OSDN Git Service

Merge tag 'memblock-v5.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rppt...
[uclinux-h8/linux.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 #ifdef HAVE_LIBBPF_SUPPORT
987 static int write_bpf_prog_info(struct feat_fd *ff,
988                                struct evlist *evlist __maybe_unused)
989 {
990         struct perf_env *env = &ff->ph->env;
991         struct rb_root *root;
992         struct rb_node *next;
993         int ret;
994
995         down_read(&env->bpf_progs.lock);
996
997         ret = do_write(ff, &env->bpf_progs.infos_cnt,
998                        sizeof(env->bpf_progs.infos_cnt));
999         if (ret < 0)
1000                 goto out;
1001
1002         root = &env->bpf_progs.infos;
1003         next = rb_first(root);
1004         while (next) {
1005                 struct bpf_prog_info_node *node;
1006                 size_t len;
1007
1008                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1009                 next = rb_next(&node->rb_node);
1010                 len = sizeof(struct perf_bpil) +
1011                         node->info_linear->data_len;
1012
1013                 /* before writing to file, translate address to offset */
1014                 bpil_addr_to_offs(node->info_linear);
1015                 ret = do_write(ff, node->info_linear, len);
1016                 /*
1017                  * translate back to address even when do_write() fails,
1018                  * so that this function never changes the data.
1019                  */
1020                 bpil_offs_to_addr(node->info_linear);
1021                 if (ret < 0)
1022                         goto out;
1023         }
1024 out:
1025         up_read(&env->bpf_progs.lock);
1026         return ret;
1027 }
1028
1029 static int write_bpf_btf(struct feat_fd *ff,
1030                          struct evlist *evlist __maybe_unused)
1031 {
1032         struct perf_env *env = &ff->ph->env;
1033         struct rb_root *root;
1034         struct rb_node *next;
1035         int ret;
1036
1037         down_read(&env->bpf_progs.lock);
1038
1039         ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1040                        sizeof(env->bpf_progs.btfs_cnt));
1041
1042         if (ret < 0)
1043                 goto out;
1044
1045         root = &env->bpf_progs.btfs;
1046         next = rb_first(root);
1047         while (next) {
1048                 struct btf_node *node;
1049
1050                 node = rb_entry(next, struct btf_node, rb_node);
1051                 next = rb_next(&node->rb_node);
1052                 ret = do_write(ff, &node->id,
1053                                sizeof(u32) * 2 + node->data_size);
1054                 if (ret < 0)
1055                         goto out;
1056         }
1057 out:
1058         up_read(&env->bpf_progs.lock);
1059         return ret;
1060 }
1061 #endif // HAVE_LIBBPF_SUPPORT
1062
1063 static int cpu_cache_level__sort(const void *a, const void *b)
1064 {
1065         struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1066         struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1067
1068         return cache_a->level - cache_b->level;
1069 }
1070
1071 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1072 {
1073         if (a->level != b->level)
1074                 return false;
1075
1076         if (a->line_size != b->line_size)
1077                 return false;
1078
1079         if (a->sets != b->sets)
1080                 return false;
1081
1082         if (a->ways != b->ways)
1083                 return false;
1084
1085         if (strcmp(a->type, b->type))
1086                 return false;
1087
1088         if (strcmp(a->size, b->size))
1089                 return false;
1090
1091         if (strcmp(a->map, b->map))
1092                 return false;
1093
1094         return true;
1095 }
1096
1097 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1098 {
1099         char path[PATH_MAX], file[PATH_MAX];
1100         struct stat st;
1101         size_t len;
1102
1103         scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1104         scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1105
1106         if (stat(file, &st))
1107                 return 1;
1108
1109         scnprintf(file, PATH_MAX, "%s/level", path);
1110         if (sysfs__read_int(file, (int *) &cache->level))
1111                 return -1;
1112
1113         scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1114         if (sysfs__read_int(file, (int *) &cache->line_size))
1115                 return -1;
1116
1117         scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1118         if (sysfs__read_int(file, (int *) &cache->sets))
1119                 return -1;
1120
1121         scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1122         if (sysfs__read_int(file, (int *) &cache->ways))
1123                 return -1;
1124
1125         scnprintf(file, PATH_MAX, "%s/type", path);
1126         if (sysfs__read_str(file, &cache->type, &len))
1127                 return -1;
1128
1129         cache->type[len] = 0;
1130         cache->type = strim(cache->type);
1131
1132         scnprintf(file, PATH_MAX, "%s/size", path);
1133         if (sysfs__read_str(file, &cache->size, &len)) {
1134                 zfree(&cache->type);
1135                 return -1;
1136         }
1137
1138         cache->size[len] = 0;
1139         cache->size = strim(cache->size);
1140
1141         scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1142         if (sysfs__read_str(file, &cache->map, &len)) {
1143                 zfree(&cache->size);
1144                 zfree(&cache->type);
1145                 return -1;
1146         }
1147
1148         cache->map[len] = 0;
1149         cache->map = strim(cache->map);
1150         return 0;
1151 }
1152
1153 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1154 {
1155         fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1156 }
1157
1158 #define MAX_CACHE_LVL 4
1159
1160 static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1161 {
1162         u32 i, cnt = 0;
1163         u32 nr, cpu;
1164         u16 level;
1165
1166         nr = cpu__max_cpu().cpu;
1167
1168         for (cpu = 0; cpu < nr; cpu++) {
1169                 for (level = 0; level < MAX_CACHE_LVL; level++) {
1170                         struct cpu_cache_level c;
1171                         int err;
1172
1173                         err = cpu_cache_level__read(&c, cpu, level);
1174                         if (err < 0)
1175                                 return err;
1176
1177                         if (err == 1)
1178                                 break;
1179
1180                         for (i = 0; i < cnt; i++) {
1181                                 if (cpu_cache_level__cmp(&c, &caches[i]))
1182                                         break;
1183                         }
1184
1185                         if (i == cnt)
1186                                 caches[cnt++] = c;
1187                         else
1188                                 cpu_cache_level__free(&c);
1189                 }
1190         }
1191         *cntp = cnt;
1192         return 0;
1193 }
1194
1195 static int write_cache(struct feat_fd *ff,
1196                        struct evlist *evlist __maybe_unused)
1197 {
1198         u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL;
1199         struct cpu_cache_level caches[max_caches];
1200         u32 cnt = 0, i, version = 1;
1201         int ret;
1202
1203         ret = build_caches(caches, &cnt);
1204         if (ret)
1205                 goto out;
1206
1207         qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1208
1209         ret = do_write(ff, &version, sizeof(u32));
1210         if (ret < 0)
1211                 goto out;
1212
1213         ret = do_write(ff, &cnt, sizeof(u32));
1214         if (ret < 0)
1215                 goto out;
1216
1217         for (i = 0; i < cnt; i++) {
1218                 struct cpu_cache_level *c = &caches[i];
1219
1220                 #define _W(v)                                   \
1221                         ret = do_write(ff, &c->v, sizeof(u32)); \
1222                         if (ret < 0)                            \
1223                                 goto out;
1224
1225                 _W(level)
1226                 _W(line_size)
1227                 _W(sets)
1228                 _W(ways)
1229                 #undef _W
1230
1231                 #define _W(v)                                           \
1232                         ret = do_write_string(ff, (const char *) c->v); \
1233                         if (ret < 0)                                    \
1234                                 goto out;
1235
1236                 _W(type)
1237                 _W(size)
1238                 _W(map)
1239                 #undef _W
1240         }
1241
1242 out:
1243         for (i = 0; i < cnt; i++)
1244                 cpu_cache_level__free(&caches[i]);
1245         return ret;
1246 }
1247
1248 static int write_stat(struct feat_fd *ff __maybe_unused,
1249                       struct evlist *evlist __maybe_unused)
1250 {
1251         return 0;
1252 }
1253
1254 static int write_sample_time(struct feat_fd *ff,
1255                              struct evlist *evlist)
1256 {
1257         int ret;
1258
1259         ret = do_write(ff, &evlist->first_sample_time,
1260                        sizeof(evlist->first_sample_time));
1261         if (ret < 0)
1262                 return ret;
1263
1264         return do_write(ff, &evlist->last_sample_time,
1265                         sizeof(evlist->last_sample_time));
1266 }
1267
1268
1269 static int memory_node__read(struct memory_node *n, unsigned long idx)
1270 {
1271         unsigned int phys, size = 0;
1272         char path[PATH_MAX];
1273         struct dirent *ent;
1274         DIR *dir;
1275
1276 #define for_each_memory(mem, dir)                                       \
1277         while ((ent = readdir(dir)))                                    \
1278                 if (strcmp(ent->d_name, ".") &&                         \
1279                     strcmp(ent->d_name, "..") &&                        \
1280                     sscanf(ent->d_name, "memory%u", &mem) == 1)
1281
1282         scnprintf(path, PATH_MAX,
1283                   "%s/devices/system/node/node%lu",
1284                   sysfs__mountpoint(), idx);
1285
1286         dir = opendir(path);
1287         if (!dir) {
1288                 pr_warning("failed: can't open memory sysfs data\n");
1289                 return -1;
1290         }
1291
1292         for_each_memory(phys, dir) {
1293                 size = max(phys, size);
1294         }
1295
1296         size++;
1297
1298         n->set = bitmap_zalloc(size);
1299         if (!n->set) {
1300                 closedir(dir);
1301                 return -ENOMEM;
1302         }
1303
1304         n->node = idx;
1305         n->size = size;
1306
1307         rewinddir(dir);
1308
1309         for_each_memory(phys, dir) {
1310                 set_bit(phys, n->set);
1311         }
1312
1313         closedir(dir);
1314         return 0;
1315 }
1316
1317 static int memory_node__sort(const void *a, const void *b)
1318 {
1319         const struct memory_node *na = a;
1320         const struct memory_node *nb = b;
1321
1322         return na->node - nb->node;
1323 }
1324
1325 static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
1326 {
1327         char path[PATH_MAX];
1328         struct dirent *ent;
1329         DIR *dir;
1330         u64 cnt = 0;
1331         int ret = 0;
1332
1333         scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1334                   sysfs__mountpoint());
1335
1336         dir = opendir(path);
1337         if (!dir) {
1338                 pr_debug2("%s: could't read %s, does this arch have topology information?\n",
1339                           __func__, path);
1340                 return -1;
1341         }
1342
1343         while (!ret && (ent = readdir(dir))) {
1344                 unsigned int idx;
1345                 int r;
1346
1347                 if (!strcmp(ent->d_name, ".") ||
1348                     !strcmp(ent->d_name, ".."))
