Merge drm/drm-fixes into drm-misc-fixes
[platform/kernel/linux-rpi.git] / tools / perf / util / header.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <inttypes.h>
4 #include "string2.h"
5 #include <sys/param.h>
6 #include <sys/types.h>
7 #include <byteswap.h>
8 #include <unistd.h>
9 #include <regex.h>
10 #include <stdio.h>
11 #include <stdlib.h>
12 #include <linux/compiler.h>
13 #include <linux/list.h>
14 #include <linux/kernel.h>
15 #include <linux/bitops.h>
16 #include <linux/string.h>
17 #include <linux/stringify.h>
18 #include <linux/zalloc.h>
19 #include <sys/stat.h>
20 #include <sys/utsname.h>
21 #include <linux/time64.h>
22 #include <dirent.h>
23 #ifdef HAVE_LIBBPF_SUPPORT
24 #include <bpf/libbpf.h>
25 #endif
26 #include <perf/cpumap.h>
27 #include <tools/libc_compat.h> // reallocarray
28
29 #include "dso.h"
30 #include "evlist.h"
31 #include "evsel.h"
32 #include "util/evsel_fprintf.h"
33 #include "header.h"
34 #include "memswap.h"
35 #include "trace-event.h"
36 #include "session.h"
37 #include "symbol.h"
38 #include "debug.h"
39 #include "cpumap.h"
40 #include "pmu.h"
41 #include "pmus.h"
42 #include "vdso.h"
43 #include "strbuf.h"
44 #include "build-id.h"
45 #include "data.h"
46 #include <api/fs/fs.h>
47 #include "asm/bug.h"
48 #include "tool.h"
49 #include "time-utils.h"
50 #include "units.h"
51 #include "util/util.h" // perf_exe()
52 #include "cputopo.h"
53 #include "bpf-event.h"
54 #include "bpf-utils.h"
55 #include "clockid.h"
56
57 #include <linux/ctype.h>
58 #include <internal/lib.h>
59
60 #ifdef HAVE_LIBTRACEEVENT
61 #include <traceevent/event-parse.h>
62 #endif
63
64 /*
65  * magic2 = "PERFILE2"
66  * must be a numerical value to let the endianness
67  * determine the memory layout. That way we are able
68  * to detect endianness when reading the perf.data file
69  * back.
70  *
71  * we check for legacy (PERFFILE) format.
72  */
73 static const char *__perf_magic1 = "PERFFILE";
74 static const u64 __perf_magic2    = 0x32454c4946524550ULL;
75 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
76
77 #define PERF_MAGIC      __perf_magic2
78
79 const char perf_version_string[] = PERF_VERSION;
80
81 struct perf_file_attr {
82         struct perf_event_attr  attr;
83         struct perf_file_section        ids;
84 };
85
86 void perf_header__set_feat(struct perf_header *header, int feat)
87 {
88         __set_bit(feat, header->adds_features);
89 }
90
91 void perf_header__clear_feat(struct perf_header *header, int feat)
92 {
93         __clear_bit(feat, header->adds_features);
94 }
95
96 bool perf_header__has_feat(const struct perf_header *header, int feat)
97 {
98         return test_bit(feat, header->adds_features);
99 }
100
101 static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
102 {
103         ssize_t ret = writen(ff->fd, buf, size);
104
105         if (ret != (ssize_t)size)
106                 return ret < 0 ? (int)ret : -1;
107         return 0;
108 }
109
110 static int __do_write_buf(struct feat_fd *ff,  const void *buf, size_t size)
111 {
112         /* struct perf_event_header::size is u16 */
113         const size_t max_size = 0xffff - sizeof(struct perf_event_header);
114         size_t new_size = ff->size;
115         void *addr;
116
117         if (size + ff->offset > max_size)
118                 return -E2BIG;
119
120         while (size > (new_size - ff->offset))
121                 new_size <<= 1;
122         new_size = min(max_size, new_size);
123
124         if (ff->size < new_size) {
125                 addr = realloc(ff->buf, new_size);
126                 if (!addr)
127                         return -ENOMEM;
128                 ff->buf = addr;
129                 ff->size = new_size;
130         }
131
132         memcpy(ff->buf + ff->offset, buf, size);
133         ff->offset += size;
134
135         return 0;
136 }
137
138 /* Return: 0 if succeeded, -ERR if failed. */
139 int do_write(struct feat_fd *ff, const void *buf, size_t size)
140 {
141         if (!ff->buf)
142                 return __do_write_fd(ff, buf, size);
143         return __do_write_buf(ff, buf, size);
144 }
145
146 /* Return: 0 if succeeded, -ERR if failed. */
147 static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
148 {
149         u64 *p = (u64 *) set;
150         int i, ret;
151
152         ret = do_write(ff, &size, sizeof(size));
153         if (ret < 0)
154                 return ret;
155
156         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
157                 ret = do_write(ff, p + i, sizeof(*p));
158                 if (ret < 0)
159                         return ret;
160         }
161
162         return 0;
163 }
164
165 /* Return: 0 if succeeded, -ERR if failed. */
166 int write_padded(struct feat_fd *ff, const void *bf,
167                  size_t count, size_t count_aligned)
168 {
169         static const char zero_buf[NAME_ALIGN];
170         int err = do_write(ff, bf, count);
171
172         if (!err)
173                 err = do_write(ff, zero_buf, count_aligned - count);
174
175         return err;
176 }
177
178 #define string_size(str)                                                \
179         (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
180
181 /* Return: 0 if succeeded, -ERR if failed. */
182 static int do_write_string(struct feat_fd *ff, const char *str)
183 {
184         u32 len, olen;
185         int ret;
186
187         olen = strlen(str) + 1;
188         len = PERF_ALIGN(olen, NAME_ALIGN);
189
190         /* write len, incl. \0 */
191         ret = do_write(ff, &len, sizeof(len));
192         if (ret < 0)
193                 return ret;
194
195         return write_padded(ff, str, olen, len);
196 }
197
198 static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
199 {
200         ssize_t ret = readn(ff->fd, addr, size);
201
202         if (ret != size)
203                 return ret < 0 ? (int)ret : -1;
204         return 0;
205 }
206
207 static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
208 {
209         if (size > (ssize_t)ff->size - ff->offset)
210                 return -1;
211
212         memcpy(addr, ff->buf + ff->offset, size);
213         ff->offset += size;
214
215         return 0;
216
217 }
218
219 static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
220 {
221         if (!ff->buf)
222                 return __do_read_fd(ff, addr, size);
223         return __do_read_buf(ff, addr, size);
224 }
225
226 static int do_read_u32(struct feat_fd *ff, u32 *addr)
227 {
228         int ret;
229
230         ret = __do_read(ff, addr, sizeof(*addr));
231         if (ret)
232                 return ret;
233
234         if (ff->ph->needs_swap)
235                 *addr = bswap_32(*addr);
236         return 0;
237 }
238
239 static int do_read_u64(struct feat_fd *ff, u64 *addr)
240 {
241         int ret;
242
243         ret = __do_read(ff, addr, sizeof(*addr));
244         if (ret)
245                 return ret;
246
247         if (ff->ph->needs_swap)
248                 *addr = bswap_64(*addr);
249         return 0;
250 }
251
252 static char *do_read_string(struct feat_fd *ff)
253 {
254         u32 len;
255         char *buf;
256
257         if (do_read_u32(ff, &len))
258                 return NULL;
259
260         buf = malloc(len);
261         if (!buf)
262                 return NULL;
263
264         if (!__do_read(ff, buf, len)) {
265                 /*
266                  * strings are padded by zeroes
267                  * thus the actual strlen of buf
268                  * may be less than len
269                  */
270                 return buf;
271         }
272
273         free(buf);
274         return NULL;
275 }
276
277 /* Return: 0 if succeeded, -ERR if failed. */
278 static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
279 {
280         unsigned long *set;
281         u64 size, *p;
282         int i, ret;
283
284         ret = do_read_u64(ff, &size);
285         if (ret)
286                 return ret;
287
288         set = bitmap_zalloc(size);
289         if (!set)
290                 return -ENOMEM;
291
292         p = (u64 *) set;
293
294         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
295                 ret = do_read_u64(ff, p + i);
296                 if (ret < 0) {
297                         free(set);
298                         return ret;
299                 }
300         }
301
302         *pset  = set;
303         *psize = size;
304         return 0;
305 }
306
307 #ifdef HAVE_LIBTRACEEVENT
308 static int write_tracing_data(struct feat_fd *ff,
309                               struct evlist *evlist)
310 {
311         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
312                 return -1;
313
314         return read_tracing_data(ff->fd, &evlist->core.entries);
315 }
316 #endif
317
318 static int write_build_id(struct feat_fd *ff,
319                           struct evlist *evlist __maybe_unused)
320 {
321         struct perf_session *session;
322         int err;
323
324         session = container_of(ff->ph, struct perf_session, header);
325
326         if (!perf_session__read_build_ids(session, true))
327                 return -1;
328
329         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
330                 return -1;
331
332         err = perf_session__write_buildid_table(session, ff);
333         if (err < 0) {
334                 pr_debug("failed to write buildid table\n");
335                 return err;
336         }
337         perf_session__cache_build_ids(session);
338
339         return 0;
340 }
341
342 static int write_hostname(struct feat_fd *ff,
343                           struct evlist *evlist __maybe_unused)
344 {
345         struct utsname uts;
346         int ret;
347
348         ret = uname(&uts);
349         if (ret < 0)
350                 return -1;
351
352         return do_write_string(ff, uts.nodename);
353 }
354
355 static int write_osrelease(struct feat_fd *ff,
356                            struct evlist *evlist __maybe_unused)
357 {
358         struct utsname uts;
359         int ret;
360
361         ret = uname(&uts);
362         if (ret < 0)
363                 return -1;
364
365         return do_write_string(ff, uts.release);
366 }
367
368 static int write_arch(struct feat_fd *ff,
369                       struct evlist *evlist __maybe_unused)
370 {
371         struct utsname uts;
372         int ret;
373
374         ret = uname(&uts);
375         if (ret < 0)
376                 return -1;
377
378         return do_write_string(ff, uts.machine);
379 }
380
381 static int write_version(struct feat_fd *ff,
382                          struct evlist *evlist __maybe_unused)
383 {
384         return do_write_string(ff, perf_version_string);
385 }
386
387 static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
388 {
389         FILE *file;
390         char *buf = NULL;
391         char *s, *p;
392         const char *search = cpuinfo_proc;
393         size_t len = 0;
394         int ret = -1;
395
396         if (!search)
397                 return -1;
398
399         file = fopen("/proc/cpuinfo", "r");
400         if (!file)
401                 return -1;
402
403         while (getline(&buf, &len, file) > 0) {
404                 ret = strncmp(buf, search, strlen(search));
405                 if (!ret)
406                         break;
407         }
408
409         if (ret) {
410                 ret = -1;
411                 goto done;
412         }
413
414         s = buf;
415
416         p = strchr(buf, ':');
417         if (p && *(p+1) == ' ' && *(p+2))
418                 s = p + 2;
419         p = strchr(s, '\n');
420         if (p)
421                 *p = '\0';
422
423         /* squash extra space characters (branding string) */
424         p = s;
425         while (*p) {
426                 if (isspace(*p)) {
427                         char *r = p + 1;
428                         char *q = skip_spaces(r);
429                         *p = ' ';
430                         if (q != (p+1))
431                                 while ((*r++ = *q++));
432                 }
433                 p++;
434         }
435         ret = do_write_string(ff, s);
436 done:
437         free(buf);
438         fclose(file);
439         return ret;
440 }
441
442 static int write_cpudesc(struct feat_fd *ff,
443                        struct evlist *evlist __maybe_unused)
444 {
445 #if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
446 #define CPUINFO_PROC    { "cpu", }
447 #elif defined(__s390__)
448 #define CPUINFO_PROC    { "vendor_id", }
449 #elif defined(__sh__)
450 #define CPUINFO_PROC    { "cpu type", }
451 #elif defined(__alpha__) || defined(__mips__)
452 #define CPUINFO_PROC    { "cpu model", }
453 #elif defined(__arm__)
454 #define CPUINFO_PROC    { "model name", "Processor", }
455 #elif defined(__arc__)
456 #define CPUINFO_PROC    { "Processor", }
457 #elif defined(__xtensa__)
458 #define CPUINFO_PROC    { "core ID", }
459 #elif defined(__loongarch__)
460 #define CPUINFO_PROC    { "Model Name", }
461 #else
462 #define CPUINFO_PROC    { "model name", }
463 #endif
464         const char *cpuinfo_procs[] = CPUINFO_PROC;
465 #undef CPUINFO_PROC
466         unsigned int i;
467
468         for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
469                 int ret;
470                 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
471                 if (ret >= 0)
472                         return ret;
473         }
474         return -1;
475 }
476
477
478 static int write_nrcpus(struct feat_fd *ff,
479                         struct evlist *evlist __maybe_unused)
480 {
481         long nr;
482         u32 nrc, nra;
483         int ret;
484
485         nrc = cpu__max_present_cpu().cpu;
486
487         nr = sysconf(_SC_NPROCESSORS_ONLN);
488         if (nr < 0)
489                 return -1;
490
491         nra = (u32)(nr & UINT_MAX);
492
493         ret = do_write(ff, &nrc, sizeof(nrc));
494         if (ret < 0)
495                 return ret;
496
497         return do_write(ff, &nra, sizeof(nra));
498 }
499
500 static int write_event_desc(struct feat_fd *ff,
501                             struct evlist *evlist)
502 {
503         struct evsel *evsel;
504         u32 nre, nri, sz;
505         int ret;
506
507         nre = evlist->core.nr_entries;
508
509         /*
510          * write number of events
511          */
512         ret = do_write(ff, &nre, sizeof(nre));
513         if (ret < 0)
514                 return ret;
515
516         /*
517          * size of perf_event_attr struct
518          */
519         sz = (u32)sizeof(evsel->core.attr);
520         ret = do_write(ff, &sz, sizeof(sz));
521         if (ret < 0)
522                 return ret;
523
524         evlist__for_each_entry(evlist, evsel) {
525                 ret = do_write(ff, &evsel->core.attr, sz);
526                 if (ret < 0)
527                         return ret;
528                 /*
529                  * write number of unique id per event
530                  * there is one id per instance of an event
531                  *
532                  * copy into an nri to be independent of the
533                  * type of ids,
534                  */
535                 nri = evsel->core.ids;
536                 ret = do_write(ff, &nri, sizeof(nri));
537                 if (ret < 0)
538                         return ret;
539
540                 /*
541                  * write event string as passed on cmdline
542                  */
543                 ret = do_write_string(ff, evsel__name(evsel));
544                 if (ret < 0)
545                         return ret;
546                 /*
547                  * write unique ids for this event
548                  */
549                 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
550                 if (ret < 0)
551                         return ret;
552         }
553         return 0;
554 }
555
556 static int write_cmdline(struct feat_fd *ff,
557                          struct evlist *evlist __maybe_unused)
558 {
559         char pbuf[MAXPATHLEN], *buf;
560         int i, ret, n;
561
562         /* actual path to perf binary */
563         buf = perf_exe(pbuf, MAXPATHLEN);
564
565         /* account for binary path */
566         n = perf_env.nr_cmdline + 1;
567
568         ret = do_write(ff, &n, sizeof(n));
569         if (ret < 0)
570                 return ret;
571
572         ret = do_write_string(ff, buf);
573         if (ret < 0)
574                 return ret;
575
576         for (i = 0 ; i < perf_env.nr_cmdline; i++) {
577                 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
578                 if (ret < 0)
579                         return ret;
580         }
581         return 0;
582 }
583
584
585 static int write_cpu_topology(struct feat_fd *ff,
586                               struct evlist *evlist __maybe_unused)
587 {
588         struct cpu_topology *tp;
589         u32 i;
590         int ret, j;
591
592         tp = cpu_topology__new();
593         if (!tp)
594                 return -1;
595
596         ret = do_write(ff, &tp->package_cpus_lists, sizeof(tp->package_cpus_lists));
597         if (ret < 0)
598                 goto done;
599
600         for (i = 0; i < tp->package_cpus_lists; i++) {
601                 ret = do_write_string(ff, tp->package_cpus_list[i]);
602                 if (ret < 0)
603                         goto done;
604         }
605         ret = do_write(ff, &tp->core_cpus_lists, sizeof(tp->core_cpus_lists));
606         if (ret < 0)
607                 goto done;
608
609         for (i = 0; i < tp->core_cpus_lists; i++) {
610                 ret = do_write_string(ff, tp->core_cpus_list[i]);
611                 if (ret < 0)
612                         break;
613         }
614
615         ret = perf_env__read_cpu_topology_map(&perf_env);
616         if (ret < 0)
617                 goto done;
618
619         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
620                 ret = do_write(ff, &perf_env.cpu[j].core_id,
621                                sizeof(perf_env.cpu[j].core_id));
622                 if (ret < 0)
623                         return ret;
624                 ret = do_write(ff, &perf_env.cpu[j].socket_id,
625                                sizeof(perf_env.cpu[j].socket_id));
626                 if (ret < 0)
627                         return ret;
628         }
629
630         if (!tp->die_cpus_lists)
631                 goto done;
632
633         ret = do_write(ff, &tp->die_cpus_lists, sizeof(tp->die_cpus_lists));
634         if (ret < 0)
635                 goto done;
636
637         for (i = 0; i < tp->die_cpus_lists; i++) {
638                 ret = do_write_string(ff, tp->die_cpus_list[i]);
639                 if (ret < 0)
640                         goto done;
641         }
642
643         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
644                 ret = do_write(ff, &perf_env.cpu[j].die_id,
645                                sizeof(perf_env.cpu[j].die_id));
646                 if (ret < 0)
647                         return ret;
648         }
649
650 done:
651         cpu_topology__delete(tp);
652         return ret;
653 }
654
655
656
657 static int write_total_mem(struct feat_fd *ff,
658                            struct evlist *evlist __maybe_unused)
659 {
660         char *buf = NULL;
661         FILE *fp;
662         size_t len = 0;
663         int ret = -1, n;
664         uint64_t mem;
665
666         fp = fopen("/proc/meminfo", "r");
667         if (!fp)
668                 return -1;
669
670         while (getline(&buf, &len, fp) > 0) {
671                 ret = strncmp(buf, "MemTotal:", 9);
672                 if (!ret)
673                         break;
674         }
675         if (!ret) {
676                 n = sscanf(buf, "%*s %"PRIu64, &mem);
677                 if (n == 1)
678                         ret = do_write(ff, &mem, sizeof(mem));
679         } else
680                 ret = -1;
681         free(buf);
682         fclose(fp);
683         return ret;
684 }
685
686 static int write_numa_topology(struct feat_fd *ff,
687                                struct evlist *evlist __maybe_unused)
688 {
689         struct numa_topology *tp;
690         int ret = -1;
691         u32 i;
692
693         tp = numa_topology__new();
694         if (!tp)
695                 return -ENOMEM;
696
697         ret = do_write(ff, &tp->nr, sizeof(u32));
698         if (ret < 0)
699                 goto err;
700
701         for (i = 0; i < tp->nr; i++) {
702                 struct numa_topology_node *n = &tp->nodes[i];
703
704                 ret = do_write(ff, &n->node, sizeof(u32));
705                 if (ret < 0)
706                         goto err;
707
708                 ret = do_write(ff, &n->mem_total, sizeof(u64));
709                 if (ret)
710                         goto err;
711
712                 ret = do_write(ff, &n->mem_free, sizeof(u64));
713                 if (ret)
714                         goto err;
715
716                 ret = do_write_string(ff, n->cpus);
717                 if (ret < 0)
718                         goto err;
719         }
720
721         ret = 0;
722
723 err:
724         numa_topology__delete(tp);
725         return ret;
726 }
727
728 /*
729  * File format:
730  *
731  * struct pmu_mappings {
732  *      u32     pmu_num;
733  *      struct pmu_map {
734  *              u32     type;
735  *              char    name[];
736  *      }[pmu_num];
737  * };
738  */
739
740 static int write_pmu_mappings(struct feat_fd *ff,
741                               struct evlist *evlist __maybe_unused)
742 {
743         struct perf_pmu *pmu = NULL;
744         u32 pmu_num = 0;
745         int ret;
746
747         /*
748          * Do a first pass to count number of pmu to avoid lseek so this
749          * works in pipe mode as well.
