?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
USB: see if URB comes from a completion handler
[linux/fpc-iii.git] / tools / perf / builtin-trace.c
blobf5aa6375e3e9add4eea3cdd81de685fa37d87c4d
1 #include <traceevent/event-parse.h>
2 #include "builtin.h"
3 #include "util/color.h"
4 #include "util/debug.h"
5 #include "util/evlist.h"
6 #include "util/machine.h"
7 #include "util/session.h"
8 #include "util/thread.h"
9 #include "util/parse-options.h"
10 #include "util/strlist.h"
11 #include "util/intlist.h"
12 #include "util/thread_map.h"
13
14 #include <libaudit.h>
15 #include <stdlib.h>
16 #include <sys/mman.h>
17 #include <linux/futex.h>
18
19 static size_t syscall_arg__scnprintf_hex(char *bf, size_t size,
20 unsigned long arg,
21 u8 arg_idx __maybe_unused,
22 u8 *arg_mask __maybe_unused)
23 {
24 return scnprintf(bf, size, "%#lx", arg);
25 }
26
27 #define SCA_HEX syscall_arg__scnprintf_hex
28
29 static size_t syscall_arg__scnprintf_whence(char *bf, size_t size,
30 unsigned long arg,
31 u8 arg_idx __maybe_unused,
32 u8 *arg_mask __maybe_unused)
33 {
34 int whence = arg;
35
36 switch (whence) {
37 #define P_WHENCE(n) case SEEK_##n: return scnprintf(bf, size, #n)
38 P_WHENCE(SET);
39 P_WHENCE(CUR);
40 P_WHENCE(END);
41 #ifdef SEEK_DATA
42 P_WHENCE(DATA);
43 #endif
44 #ifdef SEEK_HOLE
45 P_WHENCE(HOLE);
46 #endif
47 #undef P_WHENCE
48 default: break;
49 }
50
51 return scnprintf(bf, size, "%#x", whence);
52 }
53
54 #define SCA_WHENCE syscall_arg__scnprintf_whence
55
56 static size_t syscall_arg__scnprintf_mmap_prot(char *bf, size_t size,
57 unsigned long arg,
58 u8 arg_idx __maybe_unused,
59 u8 *arg_mask __maybe_unused)
60 {
61 int printed = 0, prot = arg;
62
63 if (prot == PROT_NONE)
64 return scnprintf(bf, size, "NONE");
65 #define P_MMAP_PROT(n) \
66 if (prot & PROT_##n) { \
67 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
68 prot &= ~PROT_##n; \
69 }
70
71 P_MMAP_PROT(EXEC);
72 P_MMAP_PROT(READ);
73 P_MMAP_PROT(WRITE);
74 #ifdef PROT_SEM
75 P_MMAP_PROT(SEM);
76 #endif
77 P_MMAP_PROT(GROWSDOWN);
78 P_MMAP_PROT(GROWSUP);
79 #undef P_MMAP_PROT
80
81 if (prot)
82 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", prot);
83
84 return printed;
85 }
86
87 #define SCA_MMAP_PROT syscall_arg__scnprintf_mmap_prot
88
89 static size_t syscall_arg__scnprintf_mmap_flags(char *bf, size_t size,
90 unsigned long arg, u8 arg_idx __maybe_unused,
91 u8 *arg_mask __maybe_unused)
92 {
93 int printed = 0, flags = arg;
94
95 #define P_MMAP_FLAG(n) \
96 if (flags & MAP_##n) { \
97 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
98 flags &= ~MAP_##n; \
99 }
100
101 P_MMAP_FLAG(SHARED);
102 P_MMAP_FLAG(PRIVATE);
103 #ifdef MAP_32BIT
104 P_MMAP_FLAG(32BIT);
105 #endif
106 P_MMAP_FLAG(ANONYMOUS);
107 P_MMAP_FLAG(DENYWRITE);
108 P_MMAP_FLAG(EXECUTABLE);
109 P_MMAP_FLAG(FILE);
110 P_MMAP_FLAG(FIXED);
111 P_MMAP_FLAG(GROWSDOWN);
112 #ifdef MAP_HUGETLB
113 P_MMAP_FLAG(HUGETLB);
114 #endif
115 P_MMAP_FLAG(LOCKED);
116 P_MMAP_FLAG(NONBLOCK);
117 P_MMAP_FLAG(NORESERVE);
118 P_MMAP_FLAG(POPULATE);
119 P_MMAP_FLAG(STACK);
120 #ifdef MAP_UNINITIALIZED
121 P_MMAP_FLAG(UNINITIALIZED);
122 #endif
123 #undef P_MMAP_FLAG
124
125 if (flags)
126 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
127
128 return printed;
129 }
130
131 #define SCA_MMAP_FLAGS syscall_arg__scnprintf_mmap_flags
132
133 static size_t syscall_arg__scnprintf_madvise_behavior(char *bf, size_t size,
134 unsigned long arg, u8 arg_idx __maybe_unused,
135 u8 *arg_mask __maybe_unused)
136 {
137 int behavior = arg;
138
139 switch (behavior) {
140 #define P_MADV_BHV(n) case MADV_##n: return scnprintf(bf, size, #n)
141 P_MADV_BHV(NORMAL);
142 P_MADV_BHV(RANDOM);
143 P_MADV_BHV(SEQUENTIAL);
144 P_MADV_BHV(WILLNEED);
145 P_MADV_BHV(DONTNEED);
146 P_MADV_BHV(REMOVE);
147 P_MADV_BHV(DONTFORK);
148 P_MADV_BHV(DOFORK);
149 P_MADV_BHV(HWPOISON);
150 #ifdef MADV_SOFT_OFFLINE
151 P_MADV_BHV(SOFT_OFFLINE);
152 #endif
153 P_MADV_BHV(MERGEABLE);
154 P_MADV_BHV(UNMERGEABLE);
155 #ifdef MADV_HUGEPAGE
156 P_MADV_BHV(HUGEPAGE);
157 #endif
