GNU Linux-libre 5.19-rc6-gnu
[releases.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 <stdio.h>
10 #include <stdlib.h>
11 #include <linux/compiler.h>
12 #include <linux/list.h>
13 #include <linux/kernel.h>
14 #include <linux/bitops.h>
15 #include <linux/string.h>
16 #include <linux/stringify.h>
17 #include <linux/zalloc.h>
18 #include <sys/stat.h>
19 #include <sys/utsname.h>
20 #include <linux/time64.h>
21 #include <dirent.h>
22 #ifdef HAVE_LIBBPF_SUPPORT
23 #include <bpf/libbpf.h>
24 #endif
25 #include <perf/cpumap.h>
26
27 #include "dso.h"
28 #include "evlist.h"
29 #include "evsel.h"
30 #include "util/evsel_fprintf.h"
31 #include "header.h"
32 #include "memswap.h"
33 #include "trace-event.h"
34 #include "session.h"
35 #include "symbol.h"
36 #include "debug.h"
37 #include "cpumap.h"
38 #include "pmu.h"
39 #include "vdso.h"
40 #include "strbuf.h"
41 #include "build-id.h"
42 #include "data.h"
43 #include <api/fs/fs.h>
44 #include "asm/bug.h"
45 #include "tool.h"
46 #include "time-utils.h"
47 #include "units.h"
48 #include "util/util.h" // perf_exe()
49 #include "cputopo.h"
50 #include "bpf-event.h"
51 #include "bpf-utils.h"
52 #include "clockid.h"
53 #include "pmu-hybrid.h"
54
55 #include <linux/ctype.h>
56 #include <internal/lib.h>
57
58 /*
59  * magic2 = "PERFILE2"
60  * must be a numerical value to let the endianness
61  * determine the memory layout. That way we are able
62  * to detect endianness when reading the perf.data file
63  * back.
64  *
65  * we check for legacy (PERFFILE) format.
66  */
67 static const char *__perf_magic1 = "PERFFILE";
68 static const u64 __perf_magic2    = 0x32454c4946524550ULL;
69 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
70
71 #define PERF_MAGIC      __perf_magic2
72
73 const char perf_version_string[] = PERF_VERSION;
74
75 struct perf_file_attr {
76         struct perf_event_attr  attr;
77         struct perf_file_section        ids;
78 };
79
80 void perf_header__set_feat(struct perf_header *header, int feat)
81 {
82         set_bit(feat, header->adds_features);
83 }
84
85 void perf_header__clear_feat(struct perf_header *header, int feat)
86 {
87         clear_bit(feat, header->adds_features);
88 }
89
90 bool perf_header__has_feat(const struct perf_header *header, int feat)
91 {
92         return test_bit(feat, header->adds_features);
93 }
94
95 static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
96 {
97         ssize_t ret = writen(ff->fd, buf, size);
98
99         if (ret != (ssize_t)size)
100                 return ret < 0 ? (int)ret : -1;
101         return 0;
102 }
103
104 static int __do_write_buf(struct feat_fd *ff,  const void *buf, size_t size)
105 {
106         /* struct perf_event_header::size is u16 */
107         const size_t max_size = 0xffff - sizeof(struct perf_event_header);
108         size_t new_size = ff->size;
109         void *addr;
110
111         if (size + ff->offset > max_size)
112                 return -E2BIG;
113
114         while (size > (new_size - ff->offset))
115                 new_size <<= 1;
116         new_size = min(max_size, new_size);
117
118         if (ff->size < new_size) {
119                 addr = realloc(ff->buf, new_size);
120                 if (!addr)
121                         return -ENOMEM;
122                 ff->buf = addr;
123                 ff->size = new_size;
124         }
125
126         memcpy(ff->buf + ff->offset, buf, size);
127         ff->offset += size;
128
129         return 0;
130 }
131
132 /* Return: 0 if succeeded, -ERR if failed. */
133 int do_write(struct feat_fd *ff, const void *buf, size_t size)
134 {
135         if (!ff->buf)
136                 return __do_write_fd(ff, buf, size);
137         return __do_write_buf(ff, buf, size);
138 }
139
140 /* Return: 0 if succeeded, -ERR if failed. */
141 static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
142 {
143         u64 *p = (u64 *) set;
144         int i, ret;
145
146         ret = do_write(ff, &size, sizeof(size));
147         if (ret < 0)
148                 return ret;
149
150         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
151                 ret = do_write(ff, p + i, sizeof(*p));
152                 if (ret < 0)
153                         return ret;
154         }
155
156         return 0;
157 }
158
159 /* Return: 0 if succeeded, -ERR if failed. */
160 int write_padded(struct feat_fd *ff, const void *bf,
161                  size_t count, size_t count_aligned)
162 {
163         static const char zero_buf[NAME_ALIGN];
164         int err = do_write(ff, bf, count);
165
166         if (!err)
167                 err = do_write(ff, zero_buf, count_aligned - count);
168
169         return err;
170 }
171
172 #define string_size(str)                                                \
173         (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
174
175 /* Return: 0 if succeeded, -ERR if failed. */
176 static int do_write_string(struct feat_fd *ff, const char *str)
177 {
178         u32 len, olen;
179         int ret;
180
181         olen = strlen(str) + 1;
182         len = PERF_ALIGN(olen, NAME_ALIGN);
183
184         /* write len, incl. \0 */
185         ret = do_write(ff, &len, sizeof(len));
186         if (ret < 0)
187                 return ret;
188
189         return write_padded(ff, str, olen, len);
190 }
191
192 static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
193 {
194         ssize_t ret = readn(ff->fd, addr, size);
195
196         if (ret != size)
197                 return ret < 0 ? (int)ret : -1;
198         return 0;
199 }
200
201 static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
202 {
203         if (size > (ssize_t)ff->size - ff->offset)
204                 return -1;
205
206         memcpy(addr, ff->buf + ff->offset, size);
207         ff->offset += size;
208
209         return 0;
210
211 }
212
213 static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
214 {
215         if (!ff->buf)
216                 return __do_read_fd(ff, addr, size);
217         return __do_read_buf(ff, addr, size);
218 }
219
220 static int do_read_u32(struct feat_fd *ff, u32 *addr)
221 {
222         int ret;
223
224         ret = __do_read(ff, addr, sizeof(*addr));
225         if (ret)
226                 return ret;
227
228         if (ff->ph->needs_swap)
229                 *addr = bswap_32(*addr);
230         return 0;
231 }
232
233 static int do_read_u64(struct feat_fd *ff, u64 *addr)
234 {
235         int ret;
236
237         ret = __do_read(ff, addr, sizeof(*addr));
238         if (ret)
239                 return ret;
240
241         if (ff->ph->needs_swap)
242                 *addr = bswap_64(*addr);
243         return 0;
244 }
245
246 static char *do_read_string(struct feat_fd *ff)
247 {
248         u32 len;
249         char *buf;
250
251         if (do_read_u32(ff, &len))
252                 return NULL;
253
254         buf = malloc(len);
255         if (!buf)
256                 return NULL;
257
258         if (!__do_read(ff, buf, len)) {
259                 /*
260                  * strings are padded by zeroes
261                  * thus the actual strlen of buf
262                  * may be less than len
263                  */
264                 return buf;
265         }
266
267         free(buf);
268         return NULL;
269 }
270
271 /* Return: 0 if succeeded, -ERR if failed. */
272 static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
273 {
274         unsigned long *set;
275         u64 size, *p;
276         int i, ret;
277
278         ret = do_read_u64(ff, &size);
279         if (ret)
280                 return ret;
281
282         set = bitmap_zalloc(size);
283         if (!set)
284                 return -ENOMEM;
285
286         p = (u64 *) set;
287
288         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
289                 ret = do_read_u64(ff, p + i);
290                 if (ret < 0) {
291                         free(set);
292                         return ret;
293                 }
294         }
295
296         *pset  = set;
297         *psize = size;
298         return 0;
299 }
300
301 static int write_tracing_data(struct feat_fd *ff,
302                               struct evlist *evlist)
303 {
304         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
305                 return -1;
306
307         return read_tracing_data(ff->fd, &evlist->core.entries);
308 }
309
310 static int write_build_id(struct feat_fd *ff,
311                           struct evlist *evlist __maybe_unused)
312 {
313         struct perf_session *session;
314         int err;
315
316         session = container_of(ff->ph, struct perf_session, header);
317
318         if (!perf_session__read_build_ids(session, true))
319                 return -1;
320
321         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
322                 return -1;
323
324         err = perf_session__write_buildid_table(session, ff);
325         if (err < 0) {
326                 pr_debug("failed to write buildid table\n");
327                 return err;
328         }
329         perf_session__cache_build_ids(session);
330
331         return 0;
332 }
333
334 static int write_hostname(struct feat_fd *ff,
335                           struct evlist *evlist __maybe_unused)
336 {
337         struct utsname uts;
338         int ret;
339
340         ret = uname(&uts);
341         if (ret < 0)
342                 return -1;
343
344         return do_write_string(ff, uts.nodename);
345 }
346
347 static int write_osrelease(struct feat_fd *ff,
348                            struct evlist *evlist __maybe_unused)
349 {
350         struct utsname uts;
351         int ret;
352
353         ret = uname(&uts);
354         if (ret < 0)
355                 return -1;
356
357         return do_write_string(ff, uts.release);
358 }
359
360 static int write_arch(struct feat_fd *ff,
361                       struct evlist *evlist __maybe_unused)
362 {
363         struct utsname uts;
364         int ret;
365
366         ret = uname(&uts);
367         if (ret < 0)
368                 return -1;
369
370         return do_write_string(ff, uts.machine);
371 }
372
373 static int write_version(struct feat_fd *ff,
374                          struct evlist *evlist __maybe_unused)
375 {
376         return do_write_string(ff, perf_version_string);
377 }
378
379 static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
380 {
381         FILE *file;
382         char *buf = NULL;
383         char *s, *p;
384         const char *search = cpuinfo_proc;
385         size_t len = 0;
386         int ret = -1;
387
388         if (!search)
389                 return -1;
390
391         file = fopen("/proc/cpuinfo", "r");
392         if (!file)
393                 return -1;
394
395         while (getline(&buf, &len, file) > 0) {
396                 ret = strncmp(buf, search, strlen(search));
397                 if (!ret)
398                         break;
399         }
400
401         if (ret) {
402                 ret = -1;
403                 goto done;
404         }
405
406         s = buf;
407
408         p = strchr(buf, ':');
409         if (p && *(p+1) == ' ' && *(p+2))
410                 s = p + 2;
411         p = strchr(s, '\n');
412         if (p)
413                 *p = '\0';
414
415         /* squash extra space characters (branding string) */
416         p = s;
417         while (*p) {
418                 if (isspace(*p)) {
419                         char *r = p + 1;
420                         char *q = skip_spaces(r);
421                         *p = ' ';
422                         if (q != (p+1))
423                                 while ((*r++ = *q++));
424                 }
425                 p++;
426         }
427         ret = do_write_string(ff, s);
428 done:
429         free(buf);
430         fclose(file);
431         return ret;
432 }
433
434 static int write_cpudesc(struct feat_fd *ff,
435                        struct evlist *evlist __maybe_unused)
436 {
437 #if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
438 #define CPUINFO_PROC    { "cpu", }
439 #elif defined(__s390__)
440 #define CPUINFO_PROC    { "vendor_id", }
441 #elif defined(__sh__)
442 #define CPUINFO_PROC    { "cpu type", }
443 #elif defined(__alpha__) || defined(__mips__)
444 #define CPUINFO_PROC    { "cpu model", }
445 #elif defined(__arm__)
446 #define CPUINFO_PROC    { "model name", "Processor", }
447 #elif defined(__arc__)
448 #define CPUINFO_PROC    { "Processor", }
449 #elif defined(__xtensa__)
450 #define CPUINFO_PROC    { "core ID", }
451 #else
452 #define CPUINFO_PROC    { "model name", }
453 #endif
454         const char *cpuinfo_procs[] = CPUINFO_PROC;
455 #undef CPUINFO_PROC
456         unsigned int i;
457
458         for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
459                 int ret;
460                 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
461                 if (ret >= 0)
462                         return ret;
463         }
464         return -1;
465 }
466
467
468 static int write_nrcpus(struct feat_fd *ff,
469                         struct evlist *evlist __maybe_unused)
470 {
471         long nr;
472         u32 nrc, nra;
473         int ret;
474
475         nrc = cpu__max_present_cpu().cpu;
476
477         nr = sysconf(_SC_NPROCESSORS_ONLN);
478         if (nr < 0)
479                 return -1;
480
481         nra = (u32)(nr & UINT_MAX);
482
483         ret = do_write(ff, &nrc, sizeof(nrc));
484         if (ret < 0)
485                 return ret;
486
487         return do_write(ff, &nra, sizeof(nra));
488 }
489
490 static int write_event_desc(struct feat_fd *ff,
491                             struct evlist *evlist)
492 {
493         struct evsel *evsel;
494         u32 nre, nri, sz;
495         int ret;
496
497         nre = evlist->core.nr_entries;
498
499         /*
500          * write number of events
501          */
502         ret = do_write(ff, &nre, sizeof(nre));
503         if (ret < 0)
504                 return ret;
505
506         /*
507          * size of perf_event_attr struct
508          */
509         sz = (u32)sizeof(evsel->core.attr);
510         ret = do_write(ff, &sz, sizeof(sz));
511         if (ret < 0)
512                 return ret;
513
514         evlist__for_each_entry(evlist, evsel) {
515                 ret = do_write(ff, &evsel->core.attr, sz);
516                 if (ret < 0)
517                         return ret;
518                 /*
519                  * write number of unique id per event
520                  * there is one id per instance of an event
521                  *
522                  * copy into an nri to be independent of the
523                  * type of ids,
524                  */
525                 nri = evsel->core.ids;
526                 ret = do_write(ff, &nri, sizeof(nri));
527                 if (ret < 0)
528                         return ret;
529
530                 /*
531                  * write event string as passed on cmdline
532                  */
533                 ret = do_write_string(ff, evsel__name(evsel));
534                 if (ret < 0)
535                         return ret;
536                 /*
537                  * write unique ids for this event
538                  */
539                 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
540                 if (ret < 0)
541                         return ret;
542         }
543         return 0;
544 }
545
546 static int write_cmdline(struct feat_fd *ff,
547                          struct evlist *evlist __maybe_unused)
548 {
549         char pbuf[MAXPATHLEN], *buf;
550         int i, ret, n;
551
552         /* actual path to perf binary */
553         buf = perf_exe(pbuf, MAXPATHLEN);
554
555         /* account for binary path */
556         n = perf_env.nr_cmdline + 1;
557
558         ret = do_write(ff, &n, sizeof(n));
559         if (ret < 0)
560                 return ret;
561
562         ret = do_write_string(ff, buf);
563         if (ret < 0)
564                 return ret;
565
566         for (i = 0 ; i < perf_env.nr_cmdline; i++) {
567                 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
568                 if (ret < 0)
569                         return ret;
570         }
571         return 0;
572 }
573
574
575 static int write_cpu_topology(struct feat_fd *ff,
576                               struct evlist *evlist __maybe_unused)
577 {
578         struct cpu_topology *tp;
579         u32 i;
580         int ret, j;
581
582         tp = cpu_topology__new();
583         if (!tp)
584                 return -1;
585
586         ret = do_write(ff, &tp->package_cpus_lists, sizeof(tp->package_cpus_lists));
587         if (ret < 0)
588                 goto done;
589
590         for (i = 0; i < tp->package_cpus_lists; i++) {
591                 ret = do_write_string(ff, tp->package_cpus_list[i]);
592                 if (ret < 0)
593                         goto done;
594         }
595         ret = do_write(ff, &tp->core_cpus_lists, sizeof(tp->core_cpus_lists));
596         if (ret < 0)
597                 goto done;
598
599         for (i = 0; i < tp->core_cpus_lists; i++) {
600                 ret = do_write_string(ff, tp->core_cpus_list[i]);
601                 if (ret < 0)
602                         break;
603         }
604
605         ret = perf_env__read_cpu_topology_map(&perf_env);
606         if (ret < 0)
607                 goto done;
608
609         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
610                 ret = do_write(ff, &perf_env.cpu[j].core_id,
611                                sizeof(perf_env.cpu[j].core_id));
612                 if (ret < 0)
613                         return ret;
614                 ret = do_write(ff, &perf_env.cpu[j].socket_id,
615                                sizeof(perf_env.cpu[j].socket_id));
616                 if (ret < 0)
617                         return ret;
618         }
619
620         if (!tp->die_cpus_lists)
621                 goto done;
622
623         ret = do_write(ff, &tp->die_cpus_lists, sizeof(tp->die_cpus_lists));
624         if (ret < 0)
625                 goto done;
626
627         for (i = 0; i < tp->die_cpus_lists; i++) {
628                 ret = do_write_string(ff, tp->die_cpus_list[i]);
629                 if (ret < 0)
630                         goto done;
631         }
632
633         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
634                 ret = do_write(ff, &perf_env.cpu[j].die_id,
635                                sizeof(perf_env.cpu[j].die_id));
636                 if (ret < 0)
637                         return ret;
638         }
639
640 done:
641         cpu_topology__delete(tp);
642         return ret;
643 }
644
645
646
647 static int write_total_mem(struct feat_fd *ff,
648                            struct evlist *evlist __maybe_unused)
649 {
650         char *buf = NULL;
651         FILE *fp;
652         size_t len = 0;
653         int ret = -1, n;
654         uint64_t mem;
655
656         fp = fopen("/proc/meminfo", "r");
657         if (!fp)
658                 return -1;
659
660         while (getline(&buf, &len, fp) > 0) {
661                 ret = strncmp(buf, "MemTotal:", 9);
662                 if (!ret)
663                         break;
664         }
665         if (!ret) {
666                 n = sscanf(buf, "%*s %"PRIu64, &mem);
667                 if (n == 1)
668                         ret = do_write(ff, &mem, sizeof(mem));
669         } else
670                 ret = -1;
671         free(buf);
672         fclose(fp);
673         return ret;
674 }
675
676 static int write_numa_topology(struct feat_fd *ff,
677                                struct evlist *evlist __maybe_unused)
678 {
679         struct numa_topology *tp;
680         int ret = -1;
681         u32 i;
682
683         tp = numa_topology__new();
684         if (!tp)
685                 return -ENOMEM;
686
687         ret = do_write(ff, &tp->nr, sizeof(u32));
688         if (ret < 0)
689                 goto err;
690
691         for (i = 0; i < tp->nr; i++) {
692                 struct numa_topology_node *n = &tp->nodes[i];
693
694                 ret = do_write(ff, &n->node, sizeof(u32));
695                 if (ret < 0)
696                         goto err;
697
698                 ret = do_write(ff, &n->mem_total, sizeof(u64));
699                 if (ret)
700                         goto err;
701
702                 ret = do_write(ff, &n->mem_free, sizeof(u64));
703                 if (ret)
704                         goto err;
705
706                 ret = do_write_string(ff, n->cpus);
707                 if (ret < 0)
708                         goto err;
709         }
710
711         ret = 0;
712
713 err:
714         numa_topology__delete(tp);
715         return ret;
716 }
717
718 /*
719  * File format:
720  *
721  * struct pmu_mappings {
722  *      u32     pmu_num;
723  *      struct pmu_map {
724  *              u32     type;
725  *              char    name[];
726  *      }[pmu_num];
727  * };
728  */
729
730 static int write_pmu_mappings(struct feat_fd *ff,
731                               struct evlist *evlist __maybe_unused)
732 {
733         struct perf_pmu *pmu = NULL;
734         u32 pmu_num = 0;
735         int ret;
736
737         /*
738          * Do a first pass to count number of pmu to avoid lseek so this
739          * works in pipe mode as well.
