]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - tools/perf/builtin-trace.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
perf trace: Handle multiple threads better wrt syscalls being intermixed
[mirror_ubuntu-bionic-kernel.git] / tools / perf / builtin-trace.c
CommitLineData
4e319027 1#include <traceevent/event-parse.h>
514f1c67 2#include "builtin.h"
752fde44 3#include "util/color.h"
7c304ee0 4#include "util/debug.h"
514f1c67 5#include "util/evlist.h"
752fde44 6#include "util/machine.h"
6810fc91 7#include "util/session.h"
752fde44 8#include "util/thread.h"
514f1c67 9#include "util/parse-options.h"
2ae3a312 10#include "util/strlist.h"
bdc89661 11#include "util/intlist.h"
514f1c67 12#include "util/thread_map.h"
bf2575c1 13#include "util/stat.h"
97978b3e 14#include "trace-event.h"
9aca7f17 15#include "util/parse-events.h"
514f1c67
ACM		 16
17#include <libaudit.h>
18#include <stdlib.h>
49af9e93 19#include <sys/eventfd.h>
ae685380 20#include <sys/mman.h>
f9da0b0c 21#include <linux/futex.h>
514f1c67 22
456857bd
IM		 23/* For older distros: */
24#ifndef MAP_STACK
25# define MAP_STACK 0x20000
26#endif
27
28#ifndef MADV_HWPOISON
29# define MADV_HWPOISON 100
30#endif
31
32#ifndef MADV_MERGEABLE
33# define MADV_MERGEABLE 12
34#endif
35
36#ifndef MADV_UNMERGEABLE
37# define MADV_UNMERGEABLE 13
38#endif
39
79d26a6a
BH		 40#ifndef EFD_SEMAPHORE
41# define EFD_SEMAPHORE 1
42#endif
43
77170988
ACM		 44struct tp_field {
45 int offset;
46 union {
47 u64 (*integer)(struct tp_field *field, struct perf_sample *sample);
48 void *(*pointer)(struct tp_field *field, struct perf_sample *sample);
49 };
50};
51
52#define TP_UINT_FIELD(bits) \
53static u64 tp_field__u##bits(struct tp_field *field, struct perf_sample *sample) \
54{ \
55 return *(u##bits *)(sample->raw_data + field->offset); \
56}
57
58TP_UINT_FIELD(8);
59TP_UINT_FIELD(16);
60TP_UINT_FIELD(32);
61TP_UINT_FIELD(64);
62
63#define TP_UINT_FIELD__SWAPPED(bits) \
64static u64 tp_field__swapped_u##bits(struct tp_field *field, struct perf_sample *sample) \
65{ \
66 u##bits value = *(u##bits *)(sample->raw_data + field->offset); \
67 return bswap_##bits(value);\
68}
69
70TP_UINT_FIELD__SWAPPED(16);
71TP_UINT_FIELD__SWAPPED(32);
72TP_UINT_FIELD__SWAPPED(64);
73
74static int tp_field__init_uint(struct tp_field *field,
75 struct format_field *format_field,
76 bool needs_swap)
77{
78 field->offset = format_field->offset;
79
80 switch (format_field->size) {
81 case 1:
82 field->integer = tp_field__u8;
83 break;
84 case 2:
85 field->integer = needs_swap ? tp_field__swapped_u16 : tp_field__u16;
86 break;
87 case 4:
88 field->integer = needs_swap ? tp_field__swapped_u32 : tp_field__u32;
89 break;
90 case 8:
91 field->integer = needs_swap ? tp_field__swapped_u64 : tp_field__u64;
92 break;
93 default:
94 return -1;
95 }
96
97 return 0;
98}
99
100static void *tp_field__ptr(struct tp_field *field, struct perf_sample *sample)
101{
102 return sample->raw_data + field->offset;
103}
104
105static int tp_field__init_ptr(struct tp_field *field, struct format_field *format_field)
106{
107 field->offset = format_field->offset;
108 field->pointer = tp_field__ptr;
109 return 0;
110}
111
112struct syscall_tp {
113 struct tp_field id;
114 union {
115 struct tp_field args, ret;
116 };
117};
118
119static int perf_evsel__init_tp_uint_field(struct perf_evsel *evsel,
120 struct tp_field *field,
121 const char *name)
122{
123 struct format_field *format_field = perf_evsel__field(evsel, name);
124
125 if (format_field == NULL)
126 return -1;
127
128 return tp_field__init_uint(field, format_field, evsel->needs_swap);
129}
130
131#define perf_evsel__init_sc_tp_uint_field(evsel, name) \
132 ({ struct syscall_tp *sc = evsel->priv;\
133 perf_evsel__init_tp_uint_field(evsel, &sc->name, #name); })
134
135static int perf_evsel__init_tp_ptr_field(struct perf_evsel *evsel,
136 struct tp_field *field,
137 const char *name)
138{
139 struct format_field *format_field = perf_evsel__field(evsel, name);
140
141 if (format_field == NULL)
142 return -1;
143
144 return tp_field__init_ptr(field, format_field);
145}
146
147#define perf_evsel__init_sc_tp_ptr_field(evsel, name) \
148 ({ struct syscall_tp *sc = evsel->priv;\
149 perf_evsel__init_tp_ptr_field(evsel, &sc->name, #name); })
150
151static void perf_evsel__delete_priv(struct perf_evsel *evsel)
152{
04662523 153 zfree(&evsel->priv);
77170988
ACM		 154 perf_evsel__delete(evsel);
155}
156
96695d44
NK		 157static int perf_evsel__init_syscall_tp(struct perf_evsel *evsel, void *handler)
158{
159 evsel->priv = malloc(sizeof(struct syscall_tp));
160 if (evsel->priv != NULL) {
161 if (perf_evsel__init_sc_tp_uint_field(evsel, id))
162 goto out_delete;
163
164 evsel->handler = handler;
165 return 0;
166 }
167
168 return -ENOMEM;
169
170out_delete:
04662523 171 zfree(&evsel->priv);
96695d44
NK		 172 return -ENOENT;
173}
174
ef503831 175static struct perf_evsel *perf_evsel__syscall_newtp(const char *direction, void *handler)
77170988 176{
ef503831 177 struct perf_evsel *evsel = perf_evsel__newtp("raw_syscalls", direction);
77170988 178
9aca7f17
DA		 179 /* older kernel (e.g., RHEL6) use syscalls:{enter,exit} */
180 if (evsel == NULL)
181 evsel = perf_evsel__newtp("syscalls", direction);
182
77170988 183 if (evsel) {
96695d44 184 if (perf_evsel__init_syscall_tp(evsel, handler))
77170988 185 goto out_delete;
77170988
ACM		 186 }
187
188 return evsel;
189
190out_delete:
191 perf_evsel__delete_priv(evsel);
192 return NULL;
193}
194
195#define perf_evsel__sc_tp_uint(evsel, name, sample) \
196 ({ struct syscall_tp *fields = evsel->priv; \
197 fields->name.integer(&fields->name, sample); })
198
199#define perf_evsel__sc_tp_ptr(evsel, name, sample) \
200 ({ struct syscall_tp *fields = evsel->priv; \
201 fields->name.pointer(&fields->name, sample); })
202
203static int perf_evlist__add_syscall_newtp(struct perf_evlist *evlist,
204 void *sys_enter_handler,
205 void *sys_exit_handler)
206{
207 int ret = -1;
77170988
ACM		 208 struct perf_evsel *sys_enter, *sys_exit;
209
ef503831 210 sys_enter = perf_evsel__syscall_newtp("sys_enter", sys_enter_handler);
77170988
ACM		 211 if (sys_enter == NULL)
212 goto out;
213
214 if (perf_evsel__init_sc_tp_ptr_field(sys_enter, args))
215 goto out_delete_sys_enter;
216
ef503831 217 sys_exit = perf_evsel__syscall_newtp("sys_exit", sys_exit_handler);
77170988
ACM		 218 if (sys_exit == NULL)
219 goto out_delete_sys_enter;
220
221 if (perf_evsel__init_sc_tp_uint_field(sys_exit, ret))
222 goto out_delete_sys_exit;
223
224 perf_evlist__add(evlist, sys_enter);
225 perf_evlist__add(evlist, sys_exit);
226
227 ret = 0;
228out:
229 return ret;
230
231out_delete_sys_exit:
232 perf_evsel__delete_priv(sys_exit);
233out_delete_sys_enter:
234 perf_evsel__delete_priv(sys_enter);
235 goto out;
236}
237
238
01533e97
ACM		 239struct syscall_arg {
240 unsigned long val;
75b757ca
ACM		 241 struct thread *thread;
242 struct trace *trace;
1f115cb7 243 void *parm;
01533e97
ACM		 244 u8 idx;
245 u8 mask;
246};
247
1f115cb7 248struct strarray {
03e3adc9 249 int offset;
1f115cb7
ACM		 250 int nr_entries;
251 const char **entries;
252};
253
254#define DEFINE_STRARRAY(array) struct strarray strarray__##array = { \
255 .nr_entries = ARRAY_SIZE(array), \
256 .entries = array, \
257}
258
03e3adc9
ACM		 259#define DEFINE_STRARRAY_OFFSET(array, off) struct strarray strarray__##array = { \
260 .offset = off, \
261 .nr_entries = ARRAY_SIZE(array), \
262 .entries = array, \
263}
264
975b7c2f
ACM		 265static size_t __syscall_arg__scnprintf_strarray(char *bf, size_t size,
266 const char *intfmt,
267 struct syscall_arg *arg)
1f115cb7 268{
1f115cb7 269 struct strarray *sa = arg->parm;
03e3adc9 270 int idx = arg->val - sa->offset;
1f115cb7
ACM		 271
272 if (idx < 0 || idx >= sa->nr_entries)
975b7c2f 273 return scnprintf(bf, size, intfmt, arg->val);
1f115cb7
ACM		 274
275 return scnprintf(bf, size, "%s", sa->entries[idx]);
276}
277
975b7c2f
ACM		 278static size_t syscall_arg__scnprintf_strarray(char *bf, size_t size,
279 struct syscall_arg *arg)
280{
281 return __syscall_arg__scnprintf_strarray(bf, size, "%d", arg);
282}
283
1f115cb7
ACM		 284#define SCA_STRARRAY syscall_arg__scnprintf_strarray
285
844ae5b4
ACM		 286#if defined(__i386__) || defined(__x86_64__)
287/*
288 * FIXME: Make this available to all arches as soon as the ioctl beautifier
289 * gets rewritten to support all arches.