158 #ifdef MADV_NOHUGEPAGE
159 P_MADV_BHV(NOHUGEPAGE);
160 #endif
161 #ifdef MADV_DONTDUMP
162 P_MADV_BHV(DONTDUMP);
163 #endif
164 #ifdef MADV_DODUMP
165 P_MADV_BHV(DODUMP);
166 #endif
167 #undef P_MADV_PHV
168 default: break;
169 }
170
171 return scnprintf(bf, size, "%#x", behavior);
172 }
173
174 #define SCA_MADV_BHV syscall_arg__scnprintf_madvise_behavior
175
176 static size_t syscall_arg__scnprintf_futex_op(char *bf, size_t size, unsigned long arg,
177 u8 arg_idx __maybe_unused, u8 *arg_mask)
178 {
179 enum syscall_futex_args {
180 SCF_UADDR = (1 << 0),
181 SCF_OP = (1 << 1),
182 SCF_VAL = (1 << 2),
183 SCF_TIMEOUT = (1 << 3),
184 SCF_UADDR2 = (1 << 4),
185 SCF_VAL3 = (1 << 5),
186 };
187 int op = arg;
188 int cmd = op & FUTEX_CMD_MASK;
189 size_t printed = 0;
190
191 switch (cmd) {
192 #define P_FUTEX_OP(n) case FUTEX_##n: printed = scnprintf(bf, size, #n);
193 P_FUTEX_OP(WAIT); *arg_mask |= SCF_VAL3|SCF_UADDR2; break;
194 P_FUTEX_OP(WAKE); *arg_mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
195 P_FUTEX_OP(FD); *arg_mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
196 P_FUTEX_OP(REQUEUE); *arg_mask |= SCF_VAL3|SCF_TIMEOUT; break;
197 P_FUTEX_OP(CMP_REQUEUE); *arg_mask |= SCF_TIMEOUT; break;
198 P_FUTEX_OP(CMP_REQUEUE_PI); *arg_mask |= SCF_TIMEOUT; break;
199 P_FUTEX_OP(WAKE_OP); break;
200 P_FUTEX_OP(LOCK_PI); *arg_mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
201 P_FUTEX_OP(UNLOCK_PI); *arg_mask |= SCF_VAL3|SCF_UADDR2|SCF_TIMEOUT; break;
202 P_FUTEX_OP(TRYLOCK_PI); *arg_mask |= SCF_VAL3|SCF_UADDR2; break;
203 P_FUTEX_OP(WAIT_BITSET); *arg_mask |= SCF_UADDR2; break;
204 P_FUTEX_OP(WAKE_BITSET); *arg_mask |= SCF_UADDR2; break;
205 P_FUTEX_OP(WAIT_REQUEUE_PI); break;
206 default: printed = scnprintf(bf, size, "%#x", cmd); break;
207 }
208
209 if (op & FUTEX_PRIVATE_FLAG)
210 printed += scnprintf(bf + printed, size - printed, "|PRIV");
211
212 if (op & FUTEX_CLOCK_REALTIME)
213 printed += scnprintf(bf + printed, size - printed, "|CLKRT");
214
215 return printed;
216 }
217
218 #define SCA_FUTEX_OP syscall_arg__scnprintf_futex_op
219
220 static size_t syscall_arg__scnprintf_open_flags(char *bf, size_t size,
221 unsigned long arg,
222 u8 arg_idx, u8 *arg_mask)
223 {
224 int printed = 0, flags = arg;
225
226 if (!(flags & O_CREAT))
227 *arg_mask |= 1 << (arg_idx + 1); /* Mask the mode parm */
228
229 if (flags == 0)
230 return scnprintf(bf, size, "RDONLY");
231 #define P_FLAG(n) \
232 if (flags & O_##n) { \
233 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", #n); \
234 flags &= ~O_##n; \
235 }
236
237 P_FLAG(APPEND);
238 P_FLAG(ASYNC);
239 P_FLAG(CLOEXEC);
240 P_FLAG(CREAT);
241 P_FLAG(DIRECT);
242 P_FLAG(DIRECTORY);
243 P_FLAG(EXCL);
244 P_FLAG(LARGEFILE);
245 P_FLAG(NOATIME);
246 P_FLAG(NOCTTY);
247 #ifdef O_NONBLOCK
248 P_FLAG(NONBLOCK);
249 #elif O_NDELAY
250 P_FLAG(NDELAY);
251 #endif
252 #ifdef O_PATH
253 P_FLAG(PATH);
254 #endif
255 P_FLAG(RDWR);
256 #ifdef O_DSYNC
257 if ((flags & O_SYNC) == O_SYNC)
258 printed += scnprintf(bf + printed, size - printed, "%s%s", printed ? "|" : "", "SYNC");
259 else {
260 P_FLAG(DSYNC);
261 }
262 #else
263 P_FLAG(SYNC);
264 #endif
265 P_FLAG(TRUNC);
266 P_FLAG(WRONLY);
267 #undef P_FLAG
268
269 if (flags)
270 printed += scnprintf(bf + printed, size - printed, "%s%#x", printed ? "|" : "", flags);
271
272 return printed;
273 }
274
275 #define SCA_OPEN_FLAGS syscall_arg__scnprintf_open_flags
276
277 static struct syscall_fmt {
278 const char *name;
279 const char *alias;
280 size_t (*arg_scnprintf[6])(char *bf, size_t size, unsigned long arg, u8 arg_idx, u8 *arg_mask);
281 bool errmsg;
282 bool timeout;
283 bool hexret;
284 } syscall_fmts[] = {
285 { .name = "access", .errmsg = true, },
286 { .name = "arch_prctl", .errmsg = true, .alias = "prctl", },
287 { .name = "brk", .hexret = true,
288 .arg_scnprintf = { [0] = SCA_HEX, /* brk */ }, },
289 { .name = "mmap", .hexret = true, },
290 { .name = "connect", .errmsg = true, },
291 { .name = "fstat", .errmsg = true, .alias = "newfstat", },
292 { .name = "fstatat", .errmsg = true, .alias = "newfstatat", },