1349                         continue;
1350
1351                 r = sscanf(ent->d_name, "node%u", &idx);
1352                 if (r != 1)
1353                         continue;
1354
1355                 if (WARN_ONCE(cnt >= size,
1356                         "failed to write MEM_TOPOLOGY, way too many nodes\n")) {
1357                         closedir(dir);
1358                         return -1;
1359                 }
1360
1361                 ret = memory_node__read(&nodes[cnt++], idx);
1362         }
1363
1364         *cntp = cnt;
1365         closedir(dir);
1366
1367         if (!ret)
1368                 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1369
1370         return ret;
1371 }
1372
1373 #define MAX_MEMORY_NODES 2000
1374
1375 /*
1376  * The MEM_TOPOLOGY holds physical memory map for every
1377  * node in system. The format of data is as follows:
1378  *
1379  *  0 - version          | for future changes
1380  *  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1381  * 16 - count            | number of nodes
1382  *
1383  * For each node we store map of physical indexes for
1384  * each node:
1385  *
1386  * 32 - node id          | node index
1387  * 40 - size             | size of bitmap
1388  * 48 - bitmap           | bitmap of memory indexes that belongs to node
1389  */
1390 static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1391                               struct evlist *evlist __maybe_unused)
1392 {
1393         static struct memory_node nodes[MAX_MEMORY_NODES];
1394         u64 bsize, version = 1, i, nr;
1395         int ret;
1396
1397         ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1398                               (unsigned long long *) &bsize);
1399         if (ret)
1400                 return ret;
1401
1402         ret = build_mem_topology(&nodes[0], MAX_MEMORY_NODES, &nr);
1403         if (ret)
1404                 return ret;
1405
1406         ret = do_write(ff, &version, sizeof(version));
1407         if (ret < 0)
1408                 goto out;
1409
1410         ret = do_write(ff, &bsize, sizeof(bsize));
1411         if (ret < 0)
1412                 goto out;
1413
1414         ret = do_write(ff, &nr, sizeof(nr));
1415         if (ret < 0)
1416                 goto out;
1417
1418         for (i = 0; i < nr; i++) {
1419                 struct memory_node *n = &nodes[i];
1420
1421                 #define _W(v)                                           \
1422                         ret = do_write(ff, &n->v, sizeof(n->v));        \
1423                         if (ret < 0)                                    \
1424                                 goto out;
1425
1426                 _W(node)
1427                 _W(size)
1428
1429                 #undef _W
1430
1431                 ret = do_write_bitmap(ff, n->set, n->size);
1432                 if (ret < 0)
1433                         goto out;
1434         }
1435
1436 out:
1437         return ret;
1438 }
1439
1440 static int write_compressed(struct feat_fd *ff __maybe_unused,
1441                             struct evlist *evlist __maybe_unused)
1442 {
1443         int ret;
1444
1445         ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1446         if (ret)
1447                 return ret;
1448
1449         ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1450         if (ret)
1451                 return ret;
1452
1453         ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1454         if (ret)
1455                 return ret;
1456
1457         ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1458         if (ret)
1459                 return ret;
1460
1461         return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1462 }
1463
1464 static int write_per_cpu_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1465                                   bool write_pmu)
1466 {
1467         struct perf_pmu_caps *caps = NULL;
1468         int nr_caps;
1469         int ret;
1470
1471         nr_caps = perf_pmu__caps_parse(pmu);
1472         if (nr_caps < 0)
1473                 return nr_caps;
1474
1475         ret = do_write(ff, &nr_caps, sizeof(nr_caps));
1476         if (ret < 0)
1477                 return ret;
1478
1479         list_for_each_entry(caps, &pmu->caps, list) {
1480                 ret = do_write_string(ff, caps->name);
1481                 if (ret < 0)
1482                         return ret;
1483
1484                 ret = do_write_string(ff, caps->value);
1485                 if (ret < 0)
1486                         return ret;
1487         }
1488
1489         if (write_pmu) {
1490                 ret = do_write_string(ff, pmu->name);
1491                 if (ret < 0)
1492                         return ret;
1493         }
1494
1495         return ret;
1496 }
1497
1498 static int write_cpu_pmu_caps(struct feat_fd *ff,
1499                               struct evlist *evlist __maybe_unused)
1500 {
1501         struct perf_pmu *cpu_pmu = perf_pmu__find("cpu");
1502
1503         if (!cpu_pmu)
1504                 return -ENOENT;
1505
1506         return write_per_cpu_pmu_caps(ff, cpu_pmu, false);
1507 }
1508
1509 static int write_hybrid_cpu_pmu_caps(struct feat_fd *ff,
1510                                      struct evlist *evlist __maybe_unused)
1511 {
1512         struct perf_pmu *pmu;
1513         u32 nr_pmu = perf_pmu__hybrid_pmu_num();
1514         int ret;
1515
1516         if (nr_pmu == 0)
1517                 return -ENOENT;
1518
1519         ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1520         if (ret < 0)
1521                 return ret;
1522
1523         perf_pmu__for_each_hybrid_pmu(pmu) {
1524                 ret = write_per_cpu_pmu_caps(ff, pmu, true);
1525                 if (ret < 0)
1526                         return ret;
1527         }
1528
1529         return 0;
1530 }
1531
1532 static void print_hostname(struct feat_fd *ff, FILE *fp)
1533 {
1534         fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1535 }
1536
1537 static void print_osrelease(struct feat_fd *ff, FILE *fp)
1538 {
1539         fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1540 }
1541
1542 static void print_arch(struct feat_fd *ff, FILE *fp)
1543 {
1544         fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1545 }
1546
1547 static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1548 {
1549         fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1550 }
1551
1552 static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1553 {
1554         fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1555         fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1556 }
1557
1558 static void print_version(struct feat_fd *ff, FILE *fp)
1559 {
1560         fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1561 }
1562
1563 static void print_cmdline(struct feat_fd *ff, FILE *fp)
1564 {
1565         int nr, i;
1566
1567         nr = ff->ph->env.nr_cmdline;
1568
1569         fprintf(fp, "# cmdline : ");
1570
1571         for (i = 0; i < nr; i++) {
1572                 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1573                 if (!argv_i) {
1574                         fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1575                 } else {
1576                         char *mem = argv_i;
1577                         do {
1578                                 char *quote = strchr(argv_i, '\'');
1579                                 if (!quote)
1580                                         break;
1581                                 *quote++ = '\0';
1582                                 fprintf(fp, "%s\\\'", argv_i);
1583                                 argv_i = quote;
1584                         } while (1);
1585                         fprintf(fp, "%s ", argv_i);
1586                         free(mem);
1587                 }
1588         }
1589         fputc('\n', fp);
1590 }
1591
1592 static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1593 {
1594         struct perf_header *ph = ff->ph;
1595         int cpu_nr = ph->env.nr_cpus_avail;
1596         int nr, i;
1597         char *str;
1598
1599         nr = ph->env.nr_sibling_cores;
1600         str = ph->env.sibling_cores;
1601
1602         for (i = 0; i < nr; i++) {
1603                 fprintf(fp, "# sibling sockets : %s\n", str);
1604                 str += strlen(str) + 1;
1605         }
1606
1607         if (ph->env.nr_sibling_dies) {
1608                 nr = ph->env.nr_sibling_dies;
1609                 str = ph->env.sibling_dies;
1610
1611                 for (i = 0; i < nr; i++) {
1612                         fprintf(fp, "# sibling dies    : %s\n", str);
1613                         str += strlen(str) + 1;
1614                 }
1615         }
1616
1617         nr = ph->env.nr_sibling_threads;
1618         str = ph->env.sibling_threads;
1619
1620         for (i = 0; i < nr; i++) {
1621                 fprintf(fp, "# sibling threads : %s\n", str);
1622                 str += strlen(str) + 1;
1623         }
1624
1625         if (ph->env.nr_sibling_dies) {
1626                 if (ph->env.cpu != NULL) {
1627                         for (i = 0; i < cpu_nr; i++)
1628                                 fprintf(fp, "# CPU %d: Core ID %d, "
1629                                             "Die ID %d, Socket ID %d\n",
1630                                             i, ph->env.cpu[i].core_id,
1631                                             ph->env.cpu[i].die_id,
1632                                             ph->env.cpu[i].socket_id);
1633                 } else
1634                         fprintf(fp, "# Core ID, Die ID and Socket ID "
1635                                     "information is not available\n");
1636         } else {
1637                 if (ph->env.cpu != NULL) {
1638                         for (i = 0; i < cpu_nr; i++)
1639                                 fprintf(fp, "# CPU %d: Core ID %d, "
1640                                             "Socket ID %d\n",
1641                                             i, ph->env.cpu[i].core_id,
1642                                             ph->env.cpu[i].socket_id);
1643                 } else
1644                         fprintf(fp, "# Core ID and Socket ID "
1645                                     "information is not available\n");
1646         }
1647 }
1648
1649 static void print_clockid(struct feat_fd *ff, FILE *fp)
1650 {
1651         fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1652                 ff->ph->env.clock.clockid_res_ns * 1000);
1653 }
1654
1655 static void print_clock_data(struct feat_fd *ff, FILE *fp)
1656 {
1657         struct timespec clockid_ns;
1658         char tstr[64], date[64];
1659         struct timeval tod_ns;
1660         clockid_t clockid;
1661         struct tm ltime;
1662         u64 ref;
1663
1664         if (!ff->ph->env.clock.enabled) {
1665                 fprintf(fp, "# reference time disabled\n");
1666                 return;
1667         }
1668
1669         /* Compute TOD time. */
1670         ref = ff->ph->env.clock.tod_ns;
1671         tod_ns.tv_sec = ref / NSEC_PER_SEC;
1672         ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1673         tod_ns.tv_usec = ref / NSEC_PER_USEC;
1674
1675         /* Compute clockid time. */
1676         ref = ff->ph->env.clock.clockid_ns;
1677         clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1678         ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1679         clockid_ns.tv_nsec = ref;
1680
1681         clockid = ff->ph->env.clock.clockid;
1682
1683         if (localtime_r(&tod_ns.tv_sec, &ltime) == NULL)
1684                 snprintf(tstr, sizeof(tstr), "<error>");
1685         else {
1686                 strftime(date, sizeof(date), "%F %T", &ltime);
1687                 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1688                           date, (int) tod_ns.tv_usec);
1689         }
1690
1691         fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1692         fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1693                     tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1694                     (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1695                     clockid_name(clockid));
1696 }
1697
1698 static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1699 {
1700         int i;
1701         struct hybrid_node *n;
1702
1703         fprintf(fp, "# hybrid cpu system:\n");
1704         for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1705                 n = &ff->ph->env.hybrid_nodes[i];
1706                 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1707         }
1708 }
1709
1710 static void print_dir_format(struct feat_fd *ff, FILE *fp)
1711 {
1712         struct perf_session *session;
1713         struct perf_data *data;
1714
1715         session = container_of(ff->ph, struct perf_session, header);
1716         data = session->data;
1717
1718         fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1719 }
1720
1721 #ifdef HAVE_LIBBPF_SUPPORT
1722 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1723 {
1724         struct perf_env *env = &ff->ph->env;
1725         struct rb_root *root;
1726         struct rb_node *next;
1727
1728         down_read(&env->bpf_progs.lock);
1729
1730         root = &env->bpf_progs.infos;
1731         next = rb_first(root);
1732
1733         while (next) {
1734                 struct bpf_prog_info_node *node;
1735
1736                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1737                 next = rb_next(&node->rb_node);
1738
1739                 bpf_event__print_bpf_prog_info(&node->info_linear->info,
1740                                                env, fp);
1741         }
1742
1743         up_read(&env->bpf_progs.lock);
1744 }
1745
1746 static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1747 {
1748         struct perf_env *env = &ff->ph->env;
1749         struct rb_root *root;
1750         struct rb_node *next;
1751
1752         down_read(&env->bpf_progs.lock);
1753
1754         root = &env->bpf_progs.btfs;
1755         next = rb_first(root);
1756
1757         while (next) {
1758                 struct btf_node *node;
1759
1760                 node = rb_entry(next, struct btf_node, rb_node);
1761                 next = rb_next(&node->rb_node);
1762                 fprintf(fp, "# btf info of id %u\n", node->id);
1763         }
1764
1765         up_read(&env->bpf_progs.lock);
1766 }
1767 #endif // HAVE_LIBBPF_SUPPORT
1768
1769 static void free_event_desc(struct evsel *events)
1770 {
1771         struct evsel *evsel;
1772
1773         if (!events)
1774                 return;
1775
1776         for (evsel = events; evsel->core.attr.size; evsel++) {
1777                 zfree(&evsel->name);
1778                 zfree(&evsel->core.id);
1779         }
1780
1781         free(events);
1782 }
1783
1784 static bool perf_attr_check(struct perf_event_attr *attr)
1785 {
1786         if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1787                 pr_warning("Reserved bits are set unexpectedly. "
1788                            "Please update perf tool.\n");
1789                 return false;
1790         }
1791
1792         if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1793                 pr_warning("Unknown sample type (0x%llx) is detected. "
1794                            "Please update perf tool.\n",
1795                            attr->sample_type);
1796                 return false;
1797         }
1798
1799         if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1800                 pr_warning("Unknown read format (0x%llx) is detected. "
1801                            "Please update perf tool.\n",
1802                            attr->read_format);
1803                 return false;
1804         }
1805
1806         if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1807             (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1808                 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1809                            "Please update perf tool.\n",
1810                            attr->branch_sample_type);
1811
1812                 return false;
1813         }
1814
1815         return true;
1816 }
1817
1818 static struct evsel *read_event_desc(struct feat_fd *ff)
1819 {
1820         struct evsel *evsel, *events = NULL;
1821         u64 *id;
1822         void *buf = NULL;
1823         u32 nre, sz, nr, i, j;
1824         size_t msz;
1825
1826         /* number of events */
1827         if (do_read_u32(ff, &nre))
1828                 goto error;
1829
1830         if (do_read_u32(ff, &sz))
1831                 goto error;
1832
1833         /* buffer to hold on file attr struct */
1834         buf = malloc(sz);
1835         if (!buf)
1836                 goto error;
1837
1838         /* the last event terminates with evsel->core.attr.size == 0: */
1839         events = calloc(nre + 1, sizeof(*events));
1840         if (!events)
1841                 goto error;
1842
1843         msz = sizeof(evsel->core.attr);
1844         if (sz < msz)
1845                 msz = sz;
1846
1847         for (i = 0, evsel = events; i < nre; evsel++, i++) {
1848                 evsel->core.idx = i;
1849
1850                 /*
1851                  * must read entire on-file attr struct to
1852                  * sync up with layout.