290 */
78645cf3
ACM		 291static size_t syscall_arg__scnprintf_strhexarray(char *bf, size_t size,
292 struct syscall_arg *arg)
293{
294 return __syscall_arg__scnprintf_strarray(bf, size, "%#x", arg);
295}
296
297#define SCA_STRHEXARRAY syscall_arg__scnprintf_strhexarray
844ae5b4 298#endif /* defined(__i386__) || defined(__x86_64__) */
78645cf3 299
75b757ca
ACM		 300static size_t syscall_arg__scnprintf_fd(char *bf, size_t size,
301 struct syscall_arg *arg);
302
303#define SCA_FD syscall_arg__scnprintf_fd
304
305static size_t syscall_arg__scnprintf_fd_at(char *bf, size_t size,
306 struct syscall_arg *arg)
307{
308 int fd = arg->val;
309
310 if (fd == AT_FDCWD)
311 return scnprintf(bf, size, "CWD");
312
313 return syscall_arg__scnprintf_fd(bf, size, arg);
314}
315
316#define SCA_FDAT syscall_arg__scnprintf_fd_at
317
318static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
319 struct syscall_arg *arg);
320
321#define SCA_CLOSE_FD syscall_arg__scnprintf_close_fd
322
6e7eeb51 323static size_t syscall_arg__scnprintf_hex(char *bf, size_t size,
01533e97 324 struct syscall_arg *arg)
13d4ff3e 325{
01533e97 326 return scnprintf(bf, size, "%#lx", arg->val);
13d4ff3e
ACM		 327}
328
beccb2b5
ACM		 329#define SCA_HEX syscall_arg__scnprintf_hex
330
6e7eeb51 331static size_t syscall_arg__scnprintf_mmap_prot(char *bf, size_t size,
01533e97 332 struct syscall_arg *arg)
ae685380 333{
01533e97 334 int printed = 0, prot = arg->val;
ae685380
ACM		 335
336 if (prot == PROT_NONE)
337 return scnprintf(bf, size, "NONE");
338#define P_MMAP_PROT(n) \
339 if (prot & PROT_##n) { \
340 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
341 prot &= ~PROT_##n; \
342 }
343
344 P_MMAP_PROT(EXEC);
345 P_MMAP_PROT(READ);
346 P_MMAP_PROT(WRITE);
347#ifdef PROT_SEM
348 P_MMAP_PROT(SEM);
349#endif
350 P_MMAP_PROT(GROWSDOWN);
351 P_MMAP_PROT(GROWSUP);
352#undef P_MMAP_PROT
353
354 if (prot)
355 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", prot);
356
357 return printed;
358}
359
360#define SCA_MMAP_PROT syscall_arg__scnprintf_mmap_prot
361
6e7eeb51 362static size_t syscall_arg__scnprintf_mmap_flags(char *bf, size_t size,
01533e97 363 struct syscall_arg *arg)
941557e0 364{
01533e97 365 int printed = 0, flags = arg->val;
941557e0
ACM		 366
367#define P_MMAP_FLAG(n) \
368 if (flags & MAP_##n) { \
369 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
370 flags &= ~MAP_##n; \
371 }
372
373 P_MMAP_FLAG(SHARED);
374 P_MMAP_FLAG(PRIVATE);
41817815 375#ifdef MAP_32BIT
941557e0 376 P_MMAP_FLAG(32BIT);
41817815 377#endif
941557e0
ACM		 378 P_MMAP_FLAG(ANONYMOUS);
379 P_MMAP_FLAG(DENYWRITE);
380 P_MMAP_FLAG(EXECUTABLE);
381 P_MMAP_FLAG(FILE);
382 P_MMAP_FLAG(FIXED);
383 P_MMAP_FLAG(GROWSDOWN);
f2935f3e 384#ifdef MAP_HUGETLB
941557e0 385 P_MMAP_FLAG(HUGETLB);
f2935f3e 386#endif
941557e0
ACM		 387 P_MMAP_FLAG(LOCKED);
388 P_MMAP_FLAG(NONBLOCK);
389 P_MMAP_FLAG(NORESERVE);
390 P_MMAP_FLAG(POPULATE);
391 P_MMAP_FLAG(STACK);
392#ifdef MAP_UNINITIALIZED
393 P_MMAP_FLAG(UNINITIALIZED);
394#endif
395#undef P_MMAP_FLAG
396
397 if (flags)
398 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
399
400 return printed;
401}
402
403#define SCA_MMAP_FLAGS syscall_arg__scnprintf_mmap_flags
404
86998dda
AS		 405static size_t syscall_arg__scnprintf_mremap_flags(char *bf, size_t size,
406 struct syscall_arg *arg)
407{
408 int printed = 0, flags = arg->val;
409
410#define P_MREMAP_FLAG(n) \
411 if (flags & MREMAP_##n) { \
412 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
413 flags &= ~MREMAP_##n; \
414 }
415
416 P_MREMAP_FLAG(MAYMOVE);
417#ifdef MREMAP_FIXED
418 P_MREMAP_FLAG(FIXED);
419#endif
420#undef P_MREMAP_FLAG
421
422 if (flags)
423 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
424
425 return printed;
426}
427
428#define SCA_MREMAP_FLAGS syscall_arg__scnprintf_mremap_flags
429
6e7eeb51 430static size_t syscall_arg__scnprintf_madvise_behavior(char *bf, size_t size,
01533e97 431 struct syscall_arg *arg)
9e9716d1 432{
01533e97 433 int behavior = arg->val;
9e9716d1
ACM		 434
435 switch (behavior) {
436#define P_MADV_BHV(n) case MADV_##n: return scnprintf(bf, size, #n)
437 P_MADV_BHV(NORMAL);
438 P_MADV_BHV(RANDOM);
439 P_MADV_BHV(SEQUENTIAL);
440 P_MADV_BHV(WILLNEED);
441 P_MADV_BHV(DONTNEED);
442 P_MADV_BHV(REMOVE);
443 P_MADV_BHV(DONTFORK);
444 P_MADV_BHV(DOFORK);
445 P_MADV_BHV(HWPOISON);
446#ifdef MADV_SOFT_OFFLINE
447 P_MADV_BHV(SOFT_OFFLINE);
448#endif
449 P_MADV_BHV(MERGEABLE);
450 P_MADV_BHV(UNMERGEABLE);
f2935f3e 451#ifdef MADV_HUGEPAGE
9e9716d1 452 P_MADV_BHV(HUGEPAGE);
f2935f3e
DA		 453#endif
454#ifdef MADV_NOHUGEPAGE
9e9716d1 455 P_MADV_BHV(NOHUGEPAGE);
f2935f3e 456#endif
9e9716d1
ACM		 457#ifdef MADV_DONTDUMP
458 P_MADV_BHV(DONTDUMP);
459#endif
460#ifdef MADV_DODUMP
461 P_MADV_BHV(DODUMP);
462#endif
463#undef P_MADV_PHV
464 default: break;
465 }
466
467 return scnprintf(bf, size, "%#x", behavior);
468}
469
470#define SCA_MADV_BHV syscall_arg__scnprintf_madvise_behavior
471
5cea6ff2
ACM		 472static size_t syscall_arg__scnprintf_flock(char *bf, size_t size,
473 struct syscall_arg *arg)
474{
475 int printed = 0, op = arg->val;
476
477 if (op == 0)
478 return scnprintf(bf, size, "NONE");
479#define P_CMD(cmd) \
480 if ((op & LOCK_##cmd) == LOCK_##cmd) { \
481 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #cmd); \
482 op &= ~LOCK_##cmd; \
483 }
484
485 P_CMD(SH);
486 P_CMD(EX);
487 P_CMD(NB);
488 P_CMD(UN);
489 P_CMD(MAND);
490 P_CMD(RW);
491 P_CMD(READ);
492 P_CMD(WRITE);
493#undef P_OP
494
495 if (op)
496 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", op);
497
498 return printed;
499}
500
501#define SCA_FLOCK syscall_arg__scnprintf_flock
502
01533e97 503static size_t syscall_arg__scnprintf_futex_op(char *bf, size_t size, struct syscall_arg *arg)
f9da0b0c
ACM		 504{
505 enum syscall_futex_args {
506 SCF_UADDR = (1 << 0),
507 SCF_OP = (1 << 1),
508 SCF_VAL = (1 << 2),
509 SCF_TIMEOUT = (1 << 3),
510 SCF_UADDR2 = (1 << 4),
511 SCF_VAL3 = (1 << 5),
512 };
01533e97 513 int op = arg->val;
f9da0b0c
ACM		 514 int cmd = op & FUTEX_CMD_MASK;
515 size_t printed = 0;
516
517 switch (cmd) {
518#define P_FUTEX_OP(n) case FUTEX_##n: printed = scnprintf(bf, size, #n);
01533e97
ACM		 519 P_FUTEX_OP(WAIT); arg->mask |= SCF_VAL3|SCF_UADDR2; break;
520 P_FUTEX_OP(WAKE); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
521 P_FUTEX_OP(FD); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
522 P_FUTEX_OP(REQUEUE); arg->mask |= SCF_VAL3|SCF_TIMEOUT; break;
523 P_FUTEX_OP(CMP_REQUEUE); arg->mask |= SCF_TIMEOUT; break;
524 P_FUTEX_OP(CMP_REQUEUE_PI); arg->mask |= SCF_TIMEOUT; break;
f9da0b0c 525 P_FUTEX_OP(WAKE_OP); break;
01533e97
ACM		 526 P_FUTEX_OP(LOCK_PI); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
527 P_FUTEX_OP(UNLOCK_PI); arg->mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
528 P_FUTEX_OP(TRYLOCK_PI); arg->mask |= SCF_VAL3|SCF_UADDR2; break;
529 P_FUTEX_OP(WAIT_BITSET); arg->mask |= SCF_UADDR2; break;
530 P_FUTEX_OP(WAKE_BITSET); arg->mask |= SCF_UADDR2; break;
f9da0b0c
ACM		 531 P_FUTEX_OP(WAIT_REQUEUE_PI); break;
532 default: printed = scnprintf(bf, size, "%#x", cmd); break;
533 }
534
535 if (op & FUTEX_PRIVATE_FLAG)
536 printed += scnprintf(bf + printed, size - printed, "|PRIV");
537
538 if (op & FUTEX_CLOCK_REALTIME)
539 printed += scnprintf(bf + printed, size - printed, "|CLKRT");
540
541 return printed;
542}
543
efe6b882
ACM		 544#define SCA_FUTEX_OP syscall_arg__scnprintf_futex_op
545
03e3adc9
ACM		 546static const char *epoll_ctl_ops[] = { "ADD", "DEL", "MOD", };
547static DEFINE_STRARRAY_OFFSET(epoll_ctl_ops, 1);
eac032c5 548
1f115cb7
ACM		 549static const char *itimers[] = { "REAL", "VIRTUAL", "PROF", };
550static DEFINE_STRARRAY(itimers);
551
efe6b882
ACM		 552static const char *whences[] = { "SET", "CUR", "END",
553#ifdef SEEK_DATA
554"DATA",
555#endif
556#ifdef SEEK_HOLE
557"HOLE",
558#endif
559};
560static DEFINE_STRARRAY(whences);
f9da0b0c 561
80f587d5
ACM		 562static const char *fcntl_cmds[] = {
563 "DUPFD", "GETFD", "SETFD", "GETFL", "SETFL", "GETLK", "SETLK",
564 "SETLKW", "SETOWN", "GETOWN", "SETSIG", "GETSIG", "F_GETLK64",
565 "F_SETLK64", "F_SETLKW64", "F_SETOWN_EX", "F_GETOWN_EX",
566 "F_GETOWNER_UIDS",
567};
568static DEFINE_STRARRAY(fcntl_cmds);
569
c045bf02
ACM		 570static const char *rlimit_resources[] = {
571 "CPU", "FSIZE", "DATA", "STACK", "CORE", "RSS", "NPROC", "NOFILE",
572 "MEMLOCK", "AS", "LOCKS", "SIGPENDING", "MSGQUEUE", "NICE", "RTPRIO",
573 "RTTIME",
574};
575static DEFINE_STRARRAY(rlimit_resources);
576
eb5b1b14
ACM		 577static const char *sighow[] = { "BLOCK", "UNBLOCK", "SETMASK", };
578static DEFINE_STRARRAY(sighow);
579
4f8c1b74
DA		 580static const char *clockid[] = {
581 "REALTIME", "MONOTONIC", "PROCESS_CPUTIME_ID", "THREAD_CPUTIME_ID",
582 "MONOTONIC_RAW", "REALTIME_COARSE", "MONOTONIC_COARSE",
583};
584static DEFINE_STRARRAY(clockid);
585
e10bce81
ACM		 586static const char *socket_families[] = {
587 "UNSPEC", "LOCAL", "INET", "AX25", "IPX", "APPLETALK", "NETROM",
588 "BRIDGE", "ATMPVC", "X25", "INET6", "ROSE", "DECnet", "NETBEUI",
589 "SECURITY", "KEY", "NETLINK", "PACKET", "ASH", "ECONET", "ATMSVC",
590 "RDS", "SNA", "IRDA", "PPPOX", "WANPIPE", "LLC", "IB", "CAN", "TIPC",
591 "BLUETOOTH", "IUCV", "RXRPC", "ISDN", "PHONET", "IEEE802154", "CAIF",
592 "ALG", "NFC", "VSOCK",
593};
594static DEFINE_STRARRAY(socket_families);
595
a28b24b2
ACM		 596#ifndef SOCK_TYPE_MASK
597#define SOCK_TYPE_MASK 0xf
598#endif
599
600static size_t syscall_arg__scnprintf_socket_type(char *bf, size_t size,
601 struct syscall_arg *arg)
602{
603 size_t printed;
604 int type = arg->val,
605 flags = type & ~SOCK_TYPE_MASK;
606
607 type &= SOCK_TYPE_MASK;
608 /*
609 * Can't use a strarray, MIPS may override for ABI reasons.