293 { .name = "futex", .errmsg = true,
294 .arg_scnprintf = { [1] = SCA_FUTEX_OP, /* op */ }, },
295 { .name = "ioctl", .errmsg = true,
296 .arg_scnprintf = { [2] = SCA_HEX, /* arg */ }, },
297 { .name = "lseek", .errmsg = true,
298 .arg_scnprintf = { [2] = SCA_WHENCE, /* whence */ }, },
299 { .name = "lstat", .errmsg = true, .alias = "newlstat", },
300 { .name = "madvise", .errmsg = true,
301 .arg_scnprintf = { [0] = SCA_HEX, /* start */
302 [2] = SCA_MADV_BHV, /* behavior */ }, },
303 { .name = "mmap", .hexret = true,
304 .arg_scnprintf = { [0] = SCA_HEX, /* addr */
305 [2] = SCA_MMAP_PROT, /* prot */
306 [3] = SCA_MMAP_FLAGS, /* flags */ }, },
307 { .name = "mprotect", .errmsg = true,
308 .arg_scnprintf = { [0] = SCA_HEX, /* start */
309 [2] = SCA_MMAP_PROT, /* prot */ }, },
310 { .name = "mremap", .hexret = true,
311 .arg_scnprintf = { [0] = SCA_HEX, /* addr */
312 [4] = SCA_HEX, /* new_addr */ }, },
313 { .name = "munmap", .errmsg = true,
314 .arg_scnprintf = { [0] = SCA_HEX, /* addr */ }, },
315 { .name = "open", .errmsg = true,
316 .arg_scnprintf = { [1] = SCA_OPEN_FLAGS, /* flags */ }, },
317 { .name = "open_by_handle_at", .errmsg = true,
318 .arg_scnprintf = { [2] = SCA_OPEN_FLAGS, /* flags */ }, },
319 { .name = "openat", .errmsg = true,
320 .arg_scnprintf = { [2] = SCA_OPEN_FLAGS, /* flags */ }, },
321 { .name = "poll", .errmsg = true, .timeout = true, },
322 { .name = "ppoll", .errmsg = true, .timeout = true, },
323 { .name = "pread", .errmsg = true, .alias = "pread64", },
324 { .name = "pwrite", .errmsg = true, .alias = "pwrite64", },
325 { .name = "read", .errmsg = true, },
326 { .name = "recvfrom", .errmsg = true, },
327 { .name = "select", .errmsg = true, .timeout = true, },
328 { .name = "socket", .errmsg = true, },
329 { .name = "stat", .errmsg = true, .alias = "newstat", },
330 { .name = "uname", .errmsg = true, .alias = "newuname", },
331 };
332
333 static int syscall_fmt__cmp(const void *name, const void *fmtp)
334 {
335 const struct syscall_fmt *fmt = fmtp;
336 return strcmp(name, fmt->name);
337 }
338
339 static struct syscall_fmt *syscall_fmt__find(const char *name)
340 {
341 const int nmemb = ARRAY_SIZE(syscall_fmts);
342 return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
343 }
344
345 struct syscall {
346 struct event_format *tp_format;
347 const char *name;
348 bool filtered;
349 struct syscall_fmt *fmt;
350 size_t (**arg_scnprintf)(char *bf, size_t size,
351 unsigned long arg, u8 arg_idx, u8 *args_mask);
352 };
353
354 static size_t fprintf_duration(unsigned long t, FILE *fp)
355 {
356 double duration = (double)t / NSEC_PER_MSEC;
357 size_t printed = fprintf(fp, "(");
358
359 if (duration >= 1.0)
360 printed += color_fprintf(fp, PERF_COLOR_RED, "%6.3f ms", duration);
361 else if (duration >= 0.01)
362 printed += color_fprintf(fp, PERF_COLOR_YELLOW, "%6.3f ms", duration);
363 else
364 printed += color_fprintf(fp, PERF_COLOR_NORMAL, "%6.3f ms", duration);
365 return printed + fprintf(fp, "): ");
366 }
367
368 struct thread_trace {
369 u64 entry_time;
370 u64 exit_time;
371 bool entry_pending;
372 unsigned long nr_events;
373 char *entry_str;
374 double runtime_ms;
375 };
376
377 static struct thread_trace *thread_trace__new(void)
378 {
379 return zalloc(sizeof(struct thread_trace));
380 }
381
382 static struct thread_trace *thread__trace(struct thread *thread, FILE *fp)
383 {
384 struct thread_trace *ttrace;
385
386 if (thread == NULL)
387 goto fail;
388
389 if (thread->priv == NULL)
390 thread->priv = thread_trace__new();
391
392 if (thread->priv == NULL)
393 goto fail;
394
395 ttrace = thread->priv;
396 ++ttrace->nr_events;
397
398 return ttrace;
399 fail:
400 color_fprintf(fp, PERF_COLOR_RED,
401 "WARNING: not enough memory, dropping samples!\n");
402 return NULL;
403 }
404
405 struct trace {
406 struct perf_tool tool;
407 int audit_machine;
408 struct {
409 int max;
410 struct syscall *table;
411 } syscalls;
412 struct perf_record_opts opts;
413 struct machine host;
414 u64 base_time;
415 FILE *output;
416 unsigned long nr_events;