750          */
751         while ((pmu = perf_pmus__scan(pmu)))
752                 pmu_num++;
753
754         ret = do_write(ff, &pmu_num, sizeof(pmu_num));
755         if (ret < 0)
756                 return ret;
757
758         while ((pmu = perf_pmus__scan(pmu))) {
759                 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
760                 if (ret < 0)
761                         return ret;
762
763                 ret = do_write_string(ff, pmu->name);
764                 if (ret < 0)
765                         return ret;
766         }
767
768         return 0;
769 }
770
771 /*
772  * File format:
773  *
774  * struct group_descs {
775  *      u32     nr_groups;
776  *      struct group_desc {
777  *              char    name[];
778  *              u32     leader_idx;
779  *              u32     nr_members;
780  *      }[nr_groups];
781  * };
782  */
783 static int write_group_desc(struct feat_fd *ff,
784                             struct evlist *evlist)
785 {
786         u32 nr_groups = evlist__nr_groups(evlist);
787         struct evsel *evsel;
788         int ret;
789
790         ret = do_write(ff, &nr_groups, sizeof(nr_groups));
791         if (ret < 0)
792                 return ret;
793
794         evlist__for_each_entry(evlist, evsel) {
795                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
796                         const char *name = evsel->group_name ?: "{anon_group}";
797                         u32 leader_idx = evsel->core.idx;
798                         u32 nr_members = evsel->core.nr_members;
799
800                         ret = do_write_string(ff, name);
801                         if (ret < 0)
802                                 return ret;
803
804                         ret = do_write(ff, &leader_idx, sizeof(leader_idx));
805                         if (ret < 0)
806                                 return ret;
807
808                         ret = do_write(ff, &nr_members, sizeof(nr_members));
809                         if (ret < 0)
810                                 return ret;
811                 }
812         }
813         return 0;
814 }
815
816 /*
817  * Return the CPU id as a raw string.
818  *
819  * Each architecture should provide a more precise id string that
820  * can be use to match the architecture's "mapfile".
821  */
822 char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
823 {
824         return NULL;
825 }
826
827 /* Return zero when the cpuid from the mapfile.csv matches the
828  * cpuid string generated on this platform.
829  * Otherwise return non-zero.
830  */
831 int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
832 {
833         regex_t re;
834         regmatch_t pmatch[1];
835         int match;
836
837         if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
838                 /* Warn unable to generate match particular string. */
839                 pr_info("Invalid regular expression %s\n", mapcpuid);
840                 return 1;
841         }
842
843         match = !regexec(&re, cpuid, 1, pmatch, 0);
844         regfree(&re);
845         if (match) {
846                 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
847
848                 /* Verify the entire string matched. */
849                 if (match_len == strlen(cpuid))
850                         return 0;
851         }
852         return 1;
853 }
854
855 /*
856  * default get_cpuid(): nothing gets recorded
857  * actual implementation must be in arch/$(SRCARCH)/util/header.c
858  */
859 int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
860 {
861         return ENOSYS; /* Not implemented */
862 }
863
864 static int write_cpuid(struct feat_fd *ff,
865                        struct evlist *evlist __maybe_unused)
866 {
867         char buffer[64];
868         int ret;
869
870         ret = get_cpuid(buffer, sizeof(buffer));
871         if (ret)
872                 return -1;
873
874         return do_write_string(ff, buffer);
875 }
876
877 static int write_branch_stack(struct feat_fd *ff __maybe_unused,
878                               struct evlist *evlist __maybe_unused)
879 {
880         return 0;
881 }
882
883 static int write_auxtrace(struct feat_fd *ff,
884                           struct evlist *evlist __maybe_unused)
885 {
886         struct perf_session *session;
887         int err;
888
889         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
890                 return -1;
891
892         session = container_of(ff->ph, struct perf_session, header);
893
894         err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
895         if (err < 0)
896                 pr_err("Failed to write auxtrace index\n");
897         return err;
898 }
899
900 static int write_clockid(struct feat_fd *ff,
901                          struct evlist *evlist __maybe_unused)
902 {
903         return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
904                         sizeof(ff->ph->env.clock.clockid_res_ns));
905 }
906
907 static int write_clock_data(struct feat_fd *ff,
908                             struct evlist *evlist __maybe_unused)
909 {
910         u64 *data64;
911         u32 data32;
912         int ret;
913
914         /* version */
915         data32 = 1;
916
917         ret = do_write(ff, &data32, sizeof(data32));
918         if (ret < 0)
919                 return ret;
920
921         /* clockid */
922         data32 = ff->ph->env.clock.clockid;
923
924         ret = do_write(ff, &data32, sizeof(data32));
925         if (ret < 0)
926                 return ret;
927
928         /* TOD ref time */
929         data64 = &ff->ph->env.clock.tod_ns;
930
931         ret = do_write(ff, data64, sizeof(*data64));
932         if (ret < 0)
933                 return ret;
934
935         /* clockid ref time */
936         data64 = &ff->ph->env.clock.clockid_ns;
937
938         return do_write(ff, data64, sizeof(*data64));
939 }
940
941 static int write_hybrid_topology(struct feat_fd *ff,
942                                  struct evlist *evlist __maybe_unused)
943 {
944         struct hybrid_topology *tp;
945         int ret;
946         u32 i;
947
948         tp = hybrid_topology__new();
949         if (!tp)
950                 return -ENOENT;
951
952         ret = do_write(ff, &tp->nr, sizeof(u32));
953         if (ret < 0)
954                 goto err;
955
956         for (i = 0; i < tp->nr; i++) {
957                 struct hybrid_topology_node *n = &tp->nodes[i];
958
959                 ret = do_write_string(ff, n->pmu_name);
960                 if (ret < 0)
961                         goto err;
962
963                 ret = do_write_string(ff, n->cpus);
964                 if (ret < 0)
965                         goto err;
966         }
967
968         ret = 0;
969
970 err:
971         hybrid_topology__delete(tp);
972         return ret;
973 }
974
975 static int write_dir_format(struct feat_fd *ff,
976                             struct evlist *evlist __maybe_unused)
977 {
978         struct perf_session *session;
979         struct perf_data *data;
980
981         session = container_of(ff->ph, struct perf_session, header);
982         data = session->data;
983
984         if (WARN_ON(!perf_data__is_dir(data)))
985                 return -1;
986
987         return do_write(ff, &data->dir.version, sizeof(data->dir.version));
988 }
989
990 /*
991  * Check whether a CPU is online
992  *
993  * Returns:
994  *     1 -> if CPU is online
995  *     0 -> if CPU is offline
996  *    -1 -> error case
997  */
998 int is_cpu_online(unsigned int cpu)
999 {
1000         char *str;
1001         size_t strlen;
1002         char buf[256];
1003         int status = -1;
1004         struct stat statbuf;
1005
1006         snprintf(buf, sizeof(buf),
1007                 "/sys/devices/system/cpu/cpu%d", cpu);
1008         if (stat(buf, &statbuf) != 0)
1009                 return 0;
1010
1011         /*
1012          * Check if /sys/devices/system/cpu/cpux/online file
1013          * exists. Some cases cpu0 won't have online file since
1014          * it is not expected to be turned off generally.
1015          * In kernels without CONFIG_HOTPLUG_CPU, this
1016          * file won't exist
1017          */
1018         snprintf(buf, sizeof(buf),
1019                 "/sys/devices/system/cpu/cpu%d/online", cpu);
1020         if (stat(buf, &statbuf) != 0)
1021                 return 1;
1022
1023         /*
1024          * Read online file using sysfs__read_str.
1025          * If read or open fails, return -1.
1026          * If read succeeds, return value from file
1027          * which gets stored in "str"
1028          */
1029         snprintf(buf, sizeof(buf),
1030                 "devices/system/cpu/cpu%d/online", cpu);
1031
1032         if (sysfs__read_str(buf, &str, &strlen) < 0)
1033                 return status;
1034
1035         status = atoi(str);
1036
1037         free(str);
1038         return status;
1039 }
1040
1041 #ifdef HAVE_LIBBPF_SUPPORT
1042 static int write_bpf_prog_info(struct feat_fd *ff,
1043                                struct evlist *evlist __maybe_unused)
1044 {
1045         struct perf_env *env = &ff->ph->env;
1046         struct rb_root *root;
1047         struct rb_node *next;
1048         int ret;
1049
1050         down_read(&env->bpf_progs.lock);
1051
1052         ret = do_write(ff, &env->bpf_progs.infos_cnt,
1053                        sizeof(env->bpf_progs.infos_cnt));
1054         if (ret < 0)
1055                 goto out;
1056
1057         root = &env->bpf_progs.infos;
1058         next = rb_first(root);
1059         while (next) {
1060                 struct bpf_prog_info_node *node;
1061                 size_t len;
1062
1063                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1064                 next = rb_next(&node->rb_node);
1065                 len = sizeof(struct perf_bpil) +
1066                         node->info_linear->data_len;
1067
1068                 /* before writing to file, translate address to offset */
1069                 bpil_addr_to_offs(node->info_linear);
1070                 ret = do_write(ff, node->info_linear, len);
1071                 /*
1072                  * translate back to address even when do_write() fails,
1073                  * so that this function never changes the data.
1074                  */
1075                 bpil_offs_to_addr(node->info_linear);
1076                 if (ret < 0)
1077                         goto out;
1078         }
1079 out:
1080         up_read(&env->bpf_progs.lock);
1081         return ret;
1082 }
1083
1084 static int write_bpf_btf(struct feat_fd *ff,
1085                          struct evlist *evlist __maybe_unused)
1086 {
1087         struct perf_env *env = &ff->ph->env;
1088         struct rb_root *root;
1089         struct rb_node *next;
1090         int ret;
1091
1092         down_read(&env->bpf_progs.lock);
1093
1094         ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1095                        sizeof(env->bpf_progs.btfs_cnt));
1096
1097         if (ret < 0)
1098                 goto out;
1099
1100         root = &env->bpf_progs.btfs;
1101         next = rb_first(root);
1102         while (next) {
1103                 struct btf_node *node;
1104
1105                 node = rb_entry(next, struct btf_node, rb_node);
1106                 next = rb_next(&node->rb_node);
1107                 ret = do_write(ff, &node->id,
1108                                sizeof(u32) * 2 + node->data_size);
1109                 if (ret < 0)
1110                         goto out;
1111         }
1112 out:
1113         up_read(&env->bpf_progs.lock);
1114         return ret;
1115 }
1116 #endif // HAVE_LIBBPF_SUPPORT
1117
1118 static int cpu_cache_level__sort(const void *a, const void *b)
1119 {
1120         struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1121         struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1122
1123         return cache_a->level - cache_b->level;
1124 }
1125
1126 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1127 {
1128         if (a->level != b->level)
1129                 return false;
1130
1131         if (a->line_size != b->line_size)
1132                 return false;
1133
1134         if (a->sets != b->sets)
1135                 return false;
1136
1137         if (a->ways != b->ways)
1138                 return false;
1139
1140         if (strcmp(a->type, b->type))
1141                 return false;
1142
1143         if (strcmp(a->size, b->size))
1144                 return false;
1145
1146         if (strcmp(a->map, b->map))
1147                 return false;
1148
1149         return true;
1150 }
1151
1152 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1153 {
1154         char path[PATH_MAX], file[PATH_MAX];
1155         struct stat st;
1156         size_t len;
1157
1158         scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1159         scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1160
1161         if (stat(file, &st))
1162                 return 1;
1163
1164         scnprintf(file, PATH_MAX, "%s/level", path);
1165         if (sysfs__read_int(file, (int *) &cache->level))
1166                 return -1;
1167
1168         scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1169         if (sysfs__read_int(file, (int *) &cache->line_size))
1170                 return -1;
1171
1172         scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1173         if (sysfs__read_int(file, (int *) &cache->sets))
1174                 return -1;
1175
1176         scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1177         if (sysfs__read_int(file, (int *) &cache->ways))
1178                 return -1;
1179
1180         scnprintf(file, PATH_MAX, "%s/type", path);
1181         if (sysfs__read_str(file, &cache->type, &len))
1182                 return -1;
1183
1184         cache->type[len] = 0;
1185         cache->type = strim(cache->type);
1186
1187         scnprintf(file, PATH_MAX, "%s/size", path);
1188         if (sysfs__read_str(file, &cache->size, &len)) {
1189                 zfree(&cache->type);
1190                 return -1;
1191         }
1192
1193         cache->size[len] = 0;
1194         cache->size = strim(cache->size);
1195
1196         scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1197         if (sysfs__read_str(file, &cache->map, &len)) {
1198                 zfree(&cache->size);
1199                 zfree(&cache->type);
1200                 return -1;
1201         }
1202
1203         cache->map[len] = 0;
1204         cache->map = strim(cache->map);
1205         return 0;
1206 }
1207
1208 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1209 {
1210         fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1211 }
1212
1213 /*
1214  * Build caches levels for a particular CPU from the data in
1215  * /sys/devices/system/cpu/cpu<cpu>/cache/
1216  * The cache level data is stored in caches[] from index at
1217  * *cntp.