610 */
611 switch (type) {
612#define P_SK_TYPE(n) case SOCK_##n: printed = scnprintf(bf, size, #n); break;
613 P_SK_TYPE(STREAM);
614 P_SK_TYPE(DGRAM);
615 P_SK_TYPE(RAW);
616 P_SK_TYPE(RDM);
617 P_SK_TYPE(SEQPACKET);
618 P_SK_TYPE(DCCP);
619 P_SK_TYPE(PACKET);
620#undef P_SK_TYPE
621 default:
622 printed = scnprintf(bf, size, "%#x", type);
623 }
624
625#define P_SK_FLAG(n) \
626 if (flags & SOCK_##n) { \
627 printed += scnprintf(bf + printed, size - printed, "|%s", #n); \
628 flags &= ~SOCK_##n; \
629 }
630
631 P_SK_FLAG(CLOEXEC);
632 P_SK_FLAG(NONBLOCK);
633#undef P_SK_FLAG
634
635 if (flags)
636 printed += scnprintf(bf + printed, size - printed, "|%#x", flags);
637
638 return printed;
639}
640
641#define SCA_SK_TYPE syscall_arg__scnprintf_socket_type
642
b2cc99fd
ACM		 643#ifndef MSG_PROBE
644#define MSG_PROBE 0x10
645#endif
b6e8f8f4
DA		 646#ifndef MSG_WAITFORONE
647#define MSG_WAITFORONE 0x10000
648#endif
b2cc99fd
ACM		 649#ifndef MSG_SENDPAGE_NOTLAST
650#define MSG_SENDPAGE_NOTLAST 0x20000
651#endif
652#ifndef MSG_FASTOPEN
653#define MSG_FASTOPEN 0x20000000
654#endif
655
656static size_t syscall_arg__scnprintf_msg_flags(char *bf, size_t size,
657 struct syscall_arg *arg)
658{
659 int printed = 0, flags = arg->val;
660
661 if (flags == 0)
662 return scnprintf(bf, size, "NONE");
663#define P_MSG_FLAG(n) \
664 if (flags & MSG_##n) { \
665 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
666 flags &= ~MSG_##n; \
667 }
668
669 P_MSG_FLAG(OOB);
670 P_MSG_FLAG(PEEK);
671 P_MSG_FLAG(DONTROUTE);
672 P_MSG_FLAG(TRYHARD);
673 P_MSG_FLAG(CTRUNC);
674 P_MSG_FLAG(PROBE);
675 P_MSG_FLAG(TRUNC);
676 P_MSG_FLAG(DONTWAIT);
677 P_MSG_FLAG(EOR);
678 P_MSG_FLAG(WAITALL);
679 P_MSG_FLAG(FIN);
680 P_MSG_FLAG(SYN);
681 P_MSG_FLAG(CONFIRM);
682 P_MSG_FLAG(RST);
683 P_MSG_FLAG(ERRQUEUE);
684 P_MSG_FLAG(NOSIGNAL);
685 P_MSG_FLAG(MORE);
686 P_MSG_FLAG(WAITFORONE);
687 P_MSG_FLAG(SENDPAGE_NOTLAST);
688 P_MSG_FLAG(FASTOPEN);
689 P_MSG_FLAG(CMSG_CLOEXEC);
690#undef P_MSG_FLAG
691
692 if (flags)
693 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
694
695 return printed;
696}
697
698#define SCA_MSG_FLAGS syscall_arg__scnprintf_msg_flags
699
51108999
ACM		 700static size_t syscall_arg__scnprintf_access_mode(char *bf, size_t size,
701 struct syscall_arg *arg)
702{
703 size_t printed = 0;
704 int mode = arg->val;
705
706 if (mode == F_OK) /* 0 */
707 return scnprintf(bf, size, "F");
708#define P_MODE(n) \
709 if (mode & n##_OK) { \
710 printed += scnprintf(bf + printed, size - printed, "%s", #n); \
711 mode &= ~n##_OK; \
712 }
713
714 P_MODE(R);
715 P_MODE(W);
716 P_MODE(X);
717#undef P_MODE
718
719 if (mode)
720 printed += scnprintf(bf + printed, size - printed, "|%#x", mode);
721
722 return printed;
723}
724
725#define SCA_ACCMODE syscall_arg__scnprintf_access_mode
726
be65a89a 727static size_t syscall_arg__scnprintf_open_flags(char *bf, size_t size,
01533e97 728 struct syscall_arg *arg)
be65a89a 729{
01533e97 730 int printed = 0, flags = arg->val;
be65a89a
ACM		 731
732 if (!(flags & O_CREAT))
01533e97 733 arg->mask |= 1 << (arg->idx + 1); /* Mask the mode parm */
be65a89a
ACM		 734
735 if (flags == 0)
736 return scnprintf(bf, size, "RDONLY");
737#define P_FLAG(n) \
738 if (flags & O_##n) { \
739 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
740 flags &= ~O_##n; \
741 }
742
743 P_FLAG(APPEND);
744 P_FLAG(ASYNC);
745 P_FLAG(CLOEXEC);
746 P_FLAG(CREAT);
747 P_FLAG(DIRECT);
748 P_FLAG(DIRECTORY);
749 P_FLAG(EXCL);
750 P_FLAG(LARGEFILE);
751 P_FLAG(NOATIME);
752 P_FLAG(NOCTTY);
753#ifdef O_NONBLOCK
754 P_FLAG(NONBLOCK);
755#elif O_NDELAY
756 P_FLAG(NDELAY);
757#endif
758#ifdef O_PATH
759 P_FLAG(PATH);
760#endif
761 P_FLAG(RDWR);
762#ifdef O_DSYNC
763 if ((flags & O_SYNC) == O_SYNC)
764 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", "SYNC");
765 else {
766 P_FLAG(DSYNC);
767 }
768#else
769 P_FLAG(SYNC);
770#endif
771 P_FLAG(TRUNC);
772 P_FLAG(WRONLY);
773#undef P_FLAG
774
775 if (flags)
776 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
777
778 return printed;
779}
780
781#define SCA_OPEN_FLAGS syscall_arg__scnprintf_open_flags
782
49af9e93
ACM		 783static size_t syscall_arg__scnprintf_eventfd_flags(char *bf, size_t size,
784 struct syscall_arg *arg)
785{
786 int printed = 0, flags = arg->val;
787
788 if (flags == 0)
789 return scnprintf(bf, size, "NONE");
790#define P_FLAG(n) \
791 if (flags & EFD_##n) { \
792 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
793 flags &= ~EFD_##n; \
794 }
795
796 P_FLAG(SEMAPHORE);
797 P_FLAG(CLOEXEC);
798 P_FLAG(NONBLOCK);
799#undef P_FLAG
800
801 if (flags)
802 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
803
804 return printed;
805}
806
807#define SCA_EFD_FLAGS syscall_arg__scnprintf_eventfd_flags
808
46cce19b
ACM		 809static size_t syscall_arg__scnprintf_pipe_flags(char *bf, size_t size,
810 struct syscall_arg *arg)
811{
812 int printed = 0, flags = arg->val;
813
814#define P_FLAG(n) \
815 if (flags & O_##n) { \
816 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
817 flags &= ~O_##n; \
818 }
819
820 P_FLAG(CLOEXEC);
821 P_FLAG(NONBLOCK);
822#undef P_FLAG
823
824 if (flags)
825 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
826
827 return printed;
828}
829
830#define SCA_PIPE_FLAGS syscall_arg__scnprintf_pipe_flags
831
8bad5b0a
ACM		 832static size_t syscall_arg__scnprintf_signum(char *bf, size_t size, struct syscall_arg *arg)
833{
834 int sig = arg->val;
835
836 switch (sig) {
837#define P_SIGNUM(n) case SIG##n: return scnprintf(bf, size, #n)
838 P_SIGNUM(HUP);
839 P_SIGNUM(INT);
840 P_SIGNUM(QUIT);
841 P_SIGNUM(ILL);
842 P_SIGNUM(TRAP);
843 P_SIGNUM(ABRT);
844 P_SIGNUM(BUS);
845 P_SIGNUM(FPE);
846 P_SIGNUM(KILL);
847 P_SIGNUM(USR1);
848 P_SIGNUM(SEGV);
849 P_SIGNUM(USR2);
850 P_SIGNUM(PIPE);
851 P_SIGNUM(ALRM);
852 P_SIGNUM(TERM);
8bad5b0a
ACM		 853 P_SIGNUM(CHLD);
854 P_SIGNUM(CONT);
855 P_SIGNUM(STOP);
856 P_SIGNUM(TSTP);
857 P_SIGNUM(TTIN);
858 P_SIGNUM(TTOU);
859 P_SIGNUM(URG);
860 P_SIGNUM(XCPU);
861 P_SIGNUM(XFSZ);
862 P_SIGNUM(VTALRM);
863 P_SIGNUM(PROF);
864 P_SIGNUM(WINCH);
865 P_SIGNUM(IO);
866 P_SIGNUM(PWR);
867 P_SIGNUM(SYS);
02c5bb4a
BH		 868#ifdef SIGEMT
869 P_SIGNUM(EMT);
870#endif
871#ifdef SIGSTKFLT
872 P_SIGNUM(STKFLT);
873#endif
874#ifdef SIGSWI
875 P_SIGNUM(SWI);
876#endif
8bad5b0a
ACM		 877 default: break;
878 }
879
880 return scnprintf(bf, size, "%#x", sig);
881}
882
883#define SCA_SIGNUM syscall_arg__scnprintf_signum
884
844ae5b4
ACM		 885#if defined(__i386__) || defined(__x86_64__)
886/*
887 * FIXME: Make this available to all arches.
888 */
78645cf3
ACM		 889#define TCGETS 0x5401
890
891static const char *tioctls[] = {
892 "TCGETS", "TCSETS", "TCSETSW", "TCSETSF", "TCGETA", "TCSETA", "TCSETAW",
893 "TCSETAF", "TCSBRK", "TCXONC", "TCFLSH", "TIOCEXCL", "TIOCNXCL",
894 "TIOCSCTTY", "TIOCGPGRP", "TIOCSPGRP", "TIOCOUTQ", "TIOCSTI",
895 "TIOCGWINSZ", "TIOCSWINSZ", "TIOCMGET", "TIOCMBIS", "TIOCMBIC",
896 "TIOCMSET", "TIOCGSOFTCAR", "TIOCSSOFTCAR", "FIONREAD", "TIOCLINUX",
897 "TIOCCONS", "TIOCGSERIAL", "TIOCSSERIAL", "TIOCPKT", "FIONBIO",
898 "TIOCNOTTY", "TIOCSETD", "TIOCGETD", "TCSBRKP", [0x27] = "TIOCSBRK",
899 "TIOCCBRK", "TIOCGSID", "TCGETS2", "TCSETS2", "TCSETSW2", "TCSETSF2",
900 "TIOCGRS485", "TIOCSRS485", "TIOCGPTN", "TIOCSPTLCK",
901 "TIOCGDEV||TCGETX", "TCSETX", "TCSETXF", "TCSETXW", "TIOCSIG",
902 "TIOCVHANGUP", "TIOCGPKT", "TIOCGPTLCK", "TIOCGEXCL",
903 [0x50] = "FIONCLEX", "FIOCLEX", "FIOASYNC", "TIOCSERCONFIG",
904 "TIOCSERGWILD", "TIOCSERSWILD", "TIOCGLCKTRMIOS", "TIOCSLCKTRMIOS",
905 "TIOCSERGSTRUCT", "TIOCSERGETLSR", "TIOCSERGETMULTI", "TIOCSERSETMULTI",
906 "TIOCMIWAIT", "TIOCGICOUNT", [0x60] = "FIOQSIZE",
907};
908
909static DEFINE_STRARRAY_OFFSET(tioctls, 0x5401);
844ae5b4 910#endif /* defined(__i386__) || defined(__x86_64__) */
78645cf3 911
453350dd
ACM		 912#define STRARRAY(arg, name, array) \
913 .arg_scnprintf = { [arg] = SCA_STRARRAY, }, \
914 .arg_parm = { [arg] = &strarray__##array, }
915
514f1c67
ACM		 916static struct syscall_fmt {
917 const char *name;
aec1930b 918 const char *alias;
01533e97 919 size_t (*arg_scnprintf[6])(char *bf, size_t size, struct syscall_arg *arg);
1f115cb7 920 void *arg_parm[6];
514f1c67
ACM		 921 bool errmsg;
922 bool timeout;
04b34729 923 bool hexret;
514f1c67 924} syscall_fmts[] = {
51108999
ACM		 925 { .name = "access", .errmsg = true,
926 .arg_scnprintf = { [1] = SCA_ACCMODE, /* mode */ }, },
aec1930b 927 { .name = "arch_prctl", .errmsg = true, .alias = "prctl", },
beccb2b5
ACM		 928 { .name = "brk", .hexret = true,
929 .arg_scnprintf = { [0] = SCA_HEX, /* brk */ }, },
4f8c1b74 930 { .name = "clock_gettime", .errmsg = true, STRARRAY(0, clk_id, clockid), },
75b757ca 931 { .name = "close", .errmsg = true,
48000a1a 932 .arg_scnprintf = { [0] = SCA_CLOSE_FD, /* fd */ }, },
a14bb860 933 { .name = "connect", .errmsg = true, },
75b757ca 934 { .name = "dup", .errmsg = true,
48000a1a 935 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 936 { .name = "dup2", .errmsg = true,
48000a1a 937 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 938 { .name = "dup3", .errmsg = true,
48000a1a 939 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
453350dd 940 { .name = "epoll_ctl", .errmsg = true, STRARRAY(1, op, epoll_ctl_ops), },
49af9e93
ACM		 941 { .name = "eventfd2", .errmsg = true,
942 .arg_scnprintf = { [1] = SCA_EFD_FLAGS, /* flags */ }, },
75b757ca
ACM		 943 { .name = "faccessat", .errmsg = true,
944 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
945 { .name = "fadvise64", .errmsg = true,
48000a1a 946 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 947 { .name = "fallocate", .errmsg = true,
48000a1a 948 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 949 { .name = "fchdir", .errmsg = true,
48000a1a 950 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 951 { .name = "fchmod", .errmsg = true,
48000a1a 952 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 953 { .name = "fchmodat", .errmsg = true,
48000a1a 954 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
75b757ca 955 { .name = "fchown", .errmsg = true,
48000a1a 956 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 957 { .name = "fchownat", .errmsg = true,
48000a1a 958 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
75b757ca
ACM		 959 { .name = "fcntl", .errmsg = true,
960 .arg_scnprintf = { [0] = SCA_FD, /* fd */
961 [1] = SCA_STRARRAY, /* cmd */ },
962 .arg_parm = { [1] = &strarray__fcntl_cmds, /* cmd */ }, },
963 { .name = "fdatasync", .errmsg = true,
48000a1a 964 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
5cea6ff2 965 { .name = "flock", .errmsg = true,
75b757ca
ACM		 966 .arg_scnprintf = { [0] = SCA_FD, /* fd */
967 [1] = SCA_FLOCK, /* cmd */ }, },
968 { .name = "fsetxattr", .errmsg = true,
48000a1a 969 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 970 { .name = "fstat", .errmsg = true, .alias = "newfstat",
48000a1a 971 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 972 { .name = "fstatat", .errmsg = true, .alias = "newfstatat",
48000a1a 973 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
75b757ca 974 { .name = "fstatfs", .errmsg = true,
48000a1a 975 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 976 { .name = "fsync", .errmsg = true,
48000a1a 977 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 978 { .name = "ftruncate", .errmsg = true,
48000a1a 979 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
f9da0b0c
ACM		 980 { .name = "futex", .errmsg = true,
981 .arg_scnprintf = { [1] = SCA_FUTEX_OP, /* op */ }, },
75b757ca 982 { .name = "futimesat", .errmsg = true,
48000a1a 983 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
75b757ca 984 { .name = "getdents", .errmsg = true,
48000a1a 985 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 986 { .name = "getdents64", .errmsg = true,
48000a1a 987 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
453350dd
ACM		 988 { .name = "getitimer", .errmsg = true, STRARRAY(0, which, itimers), },
989 { .name = "getrlimit", .errmsg = true, STRARRAY(0, resource, rlimit_resources), },
beccb2b5 990 { .name = "ioctl", .errmsg = true,
48000a1a 991 .arg_scnprintf = { [0] = SCA_FD, /* fd */
844ae5b4
ACM		 992#if defined(__i386__) || defined(__x86_64__)
993/*
994 * FIXME: Make this available to all arches.
