	u8 bit = 1;
1760 	struct syscall_arg arg = {
1761 		.args	= args,
1762 		.augmented = {
1763 			.size = augmented_args_size,
1764 			.args = augmented_args,
1765 		},
1766 		.idx	= 0,
1767 		.mask	= 0,
1768 		.trace  = trace,
1769 		.thread = thread,
1770 		.show_string_prefix = trace->show_string_prefix,
1771 	};
1772 	struct thread_trace *ttrace = thread__priv(thread);
1773 
1774 	/*
1775 	 * Things like fcntl will set this in its 'cmd' formatter to pick the
1776 	 * right formatter for the return value (an fd? file flags?), which is
1777 	 * not needed for syscalls that always return a given type, say an fd.
1778 	 */
1779 	ttrace->ret_scnprintf = NULL;
1780 
1781 	if (sc->args != NULL) {
1782 		struct tep_format_field *field;
1783 
1784 		for (field = sc->args; field;
1785 		     field = field->next, ++arg.idx, bit <<= 1) {
1786 			if (arg.mask & bit)
1787 				continue;
1788 
1789 			arg.fmt = &sc->arg_fmt[arg.idx];
1790 			val = syscall_arg__val(&arg, arg.idx);
1791 			/*
1792 			 * Some syscall args need some mask, most don't and
1793 			 * return val untouched.
1794 			 */
1795 			val = syscall_arg_fmt__mask_val(&sc->arg_fmt[arg.idx], &arg, val);
1796 
1797 			/*
1798  			 * Suppress this argument if its value is zero and
1799  			 * and we don't have a string associated in an
1800  			 * strarray for it.
1801  			 */
1802 			if (val == 0 &&
1803 			    !trace->show_zeros &&
1804 			    !(sc->arg_fmt &&
1805 			      (sc->arg_fmt[arg.idx].show_zero ||
1806 			       sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAY ||
1807 			       sc->arg_fmt[arg.idx].scnprintf == SCA_STRARRAYS) &&
1808 			      sc->arg_fmt[arg.idx].parm))
1809 				continue;
1810 
1811 			printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
1812 
1813 			if (trace->show_arg_names)
1814 				printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
1815 
1816 			printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx],
1817 								  bf + printed, size - printed, &arg, val);
1818 		}
1819 	} else if (IS_ERR(sc->tp_format)) {
1820 		/*
1821 		 * If we managed to read the tracepoint /format file, then we
1822 		 * may end up not having any args, like with gettid(), so only
1823 		 * print the raw args when we didn't manage to read it.
1824 		 */
1825 		while (arg.idx < sc->nr_args) {
1826 			if (arg.mask & bit)
1827 				goto next_arg;
1828 			val = syscall_arg__val(&arg, arg.idx);
1829 			if (printed)
1830 				printed += scnprintf(bf + printed, size - printed, ", ");
1831 			printed += syscall__scnprintf_name(sc, bf + printed, size - printed, &arg);
1832 			printed += syscall_arg_fmt__scnprintf_val(&sc->arg_fmt[arg.idx], bf + printed, size - printed, &arg, val);
1833 next_arg:
1834 			++arg.idx;
1835 			bit <<= 1;
1836 		}
1837 	}
1838 
1839 	return printed;
1840 }
1841 
1842 typedef int (*tracepoint_handler)(struct trace *trace, struct evsel *evsel,
1843 				  union perf_event *event,
1844 				  struct perf_sample *sample);
1845 
1846 static struct syscall *trace__syscall_info(struct trace *trace,
1847 					   struct evsel *evsel, int id)
1848 {
1849 	int err = 0;
1850 
1851 	if (id < 0) {
1852 
1853 		/*
1854 		 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1855 		 * before that, leaving at a higher verbosity level till that is
1856 		 * explained. Reproduced with plain ftrace with:
1857 		 *
1858 		 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1859 		 * grep "NR -1 " /t/trace_pipe
1860 		 *
1861 		 * After generating some load on the machine.
1862  		 */
1863 		if (verbose > 1) {
1864 			static u64 n;
1865 			fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1866 				id, perf_evsel__name(evsel), ++n);
1867 		}
1868 		return NULL;
1869 	}
1870 
1871 	err = -EINVAL;
1872 
1873 	if (id > trace->sctbl->syscalls.max_id)
1874 		goto out_cant_read;
1875 
1876 	if ((trace->syscalls.table == NULL || trace->syscalls.table[id].name == NULL) &&
1877 	    (err = trace__read_syscall_info(trace, id)) != 0)
1878 		goto out_cant_read;
1879 
1880 	if (trace->syscalls.table[id].name == NULL) {
1881 		if (trace->syscalls.table[id].nonexistent)
1882 			return NULL;
1883 		goto out_cant_read;
1884 	}
1885 
1886 	return &trace->syscalls.table[id];
1887 
1888 out_cant_read:
1889 	if (verbose > 0) {
1890 		char sbuf[STRERR_BUFSIZE];
1891 		fprintf(trace->output, "Problems reading syscall %d: %d (%s)", id, -err, str_error_r(-err, sbuf, sizeof(sbuf)));
1892 		if (id <= trace->sctbl->syscalls.max_id && trace->syscalls.table[id].name != NULL)
1893 			fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1894 		fputs(" information\n", trace->output);
1895 	}
1896 	return NULL;
1897 }
1898 
1899 static void thread__update_stats(struct thread_trace *ttrace,
1900 				 int id, struct perf_sample *sample)
1901 {
1902 	struct int_node *inode;
1903 	struct stats *stats;
1904 	u64 duration = 0;
1905 
1906 	inode = intlist__findnew(ttrace->syscall_stats, id);
1907 	if (inode == NULL)
1908 		return;
1909 
1910 	stats = inode->priv;
1911 	if (stats == NULL) {
1912 		stats = malloc(sizeof(struct stats));
1913 		if (stats == NULL)
1914 			return;
1915 		init_stats(stats);
1916 		inode->priv = stats;
1917 	}
1918 
1919 	if (ttrace->entry_time && sample->time > ttrace->entry_time)
1920 		duration = sample->time - ttrace->entry_time;
1921 
1922 	update_stats(stats, duration);
1923 }
1924 
1925 static int trace__printf_interrupted_entry(struct trace *trace)
1926 {
1927 	struct thread_trace *ttrace;
1928 	size_t printed;
1929 	int len;
1930 
1931 	if (trace->failure_only || trace->current == NULL)
1932 		return 0;
1933 
1934 	ttrace = thread__priv(trace->current);
1935 
1936 	if (!ttrace->entry_pending)
1937 		return 0;
1938 
1939 	printed  = trace__fprintf_entry_head(trace, trace->current, 0, false, ttrace->entry_time, trace->output);
1940 	printed += len = fprintf(trace->output, "%s)", ttrace->entry_str);
1941 
1942 	if (len < trace->args_alignment - 4)
1943 		printed += fprintf(trace->output, "%-*s", trace->args_alignment - 4 - len, " ");
1944 
1945 	printed += fprintf(trace->output, " ...\n");
1946 
1947 	ttrace->entry_pending = false;
1948 	++trace->nr_events_printed;
1949 
1950 	return printed;
1951 }
1952 
1953 static int trace__fprintf_sample(struct trace *trace, struct evsel *evsel,
1954 				 struct perf_sample *sample, struct thread *thread)
1955 {
1956 	int printed = 0;
1957 
1958 	if (trace->print_sample) {
1959 		double ts = (double)sample->time / NSEC_PER_MSEC;
1960 
1961 		printed += fprintf(trace->output, "%22s %10.3f %s %d/%d [%d]\n",
1962 				   perf_evsel__name(evsel), ts,
1963 				   thread__comm_str(thread),
1964 				   sample->pid, sample->tid, sample->cpu);
1965 	}
1966 
1967 	return printed;
1968 }
1969 
1970 static void *syscall__augmented_args(struct syscall *sc, struct perf_sample *sample, int *augmented_args_size, int raw_augmented_args_size)
1971 {
1972 	void *augmented_args = NULL;
1973 	/*
1974 	 * For now with BPF raw_augmented we hook into raw_syscalls:sys_enter
1975 	 * and there we get all 6 syscall args plus the tracepoint common fields
1976 	 * that gets calculated at the start and the syscall_nr (another long).
1977 	 * So we check if that is the case and if so don't look after the
1978 	 * sc->args_size but always after the full raw_syscalls:sys_enter payload,
1979 	 * which is fixed.
1980 	 *
1981 	 * We'll revisit this later to pass s->args_size to the BPF augmenter
1982 	 * (now tools/perf/examples/bpf/augmented_raw_syscalls.c, so that it
1983 	 * copies only what we need for each syscall, like what happens when we
1984 	 * use syscalls:sys_enter_NAME, so that we reduce the kernel/userspace
1985 	 * traffic to just what is needed for each syscall.
1986 	 */
1987 	int args_size = raw_augmented_args_size ?: sc->args_size;
1988 
1989 	*augmented_args_size = sample->raw_size - args_size;
1990 	if (*augmented_args_size > 0)
1991 		augmented_args = sample->raw_data + args_size;
1992 
1993 	return augmented_args;
1994 }
1995 
1996 static int trace__sys_enter(struct trace *trace, struct evsel *evsel,
1997 			    union perf_event *event __maybe_unused,
1998 			    struct perf_sample *sample)
1999 {
2000 	char *msg;
2001 	void *args;
2002 	int printed = 0;
2003 	struct thread *thread;
2004 	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2005 	int augmented_args_size = 0;
2006 	void *augmented_args = NULL;
2007 	struct syscall *sc = trace__syscall_info(trace, evsel, id);
2008 	struct thread_trace *ttrace;
2009 
2010 	if (sc == NULL)
2011 		return -1;
2012 
2013 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2014 	ttrace = thread__trace(thread, trace->output);
2015 	if (ttrace == NULL)
2016 		goto out_put;
2017 
2018 	trace__fprintf_sample(trace, evsel, sample, thread);
2019 
2020 	args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2021 
2022 	if (ttrace->entry_str == NULL) {
2023 		ttrace->entry_str = malloc(trace__entry_str_size);
2024 		if (!ttrace->entry_str)
2025 			goto out_put;
2026 	}
2027 
2028 	if (!(trace->duration_filter || trace->summary_only || trace->min_stack))
2029 		trace__printf_interrupted_entry(trace);
2030 	/*
2031 	 * If this is raw_syscalls.sys_enter, then it always comes with the 6 possible
2032 	 * arguments, even if the syscall being handled, say "openat", uses only 4 arguments
2033 	 * this breaks syscall__augmented_args() check for augmented args, as we calculate
2034 	 * syscall->args_size using each syscalls:sys_enter_NAME tracefs format file,
2035 	 * so when handling, say the openat syscall, we end up getting 6 args for the
2036 	 * raw_syscalls:sys_enter event, when we expected just 4, we end up mistakenly
2037 	 * thinking that the extra 2 u64 args are the augmented filename, so just check
2038 	 * here and avoid using augmented syscalls when the evsel is the raw_syscalls one.
2039 	 */
2040 	if (evsel != trace->syscalls.events.sys_enter)
2041 		augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2042 	ttrace->entry_time = sample->time;
2043 	msg = ttrace->entry_str;
2044 	printed += scnprintf(msg + printed, trace__entry_str_size - printed, "%s(", sc->name);
2045 
2046 	printed += syscall__scnprintf_args(sc, msg + printed, trace__entry_str_size - printed,
2047 					   args, augmented_args, augmented_args_size, trace, thread);
2048 
2049 	if (sc->is_exit) {
2050 		if (!(trace->duration_filter || trace->summary_only || trace->failure_only || trace->min_stack)) {
2051 			int alignment = 0;
2052 
2053 			trace__fprintf_entry_head(trace, thread, 0, false, ttrace->entry_time, trace->output);
2054 			printed = fprintf(trace->output, "%s)", ttrace->entry_str);
2055 			if (trace->args_alignment > printed)
2056 				alignment = trace->args_alignment - printed;
2057 			fprintf(trace->output, "%*s= ?\n", alignment, " ");
2058 		}
2059 	} else {
2060 		ttrace->entry_pending = true;
2061 		/* See trace__vfs_getname & trace__sys_exit */
2062 		ttrace->filename.pending_open = false;
2063 	}
2064 
2065 	if (trace->current != thread) {
2066 		thread__put(trace->current);
2067 		trace->current = thread__get(thread);
2068 	}
2069 	err = 0;
2070 out_put:
2071 	thread__put(thread);
2072 	return err;
2073 }
2074 
2075 static int trace__fprintf_sys_enter(struct trace *trace, struct evsel *evsel,
2076 				    struct perf_sample *sample)
2077 {
2078 	struct thread_trace *ttrace;
2079 	struct thread *thread;
2080 	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1;
2081 	struct syscall *sc = trace__syscall_info(trace, evsel, id);
2082 	char msg[1024];
2083 	void *args, *augmented_args = NULL;
2084 	int augmented_args_size;
2085 
2086 	if (sc == NULL)
2087 		return -1;
2088 
2089 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2090 	ttrace = thread__trace(thread, trace->output);
2091 	/*
2092 	 * We need to get ttrace just to make sure it is there when syscall__scnprintf_args()
2093 	 * and the rest of the beautifiers accessing it via struct syscall_arg touches it.
2094 	 */
2095 	if (ttrace == NULL)
2096 		goto out_put;
2097 
2098 	args = perf_evsel__sc_tp_ptr(evsel, args, sample);
2099 	augmented_args = syscall__augmented_args(sc, sample, &augmented_args_size, trace->raw_augmented_syscalls_args_size);
2100 	syscall__scnprintf_args(sc, msg, sizeof(msg), args, augmented_args, augmented_args_size, trace, thread);
2101 	fprintf(trace->output, "%s", msg);
2102 	err = 0;
2103 out_put:
2104 	thread__put(thread);
2105 	return err;
2106 }
2107 
2108 static int trace__resolve_callchain(struct trace *trace, struct evsel *evsel,
2109 				    struct perf_sample *sample,
2110 				    struct callchain_cursor *cursor)
2111 {
2112 	struct addr_location al;
2113 	int max_stack = evsel->core.attr.sample_max_stack ?