1853                  */
1854                 if (__do_read(ff, buf, sz))
1855                         goto error;
1856
1857                 if (ff->ph->needs_swap)
1858                         perf_event__attr_swap(buf);
1859
1860                 memcpy(&evsel->core.attr, buf, msz);
1861
1862                 if (!perf_attr_check(&evsel->core.attr))
1863                         goto error;
1864
1865                 if (do_read_u32(ff, &nr))
1866                         goto error;
1867
1868                 if (ff->ph->needs_swap)
1869                         evsel->needs_swap = true;
1870
1871                 evsel->name = do_read_string(ff);
1872                 if (!evsel->name)
1873                         goto error;
1874
1875                 if (!nr)
1876                         continue;
1877
1878                 id = calloc(nr, sizeof(*id));
1879                 if (!id)
1880                         goto error;
1881                 evsel->core.ids = nr;
1882                 evsel->core.id = id;
1883
1884                 for (j = 0 ; j < nr; j++) {
1885                         if (do_read_u64(ff, id))
1886                                 goto error;
1887                         id++;
1888                 }
1889         }
1890 out:
1891         free(buf);
1892         return events;
1893 error:
1894         free_event_desc(events);
1895         events = NULL;
1896         goto out;
1897 }
1898
1899 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
1900                                 void *priv __maybe_unused)
1901 {
1902         return fprintf(fp, ", %s = %s", name, val);
1903 }
1904
1905 static void print_event_desc(struct feat_fd *ff, FILE *fp)
1906 {
1907         struct evsel *evsel, *events;
1908         u32 j;
1909         u64 *id;
1910
1911         if (ff->events)
1912                 events = ff->events;
1913         else
1914                 events = read_event_desc(ff);
1915
1916         if (!events) {
1917                 fprintf(fp, "# event desc: not available or unable to read\n");
1918                 return;
1919         }
1920
1921         for (evsel = events; evsel->core.attr.size; evsel++) {
1922                 fprintf(fp, "# event : name = %s, ", evsel->name);
1923
1924                 if (evsel->core.ids) {
1925                         fprintf(fp, ", id = {");
1926                         for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
1927                                 if (j)
1928                                         fputc(',', fp);
1929                                 fprintf(fp, " %"PRIu64, *id);
1930                         }
1931                         fprintf(fp, " }");
1932                 }
1933
1934                 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
1935
1936                 fputc('\n', fp);
1937         }
1938
1939         free_event_desc(events);
1940         ff->events = NULL;
1941 }
1942
1943 static void print_total_mem(struct feat_fd *ff, FILE *fp)
1944 {
1945         fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
1946 }
1947
1948 static void print_numa_topology(struct feat_fd *ff, FILE *fp)
1949 {
1950         int i;
1951         struct numa_node *n;
1952
1953         for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
1954                 n = &ff->ph->env.numa_nodes[i];
1955
1956                 fprintf(fp, "# node%u meminfo  : total = %"PRIu64" kB,"
1957                             " free = %"PRIu64" kB\n",
1958                         n->node, n->mem_total, n->mem_free);
1959
1960                 fprintf(fp, "# node%u cpu list : ", n->node);
1961                 cpu_map__fprintf(n->map, fp);
1962         }
1963 }
1964
1965 static void print_cpuid(struct feat_fd *ff, FILE *fp)
1966 {
1967         fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
1968 }
1969
1970 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
1971 {
1972         fprintf(fp, "# contains samples with branch stack\n");
1973 }
1974
1975 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
1976 {
1977         fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
1978 }
1979
1980 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
1981 {
1982         fprintf(fp, "# contains stat data\n");
1983 }
1984
1985 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
1986 {
1987         int i;
1988
1989         fprintf(fp, "# CPU cache info:\n");
1990         for (i = 0; i < ff->ph->env.caches_cnt; i++) {
1991                 fprintf(fp, "#  ");
1992                 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
1993         }
1994 }
1995
1996 static void print_compressed(struct feat_fd *ff, FILE *fp)
1997 {
1998         fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
1999                 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
2000                 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2001 }
2002
2003 static void print_per_cpu_pmu_caps(FILE *fp, int nr_caps, char *cpu_pmu_caps,
2004                                    char *pmu_name)
2005 {
2006         const char *delimiter;
2007         char *str, buf[128];
2008
2009         if (!nr_caps) {
2010                 if (!pmu_name)
2011                         fprintf(fp, "# cpu pmu capabilities: not available\n");
2012                 else
2013                         fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2014                 return;
2015         }
2016
2017         if (!pmu_name)
2018                 scnprintf(buf, sizeof(buf), "# cpu pmu capabilities: ");
2019         else
2020                 scnprintf(buf, sizeof(buf), "# %s pmu capabilities: ", pmu_name);
2021
2022         delimiter = buf;
2023
2024         str = cpu_pmu_caps;
2025         while (nr_caps--) {
2026                 fprintf(fp, "%s%s", delimiter, str);
2027                 delimiter = ", ";
2028                 str += strlen(str) + 1;
2029         }
2030
2031         fprintf(fp, "\n");
2032 }
2033
2034 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2035 {
2036         print_per_cpu_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2037                                ff->ph->env.cpu_pmu_caps, NULL);
2038 }
2039
2040 static void print_hybrid_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2041 {
2042         struct hybrid_cpc_node *n;
2043
2044         for (int i = 0; i < ff->ph->env.nr_hybrid_cpc_nodes; i++) {
2045                 n = &ff->ph->env.hybrid_cpc_nodes[i];
2046                 print_per_cpu_pmu_caps(fp, n->nr_cpu_pmu_caps,
2047                                        n->cpu_pmu_caps,
2048                                        n->pmu_name);
2049         }
2050 }
2051
2052 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2053 {
2054         const char *delimiter = "# pmu mappings: ";
2055         char *str, *tmp;
2056         u32 pmu_num;
2057         u32 type;
2058
2059         pmu_num = ff->ph->env.nr_pmu_mappings;
2060         if (!pmu_num) {
2061                 fprintf(fp, "# pmu mappings: not available\n");
2062                 return;
2063         }
2064
2065         str = ff->ph->env.pmu_mappings;
2066
2067         while (pmu_num) {
2068                 type = strtoul(str, &tmp, 0);
2069                 if (*tmp != ':')
2070                         goto error;
2071
2072                 str = tmp + 1;
2073                 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2074
2075                 delimiter = ", ";
2076                 str += strlen(str) + 1;
2077                 pmu_num--;
2078         }
2079
2080         fprintf(fp, "\n");
2081
2082         if (!pmu_num)
2083                 return;
2084 error:
2085         fprintf(fp, "# pmu mappings: unable to read\n");
2086 }
2087
2088 static void print_group_desc(struct feat_fd *ff, FILE *fp)
2089 {
2090         struct perf_session *session;
2091         struct evsel *evsel;
2092         u32 nr = 0;
2093
2094         session = container_of(ff->ph, struct perf_session, header);
2095
2096         evlist__for_each_entry(session->evlist, evsel) {
2097                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2098                         fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2099
2100                         nr = evsel->core.nr_members - 1;
2101                 } else if (nr) {
2102                         fprintf(fp, ",%s", evsel__name(evsel));
2103
2104                         if (--nr == 0)
2105                                 fprintf(fp, "}\n");
2106                 }
2107         }
2108 }
2109
2110 static void print_sample_time(struct feat_fd *ff, FILE *fp)
2111 {
2112         struct perf_session *session;
2113         char time_buf[32];
2114         double d;
2115
2116         session = container_of(ff->ph, struct perf_session, header);
2117
2118         timestamp__scnprintf_usec(session->evlist->first_sample_time,
2119                                   time_buf, sizeof(time_buf));
2120         fprintf(fp, "# time of first sample : %s\n", time_buf);
2121
2122         timestamp__scnprintf_usec(session->evlist->last_sample_time,
2123                                   time_buf, sizeof(time_buf));
2124         fprintf(fp, "# time of last sample : %s\n", time_buf);
2125
2126         d = (double)(session->evlist->last_sample_time -
2127                 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2128
2129         fprintf(fp, "# sample duration : %10.3f ms\n", d);
2130 }
2131
2132 static void memory_node__fprintf(struct memory_node *n,
2133                                  unsigned long long bsize, FILE *fp)
2134 {
2135         char buf_map[100], buf_size[50];
2136         unsigned long long size;
2137
2138         size = bsize * bitmap_weight(n->set, n->size);
2139         unit_number__scnprintf(buf_size, 50, size);
2140
2141         bitmap_scnprintf(n->set, n->size, buf_map, 100);
2142         fprintf(fp, "#  %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2143 }
2144
2145 static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2146 {
2147         struct memory_node *nodes;
2148         int i, nr;
2149
2150         nodes = ff->ph->env.memory_nodes;
2151         nr    = ff->ph->env.nr_memory_nodes;
2152
2153         fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2154                 nr, ff->ph->env.memory_bsize);
2155
2156         for (i = 0; i < nr; i++) {
2157                 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2158         }
2159 }
2160
2161 static int __event_process_build_id(struct perf_record_header_build_id *bev,
2162                                     char *filename,
2163                                     struct perf_session *session)
2164 {
2165         int err = -1;
2166         struct machine *machine;
2167         u16 cpumode;
2168         struct dso *dso;
2169         enum dso_space_type dso_space;
2170
2171         machine = perf_session__findnew_machine(session, bev->pid);
2172         if (!machine)
2173                 goto out;
2174
2175         cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2176
2177         switch (cpumode) {
2178         case PERF_RECORD_MISC_KERNEL:
2179                 dso_space = DSO_SPACE__KERNEL;
2180                 break;
2181         case PERF_RECORD_MISC_GUEST_KERNEL:
2182                 dso_space = DSO_SPACE__KERNEL_GUEST;
2183                 break;
2184         case PERF_RECORD_MISC_USER:
2185         case PERF_RECORD_MISC_GUEST_USER:
2186                 dso_space = DSO_SPACE__USER;
2187                 break;
2188         default:
2189                 goto out;
2190         }
2191
2192         dso = machine__findnew_dso(machine, filename);
2193         if (dso != NULL) {
2194                 char sbuild_id[SBUILD_ID_SIZE];
2195                 struct build_id bid;
2196                 size_t size = BUILD_ID_SIZE;
2197
2198                 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2199                         size = bev->size;
2200
2201                 build_id__init(&bid, bev->data, size);
2202                 dso__set_build_id(dso, &bid);
2203
2204                 if (dso_space != DSO_SPACE__USER) {
2205                         struct kmod_path m = { .name = NULL, };
2206
2207                         if (!kmod_path__parse_name(&m, filename) && m.kmod)
2208                                 dso__set_module_info(dso, &m, machine);
2209
2210                         dso->kernel = dso_space;
2211                         free(m.name);
2212                 }
2213
2214                 build_id__sprintf(&dso->bid, sbuild_id);
2215                 pr_debug("build id event received for %s: %s [%zu]\n",
2216                          dso->long_name, sbuild_id, size);
2217                 dso__put(dso);
2218         }
2219
2220         err = 0;
2221 out:
2222         return err;
2223 }
2224
2225 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2226                                                  int input, u64 offset, u64 size)
2227 {
2228         struct perf_session *session = container_of(header, struct perf_session, header);
2229         struct {
2230                 struct perf_event_header   header;
2231                 u8                         build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2232                 char                       filename[0];
2233         } old_bev;
2234         struct perf_record_header_build_id bev;
2235         char filename[PATH_MAX];
2236         u64 limit = offset + size;
2237
2238         while (offset < limit) {
2239                 ssize_t len;
2240
2241                 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2242                         return -1;
2243
2244                 if (header->needs_swap)
2245                         perf_event_header__bswap(&old_bev.header);
2246
2247                 len = old_bev.header.size - sizeof(old_bev);
2248                 if (readn(input, filename, len) != len)
2249                         return -1;
2250
2251                 bev.header = old_bev.header;
2252
2253                 /*
2254                  * As the pid is the missing value, we need to fill
2255                  * it properly. The header.misc value give us nice hint.