1218  */
1219 int build_caches_for_cpu(u32 cpu, struct cpu_cache_level caches[], u32 *cntp)
1220 {
1221         u16 level;
1222
1223         for (level = 0; level < MAX_CACHE_LVL; level++) {
1224                 struct cpu_cache_level c;
1225                 int err;
1226                 u32 i;
1227
1228                 err = cpu_cache_level__read(&c, cpu, level);
1229                 if (err < 0)
1230                         return err;
1231
1232                 if (err == 1)
1233                         break;
1234
1235                 for (i = 0; i < *cntp; i++) {
1236                         if (cpu_cache_level__cmp(&c, &caches[i]))
1237                                 break;
1238                 }
1239
1240                 if (i == *cntp) {
1241                         caches[*cntp] = c;
1242                         *cntp = *cntp + 1;
1243                 } else
1244                         cpu_cache_level__free(&c);
1245         }
1246
1247         return 0;
1248 }
1249
1250 static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1251 {
1252         u32 nr, cpu, cnt = 0;
1253
1254         nr = cpu__max_cpu().cpu;
1255
1256         for (cpu = 0; cpu < nr; cpu++) {
1257                 int ret = build_caches_for_cpu(cpu, caches, &cnt);
1258
1259                 if (ret)
1260                         return ret;
1261         }
1262         *cntp = cnt;
1263         return 0;
1264 }
1265
1266 static int write_cache(struct feat_fd *ff,
1267                        struct evlist *evlist __maybe_unused)
1268 {
1269         u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL;
1270         struct cpu_cache_level caches[max_caches];
1271         u32 cnt = 0, i, version = 1;
1272         int ret;
1273
1274         ret = build_caches(caches, &cnt);
1275         if (ret)
1276                 goto out;
1277
1278         qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1279
1280         ret = do_write(ff, &version, sizeof(u32));
1281         if (ret < 0)
1282                 goto out;
1283
1284         ret = do_write(ff, &cnt, sizeof(u32));
1285         if (ret < 0)
1286                 goto out;
1287
1288         for (i = 0; i < cnt; i++) {
1289                 struct cpu_cache_level *c = &caches[i];
1290
1291                 #define _W(v)                                   \
1292                         ret = do_write(ff, &c->v, sizeof(u32)); \
1293                         if (ret < 0)                            \
1294                                 goto out;
1295
1296                 _W(level)
1297                 _W(line_size)
1298                 _W(sets)
1299                 _W(ways)
1300                 #undef _W
1301
1302                 #define _W(v)                                           \
1303                         ret = do_write_string(ff, (const char *) c->v); \
1304                         if (ret < 0)                                    \
1305                                 goto out;
1306
1307                 _W(type)
1308                 _W(size)
1309                 _W(map)
1310                 #undef _W
1311         }
1312
1313 out:
1314         for (i = 0; i < cnt; i++)
1315                 cpu_cache_level__free(&caches[i]);
1316         return ret;
1317 }
1318
1319 static int write_stat(struct feat_fd *ff __maybe_unused,
1320                       struct evlist *evlist __maybe_unused)
1321 {
1322         return 0;
1323 }
1324
1325 static int write_sample_time(struct feat_fd *ff,
1326                              struct evlist *evlist)
1327 {
1328         int ret;
1329
1330         ret = do_write(ff, &evlist->first_sample_time,
1331                        sizeof(evlist->first_sample_time));
1332         if (ret < 0)
1333                 return ret;
1334
1335         return do_write(ff, &evlist->last_sample_time,
1336                         sizeof(evlist->last_sample_time));
1337 }
1338
1339
1340 static int memory_node__read(struct memory_node *n, unsigned long idx)
1341 {
1342         unsigned int phys, size = 0;
1343         char path[PATH_MAX];
1344         struct dirent *ent;
1345         DIR *dir;
1346
1347 #define for_each_memory(mem, dir)                                       \
1348         while ((ent = readdir(dir)))                                    \
1349                 if (strcmp(ent->d_name, ".") &&                         \
1350                     strcmp(ent->d_name, "..") &&                        \
1351                     sscanf(ent->d_name, "memory%u", &mem) == 1)
1352
1353         scnprintf(path, PATH_MAX,
1354                   "%s/devices/system/node/node%lu",
1355                   sysfs__mountpoint(), idx);
1356
1357         dir = opendir(path);
1358         if (!dir) {
1359                 pr_warning("failed: can't open memory sysfs data\n");
1360                 return -1;
1361         }
1362
1363         for_each_memory(phys, dir) {
1364                 size = max(phys, size);
1365         }
1366
1367         size++;
1368
1369         n->set = bitmap_zalloc(size);
1370         if (!n->set) {
1371                 closedir(dir);
1372                 return -ENOMEM;
1373         }
1374
1375         n->node = idx;
1376         n->size = size;
1377
1378         rewinddir(dir);
1379
1380         for_each_memory(phys, dir) {
1381                 __set_bit(phys, n->set);
1382         }
1383
1384         closedir(dir);
1385         return 0;
1386 }
1387
1388 static void memory_node__delete_nodes(struct memory_node *nodesp, u64 cnt)
1389 {
1390         for (u64 i = 0; i < cnt; i++)
1391                 bitmap_free(nodesp[i].set);
1392
1393         free(nodesp);
1394 }
1395
1396 static int memory_node__sort(const void *a, const void *b)
1397 {
1398         const struct memory_node *na = a;
1399         const struct memory_node *nb = b;
1400
1401         return na->node - nb->node;
1402 }
1403
1404 static int build_mem_topology(struct memory_node **nodesp, u64 *cntp)
1405 {
1406         char path[PATH_MAX];
1407         struct dirent *ent;
1408         DIR *dir;
1409         int ret = 0;
1410         size_t cnt = 0, size = 0;
1411         struct memory_node *nodes = NULL;
1412
1413         scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1414                   sysfs__mountpoint());
1415
1416         dir = opendir(path);
1417         if (!dir) {
1418                 pr_debug2("%s: couldn't read %s, does this arch have topology information?\n",
1419                           __func__, path);
1420                 return -1;
1421         }
1422
1423         while (!ret && (ent = readdir(dir))) {
1424                 unsigned int idx;
1425                 int r;
1426
1427                 if (!strcmp(ent->d_name, ".") ||
1428                     !strcmp(ent->d_name, ".."))
1429                         continue;
1430
1431                 r = sscanf(ent->d_name, "node%u", &idx);
1432                 if (r != 1)
1433                         continue;
1434
1435                 if (cnt >= size) {
1436                         struct memory_node *new_nodes =
1437                                 reallocarray(nodes, cnt + 4, sizeof(*nodes));
1438
1439                         if (!new_nodes) {
1440                                 pr_err("Failed to write MEM_TOPOLOGY, size %zd nodes\n", size);
1441                                 ret = -ENOMEM;
1442                                 goto out;
1443                         }
1444                         nodes = new_nodes;
1445                         size += 4;
1446                 }
1447                 ret = memory_node__read(&nodes[cnt++], idx);
1448         }
1449 out:
1450         closedir(dir);
1451         if (!ret) {
1452                 *cntp = cnt;
1453                 *nodesp = nodes;
1454                 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1455         } else
1456                 memory_node__delete_nodes(nodes, cnt);
1457
1458         return ret;
1459 }
1460
1461 /*
1462  * The MEM_TOPOLOGY holds physical memory map for every
1463  * node in system. The format of data is as follows:
1464  *
1465  *  0 - version          | for future changes
1466  *  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1467  * 16 - count            | number of nodes
1468  *
1469  * For each node we store map of physical indexes for
1470  * each node:
1471  *
1472  * 32 - node id          | node index
1473  * 40 - size             | size of bitmap
1474  * 48 - bitmap           | bitmap of memory indexes that belongs to node
1475  */
1476 static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1477                               struct evlist *evlist __maybe_unused)
1478 {
1479         struct memory_node *nodes = NULL;
1480         u64 bsize, version = 1, i, nr = 0;
1481         int ret;
1482
1483         ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1484                               (unsigned long long *) &bsize);
1485         if (ret)
1486                 return ret;
1487
1488         ret = build_mem_topology(&nodes, &nr);
1489         if (ret)
1490                 return ret;
1491
1492         ret = do_write(ff, &version, sizeof(version));
1493         if (ret < 0)
1494                 goto out;
1495
1496         ret = do_write(ff, &bsize, sizeof(bsize));
1497         if (ret < 0)
1498                 goto out;
1499
1500         ret = do_write(ff, &nr, sizeof(nr));
1501         if (ret < 0)
1502                 goto out;
1503
1504         for (i = 0; i < nr; i++) {
1505                 struct memory_node *n = &nodes[i];
1506
1507                 #define _W(v)                                           \
1508                         ret = do_write(ff, &n->v, sizeof(n->v));        \
1509                         if (ret < 0)                                    \
1510                                 goto out;
1511
1512                 _W(node)
1513                 _W(size)
1514
1515                 #undef _W
1516
1517                 ret = do_write_bitmap(ff, n->set, n->size);
1518                 if (ret < 0)
1519                         goto out;
1520         }
1521
1522 out:
1523         memory_node__delete_nodes(nodes, nr);
1524         return ret;
1525 }
1526
1527 static int write_compressed(struct feat_fd *ff __maybe_unused,
1528                             struct evlist *evlist __maybe_unused)
1529 {
1530         int ret;
1531
1532         ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1533         if (ret)
1534                 return ret;
1535
1536         ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1537         if (ret)
1538                 return ret;
1539
1540         ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1541         if (ret)
1542                 return ret;
1543
1544         ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1545         if (ret)
1546                 return ret;
1547
1548         return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1549 }
1550
1551 static int __write_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1552                             bool write_pmu)
1553 {
1554         struct perf_pmu_caps *caps = NULL;
1555         int ret;
1556
1557         ret = do_write(ff, &pmu->nr_caps, sizeof(pmu->nr_caps));
1558         if (ret < 0)
1559                 return ret;
1560
1561         list_for_each_entry(caps, &pmu->caps, list) {
1562                 ret = do_write_string(ff, caps->name);
1563                 if (ret < 0)
1564                         return ret;
1565
1566                 ret = do_write_string(ff, caps->value);
1567                 if (ret < 0)
1568                         return ret;
1569         }
1570
1571         if (write_pmu) {
1572                 ret = do_write_string(ff, pmu->name);
1573                 if (ret < 0)
1574                         return ret;
1575         }
1576
1577         return ret;
1578 }
1579
1580 static int write_cpu_pmu_caps(struct feat_fd *ff,
1581                               struct evlist *evlist __maybe_unused)
1582 {
1583         struct perf_pmu *cpu_pmu = perf_pmus__find("cpu");
1584         int ret;
1585
1586         if (!cpu_pmu)
1587                 return -ENOENT;
1588
1589         ret = perf_pmu__caps_parse(cpu_pmu);
1590         if (ret < 0)
1591                 return ret;
1592
1593         return __write_pmu_caps(ff, cpu_pmu, false);
1594 }
1595
1596 static int write_pmu_caps(struct feat_fd *ff,
1597                           struct evlist *evlist __maybe_unused)
1598 {
1599         struct perf_pmu *pmu = NULL;
1600         int nr_pmu = 0;
1601         int ret;
1602
1603         while ((pmu = perf_pmus__scan(pmu))) {
1604                 if (!strcmp(pmu->name, "cpu")) {
1605                         /*
1606                          * The "cpu" PMU is special and covered by
1607                          * HEADER_CPU_PMU_CAPS. Note, core PMUs are
1608                          * counted/written here for ARM, s390 and Intel hybrid.
1609                          */
1610                         continue;
1611                 }
1612                 if (perf_pmu__caps_parse(pmu) <= 0)
1613                         continue;
1614                 nr_pmu++;
1615         }
1616
1617         ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1618         if (ret < 0)
1619                 return ret;
1620
1621         if (!nr_pmu)
1622                 return 0;
1623
1624         /*
1625          * Note older perf tools assume core PMUs come first, this is a property
1626          * of perf_pmus__scan.