417 struct strlist *ev_qualifier;
418 bool not_ev_qualifier;
419 struct intlist *tid_list;
420 struct intlist *pid_list;
421 bool sched;
422 bool multiple_threads;
423 double duration_filter;
424 double runtime_ms;
425 };
426
427 static bool trace__filter_duration(struct trace *trace, double t)
428 {
429 return t < (trace->duration_filter * NSEC_PER_MSEC);
430 }
431
432 static size_t trace__fprintf_tstamp(struct trace *trace, u64 tstamp, FILE *fp)
433 {
434 double ts = (double)(tstamp - trace->base_time) / NSEC_PER_MSEC;
435
436 return fprintf(fp, "%10.3f ", ts);
437 }
438
439 static bool done = false;
440
441 static void sig_handler(int sig __maybe_unused)
442 {
443 done = true;
444 }
445
446 static size_t trace__fprintf_entry_head(struct trace *trace, struct thread *thread,
447 u64 duration, u64 tstamp, FILE *fp)
448 {
449 size_t printed = trace__fprintf_tstamp(trace, tstamp, fp);
450 printed += fprintf_duration(duration, fp);
451
452 if (trace->multiple_threads)
453 printed += fprintf(fp, "%d ", thread->tid);
454
455 return printed;
456 }
457
458 static int trace__process_event(struct trace *trace, struct machine *machine,
459 union perf_event *event)
460 {
461 int ret = 0;
462
463 switch (event->header.type) {
464 case PERF_RECORD_LOST:
465 color_fprintf(trace->output, PERF_COLOR_RED,
466 "LOST %" PRIu64 " events!\n", event->lost.lost);
467 ret = machine__process_lost_event(machine, event);
468 default:
469 ret = machine__process_event(machine, event);
470 break;
471 }
472
473 return ret;
474 }
475
476 static int trace__tool_process(struct perf_tool *tool,
477 union perf_event *event,
478 struct perf_sample *sample __maybe_unused,
479 struct machine *machine)
480 {
481 struct trace *trace = container_of(tool, struct trace, tool);
482 return trace__process_event(trace, machine, event);
483 }
484
485 static int trace__symbols_init(struct trace *trace, struct perf_evlist *evlist)
486 {
487 int err = symbol__init();
488
489 if (err)
490 return err;
491
492 machine__init(&trace->host, "", HOST_KERNEL_ID);
493 machine__create_kernel_maps(&trace->host);
494
495 if (perf_target__has_task(&trace->opts.target)) {
496 err = perf_event__synthesize_thread_map(&trace->tool, evlist->threads,
497 trace__tool_process,
498 &trace->host);
499 } else {
500 err = perf_event__synthesize_threads(&trace->tool, trace__tool_process,
501 &trace->host);
502 }
503
504 if (err)
505 symbol__exit();
506
507 return err;
508 }
509
510 static int syscall__set_arg_fmts(struct syscall *sc)
511 {
512 struct format_field *field;
513 int idx = 0;
514
515 sc->arg_scnprintf = calloc(sc->tp_format->format.nr_fields - 1, sizeof(void *));
516 if (sc->arg_scnprintf == NULL)
517 return -1;
518
519 for (field = sc->tp_format->format.fields->next; field; field = field->next) {
520 if (sc->fmt && sc->fmt->arg_scnprintf[idx])
521 sc->arg_scnprintf[idx] = sc->fmt->arg_scnprintf[idx];
522 else if (field->flags & FIELD_IS_POINTER)
523 sc->arg_scnprintf[idx] = syscall_arg__scnprintf_hex;
524 ++idx;
525 }
526
527 return 0;
528 }
529
530 static int trace__read_syscall_info(struct trace *trace, int id)
531 {
532 char tp_name[128];
533 struct syscall *sc;
534 const char *name = audit_syscall_to_name(id, trace->audit_machine);
535
536 if (name == NULL)
537 return -1;
538
539 if (id > trace->syscalls.max) {
540 struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
541
542 if (nsyscalls == NULL)
543 return -1;
544
545 if (trace->syscalls.max != -1) {
546 memset(nsyscalls + trace->syscalls.max + 1, 0,
547 (id - trace->syscalls.max) * sizeof(*sc));
548 } else {
549 memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
550 }
551
552 trace->syscalls.table = nsyscalls;
553 trace->syscalls.max = id;
554 }
555
556 sc = trace->syscalls.table + id;
557 sc->name = name;
558
559 if (trace->ev_qualifier) {
560 bool in = strlist__find(trace->ev_qualifier, name) != NULL;
561
562 if (!(in ^ trace->not_ev_qualifier)) {
563 sc->filtered = true;
564 /*
565 * No need to do read tracepoint information since this will be
566 * filtered out.