2114 			evsel->core.attr.sample_max_stack :
2115 			trace->max_stack;
2116 	int err;
2117 
2118 	if (machine__resolve(trace->host, &al, sample) < 0)
2119 		return -1;
2120 
2121 	err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
2122 	addr_location__put(&al);
2123 	return err;
2124 }
2125 
2126 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
2127 {
2128 	/* TODO: user-configurable print_opts */
2129 	const unsigned int print_opts = EVSEL__PRINT_SYM |
2130 				        EVSEL__PRINT_DSO |
2131 				        EVSEL__PRINT_UNKNOWN_AS_ADDR;
2132 
2133 	return sample__fprintf_callchain(sample, 38, print_opts, &callchain_cursor, symbol_conf.bt_stop_list, trace->output);
2134 }
2135 
2136 static const char *errno_to_name(struct evsel *evsel, int err)
2137 {
2138 	struct perf_env *env = perf_evsel__env(evsel);
2139 	const char *arch_name = perf_env__arch(env);
2140 
2141 	return arch_syscalls__strerrno(arch_name, err);
2142 }
2143 
2144 static int trace__sys_exit(struct trace *trace, struct evsel *evsel,
2145 			   union perf_event *event __maybe_unused,
2146 			   struct perf_sample *sample)
2147 {
2148 	long ret;
2149 	u64 duration = 0;
2150 	bool duration_calculated = false;
2151 	struct thread *thread;
2152 	int id = perf_evsel__sc_tp_uint(evsel, id, sample), err = -1, callchain_ret = 0, printed = 0;
2153 	int alignment = trace->args_alignment;
2154 	struct syscall *sc = trace__syscall_info(trace, evsel, id);
2155 	struct thread_trace *ttrace;
2156 
2157 	if (sc == NULL)
2158 		return -1;
2159 
2160 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2161 	ttrace = thread__trace(thread, trace->output);
2162 	if (ttrace == NULL)
2163 		goto out_put;
2164 
2165 	trace__fprintf_sample(trace, evsel, sample, thread);
2166 
2167 	if (trace->summary)
2168 		thread__update_stats(ttrace, id, sample);
2169 
2170 	ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
2171 
2172 	if (!trace->fd_path_disabled && sc->is_open && ret >= 0 && ttrace->filename.pending_open) {
2173 		trace__set_fd_pathname(thread, ret, ttrace->filename.name);
2174 		ttrace->filename.pending_open = false;
2175 		++trace->stats.vfs_getname;
2176 	}
2177 
2178 	if (ttrace->entry_time) {
2179 		duration = sample->time - ttrace->entry_time;
2180 		if (trace__filter_duration(trace, duration))
2181 			goto out;
2182 		duration_calculated = true;
2183 	} else if (trace->duration_filter)
2184 		goto out;
2185 
2186 	if (sample->callchain) {
2187 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2188 		if (callchain_ret == 0) {
2189 			if (callchain_cursor.nr < trace->min_stack)
2190 				goto out;
2191 			callchain_ret = 1;
2192 		}
2193 	}
2194 
2195 	if (trace->summary_only || (ret >= 0 && trace->failure_only))
2196 		goto out;
2197 
2198 	trace__fprintf_entry_head(trace, thread, duration, duration_calculated, ttrace->entry_time, trace->output);
2199 
2200 	if (ttrace->entry_pending) {
2201 		printed = fprintf(trace->output, "%s", ttrace->entry_str);
2202 	} else {
2203 		printed += fprintf(trace->output, " ... [");
2204 		color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
2205 		printed += 9;
2206 		printed += fprintf(trace->output, "]: %s()", sc->name);
2207 	}
2208 
2209 	printed++; /* the closing ')' */
2210 
2211 	if (alignment > printed)
2212 		alignment -= printed;
2213 	else
2214 		alignment = 0;
2215 
2216 	fprintf(trace->output, ")%*s= ", alignment, " ");
2217 
2218 	if (sc->fmt == NULL) {
2219 		if (ret < 0)
2220 			goto errno_print;
2221 signed_print:
2222 		fprintf(trace->output, "%ld", ret);
2223 	} else if (ret < 0) {
2224 errno_print: {
2225 		char bf[STRERR_BUFSIZE];
2226 		const char *emsg = str_error_r(-ret, bf, sizeof(bf)),
2227 			   *e = errno_to_name(evsel, -ret);
2228 
2229 		fprintf(trace->output, "-1 %s (%s)", e, emsg);
2230 	}
2231 	} else if (ret == 0 && sc->fmt->timeout)
2232 		fprintf(trace->output, "0 (Timeout)");
2233 	else if (ttrace->ret_scnprintf) {
2234 		char bf[1024];
2235 		struct syscall_arg arg = {
2236 			.val	= ret,
2237 			.thread	= thread,
2238 			.trace	= trace,
2239 		};
2240 		ttrace->ret_scnprintf(bf, sizeof(bf), &arg);
2241 		ttrace->ret_scnprintf = NULL;
2242 		fprintf(trace->output, "%s", bf);
2243 	} else if (sc->fmt->hexret)
2244 		fprintf(trace->output, "%#lx", ret);
2245 	else if (sc->fmt->errpid) {
2246 		struct thread *child = machine__find_thread(trace->host, ret, ret);
2247 
2248 		if (child != NULL) {
2249 			fprintf(trace->output, "%ld", ret);
2250 			if (child->comm_set)
2251 				fprintf(trace->output, " (%s)", thread__comm_str(child));
2252 			thread__put(child);
2253 		}
2254 	} else
2255 		goto signed_print;
2256 
2257 	fputc('\n', trace->output);
2258 
2259 	/*
2260 	 * We only consider an 'event' for the sake of --max-events a non-filtered
2261 	 * sys_enter + sys_exit and other tracepoint events.
2262 	 */
2263 	if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
2264 		interrupted = true;
2265 
2266 	if (callchain_ret > 0)
2267 		trace__fprintf_callchain(trace, sample);
2268 	else if (callchain_ret < 0)
2269 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2270 out:
2271 	ttrace->entry_pending = false;
2272 	err = 0;
2273 out_put:
2274 	thread__put(thread);
2275 	return err;
2276 }
2277 
2278 static int trace__vfs_getname(struct trace *trace, struct evsel *evsel,
2279 			      union perf_event *event __maybe_unused,
2280 			      struct perf_sample *sample)
2281 {
2282 	struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2283 	struct thread_trace *ttrace;
2284 	size_t filename_len, entry_str_len, to_move;
2285 	ssize_t remaining_space;
2286 	char *pos;
2287 	const char *filename = perf_evsel__rawptr(evsel, sample, "pathname");
2288 
2289 	if (!thread)
2290 		goto out;
2291 
2292 	ttrace = thread__priv(thread);
2293 	if (!ttrace)
2294 		goto out_put;
2295 
2296 	filename_len = strlen(filename);
2297 	if (filename_len == 0)
2298 		goto out_put;
2299 
2300 	if (ttrace->filename.namelen < filename_len) {
2301 		char *f = realloc(ttrace->filename.name, filename_len + 1);
2302 
2303 		if (f == NULL)
2304 			goto out_put;
2305 
2306 		ttrace->filename.namelen = filename_len;
2307 		ttrace->filename.name = f;
2308 	}
2309 
2310 	strcpy(ttrace->filename.name, filename);
2311 	ttrace->filename.pending_open = true;
2312 
2313 	if (!ttrace->filename.ptr)
2314 		goto out_put;
2315 
2316 	entry_str_len = strlen(ttrace->entry_str);
2317 	remaining_space = trace__entry_str_size - entry_str_len - 1; /* \0 */
2318 	if (remaining_space <= 0)
2319 		goto out_put;
2320 
2321 	if (filename_len > (size_t)remaining_space) {
2322 		filename += filename_len - remaining_space;
2323 		filename_len = remaining_space;
2324 	}
2325 
2326 	to_move = entry_str_len - ttrace->filename.entry_str_pos + 1; /* \0 */
2327 	pos = ttrace->entry_str + ttrace->filename.entry_str_pos;
2328 	memmove(pos + filename_len, pos, to_move);
2329 	memcpy(pos, filename, filename_len);
2330 
2331 	ttrace->filename.ptr = 0;
2332 	ttrace->filename.entry_str_pos = 0;
2333 out_put:
2334 	thread__put(thread);
2335 out:
2336 	return 0;
2337 }
2338 
2339 static int trace__sched_stat_runtime(struct trace *trace, struct evsel *evsel,
2340 				     union perf_event *event __maybe_unused,
2341 				     struct perf_sample *sample)
2342 {
2343         u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
2344 	double runtime_ms = (double)runtime / NSEC_PER_MSEC;
2345 	struct thread *thread = machine__findnew_thread(trace->host,
2346 							sample->pid,
2347 							sample->tid);
2348 	struct thread_trace *ttrace = thread__trace(thread, trace->output);
2349 
2350 	if (ttrace == NULL)
2351 		goto out_dump;
2352 
2353 	ttrace->runtime_ms += runtime_ms;
2354 	trace->runtime_ms += runtime_ms;
2355 out_put:
2356 	thread__put(thread);
2357 	return 0;
2358 
2359 out_dump:
2360 	fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
2361 	       evsel->name,
2362 	       perf_evsel__strval(evsel, sample, "comm"),
2363 	       (pid_t)perf_evsel__intval(evsel, sample, "pid"),
2364 	       runtime,
2365 	       perf_evsel__intval(evsel, sample, "vruntime"));
2366 	goto out_put;
2367 }
2368 
2369 static int bpf_output__printer(enum binary_printer_ops op,
2370 			       unsigned int val, void *extra __maybe_unused, FILE *fp)
2371 {
2372 	unsigned char ch = (unsigned char)val;
2373 
2374 	switch (op) {
2375 	case BINARY_PRINT_CHAR_DATA:
2376 		return fprintf(fp, "%c", isprint(ch) ? ch : '.');
2377 	case BINARY_PRINT_DATA_BEGIN:
2378 	case BINARY_PRINT_LINE_BEGIN:
2379 	case BINARY_PRINT_ADDR:
2380 	case BINARY_PRINT_NUM_DATA:
2381 	case BINARY_PRINT_NUM_PAD:
2382 	case BINARY_PRINT_SEP:
2383 	case BINARY_PRINT_CHAR_PAD:
2384 	case BINARY_PRINT_LINE_END:
2385 	case BINARY_PRINT_DATA_END:
2386 	default:
2387 		break;
2388 	}
2389 
2390 	return 0;
2391 }
2392 
2393 static void bpf_output__fprintf(struct trace *trace,
2394 				struct perf_sample *sample)
2395 {
2396 	binary__fprintf(sample->raw_data, sample->raw_size, 8,
2397 			bpf_output__printer, NULL, trace->output);
2398 	++trace->nr_events_printed;
2399 }
2400 
2401 static size_t trace__fprintf_tp_fields(struct trace *trace, struct evsel *evsel, struct perf_sample *sample,
2402 				       struct thread *thread, void *augmented_args, int augmented_args_size)
2403 {
2404 	char bf[2048];
2405 	size_t size = sizeof(bf);
2406 	struct tep_format_field *field = evsel->tp_format->format.fields;
2407 	struct syscall_arg_fmt *arg = evsel->priv;
2408 	size_t printed = 0;
2409 	unsigned long val;
2410 	u8 bit = 1;
2411 	struct syscall_arg syscall_arg = {
2412 		.augmented = {
2413 			.size = augmented_args_size,
2414 			.args = augmented_args,
2415 		},
2416 		.idx	= 0,
2417 		.mask	= 0,
2418 		.trace  = trace,
2419 		.thread = thread,
2420 		.show_string_prefix = trace->show_string_prefix,
2421 	};
2422 
2423 	for (; field && arg; field = field->next, ++syscall_arg.idx, bit <<= 1, ++arg) {
2424 		if (syscall_arg.mask & bit)
2425 			continue;
2426 
2427 		syscall_arg.fmt = arg;
2428 		if (field->flags & TEP_FIELD_IS_ARRAY)
2429 			val = (uintptr_t)(sample->raw_data + field->offset);
2430 		else
2431 			val = format_field__intval(field, sample, evsel->needs_swap);
2432 		/*
2433 		 * Some syscall args need some mask, most don't and
2434 		 * return val untouched.
2435 		 */
2436 		val = syscall_arg_fmt__mask_val(arg, &syscall_arg, val);
2437 
2438 		/*
2439 		 * Suppress this argument if its value is zero and
2440 		 * and we don't have a string associated in an
2441 		 * strarray for it.
2442 		 */
2443 		if (val == 0 &&
2444 		    !trace->show_zeros &&
2445 		    !((arg->show_zero ||
2446 		       arg->scnprintf == SCA_STRARRAY ||
2447 		       arg->scnprintf == SCA_STRARRAYS) &&
2448 		      arg->parm))
2449 			continue;
2450 
2451 		printed += scnprintf(bf + printed, size - printed, "%s", printed ? ", " : "");
2452 
2453 		/*
2454 		 * XXX Perhaps we should have a show_tp_arg_names,
2455 		 * leaving show_arg_names just for syscalls?
2456 		 */
2457 		if (1 || trace->show_arg_names)
2458 			printed += scnprintf(bf + printed, size - printed, "%s: ", field->name);
2459 
2460 		printed += syscall_arg_fmt__scnprintf_val(arg, bf + printed, size - printed, &syscall_arg, val);
2461 	}
2462 
2463 	return printed + fprintf(trace->output, "%s", bf);
2464 }
2465 
2466 static int trace__event_handler(struct trace *trace, struct evsel *evsel,
2467 				union perf_event *event __maybe_unused,
2468 				struct perf_sample *sample)
2469 {
2470 	struct thread *thread;
2471 	int callchain_ret = 0;
2472 	/*
2473 	 * Check if we called perf_evsel__disable(evsel) due to, for instance,
2474 	 * this event's max_events having been hit and this is an entry coming
2475 	 * from the ring buffer that we should discard, since the max events
2476 	 * have already been considered/printed.