1627          */
1628         pmu = NULL;
1629         while ((pmu = perf_pmus__scan(pmu))) {
1630                 if (!strcmp(pmu->name, "cpu")) {
1631                         /* Skip as above. */
1632                         continue;
1633                 }
1634                 if (perf_pmu__caps_parse(pmu) <= 0)
1635                         continue;
1636                 ret = __write_pmu_caps(ff, pmu, true);
1637                 if (ret < 0)
1638                         return ret;
1639         }
1640         return 0;
1641 }
1642
1643 static void print_hostname(struct feat_fd *ff, FILE *fp)
1644 {
1645         fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1646 }
1647
1648 static void print_osrelease(struct feat_fd *ff, FILE *fp)
1649 {
1650         fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1651 }
1652
1653 static void print_arch(struct feat_fd *ff, FILE *fp)
1654 {
1655         fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1656 }
1657
1658 static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1659 {
1660         fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1661 }
1662
1663 static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1664 {
1665         fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1666         fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1667 }
1668
1669 static void print_version(struct feat_fd *ff, FILE *fp)
1670 {
1671         fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1672 }
1673
1674 static void print_cmdline(struct feat_fd *ff, FILE *fp)
1675 {
1676         int nr, i;
1677
1678         nr = ff->ph->env.nr_cmdline;
1679
1680         fprintf(fp, "# cmdline : ");
1681
1682         for (i = 0; i < nr; i++) {
1683                 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1684                 if (!argv_i) {
1685                         fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1686                 } else {
1687                         char *mem = argv_i;
1688                         do {
1689                                 char *quote = strchr(argv_i, '\'');
1690                                 if (!quote)
1691                                         break;
1692                                 *quote++ = '\0';
1693                                 fprintf(fp, "%s\\\'", argv_i);
1694                                 argv_i = quote;
1695                         } while (1);
1696                         fprintf(fp, "%s ", argv_i);
1697                         free(mem);
1698                 }
1699         }
1700         fputc('\n', fp);
1701 }
1702
1703 static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1704 {
1705         struct perf_header *ph = ff->ph;
1706         int cpu_nr = ph->env.nr_cpus_avail;
1707         int nr, i;
1708         char *str;
1709
1710         nr = ph->env.nr_sibling_cores;
1711         str = ph->env.sibling_cores;
1712
1713         for (i = 0; i < nr; i++) {
1714                 fprintf(fp, "# sibling sockets : %s\n", str);
1715                 str += strlen(str) + 1;
1716         }
1717
1718         if (ph->env.nr_sibling_dies) {
1719                 nr = ph->env.nr_sibling_dies;
1720                 str = ph->env.sibling_dies;
1721
1722                 for (i = 0; i < nr; i++) {
1723                         fprintf(fp, "# sibling dies    : %s\n", str);
1724                         str += strlen(str) + 1;
1725                 }
1726         }
1727
1728         nr = ph->env.nr_sibling_threads;
1729         str = ph->env.sibling_threads;
1730
1731         for (i = 0; i < nr; i++) {
1732                 fprintf(fp, "# sibling threads : %s\n", str);
1733                 str += strlen(str) + 1;
1734         }
1735
1736         if (ph->env.nr_sibling_dies) {
1737                 if (ph->env.cpu != NULL) {
1738                         for (i = 0; i < cpu_nr; i++)
1739                                 fprintf(fp, "# CPU %d: Core ID %d, "
1740                                             "Die ID %d, Socket ID %d\n",
1741                                             i, ph->env.cpu[i].core_id,
1742                                             ph->env.cpu[i].die_id,
1743                                             ph->env.cpu[i].socket_id);
1744                 } else
1745                         fprintf(fp, "# Core ID, Die ID and Socket ID "
1746                                     "information is not available\n");
1747         } else {
1748                 if (ph->env.cpu != NULL) {
1749                         for (i = 0; i < cpu_nr; i++)
1750                                 fprintf(fp, "# CPU %d: Core ID %d, "
1751                                             "Socket ID %d\n",
1752                                             i, ph->env.cpu[i].core_id,
1753                                             ph->env.cpu[i].socket_id);
1754                 } else
1755                         fprintf(fp, "# Core ID and Socket ID "
1756                                     "information is not available\n");
1757         }
1758 }
1759
1760 static void print_clockid(struct feat_fd *ff, FILE *fp)
1761 {
1762         fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1763                 ff->ph->env.clock.clockid_res_ns * 1000);
1764 }
1765
1766 static void print_clock_data(struct feat_fd *ff, FILE *fp)
1767 {
1768         struct timespec clockid_ns;
1769         char tstr[64], date[64];
1770         struct timeval tod_ns;
1771         clockid_t clockid;
1772         struct tm ltime;
1773         u64 ref;
1774
1775         if (!ff->ph->env.clock.enabled) {
1776                 fprintf(fp, "# reference time disabled\n");
1777                 return;
1778         }
1779
1780         /* Compute TOD time. */
1781         ref = ff->ph->env.clock.tod_ns;
1782         tod_ns.tv_sec = ref / NSEC_PER_SEC;
1783         ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1784         tod_ns.tv_usec = ref / NSEC_PER_USEC;
1785
1786         /* Compute clockid time. */
1787         ref = ff->ph->env.clock.clockid_ns;
1788         clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1789         ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1790         clockid_ns.tv_nsec = ref;
1791
1792         clockid = ff->ph->env.clock.clockid;
1793
1794         if (localtime_r(&tod_ns.tv_sec, &ltime) == NULL)
1795                 snprintf(tstr, sizeof(tstr), "<error>");
1796         else {
1797                 strftime(date, sizeof(date), "%F %T", &ltime);
1798                 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1799                           date, (int) tod_ns.tv_usec);
1800         }
1801
1802         fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1803         fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1804                     tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1805                     (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1806                     clockid_name(clockid));
1807 }
1808
1809 static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1810 {
1811         int i;
1812         struct hybrid_node *n;
1813
1814         fprintf(fp, "# hybrid cpu system:\n");
1815         for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1816                 n = &ff->ph->env.hybrid_nodes[i];
1817                 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1818         }
1819 }
1820
1821 static void print_dir_format(struct feat_fd *ff, FILE *fp)
1822 {
1823         struct perf_session *session;
1824         struct perf_data *data;
1825
1826         session = container_of(ff->ph, struct perf_session, header);
1827         data = session->data;
1828
1829         fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1830 }
1831
1832 #ifdef HAVE_LIBBPF_SUPPORT
1833 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1834 {
1835         struct perf_env *env = &ff->ph->env;
1836         struct rb_root *root;
1837         struct rb_node *next;
1838
1839         down_read(&env->bpf_progs.lock);
1840
1841         root = &env->bpf_progs.infos;
1842         next = rb_first(root);
1843
1844         while (next) {
1845                 struct bpf_prog_info_node *node;
1846
1847                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1848                 next = rb_next(&node->rb_node);
1849
1850                 bpf_event__print_bpf_prog_info(&node->info_linear->info,
1851                                                env, fp);
1852         }
1853
1854         up_read(&env->bpf_progs.lock);
1855 }
1856
1857 static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1858 {
1859         struct perf_env *env = &ff->ph->env;
1860         struct rb_root *root;
1861         struct rb_node *next;
1862
1863         down_read(&env->bpf_progs.lock);
1864
1865         root = &env->bpf_progs.btfs;
1866         next = rb_first(root);
1867
1868         while (next) {
1869                 struct btf_node *node;
1870
1871                 node = rb_entry(next, struct btf_node, rb_node);
1872                 next = rb_next(&node->rb_node);
1873                 fprintf(fp, "# btf info of id %u\n", node->id);
1874         }
1875
1876         up_read(&env->bpf_progs.lock);
1877 }
1878 #endif // HAVE_LIBBPF_SUPPORT
1879
1880 static void free_event_desc(struct evsel *events)
1881 {
1882         struct evsel *evsel;
1883
1884         if (!events)
1885                 return;
1886
1887         for (evsel = events; evsel->core.attr.size; evsel++) {
1888                 zfree(&evsel->name);
1889                 zfree(&evsel->core.id);
1890         }
1891
1892         free(events);
1893 }
1894
1895 static bool perf_attr_check(struct perf_event_attr *attr)
1896 {
1897         if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1898                 pr_warning("Reserved bits are set unexpectedly. "
1899                            "Please update perf tool.\n");
1900                 return false;
1901         }
1902
1903         if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1904                 pr_warning("Unknown sample type (0x%llx) is detected. "
1905                            "Please update perf tool.\n",
1906                            attr->sample_type);
1907                 return false;
1908         }
1909
1910         if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1911                 pr_warning("Unknown read format (0x%llx) is detected. "
1912                            "Please update perf tool.\n",
1913                            attr->read_format);
1914                 return false;
1915         }
1916
1917         if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1918             (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1919                 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1920                            "Please update perf tool.\n",
1921                            attr->branch_sample_type);
1922
1923                 return false;
1924         }
1925
1926         return true;
1927 }
1928
1929 static struct evsel *read_event_desc(struct feat_fd *ff)
1930 {
1931         struct evsel *evsel, *events = NULL;
1932         u64 *id;
1933         void *buf = NULL;
1934         u32 nre, sz, nr, i, j;
1935         size_t msz;
1936
1937         /* number of events */
1938         if (do_read_u32(ff, &nre))
1939                 goto error;
1940
1941         if (do_read_u32(ff, &sz))
1942                 goto error;
1943
1944         /* buffer to hold on file attr struct */
1945         buf = malloc(sz);
1946         if (!buf)
1947                 goto error;
1948
1949         /* the last event terminates with evsel->core.attr.size == 0: */
1950         events = calloc(nre + 1, sizeof(*events));
1951         if (!events)
1952                 goto error;
1953
1954         msz = sizeof(evsel->core.attr);
1955         if (sz < msz)
1956                 msz = sz;
1957
1958         for (i = 0, evsel = events; i < nre; evsel++, i++) {
1959                 evsel->core.idx = i;
1960
1961                 /*
1962                  * must read entire on-file attr struct to
1963                  * sync up with layout.
1964                  */
1965                 if (__do_read(ff, buf, sz))
1966                         goto error;
1967
1968                 if (ff->ph->needs_swap)
1969                         perf_event__attr_swap(buf);
1970
1971                 memcpy(&evsel->core.attr, buf, msz);
1972
1973                 if (!perf_attr_check(&evsel->core.attr))
1974                         goto error;
1975
1976                 if (do_read_u32(ff, &nr))
1977                         goto error;
1978
1979                 if (ff->ph->needs_swap)
1980                         evsel->needs_swap = true;
1981
1982                 evsel->name = do_read_string(ff);
1983                 if (!evsel->name)
1984                         goto error;
1985
1986                 if (!nr)
1987                         continue;
1988
1989                 id = calloc(nr, sizeof(*id));
1990                 if (!id)
1991                         goto error;
1992                 evsel->core.ids = nr;
1993                 evsel->core.id = id;
1994
1995                 for (j = 0 ; j < nr; j++) {
1996                         if (do_read_u64(ff, id))
1997                                 goto error;
1998                         id++;
1999                 }
2000         }
2001 out:
2002         free(buf);
2003         return events;
2004 error:
2005         free_event_desc(events);
2006         events = NULL;
2007         goto out;
2008 }
2009
2010 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
2011                                 void *priv __maybe_unused)
2012 {
2013         return fprintf(fp, ", %s = %s", name, val);
2014 }
2015
2016 static void print_event_desc(struct feat_fd *ff, FILE *fp)
2017 {
2018         struct evsel *evsel, *events;
2019         u32 j;
2020         u64 *id;
2021
2022         if (ff->events)
2023                 events = ff->events;
2024         else
2025                 events = read_event_desc(ff);
2026
2027         if (!events) {
2028                 fprintf(fp, "# event desc: not available or unable to read\n");
2029                 return;
2030         }
2031
2032         for (evsel = events; evsel->core.attr.size; evsel++) {
2033                 fprintf(fp, "# event : name = %s, ", evsel->name);
2034
2035                 if (evsel->core.ids) {
2036                         fprintf(fp, ", id = {");
2037                         for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
2038                                 if (j)
2039                                         fputc(',', fp);
2040                                 fprintf(fp, " %"PRIu64, *id);
2041                         }
2042                         fprintf(fp, " }");
2043                 }
2044
2045                 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
2046
2047                 fputc('\n', fp);
2048         }
2049
2050         free_event_desc(events);
2051         ff->events = NULL;
2052 }
2053
2054 static void print_total_mem(struct feat_fd *ff, FILE *fp)
2055 {
2056         fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
2057 }
2058
2059 static void print_numa_topology(struct feat_fd *ff, FILE *fp)
2060 {
2061         int i;
2062         struct numa_node *n;
2063
2064         for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
2065                 n = &ff->ph->env.numa_nodes[i];
2066
2067                 fprintf(fp, "# node%u meminfo  : total = %"PRIu64" kB,"
2068                             " free = %"PRIu64" kB\n",
2069                         n->node, n->mem_total, n->mem_free);
2070
2071                 fprintf(fp, "# node%u cpu list : ", n->node);
2072                 cpu_map__fprintf(n->map, fp);
2073         }
2074 }
2075
2076 static void print_cpuid(struct feat_fd *ff, FILE *fp)
2077 {
2078         fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
2079 }
2080
2081 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
2082 {
2083         fprintf(fp, "# contains samples with branch stack\n");
2084 }
2085
2086 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
2087 {
2088         fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
2089 }
2090
2091 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
2092 {
2093         fprintf(fp, "# contains stat data\n");
2094 }
2095
2096 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
2097 {
2098         int i;
2099
2100         fprintf(fp, "# CPU cache info:\n");
2101         for (i = 0; i < ff->ph->env.caches_cnt; i++) {
2102                 fprintf(fp, "#  ");
2103                 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
2104         }
2105 }
2106
2107 static void print_compressed(struct feat_fd *ff, FILE *fp)
2108 {
2109         fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
2110                 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
2111                 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2112 }
2113
2114 static void __print_pmu_caps(FILE *fp, int nr_caps, char **caps, char *pmu_name)
2115 {
2116         const char *delimiter = "";
2117         int i;
2118
2119         if (!nr_caps) {
2120                 fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2121                 return;
2122         }
2123
2124         fprintf(fp, "# %s pmu capabilities: ", pmu_name);
2125         for (i = 0; i < nr_caps; i++) {
2126                 fprintf(fp, "%s%s", delimiter, caps[i]);
2127                 delimiter = ", ";
2128         }
2129
2130         fprintf(fp, "\n");
2131 }
2132
2133 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2134 {
2135         __print_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2136                          ff->ph->env.cpu_pmu_caps, (char *)"cpu");
2137 }
2138
2139 static void print_pmu_caps(struct feat_fd *ff, FILE *fp)
2140 {
2141         struct pmu_caps *pmu_caps;
2142
2143         for (int i = 0; i < ff->ph->env.nr_pmus_with_caps; i++) {
2144                 pmu_caps = &ff->ph->env.pmu_caps[i];
2145                 __print_pmu_caps(fp, pmu_caps->nr_caps, pmu_caps->caps,
2146                                  pmu_caps->pmu_name);
2147         }
2148 }
2149
2150 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2151 {
2152         const char *delimiter = "# pmu mappings: ";
2153         char *str, *tmp;
2154         u32 pmu_num;
2155         u32 type;
2156
2157         pmu_num = ff->ph->env.nr_pmu_mappings;
2158         if (!pmu_num) {
2159                 fprintf(fp, "# pmu mappings: not available\n");
2160                 return;
2161         }
2162
2163         str = ff->ph->env.pmu_mappings;
2164
2165         while (pmu_num) {
2166                 type = strtoul(str, &tmp, 0);
2167                 if (*tmp != ':')
2168                         goto error;
2169
2170                 str = tmp + 1;
2171                 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2172
2173                 delimiter = ", ";
2174                 str += strlen(str) + 1;
2175                 pmu_num--;
2176         }
2177
2178         fprintf(fp, "\n");
2179
2180         if (!pmu_num)
2181                 return;
2182 error:
2183         fprintf(fp, "# pmu mappings: unable to read\n");
2184 }
2185
2186 static void print_group_desc(struct feat_fd *ff, FILE *fp)
2187 {
2188         struct perf_session *session;
2189         struct evsel *evsel;
2190         u32 nr = 0;
2191
2192         session = container_of(ff->ph, struct perf_session, header);
2193
2194         evlist__for_each_entry(session->evlist, evsel) {
2195                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2196                         fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2197
2198                         nr = evsel->core.nr_members - 1;
2199                 } else if (nr) {
2200                         fprintf(fp, ",%s", evsel__name(evsel));
2201
2202                         if (--nr == 0)
2203                                 fprintf(fp, "}\n");
2204                 }
2205         }
2206 }
2207
2208 static void print_sample_time(struct feat_fd *ff, FILE *fp)
2209 {
2210         struct perf_session *session;
2211         char time_buf[32];
2212         double d;
2213
2214         session = container_of(ff->ph, struct perf_session, header);
2215
2216         timestamp__scnprintf_usec(session->evlist->first_sample_time,
2217                                   time_buf, sizeof(time_buf));
2218         fprintf(fp, "# time of first sample : %s\n", time_buf);
2219
2220         timestamp__scnprintf_usec(session->evlist->last_sample_time,
2221                                   time_buf, sizeof(time_buf));
2222         fprintf(fp, "# time of last sample : %s\n", time_buf);
2223
2224         d = (double)(session->evlist->last_sample_time -
2225                 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2226
2227         fprintf(fp, "# sample duration : %10.3f ms\n", d);
2228 }
2229
2230 static void memory_node__fprintf(struct memory_node *n,
2231                                  unsigned long long bsize, FILE *fp)
2232 {
2233         char buf_map[100], buf_size[50];
2234         unsigned long long size;
2235
2236         size = bsize * bitmap_weight(n->set, n->size);
2237         unit_number__scnprintf(buf_size, 50, size);
2238
2239         bitmap_scnprintf(n->set, n->size, buf_map, 100);
2240         fprintf(fp, "#  %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2241 }
2242
2243 static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2244 {
2245         struct memory_node *nodes;
2246         int i, nr;
2247
2248         nodes = ff->ph->env.memory_nodes;
2249         nr    = ff->ph->env.nr_memory_nodes;
2250
2251         fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2252                 nr, ff->ph->env.memory_bsize);
2253
2254         for (i = 0; i < nr; i++) {
2255                 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2256         }
2257 }
2258
2259 static int __event_process_build_id(struct perf_record_header_build_id *bev,
2260                                     char *filename,
2261                                     struct perf_session *session)
2262 {
2263         int err = -1;
2264         struct machine *machine;
2265         u16 cpumode;
2266         struct dso *dso;
2267         enum dso_space_type dso_space;
2268
2269         machine = perf_session__findnew_machine(session, bev->pid);
2270         if (!machine)
2271                 goto out;
2272
2273         cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2274
2275         switch (cpumode) {
2276         case PERF_RECORD_MISC_KERNEL:
2277                 dso_space = DSO_SPACE__KERNEL;
2278                 break;
2279         case PERF_RECORD_MISC_GUEST_KERNEL:
2280                 dso_space = DSO_SPACE__KERNEL_GUEST;
2281                 break;
2282         case PERF_RECORD_MISC_USER:
2283         case PERF_RECORD_MISC_GUEST_USER:
2284                 dso_space = DSO_SPACE__USER;
2285                 break;
2286         default:
2287                 goto out;
2288         }
2289
2290         dso = machine__findnew_dso(machine, filename);
2291         if (dso != NULL) {
2292                 char sbuild_id[SBUILD_ID_SIZE];
2293                 struct build_id bid;
2294                 size_t size = BUILD_ID_SIZE;
2295
2296                 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2297                         size = bev->size;
2298
2299                 build_id__init(&bid, bev->data, size);
2300                 dso__set_build_id(dso, &bid);
2301                 dso->header_build_id = 1;
2302
2303                 if (dso_space != DSO_SPACE__USER) {
2304                         struct kmod_path m = { .name = NULL, };
2305
2306                         if (!kmod_path__parse_name(&m, filename) && m.kmod)
2307                                 dso__set_module_info(dso, &m, machine);
2308
2309                         dso->kernel = dso_space;
2310                         free(m.name);
2311                 }
2312
2313                 build_id__sprintf(&dso->bid, sbuild_id);
2314                 pr_debug("build id event received for %s: %s [%zu]\n",
2315                          dso->long_name, sbuild_id, size);
2316                 dso__put(dso);
2317         }
2318
2319         err = 0;
2320 out:
2321         return err;
2322 }
2323
2324 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2325                                                  int input, u64 offset, u64 size)
2326 {
2327         struct perf_session *session = container_of(header, struct perf_session, header);
2328         struct {
2329                 struct perf_event_header   header;
2330                 u8                         build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2331                 char                       filename[0];
2332         } old_bev;
2333         struct perf_record_header_build_id bev;
2334         char filename[PATH_MAX];
2335         u64 limit = offset + size;
2336
2337         while (offset < limit) {
2338                 ssize_t len;
2339
2340                 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2341                         return -1;
2342
2343                 if (header->needs_swap)
2344                         perf_event_header__bswap(&old_bev.header);
2345
2346                 len = old_bev.header.size - sizeof(old_bev);
2347                 if (readn(input, filename, len) != len)
2348                         return -1;
2349
2350                 bev.header = old_bev.header;
2351
2352                 /*
2353                  * As the pid is the missing value, we need to fill
2354                  * it properly. The header.misc value give us nice hint.