567 */
568 return 0;
569 }
570 }
571
572 sc->fmt = syscall_fmt__find(sc->name);
573
574 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
575 sc->tp_format = event_format__new("syscalls", tp_name);
576
577 if (sc->tp_format == NULL && sc->fmt && sc->fmt->alias) {
578 snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
579 sc->tp_format = event_format__new("syscalls", tp_name);
580 }
581
582 if (sc->tp_format == NULL)
583 return -1;
584
585 return syscall__set_arg_fmts(sc);
586 }
587
588 static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,
589 unsigned long *args)
590 {
591 int i = 0;
592 size_t printed = 0;
593
594 if (sc->tp_format != NULL) {
595 struct format_field *field;
596 u8 mask = 0, bit = 1;
597
598 for (field = sc->tp_format->format.fields->next; field;
599 field = field->next, ++i, bit <<= 1) {
600 if (mask & bit)
601 continue;
602
603 printed += scnprintf(bf + printed, size - printed,
604 "%s%s: ", printed ? ", " : "", field->name);
605
606 if (sc->arg_scnprintf && sc->arg_scnprintf[i]) {
607 printed += sc->arg_scnprintf[i](bf + printed, size - printed,
608 args[i], i, &mask);
609 } else {
610 printed += scnprintf(bf + printed, size - printed,
611 "%ld", args[i]);
612 }
613 }
614 } else {
615 while (i < 6) {
616 printed += scnprintf(bf + printed, size - printed,
617 "%sarg%d: %ld",
618 printed ? ", " : "", i, args[i]);
619 ++i;
620 }
621 }
622
623 return printed;
624 }
625
626 typedef int (*tracepoint_handler)(struct trace *trace, struct perf_evsel *evsel,
627 struct perf_sample *sample);
628
629 static struct syscall *trace__syscall_info(struct trace *trace,
630 struct perf_evsel *evsel,
631 struct perf_sample *sample)
632 {
633 int id = perf_evsel__intval(evsel, sample, "id");
634
635 if (id < 0) {
636
637 /*
638 * XXX: Noticed on x86_64, reproduced as far back as 3.0.36, haven't tried
639 * before that, leaving at a higher verbosity level till that is
640 * explained. Reproduced with plain ftrace with:
641 *
642 * echo 1 > /t/events/raw_syscalls/sys_exit/enable
643 * grep "NR -1 " /t/trace_pipe
644 *
645 * After generating some load on the machine.
646 */
647 if (verbose > 1) {
648 static u64 n;
649 fprintf(trace->output, "Invalid syscall %d id, skipping (%s, %" PRIu64 ") ...\n",
650 id, perf_evsel__name(evsel), ++n);
651 }
652 return NULL;
653 }
654
655 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
656 trace__read_syscall_info(trace, id))
657 goto out_cant_read;
658
659 if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
660 goto out_cant_read;
661
662 return &trace->syscalls.table[id];
663
664 out_cant_read:
665 if (verbose) {
666 fprintf(trace->output, "Problems reading syscall %d", id);
667 if (id <= trace->syscalls.max && trace->syscalls.table[id].name != NULL)
668 fprintf(trace->output, "(%s)", trace->syscalls.table[id].name);
669 fputs(" information\n", trace->output);
670 }
671 return NULL;
672 }
673
674 static int trace__sys_enter(struct trace *trace, struct perf_evsel *evsel,
675 struct perf_sample *sample)
676 {
677 char *msg;
678 void *args;
679 size_t printed = 0;
680 struct thread *thread;
681 struct syscall *sc = trace__syscall_info(trace, evsel, sample);
682 struct thread_trace *ttrace;
683
684 if (sc == NULL)
685 return -1;
686
687 if (sc->filtered)
688 return 0;
689
690 thread = machine__findnew_thread(&trace->host, sample->pid,
691 sample->tid);
692 ttrace = thread__trace(thread, trace->output);
693 if (ttrace == NULL)
694 return -1;
695
696 args = perf_evsel__rawptr(evsel, sample, "args");
697 if (args == NULL) {
698 fprintf(trace->output, "Problems reading syscall arguments\n");
699 return -1;
700 }
701
702 ttrace = thread->priv;
703
704 if (ttrace->entry_str == NULL) {
705 ttrace->entry_str = malloc(1024);
706 if (!ttrace->entry_str)
707 return -1;
708 }
709
710 ttrace->entry_time = sample->time;
711 msg = ttrace->entry_str;
712 printed += scnprintf(msg + printed, 1024 - printed, "%s(", sc->name);
713
714 printed += syscall__scnprintf_args(sc, msg + printed, 1024 - printed, args);
715
716 if (!strcmp(sc->name, "exit_group") || !strcmp(sc->name, "exit")) {
717 if (!trace->duration_filter) {
718 trace__fprintf_entry_head(trace, thread, 1, sample->time, trace->output);
719 fprintf(trace->output, "%-70s\n", ttrace->entry_str);
720 }
721 } else
722 ttrace->entry_pending = true;
723
724 return 0;
725 }
726
727 static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
728 struct perf_sample *sample)
729 {
730 int ret;
731 u64 duration = 0;
732 struct thread *thread;
733 struct syscall *sc = trace__syscall_info(trace, evsel, sample);
734 struct thread_trace *ttrace;
735
736 if (sc == NULL)
737 return -1;
738
739 if (sc->filtered)
740 return 0;
741
742 thread = machine__findnew_thread(&trace->host, sample->pid,
743 sample->tid);
744 ttrace = thread__trace(thread, trace->output);
745 if (ttrace == NULL)
746 return -1;
747
748 ret = perf_evsel__intval(evsel, sample, "ret");
749
750 ttrace = thread->priv;
751
752 ttrace->exit_time = sample->time;
753
754 if (ttrace->entry_time) {