2256                  */
2257                 bev.pid = HOST_KERNEL_ID;
2258                 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2259                     bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2260                         bev.pid = DEFAULT_GUEST_KERNEL_ID;
2261
2262                 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2263                 __event_process_build_id(&bev, filename, session);
2264
2265                 offset += bev.header.size;
2266         }
2267
2268         return 0;
2269 }
2270
2271 static int perf_header__read_build_ids(struct perf_header *header,
2272                                        int input, u64 offset, u64 size)
2273 {
2274         struct perf_session *session = container_of(header, struct perf_session, header);
2275         struct perf_record_header_build_id bev;
2276         char filename[PATH_MAX];
2277         u64 limit = offset + size, orig_offset = offset;
2278         int err = -1;
2279
2280         while (offset < limit) {
2281                 ssize_t len;
2282
2283                 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2284                         goto out;
2285
2286                 if (header->needs_swap)
2287                         perf_event_header__bswap(&bev.header);
2288
2289                 len = bev.header.size - sizeof(bev);
2290                 if (readn(input, filename, len) != len)
2291                         goto out;
2292                 /*
2293                  * The a1645ce1 changeset:
2294                  *
2295                  * "perf: 'perf kvm' tool for monitoring guest performance from host"
2296                  *
2297                  * Added a field to struct perf_record_header_build_id that broke the file
2298                  * format.
2299                  *
2300                  * Since the kernel build-id is the first entry, process the
2301                  * table using the old format if the well known
2302                  * '[kernel.kallsyms]' string for the kernel build-id has the
2303                  * first 4 characters chopped off (where the pid_t sits).
2304                  */
2305                 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2306                         if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2307                                 return -1;
2308                         return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2309                 }
2310
2311                 __event_process_build_id(&bev, filename, session);
2312
2313                 offset += bev.header.size;
2314         }
2315         err = 0;
2316 out:
2317         return err;
2318 }
2319
2320 /* Macro for features that simply need to read and store a string. */
2321 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2322 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2323 {\
2324         free(ff->ph->env.__feat_env);                \
2325         ff->ph->env.__feat_env = do_read_string(ff); \
2326         return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2327 }
2328
2329 FEAT_PROCESS_STR_FUN(hostname, hostname);
2330 FEAT_PROCESS_STR_FUN(osrelease, os_release);
2331 FEAT_PROCESS_STR_FUN(version, version);
2332 FEAT_PROCESS_STR_FUN(arch, arch);
2333 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2334 FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2335
2336 static int process_tracing_data(struct feat_fd *ff, void *data)
2337 {
2338         ssize_t ret = trace_report(ff->fd, data, false);
2339
2340         return ret < 0 ? -1 : 0;
2341 }
2342
2343 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2344 {
2345         if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2346                 pr_debug("Failed to read buildids, continuing...\n");
2347         return 0;
2348 }
2349
2350 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2351 {
2352         int ret;
2353         u32 nr_cpus_avail, nr_cpus_online;
2354
2355         ret = do_read_u32(ff, &nr_cpus_avail);
2356         if (ret)
2357                 return ret;
2358
2359         ret = do_read_u32(ff, &nr_cpus_online);
2360         if (ret)
2361                 return ret;
2362         ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2363         ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2364         return 0;
2365 }
2366
2367 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2368 {
2369         u64 total_mem;
2370         int ret;
2371
2372         ret = do_read_u64(ff, &total_mem);
2373         if (ret)
2374                 return -1;
2375         ff->ph->env.total_mem = (unsigned long long)total_mem;
2376         return 0;
2377 }
2378
2379 static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2380 {
2381         struct evsel *evsel;
2382
2383         evlist__for_each_entry(evlist, evsel) {
2384                 if (evsel->core.idx == idx)
2385                         return evsel;
2386         }
2387
2388         return NULL;
2389 }
2390
2391 static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2392 {
2393         struct evsel *evsel;
2394
2395         if (!event->name)
2396                 return;
2397
2398         evsel = evlist__find_by_index(evlist, event->core.idx);
2399         if (!evsel)
2400                 return;
2401
2402         if (evsel->name)
2403                 return;
2404
2405         evsel->name = strdup(event->name);
2406 }
2407
2408 static int
2409 process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2410 {
2411         struct perf_session *session;
2412         struct evsel *evsel, *events = read_event_desc(ff);
2413
2414         if (!events)
2415                 return 0;
2416
2417         session = container_of(ff->ph, struct perf_session, header);
2418
2419         if (session->data->is_pipe) {
2420                 /* Save events for reading later by print_event_desc,
2421                  * since they can't be read again in pipe mode. */
2422                 ff->events = events;
2423         }
2424
2425         for (evsel = events; evsel->core.attr.size; evsel++)
2426                 evlist__set_event_name(session->evlist, evsel);
2427
2428         if (!session->data->is_pipe)
2429                 free_event_desc(events);
2430
2431         return 0;
2432 }
2433
2434 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2435 {
2436         char *str, *cmdline = NULL, **argv = NULL;
2437         u32 nr, i, len = 0;
2438
2439         if (do_read_u32(ff, &nr))
2440                 return -1;
2441
2442         ff->ph->env.nr_cmdline = nr;
2443
2444         cmdline = zalloc(ff->size + nr + 1);
2445         if (!cmdline)
2446                 return -1;
2447
2448         argv = zalloc(sizeof(char *) * (nr + 1));
2449         if (!argv)
2450                 goto error;
2451
2452         for (i = 0; i < nr; i++) {
2453                 str = do_read_string(ff);
2454                 if (!str)
2455                         goto error;
2456
2457                 argv[i] = cmdline + len;
2458                 memcpy(argv[i], str, strlen(str) + 1);
2459                 len += strlen(str) + 1;
2460                 free(str);
2461         }
2462         ff->ph->env.cmdline = cmdline;
2463         ff->ph->env.cmdline_argv = (const char **) argv;
2464         return 0;
2465
2466 error:
2467         free(argv);
2468         free(cmdline);
2469         return -1;
2470 }
2471
2472 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2473 {
2474         u32 nr, i;
2475         char *str;
2476         struct strbuf sb;
2477         int cpu_nr = ff->ph->env.nr_cpus_avail;
2478         u64 size = 0;
2479         struct perf_header *ph = ff->ph;
2480         bool do_core_id_test = true;
2481
2482         ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2483         if (!ph->env.cpu)
2484                 return -1;
2485
2486         if (do_read_u32(ff, &nr))
2487                 goto free_cpu;
2488
2489         ph->env.nr_sibling_cores = nr;
2490         size += sizeof(u32);
2491         if (strbuf_init(&sb, 128) < 0)
2492                 goto free_cpu;
2493
2494         for (i = 0; i < nr; i++) {
2495                 str = do_read_string(ff);
2496                 if (!str)
2497                         goto error;
2498
2499                 /* include a NULL character at the end */
2500                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2501                         goto error;
2502                 size += string_size(str);
2503                 free(str);
2504         }
2505         ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2506
2507         if (do_read_u32(ff, &nr))
2508                 return -1;
2509
2510         ph->env.nr_sibling_threads = nr;
2511         size += sizeof(u32);
2512
2513         for (i = 0; i < nr; i++) {
2514                 str = do_read_string(ff);
2515                 if (!str)
2516                         goto error;
2517
2518                 /* include a NULL character at the end */
2519                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2520                         goto error;
2521                 size += string_size(str);
2522                 free(str);
2523         }
2524         ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2525
2526         /*
2527          * The header may be from old perf,
2528          * which doesn't include core id and socket id information.
2529          */
2530         if (ff->size <= size) {
2531                 zfree(&ph->env.cpu);
2532                 return 0;
2533         }
2534
2535         /* On s390 the socket_id number is not related to the numbers of cpus.
2536          * The socket_id number might be higher than the numbers of cpus.
2537          * This depends on the configuration.
2538          * AArch64 is the same.
2539          */
2540         if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2541                           || !strncmp(ph->env.arch, "aarch64", 7)))
2542                 do_core_id_test = false;
2543
2544         for (i = 0; i < (u32)cpu_nr; i++) {
2545                 if (do_read_u32(ff, &nr))
2546                         goto free_cpu;
2547
2548                 ph->env.cpu[i].core_id = nr;
2549                 size += sizeof(u32);
2550
2551                 if (do_read_u32(ff, &nr))
2552                         goto free_cpu;
2553
2554                 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2555                         pr_debug("socket_id number is too big."
2556                                  "You may need to upgrade the perf tool.\n");
2557                         goto free_cpu;
2558                 }
2559
2560                 ph->env.cpu[i].socket_id = nr;
2561                 size += sizeof(u32);
2562         }
2563
2564         /*
2565          * The header may be from old perf,
2566          * which doesn't include die information.