740          */
741         while ((pmu = perf_pmu__scan(pmu))) {
742                 if (!pmu->name)
743                         continue;
744                 pmu_num++;
745         }
746
747         ret = do_write(ff, &pmu_num, sizeof(pmu_num));
748         if (ret < 0)
749                 return ret;
750
751         while ((pmu = perf_pmu__scan(pmu))) {
752                 if (!pmu->name)
753                         continue;
754
755                 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
756                 if (ret < 0)
757                         return ret;
758
759                 ret = do_write_string(ff, pmu->name);
760                 if (ret < 0)
761                         return ret;
762         }
763
764         return 0;
765 }
766
767 /*
768  * File format:
769  *
770  * struct group_descs {
771  *      u32     nr_groups;
772  *      struct group_desc {
773  *              char    name[];
774  *              u32     leader_idx;
775  *              u32     nr_members;
776  *      }[nr_groups];
777  * };
778  */
779 static int write_group_desc(struct feat_fd *ff,
780                             struct evlist *evlist)
781 {
782         u32 nr_groups = evlist->core.nr_groups;
783         struct evsel *evsel;
784         int ret;
785
786         ret = do_write(ff, &nr_groups, sizeof(nr_groups));
787         if (ret < 0)
788                 return ret;
789
790         evlist__for_each_entry(evlist, evsel) {
791                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
792                         const char *name = evsel->group_name ?: "{anon_group}";
793                         u32 leader_idx = evsel->core.idx;
794                         u32 nr_members = evsel->core.nr_members;
795
796                         ret = do_write_string(ff, name);
797                         if (ret < 0)
798                                 return ret;
799
800                         ret = do_write(ff, &leader_idx, sizeof(leader_idx));
801                         if (ret < 0)
802                                 return ret;
803
804                         ret = do_write(ff, &nr_members, sizeof(nr_members));
805                         if (ret < 0)
806                                 return ret;
807                 }
808         }
809         return 0;
810 }
811
812 /*
813  * Return the CPU id as a raw string.
814  *
815  * Each architecture should provide a more precise id string that
816  * can be use to match the architecture's "mapfile".
817  */
818 char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
819 {
820         return NULL;
821 }
822
823 /* Return zero when the cpuid from the mapfile.csv matches the
824  * cpuid string generated on this platform.
825  * Otherwise return non-zero.
826  */
827 int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
828 {
829         regex_t re;
830         regmatch_t pmatch[1];
831         int match;
832
833         if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
834                 /* Warn unable to generate match particular string. */
835                 pr_info("Invalid regular expression %s\n", mapcpuid);
836                 return 1;
837         }
838
839         match = !regexec(&re, cpuid, 1, pmatch, 0);
840         regfree(&re);
841         if (match) {
842                 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
843
844                 /* Verify the entire string matched. */
845                 if (match_len == strlen(cpuid))
846                         return 0;
847         }
848         return 1;
849 }
850
851 /*
852  * default get_cpuid(): nothing gets recorded
853  * actual implementation must be in arch/$(SRCARCH)/util/header.c
854  */
855 int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
856 {
857         return ENOSYS; /* Not implemented */
858 }
859
860 static int write_cpuid(struct feat_fd *ff,
861                        struct evlist *evlist __maybe_unused)
862 {
863         char buffer[64];
864         int ret;
865
866         ret = get_cpuid(buffer, sizeof(buffer));
867         if (ret)
868                 return -1;
869
870         return do_write_string(ff, buffer);
871 }
872
873 static int write_branch_stack(struct feat_fd *ff __maybe_unused,
874                               struct evlist *evlist __maybe_unused)
875 {
876         return 0;
877 }
878
879 static int write_auxtrace(struct feat_fd *ff,
880                           struct evlist *evlist __maybe_unused)
881 {
882         struct perf_session *session;
883         int err;
884
885         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
886                 return -1;
887
888         session = container_of(ff->ph, struct perf_session, header);
889
890         err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
891         if (err < 0)
892                 pr_err("Failed to write auxtrace index\n");
893         return err;
894 }
895
896 static int write_clockid(struct feat_fd *ff,
897                          struct evlist *evlist __maybe_unused)
898 {
899         return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
900                         sizeof(ff->ph->env.clock.clockid_res_ns));
901 }
902
903 static int write_clock_data(struct feat_fd *ff,
904                             struct evlist *evlist __maybe_unused)
905 {
906         u64 *data64;
907         u32 data32;
908         int ret;
909
910         /* version */
911         data32 = 1;
912
913         ret = do_write(ff, &data32, sizeof(data32));
914         if (ret < 0)
915                 return ret;
916
917         /* clockid */
918         data32 = ff->ph->env.clock.clockid;
919
920         ret = do_write(ff, &data32, sizeof(data32));
921         if (ret < 0)
922                 return ret;
923
924         /* TOD ref time */
925         data64 = &ff->ph->env.clock.tod_ns;
926
927         ret = do_write(ff, data64, sizeof(*data64));
928         if (ret < 0)
929                 return ret;
930
931         /* clockid ref time */
932         data64 = &ff->ph->env.clock.clockid_ns;
933
934         return do_write(ff, data64, sizeof(*data64));
935 }
936
937 static int write_hybrid_topology(struct feat_fd *ff,
938                                  struct evlist *evlist __maybe_unused)
939 {
940         struct hybrid_topology *tp;
941         int ret;
942         u32 i;
943
944         tp = hybrid_topology__new();
945         if (!tp)
946                 return -ENOENT;
947
948         ret = do_write(ff, &tp->nr, sizeof(u32));
949         if (ret < 0)
950                 goto err;
951
952         for (i = 0; i < tp->nr; i++) {
953                 struct hybrid_topology_node *n = &tp->nodes[i];
954
955                 ret = do_write_string(ff, n->pmu_name);
956                 if (ret < 0)
957                         goto err;
958
959                 ret = do_write_string(ff, n->cpus);
960                 if (ret < 0)
961                         goto err;
962         }
963
964         ret = 0;
965
966 err:
967         hybrid_topology__delete(tp);
968         return ret;
969 }
970
971 static int write_dir_format(struct feat_fd *ff,
972                             struct evlist *evlist __maybe_unused)
973 {
974         struct perf_session *session;
975         struct perf_data *data;
976
977         session = container_of(ff->ph, struct perf_session, header);
978         data = session->data;
979
980         if (WARN_ON(!perf_data__is_dir(data)))
981                 return -1;
982
983         return do_write(ff, &data->dir.version, sizeof(data->dir.version));
984 }
985
986 /*
987  * Check whether a CPU is online
988  *
989  * Returns:
990  *     1 -> if CPU is online
991  *     0 -> if CPU is offline
992  *    -1 -> error case
993  */
994 int is_cpu_online(unsigned int cpu)
995 {
996         char *str;
997         size_t strlen;
998         char buf[256];
999         int status = -1;
1000         struct stat statbuf;
1001
1002         snprintf(buf, sizeof(buf),
1003                 "/sys/devices/system/cpu/cpu%d", cpu);
1004         if (stat(buf, &statbuf) != 0)
1005                 return 0;
1006
1007         /*
1008          * Check if /sys/devices/system/cpu/cpux/online file
1009          * exists. Some cases cpu0 won't have online file since
1010          * it is not expected to be turned off generally.
1011          * In kernels without CONFIG_HOTPLUG_CPU, this
1012          * file won't exist
1013          */
1014         snprintf(buf, sizeof(buf),
1015                 "/sys/devices/system/cpu/cpu%d/online", cpu);
1016         if (stat(buf, &statbuf) != 0)
1017                 return 1;
1018
1019         /*
1020          * Read online file using sysfs__read_str.
1021          * If read or open fails, return -1.
1022          * If read succeeds, return value from file
1023          * which gets stored in "str"
1024          */
1025         snprintf(buf, sizeof(buf),
1026                 "devices/system/cpu/cpu%d/online", cpu);
1027
1028         if (sysfs__read_str(buf, &str, &strlen) < 0)
1029                 return status;
1030
1031         status = atoi(str);
1032
1033         free(str);
1034         return status;
1035 }
1036
1037 #ifdef HAVE_LIBBPF_SUPPORT
1038 static int write_bpf_prog_info(struct feat_fd *ff,
1039                                struct evlist *evlist __maybe_unused)
1040 {
1041         struct perf_env *env = &ff->ph->env;
1042         struct rb_root *root;
1043         struct rb_node *next;
1044         int ret;
1045
1046         down_read(&env->bpf_progs.lock);
1047
1048         ret = do_write(ff, &env->bpf_progs.infos_cnt,
1049                        sizeof(env->bpf_progs.infos_cnt));
1050         if (ret < 0)
1051                 goto out;
1052
1053         root = &env->bpf_progs.infos;
1054         next = rb_first(root);
1055         while (next) {
1056                 struct bpf_prog_info_node *node;
1057                 size_t len;
1058
1059                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1060                 next = rb_next(&node->rb_node);
1061                 len = sizeof(struct perf_bpil) +
1062                         node->info_linear->data_len;
1063
1064                 /* before writing to file, translate address to offset */
1065                 bpil_addr_to_offs(node->info_linear);
1066                 ret = do_write(ff, node->info_linear, len);
1067                 /*
1068                  * translate back to address even when do_write() fails,
1069                  * so that this function never changes the data.