755 duration = sample->time - ttrace->entry_time;
756 if (trace__filter_duration(trace, duration))
757 goto out;
758 } else if (trace->duration_filter)
759 goto out;
760
761 trace__fprintf_entry_head(trace, thread, duration, sample->time, trace->output);
762
763 if (ttrace->entry_pending) {
764 fprintf(trace->output, "%-70s", ttrace->entry_str);
765 } else {
766 fprintf(trace->output, " ... [");
767 color_fprintf(trace->output, PERF_COLOR_YELLOW, "continued");
768 fprintf(trace->output, "]: %s()", sc->name);
769 }
770
771 if (sc->fmt == NULL) {
772 signed_print:
773 fprintf(trace->output, ") = %d", ret);
774 } else if (ret < 0 && sc->fmt->errmsg) {
775 char bf[256];
776 const char *emsg = strerror_r(-ret, bf, sizeof(bf)),
777 *e = audit_errno_to_name(-ret);
778
779 fprintf(trace->output, ") = -1 %s %s", e, emsg);
780 } else if (ret == 0 && sc->fmt->timeout)
781 fprintf(trace->output, ") = 0 Timeout");
782 else if (sc->fmt->hexret)
783 fprintf(trace->output, ") = %#x", ret);
784 else
785 goto signed_print;
786
787 fputc('\n', trace->output);
788 out:
789 ttrace->entry_pending = false;
790
791 return 0;
792 }
793
794 static int trace__sched_stat_runtime(struct trace *trace, struct perf_evsel *evsel,
795 struct perf_sample *sample)
796 {
797 u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
798 double runtime_ms = (double)runtime / NSEC_PER_MSEC;
799 struct thread *thread = machine__findnew_thread(&trace->host,
800 sample->pid,
801 sample->tid);
802 struct thread_trace *ttrace = thread__trace(thread, trace->output);
803
804 if (ttrace == NULL)
805 goto out_dump;
806
807 ttrace->runtime_ms += runtime_ms;
808 trace->runtime_ms += runtime_ms;
809 return 0;
810
811 out_dump:
812 fprintf(trace->output, "%s: comm=%s,pid=%u,runtime=%" PRIu64 ",vruntime=%" PRIu64 ")\n",
813 evsel->name,
814 perf_evsel__strval(evsel, sample, "comm"),
815 (pid_t)perf_evsel__intval(evsel, sample, "pid"),
816 runtime,
817 perf_evsel__intval(evsel, sample, "vruntime"));
818 return 0;
819 }
820
821 static bool skip_sample(struct trace *trace, struct perf_sample *sample)
822 {
823 if ((trace->pid_list && intlist__find(trace->pid_list, sample->pid)) ||
824 (trace->tid_list && intlist__find(trace->tid_list, sample->tid)))
825 return false;
826
827 if (trace->pid_list || trace->tid_list)
828 return true;
829
830 return false;
831 }
832
833 static int trace__process_sample(struct perf_tool *tool,
834 union perf_event *event __maybe_unused,
835 struct perf_sample *sample,
836 struct perf_evsel *evsel,
837 struct machine *machine __maybe_unused)
838 {
839 struct trace *trace = container_of(tool, struct trace, tool);
840 int err = 0;
841
842 tracepoint_handler handler = evsel->handler.func;
843
844 if (skip_sample(trace, sample))
845 return 0;
846
847 if (trace->base_time == 0)
848 trace->base_time = sample->time;
849
850 if (handler)
851 handler(trace, evsel, sample);
852
853 return err;
854 }
855
856 static bool
857 perf_session__has_tp(struct perf_session *session, const char *name)
858 {
859 struct perf_evsel *evsel;
860
861 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, name);
862
863 return evsel != NULL;
864 }
865
866 static int parse_target_str(struct trace *trace)
867 {
868 if (trace->opts.target.pid) {
869 trace->pid_list = intlist__new(trace->opts.target.pid);
870 if (trace->pid_list == NULL) {
871 pr_err("Error parsing process id string\n");
872 return -EINVAL;
873 }
874 }
875
876 if (trace->opts.target.tid) {
877 trace->tid_list = intlist__new(trace->opts.target.tid);
878 if (trace->tid_list == NULL) {
879 pr_err("Error parsing thread id string\n");
880 return -EINVAL;
881 }
882 }
883
884 return 0;
885 }
886
887 static int trace__run(struct trace *trace, int argc, const char **argv)
888 {
889 struct perf_evlist *evlist = perf_evlist__new();
890 struct perf_evsel *evsel;
891 int err = -1, i;
892 unsigned long before;
893 const bool forks = argc > 0;
894
895 if (evlist == NULL) {
896 fprintf(trace->output, "Not enough memory to run!\n");
897 goto out;
898 }
899
900 if (perf_evlist__add_newtp(evlist, "raw_syscalls", "sys_enter", trace__sys_enter) ||
901 perf_evlist__add_newtp(evlist, "raw_syscalls", "sys_exit", trace__sys_exit)) {
902 fprintf(trace->output, "Couldn't read the raw_syscalls tracepoints information!\n");
903 goto out_delete_evlist;
904 }
905
906 if (trace->sched &&
907 perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
908 trace__sched_stat_runtime)) {
909 fprintf(trace->output, "Couldn't read the sched_stat_runtime tracepoint information!\n");
910 goto out_delete_evlist;
911 }
912
913 err = perf_evlist__create_maps(evlist, &trace->opts.target);
914 if (err < 0) {
915 fprintf(trace->output, "Problems parsing the target to trace, check your options!\n");
916 goto out_delete_evlist;
917 }
918
919 err = trace__symbols_init(trace, evlist);
920 if (err < 0) {
921 fprintf(trace->output, "Problems initializing symbol libraries!\n");
922 goto out_delete_maps;
923 }
924
925 perf_evlist__config(evlist, &trace->opts);
926