2567          */
2568         if (ff->size <= size)
2569                 return 0;
2570
2571         if (do_read_u32(ff, &nr))
2572                 return -1;
2573
2574         ph->env.nr_sibling_dies = nr;
2575         size += sizeof(u32);
2576
2577         for (i = 0; i < nr; i++) {
2578                 str = do_read_string(ff);
2579                 if (!str)
2580                         goto error;
2581
2582                 /* include a NULL character at the end */
2583                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2584                         goto error;
2585                 size += string_size(str);
2586                 free(str);
2587         }
2588         ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2589
2590         for (i = 0; i < (u32)cpu_nr; i++) {
2591                 if (do_read_u32(ff, &nr))
2592                         goto free_cpu;
2593
2594                 ph->env.cpu[i].die_id = nr;
2595         }
2596
2597         return 0;
2598
2599 error:
2600         strbuf_release(&sb);
2601 free_cpu:
2602         zfree(&ph->env.cpu);
2603         return -1;
2604 }
2605
2606 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2607 {
2608         struct numa_node *nodes, *n;
2609         u32 nr, i;
2610         char *str;
2611
2612         /* nr nodes */
2613         if (do_read_u32(ff, &nr))
2614                 return -1;
2615
2616         nodes = zalloc(sizeof(*nodes) * nr);
2617         if (!nodes)
2618                 return -ENOMEM;
2619
2620         for (i = 0; i < nr; i++) {
2621                 n = &nodes[i];
2622
2623                 /* node number */
2624                 if (do_read_u32(ff, &n->node))
2625                         goto error;
2626
2627                 if (do_read_u64(ff, &n->mem_total))
2628                         goto error;
2629
2630                 if (do_read_u64(ff, &n->mem_free))
2631                         goto error;
2632
2633                 str = do_read_string(ff);
2634                 if (!str)
2635                         goto error;
2636
2637                 n->map = perf_cpu_map__new(str);
2638                 if (!n->map)
2639                         goto error;
2640
2641                 free(str);
2642         }
2643         ff->ph->env.nr_numa_nodes = nr;
2644         ff->ph->env.numa_nodes = nodes;
2645         return 0;
2646
2647 error:
2648         free(nodes);
2649         return -1;
2650 }
2651
2652 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2653 {
2654         char *name;
2655         u32 pmu_num;
2656         u32 type;
2657         struct strbuf sb;
2658
2659         if (do_read_u32(ff, &pmu_num))
2660                 return -1;
2661
2662         if (!pmu_num) {
2663                 pr_debug("pmu mappings not available\n");
2664                 return 0;
2665         }
2666
2667         ff->ph->env.nr_pmu_mappings = pmu_num;
2668         if (strbuf_init(&sb, 128) < 0)
2669                 return -1;
2670
2671         while (pmu_num) {
2672                 if (do_read_u32(ff, &type))
2673                         goto error;
2674
2675                 name = do_read_string(ff);
2676                 if (!name)
2677                         goto error;
2678
2679                 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2680                         goto error;
2681                 /* include a NULL character at the end */
2682                 if (strbuf_add(&sb, "", 1) < 0)
2683                         goto error;
2684
2685                 if (!strcmp(name, "msr"))
2686                         ff->ph->env.msr_pmu_type = type;
2687
2688                 free(name);
2689                 pmu_num--;
2690         }
2691         ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2692         return 0;
2693
2694 error:
2695         strbuf_release(&sb);
2696         return -1;
2697 }
2698
2699 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2700 {
2701         size_t ret = -1;
2702         u32 i, nr, nr_groups;
2703         struct perf_session *session;
2704         struct evsel *evsel, *leader = NULL;
2705         struct group_desc {
2706                 char *name;
2707                 u32 leader_idx;
2708                 u32 nr_members;
2709         } *desc;
2710
2711         if (do_read_u32(ff, &nr_groups))
2712                 return -1;
2713
2714         ff->ph->env.nr_groups = nr_groups;
2715         if (!nr_groups) {
2716                 pr_debug("group desc not available\n");
2717                 return 0;
2718         }
2719
2720         desc = calloc(nr_groups, sizeof(*desc));
2721         if (!desc)
2722                 return -1;
2723
2724         for (i = 0; i < nr_groups; i++) {
2725                 desc[i].name = do_read_string(ff);
2726                 if (!desc[i].name)
2727                         goto out_free;
2728
2729                 if (do_read_u32(ff, &desc[i].leader_idx))
2730                         goto out_free;
2731
2732                 if (do_read_u32(ff, &desc[i].nr_members))
2733                         goto out_free;
2734         }
2735
2736         /*
2737          * Rebuild group relationship based on the group_desc
2738          */
2739         session = container_of(ff->ph, struct perf_session, header);
2740         session->evlist->core.nr_groups = nr_groups;
2741
2742         i = nr = 0;
2743         evlist__for_each_entry(session->evlist, evsel) {
2744                 if (evsel->core.idx == (int) desc[i].leader_idx) {
2745                         evsel__set_leader(evsel, evsel);
2746                         /* {anon_group} is a dummy name */
2747                         if (strcmp(desc[i].name, "{anon_group}")) {
2748                                 evsel->group_name = desc[i].name;
2749                                 desc[i].name = NULL;
2750                         }
2751                         evsel->core.nr_members = desc[i].nr_members;
2752
2753                         if (i >= nr_groups || nr > 0) {
2754                                 pr_debug("invalid group desc\n");
2755                                 goto out_free;
2756                         }
2757
2758                         leader = evsel;
2759                         nr = evsel->core.nr_members - 1;
2760                         i++;
2761                 } else if (nr) {
2762                         /* This is a group member */
2763                         evsel__set_leader(evsel, leader);
2764
2765                         nr--;
2766                 }
2767         }
2768
2769         if (i != nr_groups || nr != 0) {
2770                 pr_debug("invalid group desc\n");
2771                 goto out_free;
2772         }
2773
2774         ret = 0;
2775 out_free:
2776         for (i = 0; i < nr_groups; i++)
2777                 zfree(&desc[i].name);
2778         free(desc);
2779
2780         return ret;
2781 }
2782
2783 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2784 {
2785         struct perf_session *session;
2786         int err;
2787
2788         session = container_of(ff->ph, struct perf_session, header);
2789
2790         err = auxtrace_index__process(ff->fd, ff->size, session,
2791                                       ff->ph->needs_swap);
2792         if (err < 0)
2793                 pr_err("Failed to process auxtrace index\n");
2794         return err;
2795 }
2796
2797 static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2798 {
2799         struct cpu_cache_level *caches;
2800         u32 cnt, i, version;
2801
2802         if (do_read_u32(ff, &version))
2803                 return -1;
2804
2805         if (version != 1)
2806                 return -1;
2807
2808         if (do_read_u32(ff, &cnt))
2809                 return -1;
2810
2811         caches = zalloc(sizeof(*caches) * cnt);
2812         if (!caches)
2813                 return -1;
2814
2815         for (i = 0; i < cnt; i++) {
2816                 struct cpu_cache_level c;
2817
2818                 #define _R(v)                                           \
2819                         if (do_read_u32(ff, &c.v))\
2820                                 goto out_free_caches;                   \
2821
2822                 _R(level)
2823                 _R(line_size)
2824                 _R(sets)
2825                 _R(ways)
2826                 #undef _R
2827
2828                 #define _R(v)                                   \
2829                         c.v = do_read_string(ff);               \
2830                         if (!c.v)                               \
2831                                 goto out_free_caches;
2832
2833                 _R(type)
2834                 _R(size)
2835                 _R(map)
2836                 #undef _R
2837
2838                 caches[i] = c;
2839         }
2840
2841         ff->ph->env.caches = caches;
2842         ff->ph->env.caches_cnt = cnt;
2843         return 0;
2844 out_free_caches:
2845         free(caches);
2846         return -1;
2847 }
2848
2849 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2850 {
2851         struct perf_session *session;
2852         u64 first_sample_time, last_sample_time;
2853         int ret;
2854
2855         session = container_of(ff->ph, struct perf_session, header);
2856
2857         ret = do_read_u64(ff, &first_sample_time);
2858         if (ret)
2859                 return -1;
2860
2861         ret = do_read_u64(ff, &last_sample_time);
2862         if (ret)
2863                 return -1;
2864
2865         session->evlist->first_sample_time = first_sample_time;
2866         session->evlist->last_sample_time = last_sample_time;
2867         return 0;
2868 }
2869
2870 static int process_mem_topology(struct feat_fd *ff,
2871                                 void *data __maybe_unused)
2872 {
2873         struct memory_node *nodes;
2874         u64 version, i, nr, bsize;
2875         int ret = -1;
2876
2877         if (do_read_u64(ff, &version))
2878                 return -1;
2879
2880         if (version != 1)
2881                 return -1;
2882
2883         if (do_read_u64(ff, &bsize))
2884                 return -1;
2885
2886         if (do_read_u64(ff, &nr))
2887                 return -1;
2888
2889         nodes = zalloc(sizeof(*nodes) * nr);
2890         if (!nodes)
2891                 return -1;
2892
2893         for (i = 0; i < nr; i++) {
2894                 struct memory_node n;
2895
2896                 #define _R(v)                           \
2897                         if (do_read_u64(ff, &n.v))      \
2898                                 goto out;               \
2899
2900                 _R(node)
2901                 _R(size)
2902
2903                 #undef _R
2904
2905                 if (do_read_bitmap(ff, &n.set, &n.size))
2906                         goto out;
2907
2908                 nodes[i] = n;
2909         }
2910
2911         ff->ph->env.memory_bsize    = bsize;
2912         ff->ph->env.memory_nodes    = nodes;
2913         ff->ph->env.nr_memory_nodes = nr;
2914         ret = 0;
2915
2916 out:
2917         if (ret)
2918                 free(nodes);
2919         return ret;
2920 }
2921
2922 static int process_clockid(struct feat_fd *ff,
2923                            void *data __maybe_unused)
2924 {
2925         if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
2926                 return -1;
2927
2928         return 0;
2929 }
2930
2931 static int process_clock_data(struct feat_fd *ff,
2932                               void *_data __maybe_unused)
2933 {
2934         u32 data32;
2935         u64 data64;
2936
2937         /* version */
2938         if (do_read_u32(ff, &data32))
2939                 return -1;
2940
2941         if (data32 != 1)
2942                 return -1;
2943
2944         /* clockid */
2945         if (do_read_u32(ff, &data32))
2946                 return -1;
2947
2948         ff->ph->env.clock.clockid = data32;
2949
2950         /* TOD ref time */
2951         if (do_read_u64(ff, &data64))
2952                 return -1;
2953
2954         ff->ph->env.clock.tod_ns = data64;
2955
2956         /* clockid ref time */
2957         if (do_read_u64(ff, &data64))
2958                 return -1;
2959
2960         ff->ph->env.clock.clockid_ns = data64;
2961         ff->ph->env.clock.enabled = true;
2962         return 0;
2963 }
2964
2965 static int process_hybrid_topology(struct feat_fd *ff,
2966                                    void *data __maybe_unused)
2967 {
2968         struct hybrid_node *nodes, *n;
2969         u32 nr, i;
2970
2971         /* nr nodes */
2972         if (do_read_u32(ff, &nr))
2973                 return -1;
2974
2975         nodes = zalloc(sizeof(*nodes) * nr);
2976         if (!nodes)
2977                 return -ENOMEM;
2978
2979         for (i = 0; i < nr; i++) {
2980                 n = &nodes[i];
2981
2982                 n->pmu_name = do_read_string(ff);
2983                 if (!n->pmu_name)
2984                         goto error;
2985
2986                 n->cpus = do_read_string(ff);
2987                 if (!n->cpus)
2988                         goto error;
2989         }
2990
2991         ff->ph->env.nr_hybrid_nodes = nr;
2992         ff->ph->env.hybrid_nodes = nodes;
2993         return 0;
2994
2995 error:
2996         for (i = 0; i < nr; i++) {
2997                 free(nodes[i].pmu_name);
2998                 free(nodes[i].cpus);
2999         }
3000
3001         free(nodes);
3002         return -1;
3003 }
3004
3005 static int process_dir_format(struct feat_fd *ff,
3006                               void *_data __maybe_unused)
3007 {
3008         struct perf_session *session;
3009         struct perf_data *data;
3010
3011         session = container_of(ff->ph, struct perf_session, header);
3012         data = session->data;
3013
3014         if (WARN_ON(!perf_data__is_dir(data)))
3015                 return -1;
3016
3017         return do_read_u64(ff, &data->dir.version);
3018 }
3019
3020 #ifdef HAVE_LIBBPF_SUPPORT
3021 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3022 {
3023         struct bpf_prog_info_node *info_node;
3024         struct perf_env *env = &ff->ph->env;
3025         struct perf_bpil *info_linear;
3026         u32 count, i;
3027         int err = -1;
3028
3029         if (ff->ph->needs_swap) {
3030                 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3031                 return 0;
3032         }
3033
3034         if (do_read_u32(ff, &count))
3035                 return -1;
3036
3037         down_write(&env->bpf_progs.lock);
3038
3039         for (i = 0; i < count; ++i) {
3040                 u32 info_len, data_len;
3041
3042                 info_linear = NULL;
3043                 info_node = NULL;
3044                 if (do_read_u32(ff, &info_len))
3045                         goto out;
3046                 if (do_read_u32(ff, &data_len))
3047                         goto out;
3048
3049                 if (info_len > sizeof(struct bpf_prog_info)) {
3050                         pr_warning("detected invalid bpf_prog_info\n");
3051                         goto out;
3052                 }
3053
3054                 info_linear = malloc(sizeof(struct perf_bpil) +
3055                                      data_len);
3056                 if (!info_linear)
3057                         goto out;
3058                 info_linear->info_len = sizeof(struct bpf_prog_info);
3059                 info_linear->data_len = data_len;
3060                 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3061                         goto out;
3062                 if (__do_read(ff, &info_linear->info, info_len))
3063                         goto out;
3064                 if (info_len < sizeof(struct bpf_prog_info))
3065                         memset(((void *)(&info_linear->info)) + info_len, 0,
3066                                sizeof(struct bpf_prog_info) - info_len);
3067
3068                 if (__do_read(ff, info_linear->data, data_len))
3069                         goto out;
3070
3071                 info_node = malloc(sizeof(struct bpf_prog_info_node));
3072                 if (!info_node)
3073                         goto out;
3074
3075                 /* after reading from file, translate offset to address */
3076                 bpil_offs_to_addr(info_linear);
3077                 info_node->info_linear = info_linear;
3078                 perf_env__insert_bpf_prog_info(env, info_node);
3079         }
3080
3081         up_write(&env->bpf_progs.lock);
3082         return 0;
3083 out:
3084         free(info_linear);
3085         free(info_node);
3086         up_write(&env->bpf_progs.lock);
3087         return err;
3088 }
3089
3090 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3091 {