1070                  */
1071                 bpil_offs_to_addr(node->info_linear);
1072                 if (ret < 0)
1073                         goto out;
1074         }
1075 out:
1076         up_read(&env->bpf_progs.lock);
1077         return ret;
1078 }
1079
1080 static int write_bpf_btf(struct feat_fd *ff,
1081                          struct evlist *evlist __maybe_unused)
1082 {
1083         struct perf_env *env = &ff->ph->env;
1084         struct rb_root *root;
1085         struct rb_node *next;
1086         int ret;
1087
1088         down_read(&env->bpf_progs.lock);
1089
1090         ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1091                        sizeof(env->bpf_progs.btfs_cnt));
1092
1093         if (ret < 0)
1094                 goto out;
1095
1096         root = &env->bpf_progs.btfs;
1097         next = rb_first(root);
1098         while (next) {
1099                 struct btf_node *node;
1100
1101                 node = rb_entry(next, struct btf_node, rb_node);
1102                 next = rb_next(&node->rb_node);
1103                 ret = do_write(ff, &node->id,
1104                                sizeof(u32) * 2 + node->data_size);
1105                 if (ret < 0)
1106                         goto out;
1107         }
1108 out:
1109         up_read(&env->bpf_progs.lock);
1110         return ret;
1111 }
1112 #endif // HAVE_LIBBPF_SUPPORT
1113
1114 static int cpu_cache_level__sort(const void *a, const void *b)
1115 {
1116         struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1117         struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1118
1119         return cache_a->level - cache_b->level;
1120 }
1121
1122 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1123 {
1124         if (a->level != b->level)
1125                 return false;
1126
1127         if (a->line_size != b->line_size)
1128                 return false;
1129
1130         if (a->sets != b->sets)
1131                 return false;
1132
1133         if (a->ways != b->ways)
1134                 return false;
1135
1136         if (strcmp(a->type, b->type))
1137                 return false;
1138
1139         if (strcmp(a->size, b->size))
1140                 return false;
1141
1142         if (strcmp(a->map, b->map))
1143                 return false;
1144
1145         return true;
1146 }
1147
1148 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1149 {
1150         char path[PATH_MAX], file[PATH_MAX];
1151         struct stat st;
1152         size_t len;
1153
1154         scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1155         scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1156
1157         if (stat(file, &st))
1158                 return 1;
1159
1160         scnprintf(file, PATH_MAX, "%s/level", path);
1161         if (sysfs__read_int(file, (int *) &cache->level))
1162                 return -1;
1163
1164         scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1165         if (sysfs__read_int(file, (int *) &cache->line_size))
1166                 return -1;
1167
1168         scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1169         if (sysfs__read_int(file, (int *) &cache->sets))
1170                 return -1;
1171
1172         scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1173         if (sysfs__read_int(file, (int *) &cache->ways))
1174                 return -1;
1175
1176         scnprintf(file, PATH_MAX, "%s/type", path);
1177         if (sysfs__read_str(file, &cache->type, &len))
1178                 return -1;
1179
1180         cache->type[len] = 0;
1181         cache->type = strim(cache->type);
1182
1183         scnprintf(file, PATH_MAX, "%s/size", path);
1184         if (sysfs__read_str(file, &cache->size, &len)) {
1185                 zfree(&cache->type);
1186                 return -1;
1187         }
1188
1189         cache->size[len] = 0;
1190         cache->size = strim(cache->size);
1191
1192         scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1193         if (sysfs__read_str(file, &cache->map, &len)) {
1194                 zfree(&cache->size);
1195                 zfree(&cache->type);
1196                 return -1;
1197         }
1198
1199         cache->map[len] = 0;
1200         cache->map = strim(cache->map);
1201         return 0;
1202 }
1203
1204 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1205 {
1206         fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1207 }
1208
1209 #define MAX_CACHE_LVL 4
1210
1211 static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1212 {
1213         u32 i, cnt = 0;
1214         u32 nr, cpu;
1215         u16 level;
1216
1217         nr = cpu__max_cpu().cpu;
1218
1219         for (cpu = 0; cpu < nr; cpu++) {
1220                 for (level = 0; level < MAX_CACHE_LVL; level++) {
1221                         struct cpu_cache_level c;
1222                         int err;
1223
1224                         err = cpu_cache_level__read(&c, cpu, level);
1225                         if (err < 0)
1226                                 return err;
1227
1228                         if (err == 1)
1229                                 break;
1230
1231                         for (i = 0; i < cnt; i++) {
1232                                 if (cpu_cache_level__cmp(&c, &caches[i]))
1233                                         break;
1234                         }
1235
1236                         if (i == cnt)
1237                                 caches[cnt++] = c;
1238                         else
1239                                 cpu_cache_level__free(&c);
1240                 }
1241         }
1242         *cntp = cnt;
1243         return 0;
1244 }
1245
1246 static int write_cache(struct feat_fd *ff,
1247                        struct evlist *evlist __maybe_unused)
1248 {
1249         u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL;
1250         struct cpu_cache_level caches[max_caches];
1251         u32 cnt = 0, i, version = 1;
1252         int ret;
1253
1254         ret = build_caches(caches, &cnt);
1255         if (ret)
1256                 goto out;
1257
1258         qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1259
1260         ret = do_write(ff, &version, sizeof(u32));
1261         if (ret < 0)
1262                 goto out;
1263
1264         ret = do_write(ff, &cnt, sizeof(u32));
1265         if (ret < 0)
1266                 goto out;
1267
1268         for (i = 0; i < cnt; i++) {
1269                 struct cpu_cache_level *c = &caches[i];
1270
1271                 #define _W(v)                                   \
1272                         ret = do_write(ff, &c->v, sizeof(u32)); \
1273                         if (ret < 0)                            \
1274                                 goto out;
1275
1276                 _W(level)
1277                 _W(line_size)
1278                 _W(sets)
1279                 _W(ways)
1280                 #undef _W
1281
1282                 #define _W(v)                                           \
1283                         ret = do_write_string(ff, (const char *) c->v); \
1284                         if (ret < 0)                                    \
1285                                 goto out;
1286
1287                 _W(type)
1288                 _W(size)
1289                 _W(map)
1290                 #undef _W
1291         }
1292
1293 out:
1294         for (i = 0; i < cnt; i++)
1295                 cpu_cache_level__free(&caches[i]);
1296         return ret;
1297 }
1298
1299 static int write_stat(struct feat_fd *ff __maybe_unused,
1300                       struct evlist *evlist __maybe_unused)
1301 {
1302         return 0;
1303 }
1304
1305 static int write_sample_time(struct feat_fd *ff,
1306                              struct evlist *evlist)
1307 {
1308         int ret;
1309
1310         ret = do_write(ff, &evlist->first_sample_time,
1311                        sizeof(evlist->first_sample_time));
1312         if (ret < 0)
1313                 return ret;
1314
1315         return do_write(ff, &evlist->last_sample_time,
1316                         sizeof(evlist->last_sample_time));
1317 }
1318
1319
1320 static int memory_node__read(struct memory_node *n, unsigned long idx)
1321 {
1322         unsigned int phys, size = 0;
1323         char path[PATH_MAX];
1324         struct dirent *ent;
1325         DIR *dir;
1326
1327 #define for_each_memory(mem, dir)                                       \
1328         while ((ent = readdir(dir)))                                    \
1329                 if (strcmp(ent->d_name, ".") &&                         \
1330                     strcmp(ent->d_name, "..") &&                        \
1331                     sscanf(ent->d_name, "memory%u", &mem) == 1)
1332
1333         scnprintf(path, PATH_MAX,
1334                   "%s/devices/system/node/node%lu",
1335                   sysfs__mountpoint(), idx);
1336
1337         dir = opendir(path);
1338         if (!dir) {
1339                 pr_warning("failed: can't open memory sysfs data\n");
1340                 return -1;
1341         }
1342
1343         for_each_memory(phys, dir) {
1344                 size = max(phys, size);
1345         }
1346
1347         size++;
1348
1349         n->set = bitmap_zalloc(size);
1350         if (!n->set) {
1351                 closedir(dir);
1352                 return -ENOMEM;
1353         }
1354
1355         n->node = idx;
1356         n->size = size;
1357
1358         rewinddir(dir);
1359
1360         for_each_memory(phys, dir) {
1361                 set_bit(phys, n->set);
1362         }
1363
1364         closedir(dir);
1365         return 0;
1366 }
1367
1368 static int memory_node__sort(const void *a, const void *b)
1369 {
1370         const struct memory_node *na = a;
1371         const struct memory_node *nb = b;
1372
1373         return na->node - nb->node;
1374 }
1375
1376 static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
1377 {
1378         char path[PATH_MAX];
1379         struct dirent *ent;
1380         DIR *dir;
1381         u64 cnt = 0;
1382         int ret = 0;
1383
1384         scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1385                   sysfs__mountpoint());
1386
1387         dir = opendir(path);
1388         if (!dir) {
1389                 pr_debug2("%s: couldn't read %s, does this arch have topology information?\n",
1390                           __func__, path);
1391                 return -1;
1392         }
1393
1394         while (!ret && (ent = readdir(dir))) {
1395                 unsigned int idx;
1396                 int r;
1397
1398                 if (!strcmp(ent->d_name, ".") ||
1399                     !strcmp(ent->d_name, ".."))
1400                         continue;
1401
1402                 r = sscanf(ent->d_name, "node%u", &idx);
1403                 if (r != 1)
1404                         continue;
1405
1406                 if (WARN_ONCE(cnt >= size,
1407                         "failed to write MEM_TOPOLOGY, way too many nodes\n")) {
1408                         closedir(dir);
1409                         return -1;
1410                 }
1411
1412                 ret = memory_node__read(&nodes[cnt++], idx);
1413         }
1414
1415         *cntp = cnt;
1416         closedir(dir);
1417
1418         if (!ret)
1419                 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1420
1421         return ret;
1422 }
1423
1424 #define MAX_MEMORY_NODES 2000
1425
1426 /*
1427  * The MEM_TOPOLOGY holds physical memory map for every
1428  * node in system. The format of data is as follows:
1429  *
1430  *  0 - version          | for future changes
1431  *  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1432  * 16 - count            | number of nodes
1433  *
1434  * For each node we store map of physical indexes for
1435  * each node:
1436  *
1437  * 32 - node id          | node index
1438  * 40 - size             | size of bitmap
1439  * 48 - bitmap           | bitmap of memory indexes that belongs to node
1440  */
1441 static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1442                               struct evlist *evlist __maybe_unused)
1443 {
1444         static struct memory_node nodes[MAX_MEMORY_NODES];
1445         u64 bsize, version = 1, i, nr;
1446         int ret;
1447
1448         ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1449                               (unsigned long long *) &bsize);
1450         if (ret)
1451                 return ret;
1452
1453         ret = build_mem_topology(&nodes[0], MAX_MEMORY_NODES, &nr);
1454         if (ret)
1455                 return ret;
1456
1457         ret = do_write(ff, &version, sizeof(version));
1458         if (ret < 0)
1459                 goto out;
1460
1461         ret = do_write(ff, &bsize, sizeof(bsize));
1462         if (ret < 0)
1463                 goto out;
1464
1465         ret = do_write(ff, &nr, sizeof(nr));
1466         if (ret < 0)
1467                 goto out;
1468
1469         for (i = 0; i < nr; i++) {
1470                 struct memory_node *n = &nodes[i];
1471
1472                 #define _W(v)                                           \
1473                         ret = do_write(ff, &n->v, sizeof(n->v));        \
1474                         if (ret < 0)                                    \
1475                                 goto out;
1476
1477                 _W(node)
1478                 _W(size)
1479
1480                 #undef _W
1481
1482                 ret = do_write_bitmap(ff, n->set, n->size);
1483                 if (ret < 0)
1484                         goto out;
1485         }
1486
1487 out:
1488         return ret;
1489 }
1490
1491 static int write_compressed(struct feat_fd *ff __maybe_unused,
1492                             struct evlist *evlist __maybe_unused)
1493 {
1494         int ret;
1495
1496         ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1497         if (ret)
1498                 return ret;
1499
1500         ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1501         if (ret)
1502                 return ret;
1503
1504         ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1505         if (ret)
1506                 return ret;
1507
1508         ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1509         if (ret)
1510                 return ret;
1511
1512         return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1513 }
1514
1515 static int write_per_cpu_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1516                                   bool write_pmu)
1517 {
1518         struct perf_pmu_caps *caps = NULL;
1519         int nr_caps;
1520         int ret;
1521
1522         nr_caps = perf_pmu__caps_parse(pmu);
1523         if (nr_caps < 0)
1524                 return nr_caps;
1525
1526         ret = do_write(ff, &nr_caps, sizeof(nr_caps));
1527         if (ret < 0)
1528                 return ret;
1529
1530         list_for_each_entry(caps, &pmu->caps, list) {
1531                 ret = do_write_string(ff, caps->name);
1532                 if (ret < 0)
1533                         return ret;
1534
1535                 ret = do_write_string(ff, caps->value);
1536                 if (ret < 0)
1537                         return ret;
1538         }
1539
1540         if (write_pmu) {
1541                 ret = do_write_string(ff, pmu->name);
1542                 if (ret < 0)
1543                         return ret;
1544         }
1545
1546         return ret;
1547 }
1548
1549 static int write_cpu_pmu_caps(struct feat_fd *ff,
1550                               struct evlist *evlist __maybe_unused)
1551 {
1552         struct perf_pmu *cpu_pmu = perf_pmu__find("cpu");
1553
1554         if (!cpu_pmu)
1555                 return -ENOENT;
1556
1557         return write_per_cpu_pmu_caps(ff, cpu_pmu, false);
1558 }
1559
1560 static int write_hybrid_cpu_pmu_caps(struct feat_fd *ff,
1561                                      struct evlist *evlist __maybe_unused)
1562 {
1563         struct perf_pmu *pmu;
1564         u32 nr_pmu = perf_pmu__hybrid_pmu_num();
1565         int ret;
1566
1567         if (nr_pmu == 0)
1568                 return -ENOENT;
1569
1570         ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1571         if (ret < 0)
1572                 return ret;
1573
1574         perf_pmu__for_each_hybrid_pmu(pmu) {
1575                 ret = write_per_cpu_pmu_caps(ff, pmu, true);
1576                 if (ret < 0)
1577                         return ret;
1578         }
1579
1580         return 0;
1581 }
1582
1583 static void print_hostname(struct feat_fd *ff, FILE *fp)
1584 {
1585         fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1586 }
1587
1588 static void print_osrelease(struct feat_fd *ff, FILE *fp)
1589 {
1590         fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1591 }
1592
1593 static void print_arch(struct feat_fd *ff, FILE *fp)
1594 {
1595         fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1596 }
1597
1598 static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1599 {
1600         fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1601 }
1602
1603 static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1604 {
1605         fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1606         fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1607 }
1608
1609 static void print_version(struct feat_fd *ff, FILE *fp)
1610 {
1611         fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1612 }
1613
1614 static void print_cmdline(struct feat_fd *ff, FILE *fp)
1615 {
1616         int nr, i;
1617
1618         nr = ff->ph->env.nr_cmdline;
1619
1620         fprintf(fp, "# cmdline : ");
1621
1622         for (i = 0; i < nr; i++) {
1623                 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1624                 if (!argv_i) {
1625                         fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1626                 } else {
1627                         char *mem = argv_i;
1628                         do {
1629                                 char *quote = strchr(argv_i, '\'');
1630                                 if (!quote)
1631                                         break;
1632                                 *quote++ = '\0';
1633                                 fprintf(fp, "%s\\\'", argv_i);
1634                                 argv_i = quote;
1635                         } while (1);
1636                         fprintf(fp, "%s ", argv_i);
1637                         free(mem);
1638                 }
1639         }
1640         fputc('\n', fp);
1641 }
1642
1643 static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1644 {
1645         struct perf_header *ph = ff->ph;
1646         int cpu_nr = ph->env.nr_cpus_avail;
1647         int nr, i;
1648         char *str;
1649
1650         nr = ph->env.nr_sibling_cores;
1651         str = ph->env.sibling_cores;
1652
1653         for (i = 0; i < nr; i++) {
1654                 fprintf(fp, "# sibling sockets : %s\n", str);
1655                 str += strlen(str) + 1;
1656         }
1657
1658         if (ph->env.nr_sibling_dies) {
1659                 nr = ph->env.nr_sibling_dies;
1660                 str = ph->env.sibling_dies;
1661
1662                 for (i = 0; i < nr; i++) {
1663                         fprintf(fp, "# sibling dies    : %s\n", str);
1664                         str += strlen(str) + 1;
1665                 }
1666         }
1667
1668         nr = ph->env.nr_sibling_threads;
1669         str = ph->env.sibling_threads;
1670
1671         for (i = 0; i < nr; i++) {
1672                 fprintf(fp, "# sibling threads : %s\n", str);
1673                 str += strlen(str) + 1;
1674         }
1675
1676         if (ph->env.nr_sibling_dies) {
1677                 if (ph->env.cpu != NULL) {
1678                         for (i = 0; i < cpu_nr; i++)
1679                                 fprintf(fp, "# CPU %d: Core ID %d, "
1680                                             "Die ID %d, Socket ID %d\n",
1681                                             i, ph->env.cpu[i].core_id,
1682                                             ph->env.cpu[i].die_id,
1683                                             ph->env.cpu[i].socket_id);
1684                 } else
1685                         fprintf(fp, "# Core ID, Die ID and Socket ID "
1686                                     "information is not available\n");
1687         } else {
1688                 if (ph->env.cpu != NULL) {
1689                         for (i = 0; i < cpu_nr; i++)
1690                                 fprintf(fp, "# CPU %d: Core ID %d, "
1691                                             "Socket ID %d\n",
1692                                             i, ph->env.cpu[i].core_id,
1693                                             ph->env.cpu[i].socket_id);
1694                 } else
1695                         fprintf(fp, "# Core ID and Socket ID "
1696                                     "information is not available\n");
1697         }
1698 }
1699
1700 static void print_clockid(struct feat_fd *ff, FILE *fp)
1701 {
1702         fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1703                 ff->ph->env.clock.clockid_res_ns * 1000);
1704 }
1705
1706 static void print_clock_data(struct feat_fd *ff, FILE *fp)
1707 {
1708         struct timespec clockid_ns;
1709         char tstr[64], date[64];
1710         struct timeval tod_ns;
1711         clockid_t clockid;
1712         struct tm ltime;
1713         u64 ref;
1714
1715         if (!ff->ph->env.clock.enabled) {
1716                 fprintf(fp, "# reference time disabled\n");
1717                 return;
1718         }
1719
1720         /* Compute TOD time. */
1721         ref = ff->ph->env.clock.tod_ns;
1722         tod_ns.tv_sec = ref / NSEC_PER_SEC;
1723         ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1724         tod_ns.tv_usec = ref / NSEC_PER_USEC;
1725
1726         /* Compute clockid time. */
1727         ref = ff->ph->env.clock.clockid_ns;
1728         clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1729         ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1730         clockid_ns.tv_nsec = ref;
1731
1732         clockid = ff->ph->env.clock.clockid;
1733
1734         if (localtime_r(&tod_ns.tv_sec, &ltime) == NULL)
1735                 snprintf(tstr, sizeof(tstr), "<error>");
1736         else {
1737                 strftime(date, sizeof(date), "%F %T", &ltime);
1738                 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1739                           date, (int) tod_ns.tv_usec);
1740         }
1741
1742         fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1743         fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1744                     tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1745                     (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1746                     clockid_name(clockid));
1747 }
1748
1749 static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1750 {
1751         int i;
1752         struct hybrid_node *n;
1753
1754         fprintf(fp, "# hybrid cpu system:\n");
1755         for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1756                 n = &ff->ph->env.hybrid_nodes[i];
1757                 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1758         }
1759 }
1760
1761 static void print_dir_format(struct feat_fd *ff, FILE *fp)
1762 {
1763         struct perf_session *session;
1764         struct perf_data *data;
1765
1766         session = container_of(ff->ph, struct perf_session, header);
1767         data = session->data;
1768
1769         fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1770 }
1771
1772 #ifdef HAVE_LIBBPF_SUPPORT
1773 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1774 {
1775         struct perf_env *env = &ff->ph->env;
1776         struct rb_root *root;
1777         struct rb_node *next;
1778
1779         down_read(&env->bpf_progs.lock);
1780
1781         root = &env->bpf_progs.infos;
1782         next = rb_first(root);
1783
1784         while (next) {
1785                 struct bpf_prog_info_node *node;
1786
1787                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1788                 next = rb_next(&node->rb_node);
1789
1790                 bpf_event__print_bpf_prog_info(&node->info_linear->info,
1791                                                env, fp);
1792         }
1793
1794         up_read(&env->bpf_progs.lock);
1795 }
1796
1797 static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1798 {
1799         struct perf_env *env = &ff->ph->env;
1800         struct rb_root *root;
1801         struct rb_node *next;
1802
1803         down_read(&env->bpf_progs.lock);
1804
1805         root = &env->bpf_progs.btfs;
1806         next = rb_first(root);
1807
1808         while (next) {
1809                 struct btf_node *node;
1810
1811                 node = rb_entry(next, struct btf_node, rb_node);
1812                 next = rb_next(&node->rb_node);
1813                 fprintf(fp, "# btf info of id %u\n", node->id);
1814         }
1815
1816         up_read(&env->bpf_progs.lock);
1817 }
1818 #endif // HAVE_LIBBPF_SUPPORT
1819
1820 static void free_event_desc(struct evsel *events)
1821 {
1822         struct evsel *evsel;
1823
1824         if (!events)
1825                 return;
1826
1827         for (evsel = events; evsel->core.attr.size; evsel++) {
1828                 zfree(&evsel->name);
1829                 zfree(&evsel->core.id);
1830         }
1831
1832         free(events);
1833 }
1834
1835 static bool perf_attr_check(struct perf_event_attr *attr)
1836 {
1837         if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1838                 pr_warning("Reserved bits are set unexpectedly. "
1839                            "Please update perf tool.\n");
1840                 return false;
1841         }
1842
1843         if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1844                 pr_warning("Unknown sample type (0x%llx) is detected. "
1845                            "Please update perf tool.\n",
1846                            attr->sample_type);
1847                 return false;
1848         }
1849
1850         if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1851                 pr_warning("Unknown read format (0x%llx) is detected. "
1852                            "Please update perf tool.\n",
1853                            attr->read_format);
1854                 return false;
1855         }
1856
1857         if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1858             (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1859                 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1860                            "Please update perf tool.\n",
1861                            attr->branch_sample_type);
1862
1863                 return false;
1864         }
1865
1866         return true;
1867 }
1868
1869 static struct evsel *read_event_desc(struct feat_fd *ff)
1870 {
1871         struct evsel *evsel, *events = NULL;
1872         u64 *id;
1873         void *buf = NULL;
1874         u32 nre, sz, nr, i, j;
1875         size_t msz;
1876
1877         /* number of events */
1878         if (do_read_u32(ff, &nre))
1879                 goto error;
1880
1881         if (do_read_u32(ff, &sz))
1882                 goto error;
1883
1884         /* buffer to hold on file attr struct */
1885         buf = malloc(sz);
1886         if (!buf)
1887                 goto error;
1888
1889         /* the last event terminates with evsel->core.attr.size == 0: */
1890         events = calloc(nre + 1, sizeof(*events));
1891         if (!events)
1892                 goto error;
1893
1894         msz = sizeof(evsel->core.attr);
1895         if (sz < msz)
1896                 msz = sz;
1897
1898         for (i = 0, evsel = events; i < nre; evsel++, i++) {
1899                 evsel->core.idx = i;
1900
1901                 /*
1902                  * must read entire on-file attr struct to
1903                  * sync up with layout.