995 */
78645cf3
ACM		 996 [1] = SCA_STRHEXARRAY, /* cmd */
997 [2] = SCA_HEX, /* arg */ },
998 .arg_parm = { [1] = &strarray__tioctls, /* cmd */ }, },
844ae5b4
ACM		 999#else
1000 [2] = SCA_HEX, /* arg */ }, },
1001#endif
8bad5b0a
ACM		 1002 { .name = "kill", .errmsg = true,
1003 .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
75b757ca 1004 { .name = "linkat", .errmsg = true,
48000a1a 1005 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
75b757ca
ACM		 1006 { .name = "lseek", .errmsg = true,
1007 .arg_scnprintf = { [0] = SCA_FD, /* fd */
1008 [2] = SCA_STRARRAY, /* whence */ },
1009 .arg_parm = { [2] = &strarray__whences, /* whence */ }, },
e5959683 1010 { .name = "lstat", .errmsg = true, .alias = "newlstat", },
9e9716d1
ACM		 1011 { .name = "madvise", .errmsg = true,
1012 .arg_scnprintf = { [0] = SCA_HEX, /* start */
1013 [2] = SCA_MADV_BHV, /* behavior */ }, },
75b757ca 1014 { .name = "mkdirat", .errmsg = true,
48000a1a 1015 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
75b757ca 1016 { .name = "mknodat", .errmsg = true,
48000a1a 1017 .arg_scnprintf = { [0] = SCA_FDAT, /* fd */ }, },
3d903aa7
ACM		 1018 { .name = "mlock", .errmsg = true,
1019 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
1020 { .name = "mlockall", .errmsg = true,
1021 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
beccb2b5 1022 { .name = "mmap", .hexret = true,
ae685380 1023 .arg_scnprintf = { [0] = SCA_HEX, /* addr */
941557e0 1024 [2] = SCA_MMAP_PROT, /* prot */
73faab3a
NK		 1025 [3] = SCA_MMAP_FLAGS, /* flags */
1026 [4] = SCA_FD, /* fd */ }, },
beccb2b5 1027 { .name = "mprotect", .errmsg = true,
ae685380
ACM		 1028 .arg_scnprintf = { [0] = SCA_HEX, /* start */
1029 [2] = SCA_MMAP_PROT, /* prot */ }, },
1030 { .name = "mremap", .hexret = true,
1031 .arg_scnprintf = { [0] = SCA_HEX, /* addr */
86998dda 1032 [3] = SCA_MREMAP_FLAGS, /* flags */
ae685380 1033 [4] = SCA_HEX, /* new_addr */ }, },
3d903aa7
ACM		 1034 { .name = "munlock", .errmsg = true,
1035 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
beccb2b5
ACM		 1036 { .name = "munmap", .errmsg = true,
1037 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
75b757ca 1038 { .name = "name_to_handle_at", .errmsg = true,
48000a1a 1039 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
75b757ca 1040 { .name = "newfstatat", .errmsg = true,
48000a1a 1041 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
be65a89a
ACM		 1042 { .name = "open", .errmsg = true,
1043 .arg_scnprintf = { [1] = SCA_OPEN_FLAGS, /* flags */ }, },
31cd3855 1044 { .name = "open_by_handle_at", .errmsg = true,
75b757ca
ACM		 1045 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1046 [2] = SCA_OPEN_FLAGS, /* flags */ }, },
31cd3855 1047 { .name = "openat", .errmsg = true,
75b757ca
ACM		 1048 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */
1049 [2] = SCA_OPEN_FLAGS, /* flags */ }, },
46cce19b
ACM		 1050 { .name = "pipe2", .errmsg = true,
1051 .arg_scnprintf = { [1] = SCA_PIPE_FLAGS, /* flags */ }, },
aec1930b
ACM		 1052 { .name = "poll", .errmsg = true, .timeout = true, },
1053 { .name = "ppoll", .errmsg = true, .timeout = true, },
75b757ca 1054 { .name = "pread", .errmsg = true, .alias = "pread64",
48000a1a 1055 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1056 { .name = "preadv", .errmsg = true, .alias = "pread",
48000a1a 1057 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
453350dd 1058 { .name = "prlimit64", .errmsg = true, STRARRAY(1, resource, rlimit_resources), },
75b757ca 1059 { .name = "pwrite", .errmsg = true, .alias = "pwrite64",
48000a1a 1060 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1061 { .name = "pwritev", .errmsg = true,
48000a1a 1062 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1063 { .name = "read", .errmsg = true,
48000a1a 1064 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1065 { .name = "readlinkat", .errmsg = true,
48000a1a 1066 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
75b757ca 1067 { .name = "readv", .errmsg = true,
48000a1a 1068 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
b2cc99fd
ACM		 1069 { .name = "recvfrom", .errmsg = true,
1070 .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1071 { .name = "recvmmsg", .errmsg = true,
1072 .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1073 { .name = "recvmsg", .errmsg = true,
1074 .arg_scnprintf = { [2] = SCA_MSG_FLAGS, /* flags */ }, },
75b757ca 1075 { .name = "renameat", .errmsg = true,
48000a1a 1076 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
8bad5b0a
ACM		 1077 { .name = "rt_sigaction", .errmsg = true,
1078 .arg_scnprintf = { [0] = SCA_SIGNUM, /* sig */ }, },
453350dd 1079 { .name = "rt_sigprocmask", .errmsg = true, STRARRAY(0, how, sighow), },
8bad5b0a
ACM		 1080 { .name = "rt_sigqueueinfo", .errmsg = true,
1081 .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
1082 { .name = "rt_tgsigqueueinfo", .errmsg = true,
1083 .arg_scnprintf = { [2] = SCA_SIGNUM, /* sig */ }, },
aec1930b 1084 { .name = "select", .errmsg = true, .timeout = true, },
b2cc99fd
ACM		 1085 { .name = "sendmmsg", .errmsg = true,
1086 .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
1087 { .name = "sendmsg", .errmsg = true,
1088 .arg_scnprintf = { [2] = SCA_MSG_FLAGS, /* flags */ }, },
1089 { .name = "sendto", .errmsg = true,
1090 .arg_scnprintf = { [3] = SCA_MSG_FLAGS, /* flags */ }, },
453350dd
ACM		 1091 { .name = "setitimer", .errmsg = true, STRARRAY(0, which, itimers), },
1092 { .name = "setrlimit", .errmsg = true, STRARRAY(0, resource, rlimit_resources), },
75b757ca 1093 { .name = "shutdown", .errmsg = true,
48000a1a 1094 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
e10bce81 1095 { .name = "socket", .errmsg = true,
a28b24b2
ACM		 1096 .arg_scnprintf = { [0] = SCA_STRARRAY, /* family */
1097 [1] = SCA_SK_TYPE, /* type */ },
07120aa5
ACM		 1098 .arg_parm = { [0] = &strarray__socket_families, /* family */ }, },
1099 { .name = "socketpair", .errmsg = true,
1100 .arg_scnprintf = { [0] = SCA_STRARRAY, /* family */
1101 [1] = SCA_SK_TYPE, /* type */ },
e10bce81 1102 .arg_parm = { [0] = &strarray__socket_families, /* family */ }, },
aec1930b 1103 { .name = "stat", .errmsg = true, .alias = "newstat", },
75b757ca 1104 { .name = "symlinkat", .errmsg = true,
48000a1a 1105 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
8bad5b0a
ACM		 1106 { .name = "tgkill", .errmsg = true,
1107 .arg_scnprintf = { [2] = SCA_SIGNUM, /* sig */ }, },
1108 { .name = "tkill", .errmsg = true,
1109 .arg_scnprintf = { [1] = SCA_SIGNUM, /* sig */ }, },
e5959683 1110 { .name = "uname", .errmsg = true, .alias = "newuname", },
75b757ca
ACM		 1111 { .name = "unlinkat", .errmsg = true,
1112 .arg_scnprintf = { [0] = SCA_FDAT, /* dfd */ }, },
1113 { .name = "utimensat", .errmsg = true,
1114 .arg_scnprintf = { [0] = SCA_FDAT, /* dirfd */ }, },
1115 { .name = "write", .errmsg = true,
48000a1a 1116 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
75b757ca 1117 { .name = "writev", .errmsg = true,
48000a1a 1118 .arg_scnprintf = { [0] = SCA_FD, /* fd */ }, },
514f1c67
ACM		 1119};
1120
1121static int syscall_fmt__cmp(const void *name, const void *fmtp)
1122{
1123 const struct syscall_fmt *fmt = fmtp;
1124 return strcmp(name, fmt->name);
1125}
1126
1127static struct syscall_fmt *syscall_fmt__find(const char *name)
1128{
1129 const int nmemb = ARRAY_SIZE(syscall_fmts);
1130 return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
1131}
1132
1133struct syscall {
1134 struct event_format *tp_format;
1135 const char *name;
2ae3a312 1136 bool filtered;
5089f20e 1137 bool is_exit;
514f1c67 1138 struct syscall_fmt *fmt;
01533e97 1139 size_t (**arg_scnprintf)(char *bf, size_t size, struct syscall_arg *arg);
1f115cb7 1140 void **arg_parm;
514f1c67
ACM		 1141};
1142
60c907ab
ACM		 1143static size_t fprintf_duration(unsigned long t, FILE *fp)
1144{
1145 double duration = (double)t / NSEC_PER_MSEC;
1146 size_t printed = fprintf(fp, "(");
1147
1148 if (duration >= 1.0)
1149 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
1150 else if (duration >= 0.01)
1151 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
1152 else
1153 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
c24ff998 1154 return printed + fprintf(fp, "): ");
60c907ab
ACM		 1155}
1156
752fde44
ACM		 1157struct thread_trace {
1158 u64 entry_time;
1159 u64 exit_time;
1160 bool entry_pending;
efd5745e 1161 unsigned long nr_events;
a2ea67d7 1162 unsigned long pfmaj, pfmin;
752fde44 1163 char *entry_str;
1302d88e 1164 double runtime_ms;
75b757ca
ACM		 1165 struct {
1166 int max;
1167 char **table;
1168 } paths;
bf2575c1
DA		 1169
1170 struct intlist *syscall_stats;
752fde44
ACM		 1171};
1172
1173static struct thread_trace *thread_trace__new(void)
1174{
75b757ca
ACM		 1175 struct thread_trace *ttrace = zalloc(sizeof(struct thread_trace));
1176
1177 if (ttrace)
1178 ttrace->paths.max = -1;
1179
bf2575c1
DA		 1180 ttrace->syscall_stats = intlist__new(NULL);
1181
75b757ca 1182 return ttrace;
752fde44
ACM		 1183}
1184
c24ff998 1185static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
752fde44 1186{
efd5745e
ACM		 1187 struct thread_trace *ttrace;
1188
752fde44
ACM		 1189 if (thread == NULL)
1190 goto fail;
1191
89dceb22
NK		 1192 if (thread__priv(thread) == NULL)
1193 thread__set_priv(thread, thread_trace__new());
48000a1a 1194
89dceb22 1195 if (thread__priv(thread) == NULL)
752fde44
ACM		 1196 goto fail;
1197
89dceb22 1198 ttrace = thread__priv(thread);
efd5745e
ACM		 1199 ++ttrace->nr_events;
1200
1201 return ttrace;
752fde44 1202fail:
c24ff998 1203 color_fprintf(fp, PERF_COLOR_RED,
752fde44
ACM		 1204 "WARNING: not enough memory, dropping samples!\n");
1205 return NULL;
1206}
1207
598d02c5
SF		 1208#define TRACE_PFMAJ (1 << 0)
1209#define TRACE_PFMIN (1 << 1)
1210
514f1c67 1211struct trace {
c24ff998 1212 struct perf_tool tool;
c522739d
ACM		 1213 struct {
1214 int machine;
1215 int open_id;
1216 } audit;
514f1c67
ACM		 1217 struct {
1218 int max;
1219 struct syscall *table;
1220 } syscalls;
b4006796 1221 struct record_opts opts;
8fb598e5 1222 struct machine *host;
e596663e 1223 struct thread *current;
752fde44 1224 u64 base_time;
c24ff998 1225 FILE *output;
efd5745e 1226 unsigned long nr_events;
b059efdf 1227 struct strlist *ev_qualifier;
c522739d 1228 const char *last_vfs_getname;
bdc89661
DA		 1229 struct intlist *tid_list;
1230 struct intlist *pid_list;
98eafce6
ACM		 1231 double duration_filter;
1232 double runtime_ms;
1233 struct {
1234 u64 vfs_getname,
1235 proc_getname;
1236 } stats;
1237 bool not_ev_qualifier;
1238 bool live;
1239 bool full_time;
1302d88e 1240 bool sched;
752fde44 1241 bool multiple_threads;
bf2575c1 1242 bool summary;
fd2eabaf 1243 bool summary_only;
50c95cbd 1244 bool show_comm;
c522739d 1245 bool show_tool_stats;
e281a960 1246 bool trace_syscalls;
598d02c5 1247 int trace_pgfaults;
514f1c67