927 signal(SIGCHLD, sig_handler);
928 signal(SIGINT, sig_handler);
929
930 if (forks) {
931 err = perf_evlist__prepare_workload(evlist, &trace->opts.target,
932 argv, false, false);
933 if (err < 0) {
934 fprintf(trace->output, "Couldn't run the workload!\n");
935 goto out_delete_maps;
936 }
937 }
938
939 err = perf_evlist__open(evlist);
940 if (err < 0) {
941 fprintf(trace->output, "Couldn't create the events: %s\n", strerror(errno));
942 goto out_delete_maps;
943 }
944
945 err = perf_evlist__mmap(evlist, UINT_MAX, false);
946 if (err < 0) {
947 fprintf(trace->output, "Couldn't mmap the events: %s\n", strerror(errno));
948 goto out_close_evlist;
949 }
950
951 perf_evlist__enable(evlist);
952
953 if (forks)
954 perf_evlist__start_workload(evlist);
955
956 trace->multiple_threads = evlist->threads->map[0] == -1 || evlist->threads->nr > 1;
957 again:
958 before = trace->nr_events;
959
960 for (i = 0; i < evlist->nr_mmaps; i++) {
961 union perf_event *event;
962
963 while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
964 const u32 type = event->header.type;
965 tracepoint_handler handler;
966 struct perf_sample sample;
967
968 ++trace->nr_events;
969
970 err = perf_evlist__parse_sample(evlist, event, &sample);
971 if (err) {
972 fprintf(trace->output, "Can't parse sample, err = %d, skipping...\n", err);
973 continue;
974 }
975
976 if (trace->base_time == 0)
977 trace->base_time = sample.time;
978
979 if (type != PERF_RECORD_SAMPLE) {
980 trace__process_event(trace, &trace->host, event);
981 continue;
982 }
983
984 evsel = perf_evlist__id2evsel(evlist, sample.id);
985 if (evsel == NULL) {
986 fprintf(trace->output, "Unknown tp ID %" PRIu64 ", skipping...\n", sample.id);
987 continue;
988 }
989
990 if (sample.raw_data == NULL) {
991 fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
992 perf_evsel__name(evsel), sample.tid,
993 sample.cpu, sample.raw_size);
994 continue;
995 }
996
997 handler = evsel->handler.func;
998 handler(trace, evsel, &sample);
999
1000 if (done)
1001 goto out_unmap_evlist;
1002 }
1003 }
1004
1005 if (trace->nr_events == before) {
1006 if (done)
1007 goto out_unmap_evlist;
1008
1009 poll(evlist->pollfd, evlist->nr_fds, -1);
1010 }
1011
1012 if (done)
1013 perf_evlist__disable(evlist);
1014
1015 goto again;
1016
1017 out_unmap_evlist:
1018 perf_evlist__munmap(evlist);
1019 out_close_evlist:
1020 perf_evlist__close(evlist);
1021 out_delete_maps:
1022 perf_evlist__delete_maps(evlist);
1023 out_delete_evlist:
1024 perf_evlist__delete(evlist);
1025 out:
1026 return err;
1027 }
1028
1029 static int trace__replay(struct trace *trace)
1030 {
1031 const struct perf_evsel_str_handler handlers[] = {
1032 { "raw_syscalls:sys_enter", trace__sys_enter, },
1033 { "raw_syscalls:sys_exit", trace__sys_exit, },
1034 };
1035
1036 struct perf_session *session;
1037 int err = -1;
1038
1039 trace->tool.sample = trace__process_sample;
1040 trace->tool.mmap = perf_event__process_mmap;
1041 trace->tool.comm = perf_event__process_comm;
1042 trace->tool.exit = perf_event__process_exit;
1043 trace->tool.fork = perf_event__process_fork;
1044 trace->tool.attr = perf_event__process_attr;
1045 trace->tool.tracing_data = perf_event__process_tracing_data;
1046 trace->tool.build_id = perf_event__process_build_id;
1047
1048 trace->tool.ordered_samples = true;
1049 trace->tool.ordering_requires_timestamps = true;
1050
1051 /* add tid to output */
1052 trace->multiple_threads = true;
1053
1054 if (symbol__init() < 0)
1055 return -1;
1056
1057 session = perf_session__new(input_name, O_RDONLY, 0, false,
1058 &trace->tool);
1059 if (session == NULL)
1060 return -ENOMEM;
1061
1062 err = perf_session__set_tracepoints_handlers(session, handlers);
1063 if (err)
1064 goto out;
1065
1066 if (!perf_session__has_tp(session, "raw_syscalls:sys_enter")) {
1067 pr_err("Data file does not have raw_syscalls:sys_enter events\n");
1068 goto out;
1069 }
1070
1071 if (!perf_session__has_tp(session, "raw_syscalls:sys_exit")) {
1072 pr_err("Data file does not have raw_syscalls:sys_exit events\n");
1073 goto out;
1074 }
1075
1076 err = parse_target_str(trace);
1077 if (err != 0)
1078 goto out;
1079
1080 setup_pager();
1081
1082 err = perf_session__process_events(session, &trace->tool);
1083 if (err)
1084 pr_err("Failed to process events, error %d", err);
1085
1086 out:
1087 perf_session__delete(session);
1088
1089 return err;
1090 }
1091
1092 static size_t trace__fprintf_threads_header(FILE *fp)
1093 {
1094 size_t printed;
1095
1096 printed = fprintf(fp, "\n _____________________________________________________________________\n");
1097 printed += fprintf(fp," __) Summary of events (__\n\n");
1098 printed += fprintf(fp," [ task - pid ] [ events ] [ ratio ] [ runtime ]\n");
1099 printed += fprintf(fp," _____________________________________________________________________\n\n");
1100
1101 return printed;
1102 }
1103
1104 static size_t trace__fprintf_thread_summary(struct trace *trace, FILE *fp)
1105 {
1106 size_t printed = trace__fprintf_threads_header(fp);
1107 struct rb_node *nd;