1904                  */
1905                 if (__do_read(ff, buf, sz))
1906                         goto error;
1907
1908                 if (ff->ph->needs_swap)
1909                         perf_event__attr_swap(buf);
1910
1911                 memcpy(&evsel->core.attr, buf, msz);
1912
1913                 if (!perf_attr_check(&evsel->core.attr))
1914                         goto error;
1915
1916                 if (do_read_u32(ff, &nr))
1917                         goto error;
1918
1919                 if (ff->ph->needs_swap)
1920                         evsel->needs_swap = true;
1921
1922                 evsel->name = do_read_string(ff);
1923                 if (!evsel->name)
1924                         goto error;
1925
1926                 if (!nr)
1927                         continue;
1928
1929                 id = calloc(nr, sizeof(*id));
1930                 if (!id)
1931                         goto error;
1932                 evsel->core.ids = nr;
1933                 evsel->core.id = id;
1934
1935                 for (j = 0 ; j < nr; j++) {
1936                         if (do_read_u64(ff, id))
1937                                 goto error;
1938                         id++;
1939                 }
1940         }
1941 out:
1942         free(buf);
1943         return events;
1944 error:
1945         free_event_desc(events);
1946         events = NULL;
1947         goto out;
1948 }
1949
1950 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
1951                                 void *priv __maybe_unused)
1952 {
1953         return fprintf(fp, ", %s = %s", name, val);
1954 }
1955
1956 static void print_event_desc(struct feat_fd *ff, FILE *fp)
1957 {
1958         struct evsel *evsel, *events;
1959         u32 j;
1960         u64 *id;
1961
1962         if (ff->events)
1963                 events = ff->events;
1964         else
1965                 events = read_event_desc(ff);
1966
1967         if (!events) {
1968                 fprintf(fp, "# event desc: not available or unable to read\n");
1969                 return;
1970         }
1971
1972         for (evsel = events; evsel->core.attr.size; evsel++) {
1973                 fprintf(fp, "# event : name = %s, ", evsel->name);
1974
1975                 if (evsel->core.ids) {
1976                         fprintf(fp, ", id = {");
1977                         for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
1978                                 if (j)
1979                                         fputc(',', fp);
1980                                 fprintf(fp, " %"PRIu64, *id);
1981                         }
1982                         fprintf(fp, " }");
1983                 }
1984
1985                 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
1986
1987                 fputc('\n', fp);
1988         }
1989
1990         free_event_desc(events);
1991         ff->events = NULL;
1992 }
1993
1994 static void print_total_mem(struct feat_fd *ff, FILE *fp)
1995 {
1996         fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
1997 }
1998
1999 static void print_numa_topology(struct feat_fd *ff, FILE *fp)
2000 {
2001         int i;
2002         struct numa_node *n;
2003
2004         for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
2005                 n = &ff->ph->env.numa_nodes[i];
2006
2007                 fprintf(fp, "# node%u meminfo  : total = %"PRIu64" kB,"
2008                             " free = %"PRIu64" kB\n",
2009                         n->node, n->mem_total, n->mem_free);
2010
2011                 fprintf(fp, "# node%u cpu list : ", n->node);
2012                 cpu_map__fprintf(n->map, fp);
2013         }
2014 }
2015
2016 static void print_cpuid(struct feat_fd *ff, FILE *fp)
2017 {
2018         fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
2019 }
2020
2021 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
2022 {
2023         fprintf(fp, "# contains samples with branch stack\n");
2024 }
2025
2026 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
2027 {
2028         fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
2029 }
2030
2031 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
2032 {
2033         fprintf(fp, "# contains stat data\n");
2034 }
2035
2036 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
2037 {
2038         int i;
2039
2040         fprintf(fp, "# CPU cache info:\n");
2041         for (i = 0; i < ff->ph->env.caches_cnt; i++) {
2042                 fprintf(fp, "#  ");
2043                 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
2044         }
2045 }
2046
2047 static void print_compressed(struct feat_fd *ff, FILE *fp)
2048 {
2049         fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
2050                 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
2051                 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2052 }
2053
2054 static void print_per_cpu_pmu_caps(FILE *fp, int nr_caps, char *cpu_pmu_caps,
2055                                    char *pmu_name)
2056 {
2057         const char *delimiter;
2058         char *str, buf[128];
2059
2060         if (!nr_caps) {
2061                 if (!pmu_name)
2062                         fprintf(fp, "# cpu pmu capabilities: not available\n");
2063                 else
2064                         fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2065                 return;
2066         }
2067
2068         if (!pmu_name)
2069                 scnprintf(buf, sizeof(buf), "# cpu pmu capabilities: ");
2070         else
2071                 scnprintf(buf, sizeof(buf), "# %s pmu capabilities: ", pmu_name);
2072
2073         delimiter = buf;
2074
2075         str = cpu_pmu_caps;
2076         while (nr_caps--) {
2077                 fprintf(fp, "%s%s", delimiter, str);
2078                 delimiter = ", ";
2079                 str += strlen(str) + 1;
2080         }
2081
2082         fprintf(fp, "\n");
2083 }
2084
2085 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2086 {
2087         print_per_cpu_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2088                                ff->ph->env.cpu_pmu_caps, NULL);
2089 }
2090
2091 static void print_hybrid_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2092 {
2093         struct hybrid_cpc_node *n;
2094
2095         for (int i = 0; i < ff->ph->env.nr_hybrid_cpc_nodes; i++) {
2096                 n = &ff->ph->env.hybrid_cpc_nodes[i];
2097                 print_per_cpu_pmu_caps(fp, n->nr_cpu_pmu_caps,
2098                                        n->cpu_pmu_caps,
2099                                        n->pmu_name);
2100         }
2101 }
2102
2103 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2104 {
2105         const char *delimiter = "# pmu mappings: ";
2106         char *str, *tmp;
2107         u32 pmu_num;
2108         u32 type;
2109
2110         pmu_num = ff->ph->env.nr_pmu_mappings;
2111         if (!pmu_num) {
2112                 fprintf(fp, "# pmu mappings: not available\n");
2113                 return;
2114         }
2115
2116         str = ff->ph->env.pmu_mappings;
2117
2118         while (pmu_num) {
2119                 type = strtoul(str, &tmp, 0);
2120                 if (*tmp != ':')
2121                         goto error;
2122
2123                 str = tmp + 1;
2124                 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2125
2126                 delimiter = ", ";
2127                 str += strlen(str) + 1;
2128                 pmu_num--;
2129         }
2130
2131         fprintf(fp, "\n");
2132
2133         if (!pmu_num)
2134                 return;
2135 error:
2136         fprintf(fp, "# pmu mappings: unable to read\n");
2137 }
2138
2139 static void print_group_desc(struct feat_fd *ff, FILE *fp)
2140 {
2141         struct perf_session *session;
2142         struct evsel *evsel;
2143         u32 nr = 0;
2144
2145         session = container_of(ff->ph, struct perf_session, header);
2146
2147         evlist__for_each_entry(session->evlist, evsel) {
2148                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2149                         fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2150
2151                         nr = evsel->core.nr_members - 1;
2152                 } else if (nr) {
2153                         fprintf(fp, ",%s", evsel__name(evsel));
2154
2155                         if (--nr == 0)
2156                                 fprintf(fp, "}\n");
2157                 }
2158         }
2159 }
2160
2161 static void print_sample_time(struct feat_fd *ff, FILE *fp)
2162 {
2163         struct perf_session *session;
2164         char time_buf[32];
2165         double d;
2166
2167         session = container_of(ff->ph, struct perf_session, header);
2168
2169         timestamp__scnprintf_usec(session->evlist->first_sample_time,
2170                                   time_buf, sizeof(time_buf));
2171         fprintf(fp, "# time of first sample : %s\n", time_buf);
2172
2173         timestamp__scnprintf_usec(session->evlist->last_sample_time,
2174                                   time_buf, sizeof(time_buf));
2175         fprintf(fp, "# time of last sample : %s\n", time_buf);
2176
2177         d = (double)(session->evlist->last_sample_time -
2178                 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2179
2180         fprintf(fp, "# sample duration : %10.3f ms\n", d);
2181 }
2182
2183 static void memory_node__fprintf(struct memory_node *n,
2184                                  unsigned long long bsize, FILE *fp)
2185 {
2186         char buf_map[100], buf_size[50];
2187         unsigned long long size;
2188
2189         size = bsize * bitmap_weight(n->set, n->size);
2190         unit_number__scnprintf(buf_size, 50, size);
2191
2192         bitmap_scnprintf(n->set, n->size, buf_map, 100);
2193         fprintf(fp, "#  %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2194 }
2195
2196 static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2197 {
2198         struct memory_node *nodes;
2199         int i, nr;
2200
2201         nodes = ff->ph->env.memory_nodes;
2202         nr    = ff->ph->env.nr_memory_nodes;
2203
2204         fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2205                 nr, ff->ph->env.memory_bsize);
2206
2207         for (i = 0; i < nr; i++) {
2208                 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2209         }
2210 }
2211
2212 static int __event_process_build_id(struct perf_record_header_build_id *bev,
2213                                     char *filename,
2214                                     struct perf_session *session)
2215 {
2216         int err = -1;
2217         struct machine *machine;
2218         u16 cpumode;
2219         struct dso *dso;
2220         enum dso_space_type dso_space;
2221
2222         machine = perf_session__findnew_machine(session, bev->pid);
2223         if (!machine)
2224                 goto out;
2225
2226         cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2227
2228         switch (cpumode) {
2229         case PERF_RECORD_MISC_KERNEL:
2230                 dso_space = DSO_SPACE__KERNEL;
2231                 break;
2232         case PERF_RECORD_MISC_GUEST_KERNEL:
2233                 dso_space = DSO_SPACE__KERNEL_GUEST;
2234                 break;
2235         case PERF_RECORD_MISC_USER:
2236         case PERF_RECORD_MISC_GUEST_USER:
2237                 dso_space = DSO_SPACE__USER;
2238                 break;
2239         default:
2240                 goto out;
2241         }
2242
2243         dso = machine__findnew_dso(machine, filename);
2244         if (dso != NULL) {
2245                 char sbuild_id[SBUILD_ID_SIZE];
2246                 struct build_id bid;
2247                 size_t size = BUILD_ID_SIZE;
2248
2249                 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2250                         size = bev->size;
2251
2252                 build_id__init(&bid, bev->data, size);
2253                 dso__set_build_id(dso, &bid);
2254                 dso->header_build_id = 1;
2255
2256                 if (dso_space != DSO_SPACE__USER) {
2257                         struct kmod_path m = { .name = NULL, };
2258
2259                         if (!kmod_path__parse_name(&m, filename) && m.kmod)
2260                                 dso__set_module_info(dso, &m, machine);
2261
2262                         dso->kernel = dso_space;
2263                         free(m.name);
2264                 }
2265
2266                 build_id__sprintf(&dso->bid, sbuild_id);
2267                 pr_debug("build id event received for %s: %s [%zu]\n",
2268                          dso->long_name, sbuild_id, size);
2269                 dso__put(dso);
2270         }
2271
2272         err = 0;
2273 out:
2274         return err;
2275 }
2276
2277 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2278                                                  int input, u64 offset, u64 size)
2279 {
2280         struct perf_session *session = container_of(header, struct perf_session, header);
2281         struct {
2282                 struct perf_event_header   header;
2283                 u8                         build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2284                 char                       filename[0];
2285         } old_bev;
2286         struct perf_record_header_build_id bev;
2287         char filename[PATH_MAX];
2288         u64 limit = offset + size;
2289
2290         while (offset < limit) {
2291                 ssize_t len;
2292
2293                 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2294                         return -1;
2295
2296                 if (header->needs_swap)
2297                         perf_event_header__bswap(&old_bev.header);
2298
2299                 len = old_bev.header.size - sizeof(old_bev);
2300                 if (readn(input, filename, len) != len)
2301                         return -1;
2302
2303                 bev.header = old_bev.header;
2304
2305                 /*
2306                  * As the pid is the missing value, we need to fill
2307                  * it properly. The header.misc value give us nice hint.