2477 	 */
2478 	if (evsel->disabled)
2479 		return 0;
2480 
2481 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2482 
2483 	if (sample->callchain) {
2484 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2485 		if (callchain_ret == 0) {
2486 			if (callchain_cursor.nr < trace->min_stack)
2487 				goto out;
2488 			callchain_ret = 1;
2489 		}
2490 	}
2491 
2492 	trace__printf_interrupted_entry(trace);
2493 	trace__fprintf_tstamp(trace, sample->time, trace->output);
2494 
2495 	if (trace->trace_syscalls && trace->show_duration)
2496 		fprintf(trace->output, "(         ): ");
2497 
2498 	if (thread)
2499 		trace__fprintf_comm_tid(trace, thread, trace->output);
2500 
2501 	if (evsel == trace->syscalls.events.augmented) {
2502 		int id = perf_evsel__sc_tp_uint(evsel, id, sample);
2503 		struct syscall *sc = trace__syscall_info(trace, evsel, id);
2504 
2505 		if (sc) {
2506 			fprintf(trace->output, "%s(", sc->name);
2507 			trace__fprintf_sys_enter(trace, evsel, sample);
2508 			fputc(')', trace->output);
2509 			goto newline;
2510 		}
2511 
2512 		/*
2513 		 * XXX: Not having the associated syscall info or not finding/adding
2514 		 * 	the thread should never happen, but if it does...
2515 		 * 	fall thru and print it as a bpf_output event.
2516 		 */
2517 	}
2518 
2519 	fprintf(trace->output, "%s(", evsel->name);
2520 
2521 	if (perf_evsel__is_bpf_output(evsel)) {
2522 		bpf_output__fprintf(trace, sample);
2523 	} else if (evsel->tp_format) {
2524 		if (strncmp(evsel->tp_format->name, "sys_enter_", 10) ||
2525 		    trace__fprintf_sys_enter(trace, evsel, sample)) {
2526 			if (trace->libtraceevent_print) {
2527 				event_format__fprintf(evsel->tp_format, sample->cpu,
2528 						      sample->raw_data, sample->raw_size,
2529 						      trace->output);
2530 			} else {
2531 				trace__fprintf_tp_fields(trace, evsel, sample, thread, NULL, 0);
2532 			}
2533 			++trace->nr_events_printed;
2534 
2535 			if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
2536 				evsel__disable(evsel);
2537 				evsel__close(evsel);
2538 			}
2539 		}
2540 	}
2541 
2542 newline:
2543 	fprintf(trace->output, ")\n");
2544 
2545 	if (callchain_ret > 0)
2546 		trace__fprintf_callchain(trace, sample);
2547 	else if (callchain_ret < 0)
2548 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2549 out:
2550 	thread__put(thread);
2551 	return 0;
2552 }
2553 
2554 static void print_location(FILE *f, struct perf_sample *sample,
2555 			   struct addr_location *al,
2556 			   bool print_dso, bool print_sym)
2557 {
2558 
2559 	if ((verbose > 0 || print_dso) && al->map)
2560 		fprintf(f, "%s@", al->map->dso->long_name);
2561 
2562 	if ((verbose > 0 || print_sym) && al->sym)
2563 		fprintf(f, "%s+0x%" PRIx64, al->sym->name,
2564 			al->addr - al->sym->start);
2565 	else if (al->map)
2566 		fprintf(f, "0x%" PRIx64, al->addr);
2567 	else
2568 		fprintf(f, "0x%" PRIx64, sample->addr);
2569 }
2570 
2571 static int trace__pgfault(struct trace *trace,
2572 			  struct evsel *evsel,
2573 			  union perf_event *event __maybe_unused,
2574 			  struct perf_sample *sample)
2575 {
2576 	struct thread *thread;
2577 	struct addr_location al;
2578 	char map_type = 'd';
2579 	struct thread_trace *ttrace;
2580 	int err = -1;
2581 	int callchain_ret = 0;
2582 
2583 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2584 
2585 	if (sample->callchain) {
2586 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
2587 		if (callchain_ret == 0) {
2588 			if (callchain_cursor.nr < trace->min_stack)
2589 				goto out_put;
2590 			callchain_ret = 1;
2591 		}
2592 	}
2593 
2594 	ttrace = thread__trace(thread, trace->output);
2595 	if (ttrace == NULL)
2596 		goto out_put;
2597 
2598 	if (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
2599 		ttrace->pfmaj++;
2600 	else
2601 		ttrace->pfmin++;
2602 
2603 	if (trace->summary_only)
2604 		goto out;
2605 
2606 	thread__find_symbol(thread, sample->cpumode, sample->ip, &al);
2607 
2608 	trace__fprintf_entry_head(trace, thread, 0, true, sample->time, trace->output);
2609 
2610 	fprintf(trace->output, "%sfault [",
2611 		evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
2612 		"maj" : "min");
2613 
2614 	print_location(trace->output, sample, &al, false, true);
2615 
2616 	fprintf(trace->output, "] => ");
2617 
2618 	thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2619 
2620 	if (!al.map) {
2621 		thread__find_symbol(thread, sample->cpumode, sample->addr, &al);
2622 
2623 		if (al.map)
2624 			map_type = 'x';
2625 		else
2626 			map_type = '?';
2627 	}
2628 
2629 	print_location(trace->output, sample, &al, true, false);
2630 
2631 	fprintf(trace->output, " (%c%c)\n", map_type, al.level);
2632 
2633 	if (callchain_ret > 0)
2634 		trace__fprintf_callchain(trace, sample);
2635 	else if (callchain_ret < 0)
2636 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
2637 
2638 	++trace->nr_events_printed;
2639 out:
2640 	err = 0;
2641 out_put:
2642 	thread__put(thread);
2643 	return err;
2644 }
2645 
2646 static void trace__set_base_time(struct trace *trace,
2647 				 struct evsel *evsel,
2648 				 struct perf_sample *sample)
2649 {
2650 	/*
2651 	 * BPF events were not setting PERF_SAMPLE_TIME, so be more robust
2652 	 * and don't use sample->time unconditionally, we may end up having
2653 	 * some other event in the future without PERF_SAMPLE_TIME for good
2654 	 * reason, i.e. we may not be interested in its timestamps, just in
2655 	 * it taking place, picking some piece of information when it
2656 	 * appears in our event stream (vfs_getname comes to mind).
2657 	 */
2658 	if (trace->base_time == 0 && !trace->full_time &&
2659 	    (evsel->core.attr.sample_type & PERF_SAMPLE_TIME))
2660 		trace->base_time = sample->time;
2661 }
2662 
2663 static int trace__process_sample(struct perf_tool *tool,
2664 				 union perf_event *event,
2665 				 struct perf_sample *sample,
2666 				 struct evsel *evsel,
2667 				 struct machine *machine __maybe_unused)
2668 {
2669 	struct trace *trace = container_of(tool, struct trace, tool);
2670 	struct thread *thread;
2671 	int err = 0;
2672 
2673 	tracepoint_handler handler = evsel->handler;
2674 
2675 	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
2676 	if (thread && thread__is_filtered(thread))
2677 		goto out;
2678 
2679 	trace__set_base_time(trace, evsel, sample);
2680 
2681 	if (handler) {
2682 		++trace->nr_events;
2683 		handler(trace, evsel, event, sample);
2684 	}
2685 out:
2686 	thread__put(thread);
2687 	return err;
2688 }
2689 
2690 static int trace__record(struct trace *trace, int argc, const char **argv)
2691 {
2692 	unsigned int rec_argc, i, j;
2693 	const char **rec_argv;
2694 	const char * const record_args[] = {
2695 		"record",
2696 		"-R",
2697 		"-m", "1024",
2698 		"-c", "1",
2699 	};
2700 
2701 	const char * const sc_args[] = { "-e", };
2702 	unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
2703 	const char * const majpf_args[] = { "-e", "major-faults" };
2704 	unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
2705 	const char * const minpf_args[] = { "-e", "minor-faults" };
2706 	unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
2707 
2708 	/* +1 is for the event string below */
2709 	rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
2710 		majpf_args_nr + minpf_args_nr + argc;
2711 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
2712 
2713 	if (rec_argv == NULL)
2714 		return -ENOMEM;
2715 
2716 	j = 0;
2717 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
2718 		rec_argv[j++] = record_args[i];
2719 
2720 	if (trace->trace_syscalls) {
2721 		for (i = 0; i < sc_args_nr; i++)
2722 			rec_argv[j++] = sc_args[i];
2723 
2724 		/* event string may be different for older kernels - e.g., RHEL6 */
2725 		if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2726 			rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2727 		else if (is_valid_tracepoint("syscalls:sys_enter"))
2728 			rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2729 		else {
2730 			pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2731 			free(rec_argv);
2732 			return -1;
2733 		}
2734 	}
2735 
2736 	if (trace->trace_pgfaults & TRACE_PFMAJ)
2737 		for (i = 0; i < majpf_args_nr; i++)
2738 			rec_argv[j++] = majpf_args[i];
2739 
2740 	if (trace->trace_pgfaults & TRACE_PFMIN)
2741 		for (i = 0; i < minpf_args_nr; i++)
2742 			rec_argv[j++] = minpf_args[i];
2743 
2744 	for (i = 0; i < (unsigned int)argc; i++)
2745 		rec_argv[j++] = argv[i];
2746 
2747 	return cmd_record(j, rec_argv);
2748 }
2749 
2750 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2751 
2752 static bool evlist__add_vfs_getname(struct evlist *evlist)
2753 {
2754 	bool found = false;
2755 	struct evsel *evsel, *tmp;
2756 	struct parse_events_error err = { .idx = 0, };
2757 	int ret = parse_events(evlist, "probe:vfs_getname*", &err);
2758 
2759 	if (ret)
2760 		return false;
2761 
2762 	evlist__for_each_entry_safe(evlist, evsel, tmp) {
2763 		if (!strstarts(perf_evsel__name(evsel), "probe:vfs_getname"))
2764 			continue;
2765 
2766 		if (perf_evsel__field(evsel, "pathname")) {
2767 			evsel->handler = trace__vfs_getname;
2768 			found = true;
2769 			continue;
2770 		}
2771 
2772 		list_del_init(&evsel->core.node);
2773 		evsel->evlist = NULL;
2774 		evsel__delete(evsel);
2775 	}
2776 
2777 	return found;
2778 }
2779 
2780 static struct evsel *perf_evsel__new_pgfault(u64 config)
2781 {
2782 	struct evsel *evsel;
2783 	struct perf_event_attr attr = {
2784 		.type = PERF_TYPE_SOFTWARE,
2785 		.mmap_data = 1,
2786 	};
2787 
2788 	attr.config = config;
2789 	attr.sample_period = 1;
2790 
2791 	event_attr_init(&attr);
2792 
2793 	evsel = evsel__new(&attr);
2794 	if (evsel)
2795 		evsel->handler = trace__pgfault;
2796 
2797 	return evsel;
2798 }
2799 
2800 static void trace__handle_event(struct trace *trace, union perf_event *event, struct perf_sample *sample)
2801 {
2802 	const u32 type = event->header.type;
2803 	struct evsel *evsel;
2804 
2805 	if (type != PERF_RECORD_SAMPLE) {
2806 		trace__process_event(trace, trace->host, event, sample);
2807 		return;
2808 	}
2809 
2810 	evsel = perf_evlist__id2evsel(trace->evlist, sample->id);
2811 	if (evsel == NULL) {
2812 		fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample->id);
2813 		return;
2814 	}
2815 
2816 	if (evswitch__discard(&trace->evswitch, evsel))
2817 		return;
2818 
2819 	trace__set_base_time(trace, evsel, sample);
2820 
2821 	if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT &&
2822 	    sample->raw_data == NULL) {
2823 		fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
2824 		       perf_evsel__name(evsel), sample->tid,
2825 		       sample->cpu, sample->raw_size);
2826 	} else {
2827 		tracepoint_handler handler = evsel->handler;
2828 		handler(trace, evsel, event, sample);
2829 	}
2830 
2831 	if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
2832 		interrupted = true;
2833 }
2834 
2835 static int trace__add_syscall_newtp(struct trace *trace)
2836 {
2837 	int ret = -1;
2838 	struct evlist *evlist = trace->evlist;
2839 	struct evsel *sys_enter, *sys_exit;
2840 
2841 	sys_enter = perf_evsel__raw_syscall_newtp("sys_enter", trace__sys_enter);
2842 	if (sys_enter == NULL)
2843 		goto out;
2844 
2845 	if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
2846 		goto out_delete_sys_enter;
2847 
2848 	sys_exit = perf_evsel__raw_syscall_newtp("sys_exit", trace__sys_exit);
2849 	if (sys_exit == NULL)
2850 		goto out_delete_sys_enter;
2851 
2852 	if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