ACM		 1248};
1249
97119f37 1250static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
75b757ca 1251{
89dceb22 1252 struct thread_trace *ttrace = thread__priv(thread);
75b757ca
ACM		 1253
1254 if (fd > ttrace->paths.max) {
1255 char **npath = realloc(ttrace->paths.table, (fd + 1) * sizeof(char *));
1256
1257 if (npath == NULL)
1258 return -1;
1259
1260 if (ttrace->paths.max != -1) {
1261 memset(npath + ttrace->paths.max + 1, 0,
1262 (fd - ttrace->paths.max) * sizeof(char *));
1263 } else {
1264 memset(npath, 0, (fd + 1) * sizeof(char *));
1265 }
1266
1267 ttrace->paths.table = npath;
1268 ttrace->paths.max = fd;
1269 }
1270
1271 ttrace->paths.table[fd] = strdup(pathname);
1272
1273 return ttrace->paths.table[fd] != NULL ? 0 : -1;
1274}
1275
97119f37
ACM		 1276static int thread__read_fd_path(struct thread *thread, int fd)
1277{
1278 char linkname[PATH_MAX], pathname[PATH_MAX];
1279 struct stat st;
1280 int ret;
1281
1282 if (thread->pid_ == thread->tid) {
1283 scnprintf(linkname, sizeof(linkname),
1284 "/proc/%d/fd/%d", thread->pid_, fd);
1285 } else {
1286 scnprintf(linkname, sizeof(linkname),
1287 "/proc/%d/task/%d/fd/%d", thread->pid_, thread->tid, fd);
1288 }
1289
1290 if (lstat(linkname, &st) < 0 || st.st_size + 1 > (off_t)sizeof(pathname))
1291 return -1;
1292
1293 ret = readlink(linkname, pathname, sizeof(pathname));
1294
1295 if (ret < 0 || ret > st.st_size)
1296 return -1;
1297
1298 pathname[ret] = '\0';
1299 return trace__set_fd_pathname(thread, fd, pathname);
1300}
1301
c522739d
ACM		 1302static const char *thread__fd_path(struct thread *thread, int fd,
1303 struct trace *trace)
75b757ca 1304{
89dceb22 1305 struct thread_trace *ttrace = thread__priv(thread);
75b757ca
ACM		 1306
1307 if (ttrace == NULL)
1308 return NULL;
1309
1310 if (fd < 0)
1311 return NULL;
1312
cdcd1e6b 1313 if ((fd > ttrace->paths.max || ttrace->paths.table[fd] == NULL)) {
c522739d
ACM		 1314 if (!trace->live)
1315 return NULL;
1316 ++trace->stats.proc_getname;
cdcd1e6b 1317 if (thread__read_fd_path(thread, fd))
c522739d
ACM		 1318 return NULL;
1319 }
75b757ca
ACM		 1320
1321 return ttrace->paths.table[fd];
1322}
1323
1324static size_t syscall_arg__scnprintf_fd(char *bf, size_t size,
1325 struct syscall_arg *arg)
1326{
1327 int fd = arg->val;
1328 size_t printed = scnprintf(bf, size, "%d", fd);
c522739d 1329 const char *path = thread__fd_path(arg->thread, fd, arg->trace);
75b757ca
ACM		 1330
1331 if (path)
1332 printed += scnprintf(bf + printed, size - printed, "<%s>", path);
1333
1334 return printed;
1335}
1336
1337static size_t syscall_arg__scnprintf_close_fd(char *bf, size_t size,
1338 struct syscall_arg *arg)
1339{
1340 int fd = arg->val;
1341 size_t printed = syscall_arg__scnprintf_fd(bf, size, arg);
89dceb22 1342 struct thread_trace *ttrace = thread__priv(arg->thread);
75b757ca 1343
04662523
ACM		 1344 if (ttrace && fd >= 0 && fd <= ttrace->paths.max)
1345 zfree(&ttrace->paths.table[fd]);
75b757ca
ACM		 1346
1347 return printed;
1348}
1349
ae9ed035
ACM		 1350static bool trace__filter_duration(struct trace *trace, double t)
1351{
1352 return t < (trace->duration_filter * NSEC_PER_MSEC);
1353}
1354
752fde44
ACM		 1355static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
1356{
1357 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
1358
60c907ab 1359 return fprintf(fp, "%10.3f ", ts);
752fde44
ACM		 1360}
1361
f15eb531 1362static bool done = false;
ba209f85 1363static bool interrupted = false;
f15eb531 1364
ba209f85 1365static void sig_handler(int sig)
f15eb531
NK		 1366{
1367 done = true;
ba209f85 1368 interrupted = sig == SIGINT;
f15eb531
NK		 1369}
1370
752fde44 1371static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
60c907ab 1372 u64 duration, u64 tstamp, FILE *fp)
752fde44
ACM		 1373{
1374 size_t printed = trace__fprintf_tstamp(trace, tstamp, fp);
60c907ab 1375 printed += fprintf_duration(duration, fp);
752fde44 1376
50c95cbd
ACM		 1377 if (trace->multiple_threads) {
1378 if (trace->show_comm)
1902efe7 1379 printed += fprintf(fp, "%.14s/", thread__comm_str(thread));
38051234 1380 printed += fprintf(fp, "%d ", thread->tid);
50c95cbd 1381 }
752fde44
ACM		 1382
1383 return printed;
1384}
1385
c24ff998 1386static int trace__process_event(struct trace *trace, struct machine *machine,
162f0bef 1387 union perf_event *event, struct perf_sample *sample)
752fde44
ACM		 1388{
1389 int ret = 0;
1390
1391 switch (event->header.type) {
1392 case PERF_RECORD_LOST:
c24ff998 1393 color_fprintf(trace->output, PERF_COLOR_RED,
752fde44 1394 "LOST %" PRIu64 " events!\n", event->lost.lost);
162f0bef 1395 ret = machine__process_lost_event(machine, event, sample);
752fde44 1396 default:
162f0bef 1397 ret = machine__process_event(machine, event, sample);
752fde44
ACM		 1398 break;
1399 }
1400
1401 return ret;
1402}
1403
c24ff998 1404static int trace__tool_process(struct perf_tool *tool,
752fde44 1405 union perf_event *event,
162f0bef 1406 struct perf_sample *sample,
752fde44
ACM		 1407 struct machine *machine)
1408{
c24ff998 1409 struct trace *trace = container_of(tool, struct trace, tool);
162f0bef 1410 return trace__process_event(trace, machine, event, sample);
752fde44
ACM		 1411}
1412
1413static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
1414{
0a7e6d1b 1415 int err = symbol__init(NULL);
752fde44
ACM		 1416
1417 if (err)
1418 return err;
1419
8fb598e5
DA		 1420 trace->host = machine__new_host();
1421 if (trace->host == NULL)
1422 return -ENOMEM;
752fde44 1423
a33fbd56
ACM		 1424 err = __machine__synthesize_threads(trace->host, &trace->tool, &trace->opts.target,
1425 evlist->threads, trace__tool_process, false);
752fde44
ACM		 1426 if (err)
1427 symbol__exit();
1428
1429 return err;
1430}
1431
13d4ff3e
ACM		 1432static int syscall__set_arg_fmts(struct syscall *sc)
1433{
1434 struct format_field *field;
1435 int idx = 0;
1436
1437 sc->arg_scnprintf = calloc(sc->tp_format->format.nr_fields - 1, sizeof(void *));
1438 if (sc->arg_scnprintf == NULL)
1439 return -1;
1440
1f115cb7
ACM		 1441 if (sc->fmt)
1442 sc->arg_parm = sc->fmt->arg_parm;
1443
13d4ff3e 1444 for (field = sc->tp_format->format.fields->next; field; field = field->next) {
beccb2b5
ACM		 1445 if (sc->fmt && sc->fmt->arg_scnprintf[idx])
1446 sc->arg_scnprintf[idx] = sc->fmt->arg_scnprintf[idx];
1447 else if (field->flags & FIELD_IS_POINTER)
13d4ff3e
ACM		 1448 sc->arg_scnprintf[idx] = syscall_arg__scnprintf_hex;
1449 ++idx;
1450 }
1451
1452 return 0;
1453}
1454
514f1c67
ACM		 1455static int trace__read_syscall_info(struct trace *trace, int id)
1456{
1457 char tp_name[128];
1458 struct syscall *sc;
c522739d 1459 const char *name = audit_syscall_to_name(id, trace->audit.machine);
3a531260
ACM		 1460
1461 if (name == NULL)
1462 return -1;
514f1c67
ACM		 1463
1464 if (id > trace->syscalls.max) {
1465 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
1466
1467 if (nsyscalls == NULL)
1468 return -1;
1469
1470 if (trace->syscalls.max != -1) {
1471 memset(nsyscalls + trace->syscalls.max + 1, 0,
1472 (id - trace->syscalls.max) * sizeof(*sc));
1473 } else {
1474 memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
1475 }
1476
1477 trace->syscalls.table = nsyscalls;
1478 trace->syscalls.max = id;
1479 }
1480
1481 sc = trace->syscalls.table + id;
3a531260 1482 sc->name = name;
2ae3a312 1483
b059efdf
ACM		 1484 if (trace->ev_qualifier) {
1485 bool in = strlist__find(trace->ev_qualifier, name) != NULL;
1486
1487 if (!(in ^ trace->not_ev_qualifier)) {
1488 sc->filtered = true;
1489 /*
1490 * No need to do read tracepoint information since this will be
1491 * filtered out.
1492 */
1493 return 0;
1494 }
2ae3a312
ACM		 1495 }
1496
3a531260 1497 sc->fmt = syscall_fmt__find(sc->name);
514f1c67 1498
aec1930b 1499 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
97978b3e 1500 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b
ACM		 1501
1502 if (sc->tp_format == NULL && sc->fmt && sc->fmt->alias) {
1503 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
97978b3e 1504 sc->tp_format = trace_event__tp_format("syscalls", tp_name);
aec1930b 1505 }
514f1c67 1506
13d4ff3e
ACM		 1507 if (sc->tp_format == NULL)
1508 return -1;
1509
5089f20e
ACM		 1510 sc->is_exit = !strcmp(name, "exit_group") || !strcmp(name, "exit");
1511
13d4ff3e 1512 return syscall__set_arg_fmts(sc);
514f1c67
ACM		 1513}
1514
752fde44 1515static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
75b757ca
ACM		 1516 unsigned long *args, struct trace *trace,
1517 struct thread *thread)
514f1c67 1518{
514f1c67
ACM		 1519 size_t printed = 0;
1520
1521 if (sc->tp_format != NULL) {
1522 struct format_field *field;
01533e97
ACM		 1523 u8 bit = 1;
1524 struct syscall_arg arg = {
75b757ca
ACM		 1525 .idx = 0,
1526 .mask = 0,
1527 .trace = trace,
1528 .thread = thread,
01533e97 1529 };
6e7eeb51
ACM		 1530
1531 for (field = sc->tp_format->format.fields->next; field;
01533e97
ACM		 1532 field = field->next, ++arg.idx, bit <<= 1) {
1533 if (arg.mask & bit)
6e7eeb51 1534 continue;
4aa58232
ACM		 1535 /*
1536 * Suppress this argument if its value is zero and
1537 * and we don't have a string associated in an
1538 * strarray for it.
1539 */
1540 if (args[arg.idx] == 0 &&
1541 !(sc->arg_scnprintf &&
1542 sc->arg_scnprintf[arg.idx] == SCA_STRARRAY &&
1543 sc->arg_parm[arg.idx]))
22ae5cf1
ACM		 1544 continue;
1545
752fde44 1546 printed += scnprintf(bf + printed, size - printed,
13d4ff3e 1547 "%s%s: ", printed ? ", " : "", field->name);
01533e97
ACM		 1548 if (sc->arg_scnprintf && sc->arg_scnprintf[arg.idx]) {
1549 arg.val = args[arg.idx];
1f115cb7
ACM		 1550 if (sc->arg_parm)
1551 arg.parm = sc->arg_parm[arg.idx];
01533e97
ACM		 1552 printed += sc->arg_scnprintf[arg.idx](bf + printed,
1553 size - printed, &arg);
6e7eeb51 1554 } else {
13d4ff3e 1555 printed += scnprintf(bf + printed, size - printed,
01533e97 1556 "%ld", args[arg.idx]);
6e7eeb51 1557 }
514f1c67
ACM		 1558 }
1559 } else {
01533e97
ACM		 1560 int i = 0;
1561
514f1c67 1562 while (i < 6) {
752fde44
ACM		 1563 printed += scnprintf(bf + printed, size - printed,
1564 "%sarg%d: %ld",
1565 printed ? ", " : "", i, args[i]);
514f1c67
ACM		 1566 ++i;
1567 }
1568 }
1569
1570 return printed;
1571}
1572
ba3d7dee 1573typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1574 union perf_event *event,
ba3d7dee
ACM		 1575 struct perf_sample *sample);
1576
1577static struct syscall *trace__syscall_info(struct trace *trace,
bf2575c1 1578 struct perf_evsel *evsel, int id)
ba3d7dee 1579{
ba3d7dee
ACM		 1580
1581 if (id < 0) {
adaa18bf
ACM		 1582
1583 /*
1584 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
1585 * before that, leaving at a higher verbosity level till that is
1586 * explained. Reproduced with plain ftrace with:
1587 *
1588 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
1589 * grep "NR -1 " /t/trace_pipe
1590 *
1591 * After generating some load on the machine.