2355                  */
2356                 bev.pid = HOST_KERNEL_ID;
2357                 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2358                     bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2359                         bev.pid = DEFAULT_GUEST_KERNEL_ID;
2360
2361                 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2362                 __event_process_build_id(&bev, filename, session);
2363
2364                 offset += bev.header.size;
2365         }
2366
2367         return 0;
2368 }
2369
2370 static int perf_header__read_build_ids(struct perf_header *header,
2371                                        int input, u64 offset, u64 size)
2372 {
2373         struct perf_session *session = container_of(header, struct perf_session, header);
2374         struct perf_record_header_build_id bev;
2375         char filename[PATH_MAX];
2376         u64 limit = offset + size, orig_offset = offset;
2377         int err = -1;
2378
2379         while (offset < limit) {
2380                 ssize_t len;
2381
2382                 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2383                         goto out;
2384
2385                 if (header->needs_swap)
2386                         perf_event_header__bswap(&bev.header);
2387
2388                 len = bev.header.size - sizeof(bev);
2389                 if (readn(input, filename, len) != len)
2390                         goto out;
2391                 /*
2392                  * The a1645ce1 changeset:
2393                  *
2394                  * "perf: 'perf kvm' tool for monitoring guest performance from host"
2395                  *
2396                  * Added a field to struct perf_record_header_build_id that broke the file
2397                  * format.
2398                  *
2399                  * Since the kernel build-id is the first entry, process the
2400                  * table using the old format if the well known
2401                  * '[kernel.kallsyms]' string for the kernel build-id has the
2402                  * first 4 characters chopped off (where the pid_t sits).
2403                  */
2404                 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2405                         if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2406                                 return -1;
2407                         return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2408                 }
2409
2410                 __event_process_build_id(&bev, filename, session);
2411
2412                 offset += bev.header.size;
2413         }
2414         err = 0;
2415 out:
2416         return err;
2417 }
2418
2419 /* Macro for features that simply need to read and store a string. */
2420 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2421 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2422 {\
2423         free(ff->ph->env.__feat_env);                \
2424         ff->ph->env.__feat_env = do_read_string(ff); \
2425         return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2426 }
2427
2428 FEAT_PROCESS_STR_FUN(hostname, hostname);
2429 FEAT_PROCESS_STR_FUN(osrelease, os_release);
2430 FEAT_PROCESS_STR_FUN(version, version);
2431 FEAT_PROCESS_STR_FUN(arch, arch);
2432 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2433 FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2434
2435 #ifdef HAVE_LIBTRACEEVENT
2436 static int process_tracing_data(struct feat_fd *ff, void *data)
2437 {
2438         ssize_t ret = trace_report(ff->fd, data, false);
2439
2440         return ret < 0 ? -1 : 0;
2441 }
2442 #endif
2443
2444 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2445 {
2446         if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2447                 pr_debug("Failed to read buildids, continuing...\n");
2448         return 0;
2449 }
2450
2451 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2452 {
2453         int ret;
2454         u32 nr_cpus_avail, nr_cpus_online;
2455
2456         ret = do_read_u32(ff, &nr_cpus_avail);
2457         if (ret)
2458                 return ret;
2459
2460         ret = do_read_u32(ff, &nr_cpus_online);
2461         if (ret)
2462                 return ret;
2463         ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2464         ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2465         return 0;
2466 }
2467
2468 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2469 {
2470         u64 total_mem;
2471         int ret;
2472
2473         ret = do_read_u64(ff, &total_mem);
2474         if (ret)
2475                 return -1;
2476         ff->ph->env.total_mem = (unsigned long long)total_mem;
2477         return 0;
2478 }
2479
2480 static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2481 {
2482         struct evsel *evsel;
2483
2484         evlist__for_each_entry(evlist, evsel) {
2485                 if (evsel->core.idx == idx)
2486                         return evsel;
2487         }
2488
2489         return NULL;
2490 }
2491
2492 static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2493 {
2494         struct evsel *evsel;
2495
2496         if (!event->name)
2497                 return;
2498
2499         evsel = evlist__find_by_index(evlist, event->core.idx);
2500         if (!evsel)
2501                 return;
2502
2503         if (evsel->name)
2504                 return;
2505
2506         evsel->name = strdup(event->name);
2507 }
2508
2509 static int
2510 process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2511 {
2512         struct perf_session *session;
2513         struct evsel *evsel, *events = read_event_desc(ff);
2514
2515         if (!events)
2516                 return 0;
2517
2518         session = container_of(ff->ph, struct perf_session, header);
2519
2520         if (session->data->is_pipe) {
2521                 /* Save events for reading later by print_event_desc,
2522                  * since they can't be read again in pipe mode. */
2523                 ff->events = events;
2524         }
2525
2526         for (evsel = events; evsel->core.attr.size; evsel++)
2527                 evlist__set_event_name(session->evlist, evsel);
2528
2529         if (!session->data->is_pipe)
2530                 free_event_desc(events);
2531
2532         return 0;
2533 }
2534
2535 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2536 {
2537         char *str, *cmdline = NULL, **argv = NULL;
2538         u32 nr, i, len = 0;
2539
2540         if (do_read_u32(ff, &nr))
2541                 return -1;
2542
2543         ff->ph->env.nr_cmdline = nr;
2544
2545         cmdline = zalloc(ff->size + nr + 1);
2546         if (!cmdline)
2547                 return -1;
2548
2549         argv = zalloc(sizeof(char *) * (nr + 1));
2550         if (!argv)
2551                 goto error;
2552
2553         for (i = 0; i < nr; i++) {
2554                 str = do_read_string(ff);
2555                 if (!str)
2556                         goto error;
2557
2558                 argv[i] = cmdline + len;
2559                 memcpy(argv[i], str, strlen(str) + 1);
2560                 len += strlen(str) + 1;
2561                 free(str);
2562         }
2563         ff->ph->env.cmdline = cmdline;
2564         ff->ph->env.cmdline_argv = (const char **) argv;
2565         return 0;
2566
2567 error:
2568         free(argv);
2569         free(cmdline);
2570         return -1;
2571 }
2572
2573 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2574 {
2575         u32 nr, i;
2576         char *str;
2577         struct strbuf sb;
2578         int cpu_nr = ff->ph->env.nr_cpus_avail;
2579         u64 size = 0;
2580         struct perf_header *ph = ff->ph;
2581         bool do_core_id_test = true;
2582
2583         ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2584         if (!ph->env.cpu)
2585                 return -1;
2586
2587         if (do_read_u32(ff, &nr))
2588                 goto free_cpu;
2589
2590         ph->env.nr_sibling_cores = nr;
2591         size += sizeof(u32);
2592         if (strbuf_init(&sb, 128) < 0)
2593                 goto free_cpu;
2594
2595         for (i = 0; i < nr; i++) {
2596                 str = do_read_string(ff);
2597                 if (!str)
2598                         goto error;
2599
2600                 /* include a NULL character at the end */
2601                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2602                         goto error;
2603                 size += string_size(str);
2604                 free(str);
2605         }
2606         ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2607
2608         if (do_read_u32(ff, &nr))
2609                 return -1;
2610
2611         ph->env.nr_sibling_threads = nr;
2612         size += sizeof(u32);
2613
2614         for (i = 0; i < nr; i++) {
2615                 str = do_read_string(ff);
2616                 if (!str)
2617                         goto error;
2618
2619                 /* include a NULL character at the end */
2620                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2621                         goto error;
2622                 size += string_size(str);
2623                 free(str);
2624         }
2625         ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2626
2627         /*
2628          * The header may be from old perf,
2629          * which doesn't include core id and socket id information.
2630          */
2631         if (ff->size <= size) {
2632                 zfree(&ph->env.cpu);
2633                 return 0;
2634         }
2635
2636         /* On s390 the socket_id number is not related to the numbers of cpus.
2637          * The socket_id number might be higher than the numbers of cpus.
2638          * This depends on the configuration.
2639          * AArch64 is the same.
2640          */
2641         if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2642                           || !strncmp(ph->env.arch, "aarch64", 7)))
2643                 do_core_id_test = false;
2644
2645         for (i = 0; i < (u32)cpu_nr; i++) {
2646                 if (do_read_u32(ff, &nr))
2647                         goto free_cpu;
2648
2649                 ph->env.cpu[i].core_id = nr;
2650                 size += sizeof(u32);
2651
2652                 if (do_read_u32(ff, &nr))
2653                         goto free_cpu;
2654
2655                 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2656                         pr_debug("socket_id number is too big."
2657                                  "You may need to upgrade the perf tool.\n");
2658                         goto free_cpu;
2659                 }
2660
2661                 ph->env.cpu[i].socket_id = nr;
2662                 size += sizeof(u32);
2663         }
2664
2665         /*
2666          * The header may be from old perf,
2667          * which doesn't include die information.
2668          */
2669         if (ff->size <= size)
2670                 return 0;
2671
2672         if (do_read_u32(ff, &nr))
2673                 return -1;
2674
2675         ph->env.nr_sibling_dies = nr;
2676         size += sizeof(u32);
2677
2678         for (i = 0; i < nr; i++) {
2679                 str = do_read_string(ff);
2680                 if (!str)
2681                         goto error;
2682
2683                 /* include a NULL character at the end */
2684                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2685                         goto error;
2686                 size += string_size(str);
2687                 free(str);
2688         }
2689         ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2690
2691         for (i = 0; i < (u32)cpu_nr; i++) {
2692                 if (do_read_u32(ff, &nr))
2693                         goto free_cpu;
2694
2695                 ph->env.cpu[i].die_id = nr;
2696         }
2697
2698         return 0;
2699
2700 error:
2701         strbuf_release(&sb);
2702 free_cpu:
2703         zfree(&ph->env.cpu);
2704         return -1;
2705 }
2706
2707 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2708 {
2709         struct numa_node *nodes, *n;
2710         u32 nr, i;
2711         char *str;
2712
2713         /* nr nodes */
2714         if (do_read_u32(ff, &nr))
2715                 return -1;
2716
2717         nodes = zalloc(sizeof(*nodes) * nr);
2718         if (!nodes)
2719                 return -ENOMEM;
2720
2721         for (i = 0; i < nr; i++) {
2722                 n = &nodes[i];
2723
2724                 /* node number */
2725                 if (do_read_u32(ff, &n->node))
2726                         goto error;
2727
2728                 if (do_read_u64(ff, &n->mem_total))
2729                         goto error;
2730
2731                 if (do_read_u64(ff, &n->mem_free))
2732                         goto error;
2733
2734                 str = do_read_string(ff);
2735                 if (!str)
2736                         goto error;
2737
2738                 n->map = perf_cpu_map__new(str);
2739                 if (!n->map)
2740                         goto error;
2741
2742                 free(str);
2743         }
2744         ff->ph->env.nr_numa_nodes = nr;
2745         ff->ph->env.numa_nodes = nodes;
2746         return 0;
2747
2748 error:
2749         free(nodes);
2750         return -1;
2751 }
2752
2753 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2754 {
2755         char *name;
2756         u32 pmu_num;
2757         u32 type;
2758         struct strbuf sb;
2759
2760         if (do_read_u32(ff, &pmu_num))
2761                 return -1;
2762
2763         if (!pmu_num) {
2764                 pr_debug("pmu mappings not available\n");
2765                 return 0;
2766         }
2767
2768         ff->ph->env.nr_pmu_mappings = pmu_num;
2769         if (strbuf_init(&sb, 128) < 0)
2770                 return -1;
2771
2772         while (pmu_num) {
2773                 if (do_read_u32(ff, &type))
2774                         goto error;
2775
2776                 name = do_read_string(ff);
2777                 if (!name)
2778                         goto error;
2779
2780                 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2781                         goto error;
2782                 /* include a NULL character at the end */
2783                 if (strbuf_add(&sb, "", 1) < 0)
2784                         goto error;
2785
2786                 if (!strcmp(name, "msr"))
2787                         ff->ph->env.msr_pmu_type = type;
2788
2789                 free(name);
2790                 pmu_num--;
2791         }
2792         ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2793         return 0;
2794
2795 error:
2796         strbuf_release(&sb);
2797         return -1;
2798 }
2799
2800 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2801 {
2802         size_t ret = -1;
2803         u32 i, nr, nr_groups;
2804         struct perf_session *session;
2805         struct evsel *evsel, *leader = NULL;
2806         struct group_desc {
2807                 char *name;
2808                 u32 leader_idx;
2809                 u32 nr_members;
2810         } *desc;
2811
2812         if (do_read_u32(ff, &nr_groups))
2813                 return -1;
2814
2815         ff->ph->env.nr_groups = nr_groups;
2816         if (!nr_groups) {
2817                 pr_debug("group desc not available\n");
2818                 return 0;
2819         }
2820
2821         desc = calloc(nr_groups, sizeof(*desc));
2822         if (!desc)
2823                 return -1;
2824
2825         for (i = 0; i < nr_groups; i++) {
2826                 desc[i].name = do_read_string(ff);
2827                 if (!desc[i].name)
2828                         goto out_free;
2829
2830                 if (do_read_u32(ff, &desc[i].leader_idx))
2831                         goto out_free;
2832
2833                 if (do_read_u32(ff, &desc[i].nr_members))
2834                         goto out_free;
2835         }
2836
2837         /*
2838          * Rebuild group relationship based on the group_desc
2839          */
2840         session = container_of(ff->ph, struct perf_session, header);
2841
2842         i = nr = 0;
2843         evlist__for_each_entry(session->evlist, evsel) {
2844                 if (i < nr_groups && evsel->core.idx == (int) desc[i].leader_idx) {
2845                         evsel__set_leader(evsel, evsel);
2846                         /* {anon_group} is a dummy name */
2847                         if (strcmp(desc[i].name, "{anon_group}")) {
2848                                 evsel->group_name = desc[i].name;
2849                                 desc[i].name = NULL;
2850                         }
2851                         evsel->core.nr_members = desc[i].nr_members;
2852
2853                         if (i >= nr_groups || nr > 0) {
2854                                 pr_debug("invalid group desc\n");
2855                                 goto out_free;
2856                         }
2857
2858                         leader = evsel;
2859                         nr = evsel->core.nr_members - 1;
2860                         i++;
2861                 } else if (nr) {
2862                         /* This is a group member */
2863                         evsel__set_leader(evsel, leader);
2864
2865                         nr--;
2866                 }
2867         }
2868
2869         if (i != nr_groups || nr != 0) {
2870                 pr_debug("invalid group desc\n");
2871                 goto out_free;
2872         }
2873
2874         ret = 0;
2875 out_free:
2876         for (i = 0; i < nr_groups; i++)
2877                 zfree(&desc[i].name);
2878         free(desc);
2879
2880         return ret;
2881 }
2882
2883 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2884 {
2885         struct perf_session *session;
2886         int err;
2887
2888         session = container_of(ff->ph, struct perf_session, header);
2889
2890         err = auxtrace_index__process(ff->fd, ff->size, session,
2891                                       ff->ph->needs_swap);
2892         if (err < 0)
2893                 pr_err("Failed to process auxtrace index\n");
2894         return err;
2895 }
2896
2897 static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2898 {
2899         struct cpu_cache_level *caches;
2900         u32 cnt, i, version;
2901
2902         if (do_read_u32(ff, &version))
2903                 return -1;
2904
2905         if (version != 1)
2906                 return -1;
2907
2908         if (do_read_u32(ff, &cnt))
2909                 return -1;
2910
2911         caches = zalloc(sizeof(*caches) * cnt);
2912         if (!caches)
2913                 return -1;
2914
2915         for (i = 0; i < cnt; i++) {
2916                 struct cpu_cache_level c;
2917
2918                 #define _R(v)                                           \
2919                         if (do_read_u32(ff, &c.v))\
2920                                 goto out_free_caches;                   \
2921
2922                 _R(level)
2923                 _R(line_size)
2924                 _R(sets)
2925                 _R(ways)
2926                 #undef _R
2927
2928                 #define _R(v)                                   \
2929                         c.v = do_read_string(ff);               \
2930                         if (!c.v)                               \
2931                                 goto out_free_caches;
2932
2933                 _R(type)
2934                 _R(size)
2935                 _R(map)
2936                 #undef _R
2937
2938                 caches[i] = c;
2939         }
2940
2941         ff->ph->env.caches = caches;
2942         ff->ph->env.caches_cnt = cnt;
2943         return 0;
2944 out_free_caches:
2945         free(caches);
2946         return -1;
2947 }
2948
2949 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2950 {
2951         struct perf_session *session;
2952         u64 first_sample_time, last_sample_time;
2953         int ret;
2954
2955         session = container_of(ff->ph, struct perf_session, header);
2956
2957         ret = do_read_u64(ff, &first_sample_time);
2958         if (ret)
2959                 return -1;
2960
2961         ret = do_read_u64(ff, &last_sample_time);
2962         if (ret)
2963                 return -1;
2964
2965         session->evlist->first_sample_time = first_sample_time;
2966         session->evlist->last_sample_time = last_sample_time;