3092         struct perf_env *env = &ff->ph->env;
3093         struct btf_node *node = NULL;
3094         u32 count, i;
3095         int err = -1;
3096
3097         if (ff->ph->needs_swap) {
3098                 pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3099                 return 0;
3100         }
3101
3102         if (do_read_u32(ff, &count))
3103                 return -1;
3104
3105         down_write(&env->bpf_progs.lock);
3106
3107         for (i = 0; i < count; ++i) {
3108                 u32 id, data_size;
3109
3110                 if (do_read_u32(ff, &id))
3111                         goto out;
3112                 if (do_read_u32(ff, &data_size))
3113                         goto out;
3114
3115                 node = malloc(sizeof(struct btf_node) + data_size);
3116                 if (!node)
3117                         goto out;
3118
3119                 node->id = id;
3120                 node->data_size = data_size;
3121
3122                 if (__do_read(ff, node->data, data_size))
3123                         goto out;
3124
3125                 perf_env__insert_btf(env, node);
3126                 node = NULL;
3127         }
3128
3129         err = 0;
3130 out:
3131         up_write(&env->bpf_progs.lock);
3132         free(node);
3133         return err;
3134 }
3135 #endif // HAVE_LIBBPF_SUPPORT
3136
3137 static int process_compressed(struct feat_fd *ff,
3138                               void *data __maybe_unused)
3139 {
3140         if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3141                 return -1;
3142
3143         if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3144                 return -1;
3145
3146         if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3147                 return -1;
3148
3149         if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3150                 return -1;
3151
3152         if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3153                 return -1;
3154
3155         return 0;
3156 }
3157
3158 static int process_per_cpu_pmu_caps(struct feat_fd *ff, int *nr_cpu_pmu_caps,
3159                                     char **cpu_pmu_caps,
3160                                     unsigned int *max_branches)
3161 {
3162         char *name, *value;
3163         struct strbuf sb;
3164         u32 nr_caps;
3165
3166         if (do_read_u32(ff, &nr_caps))
3167                 return -1;
3168
3169         if (!nr_caps) {
3170                 pr_debug("cpu pmu capabilities not available\n");
3171                 return 0;
3172         }
3173
3174         *nr_cpu_pmu_caps = nr_caps;
3175
3176         if (strbuf_init(&sb, 128) < 0)
3177                 return -1;
3178
3179         while (nr_caps--) {
3180                 name = do_read_string(ff);
3181                 if (!name)
3182                         goto error;
3183
3184                 value = do_read_string(ff);
3185                 if (!value)
3186                         goto free_name;
3187
3188                 if (strbuf_addf(&sb, "%s=%s", name, value) < 0)
3189                         goto free_value;
3190
3191                 /* include a NULL character at the end */
3192                 if (strbuf_add(&sb, "", 1) < 0)
3193                         goto free_value;
3194
3195                 if (!strcmp(name, "branches"))
3196                         *max_branches = atoi(value);
3197
3198                 free(value);
3199                 free(name);
3200         }
3201         *cpu_pmu_caps = strbuf_detach(&sb, NULL);
3202         return 0;
3203
3204 free_value:
3205         free(value);
3206 free_name:
3207         free(name);
3208 error:
3209         strbuf_release(&sb);
3210         return -1;
3211 }
3212
3213 static int process_cpu_pmu_caps(struct feat_fd *ff,
3214                                 void *data __maybe_unused)
3215 {
3216         return process_per_cpu_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3217                                         &ff->ph->env.cpu_pmu_caps,
3218                                         &ff->ph->env.max_branches);
3219 }
3220
3221 static int process_hybrid_cpu_pmu_caps(struct feat_fd *ff,
3222                                        void *data __maybe_unused)
3223 {
3224         struct hybrid_cpc_node *nodes;
3225         u32 nr_pmu, i;
3226         int ret;
3227
3228         if (do_read_u32(ff, &nr_pmu))
3229                 return -1;
3230
3231         if (!nr_pmu) {
3232                 pr_debug("hybrid cpu pmu capabilities not available\n");
3233                 return 0;
3234         }
3235
3236         nodes = zalloc(sizeof(*nodes) * nr_pmu);
3237         if (!nodes)
3238                 return -ENOMEM;
3239
3240         for (i = 0; i < nr_pmu; i++) {
3241                 struct hybrid_cpc_node *n = &nodes[i];
3242
3243                 ret = process_per_cpu_pmu_caps(ff, &n->nr_cpu_pmu_caps,
3244                                                &n->cpu_pmu_caps,
3245                                                &n->max_branches);
3246                 if (ret)
3247                         goto err;
3248
3249                 n->pmu_name = do_read_string(ff);
3250                 if (!n->pmu_name) {
3251                         ret = -1;
3252                         goto err;
3253                 }
3254         }
3255
3256         ff->ph->env.nr_hybrid_cpc_nodes = nr_pmu;
3257         ff->ph->env.hybrid_cpc_nodes = nodes;
3258         return 0;
3259
3260 err:
3261         for (i = 0; i < nr_pmu; i++) {
3262                 free(nodes[i].cpu_pmu_caps);
3263                 free(nodes[i].pmu_name);
3264         }
3265
3266         free(nodes);
3267         return ret;
3268 }
3269
3270 #define FEAT_OPR(n, func, __full_only) \
3271         [HEADER_##n] = {                                        \
3272                 .name       = __stringify(n),                   \
3273                 .write      = write_##func,                     \
3274                 .print      = print_##func,                     \
3275                 .full_only  = __full_only,                      \
3276                 .process    = process_##func,                   \
3277                 .synthesize = true                              \
3278         }
3279
3280 #define FEAT_OPN(n, func, __full_only) \
3281         [HEADER_##n] = {                                        \
3282                 .name       = __stringify(n),                   \
3283                 .write      = write_##func,                     \
3284                 .print      = print_##func,                     \
3285                 .full_only  = __full_only,                      \
3286                 .process    = process_##func                    \
3287         }
3288
3289 /* feature_ops not implemented: */
3290 #define print_tracing_data      NULL
3291 #define print_build_id          NULL
3292
3293 #define process_branch_stack    NULL
3294 #define process_stat            NULL
3295
3296 // Only used in util/synthetic-events.c
3297 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3298
3299 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3300         FEAT_OPN(TRACING_DATA,  tracing_data,   false),
3301         FEAT_OPN(BUILD_ID,      build_id,       false),
3302         FEAT_OPR(HOSTNAME,      hostname,       false),
3303         FEAT_OPR(OSRELEASE,     osrelease,      false),
3304         FEAT_OPR(VERSION,       version,        false),
3305         FEAT_OPR(ARCH,          arch,           false),
3306         FEAT_OPR(NRCPUS,        nrcpus,         false),
3307         FEAT_OPR(CPUDESC,       cpudesc,        false),
3308         FEAT_OPR(CPUID,         cpuid,          false),
3309         FEAT_OPR(TOTAL_MEM,     total_mem,      false),
3310         FEAT_OPR(EVENT_DESC,    event_desc,     false),
3311         FEAT_OPR(CMDLINE,       cmdline,        false),
3312         FEAT_OPR(CPU_TOPOLOGY,  cpu_topology,   true),
3313         FEAT_OPR(NUMA_TOPOLOGY, numa_topology,  true),
3314         FEAT_OPN(BRANCH_STACK,  branch_stack,   false),
3315         FEAT_OPR(PMU_MAPPINGS,  pmu_mappings,   false),
3316         FEAT_OPR(GROUP_DESC,    group_desc,     false),
3317         FEAT_OPN(AUXTRACE,      auxtrace,       false),
3318         FEAT_OPN(STAT,          stat,           false),
3319         FEAT_OPN(CACHE,         cache,          true),
3320         FEAT_OPR(SAMPLE_TIME,   sample_time,    false),
3321         FEAT_OPR(MEM_TOPOLOGY,  mem_topology,   true),
3322         FEAT_OPR(CLOCKID,       clockid,        false),
3323         FEAT_OPN(DIR_FORMAT,    dir_format,     false),
3324 #ifdef HAVE_LIBBPF_SUPPORT
3325         FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
3326         FEAT_OPR(BPF_BTF,       bpf_btf,        false),
3327 #endif
3328         FEAT_OPR(COMPRESSED,    compressed,     false),
3329         FEAT_OPR(CPU_PMU_CAPS,  cpu_pmu_caps,   false),
3330         FEAT_OPR(CLOCK_DATA,    clock_data,     false),
3331         FEAT_OPN(HYBRID_TOPOLOGY,       hybrid_topology,        true),
3332         FEAT_OPR(HYBRID_CPU_PMU_CAPS,   hybrid_cpu_pmu_caps,    false),
3333 };
3334
3335 struct header_print_data {
3336         FILE *fp;
3337         bool full; /* extended list of headers */
3338 };
3339
3340 static int perf_file_section__fprintf_info(struct perf_file_section *section,
3341                                            struct perf_header *ph,
3342                                            int feat, int fd, void *data)
3343 {
3344         struct header_print_data *hd = data;
3345         struct feat_fd ff;
3346
3347         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3348                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3349                                 "%d, continuing...\n", section->offset, feat);
3350                 return 0;
3351         }
3352         if (feat >= HEADER_LAST_FEATURE) {
3353                 pr_warning("unknown feature %d\n", feat);
3354                 return 0;
3355         }
3356         if (!feat_ops[feat].print)
3357                 return 0;
3358
3359         ff = (struct  feat_fd) {
3360                 .fd = fd,
3361                 .ph = ph,
3362         };
3363
3364         if (!feat_ops[feat].full_only || hd->full)
3365                 feat_ops[feat].print(&ff, hd->fp);
3366         else
3367                 fprintf(hd->fp, "# %s info available, use -I to display\n",
3368                         feat_ops[feat].name);
3369
3370         return 0;
3371 }
3372
3373 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3374 {
3375         struct header_print_data hd;
3376         struct perf_header *header = &session->header;
3377         int fd = perf_data__fd(session->data);
3378         struct stat st;
3379         time_t stctime;
3380         int ret, bit;
3381
3382         hd.fp = fp;
3383         hd.full = full;
3384
3385         ret = fstat(fd, &st);
3386         if (ret == -1)
3387                 return -1;
3388
3389         stctime = st.st_mtime;
3390         fprintf(fp, "# captured on    : %s", ctime(&stctime));
3391
3392         fprintf(fp, "# header version : %u\n", header->version);
3393         fprintf(fp, "# data offset    : %" PRIu64 "\n", header->data_offset);
3394         fprintf(fp, "# data size      : %" PRIu64 "\n", header->data_size);
3395         fprintf(fp, "# feat offset    : %" PRIu64 "\n", header->feat_offset);
3396
3397         perf_header__process_sections(header, fd, &hd,
3398                                       perf_file_section__fprintf_info);
3399
3400         if (session->data->is_pipe)
3401                 return 0;
3402
3403         fprintf(fp, "# missing features: ");
3404         for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3405                 if (bit)
3406                         fprintf(fp, "%s ", feat_ops[bit].name);
3407         }
3408
3409         fprintf(fp, "\n");
3410         return 0;
3411 }
3412
3413 static int do_write_feat(struct feat_fd *ff, int type,
3414                          struct perf_file_section **p,
3415                          struct evlist *evlist)
3416 {
3417         int err;
3418         int ret = 0;
3419
3420         if (perf_header__has_feat(ff->ph, type)) {
3421                 if (!feat_ops[type].write)
3422                         return -1;
3423
3424                 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3425                         return -1;
3426
3427                 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3428
3429                 err = feat_ops[type].write(ff, evlist);
3430                 if (err < 0) {
3431                         pr_debug("failed to write feature %s\n", feat_ops[type].name);
3432
3433                         /* undo anything written */
3434                         lseek(ff->fd, (*p)->offset, SEEK_SET);
3435
3436                         return -1;
3437                 }
3438                 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3439                 (*p)++;
3440         }
3441         return ret;
3442 }
3443
3444 static int perf_header__adds_write(struct perf_header *header,
3445                                    struct evlist *evlist, int fd)
3446 {
3447         int nr_sections;
3448         struct feat_fd ff;
3449         struct perf_file_section *feat_sec, *p;
3450         int sec_size;
3451         u64 sec_start;
3452         int feat;
3453         int err;
3454
3455         ff = (struct feat_fd){
3456                 .fd  = fd,
3457                 .ph = header,
3458         };
3459
3460         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3461         if (!nr_sections)
3462                 return 0;
3463
3464         feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3465         if (feat_sec == NULL)
3466                 return -ENOMEM;
3467
3468         sec_size = sizeof(*feat_sec) * nr_sections;
3469
3470         sec_start = header->feat_offset;
3471         lseek(fd, sec_start + sec_size, SEEK_SET);
3472
3473         for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3474                 if (do_write_feat(&ff, feat, &p, evlist))
3475                         perf_header__clear_feat(header, feat);
3476         }
3477
3478         lseek(fd, sec_start, SEEK_SET);
3479         /*
3480          * may write more than needed due to dropped feature, but
3481          * this is okay, reader will skip the missing entries
3482          */
3483         err = do_write(&ff, feat_sec, sec_size);
3484         if (err < 0)
3485                 pr_debug("failed to write feature section\n");
3486         free(feat_sec);
3487         return err;
3488 }
3489
3490 int perf_header__write_pipe(int fd)
3491 {
3492         struct perf_pipe_file_header f_header;
3493         struct feat_fd ff;
3494         int err;
3495
3496         ff = (struct feat_fd){ .fd = fd };
3497
3498         f_header = (struct perf_pipe_file_header){
3499                 .magic     = PERF_MAGIC,
3500                 .size      = sizeof(f_header),
3501         };
3502
3503         err = do_write(&ff, &f_header, sizeof(f_header));
3504         if (err < 0) {
3505                 pr_debug("failed to write perf pipe header\n");
3506                 return err;
3507         }
3508
3509         return 0;
3510 }
3511
3512 int perf_session__write_header(struct perf_session *session,
3513                                struct evlist *evlist,
3514                                int fd, bool at_exit)
3515 {
3516         struct perf_file_header f_header;
3517         struct perf_file_attr   f_attr;
3518         struct perf_header *header = &session->header;
3519         struct evsel *evsel;
3520         struct feat_fd ff;
3521         u64 attr_offset;
3522         int err;
3523
3524         ff = (struct feat_fd){ .fd = fd};
3525         lseek(fd, sizeof(f_header), SEEK_SET);
3526
3527         evlist__for_each_entry(session->evlist, evsel) {
3528                 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3529                 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3530                 if (err < 0) {
3531                         pr_debug("failed to write perf header\n");
3532                         return err;
3533                 }
3534         }
3535
3536         attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3537
3538         evlist__for_each_entry(evlist, evsel) {
3539                 if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3540                         /*
3541                          * We are likely in "perf inject" and have read
3542                          * from an older file. Update attr size so that
3543                          * reader gets the right offset to the ids.