1592 */
1593 if (verbose > 1) {
1594 static u64 n;
1595 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
1596 id, perf_evsel__name(evsel), ++n);
1597 }
ba3d7dee
ACM		 1598 return NULL;
1599 }
1600
1601 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
1602 trace__read_syscall_info(trace, id))
1603 goto out_cant_read;
1604
1605 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
1606 goto out_cant_read;
1607
1608 return &trace->syscalls.table[id];
1609
1610out_cant_read:
7c304ee0
ACM		 1611 if (verbose) {
1612 fprintf(trace->output, "Problems reading syscall %d", id);
1613 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
1614 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
1615 fputs(" information\n", trace->output);
1616 }
ba3d7dee
ACM		 1617 return NULL;
1618}
1619
bf2575c1
DA		 1620static void thread__update_stats(struct thread_trace *ttrace,
1621 int id, struct perf_sample *sample)
1622{
1623 struct int_node *inode;
1624 struct stats *stats;
1625 u64 duration = 0;
1626
1627 inode = intlist__findnew(ttrace->syscall_stats, id);
1628 if (inode == NULL)
1629 return;
1630
1631 stats = inode->priv;
1632 if (stats == NULL) {
1633 stats = malloc(sizeof(struct stats));
1634 if (stats == NULL)
1635 return;
1636 init_stats(stats);
1637 inode->priv = stats;
1638 }
1639
1640 if (ttrace->entry_time && sample->time > ttrace->entry_time)
1641 duration = sample->time - ttrace->entry_time;
1642
1643 update_stats(stats, duration);
1644}
1645
e596663e
ACM		 1646static int trace__printf_interrupted_entry(struct trace *trace, struct perf_sample *sample)
1647{
1648 struct thread_trace *ttrace;
1649 u64 duration;
1650 size_t printed;
1651
1652 if (trace->current == NULL)
1653 return 0;
1654
1655 ttrace = thread__priv(trace->current);
1656
1657 if (!ttrace->entry_pending)
1658 return 0;
1659
1660 duration = sample->time - ttrace->entry_time;
1661
1662 printed = trace__fprintf_entry_head(trace, trace->current, duration, sample->time, trace->output);
1663 printed += fprintf(trace->output, "%-70s) ...\n", ttrace->entry_str);
1664 ttrace->entry_pending = false;
1665
1666 return printed;
1667}
1668
ba3d7dee 1669static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1670 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 1671 struct perf_sample *sample)
1672{
752fde44 1673 char *msg;
ba3d7dee 1674 void *args;
752fde44 1675 size_t printed = 0;
2ae3a312 1676 struct thread *thread;
77170988 1677 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
bf2575c1 1678 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 1679 struct thread_trace *ttrace;
1680
1681 if (sc == NULL)
1682 return -1;
ba3d7dee 1683
2ae3a312
ACM		 1684 if (sc->filtered)
1685 return 0;
1686
8fb598e5 1687 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 1688 ttrace = thread__trace(thread, trace->output);
2ae3a312 1689 if (ttrace == NULL)
ba3d7dee
ACM		 1690 return -1;
1691
77170988 1692 args = perf_evsel__sc_tp_ptr(evsel, args, sample);
752fde44
ACM		 1693
1694 if (ttrace->entry_str == NULL) {
1695 ttrace->entry_str = malloc(1024);
1696 if (!ttrace->entry_str)
1697 return -1;
1698 }
1699
e596663e
ACM		 1700 printed += trace__printf_interrupted_entry(trace, sample);
1701
752fde44
ACM		 1702 ttrace->entry_time = sample->time;
1703 msg = ttrace->entry_str;
1704 printed += scnprintf(msg + printed, 1024 - printed, "%s(", sc->name);
1705
75b757ca
ACM		 1706 printed += syscall__scnprintf_args(sc, msg + printed, 1024 - printed,
1707 args, trace, thread);
752fde44 1708
5089f20e 1709 if (sc->is_exit) {
fd2eabaf 1710 if (!trace->duration_filter && !trace->summary_only) {
c24ff998
ACM		 1711 trace__fprintf_entry_head(trace, thread, 1, sample->time, trace->output);
1712 fprintf(trace->output, "%-70s\n", ttrace->entry_str);
ae9ed035 1713 }
752fde44
ACM		 1714 } else
1715 ttrace->entry_pending = true;
ba3d7dee 1716
e596663e
ACM		 1717 trace->current = thread;
1718
ba3d7dee
ACM		 1719 return 0;
1720}
1721
1722static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1723 union perf_event *event __maybe_unused,
ba3d7dee
ACM		 1724 struct perf_sample *sample)
1725{
2c82c3ad 1726 long ret;
60c907ab 1727 u64 duration = 0;
2ae3a312 1728 struct thread *thread;
77170988 1729 int id = perf_evsel__sc_tp_uint(evsel, id, sample);
bf2575c1 1730 struct syscall *sc = trace__syscall_info(trace, evsel, id);
2ae3a312
ACM		 1731 struct thread_trace *ttrace;
1732
1733 if (sc == NULL)
1734 return -1;
ba3d7dee 1735
2ae3a312
ACM		 1736 if (sc->filtered)
1737 return 0;
1738
8fb598e5 1739 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
c24ff998 1740 ttrace = thread__trace(thread, trace->output);
2ae3a312 1741 if (ttrace == NULL)
ba3d7dee
ACM		 1742 return -1;
1743
bf2575c1
DA		 1744 if (trace->summary)
1745 thread__update_stats(ttrace, id, sample);
1746
77170988 1747 ret = perf_evsel__sc_tp_uint(evsel, ret, sample);
ba3d7dee 1748
c522739d
ACM		 1749 if (id == trace->audit.open_id && ret >= 0 && trace->last_vfs_getname) {
1750 trace__set_fd_pathname(thread, ret, trace->last_vfs_getname);
1751 trace->last_vfs_getname = NULL;
1752 ++trace->stats.vfs_getname;
1753 }
1754
752fde44
ACM		 1755 ttrace->exit_time = sample->time;
1756
ae9ed035 1757 if (ttrace->entry_time) {
60c907ab 1758 duration = sample->time - ttrace->entry_time;
ae9ed035
ACM		 1759 if (trace__filter_duration(trace, duration))
1760 goto out;
1761 } else if (trace->duration_filter)
1762 goto out;
60c907ab 1763
fd2eabaf
DA		 1764 if (trace->summary_only)
1765 goto out;
1766
c24ff998 1767 trace__fprintf_entry_head(trace, thread, duration, sample->time, trace->output);
752fde44
ACM		 1768
1769 if (ttrace->entry_pending) {
c24ff998 1770 fprintf(trace->output, "%-70s", ttrace->entry_str);
752fde44 1771 } else {
c24ff998
ACM		 1772 fprintf(trace->output, " ... [");
1773 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
1774 fprintf(trace->output, "]: %s()", sc->name);
752fde44
ACM		 1775 }
1776
da3c9a44
ACM		 1777 if (sc->fmt == NULL) {
1778signed_print:
2c82c3ad 1779 fprintf(trace->output, ") = %ld", ret);
da3c9a44 1780 } else if (ret < 0 && sc->fmt->errmsg) {
942a91ed 1781 char bf[STRERR_BUFSIZE];
ba3d7dee
ACM		 1782 const char *emsg = strerror_r(-ret, bf, sizeof(bf)),
1783 *e = audit_errno_to_name(-ret);
1784
c24ff998 1785 fprintf(trace->output, ") = -1 %s %s", e, emsg);
da3c9a44 1786 } else if (ret == 0 && sc->fmt->timeout)
c24ff998 1787 fprintf(trace->output, ") = 0 Timeout");
04b34729 1788 else if (sc->fmt->hexret)
2c82c3ad 1789 fprintf(trace->output, ") = %#lx", ret);
ba3d7dee 1790 else
da3c9a44 1791 goto signed_print;
ba3d7dee 1792
c24ff998 1793 fputc('\n', trace->output);
ae9ed035 1794out:
752fde44
ACM		 1795 ttrace->entry_pending = false;
1796
ba3d7dee
ACM		 1797 return 0;
1798}
1799
c522739d 1800static int trace__vfs_getname(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1801 union perf_event *event __maybe_unused,
c522739d
ACM		 1802 struct perf_sample *sample)
1803{
1804 trace->last_vfs_getname = perf_evsel__rawptr(evsel, sample, "pathname");
1805 return 0;
1806}
1807
1302d88e 1808static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
0c82adcf 1809 union perf_event *event __maybe_unused,
1302d88e
ACM		 1810 struct perf_sample *sample)
1811{
1812 u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
1813 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
8fb598e5 1814 struct thread *thread = machine__findnew_thread(trace->host,
314add6b
AH		 1815 sample->pid,
1816 sample->tid);
c24ff998 1817 struct thread_trace *ttrace = thread__trace(thread, trace->output);
1302d88e
ACM		 1818
1819 if (ttrace == NULL)
1820 goto out_dump;
1821
1822 ttrace->runtime_ms += runtime_ms;
1823 trace->runtime_ms += runtime_ms;
1824 return 0;
1825
1826out_dump:
c24ff998 1827 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
1302d88e
ACM		 1828 evsel->name,
1829 perf_evsel__strval(evsel, sample, "comm"),
1830 (pid_t)perf_evsel__intval(evsel, sample, "pid"),
1831 runtime,
1832 perf_evsel__intval(evsel, sample, "vruntime"));
1833 return 0;
1834}
1835
598d02c5
SF		 1836static void print_location(FILE *f, struct perf_sample *sample,
1837 struct addr_location *al,
1838 bool print_dso, bool print_sym)
1839{
1840
1841 if ((verbose || print_dso) && al->map)
1842 fprintf(f, "%s@", al->map->dso->long_name);
1843
1844 if ((verbose || print_sym) && al->sym)
4414a3c5 1845 fprintf(f, "%s+0x%" PRIx64, al->sym->name,
598d02c5
SF		 1846 al->addr - al->sym->start);
1847 else if (al->map)
4414a3c5 1848 fprintf(f, "0x%" PRIx64, al->addr);
598d02c5 1849 else
4414a3c5 1850 fprintf(f, "0x%" PRIx64, sample->addr);
598d02c5
SF		 1851}
1852
1853static int trace__pgfault(struct trace *trace,
1854 struct perf_evsel *evsel,
1855 union perf_event *event,
1856 struct perf_sample *sample)
1857{
1858 struct thread *thread;
1859 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1860 struct addr_location al;
1861 char map_type = 'd';
a2ea67d7 1862 struct thread_trace *ttrace;
598d02c5
SF		 1863
1864 thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
a2ea67d7
SF		 1865 ttrace = thread__trace(thread, trace->output);
1866 if (ttrace == NULL)
1867 return -1;
1868
1869 if (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ)
1870 ttrace->pfmaj++;
1871 else
1872 ttrace->pfmin++;
1873
1874 if (trace->summary_only)
1875 return 0;
598d02c5 1876
bb871a9c 1877 thread__find_addr_location(thread, cpumode, MAP__FUNCTION,
598d02c5
SF		 1878 sample->ip, &al);
1879
1880 trace__fprintf_entry_head(trace, thread, 0, sample->time, trace->output);
1881
1882 fprintf(trace->output, "%sfault [",
1883 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
1884 "maj" : "min");
1885
1886 print_location(trace->output, sample, &al, false, true);
1887
1888 fprintf(trace->output, "] => ");
1889
bb871a9c 1890 thread__find_addr_location(thread, cpumode, MAP__VARIABLE,
598d02c5
SF		 1891 sample->addr, &al);
1892
1893 if (!al.map) {
bb871a9c 1894 thread__find_addr_location(thread, cpumode,
598d02c5
SF		 1895 MAP__FUNCTION, sample->addr, &al);
1896
1897 if (al.map)
1898 map_type = 'x';
1899 else
1900 map_type = '?';
1901 }
1902
1903 print_location(trace->output, sample, &al, true, false);
1904
1905 fprintf(trace->output, " (%c%c)\n", map_type, al.level);
1906
1907 return 0;
1908}
1909
bdc89661
DA		 1910static bool skip_sample(struct trace *trace, struct perf_sample *sample)
1911{
1912 if ((trace->pid_list && intlist__find(trace->pid_list, sample->pid)) ||
1913 (trace->tid_list && intlist__find(trace->tid_list, sample->tid)))
1914 return false;
1915
1916 if (trace->pid_list || trace->tid_list)
1917 return true;
1918
1919 return false;
1920}
1921
6810fc91 1922static int trace__process_sample(struct perf_tool *tool,
0c82adcf 1923 union perf_event *event,
6810fc91
DA		 1924 struct perf_sample *sample,
1925 struct perf_evsel *evsel,
1926 struct machine *machine __maybe_unused)
1927{
1928 struct trace *trace = container_of(tool, struct trace, tool);
1929 int err = 0;
1930
744a9719 1931 tracepoint_handler handler = evsel->handler;
6810fc91 1932
bdc89661
DA		 1933 if (skip_sample(trace, sample))
1934 return 0;
1935
4bb09192 1936 if (!trace->full_time && trace->base_time == 0)
6810fc91
DA		 1937 trace->base_time = sample->time;
1938
3160565f
DA		 1939 if (handler) {
1940 ++trace->nr_events;
0c82adcf 1941 handler(trace, evsel, event, sample);
3160565f 1942 }
6810fc91
DA		 1943
1944 return err;
1945}
1946
bdc89661
DA		 1947static int parse_target_str(struct trace *trace)
1948{
1949 if (trace->opts.target.pid) {
1950 trace->pid_list = intlist__new(trace->opts.target.pid);
1951 if (trace->pid_list == NULL) {
1952 pr_err("Error parsing process id string\n");
1953 return -EINVAL;
1954 }
1955 }
1956
1957 if (trace->opts.target.tid) {
1958 trace->tid_list = intlist__new(trace->opts.target.tid);