2308                  */
2309                 bev.pid = HOST_KERNEL_ID;
2310                 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2311                     bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2312                         bev.pid = DEFAULT_GUEST_KERNEL_ID;
2313
2314                 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2315                 __event_process_build_id(&bev, filename, session);
2316
2317                 offset += bev.header.size;
2318         }
2319
2320         return 0;
2321 }
2322
2323 static int perf_header__read_build_ids(struct perf_header *header,
2324                                        int input, u64 offset, u64 size)
2325 {
2326         struct perf_session *session = container_of(header, struct perf_session, header);
2327         struct perf_record_header_build_id bev;
2328         char filename[PATH_MAX];
2329         u64 limit = offset + size, orig_offset = offset;
2330         int err = -1;
2331
2332         while (offset < limit) {
2333                 ssize_t len;
2334
2335                 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2336                         goto out;
2337
2338                 if (header->needs_swap)
2339                         perf_event_header__bswap(&bev.header);
2340
2341                 len = bev.header.size - sizeof(bev);
2342                 if (readn(input, filename, len) != len)
2343                         goto out;
2344                 /*
2345                  * The a1645ce1 changeset:
2346                  *
2347                  * "perf: 'perf kvm' tool for monitoring guest performance from host"
2348                  *
2349                  * Added a field to struct perf_record_header_build_id that broke the file
2350                  * format.
2351                  *
2352                  * Since the kernel build-id is the first entry, process the
2353                  * table using the old format if the well known
2354                  * '[kernel.kallsyms]' string for the kernel build-id has the
2355                  * first 4 characters chopped off (where the pid_t sits).
2356                  */
2357                 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2358                         if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2359                                 return -1;
2360                         return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2361                 }
2362
2363                 __event_process_build_id(&bev, filename, session);
2364
2365                 offset += bev.header.size;
2366         }
2367         err = 0;
2368 out:
2369         return err;
2370 }
2371
2372 /* Macro for features that simply need to read and store a string. */
2373 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2374 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2375 {\
2376         free(ff->ph->env.__feat_env);                \
2377         ff->ph->env.__feat_env = do_read_string(ff); \
2378         return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2379 }
2380
2381 FEAT_PROCESS_STR_FUN(hostname, hostname);
2382 FEAT_PROCESS_STR_FUN(osrelease, os_release);
2383 FEAT_PROCESS_STR_FUN(version, version);
2384 FEAT_PROCESS_STR_FUN(arch, arch);
2385 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2386 FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2387
2388 static int process_tracing_data(struct feat_fd *ff, void *data)
2389 {
2390         ssize_t ret = trace_report(ff->fd, data, false);
2391
2392         return ret < 0 ? -1 : 0;
2393 }
2394
2395 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2396 {
2397         if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2398                 pr_debug("Failed to read buildids, continuing...\n");
2399         return 0;
2400 }
2401
2402 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2403 {
2404         int ret;
2405         u32 nr_cpus_avail, nr_cpus_online;
2406
2407         ret = do_read_u32(ff, &nr_cpus_avail);
2408         if (ret)
2409                 return ret;
2410
2411         ret = do_read_u32(ff, &nr_cpus_online);
2412         if (ret)
2413                 return ret;
2414         ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2415         ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2416         return 0;
2417 }
2418
2419 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2420 {
2421         u64 total_mem;
2422         int ret;
2423
2424         ret = do_read_u64(ff, &total_mem);
2425         if (ret)
2426                 return -1;
2427         ff->ph->env.total_mem = (unsigned long long)total_mem;
2428         return 0;
2429 }
2430
2431 static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2432 {
2433         struct evsel *evsel;
2434
2435         evlist__for_each_entry(evlist, evsel) {
2436                 if (evsel->core.idx == idx)
2437                         return evsel;
2438         }
2439
2440         return NULL;
2441 }
2442
2443 static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2444 {
2445         struct evsel *evsel;
2446
2447         if (!event->name)
2448                 return;
2449
2450         evsel = evlist__find_by_index(evlist, event->core.idx);
2451         if (!evsel)
2452                 return;
2453
2454         if (evsel->name)
2455                 return;
2456
2457         evsel->name = strdup(event->name);
2458 }
2459
2460 static int
2461 process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2462 {
2463         struct perf_session *session;
2464         struct evsel *evsel, *events = read_event_desc(ff);
2465
2466         if (!events)
2467                 return 0;
2468
2469         session = container_of(ff->ph, struct perf_session, header);
2470
2471         if (session->data->is_pipe) {
2472                 /* Save events for reading later by print_event_desc,
2473                  * since they can't be read again in pipe mode. */
2474                 ff->events = events;
2475         }
2476
2477         for (evsel = events; evsel->core.attr.size; evsel++)
2478                 evlist__set_event_name(session->evlist, evsel);
2479
2480         if (!session->data->is_pipe)
2481                 free_event_desc(events);
2482
2483         return 0;
2484 }
2485
2486 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2487 {
2488         char *str, *cmdline = NULL, **argv = NULL;
2489         u32 nr, i, len = 0;
2490
2491         if (do_read_u32(ff, &nr))
2492                 return -1;
2493
2494         ff->ph->env.nr_cmdline = nr;
2495
2496         cmdline = zalloc(ff->size + nr + 1);
2497         if (!cmdline)
2498                 return -1;
2499
2500         argv = zalloc(sizeof(char *) * (nr + 1));
2501         if (!argv)
2502                 goto error;
2503
2504         for (i = 0; i < nr; i++) {
2505                 str = do_read_string(ff);
2506                 if (!str)
2507                         goto error;
2508
2509                 argv[i] = cmdline + len;
2510                 memcpy(argv[i], str, strlen(str) + 1);
2511                 len += strlen(str) + 1;
2512                 free(str);
2513         }
2514         ff->ph->env.cmdline = cmdline;
2515         ff->ph->env.cmdline_argv = (const char **) argv;
2516         return 0;
2517
2518 error:
2519         free(argv);
2520         free(cmdline);
2521         return -1;
2522 }
2523
2524 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2525 {
2526         u32 nr, i;
2527         char *str;
2528         struct strbuf sb;
2529         int cpu_nr = ff->ph->env.nr_cpus_avail;
2530         u64 size = 0;
2531         struct perf_header *ph = ff->ph;
2532         bool do_core_id_test = true;
2533
2534         ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2535         if (!ph->env.cpu)
2536                 return -1;
2537
2538         if (do_read_u32(ff, &nr))
2539                 goto free_cpu;
2540
2541         ph->env.nr_sibling_cores = nr;
2542         size += sizeof(u32);
2543         if (strbuf_init(&sb, 128) < 0)
2544                 goto free_cpu;
2545
2546         for (i = 0; i < nr; i++) {
2547                 str = do_read_string(ff);
2548                 if (!str)
2549                         goto error;
2550
2551                 /* include a NULL character at the end */
2552                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2553                         goto error;
2554                 size += string_size(str);
2555                 free(str);
2556         }
2557         ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2558
2559         if (do_read_u32(ff, &nr))
2560                 return -1;
2561
2562         ph->env.nr_sibling_threads = nr;
2563         size += sizeof(u32);
2564
2565         for (i = 0; i < nr; i++) {
2566                 str = do_read_string(ff);
2567                 if (!str)
2568                         goto error;
2569
2570                 /* include a NULL character at the end */
2571                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2572                         goto error;
2573                 size += string_size(str);
2574                 free(str);
2575         }
2576         ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2577
2578         /*
2579          * The header may be from old perf,
2580          * which doesn't include core id and socket id information.
2581          */
2582         if (ff->size <= size) {
2583                 zfree(&ph->env.cpu);
2584                 return 0;
2585         }
2586
2587         /* On s390 the socket_id number is not related to the numbers of cpus.
2588          * The socket_id number might be higher than the numbers of cpus.
2589          * This depends on the configuration.
2590          * AArch64 is the same.
2591          */
2592         if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2593                           || !strncmp(ph->env.arch, "aarch64", 7)))
2594                 do_core_id_test = false;
2595
2596         for (i = 0; i < (u32)cpu_nr; i++) {
2597                 if (do_read_u32(ff, &nr))
2598                         goto free_cpu;
2599
2600                 ph->env.cpu[i].core_id = nr;
2601                 size += sizeof(u32);
2602
2603                 if (do_read_u32(ff, &nr))
2604                         goto free_cpu;
2605
2606                 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2607                         pr_debug("socket_id number is too big."
2608                                  "You may need to upgrade the perf tool.\n");
2609                         goto free_cpu;
2610                 }
2611
2612                 ph->env.cpu[i].socket_id = nr;
2613                 size += sizeof(u32);
2614         }
2615
2616         /*
2617          * The header may be from old perf,
2618          * which doesn't include die information.
2619          */
2620         if (ff->size <= size)
2621                 return 0;
2622
2623         if (do_read_u32(ff, &nr))
2624                 return -1;
2625
2626         ph->env.nr_sibling_dies = nr;
2627         size += sizeof(u32);
2628
2629         for (i = 0; i < nr; i++) {
2630                 str = do_read_string(ff);
2631                 if (!str)
2632                         goto error;
2633
2634                 /* include a NULL character at the end */
2635                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2636                         goto error;
2637                 size += string_size(str);
2638                 free(str);
2639         }
2640         ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2641
2642         for (i = 0; i < (u32)cpu_nr; i++) {
2643                 if (do_read_u32(ff, &nr))
2644                         goto free_cpu;
2645
2646                 ph->env.cpu[i].die_id = nr;
2647         }
2648
2649         return 0;
2650
2651 error:
2652         strbuf_release(&sb);
2653 free_cpu:
2654         zfree(&ph->env.cpu);
2655         return -1;
2656 }
2657
2658 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2659 {
2660         struct numa_node *nodes, *n;
2661         u32 nr, i;
2662         char *str;
2663
2664         /* nr nodes */
2665         if (do_read_u32(ff, &nr))
2666                 return -1;
2667
2668         nodes = zalloc(sizeof(*nodes) * nr);
2669         if (!nodes)
2670                 return -ENOMEM;
2671
2672         for (i = 0; i < nr; i++) {
2673                 n = &nodes[i];
2674
2675                 /* node number */
2676                 if (do_read_u32(ff, &n->node))
2677                         goto error;
2678
2679                 if (do_read_u64(ff, &n->mem_total))
2680                         goto error;
2681
2682                 if (do_read_u64(ff, &n->mem_free))
2683                         goto error;
2684
2685                 str = do_read_string(ff);
2686                 if (!str)
2687                         goto error;
2688
2689                 n->map = perf_cpu_map__new(str);
2690                 if (!n->map)
2691                         goto error;
2692
2693                 free(str);
2694         }
2695         ff->ph->env.nr_numa_nodes = nr;
2696         ff->ph->env.numa_nodes = nodes;
2697         return 0;
2698
2699 error:
2700         free(nodes);
2701         return -1;
2702 }
2703
2704 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2705 {
2706         char *name;
2707         u32 pmu_num;
2708         u32 type;
2709         struct strbuf sb;
2710
2711         if (do_read_u32(ff, &pmu_num))
2712                 return -1;
2713
2714         if (!pmu_num) {
2715                 pr_debug("pmu mappings not available\n");
2716                 return 0;
2717         }
2718
2719         ff->ph->env.nr_pmu_mappings = pmu_num;
2720         if (strbuf_init(&sb, 128) < 0)
2721                 return -1;
2722
2723         while (pmu_num) {
2724                 if (do_read_u32(ff, &type))
2725                         goto error;
2726
2727                 name = do_read_string(ff);
2728                 if (!name)
2729                         goto error;
2730
2731                 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2732                         goto error;
2733                 /* include a NULL character at the end */
2734                 if (strbuf_add(&sb, "", 1) < 0)
2735                         goto error;
2736
2737                 if (!strcmp(name, "msr"))
2738                         ff->ph->env.msr_pmu_type = type;
2739
2740                 free(name);
2741                 pmu_num--;
2742         }
2743         ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2744         return 0;
2745
2746 error:
2747         strbuf_release(&sb);
2748         return -1;
2749 }
2750
2751 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2752 {
2753         size_t ret = -1;
2754         u32 i, nr, nr_groups;
2755         struct perf_session *session;
2756         struct evsel *evsel, *leader = NULL;
2757         struct group_desc {
2758                 char *name;
2759                 u32 leader_idx;
2760                 u32 nr_members;
2761         } *desc;
2762
2763         if (do_read_u32(ff, &nr_groups))
2764                 return -1;
2765
2766         ff->ph->env.nr_groups = nr_groups;
2767         if (!nr_groups) {
2768                 pr_debug("group desc not available\n");
2769                 return 0;
2770         }
2771
2772         desc = calloc(nr_groups, sizeof(*desc));
2773         if (!desc)
2774                 return -1;
2775
2776         for (i = 0; i < nr_groups; i++) {
2777                 desc[i].name = do_read_string(ff);
2778                 if (!desc[i].name)
2779                         goto out_free;
2780
2781                 if (do_read_u32(ff, &desc[i].leader_idx))
2782                         goto out_free;
2783
2784                 if (do_read_u32(ff, &desc[i].nr_members))
2785                         goto out_free;
2786         }
2787
2788         /*
2789          * Rebuild group relationship based on the group_desc
2790          */
2791         session = container_of(ff->ph, struct perf_session, header);
2792         session->evlist->core.nr_groups = nr_groups;
2793
2794         i = nr = 0;
2795         evlist__for_each_entry(session->evlist, evsel) {
2796                 if (evsel->core.idx == (int) desc[i].leader_idx) {
2797                         evsel__set_leader(evsel, evsel);
2798                         /* {anon_group} is a dummy name */
2799                         if (strcmp(desc[i].name, "{anon_group}")) {
2800                                 evsel->group_name = desc[i].name;
2801                                 desc[i].name = NULL;
2802                         }
2803                         evsel->core.nr_members = desc[i].nr_members;
2804
2805                         if (i >= nr_groups || nr > 0) {
2806                                 pr_debug("invalid group desc\n");
2807                                 goto out_free;
2808                         }
2809
2810                         leader = evsel;
2811                         nr = evsel->core.nr_members - 1;
2812                         i++;
2813                 } else if (nr) {
2814                         /* This is a group member */
2815                         evsel__set_leader(evsel, leader);
2816
2817                         nr--;
2818                 }
2819         }
2820
2821         if (i != nr_groups || nr != 0) {
2822                 pr_debug("invalid group desc\n");
2823                 goto out_free;
2824         }
2825
2826         ret = 0;
2827 out_free:
2828         for (i = 0; i < nr_groups; i++)
2829                 zfree(&desc[i].name);
2830         free(desc);
2831
2832         return ret;
2833 }
2834
2835 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2836 {
2837         struct perf_session *session;
2838         int err;
2839
2840         session = container_of(ff->ph, struct perf_session, header);
2841
2842         err = auxtrace_index__process(ff->fd, ff->size, session,
2843                                       ff->ph->needs_swap);
2844         if (err < 0)
2845                 pr_err("Failed to process auxtrace index\n");
2846         return err;
2847 }
2848
2849 static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2850 {
2851         struct cpu_cache_level *caches;
2852         u32 cnt, i, version;
2853
2854         if (do_read_u32(ff, &version))
2855                 return -1;
2856
2857         if (version != 1)
2858                 return -1;
2859
2860         if (do_read_u32(ff, &cnt))
2861                 return -1;
2862
2863         caches = zalloc(sizeof(*caches) * cnt);
2864         if (!caches)
2865                 return -1;
2866
2867         for (i = 0; i < cnt; i++) {
2868                 struct cpu_cache_level c;
2869
2870                 #define _R(v)                                           \
2871                         if (do_read_u32(ff, &c.v))\
2872                                 goto out_free_caches;                   \
2873
2874                 _R(level)
2875                 _R(line_size)
2876                 _R(sets)
2877                 _R(ways)
2878                 #undef _R
2879
2880                 #define _R(v)                                   \
2881                         c.v = do_read_string(ff);               \
2882                         if (!c.v)                               \
2883                                 goto out_free_caches;
2884
2885                 _R(type)
2886                 _R(size)
2887                 _R(map)
2888                 #undef _R
2889
2890                 caches[i] = c;
2891         }
2892
2893         ff->ph->env.caches = caches;
2894         ff->ph->env.caches_cnt = cnt;
2895         return 0;
2896 out_free_caches:
2897         free(caches);
2898         return -1;
2899 }
2900
2901 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2902 {
2903         struct perf_session *session;
2904         u64 first_sample_time, last_sample_time;
2905         int ret;
2906
2907         session = container_of(ff->ph, struct perf_session, header);
2908
2909         ret = do_read_u64(ff, &first_sample_time);
2910         if (ret)
2911                 return -1;
2912
2913         ret = do_read_u64(ff, &last_sample_time);
2914         if (ret)