3544                          */
3545                         evsel->core.attr.size = sizeof(evsel->core.attr);
3546                 }
3547                 f_attr = (struct perf_file_attr){
3548                         .attr = evsel->core.attr,
3549                         .ids  = {
3550                                 .offset = evsel->id_offset,
3551                                 .size   = evsel->core.ids * sizeof(u64),
3552                         }
3553                 };
3554                 err = do_write(&ff, &f_attr, sizeof(f_attr));
3555                 if (err < 0) {
3556                         pr_debug("failed to write perf header attribute\n");
3557                         return err;
3558                 }
3559         }
3560
3561         if (!header->data_offset)
3562                 header->data_offset = lseek(fd, 0, SEEK_CUR);
3563         header->feat_offset = header->data_offset + header->data_size;
3564
3565         if (at_exit) {
3566                 err = perf_header__adds_write(header, evlist, fd);
3567                 if (err < 0)
3568                         return err;
3569         }
3570
3571         f_header = (struct perf_file_header){
3572                 .magic     = PERF_MAGIC,
3573                 .size      = sizeof(f_header),
3574                 .attr_size = sizeof(f_attr),
3575                 .attrs = {
3576                         .offset = attr_offset,
3577                         .size   = evlist->core.nr_entries * sizeof(f_attr),
3578                 },
3579                 .data = {
3580                         .offset = header->data_offset,
3581                         .size   = header->data_size,
3582                 },
3583                 /* event_types is ignored, store zeros */
3584         };
3585
3586         memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3587
3588         lseek(fd, 0, SEEK_SET);
3589         err = do_write(&ff, &f_header, sizeof(f_header));
3590         if (err < 0) {
3591                 pr_debug("failed to write perf header\n");
3592                 return err;
3593         }
3594         lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3595
3596         return 0;
3597 }
3598
3599 static int perf_header__getbuffer64(struct perf_header *header,
3600                                     int fd, void *buf, size_t size)
3601 {
3602         if (readn(fd, buf, size) <= 0)
3603                 return -1;
3604
3605         if (header->needs_swap)
3606                 mem_bswap_64(buf, size);
3607
3608         return 0;
3609 }
3610
3611 int perf_header__process_sections(struct perf_header *header, int fd,
3612                                   void *data,
3613                                   int (*process)(struct perf_file_section *section,
3614                                                  struct perf_header *ph,
3615                                                  int feat, int fd, void *data))
3616 {
3617         struct perf_file_section *feat_sec, *sec;
3618         int nr_sections;
3619         int sec_size;
3620         int feat;
3621         int err;
3622
3623         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3624         if (!nr_sections)
3625                 return 0;
3626
3627         feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3628         if (!feat_sec)
3629                 return -1;
3630
3631         sec_size = sizeof(*feat_sec) * nr_sections;
3632
3633         lseek(fd, header->feat_offset, SEEK_SET);
3634
3635         err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3636         if (err < 0)
3637                 goto out_free;
3638
3639         for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3640                 err = process(sec++, header, feat, fd, data);
3641                 if (err < 0)
3642                         goto out_free;
3643         }
3644         err = 0;
3645 out_free:
3646         free(feat_sec);
3647         return err;
3648 }
3649
3650 static const int attr_file_abi_sizes[] = {
3651         [0] = PERF_ATTR_SIZE_VER0,
3652         [1] = PERF_ATTR_SIZE_VER1,
3653         [2] = PERF_ATTR_SIZE_VER2,
3654         [3] = PERF_ATTR_SIZE_VER3,
3655         [4] = PERF_ATTR_SIZE_VER4,
3656         0,
3657 };
3658
3659 /*
3660  * In the legacy file format, the magic number is not used to encode endianness.
3661  * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3662  * on ABI revisions, we need to try all combinations for all endianness to
3663  * detect the endianness.
3664  */
3665 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3666 {
3667         uint64_t ref_size, attr_size;
3668         int i;
3669
3670         for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3671                 ref_size = attr_file_abi_sizes[i]
3672                          + sizeof(struct perf_file_section);
3673                 if (hdr_sz != ref_size) {
3674                         attr_size = bswap_64(hdr_sz);
3675                         if (attr_size != ref_size)
3676                                 continue;
3677
3678                         ph->needs_swap = true;
3679                 }
3680                 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3681                          i,
3682                          ph->needs_swap);
3683                 return 0;
3684         }
3685         /* could not determine endianness */
3686         return -1;
3687 }
3688
3689 #define PERF_PIPE_HDR_VER0      16
3690
3691 static const size_t attr_pipe_abi_sizes[] = {
3692         [0] = PERF_PIPE_HDR_VER0,
3693         0,
3694 };
3695
3696 /*
3697  * In the legacy pipe format, there is an implicit assumption that endianness
3698  * between host recording the samples, and host parsing the samples is the
3699  * same. This is not always the case given that the pipe output may always be
3700  * redirected into a file and analyzed on a different machine with possibly a
3701  * different endianness and perf_event ABI revisions in the perf tool itself.
3702  */
3703 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3704 {
3705         u64 attr_size;
3706         int i;
3707
3708         for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3709                 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3710                         attr_size = bswap_64(hdr_sz);
3711                         if (attr_size != hdr_sz)
3712                                 continue;
3713
3714                         ph->needs_swap = true;
3715                 }
3716                 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3717                 return 0;
3718         }
3719         return -1;
3720 }
3721
3722 bool is_perf_magic(u64 magic)
3723 {
3724         if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3725                 || magic == __perf_magic2
3726                 || magic == __perf_magic2_sw)
3727                 return true;
3728
3729         return false;
3730 }
3731
3732 static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3733                               bool is_pipe, struct perf_header *ph)
3734 {
3735         int ret;
3736
3737         /* check for legacy format */
3738         ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3739         if (ret == 0) {
3740                 ph->version = PERF_HEADER_VERSION_1;
3741                 pr_debug("legacy perf.data format\n");
3742                 if (is_pipe)
3743                         return try_all_pipe_abis(hdr_sz, ph);
3744
3745                 return try_all_file_abis(hdr_sz, ph);
3746         }
3747         /*
3748          * the new magic number serves two purposes:
3749          * - unique number to identify actual perf.data files
3750          * - encode endianness of file
3751          */
3752         ph->version = PERF_HEADER_VERSION_2;
3753
3754         /* check magic number with one endianness */
3755         if (magic == __perf_magic2)
3756                 return 0;
3757
3758         /* check magic number with opposite endianness */
3759         if (magic != __perf_magic2_sw)
3760                 return -1;
3761
3762         ph->needs_swap = true;
3763
3764         return 0;
3765 }
3766
3767 int perf_file_header__read(struct perf_file_header *header,
3768                            struct perf_header *ph, int fd)
3769 {
3770         ssize_t ret;
3771
3772         lseek(fd, 0, SEEK_SET);
3773
3774         ret = readn(fd, header, sizeof(*header));
3775         if (ret <= 0)
3776                 return -1;
3777
3778         if (check_magic_endian(header->magic,
3779                                header->attr_size, false, ph) < 0) {
3780                 pr_debug("magic/endian check failed\n");
3781                 return -1;
3782         }
3783
3784         if (ph->needs_swap) {
3785                 mem_bswap_64(header, offsetof(struct perf_file_header,
3786                              adds_features));
3787         }
3788
3789         if (header->size != sizeof(*header)) {
3790                 /* Support the previous format */
3791                 if (header->size == offsetof(typeof(*header), adds_features))
3792                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3793                 else
3794                         return -1;
3795         } else if (ph->needs_swap) {
3796                 /*
3797                  * feature bitmap is declared as an array of unsigned longs --
3798                  * not good since its size can differ between the host that
3799                  * generated the data file and the host analyzing the file.
3800                  *
3801                  * We need to handle endianness, but we don't know the size of
3802                  * the unsigned long where the file was generated. Take a best
3803                  * guess at determining it: try 64-bit swap first (ie., file
3804                  * created on a 64-bit host), and check if the hostname feature
3805                  * bit is set (this feature bit is forced on as of fbe96f2).
3806                  * If the bit is not, undo the 64-bit swap and try a 32-bit
3807                  * swap. If the hostname bit is still not set (e.g., older data
3808                  * file), punt and fallback to the original behavior --
3809                  * clearing all feature bits and setting buildid.