2853 		goto out_delete_sys_exit;
2854 
2855 	perf_evsel__config_callchain(sys_enter, &trace->opts, &callchain_param);
2856 	perf_evsel__config_callchain(sys_exit, &trace->opts, &callchain_param);
2857 
2858 	evlist__add(evlist, sys_enter);
2859 	evlist__add(evlist, sys_exit);
2860 
2861 	if (callchain_param.enabled && !trace->kernel_syscallchains) {
2862 		/*
2863 		 * We're interested only in the user space callchain
2864 		 * leading to the syscall, allow overriding that for
2865 		 * debugging reasons using --kernel_syscall_callchains
2866 		 */
2867 		sys_exit->core.attr.exclude_callchain_kernel = 1;
2868 	}
2869 
2870 	trace->syscalls.events.sys_enter = sys_enter;
2871 	trace->syscalls.events.sys_exit  = sys_exit;
2872 
2873 	ret = 0;
2874 out:
2875 	return ret;
2876 
2877 out_delete_sys_exit:
2878 	evsel__delete_priv(sys_exit);
2879 out_delete_sys_enter:
2880 	evsel__delete_priv(sys_enter);
2881 	goto out;
2882 }
2883 
2884 static int trace__set_ev_qualifier_tp_filter(struct trace *trace)
2885 {
2886 	int err = -1;
2887 	struct evsel *sys_exit;
2888 	char *filter = asprintf_expr_inout_ints("id", !trace->not_ev_qualifier,
2889 						trace->ev_qualifier_ids.nr,
2890 						trace->ev_qualifier_ids.entries);
2891 
2892 	if (filter == NULL)
2893 		goto out_enomem;
2894 
2895 	if (!perf_evsel__append_tp_filter(trace->syscalls.events.sys_enter,
2896 					  filter)) {
2897 		sys_exit = trace->syscalls.events.sys_exit;
2898 		err = perf_evsel__append_tp_filter(sys_exit, filter);
2899 	}
2900 
2901 	free(filter);
2902 out:
2903 	return err;
2904 out_enomem:
2905 	errno = ENOMEM;
2906 	goto out;
2907 }
2908 
2909 #ifdef HAVE_LIBBPF_SUPPORT
2910 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace, const char *name)
2911 {
2912 	if (trace->bpf_obj == NULL)
2913 		return NULL;
2914 
2915 	return bpf_object__find_program_by_title(trace->bpf_obj, name);
2916 }
2917 
2918 static struct bpf_program *trace__find_syscall_bpf_prog(struct trace *trace, struct syscall *sc,
2919 							const char *prog_name, const char *type)
2920 {
2921 	struct bpf_program *prog;
2922 
2923 	if (prog_name == NULL) {
2924 		char default_prog_name[256];
2925 		scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->name);
2926 		prog = trace__find_bpf_program_by_title(trace, default_prog_name);
2927 		if (prog != NULL)
2928 			goto out_found;
2929 		if (sc->fmt && sc->fmt->alias) {
2930 			scnprintf(default_prog_name, sizeof(default_prog_name), "!syscalls:sys_%s_%s", type, sc->fmt->alias);
2931 			prog = trace__find_bpf_program_by_title(trace, default_prog_name);
2932 			if (prog != NULL)
2933 				goto out_found;
2934 		}
2935 		goto out_unaugmented;
2936 	}
2937 
2938 	prog = trace__find_bpf_program_by_title(trace, prog_name);
2939 
2940 	if (prog != NULL) {
2941 out_found:
2942 		return prog;
2943 	}
2944 
2945 	pr_debug("Couldn't find BPF prog \"%s\" to associate with syscalls:sys_%s_%s, not augmenting it\n",
2946 		 prog_name, type, sc->name);
2947 out_unaugmented:
2948 	return trace->syscalls.unaugmented_prog;
2949 }
2950 
2951 static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
2952 {
2953 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2954 
2955 	if (sc == NULL)
2956 		return;
2957 
2958 	sc->bpf_prog.sys_enter = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_enter : NULL, "enter");
2959 	sc->bpf_prog.sys_exit  = trace__find_syscall_bpf_prog(trace, sc, sc->fmt ? sc->fmt->bpf_prog_name.sys_exit  : NULL,  "exit");
2960 }
2961 
2962 static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
2963 {
2964 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2965 	return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
2966 }
2967 
2968 static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
2969 {
2970 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2971 	return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
2972 }
2973 
2974 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
2975 {
2976 	struct syscall *sc = trace__syscall_info(trace, NULL, id);
2977 	int arg = 0;
2978 
2979 	if (sc == NULL)
2980 		goto out;
2981 
2982 	for (; arg < sc->nr_args; ++arg) {
2983 		entry->string_args_len[arg] = 0;
2984 		if (sc->arg_fmt[arg].scnprintf == SCA_FILENAME) {
2985 			/* Should be set like strace -s strsize */
2986 			entry->string_args_len[arg] = PATH_MAX;
2987 		}
2988 	}
2989 out:
2990 	for (; arg < 6; ++arg)
2991 		entry->string_args_len[arg] = 0;
2992 }
2993 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace)
2994 {
2995 	int fd = bpf_map__fd(trace->syscalls.map);
2996 	struct bpf_map_syscall_entry value = {
2997 		.enabled = !trace->not_ev_qualifier,
2998 	};
2999 	int err = 0;
3000 	size_t i;
3001 
3002 	for (i = 0; i < trace->ev_qualifier_ids.nr; ++i) {
3003 		int key = trace->ev_qualifier_ids.entries[i];
3004 
3005 		if (value.enabled) {
3006 			trace__init_bpf_map_syscall_args(trace, key, &value);
3007 			trace__init_syscall_bpf_progs(trace, key);
3008 		}
3009 
3010 		err = bpf_map_update_elem(fd, &key, &value, BPF_EXIST);
3011 		if (err)
3012 			break;
3013 	}
3014 
3015 	return err;
3016 }
3017 
3018 static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
3019 {
3020 	int fd = bpf_map__fd(trace->syscalls.map);
3021 	struct bpf_map_syscall_entry value = {
3022 		.enabled = enabled,
3023 	};
3024 	int err = 0, key;
3025 
3026 	for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3027 		if (enabled)
3028 			trace__init_bpf_map_syscall_args(trace, key, &value);
3029 
3030 		err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
3031 		if (err)
3032 			break;
3033 	}
3034 
3035 	return err;
3036 }
3037 
3038 static int trace__init_syscalls_bpf_map(struct trace *trace)
3039 {
3040 	bool enabled = true;
3041 
3042 	if (trace->ev_qualifier_ids.nr)
3043 		enabled = trace->not_ev_qualifier;
3044 
3045 	return __trace__init_syscalls_bpf_map(trace, enabled);
3046 }
3047 
3048 static struct bpf_program *trace__find_usable_bpf_prog_entry(struct trace *trace, struct syscall *sc)
3049 {
3050 	struct tep_format_field *field, *candidate_field;
3051 	int id;
3052 
3053 	/*
3054 	 * We're only interested in syscalls that have a pointer:
3055 	 */
3056 	for (field = sc->args; field; field = field->next) {
3057 		if (field->flags & TEP_FIELD_IS_POINTER)
3058 			goto try_to_find_pair;
3059 	}
3060 
3061 	return NULL;
3062 
3063 try_to_find_pair:
3064 	for (id = 0; id < trace->sctbl->syscalls.nr_entries; ++id) {
3065 		struct syscall *pair = trace__syscall_info(trace, NULL, id);
3066 		struct bpf_program *pair_prog;
3067 		bool is_candidate = false;
3068 
3069 		if (pair == NULL || pair == sc ||
3070 		    pair->bpf_prog.sys_enter == trace->syscalls.unaugmented_prog)
3071 			continue;
3072 
3073 		for (field = sc->args, candidate_field = pair->args;
3074 		     field && candidate_field; field = field->next, candidate_field = candidate_field->next) {
3075 			bool is_pointer = field->flags & TEP_FIELD_IS_POINTER,
3076 			     candidate_is_pointer = candidate_field->flags & TEP_FIELD_IS_POINTER;
3077 
3078 			if (is_pointer) {
3079 			       if (!candidate_is_pointer) {
3080 					// The candidate just doesn't copies our pointer arg, might copy other pointers we want.
3081 					continue;
3082 			       }
3083 			} else {
3084 				if (candidate_is_pointer) {
3085 					// The candidate might copy a pointer we don't have, skip it.
3086 					goto next_candidate;
3087 				}
3088 				continue;
3089 			}
3090 
3091 			if (strcmp(field->type, candidate_field->type))
3092 				goto next_candidate;
3093 
3094 			is_candidate = true;
3095 		}
3096 
3097 		if (!is_candidate)
3098 			goto next_candidate;
3099 
3100 		/*
3101 		 * Check if the tentative pair syscall augmenter has more pointers, if it has,
3102 		 * then it may be collecting that and we then can't use it, as it would collect
3103 		 * more than what is common to the two syscalls.
3104 		 */
3105 		if (candidate_field) {
3106 			for (candidate_field = candidate_field->next; candidate_field; candidate_field = candidate_field->next)
3107 				if (candidate_field->flags & TEP_FIELD_IS_POINTER)
3108 					goto next_candidate;
3109 		}
3110 
3111 		pair_prog = pair->bpf_prog.sys_enter;
3112 		/*
3113 		 * If the pair isn't enabled, then its bpf_prog.sys_enter will not
3114 		 * have been searched for, so search it here and if it returns the
3115 		 * unaugmented one, then ignore it, otherwise we'll reuse that BPF
3116 		 * program for a filtered syscall on a non-filtered one.
3117 		 *
3118 		 * For instance, we have "!syscalls:sys_enter_renameat" and that is
3119 		 * useful for "renameat2".
3120 		 */
3121 		if (pair_prog == NULL) {
3122 			pair_prog = trace__find_syscall_bpf_prog(trace, pair, pair->fmt ? pair->fmt->bpf_prog_name.sys_enter : NULL, "enter");
3123 			if (pair_prog == trace->syscalls.unaugmented_prog)
3124 				goto next_candidate;
3125 		}
3126 
3127 		pr_debug("Reusing \"%s\" BPF sys_enter augmenter for \"%s\"\n", pair->name, sc->name);
3128 		return pair_prog;
3129 	next_candidate:
3130 		continue;
3131 	}
3132 
3133 	return NULL;
3134 }
3135 
3136 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
3137 {
3138 	int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
3139 	    map_exit_fd  = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
3140 	int err = 0, key;
3141 
3142 	for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3143 		int prog_fd;
3144 
3145 		if (!trace__syscall_enabled(trace, key))
3146 			continue;
3147 
3148 		trace__init_syscall_bpf_progs(trace, key);
3149 
3150 		// It'll get at least the "!raw_syscalls:unaugmented"
3151 		prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
3152 		err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3153 		if (err)
3154 			break;
3155 		prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
3156 		err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
3157 		if (err)
3158 			break;
3159 	}
3160 
3161 	/*
3162 	 * Now lets do a second pass looking for enabled syscalls without
3163 	 * an augmenter that have a signature that is a superset of another
3164 	 * syscall with an augmenter so that we can auto-reuse it.
3165 	 *
3166 	 * I.e. if we have an augmenter for the "open" syscall that has
3167 	 * this signature:
3168 	 *
3169 	 *   int open(const char *pathname, int flags, mode_t mode);
3170 	 *
3171 	 * I.e. that will collect just the first string argument, then we
3172 	 * can reuse it for the 'creat' syscall, that has this signature:
3173 	 *
3174 	 *   int creat(const char *pathname, mode_t mode);
3175 	 *
3176 	 * and for:
3177 	 *
3178 	 *   int stat(const char *pathname, struct stat *statbuf);
3179 	 *   int lstat(const char *pathname, struct stat *statbuf);
3180 	 *
3181 	 * Because the 'open' augmenter will collect the first arg as a string,
3182 	 * and leave alone all the other args, which already helps with
3183 	 * beautifying 'stat' and 'lstat''s pathname arg.
3184 	 *
3185 	 * Then, in time, when 'stat' gets an augmenter that collects both
3186 	 * first and second arg (this one on the raw_syscalls:sys_exit prog
3187 	 * array tail call, then that one will be used.
3188 	 */
3189 	for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
3190 		struct syscall *sc = trace__syscall_info(trace, NULL, key);
3191 		struct bpf_program *pair_prog;
3192 		int prog_fd;
3193 
3194 		if (sc == NULL || sc->bpf_prog.sys_enter == NULL)
3195 			continue;
3196 
3197 		/*
3198 		 * For now we're just reusing the sys_enter prog, and if it
3199 		 * already has an augmenter, we don't need to find one.