2915                 return -1;
2916
2917         session->evlist->first_sample_time = first_sample_time;
2918         session->evlist->last_sample_time = last_sample_time;
2919         return 0;
2920 }
2921
2922 static int process_mem_topology(struct feat_fd *ff,
2923                                 void *data __maybe_unused)
2924 {
2925         struct memory_node *nodes;
2926         u64 version, i, nr, bsize;
2927         int ret = -1;
2928
2929         if (do_read_u64(ff, &version))
2930                 return -1;
2931
2932         if (version != 1)
2933                 return -1;
2934
2935         if (do_read_u64(ff, &bsize))
2936                 return -1;
2937
2938         if (do_read_u64(ff, &nr))
2939                 return -1;
2940
2941         nodes = zalloc(sizeof(*nodes) * nr);
2942         if (!nodes)
2943                 return -1;
2944
2945         for (i = 0; i < nr; i++) {
2946                 struct memory_node n;
2947
2948                 #define _R(v)                           \
2949                         if (do_read_u64(ff, &n.v))      \
2950                                 goto out;               \
2951
2952                 _R(node)
2953                 _R(size)
2954
2955                 #undef _R
2956
2957                 if (do_read_bitmap(ff, &n.set, &n.size))
2958                         goto out;
2959
2960                 nodes[i] = n;
2961         }
2962
2963         ff->ph->env.memory_bsize    = bsize;
2964         ff->ph->env.memory_nodes    = nodes;
2965         ff->ph->env.nr_memory_nodes = nr;
2966         ret = 0;
2967
2968 out:
2969         if (ret)
2970                 free(nodes);
2971         return ret;
2972 }
2973
2974 static int process_clockid(struct feat_fd *ff,
2975                            void *data __maybe_unused)
2976 {
2977         if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
2978                 return -1;
2979
2980         return 0;
2981 }
2982
2983 static int process_clock_data(struct feat_fd *ff,
2984                               void *_data __maybe_unused)
2985 {
2986         u32 data32;
2987         u64 data64;
2988
2989         /* version */
2990         if (do_read_u32(ff, &data32))
2991                 return -1;
2992
2993         if (data32 != 1)
2994                 return -1;
2995
2996         /* clockid */
2997         if (do_read_u32(ff, &data32))
2998                 return -1;
2999
3000         ff->ph->env.clock.clockid = data32;
3001
3002         /* TOD ref time */
3003         if (do_read_u64(ff, &data64))
3004                 return -1;
3005
3006         ff->ph->env.clock.tod_ns = data64;
3007
3008         /* clockid ref time */
3009         if (do_read_u64(ff, &data64))
3010                 return -1;
3011
3012         ff->ph->env.clock.clockid_ns = data64;
3013         ff->ph->env.clock.enabled = true;
3014         return 0;
3015 }
3016
3017 static int process_hybrid_topology(struct feat_fd *ff,
3018                                    void *data __maybe_unused)
3019 {
3020         struct hybrid_node *nodes, *n;
3021         u32 nr, i;
3022
3023         /* nr nodes */
3024         if (do_read_u32(ff, &nr))
3025                 return -1;
3026
3027         nodes = zalloc(sizeof(*nodes) * nr);
3028         if (!nodes)
3029                 return -ENOMEM;
3030
3031         for (i = 0; i < nr; i++) {
3032                 n = &nodes[i];
3033
3034                 n->pmu_name = do_read_string(ff);
3035                 if (!n->pmu_name)
3036                         goto error;
3037
3038                 n->cpus = do_read_string(ff);
3039                 if (!n->cpus)
3040                         goto error;
3041         }
3042
3043         ff->ph->env.nr_hybrid_nodes = nr;
3044         ff->ph->env.hybrid_nodes = nodes;
3045         return 0;
3046
3047 error:
3048         for (i = 0; i < nr; i++) {
3049                 free(nodes[i].pmu_name);
3050                 free(nodes[i].cpus);
3051         }
3052
3053         free(nodes);
3054         return -1;
3055 }
3056
3057 static int process_dir_format(struct feat_fd *ff,
3058                               void *_data __maybe_unused)
3059 {
3060         struct perf_session *session;
3061         struct perf_data *data;
3062
3063         session = container_of(ff->ph, struct perf_session, header);
3064         data = session->data;
3065
3066         if (WARN_ON(!perf_data__is_dir(data)))
3067                 return -1;
3068
3069         return do_read_u64(ff, &data->dir.version);
3070 }
3071
3072 #ifdef HAVE_LIBBPF_SUPPORT
3073 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3074 {
3075         struct bpf_prog_info_node *info_node;
3076         struct perf_env *env = &ff->ph->env;
3077         struct perf_bpil *info_linear;
3078         u32 count, i;
3079         int err = -1;
3080
3081         if (ff->ph->needs_swap) {
3082                 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3083                 return 0;
3084         }
3085
3086         if (do_read_u32(ff, &count))
3087                 return -1;
3088
3089         down_write(&env->bpf_progs.lock);
3090
3091         for (i = 0; i < count; ++i) {
3092                 u32 info_len, data_len;
3093
3094                 info_linear = NULL;
3095                 info_node = NULL;
3096                 if (do_read_u32(ff, &info_len))
3097                         goto out;
3098                 if (do_read_u32(ff, &data_len))
3099                         goto out;
3100
3101                 if (info_len > sizeof(struct bpf_prog_info)) {
3102                         pr_warning("detected invalid bpf_prog_info\n");
3103                         goto out;
3104                 }
3105
3106                 info_linear = malloc(sizeof(struct perf_bpil) +
3107                                      data_len);
3108                 if (!info_linear)
3109                         goto out;
3110                 info_linear->info_len = sizeof(struct bpf_prog_info);
3111                 info_linear->data_len = data_len;
3112                 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3113                         goto out;
3114                 if (__do_read(ff, &info_linear->info, info_len))
3115                         goto out;
3116                 if (info_len < sizeof(struct bpf_prog_info))
3117                         memset(((void *)(&info_linear->info)) + info_len, 0,
3118                                sizeof(struct bpf_prog_info) - info_len);
3119
3120                 if (__do_read(ff, info_linear->data, data_len))
3121                         goto out;
3122
3123                 info_node = malloc(sizeof(struct bpf_prog_info_node));
3124                 if (!info_node)
3125                         goto out;
3126
3127                 /* after reading from file, translate offset to address */
3128                 bpil_offs_to_addr(info_linear);
3129                 info_node->info_linear = info_linear;
3130                 perf_env__insert_bpf_prog_info(env, info_node);
3131         }
3132
3133         up_write(&env->bpf_progs.lock);
3134         return 0;
3135 out:
3136         free(info_linear);
3137         free(info_node);
3138         up_write(&env->bpf_progs.lock);
3139         return err;
3140 }
3141
3142 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3143 {
3144         struct perf_env *env = &ff->ph->env;
3145         struct btf_node *node = NULL;
3146         u32 count, i;
3147         int err = -1;
3148
3149         if (ff->ph->needs_swap) {
3150                 pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3151                 return 0;
3152         }
3153
3154         if (do_read_u32(ff, &count))
3155                 return -1;
3156
3157         down_write(&env->bpf_progs.lock);
3158
3159         for (i = 0; i < count; ++i) {
3160                 u32 id, data_size;
3161
3162                 if (do_read_u32(ff, &id))
3163                         goto out;
3164                 if (do_read_u32(ff, &data_size))
3165                         goto out;
3166
3167                 node = malloc(sizeof(struct btf_node) + data_size);
3168                 if (!node)
3169                         goto out;
3170
3171                 node->id = id;
3172                 node->data_size = data_size;
3173
3174                 if (__do_read(ff, node->data, data_size))
3175                         goto out;
3176
3177                 perf_env__insert_btf(env, node);
3178                 node = NULL;
3179         }
3180
3181         err = 0;
3182 out:
3183         up_write(&env->bpf_progs.lock);
3184         free(node);
3185         return err;
3186 }
3187 #endif // HAVE_LIBBPF_SUPPORT
3188
3189 static int process_compressed(struct feat_fd *ff,
3190                               void *data __maybe_unused)
3191 {
3192         if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3193                 return -1;
3194
3195         if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3196                 return -1;
3197
3198         if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3199                 return -1;
3200
3201         if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3202                 return -1;
3203
3204         if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3205                 return -1;
3206
3207         return 0;
3208 }
3209
3210 static int process_per_cpu_pmu_caps(struct feat_fd *ff, int *nr_cpu_pmu_caps,
3211                                     char **cpu_pmu_caps,
3212                                     unsigned int *max_branches)
3213 {
3214         char *name, *value;
3215         struct strbuf sb;
3216         u32 nr_caps;
3217
3218         if (do_read_u32(ff, &nr_caps))
3219                 return -1;
3220
3221         if (!nr_caps) {
3222                 pr_debug("cpu pmu capabilities not available\n");
3223                 return 0;
3224         }
3225
3226         *nr_cpu_pmu_caps = nr_caps;
3227
3228         if (strbuf_init(&sb, 128) < 0)
3229                 return -1;
3230
3231         while (nr_caps--) {
3232                 name = do_read_string(ff);
3233                 if (!name)
3234                         goto error;
3235
3236                 value = do_read_string(ff);
3237                 if (!value)
3238                         goto free_name;
3239
3240                 if (strbuf_addf(&sb, "%s=%s", name, value) < 0)
3241                         goto free_value;
3242
3243                 /* include a NULL character at the end */
3244                 if (strbuf_add(&sb, "", 1) < 0)
3245                         goto free_value;
3246
3247                 if (!strcmp(name, "branches"))
3248                         *max_branches = atoi(value);
3249
3250                 free(value);
3251                 free(name);
3252         }
3253         *cpu_pmu_caps = strbuf_detach(&sb, NULL);
3254         return 0;
3255
3256 free_value:
3257         free(value);
3258 free_name:
3259         free(name);
3260 error:
3261         strbuf_release(&sb);
3262         return -1;
3263 }
3264
3265 static int process_cpu_pmu_caps(struct feat_fd *ff,
3266                                 void *data __maybe_unused)
3267 {
3268         return process_per_cpu_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3269                                         &ff->ph->env.cpu_pmu_caps,
3270                                         &ff->ph->env.max_branches);
3271 }
3272
3273 static int process_hybrid_cpu_pmu_caps(struct feat_fd *ff,
3274                                        void *data __maybe_unused)
3275 {
3276         struct hybrid_cpc_node *nodes;
3277         u32 nr_pmu, i;
3278         int ret;
3279
3280         if (do_read_u32(ff, &nr_pmu))
3281                 return -1;
3282
3283         if (!nr_pmu) {
3284                 pr_debug("hybrid cpu pmu capabilities not available\n");
3285                 return 0;
3286         }
3287
3288         nodes = zalloc(sizeof(*nodes) * nr_pmu);
3289         if (!nodes)
3290                 return -ENOMEM;
3291
3292         for (i = 0; i < nr_pmu; i++) {
3293                 struct hybrid_cpc_node *n = &nodes[i];
3294
3295                 ret = process_per_cpu_pmu_caps(ff, &n->nr_cpu_pmu_caps,
3296                                                &n->cpu_pmu_caps,
3297                                                &n->max_branches);
3298                 if (ret)
3299                         goto err;
3300
3301                 n->pmu_name = do_read_string(ff);
3302                 if (!n->pmu_name) {
3303                         ret = -1;
3304                         goto err;
3305                 }
3306         }
3307
3308         ff->ph->env.nr_hybrid_cpc_nodes = nr_pmu;
3309         ff->ph->env.hybrid_cpc_nodes = nodes;
3310         return 0;
3311
3312 err:
3313         for (i = 0; i < nr_pmu; i++) {
3314                 free(nodes[i].cpu_pmu_caps);
3315                 free(nodes[i].pmu_name);
3316         }
3317
3318         free(nodes);
3319         return ret;
3320 }
3321
3322 #define FEAT_OPR(n, func, __full_only) \
3323         [HEADER_##n] = {                                        \
3324                 .name       = __stringify(n),                   \
3325                 .write      = write_##func,                     \
3326                 .print      = print_##func,                     \
3327                 .full_only  = __full_only,                      \
3328                 .process    = process_##func,                   \
3329                 .synthesize = true                              \
3330         }
3331
3332 #define FEAT_OPN(n, func, __full_only) \
3333         [HEADER_##n] = {                                        \
3334                 .name       = __stringify(n),                   \
3335                 .write      = write_##func,                     \
3336                 .print      = print_##func,                     \
3337                 .full_only  = __full_only,                      \
3338                 .process    = process_##func                    \
3339         }
3340
3341 /* feature_ops not implemented: */
3342 #define print_tracing_data      NULL
3343 #define print_build_id          NULL
3344
3345 #define process_branch_stack    NULL
3346 #define process_stat            NULL
3347
3348 // Only used in util/synthetic-events.c
3349 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3350
3351 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3352         FEAT_OPN(TRACING_DATA,  tracing_data,   false),
3353         FEAT_OPN(BUILD_ID,      build_id,       false),
3354         FEAT_OPR(HOSTNAME,      hostname,       false),
3355         FEAT_OPR(OSRELEASE,     osrelease,      false),
3356         FEAT_OPR(VERSION,       version,        false),
3357         FEAT_OPR(ARCH,          arch,           false),
3358         FEAT_OPR(NRCPUS,        nrcpus,         false),
3359         FEAT_OPR(CPUDESC,       cpudesc,        false),
3360         FEAT_OPR(CPUID,         cpuid,          false),
3361         FEAT_OPR(TOTAL_MEM,     total_mem,      false),
3362         FEAT_OPR(EVENT_DESC,    event_desc,     false),
3363         FEAT_OPR(CMDLINE,       cmdline,        false),
3364         FEAT_OPR(CPU_TOPOLOGY,  cpu_topology,   true),
3365         FEAT_OPR(NUMA_TOPOLOGY, numa_topology,  true),
3366         FEAT_OPN(BRANCH_STACK,  branch_stack,   false),
3367         FEAT_OPR(PMU_MAPPINGS,  pmu_mappings,   false),
3368         FEAT_OPR(GROUP_DESC,    group_desc,     false),
3369         FEAT_OPN(AUXTRACE,      auxtrace,       false),
3370         FEAT_OPN(STAT,          stat,           false),
3371         FEAT_OPN(CACHE,         cache,          true),
3372         FEAT_OPR(SAMPLE_TIME,   sample_time,    false),
3373         FEAT_OPR(MEM_TOPOLOGY,  mem_topology,   true),
3374         FEAT_OPR(CLOCKID,       clockid,        false),
3375         FEAT_OPN(DIR_FORMAT,    dir_format,     false),
3376 #ifdef HAVE_LIBBPF_SUPPORT
3377         FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
3378         FEAT_OPR(BPF_BTF,       bpf_btf,        false),
3379 #endif
3380         FEAT_OPR(COMPRESSED,    compressed,     false),
3381         FEAT_OPR(CPU_PMU_CAPS,  cpu_pmu_caps,   false),
3382         FEAT_OPR(CLOCK_DATA,    clock_data,     false),
3383         FEAT_OPN(HYBRID_TOPOLOGY,       hybrid_topology,        true),
3384         FEAT_OPR(HYBRID_CPU_PMU_CAPS,   hybrid_cpu_pmu_caps,    false),
3385 };
3386
3387 struct header_print_data {
3388         FILE *fp;
3389         bool full; /* extended list of headers */
3390 };
3391
3392 static int perf_file_section__fprintf_info(struct perf_file_section *section,
3393                                            struct perf_header *ph,
3394                                            int feat, int fd, void *data)
3395 {
3396         struct header_print_data *hd = data;
3397         struct feat_fd ff;
3398
3399         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3400                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3401                                 "%d, continuing...\n", section->offset, feat);
3402                 return 0;
3403         }
3404         if (feat >= HEADER_LAST_FEATURE) {
3405                 pr_warning("unknown feature %d\n", feat);
3406                 return 0;
3407         }
3408         if (!feat_ops[feat].print)
3409                 return 0;
3410
3411         ff = (struct  feat_fd) {
3412                 .fd = fd,
3413                 .ph = ph,
3414         };
3415
3416         if (!feat_ops[feat].full_only || hd->full)
3417                 feat_ops[feat].print(&ff, hd->fp);
3418         else
3419                 fprintf(hd->fp, "# %s info available, use -I to display\n",
3420                         feat_ops[feat].name);
3421
3422         return 0;
3423 }
3424
3425 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3426 {
3427         struct header_print_data hd;
3428         struct perf_header *header = &session->header;
3429         int fd = perf_data__fd(session->data);
3430         struct stat st;
3431         time_t stctime;
3432         int ret, bit;
3433
3434         hd.fp = fp;
3435         hd.full = full;
3436
3437         ret = fstat(fd, &st);
3438         if (ret == -1)
3439                 return -1;
3440
3441         stctime = st.st_mtime;
3442         fprintf(fp, "# captured on    : %s", ctime(&stctime));
3443
3444         fprintf(fp, "# header version : %u\n", header->version);
3445         fprintf(fp, "# data offset    : %" PRIu64 "\n", header->data_offset);
3446         fprintf(fp, "# data size      : %" PRIu64 "\n", header->data_size);
3447         fprintf(fp, "# feat offset    : %" PRIu64 "\n", header->feat_offset);
3448
3449         perf_header__process_sections(header, fd, &hd,
3450                                       perf_file_section__fprintf_info);
3451
3452         if (session->data->is_pipe)
3453                 return 0;
3454
3455         fprintf(fp, "# missing features: ");
3456         for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3457                 if (bit)
3458                         fprintf(fp, "%s ", feat_ops[bit].name);
3459         }
3460
3461         fprintf(fp, "\n");
3462         return 0;
3463 }
3464
3465 struct header_fw {
3466         struct feat_writer      fw;
3467         struct feat_fd          *ff;
3468 };
3469
3470 static int feat_writer_cb(struct feat_writer *fw, void *buf, size_t sz)
3471 {
3472         struct header_fw *h = container_of(fw, struct header_fw, fw);
3473
3474         return do_write(h->ff, buf, sz);
3475 }
3476
3477 static int do_write_feat(struct feat_fd *ff, int type,
3478                          struct perf_file_section **p,
3479                          struct evlist *evlist,
3480                          struct feat_copier *fc)
3481 {
3482         int err;
3483         int ret = 0;
3484
3485         if (perf_header__has_feat(ff->ph, type)) {
3486                 if (!feat_ops[type].write)
3487                         return -1;
3488
3489                 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3490                         return -1;
3491
3492                 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3493
3494                 /*
3495                  * Hook to let perf inject copy features sections from the input
3496                  * file.