2967         return 0;
2968 }
2969
2970 static int process_mem_topology(struct feat_fd *ff,
2971                                 void *data __maybe_unused)
2972 {
2973         struct memory_node *nodes;
2974         u64 version, i, nr, bsize;
2975         int ret = -1;
2976
2977         if (do_read_u64(ff, &version))
2978                 return -1;
2979
2980         if (version != 1)
2981                 return -1;
2982
2983         if (do_read_u64(ff, &bsize))
2984                 return -1;
2985
2986         if (do_read_u64(ff, &nr))
2987                 return -1;
2988
2989         nodes = zalloc(sizeof(*nodes) * nr);
2990         if (!nodes)
2991                 return -1;
2992
2993         for (i = 0; i < nr; i++) {
2994                 struct memory_node n;
2995
2996                 #define _R(v)                           \
2997                         if (do_read_u64(ff, &n.v))      \
2998                                 goto out;               \
2999
3000                 _R(node)
3001                 _R(size)
3002
3003                 #undef _R
3004
3005                 if (do_read_bitmap(ff, &n.set, &n.size))
3006                         goto out;
3007
3008                 nodes[i] = n;
3009         }
3010
3011         ff->ph->env.memory_bsize    = bsize;
3012         ff->ph->env.memory_nodes    = nodes;
3013         ff->ph->env.nr_memory_nodes = nr;
3014         ret = 0;
3015
3016 out:
3017         if (ret)
3018                 free(nodes);
3019         return ret;
3020 }
3021
3022 static int process_clockid(struct feat_fd *ff,
3023                            void *data __maybe_unused)
3024 {
3025         if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
3026                 return -1;
3027
3028         return 0;
3029 }
3030
3031 static int process_clock_data(struct feat_fd *ff,
3032                               void *_data __maybe_unused)
3033 {
3034         u32 data32;
3035         u64 data64;
3036
3037         /* version */
3038         if (do_read_u32(ff, &data32))
3039                 return -1;
3040
3041         if (data32 != 1)
3042                 return -1;
3043
3044         /* clockid */
3045         if (do_read_u32(ff, &data32))
3046                 return -1;
3047
3048         ff->ph->env.clock.clockid = data32;
3049
3050         /* TOD ref time */
3051         if (do_read_u64(ff, &data64))
3052                 return -1;
3053
3054         ff->ph->env.clock.tod_ns = data64;
3055
3056         /* clockid ref time */
3057         if (do_read_u64(ff, &data64))
3058                 return -1;
3059
3060         ff->ph->env.clock.clockid_ns = data64;
3061         ff->ph->env.clock.enabled = true;
3062         return 0;
3063 }
3064
3065 static int process_hybrid_topology(struct feat_fd *ff,
3066                                    void *data __maybe_unused)
3067 {
3068         struct hybrid_node *nodes, *n;
3069         u32 nr, i;
3070
3071         /* nr nodes */
3072         if (do_read_u32(ff, &nr))
3073                 return -1;
3074
3075         nodes = zalloc(sizeof(*nodes) * nr);
3076         if (!nodes)
3077                 return -ENOMEM;
3078
3079         for (i = 0; i < nr; i++) {
3080                 n = &nodes[i];
3081
3082                 n->pmu_name = do_read_string(ff);
3083                 if (!n->pmu_name)
3084                         goto error;
3085
3086                 n->cpus = do_read_string(ff);
3087                 if (!n->cpus)
3088                         goto error;
3089         }
3090
3091         ff->ph->env.nr_hybrid_nodes = nr;
3092         ff->ph->env.hybrid_nodes = nodes;
3093         return 0;
3094
3095 error:
3096         for (i = 0; i < nr; i++) {
3097                 free(nodes[i].pmu_name);
3098                 free(nodes[i].cpus);
3099         }
3100
3101         free(nodes);
3102         return -1;
3103 }
3104
3105 static int process_dir_format(struct feat_fd *ff,
3106                               void *_data __maybe_unused)
3107 {
3108         struct perf_session *session;
3109         struct perf_data *data;
3110
3111         session = container_of(ff->ph, struct perf_session, header);
3112         data = session->data;
3113
3114         if (WARN_ON(!perf_data__is_dir(data)))
3115                 return -1;
3116
3117         return do_read_u64(ff, &data->dir.version);
3118 }
3119
3120 #ifdef HAVE_LIBBPF_SUPPORT
3121 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3122 {
3123         struct bpf_prog_info_node *info_node;
3124         struct perf_env *env = &ff->ph->env;
3125         struct perf_bpil *info_linear;
3126         u32 count, i;
3127         int err = -1;
3128
3129         if (ff->ph->needs_swap) {
3130                 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3131                 return 0;
3132         }
3133
3134         if (do_read_u32(ff, &count))
3135                 return -1;
3136
3137         down_write(&env->bpf_progs.lock);
3138
3139         for (i = 0; i < count; ++i) {
3140                 u32 info_len, data_len;
3141
3142                 info_linear = NULL;
3143                 info_node = NULL;
3144                 if (do_read_u32(ff, &info_len))
3145                         goto out;
3146                 if (do_read_u32(ff, &data_len))
3147                         goto out;
3148
3149                 if (info_len > sizeof(struct bpf_prog_info)) {
3150                         pr_warning("detected invalid bpf_prog_info\n");
3151                         goto out;
3152                 }
3153
3154                 info_linear = malloc(sizeof(struct perf_bpil) +
3155                                      data_len);
3156                 if (!info_linear)
3157                         goto out;
3158                 info_linear->info_len = sizeof(struct bpf_prog_info);
3159                 info_linear->data_len = data_len;
3160                 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3161                         goto out;
3162                 if (__do_read(ff, &info_linear->info, info_len))
3163                         goto out;
3164                 if (info_len < sizeof(struct bpf_prog_info))
3165                         memset(((void *)(&info_linear->info)) + info_len, 0,
3166                                sizeof(struct bpf_prog_info) - info_len);
3167
3168                 if (__do_read(ff, info_linear->data, data_len))
3169                         goto out;
3170
3171                 info_node = malloc(sizeof(struct bpf_prog_info_node));
3172                 if (!info_node)
3173                         goto out;
3174
3175                 /* after reading from file, translate offset to address */
3176                 bpil_offs_to_addr(info_linear);
3177                 info_node->info_linear = info_linear;
3178                 perf_env__insert_bpf_prog_info(env, info_node);
3179         }
3180
3181         up_write(&env->bpf_progs.lock);
3182         return 0;
3183 out:
3184         free(info_linear);
3185         free(info_node);
3186         up_write(&env->bpf_progs.lock);
3187         return err;
3188 }
3189
3190 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3191 {
3192         struct perf_env *env = &ff->ph->env;
3193         struct btf_node *node = NULL;
3194         u32 count, i;
3195         int err = -1;
3196
3197         if (ff->ph->needs_swap) {
3198                 pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3199                 return 0;
3200         }
3201
3202         if (do_read_u32(ff, &count))
3203                 return -1;
3204
3205         down_write(&env->bpf_progs.lock);
3206
3207         for (i = 0; i < count; ++i) {
3208                 u32 id, data_size;
3209
3210                 if (do_read_u32(ff, &id))
3211                         goto out;
3212                 if (do_read_u32(ff, &data_size))
3213                         goto out;
3214
3215                 node = malloc(sizeof(struct btf_node) + data_size);
3216                 if (!node)
3217                         goto out;
3218
3219                 node->id = id;
3220                 node->data_size = data_size;
3221
3222                 if (__do_read(ff, node->data, data_size))
3223                         goto out;
3224
3225                 perf_env__insert_btf(env, node);
3226                 node = NULL;
3227         }
3228
3229         err = 0;
3230 out:
3231         up_write(&env->bpf_progs.lock);
3232         free(node);
3233         return err;
3234 }
3235 #endif // HAVE_LIBBPF_SUPPORT
3236
3237 static int process_compressed(struct feat_fd *ff,
3238                               void *data __maybe_unused)
3239 {
3240         if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3241                 return -1;
3242
3243         if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3244                 return -1;
3245
3246         if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3247                 return -1;
3248
3249         if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3250                 return -1;
3251
3252         if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3253                 return -1;
3254
3255         return 0;
3256 }
3257
3258 static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
3259                               char ***caps, unsigned int *max_branches)
3260 {
3261         char *name, *value, *ptr;
3262         u32 nr_pmu_caps, i;
3263
3264         *nr_caps = 0;
3265         *caps = NULL;
3266
3267         if (do_read_u32(ff, &nr_pmu_caps))
3268                 return -1;
3269
3270         if (!nr_pmu_caps)
3271                 return 0;
3272
3273         *caps = zalloc(sizeof(char *) * nr_pmu_caps);
3274         if (!*caps)
3275                 return -1;
3276
3277         for (i = 0; i < nr_pmu_caps; i++) {
3278                 name = do_read_string(ff);
3279                 if (!name)
3280                         goto error;
3281
3282                 value = do_read_string(ff);
3283                 if (!value)
3284                         goto free_name;
3285
3286                 if (asprintf(&ptr, "%s=%s", name, value) < 0)
3287                         goto free_value;
3288
3289                 (*caps)[i] = ptr;
3290
3291                 if (!strcmp(name, "branches"))
3292                         *max_branches = atoi(value);
3293
3294                 free(value);
3295                 free(name);
3296         }
3297         *nr_caps = nr_pmu_caps;
3298         return 0;
3299
3300 free_value:
3301         free(value);
3302 free_name:
3303         free(name);
3304 error:
3305         for (; i > 0; i--)
3306                 free((*caps)[i - 1]);
3307         free(*caps);
3308         *caps = NULL;
3309         *nr_caps = 0;
3310         return -1;
3311 }
3312
3313 static int process_cpu_pmu_caps(struct feat_fd *ff,
3314                                 void *data __maybe_unused)
3315 {
3316         int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3317                                      &ff->ph->env.cpu_pmu_caps,
3318                                      &ff->ph->env.max_branches);
3319
3320         if (!ret && !ff->ph->env.cpu_pmu_caps)
3321                 pr_debug("cpu pmu capabilities not available\n");
3322         return ret;
3323 }
3324
3325 static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused)
3326 {
3327         struct pmu_caps *pmu_caps;
3328         u32 nr_pmu, i;
3329         int ret;
3330         int j;
3331
3332         if (do_read_u32(ff, &nr_pmu))
3333                 return -1;
3334
3335         if (!nr_pmu) {
3336                 pr_debug("pmu capabilities not available\n");
3337                 return 0;
3338         }
3339
3340         pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
3341         if (!pmu_caps)
3342                 return -ENOMEM;
3343
3344         for (i = 0; i < nr_pmu; i++) {
3345                 ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps,
3346                                          &pmu_caps[i].caps,
3347                                          &pmu_caps[i].max_branches);
3348                 if (ret)
3349                         goto err;
3350
3351                 pmu_caps[i].pmu_name = do_read_string(ff);
3352                 if (!pmu_caps[i].pmu_name) {
3353                         ret = -1;
3354                         goto err;
3355                 }
3356                 if (!pmu_caps[i].nr_caps) {
3357                         pr_debug("%s pmu capabilities not available\n",
3358                                  pmu_caps[i].pmu_name);
3359                 }
3360         }
3361
3362         ff->ph->env.nr_pmus_with_caps = nr_pmu;
3363         ff->ph->env.pmu_caps = pmu_caps;
3364         return 0;
3365
3366 err:
3367         for (i = 0; i < nr_pmu; i++) {
3368                 for (j = 0; j < pmu_caps[i].nr_caps; j++)
3369                         free(pmu_caps[i].caps[j]);
3370                 free(pmu_caps[i].caps);
3371                 free(pmu_caps[i].pmu_name);
3372         }
3373
3374         free(pmu_caps);
3375         return ret;
3376 }
3377
3378 #define FEAT_OPR(n, func, __full_only) \
3379         [HEADER_##n] = {                                        \
3380                 .name       = __stringify(n),                   \
3381                 .write      = write_##func,                     \
3382                 .print      = print_##func,                     \
3383                 .full_only  = __full_only,                      \
3384                 .process    = process_##func,                   \
3385                 .synthesize = true                              \
3386         }
3387
3388 #define FEAT_OPN(n, func, __full_only) \
3389         [HEADER_##n] = {                                        \
3390                 .name       = __stringify(n),                   \
3391                 .write      = write_##func,                     \
3392                 .print      = print_##func,                     \
3393                 .full_only  = __full_only,                      \
3394                 .process    = process_##func                    \
3395         }
3396
3397 /* feature_ops not implemented: */
3398 #define print_tracing_data      NULL
3399 #define print_build_id          NULL
3400
3401 #define process_branch_stack    NULL
3402 #define process_stat            NULL
3403
3404 // Only used in util/synthetic-events.c
3405 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3406
3407 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3408 #ifdef HAVE_LIBTRACEEVENT
3409         FEAT_OPN(TRACING_DATA,  tracing_data,   false),
3410 #endif
3411         FEAT_OPN(BUILD_ID,      build_id,       false),
3412         FEAT_OPR(HOSTNAME,      hostname,       false),
3413         FEAT_OPR(OSRELEASE,     osrelease,      false),
3414         FEAT_OPR(VERSION,       version,        false),
3415         FEAT_OPR(ARCH,          arch,           false),
3416         FEAT_OPR(NRCPUS,        nrcpus,         false),
3417         FEAT_OPR(CPUDESC,       cpudesc,        false),
3418         FEAT_OPR(CPUID,         cpuid,          false),
3419         FEAT_OPR(TOTAL_MEM,     total_mem,      false),
3420         FEAT_OPR(EVENT_DESC,    event_desc,     false),
3421         FEAT_OPR(CMDLINE,       cmdline,        false),
3422         FEAT_OPR(CPU_TOPOLOGY,  cpu_topology,   true),
3423         FEAT_OPR(NUMA_TOPOLOGY, numa_topology,  true),
3424         FEAT_OPN(BRANCH_STACK,  branch_stack,   false),
3425         FEAT_OPR(PMU_MAPPINGS,  pmu_mappings,   false),
3426         FEAT_OPR(GROUP_DESC,    group_desc,     false),
3427         FEAT_OPN(AUXTRACE,      auxtrace,       false),
3428         FEAT_OPN(STAT,          stat,           false),
3429         FEAT_OPN(CACHE,         cache,          true),
3430         FEAT_OPR(SAMPLE_TIME,   sample_time,    false),
3431         FEAT_OPR(MEM_TOPOLOGY,  mem_topology,   true),
3432         FEAT_OPR(CLOCKID,       clockid,        false),
3433         FEAT_OPN(DIR_FORMAT,    dir_format,     false),
3434 #ifdef HAVE_LIBBPF_SUPPORT
3435         FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
3436         FEAT_OPR(BPF_BTF,       bpf_btf,        false),
3437 #endif
3438         FEAT_OPR(COMPRESSED,    compressed,     false),
3439         FEAT_OPR(CPU_PMU_CAPS,  cpu_pmu_caps,   false),
3440         FEAT_OPR(CLOCK_DATA,    clock_data,     false),
3441         FEAT_OPN(HYBRID_TOPOLOGY,       hybrid_topology,        true),
3442         FEAT_OPR(PMU_CAPS,      pmu_caps,       false),
3443 };
3444
3445 struct header_print_data {
3446         FILE *fp;
3447         bool full; /* extended list of headers */
3448 };
3449
3450 static int perf_file_section__fprintf_info(struct perf_file_section *section,
3451                                            struct perf_header *ph,
3452                                            int feat, int fd, void *data)
3453 {
3454         struct header_print_data *hd = data;
3455         struct feat_fd ff;
3456
3457         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3458                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3459                                 "%d, continuing...\n", section->offset, feat);
3460                 return 0;
3461         }
3462         if (feat >= HEADER_LAST_FEATURE) {
3463                 pr_warning("unknown feature %d\n", feat);
3464                 return 0;
3465         }
3466         if (!feat_ops[feat].print)
3467                 return 0;
3468
3469         ff = (struct  feat_fd) {
3470                 .fd = fd,
3471                 .ph = ph,
3472         };
3473
3474         if (!feat_ops[feat].full_only || hd->full)
3475                 feat_ops[feat].print(&ff, hd->fp);
3476         else
3477                 fprintf(hd->fp, "# %s info available, use -I to display\n",
3478                         feat_ops[feat].name);
3479
3480         return 0;
3481 }
3482
3483 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3484 {
3485         struct header_print_data hd;
3486         struct perf_header *header = &session->header;
3487         int fd = perf_data__fd(session->data);
3488         struct stat st;
3489         time_t stctime;
3490         int ret, bit;
3491
3492         hd.fp = fp;
3493         hd.full = full;
3494
3495         ret = fstat(fd, &st);
3496         if (ret == -1)
3497                 return -1;
3498
3499         stctime = st.st_mtime;
3500         fprintf(fp, "# captured on    : %s", ctime(&stctime));
3501
3502         fprintf(fp, "# header version : %u\n", header->version);
3503         fprintf(fp, "# data offset    : %" PRIu64 "\n", header->data_offset);
3504         fprintf(fp, "# data size      : %" PRIu64 "\n", header->data_size);
3505         fprintf(fp, "# feat offset    : %" PRIu64 "\n", header->feat_offset);
3506
3507         perf_header__process_sections(header, fd, &hd,
3508                                       perf_file_section__fprintf_info);
3509
3510         if (session->data->is_pipe)
3511                 return 0;
3512
3513         fprintf(fp, "# missing features: ");
3514         for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3515                 if (bit)
3516                         fprintf(fp, "%s ", feat_ops[bit].name);
3517         }
3518
3519         fprintf(fp, "\n");
3520         return 0;
3521 }
3522
3523 struct header_fw {
3524         struct feat_writer      fw;
3525         struct feat_fd          *ff;
3526 };
3527
3528 static int feat_writer_cb(struct feat_writer *fw, void *buf, size_t sz)
3529 {
3530         struct header_fw *h = container_of(fw, struct header_fw, fw);
3531
3532         return do_write(h->ff, buf, sz);
3533 }
3534
3535 static int do_write_feat(struct feat_fd *ff, int type,
3536                          struct perf_file_section **p,
3537                          struct evlist *evlist,
3538                          struct feat_copier *fc)
3539 {
3540         int err;
3541         int ret = 0;
3542
3543         if (perf_header__has_feat(ff->ph, type)) {
3544                 if (!feat_ops[type].write)
3545                         return -1;
3546
3547                 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3548                         return -1;
3549
3550                 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3551
3552                 /*
3553                  * Hook to let perf inject copy features sections from the input
3554                  * file.