1959 if (trace->tid_list == NULL) {
1960 pr_err("Error parsing thread id string\n");
1961 return -EINVAL;
1962 }
1963 }
1964
1965 return 0;
1966}
1967
1e28fe0a 1968static int trace__record(struct trace *trace, int argc, const char **argv)
5e2485b1
DA		 1969{
1970 unsigned int rec_argc, i, j;
1971 const char **rec_argv;
1972 const char * const record_args[] = {
1973 "record",
1974 "-R",
1975 "-m", "1024",
1976 "-c", "1",
5e2485b1
DA		 1977 };
1978
1e28fe0a
SF		 1979 const char * const sc_args[] = { "-e", };
1980 unsigned int sc_args_nr = ARRAY_SIZE(sc_args);
1981 const char * const majpf_args[] = { "-e", "major-faults" };
1982 unsigned int majpf_args_nr = ARRAY_SIZE(majpf_args);
1983 const char * const minpf_args[] = { "-e", "minor-faults" };
1984 unsigned int minpf_args_nr = ARRAY_SIZE(minpf_args);
1985
9aca7f17 1986 /* +1 is for the event string below */
1e28fe0a
SF		 1987 rec_argc = ARRAY_SIZE(record_args) + sc_args_nr + 1 +
1988 majpf_args_nr + minpf_args_nr + argc;
5e2485b1
DA		 1989 rec_argv = calloc(rec_argc + 1, sizeof(char *));
1990
1991 if (rec_argv == NULL)
1992 return -ENOMEM;
1993
1e28fe0a 1994 j = 0;
5e2485b1 1995 for (i = 0; i < ARRAY_SIZE(record_args); i++)
1e28fe0a
SF		 1996 rec_argv[j++] = record_args[i];
1997
e281a960
SF		 1998 if (trace->trace_syscalls) {
1999 for (i = 0; i < sc_args_nr; i++)
2000 rec_argv[j++] = sc_args[i];
2001
2002 /* event string may be different for older kernels - e.g., RHEL6 */
2003 if (is_valid_tracepoint("raw_syscalls:sys_enter"))
2004 rec_argv[j++] = "raw_syscalls:sys_enter,raw_syscalls:sys_exit";
2005 else if (is_valid_tracepoint("syscalls:sys_enter"))
2006 rec_argv[j++] = "syscalls:sys_enter,syscalls:sys_exit";
2007 else {
2008 pr_err("Neither raw_syscalls nor syscalls events exist.\n");
2009 return -1;
2010 }
9aca7f17 2011 }
9aca7f17 2012
1e28fe0a
SF		 2013 if (trace->trace_pgfaults & TRACE_PFMAJ)
2014 for (i = 0; i < majpf_args_nr; i++)
2015 rec_argv[j++] = majpf_args[i];
2016
2017 if (trace->trace_pgfaults & TRACE_PFMIN)
2018 for (i = 0; i < minpf_args_nr; i++)
2019 rec_argv[j++] = minpf_args[i];
2020
2021 for (i = 0; i < (unsigned int)argc; i++)
2022 rec_argv[j++] = argv[i];
5e2485b1 2023
1e28fe0a 2024 return cmd_record(j, rec_argv, NULL);
5e2485b1
DA		 2025}
2026
bf2575c1
DA		 2027static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp);
2028
c522739d
ACM		 2029static void perf_evlist__add_vfs_getname(struct perf_evlist *evlist)
2030{
ef503831 2031 struct perf_evsel *evsel = perf_evsel__newtp("probe", "vfs_getname");
c522739d
ACM		 2032 if (evsel == NULL)
2033 return;
2034
2035 if (perf_evsel__field(evsel, "pathname") == NULL) {
2036 perf_evsel__delete(evsel);
2037 return;
2038 }
2039
744a9719 2040 evsel->handler = trace__vfs_getname;
c522739d
ACM		 2041 perf_evlist__add(evlist, evsel);
2042}
2043
598d02c5
SF		 2044static int perf_evlist__add_pgfault(struct perf_evlist *evlist,
2045 u64 config)
2046{
2047 struct perf_evsel *evsel;
2048 struct perf_event_attr attr = {
2049 .type = PERF_TYPE_SOFTWARE,
2050 .mmap_data = 1,
598d02c5
SF		 2051 };
2052
2053 attr.config = config;
0524798c 2054 attr.sample_period = 1;
598d02c5
SF		 2055
2056 event_attr_init(&attr);
2057
2058 evsel = perf_evsel__new(&attr);
2059 if (!evsel)
2060 return -ENOMEM;
2061
2062 evsel->handler = trace__pgfault;
2063 perf_evlist__add(evlist, evsel);
2064
2065 return 0;
2066}
2067
f15eb531 2068static int trace__run(struct trace *trace, int argc, const char **argv)
514f1c67 2069{
334fe7a3 2070 struct perf_evlist *evlist = perf_evlist__new();
ba3d7dee 2071 struct perf_evsel *evsel;
efd5745e
ACM		 2072 int err = -1, i;
2073 unsigned long before;
f15eb531 2074 const bool forks = argc > 0;
46fb3c21 2075 bool draining = false;
514f1c67 2076
75b757ca
ACM		 2077 trace->live = true;
2078
514f1c67 2079 if (evlist == NULL) {
c24ff998 2080 fprintf(trace->output, "Not enough memory to run!\n");
514f1c67
ACM		 2081 goto out;
2082 }
2083
e281a960
SF		 2084 if (trace->trace_syscalls &&
2085 perf_evlist__add_syscall_newtp(evlist, trace__sys_enter,
2086 trace__sys_exit))
801c67b0 2087 goto out_error_raw_syscalls;
514f1c67 2088
e281a960
SF		 2089 if (trace->trace_syscalls)
2090 perf_evlist__add_vfs_getname(evlist);
c522739d 2091
598d02c5 2092 if ((trace->trace_pgfaults & TRACE_PFMAJ) &&
e2726d99 2093 perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MAJ)) {
5ed08dae 2094 goto out_error_mem;
e2726d99 2095 }
598d02c5
SF		 2096
2097 if ((trace->trace_pgfaults & TRACE_PFMIN) &&
2098 perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MIN))
5ed08dae 2099 goto out_error_mem;
598d02c5 2100
1302d88e 2101 if (trace->sched &&
2cc990ba
ACM		 2102 perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
2103 trace__sched_stat_runtime))
2104 goto out_error_sched_stat_runtime;
1302d88e 2105
514f1c67
ACM		 2106 err = perf_evlist__create_maps(evlist, &trace->opts.target);
2107 if (err < 0) {
c24ff998 2108 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
514f1c67
ACM		 2109 goto out_delete_evlist;
2110 }
2111
752fde44
ACM		 2112 err = trace__symbols_init(trace, evlist);
2113 if (err < 0) {
c24ff998 2114 fprintf(trace->output, "Problems initializing symbol libraries!\n");
03ad9747 2115 goto out_delete_evlist;
752fde44
ACM		 2116 }
2117
f77a9518 2118 perf_evlist__config(evlist, &trace->opts);
514f1c67 2119
f15eb531
NK		 2120 signal(SIGCHLD, sig_handler);
2121 signal(SIGINT, sig_handler);
2122
2123 if (forks) {
6ef73ec4 2124 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
735f7e0b 2125 argv, false, NULL);
f15eb531 2126 if (err < 0) {
c24ff998 2127 fprintf(trace->output, "Couldn't run the workload!\n");
03ad9747 2128 goto out_delete_evlist;
f15eb531
NK		 2129 }
2130 }
2131
514f1c67 2132 err = perf_evlist__open(evlist);
a8f23d8f
ACM		 2133 if (err < 0)
2134 goto out_error_open;
514f1c67 2135
f885037e 2136 err = perf_evlist__mmap(evlist, trace->opts.mmap_pages, false);
e09b18d4
ACM		 2137 if (err < 0)
2138 goto out_error_mmap;
514f1c67 2139
f15eb531
NK		 2140 if (forks)
2141 perf_evlist__start_workload(evlist);
f7aa222f
ACM		 2142 else
2143 perf_evlist__enable(evlist);
f15eb531 2144
42052bea
ACM		 2145 trace->multiple_threads = evlist->threads->map[0] == -1 ||
2146 evlist->threads->nr > 1 ||
2147 perf_evlist__first(evlist)->attr.inherit;
514f1c67 2148again:
efd5745e 2149 before = trace->nr_events;
514f1c67
ACM		 2150
2151 for (i = 0; i < evlist->nr_mmaps; i++) {
2152 union perf_event *event;
2153
2154 while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
2155 const u32 type = event->header.type;
ba3d7dee 2156 tracepoint_handler handler;
514f1c67 2157 struct perf_sample sample;
514f1c67 2158
efd5745e 2159 ++trace->nr_events;
514f1c67 2160
514f1c67
ACM		 2161 err = perf_evlist__parse_sample(evlist, event, &sample);
2162 if (err) {
c24ff998 2163 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
8e50d384 2164 goto next_event;
514f1c67
ACM		 2165 }
2166
4bb09192 2167 if (!trace->full_time && trace->base_time == 0)
752fde44
ACM		 2168 trace->base_time = sample.time;
2169
2170 if (type != PERF_RECORD_SAMPLE) {
162f0bef 2171 trace__process_event(trace, trace->host, event, &sample);
752fde44
ACM		 2172 continue;
2173 }
2174
514f1c67
ACM		 2175 evsel = perf_evlist__id2evsel(evlist, sample.id);
2176 if (evsel == NULL) {
c24ff998 2177 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample.id);
8e50d384 2178 goto next_event;
514f1c67
ACM		 2179 }
2180
598d02c5
SF		 2181 if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
2182 sample.raw_data == NULL) {
c24ff998 2183 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
fc551f8d
ACM		 2184 perf_evsel__name(evsel), sample.tid,
2185 sample.cpu, sample.raw_size);
8e50d384 2186 goto next_event;
fc551f8d
ACM		 2187 }
2188
744a9719 2189 handler = evsel->handler;
0c82adcf 2190 handler(trace, evsel, event, &sample);
8e50d384
ZZ		 2191next_event:
2192 perf_evlist__mmap_consume(evlist, i);
20c5f10e 2193
ba209f85
ACM		 2194 if (interrupted)
2195 goto out_disable;
514f1c67
ACM		 2196 }
2197 }
2198
efd5745e 2199 if (trace->nr_events == before) {
ba209f85 2200 int timeout = done ? 100 : -1;
f15eb531 2201
46fb3c21
ACM		 2202 if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
2203 if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP) == 0)
2204 draining = true;
2205
ba209f85 2206 goto again;
46fb3c21 2207 }
ba209f85
ACM		 2208 } else {
2209 goto again;
f15eb531
NK		 2210 }
2211
ba209f85
ACM		 2212out_disable:
2213 perf_evlist__disable(evlist);
514f1c67 2214
c522739d
ACM		 2215 if (!err) {
2216 if (trace->summary)
2217 trace__fprintf_thread_summary(trace, trace->output);
2218
2219 if (trace->show_tool_stats) {
2220 fprintf(trace->output, "Stats:\n "
2221 " vfs_getname : %" PRIu64 "\n"
2222 " proc_getname: %" PRIu64 "\n",
2223 trace->stats.vfs_getname,
2224 trace->stats.proc_getname);
2225 }
2226 }
bf2575c1 2227
514f1c67
ACM		 2228out_delete_evlist:
2229 perf_evlist__delete(evlist);
2230out:
75b757ca 2231 trace->live = false;
514f1c67 2232 return err;
6ef068cb
ACM		 2233{
2234 char errbuf[BUFSIZ];
a8f23d8f 2235
2cc990ba
ACM		 2236out_error_sched_stat_runtime:
2237 debugfs__strerror_open_tp(errno, errbuf, sizeof(errbuf), "sched", "sched_stat_runtime");
2238 goto out_error;
2239
801c67b0 2240out_error_raw_syscalls:
2cc990ba 2241 debugfs__strerror_open_tp(errno, errbuf, sizeof(errbuf), "raw_syscalls", "sys_(enter|exit)");
a8f23d8f
ACM		 2242 goto out_error;
2243
e09b18d4
ACM		 2244out_error_mmap:
2245 perf_evlist__strerror_mmap(evlist, errno, errbuf, sizeof(errbuf));
2246 goto out_error;
2247
a8f23d8f
ACM		 2248out_error_open:
2249 perf_evlist__strerror_open(evlist, errno, errbuf, sizeof(errbuf));
2250
2251out_error:
6ef068cb 2252 fprintf(trace->output, "%s\n", errbuf);
87f91868 2253 goto out_delete_evlist;
514f1c67 2254}
5ed08dae
ACM		 2255out_error_mem:
2256 fprintf(trace->output, "Not enough memory to run!\n");
2257 goto out_delete_evlist;
a8f23d8f 2258}
514f1c67 2259
6810fc91
DA		 2260static int trace__replay(struct trace *trace)
2261{
2262 const struct perf_evsel_str_handler handlers[] = {
c522739d 2263 { "probe:vfs_getname", trace__vfs_getname, },
6810fc91 2264 };
f5fc1412
JO		 2265 struct perf_data_file file = {
2266 .path = input_name,
2267 .mode = PERF_DATA_MODE_READ,
2268 };
6810fc91 2269 struct perf_session *session;
003824e8 2270 struct perf_evsel *evsel;
6810fc91
DA		 2271 int err = -1;
2272
2273 trace->tool.sample = trace__process_sample;
2274 trace->tool.mmap = perf_event__process_mmap;
384c671e 2275 trace->tool.mmap2 = perf_event__process_mmap2;
6810fc91
DA		 2276 trace->tool.comm = perf_event__process_comm;
2277 trace->tool.exit = perf_event__process_exit;
2278 trace->tool.fork = perf_event__process_fork;
2279 trace->tool.attr = perf_event__process_attr;
2280 trace->tool.tracing_data = perf_event__process_tracing_data;
2281 trace->tool.build_id = perf_event__process_build_id;
2282
0a8cb85c 2283 trace->tool.ordered_events = true;
6810fc91
DA		 2284 trace->tool.ordering_requires_timestamps = true;
2285
2286 /* add tid to output */
2287 trace->multiple_threads = true;
2288
f5fc1412 2289 session = perf_session__new(&file, false, &trace->tool);
6810fc91 2290 if (session == NULL)
52e02834 2291 return -1;
6810fc91 2292
0a7e6d1b 2293 if (symbol__init(&session->header.env) < 0)
cb2ffae2
NK		 2294 goto out;
2295
8fb598e5
DA		 2296 trace->host = &session->machines.host;
2297
6810fc91
DA		 2298 err = perf_session__set_tracepoints_handlers(session, handlers);
2299 if (err)
2300 goto out;
2301
003824e8
NK		 2302 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2303 "raw_syscalls:sys_enter");
9aca7f17
DA		 2304 /* older kernels have syscalls tp versus raw_syscalls */
2305 if (evsel == NULL)