3200 		 */
3201 		if (sc->bpf_prog.sys_enter != trace->syscalls.unaugmented_prog)
3202 			continue;
3203 
3204 		/*
3205 		 * Look at all the other syscalls for one that has a signature
3206 		 * that is close enough that we can share:
3207 		 */
3208 		pair_prog = trace__find_usable_bpf_prog_entry(trace, sc);
3209 		if (pair_prog == NULL)
3210 			continue;
3211 
3212 		sc->bpf_prog.sys_enter = pair_prog;
3213 
3214 		/*
3215 		 * Update the BPF_MAP_TYPE_PROG_SHARED for raw_syscalls:sys_enter
3216 		 * with the fd for the program we're reusing:
3217 		 */
3218 		prog_fd = bpf_program__fd(sc->bpf_prog.sys_enter);
3219 		err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
3220 		if (err)
3221 			break;
3222 	}
3223 
3224 
3225 	return err;
3226 }
3227 
3228 static void trace__delete_augmented_syscalls(struct trace *trace)
3229 {
3230 	struct evsel *evsel, *tmp;
3231 
3232 	evlist__remove(trace->evlist, trace->syscalls.events.augmented);
3233 	evsel__delete(trace->syscalls.events.augmented);
3234 	trace->syscalls.events.augmented = NULL;
3235 
3236 	evlist__for_each_entry_safe(trace->evlist, tmp, evsel) {
3237 		if (evsel->bpf_obj == trace->bpf_obj) {
3238 			evlist__remove(trace->evlist, evsel);
3239 			evsel__delete(evsel);
3240 		}
3241 
3242 	}
3243 
3244 	bpf_object__close(trace->bpf_obj);
3245 	trace->bpf_obj = NULL;
3246 }
3247 #else // HAVE_LIBBPF_SUPPORT
3248 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
3249 {
3250 	return 0;
3251 }
3252 
3253 static int trace__init_syscalls_bpf_map(struct trace *trace __maybe_unused)
3254 {
3255 	return 0;
3256 }
3257 
3258 static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace __maybe_unused,
3259 							    const char *name __maybe_unused)
3260 {
3261 	return NULL;
3262 }
3263 
3264 static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
3265 {
3266 	return 0;
3267 }
3268 
3269 static void trace__delete_augmented_syscalls(struct trace *trace __maybe_unused)
3270 {
3271 }
3272 #endif // HAVE_LIBBPF_SUPPORT
3273 
3274 static bool trace__only_augmented_syscalls_evsels(struct trace *trace)
3275 {
3276 	struct evsel *evsel;
3277 
3278 	evlist__for_each_entry(trace->evlist, evsel) {
3279 		if (evsel == trace->syscalls.events.augmented ||
3280 		    evsel->bpf_obj == trace->bpf_obj)
3281 			continue;
3282 
3283 		return false;
3284 	}
3285 
3286 	return true;
3287 }
3288 
3289 static int trace__set_ev_qualifier_filter(struct trace *trace)
3290 {
3291 	if (trace->syscalls.map)
3292 		return trace__set_ev_qualifier_bpf_filter(trace);
3293 	if (trace->syscalls.events.sys_enter)
3294 		return trace__set_ev_qualifier_tp_filter(trace);
3295 	return 0;
3296 }
3297 
3298 static int bpf_map__set_filter_pids(struct bpf_map *map __maybe_unused,
3299 				    size_t npids __maybe_unused, pid_t *pids __maybe_unused)
3300 {
3301 	int err = 0;
3302 #ifdef HAVE_LIBBPF_SUPPORT
3303 	bool value = true;
3304 	int map_fd = bpf_map__fd(map);
3305 	size_t i;
3306 
3307 	for (i = 0; i < npids; ++i) {
3308 		err = bpf_map_update_elem(map_fd, &pids[i], &value, BPF_ANY);
3309 		if (err)
3310 			break;
3311 	}
3312 #endif
3313 	return err;
3314 }
3315 
3316 static int trace__set_filter_loop_pids(struct trace *trace)
3317 {
3318 	unsigned int nr = 1, err;
3319 	pid_t pids[32] = {
3320 		getpid(),
3321 	};
3322 	struct thread *thread = machine__find_thread(trace->host, pids[0], pids[0]);
3323 
3324 	while (thread && nr < ARRAY_SIZE(pids)) {
3325 		struct thread *parent = machine__find_thread(trace->host, thread->ppid, thread->ppid);
3326 
3327 		if (parent == NULL)
3328 			break;
3329 
3330 		if (!strcmp(thread__comm_str(parent), "sshd") ||
3331 		    strstarts(thread__comm_str(parent), "gnome-terminal")) {
3332 			pids[nr++] = parent->tid;
3333 			break;
3334 		}
3335 		thread = parent;
3336 	}
3337 
3338 	err = perf_evlist__set_tp_filter_pids(trace->evlist, nr, pids);
3339 	if (!err && trace->filter_pids.map)
3340 		err = bpf_map__set_filter_pids(trace->filter_pids.map, nr, pids);
3341 
3342 	return err;
3343 }
3344 
3345 static int trace__set_filter_pids(struct trace *trace)
3346 {
3347 	int err = 0;
3348 	/*
3349 	 * Better not use !target__has_task() here because we need to cover the
3350 	 * case where no threads were specified in the command line, but a
3351 	 * workload was, and in that case we will fill in the thread_map when
3352 	 * we fork the workload in perf_evlist__prepare_workload.
3353 	 */
3354 	if (trace->filter_pids.nr > 0) {
3355 		err = perf_evlist__set_tp_filter_pids(trace->evlist, trace->filter_pids.nr,
3356 						      trace->filter_pids.entries);
3357 		if (!err && trace->filter_pids.map) {
3358 			err = bpf_map__set_filter_pids(trace->filter_pids.map, trace->filter_pids.nr,
3359 						       trace->filter_pids.entries);
3360 		}
3361 	} else if (perf_thread_map__pid(trace->evlist->core.threads, 0) == -1) {
3362 		err = trace__set_filter_loop_pids(trace);
3363 	}
3364 
3365 	return err;
3366 }
3367 
3368 static int __trace__deliver_event(struct trace *trace, union perf_event *event)
3369 {
3370 	struct evlist *evlist = trace->evlist;
3371 	struct perf_sample sample;
3372 	int err;
3373 
3374 	err = perf_evlist__parse_sample(evlist, event, &sample);
3375 	if (err)
3376 		fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
3377 	else
3378 		trace__handle_event(trace, event, &sample);
3379 
3380 	return 0;
3381 }
3382 
3383 static int __trace__flush_events(struct trace *trace)
3384 {
3385 	u64 first = ordered_events__first_time(&trace->oe.data);
3386 	u64 flush = trace->oe.last - NSEC_PER_SEC;
3387 
3388 	/* Is there some thing to flush.. */
3389 	if (first && first < flush)
3390 		return ordered_events__flush_time(&trace->oe.data, flush);
3391 
3392 	return 0;
3393 }
3394 
3395 static int trace__flush_events(struct trace *trace)
3396 {
3397 	return !trace->sort_events ? 0 : __trace__flush_events(trace);
3398 }
3399 
3400 static int trace__deliver_event(struct trace *trace, union perf_event *event)
3401 {
3402 	int err;
3403 
3404 	if (!trace->sort_events)
3405 		return __trace__deliver_event(trace, event);
3406 
3407 	err = perf_evlist__parse_sample_timestamp(trace->evlist, event, &trace->oe.last);
3408 	if (err && err != -1)
3409 		return err;
3410 
3411 	err = ordered_events__queue(&trace->oe.data, event, trace->oe.last, 0);
3412 	if (err)
3413 		return err;
3414 
3415 	return trace__flush_events(trace);
3416 }
3417 
3418 static int ordered_events__deliver_event(struct ordered_events *oe,
3419 					 struct ordered_event *event)
3420 {
3421 	struct trace *trace = container_of(oe, struct trace, oe.data);
3422 
3423 	return __trace__deliver_event(trace, event->event);
3424 }
3425 
3426 static int trace__run(struct trace *trace, int argc, const char **argv)
3427 {
3428 	struct evlist *evlist = trace->evlist;
3429 	struct evsel *evsel, *pgfault_maj = NULL, *pgfault_min = NULL;
3430 	int err = -1, i;
3431 	unsigned long before;
3432 	const bool forks = argc > 0;
3433 	bool draining = false;
3434 
3435 	trace->live = true;
3436 
3437 	if (!trace->raw_augmented_syscalls) {
3438 		if (trace->trace_syscalls && trace__add_syscall_newtp(trace))
3439 			goto out_error_raw_syscalls;
3440 
3441 		if (trace->trace_syscalls)
3442 			trace->vfs_getname = evlist__add_vfs_getname(evlist);
3443 	}
3444 
3445 	if ((trace->trace_pgfaults & TRACE_PFMAJ)) {
3446 		pgfault_maj = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MAJ);
3447 		if (pgfault_maj == NULL)
3448 			goto out_error_mem;
3449 		perf_evsel__config_callchain(pgfault_maj, &trace->opts, &callchain_param);
3450 		evlist__add(evlist, pgfault_maj);
3451 	}
3452 
3453 	if ((trace->trace_pgfaults & TRACE_PFMIN)) {
3454 		pgfault_min = perf_evsel__new_pgfault(PERF_COUNT_SW_PAGE_FAULTS_MIN);
3455 		if (pgfault_min == NULL)
3456 			goto out_error_mem;
3457 		perf_evsel__config_callchain(pgfault_min, &trace->opts, &callchain_param);
3458 		evlist__add(evlist, pgfault_min);
3459 	}
3460 
3461 	if (trace->sched &&
3462 	    perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
3463 				   trace__sched_stat_runtime))
3464 		goto out_error_sched_stat_runtime;
3465 	/*
3466 	 * If a global cgroup was set, apply it to all the events without an
3467 	 * explicit cgroup. I.e.:
3468 	 *
3469 	 * 	trace -G A -e sched:*switch
3470 	 *
3471 	 * Will set all raw_syscalls:sys_{enter,exit}, pgfault, vfs_getname, etc
3472 	 * _and_ sched:sched_switch to the 'A' cgroup, while:
3473 	 *
3474 	 * trace -e sched:*switch -G A
3475 	 *
3476 	 * will only set the sched:sched_switch event to the 'A' cgroup, all the
3477 	 * other events (raw_syscalls:sys_{enter,exit}, etc are left "without"
3478 	 * a cgroup (on the root cgroup, sys wide, etc).
3479 	 *
3480 	 * Multiple cgroups:
3481 	 *
3482 	 * trace -G A -e sched:*switch -G B
3483 	 *
3484 	 * the syscall ones go to the 'A' cgroup, the sched:sched_switch goes
3485 	 * to the 'B' cgroup.
3486 	 *
3487 	 * evlist__set_default_cgroup() grabs a reference of the passed cgroup
3488 	 * only for the evsels still without a cgroup, i.e. evsel->cgroup == NULL.
3489 	 */
3490 	if (trace->cgroup)
3491 		evlist__set_default_cgroup(trace->evlist, trace->cgroup);
3492 
3493 	err = perf_evlist__create_maps(evlist, &trace->opts.target);
3494 	if (err < 0) {
3495 		fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
3496 		goto out_delete_evlist;
3497 	}
3498 
3499 	err = trace__symbols_init(trace, evlist);
3500 	if (err < 0) {
3501 		fprintf(trace->output, "Problems initializing symbol libraries!\n");
3502 		goto out_delete_evlist;
3503 	}
3504 
3505 	perf_evlist__config(evlist, &trace->opts, &callchain_param);
3506 
3507 	signal(SIGCHLD, sig_handler);
3508 	signal(SIGINT, sig_handler);
3509 
3510 	if (forks) {
3511 		err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
3512 						    argv, false, NULL);
3513 		if (err < 0) {
3514 			fprintf(trace->output, "Couldn't run the workload!\n");
3515 			goto out_delete_evlist;
3516 		}
3517 	}
3518 
3519 	err = evlist__open(evlist);
3520 	if (err < 0)
3521 		goto out_error_open;
3522 
3523 	err = bpf__apply_obj_config();
3524 	if (err) {
3525 		char errbuf[BUFSIZ];
3526 
3527 		bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf));
3528 		pr_err("ERROR: Apply config to BPF failed: %s\n",
3529 			 errbuf);
3530 		goto out_error_open;
3531 	}
3532 
3533 	err = trace__set_filter_pids(trace);
3534 	if (err < 0)
3535 		goto out_error_mem;
3536 
3537 	if (trace->syscalls.map)
3538 		trace__init_syscalls_bpf_map(trace);
3539 
3540 	if (trace->syscalls.prog_array.sys_enter)
3541 		trace__init_syscalls_bpf_prog_array_maps(trace);
3542 
3543 	if (trace->ev_qualifier_ids.nr > 0) {
3544 		err = trace__set_ev_qualifier_filter(trace);
3545 		if (err < 0)
3546 			goto out_errno;
3547 
3548 		if (trace->syscalls.events.sys_exit) {
3549 			pr_debug("event qualifier tracepoint filter: %s\n",
3550 				 trace->syscalls.events.sys_exit->filter);
3551 		}
3552 	}
3553 
3554 	/*
3555 	 * If the "close" syscall is not traced, then we will not have the
3556 	 * opportunity to, in syscall_arg__scnprintf_close_fd() invalidate the
3557 	 * fd->pathname table and were ending up showing the last value set by
3558 	 * syscalls opening a pathname and associating it with a descriptor or
3559 	 * reading it from /proc/pid/fd/ in cases where that doesn't make
3560 	 * sense.
3561 	 *
3562 	 *  So just disable this beautifier (SCA_FD, SCA_FDAT) when 'close' is
3563 	 *  not in use.
3564 	 */
3565 	trace->fd_path_disabled = !trace__syscall_enabled(trace, syscalltbl__id(trace->sctbl, "close"));
3566 
3567 	err = perf_evlist__apply_filters(evlist, &evsel);
3568 	if (err < 0)
3569 		goto out_error_apply_filters;
3570 
3571 	if (trace->dump.map)
3572 		bpf_map__fprintf(trace->dump.map, trace->output);
3573 
3574 	err = evlist__mmap(evlist, trace->opts.mmap_pages);
3575 	if (err < 0)
3576 		goto out_error_mmap;
3577 
3578 	if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
3579 		evlist__enable(evlist);
3580 
3581 	if (forks)
3582 		perf_evlist__start_workload(evlist);
3583 
3584 	if (trace->opts.initial_delay) {
3585 		usleep(trace->opts.initial_delay * 1000);
3586 		evlist__enable(evlist);
3587 	}
3588 
3589 	trace->multiple_threads = perf_thread_map__pid(evlist->core.threads, 0) == -1 ||
3590 				  evlist->core.threads->nr > 1 ||
3591 				  evlist__first(evlist)->core.attr.inherit;
3592 
3593 	/*
3594 	 * Now that we already used evsel->core.attr to ask the kernel to setup the
3595 	 * events, lets reuse evsel->core.attr.sample_max_stack as the limit in
3596 	 * trace__resolve_callchain(), allowing per-event max-stack settings
3597 	 * to override an explicitly set --max-stack global setting.