3555                  */
3556                 if (fc && fc->copy) {
3557                         struct header_fw h = {
3558                                 .fw.write = feat_writer_cb,
3559                                 .ff = ff,
3560                         };
3561
3562                         /* ->copy() returns 0 if the feature was not copied */
3563                         err = fc->copy(fc, type, &h.fw);
3564                 } else {
3565                         err = 0;
3566                 }
3567                 if (!err)
3568                         err = feat_ops[type].write(ff, evlist);
3569                 if (err < 0) {
3570                         pr_debug("failed to write feature %s\n", feat_ops[type].name);
3571
3572                         /* undo anything written */
3573                         lseek(ff->fd, (*p)->offset, SEEK_SET);
3574
3575                         return -1;
3576                 }
3577                 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3578                 (*p)++;
3579         }
3580         return ret;
3581 }
3582
3583 static int perf_header__adds_write(struct perf_header *header,
3584                                    struct evlist *evlist, int fd,
3585                                    struct feat_copier *fc)
3586 {
3587         int nr_sections;
3588         struct feat_fd ff;
3589         struct perf_file_section *feat_sec, *p;
3590         int sec_size;
3591         u64 sec_start;
3592         int feat;
3593         int err;
3594
3595         ff = (struct feat_fd){
3596                 .fd  = fd,
3597                 .ph = header,
3598         };
3599
3600         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3601         if (!nr_sections)
3602                 return 0;
3603
3604         feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3605         if (feat_sec == NULL)
3606                 return -ENOMEM;
3607
3608         sec_size = sizeof(*feat_sec) * nr_sections;
3609
3610         sec_start = header->feat_offset;
3611         lseek(fd, sec_start + sec_size, SEEK_SET);
3612
3613         for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3614                 if (do_write_feat(&ff, feat, &p, evlist, fc))
3615                         perf_header__clear_feat(header, feat);
3616         }
3617
3618         lseek(fd, sec_start, SEEK_SET);
3619         /*
3620          * may write more than needed due to dropped feature, but
3621          * this is okay, reader will skip the missing entries
3622          */
3623         err = do_write(&ff, feat_sec, sec_size);
3624         if (err < 0)
3625                 pr_debug("failed to write feature section\n");
3626         free(feat_sec);
3627         return err;
3628 }
3629
3630 int perf_header__write_pipe(int fd)
3631 {
3632         struct perf_pipe_file_header f_header;
3633         struct feat_fd ff;
3634         int err;
3635
3636         ff = (struct feat_fd){ .fd = fd };
3637
3638         f_header = (struct perf_pipe_file_header){
3639                 .magic     = PERF_MAGIC,
3640                 .size      = sizeof(f_header),
3641         };
3642
3643         err = do_write(&ff, &f_header, sizeof(f_header));
3644         if (err < 0) {
3645                 pr_debug("failed to write perf pipe header\n");
3646                 return err;
3647         }
3648
3649         return 0;
3650 }
3651
3652 static int perf_session__do_write_header(struct perf_session *session,
3653                                          struct evlist *evlist,
3654                                          int fd, bool at_exit,
3655                                          struct feat_copier *fc)
3656 {
3657         struct perf_file_header f_header;
3658         struct perf_file_attr   f_attr;
3659         struct perf_header *header = &session->header;
3660         struct evsel *evsel;
3661         struct feat_fd ff;
3662         u64 attr_offset;
3663         int err;
3664
3665         ff = (struct feat_fd){ .fd = fd};
3666         lseek(fd, sizeof(f_header), SEEK_SET);
3667
3668         evlist__for_each_entry(session->evlist, evsel) {
3669                 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3670                 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3671                 if (err < 0) {
3672                         pr_debug("failed to write perf header\n");
3673                         return err;
3674                 }
3675         }
3676
3677         attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3678
3679         evlist__for_each_entry(evlist, evsel) {
3680                 if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3681                         /*
3682                          * We are likely in "perf inject" and have read
3683                          * from an older file. Update attr size so that
3684                          * reader gets the right offset to the ids.
3685                          */
3686                         evsel->core.attr.size = sizeof(evsel->core.attr);
3687                 }
3688                 f_attr = (struct perf_file_attr){
3689                         .attr = evsel->core.attr,
3690                         .ids  = {
3691                                 .offset = evsel->id_offset,
3692                                 .size   = evsel->core.ids * sizeof(u64),
3693                         }
3694                 };
3695                 err = do_write(&ff, &f_attr, sizeof(f_attr));
3696                 if (err < 0) {
3697                         pr_debug("failed to write perf header attribute\n");
3698                         return err;
3699                 }
3700         }
3701
3702         if (!header->data_offset)
3703                 header->data_offset = lseek(fd, 0, SEEK_CUR);
3704         header->feat_offset = header->data_offset + header->data_size;
3705
3706         if (at_exit) {
3707                 err = perf_header__adds_write(header, evlist, fd, fc);
3708                 if (err < 0)
3709                         return err;
3710         }
3711
3712         f_header = (struct perf_file_header){
3713                 .magic     = PERF_MAGIC,
3714                 .size      = sizeof(f_header),
3715                 .attr_size = sizeof(f_attr),
3716                 .attrs = {
3717                         .offset = attr_offset,
3718                         .size   = evlist->core.nr_entries * sizeof(f_attr),
3719                 },
3720                 .data = {
3721                         .offset = header->data_offset,
3722                         .size   = header->data_size,
3723                 },
3724                 /* event_types is ignored, store zeros */
3725         };
3726
3727         memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3728
3729         lseek(fd, 0, SEEK_SET);
3730         err = do_write(&ff, &f_header, sizeof(f_header));
3731         if (err < 0) {
3732                 pr_debug("failed to write perf header\n");
3733                 return err;
3734         }
3735         lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3736
3737         return 0;
3738 }
3739
3740 int perf_session__write_header(struct perf_session *session,
3741                                struct evlist *evlist,
3742                                int fd, bool at_exit)
3743 {
3744         return perf_session__do_write_header(session, evlist, fd, at_exit, NULL);
3745 }
3746
3747 size_t perf_session__data_offset(const struct evlist *evlist)
3748 {
3749         struct evsel *evsel;
3750         size_t data_offset;
3751
3752         data_offset = sizeof(struct perf_file_header);
3753         evlist__for_each_entry(evlist, evsel) {
3754                 data_offset += evsel->core.ids * sizeof(u64);
3755         }
3756         data_offset += evlist->core.nr_entries * sizeof(struct perf_file_attr);
3757
3758         return data_offset;
3759 }
3760
3761 int perf_session__inject_header(struct perf_session *session,
3762                                 struct evlist *evlist,
3763                                 int fd,
3764                                 struct feat_copier *fc)
3765 {
3766         return perf_session__do_write_header(session, evlist, fd, true, fc);
3767 }
3768
3769 static int perf_header__getbuffer64(struct perf_header *header,
3770                                     int fd, void *buf, size_t size)
3771 {
3772         if (readn(fd, buf, size) <= 0)
3773                 return -1;
3774
3775         if (header->needs_swap)
3776                 mem_bswap_64(buf, size);
3777
3778         return 0;
3779 }
3780
3781 int perf_header__process_sections(struct perf_header *header, int fd,
3782                                   void *data,
3783                                   int (*process)(struct perf_file_section *section,
3784                                                  struct perf_header *ph,
3785                                                  int feat, int fd, void *data))
3786 {
3787         struct perf_file_section *feat_sec, *sec;
3788         int nr_sections;
3789         int sec_size;
3790         int feat;
3791         int err;
3792
3793         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3794         if (!nr_sections)
3795                 return 0;
3796
3797         feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3798         if (!feat_sec)
3799                 return -1;
3800
3801         sec_size = sizeof(*feat_sec) * nr_sections;
3802
3803         lseek(fd, header->feat_offset, SEEK_SET);
3804
3805         err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3806         if (err < 0)
3807                 goto out_free;
3808
3809         for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3810                 err = process(sec++, header, feat, fd, data);
3811                 if (err < 0)
3812                         goto out_free;
3813         }
3814         err = 0;
3815 out_free:
3816         free(feat_sec);
3817         return err;
3818 }
3819
3820 static const int attr_file_abi_sizes[] = {
3821         [0] = PERF_ATTR_SIZE_VER0,
3822         [1] = PERF_ATTR_SIZE_VER1,
3823         [2] = PERF_ATTR_SIZE_VER2,
3824         [3] = PERF_ATTR_SIZE_VER3,
3825         [4] = PERF_ATTR_SIZE_VER4,
3826         0,
3827 };
3828
3829 /*
3830  * In the legacy file format, the magic number is not used to encode endianness.
3831  * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3832  * on ABI revisions, we need to try all combinations for all endianness to
3833  * detect the endianness.
3834  */
3835 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3836 {
3837         uint64_t ref_size, attr_size;
3838         int i;
3839
3840         for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3841                 ref_size = attr_file_abi_sizes[i]
3842                          + sizeof(struct perf_file_section);
3843                 if (hdr_sz != ref_size) {
3844                         attr_size = bswap_64(hdr_sz);
3845                         if (attr_size != ref_size)
3846                                 continue;
3847
3848                         ph->needs_swap = true;
3849                 }
3850                 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3851                          i,
3852                          ph->needs_swap);
3853                 return 0;
3854         }
3855         /* could not determine endianness */
3856         return -1;
3857 }
3858
3859 #define PERF_PIPE_HDR_VER0      16
3860
3861 static const size_t attr_pipe_abi_sizes[] = {
3862         [0] = PERF_PIPE_HDR_VER0,
3863         0,
3864 };
3865
3866 /*
3867  * In the legacy pipe format, there is an implicit assumption that endianness
3868  * between host recording the samples, and host parsing the samples is the
3869  * same. This is not always the case given that the pipe output may always be
3870  * redirected into a file and analyzed on a different machine with possibly a
3871  * different endianness and perf_event ABI revisions in the perf tool itself.
3872  */
3873 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3874 {
3875         u64 attr_size;
3876         int i;
3877
3878         for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3879                 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3880                         attr_size = bswap_64(hdr_sz);
3881                         if (attr_size != hdr_sz)
3882                                 continue;
3883
3884                         ph->needs_swap = true;
3885                 }
3886                 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3887                 return 0;
3888         }
3889         return -1;
3890 }
3891
3892 bool is_perf_magic(u64 magic)
3893 {
3894         if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3895                 || magic == __perf_magic2
3896                 || magic == __perf_magic2_sw)
3897                 return true;
3898
3899         return false;
3900 }
3901
3902 static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3903                               bool is_pipe, struct perf_header *ph)
3904 {
3905         int ret;
3906
3907         /* check for legacy format */
3908         ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3909         if (ret == 0) {
3910                 ph->version = PERF_HEADER_VERSION_1;
3911                 pr_debug("legacy perf.data format\n");
3912                 if (is_pipe)
3913                         return try_all_pipe_abis(hdr_sz, ph);
3914
3915                 return try_all_file_abis(hdr_sz, ph);
3916         }
3917         /*
3918          * the new magic number serves two purposes:
3919          * - unique number to identify actual perf.data files
3920          * - encode endianness of file
3921          */
3922         ph->version = PERF_HEADER_VERSION_2;
3923
3924         /* check magic number with one endianness */
3925         if (magic == __perf_magic2)
3926                 return 0;
3927
3928         /* check magic number with opposite endianness */
3929         if (magic != __perf_magic2_sw)
3930                 return -1;
3931
3932         ph->needs_swap = true;
3933
3934         return 0;
3935 }
3936
3937 int perf_file_header__read(struct perf_file_header *header,
3938                            struct perf_header *ph, int fd)
3939 {
3940         ssize_t ret;
3941
3942         lseek(fd, 0, SEEK_SET);
3943
3944         ret = readn(fd, header, sizeof(*header));
3945         if (ret <= 0)
3946                 return -1;
3947
3948         if (check_magic_endian(header->magic,
3949                                header->attr_size, false, ph) < 0) {
3950                 pr_debug("magic/endian check failed\n");
3951                 return -1;
3952         }
3953
3954         if (ph->needs_swap) {
3955                 mem_bswap_64(header, offsetof(struct perf_file_header,
3956                              adds_features));
3957         }
3958
3959         if (header->size != sizeof(*header)) {
3960                 /* Support the previous format */
3961                 if (header->size == offsetof(typeof(*header), adds_features))
3962                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3963                 else
3964                         return -1;
3965         } else if (ph->needs_swap) {
3966                 /*
3967                  * feature bitmap is declared as an array of unsigned longs --
3968                  * not good since its size can differ between the host that
3969                  * generated the data file and the host analyzing the file.
3970                  *
3971                  * We need to handle endianness, but we don't know the size of
3972                  * the unsigned long where the file was generated. Take a best
3973                  * guess at determining it: try 64-bit swap first (ie., file
3974                  * created on a 64-bit host), and check if the hostname feature
3975                  * bit is set (this feature bit is forced on as of fbe96f2).
3976                  * If the bit is not, undo the 64-bit swap and try a 32-bit
3977                  * swap. If the hostname bit is still not set (e.g., older data
3978                  * file), punt and fallback to the original behavior --
3979                  * clearing all feature bits and setting buildid.