2306 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2307 "syscalls:sys_enter");
003824e8 2308
e281a960
SF		 2309 if (evsel &&
2310 (perf_evsel__init_syscall_tp(evsel, trace__sys_enter) < 0 ||
2311 perf_evsel__init_sc_tp_ptr_field(evsel, args))) {
003824e8
NK		 2312 pr_err("Error during initialize raw_syscalls:sys_enter event\n");
2313 goto out;
2314 }
2315
2316 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2317 "raw_syscalls:sys_exit");
9aca7f17
DA		 2318 if (evsel == NULL)
2319 evsel = perf_evlist__find_tracepoint_by_name(session->evlist,
2320 "syscalls:sys_exit");
e281a960
SF		 2321 if (evsel &&
2322 (perf_evsel__init_syscall_tp(evsel, trace__sys_exit) < 0 ||
2323 perf_evsel__init_sc_tp_uint_field(evsel, ret))) {
003824e8 2324 pr_err("Error during initialize raw_syscalls:sys_exit event\n");
6810fc91
DA		 2325 goto out;
2326 }
2327
1e28fe0a
SF		 2328 evlist__for_each(session->evlist, evsel) {
2329 if (evsel->attr.type == PERF_TYPE_SOFTWARE &&
2330 (evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ||
2331 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
2332 evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS))
2333 evsel->handler = trace__pgfault;
2334 }
2335
bdc89661
DA		 2336 err = parse_target_str(trace);
2337 if (err != 0)
2338 goto out;
2339
6810fc91
DA		 2340 setup_pager();
2341
2342 err = perf_session__process_events(session, &trace->tool);
2343 if (err)
2344 pr_err("Failed to process events, error %d", err);
2345
bf2575c1
DA		 2346 else if (trace->summary)
2347 trace__fprintf_thread_summary(trace, trace->output);
2348
6810fc91
DA		 2349out:
2350 perf_session__delete(session);
2351
2352 return err;
2353}
2354
1302d88e
ACM		 2355static size_t trace__fprintf_threads_header(FILE *fp)
2356{
2357 size_t printed;
2358
99ff7150 2359 printed = fprintf(fp, "\n Summary of events:\n\n");
bf2575c1
DA		 2360
2361 return printed;
2362}
2363
2364static size_t thread__dump_stats(struct thread_trace *ttrace,
2365 struct trace *trace, FILE *fp)
2366{
2367 struct stats *stats;
2368 size_t printed = 0;
2369 struct syscall *sc;
2370 struct int_node *inode = intlist__first(ttrace->syscall_stats);
2371
2372 if (inode == NULL)
2373 return 0;
2374
2375 printed += fprintf(fp, "\n");
2376
27a778b5
PE		 2377 printed += fprintf(fp, " syscall calls min avg max stddev\n");
2378 printed += fprintf(fp, " (msec) (msec) (msec) (%%)\n");
2379 printed += fprintf(fp, " --------------- -------- --------- --------- --------- ------\n");
99ff7150 2380
bf2575c1
DA		 2381 /* each int_node is a syscall */
2382 while (inode) {
2383 stats = inode->priv;
2384 if (stats) {
2385 double min = (double)(stats->min) / NSEC_PER_MSEC;
2386 double max = (double)(stats->max) / NSEC_PER_MSEC;
2387 double avg = avg_stats(stats);
2388 double pct;
2389 u64 n = (u64) stats->n;
2390
2391 pct = avg ? 100.0 * stddev_stats(stats)/avg : 0.0;
2392 avg /= NSEC_PER_MSEC;
2393
2394 sc = &trace->syscalls.table[inode->i];
99ff7150 2395 printed += fprintf(fp, " %-15s", sc->name);
27a778b5 2396 printed += fprintf(fp, " %8" PRIu64 " %9.3f %9.3f",
7f7a4138 2397 n, min, avg);
27a778b5 2398 printed += fprintf(fp, " %9.3f %9.2f%%\n", max, pct);
bf2575c1
DA		 2399 }
2400
2401 inode = intlist__next(inode);
2402 }
2403
2404 printed += fprintf(fp, "\n\n");
1302d88e
ACM		 2405
2406 return printed;
2407}
2408
896cbb56
DA		 2409/* struct used to pass data to per-thread function */
2410struct summary_data {
2411 FILE *fp;
2412 struct trace *trace;
2413 size_t printed;
2414};
2415
2416static int trace__fprintf_one_thread(struct thread *thread, void *priv)
2417{
2418 struct summary_data *data = priv;
2419 FILE *fp = data->fp;
2420 size_t printed = data->printed;
2421 struct trace *trace = data->trace;
89dceb22 2422 struct thread_trace *ttrace = thread__priv(thread);
896cbb56
DA		 2423 double ratio;
2424
2425 if (ttrace == NULL)
2426 return 0;
2427
2428 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
2429
15e65c69 2430 printed += fprintf(fp, " %s (%d), ", thread__comm_str(thread), thread->tid);
99ff7150 2431 printed += fprintf(fp, "%lu events, ", ttrace->nr_events);
15e65c69 2432 printed += fprintf(fp, "%.1f%%", ratio);
a2ea67d7
SF		 2433 if (ttrace->pfmaj)
2434 printed += fprintf(fp, ", %lu majfaults", ttrace->pfmaj);
2435 if (ttrace->pfmin)
2436 printed += fprintf(fp, ", %lu minfaults", ttrace->pfmin);
99ff7150 2437 printed += fprintf(fp, ", %.3f msec\n", ttrace->runtime_ms);
bf2575c1 2438 printed += thread__dump_stats(ttrace, trace, fp);
896cbb56
DA		 2439
2440 data->printed += printed;
2441
2442 return 0;
2443}
2444
1302d88e
ACM		 2445static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
2446{
896cbb56
DA		 2447 struct summary_data data = {
2448 .fp = fp,
2449 .trace = trace
2450 };
2451 data.printed = trace__fprintf_threads_header(fp);
1302d88e 2452
896cbb56
DA		 2453 machine__for_each_thread(trace->host, trace__fprintf_one_thread, &data);
2454
2455 return data.printed;
1302d88e
ACM		 2456}
2457
ae9ed035
ACM		 2458static int trace__set_duration(const struct option *opt, const char *str,
2459 int unset __maybe_unused)
2460{
2461 struct trace *trace = opt->value;
2462
2463 trace->duration_filter = atof(str);
2464 return 0;
2465}
2466
c24ff998
ACM		 2467static int trace__open_output(struct trace *trace, const char *filename)
2468{
2469 struct stat st;
2470
2471 if (!stat(filename, &st) && st.st_size) {
2472 char oldname[PATH_MAX];
2473
2474 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
2475 unlink(oldname);
2476 rename(filename, oldname);
2477 }
2478
2479 trace->output = fopen(filename, "w");
2480
2481 return trace->output == NULL ? -errno : 0;
2482}
2483
598d02c5
SF		 2484static int parse_pagefaults(const struct option *opt, const char *str,
2485 int unset __maybe_unused)
2486{
2487 int *trace_pgfaults = opt->value;
2488
2489 if (strcmp(str, "all") == 0)
2490 *trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
2491 else if (strcmp(str, "maj") == 0)
2492 *trace_pgfaults |= TRACE_PFMAJ;
2493 else if (strcmp(str, "min") == 0)
2494 *trace_pgfaults |= TRACE_PFMIN;
2495 else
2496 return -1;
2497
2498 return 0;
2499}
2500
514f1c67
ACM		 2501int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
2502{
2503 const char * const trace_usage[] = {
f15eb531
NK		 2504 "perf trace [<options>] [<command>]",
2505 "perf trace [<options>] -- <command> [<options>]",
5e2485b1
DA		 2506 "perf trace record [<options>] [<command>]",
2507 "perf trace record [<options>] -- <command> [<options>]",
514f1c67
ACM		 2508 NULL
2509 };
2510 struct trace trace = {
c522739d
ACM		 2511 .audit = {
2512 .machine = audit_detect_machine(),
2513 .open_id = audit_name_to_syscall("open", trace.audit.machine),
2514 },
514f1c67
ACM		 2515 .syscalls = {
2516 . max = -1,
2517 },
2518 .opts = {
2519 .target = {
2520 .uid = UINT_MAX,
2521 .uses_mmap = true,
2522 },
2523 .user_freq = UINT_MAX,
2524 .user_interval = ULLONG_MAX,
509051ea 2525 .no_buffering = true,
38d5447d 2526 .mmap_pages = UINT_MAX,
514f1c67 2527 },
c24ff998 2528 .output = stdout,
50c95cbd 2529 .show_comm = true,
e281a960 2530 .trace_syscalls = true,
514f1c67 2531 };
c24ff998 2532 const char *output_name = NULL;
2ae3a312 2533 const char *ev_qualifier_str = NULL;
514f1c67 2534 const struct option trace_options[] = {
50c95cbd
ACM		 2535 OPT_BOOLEAN(0, "comm", &trace.show_comm,
2536 "show the thread COMM next to its id"),
c522739d 2537 OPT_BOOLEAN(0, "tool_stats", &trace.show_tool_stats, "show tool stats"),
2ae3a312
ACM		 2538 OPT_STRING('e', "expr", &ev_qualifier_str, "expr",
2539 "list of events to trace"),
c24ff998 2540 OPT_STRING('o', "output", &output_name, "file", "output file name"),
6810fc91 2541 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
514f1c67
ACM		 2542 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
2543 "trace events on existing process id"),
ac9be8ee 2544 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
514f1c67 2545 "trace events on existing thread id"),
ac9be8ee 2546 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
514f1c67 2547 "system-wide collection from all CPUs"),
ac9be8ee 2548 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
514f1c67 2549 "list of cpus to monitor"),
6810fc91 2550 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
514f1c67 2551 "child tasks do not inherit counters"),
994a1f78
JO		 2552 OPT_CALLBACK('m', "mmap-pages", &trace.opts.mmap_pages, "pages",
2553 "number of mmap data pages",
2554 perf_evlist__parse_mmap_pages),
ac9be8ee 2555 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
514f1c67 2556 "user to profile"),
ae9ed035
ACM		 2557 OPT_CALLBACK(0, "duration", &trace, "float",
2558 "show only events with duration > N.M ms",
2559 trace__set_duration),
1302d88e 2560 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
7c304ee0 2561 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
4bb09192
DA		 2562 OPT_BOOLEAN('T', "time", &trace.full_time,
2563 "Show full timestamp, not time relative to first start"),
fd2eabaf
DA		 2564 OPT_BOOLEAN('s', "summary", &trace.summary_only,
2565 "Show only syscall summary with statistics"),
2566 OPT_BOOLEAN('S', "with-summary", &trace.summary,
2567 "Show all syscalls and summary with statistics"),
598d02c5
SF		 2568 OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
2569 "Trace pagefaults", parse_pagefaults, "maj"),
e281a960 2570 OPT_BOOLEAN(0, "syscalls", &trace.trace_syscalls, "Trace syscalls"),
514f1c67
ACM		 2571 OPT_END()
2572 };
2573 int err;
32caf0d1 2574 char bf[BUFSIZ];
514f1c67 2575
1e28fe0a
SF		 2576 argc = parse_options(argc, argv, trace_options, trace_usage,
2577 PARSE_OPT_STOP_AT_NON_OPTION);
fd2eabaf 2578
598d02c5
SF		 2579 if (trace.trace_pgfaults) {
2580 trace.opts.sample_address = true;
2581 trace.opts.sample_time = true;
2582 }
2583
1e28fe0a
SF		 2584 if ((argc >= 1) && (strcmp(argv[0], "record") == 0))
2585 return trace__record(&trace, argc-1, &argv[1]);
2586
2587 /* summary_only implies summary option, but don't overwrite summary if set */
2588 if (trace.summary_only)
2589 trace.summary = trace.summary_only;
2590
e281a960
SF		 2591 if (!trace.trace_syscalls && !trace.trace_pgfaults) {
2592 pr_err("Please specify something to trace.\n");
2593 return -1;
2594 }
2595
c24ff998
ACM		 2596 if (output_name != NULL) {
2597 err = trace__open_output(&trace, output_name);
2598 if (err < 0) {
2599 perror("failed to create output file");
2600 goto out;
2601 }
2602 }
2603
2ae3a312 2604 if (ev_qualifier_str != NULL) {
b059efdf
ACM		 2605 const char *s = ev_qualifier_str;
2606
2607 trace.not_ev_qualifier = *s == '!';
2608 if (trace.not_ev_qualifier)
2609 ++s;
2610 trace.ev_qualifier = strlist__new(true, s);
2ae3a312 2611 if (trace.ev_qualifier == NULL) {
c24ff998
ACM		 2612 fputs("Not enough memory to parse event qualifier",
2613 trace.output);
2614 err = -ENOMEM;
2615 goto out_close;
2ae3a312
ACM		 2616 }
2617 }
2618
602ad878 2619 err = target__validate(&trace.opts.target);
32caf0d1 2620 if (err) {
602ad878 2621 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 2622 fprintf(trace.output, "%s", bf);
2623 goto out_close;
32caf0d1
NK		 2624 }
2625
602ad878 2626 err = target__parse_uid(&trace.opts.target);
514f1c67 2627 if (err) {
602ad878 2628 target__strerror(&trace.opts.target, err, bf, sizeof(bf));
c24ff998
ACM		 2629 fprintf(trace.output, "%s", bf);
2630 goto out_close;
514f1c67
ACM		 2631 }
2632
602ad878 2633 if (!argc && target__none(&trace.opts.target))
ee76120e
NK		 2634 trace.opts.target.system_wide = true;
2635
6810fc91
DA		 2636 if (input_name)
2637 err = trace__replay(&trace);
2638 else
2639 err = trace__run(&trace, argc, argv);
1302d88e 2640
c24ff998
ACM		 2641out_close:
2642 if (output_name != NULL)
2643 fclose(trace.output);
2644out:
1302d88e 2645 return err;
514f1c67 2646}
ubuntu-bionic-kernel mirror