3598 	 */
3599 	evlist__for_each_entry(evlist, evsel) {
3600 		if (evsel__has_callchain(evsel) &&
3601 		    evsel->core.attr.sample_max_stack == 0)
3602 			evsel->core.attr.sample_max_stack = trace->max_stack;
3603 	}
3604 again:
3605 	before = trace->nr_events;
3606 
3607 	for (i = 0; i < evlist->core.nr_mmaps; i++) {
3608 		union perf_event *event;
3609 		struct mmap *md;
3610 
3611 		md = &evlist->mmap[i];
3612 		if (perf_mmap__read_init(md) < 0)
3613 			continue;
3614 
3615 		while ((event = perf_mmap__read_event(md)) != NULL) {
3616 			++trace->nr_events;
3617 
3618 			err = trace__deliver_event(trace, event);
3619 			if (err)
3620 				goto out_disable;
3621 
3622 			perf_mmap__consume(md);
3623 
3624 			if (interrupted)
3625 				goto out_disable;
3626 
3627 			if (done && !draining) {
3628 				evlist__disable(evlist);
3629 				draining = true;
3630 			}
3631 		}
3632 		perf_mmap__read_done(md);
3633 	}
3634 
3635 	if (trace->nr_events == before) {
3636 		int timeout = done ? 100 : -1;
3637 
3638 		if (!draining && evlist__poll(evlist, timeout) > 0) {
3639 			if (evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
3640 				draining = true;
3641 
3642 			goto again;
3643 		} else {
3644 			if (trace__flush_events(trace))
3645 				goto out_disable;
3646 		}
3647 	} else {
3648 		goto again;
3649 	}
3650 
3651 out_disable:
3652 	thread__zput(trace->current);
3653 
3654 	evlist__disable(evlist);
3655 
3656 	if (trace->sort_events)
3657 		ordered_events__flush(&trace->oe.data, OE_FLUSH__FINAL);
3658 
3659 	if (!err) {
3660 		if (trace->summary)
3661 			trace__fprintf_thread_summary(trace, trace->output);
3662 
3663 		if (trace->show_tool_stats) {
3664 			fprintf(trace->output, "Stats:\n "
3665 					       " vfs_getname : %" PRIu64 "\n"
3666 					       " proc_getname: %" PRIu64 "\n",
3667 				trace->stats.vfs_getname,
3668 				trace->stats.proc_getname);
3669 		}
3670 	}
3671 
3672 out_delete_evlist:
3673 	trace__symbols__exit(trace);
3674 
3675 	evlist__delete(evlist);
3676 	cgroup__put(trace->cgroup);
3677 	trace->evlist = NULL;
3678 	trace->live = false;
3679 	return err;
3680 {
3681 	char errbuf[BUFSIZ];
3682 
3683 out_error_sched_stat_runtime:
3684 	tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
3685 	goto out_error;
3686 
3687 out_error_raw_syscalls:
3688 	tracing_path__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
3689 	goto out_error;
3690 
3691 out_error_mmap:
3692 	perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
3693 	goto out_error;
3694 
3695 out_error_open:
3696 	perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
3697 
3698 out_error:
3699 	fprintf(trace->output, "%s\n", errbuf);
3700 	goto out_delete_evlist;
3701 
3702 out_error_apply_filters:
3703 	fprintf(trace->output,
3704 		"Failed to set filter \"%s\" on event %s with %d (%s)\n",
3705 		evsel->filter, perf_evsel__name(evsel), errno,
3706 		str_error_r(errno, errbuf, sizeof(errbuf)));
3707 	goto out_delete_evlist;
3708 }
3709 out_error_mem:
3710 	fprintf(trace->output, "Not enough memory to run!\n");
3711 	goto out_delete_evlist;
3712 
3713 out_errno:
3714 	fprintf(trace->output, "errno=%d,%s\n", errno, strerror(errno));
3715 	goto out_delete_evlist;
3716 }
3717 
3718 static int trace__replay(struct trace *trace)
3719 {
3720 	const struct evsel_str_handler handlers[] = {
3721 		{ "probe:vfs_getname",	     trace__vfs_getname, },
3722 	};
3723 	struct perf_data data = {
3724 		.path  = input_name,
3725 		.mode  = PERF_DATA_MODE_READ,
3726 		.force = trace->force,
3727 	};
3728 	struct perf_session *session;
3729 	struct evsel *evsel;
3730 	int err = -1;
3731 
3732 	trace->tool.sample	  = trace__process_sample;
3733 	trace->tool.mmap	  = perf_event__process_mmap;
3734 	trace->tool.mmap2	  = perf_event__process_mmap2;
3735 	trace->tool.comm	  = perf_event__process_comm;
3736 	trace->tool.exit	  = perf_event__process_exit;
3737 	trace->tool.fork	  = perf_event__process_fork;
3738 	trace->tool.attr	  = perf_event__process_attr;
3739 	trace->tool.tracing_data  = perf_event__process_tracing_data;
3740 	trace->tool.build_id	  = perf_event__process_build_id;
3741 	trace->tool.namespaces	  = perf_event__process_namespaces;
3742 
3743 	trace->tool.ordered_events = true;
3744 	trace->tool.ordering_requires_timestamps = true;
3745 
3746 	/* add tid to output */
3747 	trace->multiple_threads = true;
3748 
3749 	session = perf_session__new(&data, false, &trace->tool);
3750 	if (IS_ERR(session))
3751 		return PTR_ERR(session);
3752 
3753 	if (trace->opts.target.pid)
3754 		symbol_conf.pid_list_str = strdup(trace->opts.target.pid);
3755 
3756 	if (trace->opts.target.tid)
3757 		symbol_conf.tid_list_str = strdup(trace->opts.target.tid);
3758 
3759 	if (symbol__init(&session->header.env) < 0)
3760 		goto out;
3761 
3762 	trace->host = &session->machines.host;
3763 
3764 	err = perf_session__set_tracepoints_handlers(session, handlers);
3765 	if (err)
3766 		goto out;
3767 
3768 	evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3769 						     "raw_syscalls:sys_enter");
3770 	/* older kernels have syscalls tp versus raw_syscalls */
3771 	if (evsel == NULL)
3772 		evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3773 							     "syscalls:sys_enter");
3774 
3775 	if (evsel &&
3776 	    (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_enter) < 0 ||
3777 	    perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
3778 		pr_err("Error during initialize raw_syscalls:sys_enter event\n");
3779 		goto out;
3780 	}
3781 
3782 	evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3783 						     "raw_syscalls:sys_exit");
3784 	if (evsel == NULL)
3785 		evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
3786 							     "syscalls:sys_exit");
3787 	if (evsel &&
3788 	    (perf_evsel__init_raw_syscall_tp(evsel, trace__sys_exit) < 0 ||
3789 	    perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
3790 		pr_err("Error during initialize raw_syscalls:sys_exit event\n");
3791 		goto out;
3792 	}
3793 
3794 	evlist__for_each_entry(session->evlist, evsel) {
3795 		if (evsel->core.attr.type == PERF_TYPE_SOFTWARE &&
3796 		    (evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
3797 		     evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
3798 		     evsel->core.attr.config == PERF_COUNT_SW_PAGE_FAULTS))
3799 			evsel->handler = trace__pgfault;
3800 	}
3801 
3802 	setup_pager();
3803 
3804 	err = perf_session__process_events(session);
3805 	if (err)
3806 		pr_err("Failed to process events, error %d", err);
3807 
3808 	else if (trace->summary)
3809 		trace__fprintf_thread_summary(trace, trace->output);
3810 
3811 out:
3812 	perf_session__delete(session);
3813 
3814 	return err;
3815 }
3816 
3817 static size_t trace__fprintf_threads_header(FILE *fp)
3818 {
3819 	size_t printed;
3820 
3821 	printed  = fprintf(fp, "\n Summary of events:\n\n");
3822 
3823 	return printed;
3824 }
3825 
3826 DEFINE_RESORT_RB(syscall_stats, a->msecs > b->msecs,
3827 	struct stats 	*stats;
3828 	double		msecs;
3829 	int		syscall;
3830 )
3831 {
3832 	struct int_node *source = rb_entry(nd, struct int_node, rb_node);
3833 	struct stats *stats = source->priv;
3834 
3835 	entry->syscall = source->i;
3836 	entry->stats   = stats;
3837 	entry->msecs   = stats ? (u64)stats->n * (avg_stats(stats) / NSEC_PER_MSEC) : 0;
3838 }
3839 
3840 static size_t thread__dump_stats(struct thread_trace *ttrace,
3841 				 struct trace *trace, FILE *fp)
3842 {
3843 	size_t printed = 0;
3844 	struct syscall *sc;
3845 	struct rb_node *nd;
3846 	DECLARE_RESORT_RB_INTLIST(syscall_stats, ttrace->syscall_stats);
3847 
3848 	if (syscall_stats == NULL)
3849 		return 0;
3850 
3851 	printed += fprintf(fp, "\n");
3852 
3853 	printed += fprintf(fp, "   syscall            calls    total       min       avg       max      stddev\n");
3854 	printed += fprintf(fp, "                               (msec)    (msec)    (msec)    (msec)        (%%)\n");
3855 	printed += fprintf(fp, "   --------------- -------- --------- --------- --------- ---------     ------\n");
3856 
3857 	resort_rb__for_each_entry(nd, syscall_stats) {
3858 		struct stats *stats = syscall_stats_entry->stats;
3859 		if (stats) {
3860 			double min = (double)(stats->min) / NSEC_PER_MSEC;
3861 			double max = (double)(stats->max) / NSEC_PER_MSEC;
3862 			double avg = avg_stats(stats);
3863 			double pct;
3864 			u64 n = (u64) stats->n;
3865 
3866 			pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
3867 			avg /= NSEC_PER_MSEC;
3868 
3869 			sc = &trace->syscalls.table[syscall_stats_entry->syscall];
3870 			printed += fprintf(fp, "   %-15s", sc->name);
3871 			printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f %9.3f",
3872 					   n, syscall_stats_entry->msecs, min, avg);
3873 			printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
3874 		}
3875 	}
3876 
3877 	resort_rb__delete(syscall_stats);
3878 	printed += fprintf(fp, "\n\n");
3879 
3880 	return printed;
3881 }
3882 
3883 static size_t trace__fprintf_thread(FILE *fp, struct thread *thread, struct trace *trace)
3884 {
3885 	size_t printed = 0;
3886 	struct thread_trace *ttrace = thread__priv(thread);
3887 	double ratio;
3888 
3889 	if (ttrace == NULL)
3890 		return 0;
3891 
3892 	ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
3893 
3894 	printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
3895 	printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
3896 	printed += fprintf(fp, "%.1f%%", ratio);
3897 	if (ttrace->pfmaj)
3898 		printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
3899 	if (ttrace->pfmin)
3900 		printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
3901 	if (trace->sched)
3902 		printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
3903 	else if (fputc('\n', fp) != EOF)
3904 		++printed;
3905 
3906 	printed += thread__dump_stats(ttrace, trace, fp);
3907 
3908 	return printed;
3909 }
3910 
3911 static unsigned long thread__nr_events(struct thread_trace *ttrace)
3912 {
3913 	return ttrace ? ttrace->nr_events : 0;
3914 }
3915 
3916 DEFINE_RESORT_RB(threads, (thread__nr_events(a->thread->priv) < thread__nr_events(b->thread->priv)),
3917 	struct thread *thread;
3918 )
3919 {
3920 	entry->thread = rb_entry(nd, struct thread, rb_node);
3921 }
3922 
3923 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
3924 {
3925 	size_t printed = trace__fprintf_threads_header(fp);
3926 	struct rb_node *nd;
3927 	int i;
3928 
3929 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3930 		DECLARE_RESORT_RB_MACHINE_THREADS(threads, trace->host, i);
3931 
3932 		if (threads == NULL) {
3933 			fprintf(fp, "%s", "Error sorting output by nr_events!\n");
3934 			return 0;
3935 		}
3936 
3937 		resort_rb__for_each_entry(nd, threads)
3938 			printed += trace__fprintf_thread(fp, threads_entry->thread, trace);
3939 
3940 		resort_rb__delete(threads);
3941 	}
3942 	return printed;
3943 }
3944 
3945 static int trace__set_duration(const struct option *opt, const char *str,
3946 			       int unset __maybe_unused)
3947 {
3948 	struct trace *trace = opt->value;
3949 
3950 	trace->duration_filter = atof(str);
3951 	return 0;
3952 }
3953 
3954 static int trace__set_filter_pids_from_option(const struct option *opt, const char *str,
3955 					      int unset __maybe_unused)
3956 {
3957 	int ret = -1;
3958 	size_t i;
3959 	struct trace *trace = opt->value;
3960 	/*
3961 	 * FIXME: introduce a intarray class, plain parse csv and create a
3962 	 * { int nr, int entries[] } struct...
3963 	 */
3964 	struct intlist *list = intlist__new(str);
3965 
3966 	if (list == NULL)
3967 		return -1;
3968 
3969 	i = trace->filter_pids.nr = intlist__nr_entries(list) + 1;
3970 	trace->filter_pids.entries = calloc(i, sizeof(pid_t));
3971 
3972 	if (trace->filter_pids.entries == NULL)
3973 		goto out;
3974 
3975 	trace->filter_pids.entries[0] = getpid();
3976 
3977 	for (i = 1; i < trace->filter_pids.nr; ++i)
3978 		trace->filter_pids.entries[i] = intlist__entry(list, i - 1)->i;
3979 
3980 	intlist__delete(list);
3981 	ret = 0;
3982 out:
3983 	return ret;
3984 }
3985 
3986 static int trace__open_output(struct trace *trace, const char *filename)
3987 {
3988 	struct stat st;
3989 
3990 	if (!stat(filename, &st) && st.st_size) {
3991 		char oldname[PATH_MAX];
3992 
3993 		scnprintf(oldname, sizeof(oldname), "%s.old", filename);
3994 		unlink(oldname);
3995 		rename(filename, oldname);
3996 	}
3997 
3998 	trace->output = fopen(filename, "w");
3999 
4000 	return trace->output == NULL ? -errno : 0;
4001 }
4002 
4003 static int parse_pagefaults(const struct option *opt, const char *str,
4004 			    int unset __maybe_unused)
4005 {
4006 	int *trace_pgfaults = opt->value;
4007 
4008 	if (strcmp(str, "all") == 0)
4009 		*trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
4010 	else if (strcmp(str, "maj") == 0)
4011 		*trace_pgfaults |= TRACE_PFMAJ;
4012 	else if (strcmp(str, "min") == 0)
4013 		*trace_pgfaults |= TRACE_PFMIN;
4014 	else
4015 		return -1;
4016 
4017 	return 0;
4018 }
4019 
4020 static void evlist__set_default_evsel_handler(struct evlist *evlist, void *handler)
4021 {
4022 	struct evsel *evsel;
4023 
4024 	evlist__for_each_entry(evlist, evsel) {
4025 		if (evsel->handler == NULL)
4026 			evsel->handler = handler;
4027 	}
4028 }
4029 
4030 static int evlist__set_syscall_tp_fields(struct evlist *evlist)
4031 {
4032 	struct evsel *evsel;
4033 
4034 	evlist__for_each_entry(evlist, evsel) {
4035 		if (evsel->priv || !evsel->tp_format)
4036 			continue;
4037 
4038 		if (strcmp(evsel->tp_format->system, "syscalls")) {
4039 			perf_evsel__init_tp_arg_scnprintf(evsel);
4040 			continue;
4041 		}
4042 
4043 		if (perf_evsel__init_syscall_tp(evsel))
4044 			return -1;
4045 
4046 		if (!strncmp(evsel->tp_format->name, "sys_enter_", 10)) {
4047 			struct syscall_tp *sc = evsel->priv;
4048 
4049 			if (__tp_field__init_ptr(&sc->args, sc->id.offset + sizeof(u64)))
4050 				return -1;
4051 		} else if (!strncmp(evsel->tp_format->name, "sys_exit_", 9)) {
4052 			struct syscall_tp *sc = evsel->priv;
4053 
4054 			if (__tp_field__init_uint(&sc->ret, sizeof(u64), sc->id.offset + sizeof(u64), evsel->needs_swap))
4055 				return -1;
4056 		}
4057 	}
4058 
4059 	return 0;
4060 }
4061 
4062 /*
4063  * XXX: Hackish, just splitting the combined -e+--event (syscalls
4064  * (raw_syscalls:{sys_{enter,exit}} + events (tracepoints, HW, SW, etc) to use
4065  * existing facilities unchanged (trace->ev_qualifier + parse_options()).