3980                  */
3981                 mem_bswap_64(&header->adds_features,
3982                             BITS_TO_U64(HEADER_FEAT_BITS));
3983
3984                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3985                         /* unswap as u64 */
3986                         mem_bswap_64(&header->adds_features,
3987                                     BITS_TO_U64(HEADER_FEAT_BITS));
3988
3989                         /* unswap as u32 */
3990                         mem_bswap_32(&header->adds_features,
3991                                     BITS_TO_U32(HEADER_FEAT_BITS));
3992                 }
3993
3994                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3995                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3996                         __set_bit(HEADER_BUILD_ID, header->adds_features);
3997                 }
3998         }
3999
4000         memcpy(&ph->adds_features, &header->adds_features,
4001                sizeof(ph->adds_features));
4002
4003         ph->data_offset  = header->data.offset;
4004         ph->data_size    = header->data.size;
4005         ph->feat_offset  = header->data.offset + header->data.size;
4006         return 0;
4007 }
4008
4009 static int perf_file_section__process(struct perf_file_section *section,
4010                                       struct perf_header *ph,
4011                                       int feat, int fd, void *data)
4012 {
4013         struct feat_fd fdd = {
4014                 .fd     = fd,
4015                 .ph     = ph,
4016                 .size   = section->size,
4017                 .offset = section->offset,
4018         };
4019
4020         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
4021                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
4022                           "%d, continuing...\n", section->offset, feat);
4023                 return 0;
4024         }
4025
4026         if (feat >= HEADER_LAST_FEATURE) {
4027                 pr_debug("unknown feature %d, continuing...\n", feat);
4028                 return 0;
4029         }
4030
4031         if (!feat_ops[feat].process)
4032                 return 0;
4033
4034         return feat_ops[feat].process(&fdd, data);
4035 }
4036
4037 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
4038                                        struct perf_header *ph,
4039                                        struct perf_data* data,
4040                                        bool repipe, int repipe_fd)
4041 {
4042         struct feat_fd ff = {
4043                 .fd = repipe_fd,
4044                 .ph = ph,
4045         };
4046         ssize_t ret;
4047
4048         ret = perf_data__read(data, header, sizeof(*header));
4049         if (ret <= 0)
4050                 return -1;
4051
4052         if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
4053                 pr_debug("endian/magic failed\n");
4054                 return -1;
4055         }
4056
4057         if (ph->needs_swap)
4058                 header->size = bswap_64(header->size);
4059
4060         if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
4061                 return -1;
4062
4063         return 0;
4064 }
4065
4066 static int perf_header__read_pipe(struct perf_session *session, int repipe_fd)
4067 {
4068         struct perf_header *header = &session->header;
4069         struct perf_pipe_file_header f_header;
4070
4071         if (perf_file_header__read_pipe(&f_header, header, session->data,
4072                                         session->repipe, repipe_fd) < 0) {
4073                 pr_debug("incompatible file format\n");
4074                 return -EINVAL;
4075         }
4076
4077         return f_header.size == sizeof(f_header) ? 0 : -1;
4078 }
4079
4080 static int read_attr(int fd, struct perf_header *ph,
4081                      struct perf_file_attr *f_attr)
4082 {
4083         struct perf_event_attr *attr = &f_attr->attr;
4084         size_t sz, left;
4085         size_t our_sz = sizeof(f_attr->attr);
4086         ssize_t ret;
4087
4088         memset(f_attr, 0, sizeof(*f_attr));
4089
4090         /* read minimal guaranteed structure */
4091         ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
4092         if (ret <= 0) {
4093                 pr_debug("cannot read %d bytes of header attr\n",
4094                          PERF_ATTR_SIZE_VER0);
4095                 return -1;
4096         }
4097
4098         /* on file perf_event_attr size */
4099         sz = attr->size;
4100
4101         if (ph->needs_swap)
4102                 sz = bswap_32(sz);
4103
4104         if (sz == 0) {
4105                 /* assume ABI0 */
4106                 sz =  PERF_ATTR_SIZE_VER0;
4107         } else if (sz > our_sz) {
4108                 pr_debug("file uses a more recent and unsupported ABI"
4109                          " (%zu bytes extra)\n", sz - our_sz);
4110                 return -1;
4111         }
4112         /* what we have not yet read and that we know about */
4113         left = sz - PERF_ATTR_SIZE_VER0;
4114         if (left) {
4115                 void *ptr = attr;
4116                 ptr += PERF_ATTR_SIZE_VER0;
4117
4118                 ret = readn(fd, ptr, left);
4119         }
4120         /* read perf_file_section, ids are read in caller */
4121         ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
4122
4123         return ret <= 0 ? -1 : 0;
4124 }
4125
4126 #ifdef HAVE_LIBTRACEEVENT
4127 static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
4128 {
4129         struct tep_event *event;
4130         char bf[128];
4131
4132         /* already prepared */
4133         if (evsel->tp_format)
4134                 return 0;
4135
4136         if (pevent == NULL) {
4137                 pr_debug("broken or missing trace data\n");
4138                 return -1;
4139         }
4140
4141         event = tep_find_event(pevent, evsel->core.attr.config);
4142         if (event == NULL) {
4143                 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
4144                 return -1;
4145         }
4146
4147         if (!evsel->name) {
4148                 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
4149                 evsel->name = strdup(bf);
4150                 if (evsel->name == NULL)
4151                         return -1;
4152         }
4153
4154         evsel->tp_format = event;
4155         return 0;
4156 }
4157
4158 static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
4159 {
4160         struct evsel *pos;
4161
4162         evlist__for_each_entry(evlist, pos) {
4163                 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
4164                     evsel__prepare_tracepoint_event(pos, pevent))
4165                         return -1;
4166         }
4167
4168         return 0;
4169 }
4170 #endif
4171
4172 int perf_session__read_header(struct perf_session *session, int repipe_fd)
4173 {
4174         struct perf_data *data = session->data;
4175         struct perf_header *header = &session->header;
4176         struct perf_file_header f_header;
4177         struct perf_file_attr   f_attr;
4178         u64                     f_id;
4179         int nr_attrs, nr_ids, i, j, err;
4180         int fd = perf_data__fd(data);
4181
4182         session->evlist = evlist__new();
4183         if (session->evlist == NULL)
4184                 return -ENOMEM;
4185
4186         session->evlist->env = &header->env;
4187         session->machines.host.env = &header->env;
4188
4189         /*
4190          * We can read 'pipe' data event from regular file,
4191          * check for the pipe header regardless of source.
4192          */
4193         err = perf_header__read_pipe(session, repipe_fd);
4194         if (!err || perf_data__is_pipe(data)) {
4195                 data->is_pipe = true;
4196                 return err;
4197         }
4198
4199         if (perf_file_header__read(&f_header, header, fd) < 0)
4200                 return -EINVAL;
4201
4202         if (header->needs_swap && data->in_place_update) {
4203                 pr_err("In-place update not supported when byte-swapping is required\n");
4204                 return -EINVAL;
4205         }
4206
4207         /*
4208          * Sanity check that perf.data was written cleanly; data size is
4209          * initialized to 0 and updated only if the on_exit function is run.
4210          * If data size is still 0 then the file contains only partial
4211          * information.  Just warn user and process it as much as it can.
4212          */
4213         if (f_header.data.size == 0) {
4214                 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4215                            "Was the 'perf record' command properly terminated?\n",
4216                            data->file.path);
4217         }
4218
4219         if (f_header.attr_size == 0) {
4220                 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4221                        "Was the 'perf record' command properly terminated?\n",
4222                        data->file.path);
4223                 return -EINVAL;
4224         }
4225
4226         nr_attrs = f_header.attrs.size / f_header.attr_size;
4227         lseek(fd, f_header.attrs.offset, SEEK_SET);
4228
4229         for (i = 0; i < nr_attrs; i++) {
4230                 struct evsel *evsel;
4231                 off_t tmp;
4232
4233                 if (read_attr(fd, header, &f_attr) < 0)
4234                         goto out_errno;
4235
4236                 if (header->needs_swap) {
4237                         f_attr.ids.size   = bswap_64(f_attr.ids.size);
4238                         f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4239                         perf_event__attr_swap(&f_attr.attr);
4240                 }
4241
4242                 tmp = lseek(fd, 0, SEEK_CUR);
4243                 evsel = evsel__new(&f_attr.attr);
4244
4245                 if (evsel == NULL)
4246                         goto out_delete_evlist;
4247
4248                 evsel->needs_swap = header->needs_swap;
4249                 /*
4250                  * Do it before so that if perf_evsel__alloc_id fails, this
4251                  * entry gets purged too at evlist__delete().
4252                  */
4253                 evlist__add(session->evlist, evsel);
4254
4255                 nr_ids = f_attr.ids.size / sizeof(u64);
4256                 /*
4257                  * We don't have the cpu and thread maps on the header, so
4258                  * for allocating the perf_sample_id table we fake 1 cpu and
4259                  * hattr->ids threads.
4260                  */
4261                 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4262                         goto out_delete_evlist;
4263
4264                 lseek(fd, f_attr.ids.offset, SEEK_SET);
4265
4266                 for (j = 0; j < nr_ids; j++) {
4267                         if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4268                                 goto out_errno;
4269
4270                         perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4271                 }
4272
4273                 lseek(fd, tmp, SEEK_SET);
4274         }
4275
4276 #ifdef HAVE_LIBTRACEEVENT
4277         perf_header__process_sections(header, fd, &session->tevent,
4278                                       perf_file_section__process);
4279
4280         if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4281                 goto out_delete_evlist;
4282 #else
4283         perf_header__process_sections(header, fd, NULL, perf_file_section__process);
4284 #endif
4285
4286         return 0;
4287 out_errno:
4288         return -errno;
4289
4290 out_delete_evlist:
4291         evlist__delete(session->evlist);
4292         session->evlist = NULL;
4293         return -ENOMEM;
4294 }
4295
4296 int perf_event__process_feature(struct perf_session *session,
4297                                 union perf_event *event)
4298 {
4299         struct perf_tool *tool = session->tool;
4300         struct feat_fd ff = { .fd = 0 };
4301         struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4302         int type = fe->header.type;
4303         u64 feat = fe->feat_id;
4304         int ret = 0;
4305
4306         if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4307                 pr_warning("invalid record type %d in pipe-mode\n", type);
4308                 return 0;
4309         }
4310         if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4311                 pr_warning("invalid record type %d in pipe-mode\n", type);
4312                 return -1;
4313         }
4314
4315         if (!feat_ops[feat].process)
4316                 return 0;
4317
4318         ff.buf  = (void *)fe->data;
4319         ff.size = event->header.size - sizeof(*fe);
4320         ff.ph = &session->header;
4321
4322         if (feat_ops[feat].process(&ff, NULL)) {
4323                 ret = -1;
4324                 goto out;
4325         }
4326
4327         if (!feat_ops[feat].print || !tool->show_feat_hdr)
4328                 goto out;
4329
4330         if (!feat_ops[feat].full_only ||
4331             tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4332                 feat_ops[feat].print(&ff, stdout);
4333         } else {
4334                 fprintf(stdout, "# %s info available, use -I to display\n",
4335                         feat_ops[feat].name);
4336         }
4337 out:
4338         free_event_desc(ff.events);
4339         return ret;
4340 }
4341
4342 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4343 {
4344         struct perf_record_event_update *ev = &event->event_update;
4345         struct perf_cpu_map *map;
4346         size_t ret;
4347
4348         ret = fprintf(fp, "\n... id:    %" PRI_lu64 "\n", ev->id);
4349
4350         switch (ev->type) {
4351         case PERF_EVENT_UPDATE__SCALE:
4352                 ret += fprintf(fp, "... scale: %f\n", ev->scale.scale);
4353                 break;
4354         case PERF_EVENT_UPDATE__UNIT:
4355                 ret += fprintf(fp, "... unit:  %s\n", ev->unit);
4356                 break;
4357         case PERF_EVENT_UPDATE__NAME:
4358                 ret += fprintf(fp, "... name:  %s\n", ev->name);
4359                 break;
4360         case PERF_EVENT_UPDATE__CPUS:
4361                 ret += fprintf(fp, "... ");
4362
4363                 map = cpu_map__new_data(&ev->cpus.cpus);
4364                 if (map)
4365                         ret += cpu_map__fprintf(map, fp);
4366                 else
4367                         ret += fprintf(fp, "failed to get cpus\n");
4368                 break;
4369         default:
4370                 ret += fprintf(fp, "... unknown type\n");
4371                 break;
4372         }
4373
4374         return ret;
4375 }
4376
4377 int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4378                              union perf_event *event,
4379                              struct evlist **pevlist)
4380 {
4381         u32 i, n_ids;
4382         u64 *ids;
4383         struct evsel *evsel;
4384         struct evlist *evlist = *pevlist;
4385
4386         if (evlist == NULL) {
4387                 *pevlist = evlist = evlist__new();
4388                 if (evlist == NULL)
4389                         return -ENOMEM;
4390         }
4391
4392         evsel = evsel__new(&event->attr.attr);
4393         if (evsel == NULL)
4394                 return -ENOMEM;
4395
4396         evlist__add(evlist, evsel);
4397
4398         n_ids = event->header.size - sizeof(event->header) - event->attr.attr.size;
4399         n_ids = n_ids / sizeof(u64);
4400         /*
4401          * We don't have the cpu and thread maps on the header, so
4402          * for allocating the perf_sample_id table we fake 1 cpu and
4403          * hattr->ids threads.
4404          */
4405         if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4406                 return -ENOMEM;
4407
4408         ids = perf_record_header_attr_id(event);
4409         for (i = 0; i < n_ids; i++) {
4410                 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, ids[i]);
4411         }
4412
4413         return 0;
4414 }
4415
4416 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4417                                      union perf_event *event,
4418                                      struct evlist **pevlist)
4419 {
4420         struct perf_record_event_update *ev = &event->event_update;
4421         struct evlist *evlist;
4422         struct evsel *evsel;
4423         struct perf_cpu_map *map;
4424
4425         if (dump_trace)
4426                 perf_event__fprintf_event_update(event, stdout);
4427
4428         if (!pevlist || *pevlist == NULL)
4429                 return -EINVAL;
4430
4431         evlist = *pevlist;
4432
4433         evsel = evlist__id2evsel(evlist, ev->id);
4434         if (evsel == NULL)
4435                 return -EINVAL;
4436
4437         switch (ev->type) {
4438         case PERF_EVENT_UPDATE__UNIT:
4439                 free((char *)evsel->unit);
4440                 evsel->unit = strdup(ev->unit);
4441                 break;
4442         case PERF_EVENT_UPDATE__NAME:
4443                 free(evsel->name);
4444                 evsel->name = strdup(ev->name);
4445                 break;
4446         case PERF_EVENT_UPDATE__SCALE:
4447                 evsel->scale = ev->scale.scale;
4448                 break;
4449         case PERF_EVENT_UPDATE__CPUS:
4450                 map = cpu_map__new_data(&ev->cpus.cpus);
4451                 if (map) {
4452                         perf_cpu_map__put(evsel->core.own_cpus);
4453                         evsel->core.own_cpus = map;
4454                 } else
4455                         pr_err("failed to get event_update cpus\n");
4456         default:
4457                 break;
4458         }
4459
4460         return 0;
4461 }
4462
4463 #ifdef HAVE_LIBTRACEEVENT
4464 int perf_event__process_tracing_data(struct perf_session *session,
4465                                      union perf_event *event)
4466 {
4467         ssize_t size_read, padding, size = event->tracing_data.size;
4468         int fd = perf_data__fd(session->data);
4469         char buf[BUFSIZ];
4470
4471         /*
4472          * The pipe fd is already in proper place and in any case
4473          * we can't move it, and we'd screw the case where we read
4474          * 'pipe' data from regular file. The trace_report reads
4475          * data from 'fd' so we need to set it directly behind the
4476          * event, where the tracing data starts.
4477          */
4478         if (!perf_data__is_pipe(session->data)) {
4479                 off_t offset = lseek(fd, 0, SEEK_CUR);
4480
4481                 /* setup for reading amidst mmap */
4482                 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4483                       SEEK_SET);
4484         }
4485
4486         size_read = trace_report(fd, &session->tevent,
4487                                  session->repipe);
4488         padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4489
4490         if (readn(fd, buf, padding) < 0) {
4491                 pr_err("%s: reading input file", __func__);
4492                 return -1;
4493         }
4494         if (session->repipe) {
4495                 int retw = write(STDOUT_FILENO, buf, padding);
4496                 if (retw <= 0 || retw != padding) {
4497                         pr_err("%s: repiping tracing data padding", __func__);
4498                         return -1;
4499                 }
4500         }
4501
4502         if (size_read + padding != size) {
4503                 pr_err("%s: tracing data size mismatch", __func__);
4504                 return -1;
4505         }
4506
4507         evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4508
4509         return size_read + padding;
4510 }
4511 #endif
4512
4513 int perf_event__process_build_id(struct perf_session *session,
4514                                  union perf_event *event)
4515 {
4516         __event_process_build_id(&event->build_id,
4517                                  event->build_id.filename,
4518                                  session);
4519         return 0;
4520 }