3497                  */
3498                 if (fc && fc->copy) {
3499                         struct header_fw h = {
3500                                 .fw.write = feat_writer_cb,
3501                                 .ff = ff,
3502                         };
3503
3504                         /* ->copy() returns 0 if the feature was not copied */
3505                         err = fc->copy(fc, type, &h.fw);
3506                 } else {
3507                         err = 0;
3508                 }
3509                 if (!err)
3510                         err = feat_ops[type].write(ff, evlist);
3511                 if (err < 0) {
3512                         pr_debug("failed to write feature %s\n", feat_ops[type].name);
3513
3514                         /* undo anything written */
3515                         lseek(ff->fd, (*p)->offset, SEEK_SET);
3516
3517                         return -1;
3518                 }
3519                 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3520                 (*p)++;
3521         }
3522         return ret;
3523 }
3524
3525 static int perf_header__adds_write(struct perf_header *header,
3526                                    struct evlist *evlist, int fd,
3527                                    struct feat_copier *fc)
3528 {
3529         int nr_sections;
3530         struct feat_fd ff;
3531         struct perf_file_section *feat_sec, *p;
3532         int sec_size;
3533         u64 sec_start;
3534         int feat;
3535         int err;
3536
3537         ff = (struct feat_fd){
3538                 .fd  = fd,
3539                 .ph = header,
3540         };
3541
3542         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3543         if (!nr_sections)
3544                 return 0;
3545
3546         feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3547         if (feat_sec == NULL)
3548                 return -ENOMEM;
3549
3550         sec_size = sizeof(*feat_sec) * nr_sections;
3551
3552         sec_start = header->feat_offset;
3553         lseek(fd, sec_start + sec_size, SEEK_SET);
3554
3555         for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3556                 if (do_write_feat(&ff, feat, &p, evlist, fc))
3557                         perf_header__clear_feat(header, feat);
3558         }
3559
3560         lseek(fd, sec_start, SEEK_SET);
3561         /*
3562          * may write more than needed due to dropped feature, but
3563          * this is okay, reader will skip the missing entries
3564          */
3565         err = do_write(&ff, feat_sec, sec_size);
3566         if (err < 0)
3567                 pr_debug("failed to write feature section\n");
3568         free(feat_sec);
3569         return err;
3570 }
3571
3572 int perf_header__write_pipe(int fd)
3573 {
3574         struct perf_pipe_file_header f_header;
3575         struct feat_fd ff;
3576         int err;
3577
3578         ff = (struct feat_fd){ .fd = fd };
3579
3580         f_header = (struct perf_pipe_file_header){
3581                 .magic     = PERF_MAGIC,
3582                 .size      = sizeof(f_header),
3583         };
3584
3585         err = do_write(&ff, &f_header, sizeof(f_header));
3586         if (err < 0) {
3587                 pr_debug("failed to write perf pipe header\n");
3588                 return err;
3589         }
3590
3591         return 0;
3592 }
3593
3594 static int perf_session__do_write_header(struct perf_session *session,
3595                                          struct evlist *evlist,
3596                                          int fd, bool at_exit,
3597                                          struct feat_copier *fc)
3598 {
3599         struct perf_file_header f_header;
3600         struct perf_file_attr   f_attr;
3601         struct perf_header *header = &session->header;
3602         struct evsel *evsel;
3603         struct feat_fd ff;
3604         u64 attr_offset;
3605         int err;
3606
3607         ff = (struct feat_fd){ .fd = fd};
3608         lseek(fd, sizeof(f_header), SEEK_SET);
3609
3610         evlist__for_each_entry(session->evlist, evsel) {
3611                 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3612                 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3613                 if (err < 0) {
3614                         pr_debug("failed to write perf header\n");
3615                         return err;
3616                 }
3617         }
3618
3619         attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3620
3621         evlist__for_each_entry(evlist, evsel) {
3622                 if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3623                         /*
3624                          * We are likely in "perf inject" and have read
3625                          * from an older file. Update attr size so that
3626                          * reader gets the right offset to the ids.
3627                          */
3628                         evsel->core.attr.size = sizeof(evsel->core.attr);
3629                 }
3630                 f_attr = (struct perf_file_attr){
3631                         .attr = evsel->core.attr,
3632                         .ids  = {
3633                                 .offset = evsel->id_offset,
3634                                 .size   = evsel->core.ids * sizeof(u64),
3635                         }
3636                 };
3637                 err = do_write(&ff, &f_attr, sizeof(f_attr));
3638                 if (err < 0) {
3639                         pr_debug("failed to write perf header attribute\n");
3640                         return err;
3641                 }
3642         }
3643
3644         if (!header->data_offset)
3645                 header->data_offset = lseek(fd, 0, SEEK_CUR);
3646         header->feat_offset = header->data_offset + header->data_size;
3647
3648         if (at_exit) {
3649                 err = perf_header__adds_write(header, evlist, fd, fc);
3650                 if (err < 0)
3651                         return err;
3652         }
3653
3654         f_header = (struct perf_file_header){
3655                 .magic     = PERF_MAGIC,
3656                 .size      = sizeof(f_header),
3657                 .attr_size = sizeof(f_attr),
3658                 .attrs = {
3659                         .offset = attr_offset,
3660                         .size   = evlist->core.nr_entries * sizeof(f_attr),
3661                 },
3662                 .data = {
3663                         .offset = header->data_offset,
3664                         .size   = header->data_size,
3665                 },
3666                 /* event_types is ignored, store zeros */
3667         };
3668
3669         memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3670
3671         lseek(fd, 0, SEEK_SET);
3672         err = do_write(&ff, &f_header, sizeof(f_header));
3673         if (err < 0) {
3674                 pr_debug("failed to write perf header\n");
3675                 return err;
3676         }
3677         lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3678
3679         return 0;
3680 }
3681
3682 int perf_session__write_header(struct perf_session *session,
3683                                struct evlist *evlist,
3684                                int fd, bool at_exit)
3685 {
3686         return perf_session__do_write_header(session, evlist, fd, at_exit, NULL);
3687 }
3688
3689 size_t perf_session__data_offset(const struct evlist *evlist)
3690 {
3691         struct evsel *evsel;
3692         size_t data_offset;
3693
3694         data_offset = sizeof(struct perf_file_header);
3695         evlist__for_each_entry(evlist, evsel) {
3696                 data_offset += evsel->core.ids * sizeof(u64);
3697         }
3698         data_offset += evlist->core.nr_entries * sizeof(struct perf_file_attr);
3699
3700         return data_offset;
3701 }
3702
3703 int perf_session__inject_header(struct perf_session *session,
3704                                 struct evlist *evlist,
3705                                 int fd,
3706                                 struct feat_copier *fc)
3707 {
3708         return perf_session__do_write_header(session, evlist, fd, true, fc);
3709 }
3710
3711 static int perf_header__getbuffer64(struct perf_header *header,
3712                                     int fd, void *buf, size_t size)
3713 {
3714         if (readn(fd, buf, size) <= 0)
3715                 return -1;
3716
3717         if (header->needs_swap)
3718                 mem_bswap_64(buf, size);
3719
3720         return 0;
3721 }
3722
3723 int perf_header__process_sections(struct perf_header *header, int fd,
3724                                   void *data,
3725                                   int (*process)(struct perf_file_section *section,
3726                                                  struct perf_header *ph,
3727                                                  int feat, int fd, void *data))
3728 {
3729         struct perf_file_section *feat_sec, *sec;
3730         int nr_sections;
3731         int sec_size;
3732         int feat;
3733         int err;
3734
3735         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3736         if (!nr_sections)
3737                 return 0;
3738
3739         feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3740         if (!feat_sec)
3741                 return -1;
3742
3743         sec_size = sizeof(*feat_sec) * nr_sections;
3744
3745         lseek(fd, header->feat_offset, SEEK_SET);
3746
3747         err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3748         if (err < 0)
3749                 goto out_free;
3750
3751         for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3752                 err = process(sec++, header, feat, fd, data);
3753                 if (err < 0)
3754                         goto out_free;
3755         }
3756         err = 0;
3757 out_free:
3758         free(feat_sec);
3759         return err;
3760 }
3761
3762 static const int attr_file_abi_sizes[] = {
3763         [0] = PERF_ATTR_SIZE_VER0,
3764         [1] = PERF_ATTR_SIZE_VER1,
3765         [2] = PERF_ATTR_SIZE_VER2,
3766         [3] = PERF_ATTR_SIZE_VER3,
3767         [4] = PERF_ATTR_SIZE_VER4,
3768         0,
3769 };
3770
3771 /*
3772  * In the legacy file format, the magic number is not used to encode endianness.
3773  * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3774  * on ABI revisions, we need to try all combinations for all endianness to
3775  * detect the endianness.
3776  */
3777 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3778 {
3779         uint64_t ref_size, attr_size;
3780         int i;
3781
3782         for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3783                 ref_size = attr_file_abi_sizes[i]
3784                          + sizeof(struct perf_file_section);
3785                 if (hdr_sz != ref_size) {
3786                         attr_size = bswap_64(hdr_sz);
3787                         if (attr_size != ref_size)
3788                                 continue;
3789
3790                         ph->needs_swap = true;
3791                 }
3792                 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3793                          i,
3794                          ph->needs_swap);
3795                 return 0;
3796         }
3797         /* could not determine endianness */
3798         return -1;
3799 }
3800
3801 #define PERF_PIPE_HDR_VER0      16
3802
3803 static const size_t attr_pipe_abi_sizes[] = {
3804         [0] = PERF_PIPE_HDR_VER0,
3805         0,
3806 };
3807
3808 /*
3809  * In the legacy pipe format, there is an implicit assumption that endianness
3810  * between host recording the samples, and host parsing the samples is the
3811  * same. This is not always the case given that the pipe output may always be
3812  * redirected into a file and analyzed on a different machine with possibly a
3813  * different endianness and perf_event ABI revisions in the perf tool itself.
3814  */
3815 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3816 {
3817         u64 attr_size;
3818         int i;
3819
3820         for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3821                 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3822                         attr_size = bswap_64(hdr_sz);
3823                         if (attr_size != hdr_sz)
3824                                 continue;
3825
3826                         ph->needs_swap = true;
3827                 }
3828                 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3829                 return 0;
3830         }
3831         return -1;
3832 }
3833
3834 bool is_perf_magic(u64 magic)
3835 {
3836         if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3837                 || magic == __perf_magic2
3838                 || magic == __perf_magic2_sw)
3839                 return true;
3840
3841         return false;
3842 }
3843
3844 static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3845                               bool is_pipe, struct perf_header *ph)
3846 {
3847         int ret;
3848
3849         /* check for legacy format */
3850         ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3851         if (ret == 0) {
3852                 ph->version = PERF_HEADER_VERSION_1;
3853                 pr_debug("legacy perf.data format\n");
3854                 if (is_pipe)
3855                         return try_all_pipe_abis(hdr_sz, ph);
3856
3857                 return try_all_file_abis(hdr_sz, ph);
3858         }
3859         /*
3860          * the new magic number serves two purposes:
3861          * - unique number to identify actual perf.data files
3862          * - encode endianness of file
3863          */
3864         ph->version = PERF_HEADER_VERSION_2;
3865
3866         /* check magic number with one endianness */
3867         if (magic == __perf_magic2)
3868                 return 0;
3869
3870         /* check magic number with opposite endianness */
3871         if (magic != __perf_magic2_sw)
3872                 return -1;
3873
3874         ph->needs_swap = true;
3875
3876         return 0;
3877 }
3878
3879 int perf_file_header__read(struct perf_file_header *header,
3880                            struct perf_header *ph, int fd)
3881 {
3882         ssize_t ret;
3883
3884         lseek(fd, 0, SEEK_SET);
3885
3886         ret = readn(fd, header, sizeof(*header));
3887         if (ret <= 0)
3888                 return -1;
3889
3890         if (check_magic_endian(header->magic,
3891                                header->attr_size, false, ph) < 0) {
3892                 pr_debug("magic/endian check failed\n");
3893                 return -1;
3894         }
3895
3896         if (ph->needs_swap) {
3897                 mem_bswap_64(header, offsetof(struct perf_file_header,
3898                              adds_features));
3899         }
3900
3901         if (header->size != sizeof(*header)) {
3902                 /* Support the previous format */
3903                 if (header->size == offsetof(typeof(*header), adds_features))
3904                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3905                 else
3906                         return -1;
3907         } else if (ph->needs_swap) {
3908                 /*
3909                  * feature bitmap is declared as an array of unsigned longs --
3910                  * not good since its size can differ between the host that
3911                  * generated the data file and the host analyzing the file.
3912                  *
3913                  * We need to handle endianness, but we don't know the size of
3914                  * the unsigned long where the file was generated. Take a best
3915                  * guess at determining it: try 64-bit swap first (ie., file
3916                  * created on a 64-bit host), and check if the hostname feature
3917                  * bit is set (this feature bit is forced on as of fbe96f2).
3918                  * If the bit is not, undo the 64-bit swap and try a 32-bit
3919                  * swap. If the hostname bit is still not set (e.g., older data
3920                  * file), punt and fallback to the original behavior --
3921                  * clearing all feature bits and setting buildid.