4066  *
4067  * It'd be better to introduce a parse_options() variant that would return a
4068  * list with the terms it didn't match to an event...
4069  */
4070 static int trace__parse_events_option(const struct option *opt, const char *str,
4071 				      int unset __maybe_unused)
4072 {
4073 	struct trace *trace = (struct trace *)opt->value;
4074 	const char *s = str;
4075 	char *sep = NULL, *lists[2] = { NULL, NULL, };
4076 	int len = strlen(str) + 1, err = -1, list, idx;
4077 	char *strace_groups_dir = system_path(STRACE_GROUPS_DIR);
4078 	char group_name[PATH_MAX];
4079 	struct syscall_fmt *fmt;
4080 
4081 	if (strace_groups_dir == NULL)
4082 		return -1;
4083 
4084 	if (*s == '!') {
4085 		++s;
4086 		trace->not_ev_qualifier = true;
4087 	}
4088 
4089 	while (1) {
4090 		if ((sep = strchr(s, ',')) != NULL)
4091 			*sep = '\0';
4092 
4093 		list = 0;
4094 		if (syscalltbl__id(trace->sctbl, s) >= 0 ||
4095 		    syscalltbl__strglobmatch_first(trace->sctbl, s, &idx) >= 0) {
4096 			list = 1;
4097 			goto do_concat;
4098 		}
4099 
4100 		fmt = syscall_fmt__find_by_alias(s);
4101 		if (fmt != NULL) {
4102 			list = 1;
4103 			s = fmt->name;
4104 		} else {
4105 			path__join(group_name, sizeof(group_name), strace_groups_dir, s);
4106 			if (access(group_name, R_OK) == 0)
4107 				list = 1;
4108 		}
4109 do_concat:
4110 		if (lists[list]) {
4111 			sprintf(lists[list] + strlen(lists[list]), ",%s", s);
4112 		} else {
4113 			lists[list] = malloc(len);
4114 			if (lists[list] == NULL)
4115 				goto out;
4116 			strcpy(lists[list], s);
4117 		}
4118 
4119 		if (!sep)
4120 			break;
4121 
4122 		*sep = ',';
4123 		s = sep + 1;
4124 	}
4125 
4126 	if (lists[1] != NULL) {
4127 		struct strlist_config slist_config = {
4128 			.dirname = strace_groups_dir,
4129 		};
4130 
4131 		trace->ev_qualifier = strlist__new(lists[1], &slist_config);
4132 		if (trace->ev_qualifier == NULL) {
4133 			fputs("Not enough memory to parse event qualifier", trace->output);
4134 			goto out;
4135 		}
4136 
4137 		if (trace__validate_ev_qualifier(trace))
4138 			goto out;
4139 		trace->trace_syscalls = true;
4140 	}
4141 
4142 	err = 0;
4143 
4144 	if (lists[0]) {
4145 		struct option o = OPT_CALLBACK('e', "event", &trace->evlist, "event",
4146 					       "event selector. use 'perf list' to list available events",
4147 					       parse_events_option);
4148 		err = parse_events_option(&o, lists[0], 0);
4149 	}
4150 out:
4151 	if (sep)
4152 		*sep = ',';
4153 
4154 	return err;
4155 }
4156 
4157 static int trace__parse_cgroups(const struct option *opt, const char *str, int unset)
4158 {
4159 	struct trace *trace = opt->value;
4160 
4161 	if (!list_empty(&trace->evlist->core.entries))
4162 		return parse_cgroups(opt, str, unset);
4163 
4164 	trace->cgroup = evlist__findnew_cgroup(trace->evlist, str);
4165 
4166 	return 0;
4167 }
4168 
4169 static struct bpf_map *trace__find_bpf_map_by_name(struct trace *trace, const char *name)
4170 {
4171 	if (trace->bpf_obj == NULL)
4172 		return NULL;
4173 
4174 	return bpf_object__find_map_by_name(trace->bpf_obj, name);
4175 }
4176 
4177 static void trace__set_bpf_map_filtered_pids(struct trace *trace)
4178 {
4179 	trace->filter_pids.map = trace__find_bpf_map_by_name(trace, "pids_filtered");
4180 }
4181 
4182 static void trace__set_bpf_map_syscalls(struct trace *trace)
4183 {
4184 	trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
4185 	trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
4186 	trace->syscalls.prog_array.sys_exit  = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
4187 }
4188 
4189 static int trace__config(const char *var, const char *value, void *arg)
4190 {
4191 	struct trace *trace = arg;
4192 	int err = 0;
4193 
4194 	if (!strcmp(var, "trace.add_events")) {
4195 		trace->perfconfig_events = strdup(value);
4196 		if (trace->perfconfig_events == NULL) {
4197 			pr_err("Not enough memory for %s\n", "trace.add_events");
4198 			return -1;
4199 		}
4200 	} else if (!strcmp(var, "trace.show_timestamp")) {
4201 		trace->show_tstamp = perf_config_bool(var, value);
4202 	} else if (!strcmp(var, "trace.show_duration")) {
4203 		trace->show_duration = perf_config_bool(var, value);
4204 	} else if (!strcmp(var, "trace.show_arg_names")) {
4205 		trace->show_arg_names = perf_config_bool(var, value);
4206 		if (!trace->show_arg_names)
4207 			trace->show_zeros = true;
4208 	} else if (!strcmp(var, "trace.show_zeros")) {
4209 		bool new_show_zeros = perf_config_bool(var, value);
4210 		if (!trace->show_arg_names && !new_show_zeros) {
4211 			pr_warning("trace.show_zeros has to be set when trace.show_arg_names=no\n");
4212 			goto out;
4213 		}
4214 		trace->show_zeros = new_show_zeros;
4215 	} else if (!strcmp(var, "trace.show_prefix")) {
4216 		trace->show_string_prefix = perf_config_bool(var, value);
4217 	} else if (!strcmp(var, "trace.no_inherit")) {
4218 		trace->opts.no_inherit = perf_config_bool(var, value);
4219 	} else if (!strcmp(var, "trace.args_alignment")) {
4220 		int args_alignment = 0;
4221 		if (perf_config_int(&args_alignment, var, value) == 0)
4222 			trace->args_alignment = args_alignment;
4223 	} else if (!strcmp(var, "trace.tracepoint_beautifiers")) {
4224 		if (strcasecmp(value, "libtraceevent") == 0)
4225 			trace->libtraceevent_print = true;
4226 		else if (strcasecmp(value, "libbeauty") == 0)
4227 			trace->libtraceevent_print = false;
4228 	}
4229 out:
4230 	return err;
4231 }
4232 
4233 int cmd_trace(int argc, const char **argv)
4234 {
4235 	const char *trace_usage[] = {
4236 		"perf trace [<options>] [<command>]",
4237 		"perf trace [<options>] -- <command> [<options>]",
4238 		"perf trace record [<options>] [<command>]",
4239 		"perf trace record [<options>] -- <command> [<options>]",
4240 		NULL
4241 	};
4242 	struct trace trace = {
4243 		.opts = {
4244 			.target = {
4245 				.uid	   = UINT_MAX,
4246 				.uses_mmap = true,
4247 			},
4248 			.user_freq     = UINT_MAX,
4249 			.user_interval = ULLONG_MAX,
4250 			.no_buffering  = true,
4251 			.mmap_pages    = UINT_MAX,
4252 		},
4253 		.output = stderr,
4254 		.show_comm = true,
4255 		.show_tstamp = true,
4256 		.show_duration = true,
4257 		.show_arg_names = true,
4258 		.args_alignment = 70,
4259 		.trace_syscalls = false,
4260 		.kernel_syscallchains = false,
4261 		.max_stack = UINT_MAX,
4262 		.max_events = ULONG_MAX,
4263 	};
4264 	const char *map_dump_str = NULL;
4265 	const char *output_name = NULL;
4266 	const struct option trace_options[] = {
4267 	OPT_CALLBACK('e', "event", &trace, "event",
4268 		     "event/syscall selector. use 'perf list' to list available events",
4269 		     trace__parse_events_option),
4270 	OPT_BOOLEAN(0, "comm", &trace.show_comm,
4271 		    "show the thread COMM next to its id"),
4272 	OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
4273 	OPT_CALLBACK(0, "expr", &trace, "expr", "list of syscalls/events to trace",
4274 		     trace__parse_events_option),
4275 	OPT_STRING('o', "output", &output_name, "file", "output file name"),
4276 	OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
4277 	OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
4278 		    "trace events on existing process id"),
4279 	OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
4280 		    "trace events on existing thread id"),
4281 	OPT_CALLBACK(0, "filter-pids", &trace, "CSV list of pids",
4282 		     "pids to filter (by the kernel)", trace__set_filter_pids_from_option),
4283 	OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
4284 		    "system-wide collection from all CPUs"),
4285 	OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
4286 		    "list of cpus to monitor"),
4287 	OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
4288 		    "child tasks do not inherit counters"),
4289 	OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
4290 		     "number of mmap data pages",
4291 		     perf_evlist__parse_mmap_pages),
4292 	OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
4293 		   "user to profile"),
4294 	OPT_CALLBACK(0, "duration", &trace, "float",
4295 		     "show only events with duration > N.M ms",
4296 		     trace__set_duration),
4297 #ifdef HAVE_LIBBPF_SUPPORT
4298 	OPT_STRING(0, "map-dump", &map_dump_str, "BPF map", "BPF map to periodically dump"),
4299 #endif
4300 	OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
4301 	OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4302 	OPT_BOOLEAN('T', "time", &trace.full_time,
4303 		    "Show full timestamp, not time relative to first start"),
4304 	OPT_BOOLEAN(0, "failure", &trace.failure_only,
4305 		    "Show only syscalls that failed"),
4306 	OPT_BOOLEAN('s', "summary", &trace.summary_only,
4307 		    "Show only syscall summary with statistics"),
4308 	OPT_BOOLEAN('S', "with-summary", &trace.summary,
4309 		    "Show all syscalls and summary with statistics"),
4310 	OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
4311 		     "Trace pagefaults", parse_pagefaults, "maj"),
4312 	OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
4313 	OPT_BOOLEAN('f', "force", &trace.force, "don't complain, do it"),
4314 	OPT_CALLBACK(0, "call-graph", &trace.opts,
4315 		     "record_mode[,record_size]", record_callchain_help,
4316 		     &record_parse_callchain_opt),
4317 	OPT_BOOLEAN(0, "libtraceevent_print", &trace.libtraceevent_print,
4318 		    "Use libtraceevent to print the tracepoint arguments."),
4319 	OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
4320 		    "Show the kernel callchains on the syscall exit path"),
4321 	OPT_ULONG(0, "max-events", &trace.max_events,
4322 		"Set the maximum number of events to print, exit after that is reached. "),
4323 	OPT_UINTEGER(0, "min-stack", &trace.min_stack,
4324 		     "Set the minimum stack depth when parsing the callchain, "
4325 		     "anything below the specified depth will be ignored."),
4326 	OPT_UINTEGER(0, "max-stack", &trace.max_stack,
4327 		     "Set the maximum stack depth when parsing the callchain, "
4328 		     "anything beyond the specified depth will be ignored. "
4329 		     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
4330 	OPT_BOOLEAN(0, "sort-events", &trace.sort_events,
4331 			"Sort batch of events before processing, use if getting out of order events"),
4332 	OPT_BOOLEAN(0, "print-sample", &trace.print_sample,
4333 			"print the PERF_RECORD_SAMPLE PERF_SAMPLE_ info, for debugging"),
4334 	OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
4335 			"per thread proc mmap processing timeout in ms"),
4336 	OPT_CALLBACK('G', "cgroup", &trace, "name", "monitor event in cgroup name only",
4337 		     trace__parse_cgroups),
4338 	OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
4339 		     "ms to wait before starting measurement after program "
4340 		     "start"),
4341 	OPTS_EVSWITCH(&trace.evswitch),
4342 	OPT_END()
4343 	};
4344 	bool __maybe_unused max_stack_user_set = true;
4345 	bool mmap_pages_user_set = true;
4346 	struct evsel *evsel;
4347 	const char * const trace_subcommands[] = { "record", NULL };
4348 	int err = -1;
4349 	char bf[BUFSIZ];
4350 
4351 	signal(SIGSEGV, sighandler_dump_stack);
4352 	signal(SIGFPE, sighandler_dump_stack);
4353 
4354 	trace.evlist = evlist__new();
4355 	trace.sctbl = syscalltbl__new();
4356 
4357 	if (trace.evlist == NULL || trace.sctbl == NULL) {
4358 		pr_err("Not enough memory to run!\n");
4359 		err = -ENOMEM;
4360 		goto out;
4361 	}
4362 
4363 	/*
4364 	 * Parsing .perfconfig may entail creating a BPF event, that may need
4365 	 * to create BPF maps, so bump RLIM_MEMLOCK as the default 64K setting
4366 	 * is too small. This affects just this process, not touching the
4367 	 * global setting. If it fails we'll get something in 'perf trace -v'
4368 	 * to help diagnose the problem.