1108
1109 for (nd = rb_first(&trace->host.threads); nd; nd = rb_next(nd)) {
1110 struct thread *thread = rb_entry(nd, struct thread, rb_node);
1111 struct thread_trace *ttrace = thread->priv;
1112 const char *color;
1113 double ratio;
1114
1115 if (ttrace == NULL)
1116 continue;
1117
1118 ratio = (double)ttrace->nr_events / trace->nr_events * 100.0;
1119
1120 color = PERF_COLOR_NORMAL;
1121 if (ratio > 50.0)
1122 color = PERF_COLOR_RED;
1123 else if (ratio > 25.0)
1124 color = PERF_COLOR_GREEN;
1125 else if (ratio > 5.0)
1126 color = PERF_COLOR_YELLOW;
1127
1128 printed += color_fprintf(fp, color, "%20s", thread->comm);
1129 printed += fprintf(fp, " - %-5d :%11lu [", thread->tid, ttrace->nr_events);
1130 printed += color_fprintf(fp, color, "%5.1f%%", ratio);
1131 printed += fprintf(fp, " ] %10.3f ms\n", ttrace->runtime_ms);
1132 }
1133
1134 return printed;
1135 }
1136
1137 static int trace__set_duration(const struct option *opt, const char *str,
1138 int unset __maybe_unused)
1139 {
1140 struct trace *trace = opt->value;
1141
1142 trace->duration_filter = atof(str);
1143 return 0;
1144 }
1145
1146 static int trace__open_output(struct trace *trace, const char *filename)
1147 {
1148 struct stat st;
1149
1150 if (!stat(filename, &st) && st.st_size) {
1151 char oldname[PATH_MAX];
1152
1153 scnprintf(oldname, sizeof(oldname), "%s.old", filename);
1154 unlink(oldname);
1155 rename(filename, oldname);
1156 }
1157
1158 trace->output = fopen(filename, "w");
1159
1160 return trace->output == NULL ? -errno : 0;
1161 }
1162
1163 int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
1164 {
1165 const char * const trace_usage[] = {
1166 "perf trace [<options>] [<command>]",
1167 "perf trace [<options>] -- <command> [<options>]",
1168 NULL
1169 };
1170 struct trace trace = {
1171 .audit_machine = audit_detect_machine(),
1172 .syscalls = {
1173 . max = -1,
1174 },
1175 .opts = {
1176 .target = {
1177 .uid = UINT_MAX,
1178 .uses_mmap = true,
1179 },
1180 .user_freq = UINT_MAX,
1181 .user_interval = ULLONG_MAX,
1182 .no_delay = true,
1183 .mmap_pages = 1024,
1184 },
1185 .output = stdout,
1186 };
1187 const char *output_name = NULL;
1188 const char *ev_qualifier_str = NULL;
1189 const struct option trace_options[] = {
1190 OPT_STRING('e', "expr", &ev_qualifier_str, "expr",
1191 "list of events to trace"),
1192 OPT_STRING('o', "output", &output_name, "file", "output file name"),
1193 OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
1194 OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
1195 "trace events on existing process id"),
1196 OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
1197 "trace events on existing thread id"),
1198 OPT_BOOLEAN('a', "all-cpus", &trace.opts.target.system_wide,
1199 "system-wide collection from all CPUs"),
1200 OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
1201 "list of cpus to monitor"),
1202 OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
1203 "child tasks do not inherit counters"),
1204 OPT_UINTEGER('m', "mmap-pages", &trace.opts.mmap_pages,
1205 "number of mmap data pages"),
1206 OPT_STRING('u', "uid", &trace.opts.target.uid_str, "user",
1207 "user to profile"),
1208 OPT_CALLBACK(0, "duration", &trace, "float",
1209 "show only events with duration > N.M ms",
1210 trace__set_duration),
1211 OPT_BOOLEAN(0, "sched", &trace.sched, "show blocking scheduler events"),
1212 OPT_INCR('v', "verbose", &verbose, "be more verbose"),
1213 OPT_END()
1214 };
1215 int err;
1216 char bf[BUFSIZ];
1217
1218 argc = parse_options(argc, argv, trace_options, trace_usage, 0);
1219
1220 if (output_name != NULL) {
1221 err = trace__open_output(&trace, output_name);
1222 if (err < 0) {
1223 perror("failed to create output file");
1224 goto out;
1225 }
1226 }
1227
1228 if (ev_qualifier_str != NULL) {
1229 const char *s = ev_qualifier_str;
1230
1231 trace.not_ev_qualifier = *s == '!';
1232 if (trace.not_ev_qualifier)
1233 ++s;
1234 trace.ev_qualifier = strlist__new(true, s);
1235 if (trace.ev_qualifier == NULL) {
1236 fputs("Not enough memory to parse event qualifier",
1237 trace.output);
1238 err = -ENOMEM;
1239 goto out_close;
1240 }
1241 }
1242
1243 err = perf_target__validate(&trace.opts.target);
1244 if (err) {
1245 perf_target__strerror(&trace.opts.target, err, bf, sizeof(bf));
1246 fprintf(trace.output, "%s", bf);
1247 goto out_close;
1248 }
1249
1250 err = perf_target__parse_uid(&trace.opts.target);
1251 if (err) {
1252 perf_target__strerror(&trace.opts.target, err, bf, sizeof(bf));
1253 fprintf(trace.output, "%s", bf);
1254 goto out_close;
1255 }
1256
1257 if (!argc && perf_target__none(&trace.opts.target))
1258 trace.opts.target.system_wide = true;
1259
1260 if (input_name)
1261 err = trace__replay(&trace);
1262 else
1263 err = trace__run(&trace, argc, argv);
1264
1265 if (trace.sched && !err)
1266 trace__fprintf_thread_summary(&trace, trace.output);
1267
1268 out_close:
1269 if (output_name != NULL)
1270 fclose(trace.output);
1271 out:
1272 return err;
1273 }
Linux with added FPC-III support