3922                  */
3923                 mem_bswap_64(&header->adds_features,
3924                             BITS_TO_U64(HEADER_FEAT_BITS));
3925
3926                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3927                         /* unswap as u64 */
3928                         mem_bswap_64(&header->adds_features,
3929                                     BITS_TO_U64(HEADER_FEAT_BITS));
3930
3931                         /* unswap as u32 */
3932                         mem_bswap_32(&header->adds_features,
3933                                     BITS_TO_U32(HEADER_FEAT_BITS));
3934                 }
3935
3936                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3937                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3938                         set_bit(HEADER_BUILD_ID, header->adds_features);
3939                 }
3940         }
3941
3942         memcpy(&ph->adds_features, &header->adds_features,
3943                sizeof(ph->adds_features));
3944
3945         ph->data_offset  = header->data.offset;
3946         ph->data_size    = header->data.size;
3947         ph->feat_offset  = header->data.offset + header->data.size;
3948         return 0;
3949 }
3950
3951 static int perf_file_section__process(struct perf_file_section *section,
3952                                       struct perf_header *ph,
3953                                       int feat, int fd, void *data)
3954 {
3955         struct feat_fd fdd = {
3956                 .fd     = fd,
3957                 .ph     = ph,
3958                 .size   = section->size,
3959                 .offset = section->offset,
3960         };
3961
3962         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3963                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3964                           "%d, continuing...\n", section->offset, feat);
3965                 return 0;
3966         }
3967
3968         if (feat >= HEADER_LAST_FEATURE) {
3969                 pr_debug("unknown feature %d, continuing...\n", feat);
3970                 return 0;
3971         }
3972
3973         if (!feat_ops[feat].process)
3974                 return 0;
3975
3976         return feat_ops[feat].process(&fdd, data);
3977 }
3978
3979 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
3980                                        struct perf_header *ph,
3981                                        struct perf_data* data,
3982                                        bool repipe, int repipe_fd)
3983 {
3984         struct feat_fd ff = {
3985                 .fd = repipe_fd,
3986                 .ph = ph,
3987         };
3988         ssize_t ret;
3989
3990         ret = perf_data__read(data, header, sizeof(*header));
3991         if (ret <= 0)
3992                 return -1;
3993
3994         if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
3995                 pr_debug("endian/magic failed\n");
3996                 return -1;
3997         }
3998
3999         if (ph->needs_swap)
4000                 header->size = bswap_64(header->size);
4001
4002         if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
4003                 return -1;
4004
4005         return 0;
4006 }
4007
4008 static int perf_header__read_pipe(struct perf_session *session, int repipe_fd)
4009 {
4010         struct perf_header *header = &session->header;
4011         struct perf_pipe_file_header f_header;
4012
4013         if (perf_file_header__read_pipe(&f_header, header, session->data,
4014                                         session->repipe, repipe_fd) < 0) {
4015                 pr_debug("incompatible file format\n");
4016                 return -EINVAL;
4017         }
4018
4019         return f_header.size == sizeof(f_header) ? 0 : -1;
4020 }
4021
4022 static int read_attr(int fd, struct perf_header *ph,
4023                      struct perf_file_attr *f_attr)
4024 {
4025         struct perf_event_attr *attr = &f_attr->attr;
4026         size_t sz, left;
4027         size_t our_sz = sizeof(f_attr->attr);
4028         ssize_t ret;
4029
4030         memset(f_attr, 0, sizeof(*f_attr));
4031
4032         /* read minimal guaranteed structure */
4033         ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
4034         if (ret <= 0) {
4035                 pr_debug("cannot read %d bytes of header attr\n",
4036                          PERF_ATTR_SIZE_VER0);
4037                 return -1;
4038         }
4039
4040         /* on file perf_event_attr size */
4041         sz = attr->size;
4042
4043         if (ph->needs_swap)
4044                 sz = bswap_32(sz);
4045
4046         if (sz == 0) {
4047                 /* assume ABI0 */
4048                 sz =  PERF_ATTR_SIZE_VER0;
4049         } else if (sz > our_sz) {
4050                 pr_debug("file uses a more recent and unsupported ABI"
4051                          " (%zu bytes extra)\n", sz - our_sz);
4052                 return -1;
4053         }
4054         /* what we have not yet read and that we know about */
4055         left = sz - PERF_ATTR_SIZE_VER0;
4056         if (left) {
4057                 void *ptr = attr;
4058                 ptr += PERF_ATTR_SIZE_VER0;
4059
4060                 ret = readn(fd, ptr, left);
4061         }
4062         /* read perf_file_section, ids are read in caller */
4063         ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
4064
4065         return ret <= 0 ? -1 : 0;
4066 }
4067
4068 static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
4069 {
4070         struct tep_event *event;
4071         char bf[128];
4072
4073         /* already prepared */
4074         if (evsel->tp_format)
4075                 return 0;
4076
4077         if (pevent == NULL) {
4078                 pr_debug("broken or missing trace data\n");
4079                 return -1;
4080         }
4081
4082         event = tep_find_event(pevent, evsel->core.attr.config);
4083         if (event == NULL) {
4084                 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
4085                 return -1;
4086         }
4087
4088         if (!evsel->name) {
4089                 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
4090                 evsel->name = strdup(bf);
4091                 if (evsel->name == NULL)
4092                         return -1;
4093         }
4094
4095         evsel->tp_format = event;
4096         return 0;
4097 }
4098
4099 static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
4100 {
4101         struct evsel *pos;
4102
4103         evlist__for_each_entry(evlist, pos) {
4104                 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
4105                     evsel__prepare_tracepoint_event(pos, pevent))
4106                         return -1;
4107         }
4108
4109         return 0;
4110 }
4111
4112 int perf_session__read_header(struct perf_session *session, int repipe_fd)
4113 {
4114         struct perf_data *data = session->data;
4115         struct perf_header *header = &session->header;
4116         struct perf_file_header f_header;
4117         struct perf_file_attr   f_attr;
4118         u64                     f_id;
4119         int nr_attrs, nr_ids, i, j, err;
4120         int fd = perf_data__fd(data);
4121
4122         session->evlist = evlist__new();
4123         if (session->evlist == NULL)
4124                 return -ENOMEM;
4125
4126         session->evlist->env = &header->env;
4127         session->machines.host.env = &header->env;
4128
4129         /*
4130          * We can read 'pipe' data event from regular file,
4131          * check for the pipe header regardless of source.
4132          */
4133         err = perf_header__read_pipe(session, repipe_fd);
4134         if (!err || perf_data__is_pipe(data)) {
4135                 data->is_pipe = true;
4136                 return err;
4137         }
4138
4139         if (perf_file_header__read(&f_header, header, fd) < 0)
4140                 return -EINVAL;
4141
4142         if (header->needs_swap && data->in_place_update) {
4143                 pr_err("In-place update not supported when byte-swapping is required\n");
4144                 return -EINVAL;
4145         }
4146
4147         /*
4148          * Sanity check that perf.data was written cleanly; data size is
4149          * initialized to 0 and updated only if the on_exit function is run.
4150          * If data size is still 0 then the file contains only partial
4151          * information.  Just warn user and process it as much as it can.
4152          */
4153         if (f_header.data.size == 0) {
4154                 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4155                            "Was the 'perf record' command properly terminated?\n",
4156                            data->file.path);
4157         }
4158
4159         if (f_header.attr_size == 0) {
4160                 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4161                        "Was the 'perf record' command properly terminated?\n",
4162                        data->file.path);
4163                 return -EINVAL;
4164         }
4165
4166         nr_attrs = f_header.attrs.size / f_header.attr_size;
4167         lseek(fd, f_header.attrs.offset, SEEK_SET);
4168
4169         for (i = 0; i < nr_attrs; i++) {
4170                 struct evsel *evsel;
4171                 off_t tmp;
4172
4173                 if (read_attr(fd, header, &f_attr) < 0)
4174                         goto out_errno;
4175
4176                 if (header->needs_swap) {
4177                         f_attr.ids.size   = bswap_64(f_attr.ids.size);
4178                         f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4179                         perf_event__attr_swap(&f_attr.attr);
4180                 }
4181
4182                 tmp = lseek(fd, 0, SEEK_CUR);
4183                 evsel = evsel__new(&f_attr.attr);
4184
4185                 if (evsel == NULL)
4186                         goto out_delete_evlist;
4187
4188                 evsel->needs_swap = header->needs_swap;
4189                 /*
4190                  * Do it before so that if perf_evsel__alloc_id fails, this
4191                  * entry gets purged too at evlist__delete().
4192                  */
4193                 evlist__add(session->evlist, evsel);
4194
4195                 nr_ids = f_attr.ids.size / sizeof(u64);
4196                 /*
4197                  * We don't have the cpu and thread maps on the header, so
4198                  * for allocating the perf_sample_id table we fake 1 cpu and
4199                  * hattr->ids threads.
4200                  */
4201                 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4202                         goto out_delete_evlist;
4203
4204                 lseek(fd, f_attr.ids.offset, SEEK_SET);
4205
4206                 for (j = 0; j < nr_ids; j++) {
4207                         if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4208                                 goto out_errno;
4209
4210                         perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4211                 }
4212
4213                 lseek(fd, tmp, SEEK_SET);
4214         }
4215
4216         perf_header__process_sections(header, fd, &session->tevent,
4217                                       perf_file_section__process);
4218
4219         if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4220                 goto out_delete_evlist;
4221
4222         return 0;
4223 out_errno:
4224         return -errno;
4225
4226 out_delete_evlist:
4227         evlist__delete(session->evlist);
4228         session->evlist = NULL;
4229         return -ENOMEM;
4230 }
4231
4232 int perf_event__process_feature(struct perf_session *session,
4233                                 union perf_event *event)
4234 {
4235         struct perf_tool *tool = session->tool;
4236         struct feat_fd ff = { .fd = 0 };
4237         struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4238         int type = fe->header.type;
4239         u64 feat = fe->feat_id;
4240         int ret = 0;
4241
4242         if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4243                 pr_warning("invalid record type %d in pipe-mode\n", type);
4244                 return 0;
4245         }
4246         if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4247                 pr_warning("invalid record type %d in pipe-mode\n", type);
4248                 return -1;
4249         }
4250
4251         if (!feat_ops[feat].process)
4252                 return 0;
4253
4254         ff.buf  = (void *)fe->data;
4255         ff.size = event->header.size - sizeof(*fe);
4256         ff.ph = &session->header;
4257
4258         if (feat_ops[feat].process(&ff, NULL)) {
4259                 ret = -1;
4260                 goto out;
4261         }
4262
4263         if (!feat_ops[feat].print || !tool->show_feat_hdr)
4264                 goto out;
4265
4266         if (!feat_ops[feat].full_only ||
4267             tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4268                 feat_ops[feat].print(&ff, stdout);
4269         } else {
4270                 fprintf(stdout, "# %s info available, use -I to display\n",
4271                         feat_ops[feat].name);
4272         }
4273 out:
4274         free_event_desc(ff.events);
4275         return ret;
4276 }
4277
4278 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4279 {
4280         struct perf_record_event_update *ev = &event->event_update;
4281         struct perf_record_event_update_scale *ev_scale;
4282         struct perf_record_event_update_cpus *ev_cpus;
4283         struct perf_cpu_map *map;
4284         size_t ret;
4285
4286         ret = fprintf(fp, "\n... id:    %" PRI_lu64 "\n", ev->id);
4287
4288         switch (ev->type) {
4289         case PERF_EVENT_UPDATE__SCALE:
4290                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4291                 ret += fprintf(fp, "... scale: %f\n", ev_scale->scale);
4292                 break;
4293         case PERF_EVENT_UPDATE__UNIT:
4294                 ret += fprintf(fp, "... unit:  %s\n", ev->data);
4295                 break;
4296         case PERF_EVENT_UPDATE__NAME:
4297                 ret += fprintf(fp, "... name:  %s\n", ev->data);
4298                 break;
4299         case PERF_EVENT_UPDATE__CPUS:
4300                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4301                 ret += fprintf(fp, "... ");
4302
4303                 map = cpu_map__new_data(&ev_cpus->cpus);
4304                 if (map)
4305                         ret += cpu_map__fprintf(map, fp);
4306                 else
4307                         ret += fprintf(fp, "failed to get cpus\n");
4308                 break;
4309         default:
4310                 ret += fprintf(fp, "... unknown type\n");
4311                 break;
4312         }
4313
4314         return ret;
4315 }
4316
4317 int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4318                              union perf_event *event,
4319                              struct evlist **pevlist)
4320 {
4321         u32 i, ids, n_ids;
4322         struct evsel *evsel;
4323         struct evlist *evlist = *pevlist;
4324
4325         if (evlist == NULL) {
4326                 *pevlist = evlist = evlist__new();
4327                 if (evlist == NULL)
4328                         return -ENOMEM;
4329         }
4330
4331         evsel = evsel__new(&event->attr.attr);
4332         if (evsel == NULL)
4333                 return -ENOMEM;
4334
4335         evlist__add(evlist, evsel);
4336
4337         ids = event->header.size;
4338         ids -= (void *)&event->attr.id - (void *)event;
4339         n_ids = ids / sizeof(u64);
4340         /*
4341          * We don't have the cpu and thread maps on the header, so
4342          * for allocating the perf_sample_id table we fake 1 cpu and
4343          * hattr->ids threads.
4344          */
4345         if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4346                 return -ENOMEM;
4347
4348         for (i = 0; i < n_ids; i++) {
4349                 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, event->attr.id[i]);
4350         }
4351
4352         return 0;
4353 }
4354
4355 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4356                                      union perf_event *event,
4357                                      struct evlist **pevlist)
4358 {
4359         struct perf_record_event_update *ev = &event->event_update;
4360         struct perf_record_event_update_scale *ev_scale;
4361         struct perf_record_event_update_cpus *ev_cpus;
4362         struct evlist *evlist;
4363         struct evsel *evsel;
4364         struct perf_cpu_map *map;
4365
4366         if (!pevlist || *pevlist == NULL)
4367                 return -EINVAL;
4368
4369         evlist = *pevlist;
4370
4371         evsel = evlist__id2evsel(evlist, ev->id);
4372         if (evsel == NULL)
4373                 return -EINVAL;
4374
4375         switch (ev->type) {
4376         case PERF_EVENT_UPDATE__UNIT:
4377                 free((char *)evsel->unit);
4378                 evsel->unit = strdup(ev->data);
4379                 break;
4380         case PERF_EVENT_UPDATE__NAME:
4381                 free(evsel->name);
4382                 evsel->name = strdup(ev->data);
4383                 break;
4384         case PERF_EVENT_UPDATE__SCALE:
4385                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4386                 evsel->scale = ev_scale->scale;
4387                 break;
4388         case PERF_EVENT_UPDATE__CPUS:
4389                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4390                 map = cpu_map__new_data(&ev_cpus->cpus);
4391                 if (map) {
4392                         perf_cpu_map__put(evsel->core.own_cpus);
4393                         evsel->core.own_cpus = map;
4394                 } else
4395                         pr_err("failed to get event_update cpus\n");
4396         default:
4397                 break;
4398         }
4399
4400         return 0;
4401 }
4402
4403 int perf_event__process_tracing_data(struct perf_session *session,
4404                                      union perf_event *event)
4405 {
4406         ssize_t size_read, padding, size = event->tracing_data.size;
4407         int fd = perf_data__fd(session->data);
4408         char buf[BUFSIZ];
4409
4410         /*
4411          * The pipe fd is already in proper place and in any case
4412          * we can't move it, and we'd screw the case where we read
4413          * 'pipe' data from regular file. The trace_report reads
4414          * data from 'fd' so we need to set it directly behind the
4415          * event, where the tracing data starts.
4416          */
4417         if (!perf_data__is_pipe(session->data)) {
4418                 off_t offset = lseek(fd, 0, SEEK_CUR);
4419
4420                 /* setup for reading amidst mmap */
4421                 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4422                       SEEK_SET);
4423         }
4424
4425         size_read = trace_report(fd, &session->tevent,
4426                                  session->repipe);
4427         padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4428
4429         if (readn(fd, buf, padding) < 0) {
4430                 pr_err("%s: reading input file", __func__);
4431                 return -1;
4432         }
4433         if (session->repipe) {
4434                 int retw = write(STDOUT_FILENO, buf, padding);
4435                 if (retw <= 0 || retw != padding) {
4436                         pr_err("%s: repiping tracing data padding", __func__);
4437                         return -1;
4438                 }
4439         }
4440
4441         if (size_read + padding != size) {
4442                 pr_err("%s: tracing data size mismatch", __func__);
4443                 return -1;
4444         }
4445
4446         evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4447
4448         return size_read + padding;
4449 }
4450
4451 int perf_event__process_build_id(struct perf_session *session,
4452                                  union perf_event *event)
4453 {
4454         __event_process_build_id(&event->build_id,
4455                                  event->build_id.filename,
4456                                  session);
4457         return 0;
4458 }