4369 	 */
4370 	rlimit__bump_memlock();
4371 
4372 	err = perf_config(trace__config, &trace);
4373 	if (err)
4374 		goto out;
4375 
4376 	argc = parse_options_subcommand(argc, argv, trace_options, trace_subcommands,
4377 				 trace_usage, PARSE_OPT_STOP_AT_NON_OPTION);
4378 
4379 	/*
4380 	 * Here we already passed thru trace__parse_events_option() and it has
4381 	 * already figured out if -e syscall_name, if not but if --event
4382 	 * foo:bar was used, the user is interested _just_ in those, say,
4383 	 * tracepoint events, not in the strace-like syscall-name-based mode.
4384 	 *
4385 	 * This is important because we need to check if strace-like mode is
4386 	 * needed to decided if we should filter out the eBPF
4387 	 * __augmented_syscalls__ code, if it is in the mix, say, via
4388 	 * .perfconfig trace.add_events, and filter those out.
4389 	 */
4390 	if (!trace.trace_syscalls && !trace.trace_pgfaults &&
4391 	    trace.evlist->core.nr_entries == 0 /* Was --events used? */) {
4392 		trace.trace_syscalls = true;
4393 	}
4394 	/*
4395 	 * Now that we have --verbose figured out, lets see if we need to parse
4396 	 * events from .perfconfig, so that if those events fail parsing, say some
4397 	 * BPF program fails, then we'll be able to use --verbose to see what went
4398 	 * wrong in more detail.
4399 	 */
4400 	if (trace.perfconfig_events != NULL) {
4401 		struct parse_events_error parse_err = { .idx = 0, };
4402 
4403 		err = parse_events(trace.evlist, trace.perfconfig_events, &parse_err);
4404 		if (err) {
4405 			parse_events_print_error(&parse_err, trace.perfconfig_events);
4406 			goto out;
4407 		}
4408 	}
4409 
4410 	if ((nr_cgroups || trace.cgroup) && !trace.opts.target.system_wide) {
4411 		usage_with_options_msg(trace_usage, trace_options,
4412 				       "cgroup monitoring only available in system-wide mode");
4413 	}
4414 
4415 	evsel = bpf__setup_output_event(trace.evlist, "__augmented_syscalls__");
4416 	if (IS_ERR(evsel)) {
4417 		bpf__strerror_setup_output_event(trace.evlist, PTR_ERR(evsel), bf, sizeof(bf));
4418 		pr_err("ERROR: Setup trace syscalls enter failed: %s\n", bf);
4419 		goto out;
4420 	}
4421 
4422 	if (evsel) {
4423 		trace.syscalls.events.augmented = evsel;
4424 
4425 		evsel = perf_evlist__find_tracepoint_by_name(trace.evlist, "raw_syscalls:sys_enter");
4426 		if (evsel == NULL) {
4427 			pr_err("ERROR: raw_syscalls:sys_enter not found in the augmented BPF object\n");
4428 			goto out;
4429 		}
4430 
4431 		if (evsel->bpf_obj == NULL) {
4432 			pr_err("ERROR: raw_syscalls:sys_enter not associated to a BPF object\n");
4433 			goto out;
4434 		}
4435 
4436 		trace.bpf_obj = evsel->bpf_obj;
4437 
4438 		/*
4439 		 * If we have _just_ the augmenter event but don't have a
4440 		 * explicit --syscalls, then assume we want all strace-like
4441 		 * syscalls:
4442 		 */
4443 		if (!trace.trace_syscalls && trace__only_augmented_syscalls_evsels(&trace))
4444 			trace.trace_syscalls = true;
4445 		/*
4446 		 * So, if we have a syscall augmenter, but trace_syscalls, aka
4447 		 * strace-like syscall tracing is not set, then we need to trow
4448 		 * away the augmenter, i.e. all the events that were created
4449 		 * from that BPF object file.
4450 		 *
4451 		 * This is more to fix the current .perfconfig trace.add_events
4452 		 * style of setting up the strace-like eBPF based syscall point
4453 		 * payload augmenter.
4454 		 *
4455 		 * All this complexity will be avoided by adding an alternative
4456 		 * to trace.add_events in the form of
4457 		 * trace.bpf_augmented_syscalls, that will be only parsed if we
4458 		 * need it.
4459 		 *
4460 		 * .perfconfig trace.add_events is still useful if we want, for
4461 		 * instance, have msr_write.msr in some .perfconfig profile based
4462 		 * 'perf trace --config determinism.profile' mode, where for some
4463 		 * particular goal/workload type we want a set of events and
4464 		 * output mode (with timings, etc) instead of having to add
4465 		 * all via the command line.
4466 		 *
4467 		 * Also --config to specify an alternate .perfconfig file needs
4468 		 * to be implemented.
4469 		 */
4470 		if (!trace.trace_syscalls) {
4471 			trace__delete_augmented_syscalls(&trace);
4472 		} else {
4473 			trace__set_bpf_map_filtered_pids(&trace);
4474 			trace__set_bpf_map_syscalls(&trace);
4475 			trace.syscalls.unaugmented_prog = trace__find_bpf_program_by_title(&trace, "!raw_syscalls:unaugmented");
4476 		}
4477 	}
4478 
4479 	err = bpf__setup_stdout(trace.evlist);
4480 	if (err) {
4481 		bpf__strerror_setup_stdout(trace.evlist, err, bf, sizeof(bf));
4482 		pr_err("ERROR: Setup BPF stdout failed: %s\n", bf);
4483 		goto out;
4484 	}
4485 
4486 	err = -1;
4487 
4488 	if (map_dump_str) {
4489 		trace.dump.map = trace__find_bpf_map_by_name(&trace, map_dump_str);
4490 		if (trace.dump.map == NULL) {
4491 			pr_err("ERROR: BPF map \"%s\" not found\n", map_dump_str);
4492 			goto out;
4493 		}
4494 	}
4495 
4496 	if (trace.trace_pgfaults) {
4497 		trace.opts.sample_address = true;
4498 		trace.opts.sample_time = true;
4499 	}
4500 
4501 	if (trace.opts.mmap_pages == UINT_MAX)
4502 		mmap_pages_user_set = false;
4503 
4504 	if (trace.max_stack == UINT_MAX) {
4505 		trace.max_stack = input_name ? PERF_MAX_STACK_DEPTH : sysctl__max_stack();
4506 		max_stack_user_set = false;
4507 	}
4508 
4509 #ifdef HAVE_DWARF_UNWIND_SUPPORT
4510 	if ((trace.min_stack || max_stack_user_set) && !callchain_param.enabled) {
4511 		record_opts__parse_callchain(&trace.opts, &callchain_param, "dwarf", false);
4512 	}
4513 #endif
4514 
4515 	if (callchain_param.enabled) {
4516 		if (!mmap_pages_user_set && geteuid() == 0)
4517 			trace.opts.mmap_pages = perf_event_mlock_kb_in_pages() * 4;
4518 
4519 		symbol_conf.use_callchain = true;
4520 	}
4521 
4522 	if (trace.evlist->core.nr_entries > 0) {
4523 		evlist__set_default_evsel_handler(trace.evlist, trace__event_handler);
4524 		if (evlist__set_syscall_tp_fields(trace.evlist)) {
4525 			perror("failed to set syscalls:* tracepoint fields");
4526 			goto out;
4527 		}
4528 	}
4529 
4530 	if (trace.sort_events) {
4531 		ordered_events__init(&trace.oe.data, ordered_events__deliver_event, &trace);
4532 		ordered_events__set_copy_on_queue(&trace.oe.data, true);
4533 	}
4534 
4535 	/*
4536 	 * If we are augmenting syscalls, then combine what we put in the
4537 	 * __augmented_syscalls__ BPF map with what is in the
4538 	 * syscalls:sys_exit_FOO tracepoints, i.e. just like we do without BPF,
4539 	 * combining raw_syscalls:sys_enter with raw_syscalls:sys_exit.
4540 	 *
4541 	 * We'll switch to look at two BPF maps, one for sys_enter and the
4542 	 * other for sys_exit when we start augmenting the sys_exit paths with
4543 	 * buffers that are being copied from kernel to userspace, think 'read'
4544 	 * syscall.
4545 	 */
4546 	if (trace.syscalls.events.augmented) {
4547 		evlist__for_each_entry(trace.evlist, evsel) {
4548 			bool raw_syscalls_sys_exit = strcmp(perf_evsel__name(evsel), "raw_syscalls:sys_exit") == 0;
4549 
4550 			if (raw_syscalls_sys_exit) {
4551 				trace.raw_augmented_syscalls = true;
4552 				goto init_augmented_syscall_tp;
4553 			}
4554 
4555 			if (trace.syscalls.events.augmented->priv == NULL &&
4556 			    strstr(perf_evsel__name(evsel), "syscalls:sys_enter")) {
4557 				struct evsel *augmented = trace.syscalls.events.augmented;
4558 				if (perf_evsel__init_augmented_syscall_tp(augmented, evsel) ||
4559 				    perf_evsel__init_augmented_syscall_tp_args(augmented))
4560 					goto out;
4561 				/*
4562 				 * Augmented is __augmented_syscalls__ BPF_OUTPUT event
4563 				 * Above we made sure we can get from the payload the tp fields
4564 				 * that we get from syscalls:sys_enter tracefs format file.
4565 				 */
4566 				augmented->handler = trace__sys_enter;
4567 				/*
4568 				 * Now we do the same for the *syscalls:sys_enter event so that
4569 				 * if we handle it directly, i.e. if the BPF prog returns 0 so
4570 				 * as not to filter it, then we'll handle it just like we would
4571 				 * for the BPF_OUTPUT one:
4572 				 */
4573 				if (perf_evsel__init_augmented_syscall_tp(evsel, evsel) ||
4574 				    perf_evsel__init_augmented_syscall_tp_args(evsel))
4575 					goto out;
4576 				evsel->handler = trace__sys_enter;
4577 			}
4578 
4579 			if (strstarts(perf_evsel__name(evsel), "syscalls:sys_exit_")) {
4580 				struct syscall_tp *sc;
4581 init_augmented_syscall_tp:
4582 				if (perf_evsel__init_augmented_syscall_tp(evsel, evsel))
4583 					goto out;
4584 				sc = evsel->priv;
4585 				/*
4586 				 * For now with BPF raw_augmented we hook into
4587 				 * raw_syscalls:sys_enter and there we get all
4588 				 * 6 syscall args plus the tracepoint common
4589 				 * fields and the syscall_nr (another long).
4590 				 * So we check if that is the case and if so
4591 				 * don't look after the sc->args_size but
4592 				 * always after the full raw_syscalls:sys_enter
4593 				 * payload, which is fixed.
4594 				 *
4595 				 * We'll revisit this later to pass
4596 				 * s->args_size to the BPF augmenter (now
4597 				 * tools/perf/examples/bpf/augmented_raw_syscalls.c,
4598 				 * so that it copies only what we need for each
4599 				 * syscall, like what happens when we use
4600 				 * syscalls:sys_enter_NAME, so that we reduce
4601 				 * the kernel/userspace traffic to just what is
4602 				 * needed for each syscall.
4603 				 */
4604 				if (trace.raw_augmented_syscalls)
4605 					trace.raw_augmented_syscalls_args_size = (6 + 1) * sizeof(long) + sc->id.offset;
4606 				perf_evsel__init_augmented_syscall_tp_ret(evsel);
4607 				evsel->handler = trace__sys_exit;
4608 			}
4609 		}
4610 	}
4611 
4612 	if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
4613 		return trace__record(&trace, argc-1, &argv[1]);
4614 
4615 	/* summary_only implies summary option, but don't overwrite summary if set */
4616 	if (trace.summary_only)
4617 		trace.summary = trace.summary_only;
4618 
4619 	if (output_name != NULL) {
4620 		err = trace__open_output(&trace, output_name);
4621 		if (err < 0) {
4622 			perror("failed to create output file");
4623 			goto out;
4624 		}
4625 	}
4626 
4627 	err = evswitch__init(&trace.evswitch, trace.evlist, stderr);
4628 	if (err)
4629 		goto out_close;
4630 
4631 	err = target__validate(&trace.opts.target);
4632 	if (err) {
4633 		target__strerror(&trace.opts.target, err, bf, sizeof(bf));
4634 		fprintf(trace.output, "%s", bf);
4635 		goto out_close;
4636 	}
4637 
4638 	err = target__parse_uid(&trace.opts.target);
4639 	if (err) {
4640 		target__strerror(&trace.opts.target, err, bf, sizeof(bf));
4641 		fprintf(trace.output, "%s", bf);
4642 		goto out_close;
4643 	}
4644 
4645 	if (!argc && target__none(&trace.opts.target))
4646 		trace.opts.target.system_wide = true;
4647 
4648 	if (input_name)
4649 		err = trace__replay(&trace);
4650 	else
4651 		err = trace__run(&trace, argc, argv);
4652 
4653 out_close:
4654 	if (output_name != NULL)
4655 		fclose(trace.output);
4656 out:
4657 	zfree(&trace.perfconfig_events);
4658 	return err;
4659 }
4660