3810                  */
3811                 mem_bswap_64(&header->adds_features,
3812                             BITS_TO_U64(HEADER_FEAT_BITS));
3813
3814                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3815                         /* unswap as u64 */
3816                         mem_bswap_64(&header->adds_features,
3817                                     BITS_TO_U64(HEADER_FEAT_BITS));
3818
3819                         /* unswap as u32 */
3820                         mem_bswap_32(&header->adds_features,
3821                                     BITS_TO_U32(HEADER_FEAT_BITS));
3822                 }
3823
3824                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3825                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3826                         set_bit(HEADER_BUILD_ID, header->adds_features);
3827                 }
3828         }
3829
3830         memcpy(&ph->adds_features, &header->adds_features,
3831                sizeof(ph->adds_features));
3832
3833         ph->data_offset  = header->data.offset;
3834         ph->data_size    = header->data.size;
3835         ph->feat_offset  = header->data.offset + header->data.size;
3836         return 0;
3837 }
3838
3839 static int perf_file_section__process(struct perf_file_section *section,
3840                                       struct perf_header *ph,
3841                                       int feat, int fd, void *data)
3842 {
3843         struct feat_fd fdd = {
3844                 .fd     = fd,
3845                 .ph     = ph,
3846                 .size   = section->size,
3847                 .offset = section->offset,
3848         };
3849
3850         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3851                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3852                           "%d, continuing...\n", section->offset, feat);
3853                 return 0;
3854         }
3855
3856         if (feat >= HEADER_LAST_FEATURE) {
3857                 pr_debug("unknown feature %d, continuing...\n", feat);
3858                 return 0;
3859         }
3860
3861         if (!feat_ops[feat].process)
3862                 return 0;
3863
3864         return feat_ops[feat].process(&fdd, data);
3865 }
3866
3867 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
3868                                        struct perf_header *ph,
3869                                        struct perf_data* data,
3870                                        bool repipe, int repipe_fd)
3871 {
3872         struct feat_fd ff = {
3873                 .fd = repipe_fd,
3874                 .ph = ph,
3875         };
3876         ssize_t ret;
3877
3878         ret = perf_data__read(data, header, sizeof(*header));
3879         if (ret <= 0)
3880                 return -1;
3881
3882         if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
3883                 pr_debug("endian/magic failed\n");
3884                 return -1;
3885         }
3886
3887         if (ph->needs_swap)
3888                 header->size = bswap_64(header->size);
3889
3890         if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
3891                 return -1;
3892
3893         return 0;
3894 }
3895
3896 static int perf_header__read_pipe(struct perf_session *session, int repipe_fd)
3897 {
3898         struct perf_header *header = &session->header;
3899         struct perf_pipe_file_header f_header;
3900
3901         if (perf_file_header__read_pipe(&f_header, header, session->data,
3902                                         session->repipe, repipe_fd) < 0) {
3903                 pr_debug("incompatible file format\n");
3904                 return -EINVAL;
3905         }
3906
3907         return f_header.size == sizeof(f_header) ? 0 : -1;
3908 }
3909
3910 static int read_attr(int fd, struct perf_header *ph,
3911                      struct perf_file_attr *f_attr)
3912 {
3913         struct perf_event_attr *attr = &f_attr->attr;
3914         size_t sz, left;
3915         size_t our_sz = sizeof(f_attr->attr);
3916         ssize_t ret;
3917
3918         memset(f_attr, 0, sizeof(*f_attr));
3919
3920         /* read minimal guaranteed structure */
3921         ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
3922         if (ret <= 0) {
3923                 pr_debug("cannot read %d bytes of header attr\n",
3924                          PERF_ATTR_SIZE_VER0);
3925                 return -1;
3926         }
3927
3928         /* on file perf_event_attr size */
3929         sz = attr->size;
3930
3931         if (ph->needs_swap)
3932                 sz = bswap_32(sz);
3933
3934         if (sz == 0) {
3935                 /* assume ABI0 */
3936                 sz =  PERF_ATTR_SIZE_VER0;
3937         } else if (sz > our_sz) {
3938                 pr_debug("file uses a more recent and unsupported ABI"
3939                          " (%zu bytes extra)\n", sz - our_sz);
3940                 return -1;
3941         }
3942         /* what we have not yet read and that we know about */
3943         left = sz - PERF_ATTR_SIZE_VER0;
3944         if (left) {
3945                 void *ptr = attr;
3946                 ptr += PERF_ATTR_SIZE_VER0;
3947
3948                 ret = readn(fd, ptr, left);
3949         }
3950         /* read perf_file_section, ids are read in caller */
3951         ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
3952
3953         return ret <= 0 ? -1 : 0;
3954 }
3955
3956 static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
3957 {
3958         struct tep_event *event;
3959         char bf[128];
3960
3961         /* already prepared */
3962         if (evsel->tp_format)
3963                 return 0;
3964
3965         if (pevent == NULL) {
3966                 pr_debug("broken or missing trace data\n");
3967                 return -1;
3968         }
3969
3970         event = tep_find_event(pevent, evsel->core.attr.config);
3971         if (event == NULL) {
3972                 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
3973                 return -1;
3974         }
3975
3976         if (!evsel->name) {
3977                 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
3978                 evsel->name = strdup(bf);
3979                 if (evsel->name == NULL)
3980                         return -1;
3981         }
3982
3983         evsel->tp_format = event;
3984         return 0;
3985 }
3986
3987 static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
3988 {
3989         struct evsel *pos;
3990
3991         evlist__for_each_entry(evlist, pos) {
3992                 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
3993                     evsel__prepare_tracepoint_event(pos, pevent))
3994                         return -1;
3995         }
3996
3997         return 0;
3998 }
3999
4000 int perf_session__read_header(struct perf_session *session, int repipe_fd)
4001 {
4002         struct perf_data *data = session->data;
4003         struct perf_header *header = &session->header;
4004         struct perf_file_header f_header;
4005         struct perf_file_attr   f_attr;
4006         u64                     f_id;
4007         int nr_attrs, nr_ids, i, j, err;
4008         int fd = perf_data__fd(data);
4009
4010         session->evlist = evlist__new();
4011         if (session->evlist == NULL)
4012                 return -ENOMEM;
4013
4014         session->evlist->env = &header->env;
4015         session->machines.host.env = &header->env;
4016
4017         /*
4018          * We can read 'pipe' data event from regular file,
4019          * check for the pipe header regardless of source.
4020          */
4021         err = perf_header__read_pipe(session, repipe_fd);
4022         if (!err || perf_data__is_pipe(data)) {
4023                 data->is_pipe = true;
4024                 return err;
4025         }
4026
4027         if (perf_file_header__read(&f_header, header, fd) < 0)
4028                 return -EINVAL;
4029
4030         if (header->needs_swap && data->in_place_update) {
4031                 pr_err("In-place update not supported when byte-swapping is required\n");
4032                 return -EINVAL;
4033         }
4034
4035         /*
4036          * Sanity check that perf.data was written cleanly; data size is
4037          * initialized to 0 and updated only if the on_exit function is run.
4038          * If data size is still 0 then the file contains only partial
4039          * information.  Just warn user and process it as much as it can.
4040          */
4041         if (f_header.data.size == 0) {
4042                 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4043                            "Was the 'perf record' command properly terminated?\n",
4044                            data->file.path);
4045         }
4046
4047         if (f_header.attr_size == 0) {
4048                 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4049                        "Was the 'perf record' command properly terminated?\n",
4050                        data->file.path);
4051                 return -EINVAL;
4052         }
4053
4054         nr_attrs = f_header.attrs.size / f_header.attr_size;
4055         lseek(fd, f_header.attrs.offset, SEEK_SET);
4056
4057         for (i = 0; i < nr_attrs; i++) {
4058                 struct evsel *evsel;
4059                 off_t tmp;
4060
4061                 if (read_attr(fd, header, &f_attr) < 0)
4062                         goto out_errno;
4063
4064                 if (header->needs_swap) {
4065                         f_attr.ids.size   = bswap_64(f_attr.ids.size);
4066                         f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4067                         perf_event__attr_swap(&f_attr.attr);
4068                 }
4069
4070                 tmp = lseek(fd, 0, SEEK_CUR);
4071                 evsel = evsel__new(&f_attr.attr);
4072
4073                 if (evsel == NULL)
4074                         goto out_delete_evlist;
4075
4076                 evsel->needs_swap = header->needs_swap;
4077                 /*
4078                  * Do it before so that if perf_evsel__alloc_id fails, this
4079                  * entry gets purged too at evlist__delete().
4080                  */
4081                 evlist__add(session->evlist, evsel);
4082
4083                 nr_ids = f_attr.ids.size / sizeof(u64);
4084                 /*
4085                  * We don't have the cpu and thread maps on the header, so
4086                  * for allocating the perf_sample_id table we fake 1 cpu and
4087                  * hattr->ids threads.
4088                  */
4089                 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4090                         goto out_delete_evlist;
4091
4092                 lseek(fd, f_attr.ids.offset, SEEK_SET);
4093
4094                 for (j = 0; j < nr_ids; j++) {
4095                         if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4096                                 goto out_errno;
4097
4098                         perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4099                 }
4100
4101                 lseek(fd, tmp, SEEK_SET);
4102         }
4103
4104         perf_header__process_sections(header, fd, &session->tevent,
4105                                       perf_file_section__process);
4106
4107         if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4108                 goto out_delete_evlist;
4109
4110         return 0;
4111 out_errno:
4112         return -errno;
4113
4114 out_delete_evlist:
4115         evlist__delete(session->evlist);
4116         session->evlist = NULL;
4117         return -ENOMEM;
4118 }
4119
4120 int perf_event__process_feature(struct perf_session *session,
4121                                 union perf_event *event)
4122 {
4123         struct perf_tool *tool = session->tool;
4124         struct feat_fd ff = { .fd = 0 };
4125         struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4126         int type = fe->header.type;
4127         u64 feat = fe->feat_id;
4128         int ret = 0;
4129
4130         if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4131                 pr_warning("invalid record type %d in pipe-mode\n", type);
4132                 return 0;
4133         }
4134         if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4135                 pr_warning("invalid record type %d in pipe-mode\n", type);
4136                 return -1;
4137         }
4138
4139         if (!feat_ops[feat].process)
4140                 return 0;
4141
4142         ff.buf  = (void *)fe->data;
4143         ff.size = event->header.size - sizeof(*fe);
4144         ff.ph = &session->header;
4145
4146         if (feat_ops[feat].process(&ff, NULL)) {
4147                 ret = -1;
4148                 goto out;
4149         }
4150
4151         if (!feat_ops[feat].print || !tool->show_feat_hdr)
4152                 goto out;
4153
4154         if (!feat_ops[feat].full_only ||
4155             tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4156                 feat_ops[feat].print(&ff, stdout);
4157         } else {
4158                 fprintf(stdout, "# %s info available, use -I to display\n",
4159                         feat_ops[feat].name);
4160         }
4161 out:
4162         free_event_desc(ff.events);
4163         return ret;
4164 }
4165
4166 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4167 {
4168         struct perf_record_event_update *ev = &event->event_update;
4169         struct perf_record_event_update_scale *ev_scale;
4170         struct perf_record_event_update_cpus *ev_cpus;
4171         struct perf_cpu_map *map;
4172         size_t ret;
4173
4174         ret = fprintf(fp, "\n... id:    %" PRI_lu64 "\n", ev->id);
4175
4176         switch (ev->type) {
4177         case PERF_EVENT_UPDATE__SCALE:
4178                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4179                 ret += fprintf(fp, "... scale: %f\n", ev_scale->scale);
4180                 break;
4181         case PERF_EVENT_UPDATE__UNIT:
4182                 ret += fprintf(fp, "... unit:  %s\n", ev->data);
4183                 break;
4184         case PERF_EVENT_UPDATE__NAME:
4185                 ret += fprintf(fp, "... name:  %s\n", ev->data);
4186                 break;
4187         case PERF_EVENT_UPDATE__CPUS:
4188                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4189                 ret += fprintf(fp, "... ");
4190
4191                 map = cpu_map__new_data(&ev_cpus->cpus);
4192                 if (map)
4193                         ret += cpu_map__fprintf(map, fp);
4194                 else
4195                         ret += fprintf(fp, "failed to get cpus\n");
4196                 break;
4197         default:
4198                 ret += fprintf(fp, "... unknown type\n");
4199                 break;
4200         }
4201
4202         return ret;
4203 }
4204
4205 int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4206                              union perf_event *event,
4207                              struct evlist **pevlist)
4208 {
4209         u32 i, ids, n_ids;
4210         struct evsel *evsel;
4211         struct evlist *evlist = *pevlist;
4212
4213         if (evlist == NULL) {
4214                 *pevlist = evlist = evlist__new();
4215                 if (evlist == NULL)
4216                         return -ENOMEM;
4217         }
4218
4219         evsel = evsel__new(&event->attr.attr);
4220         if (evsel == NULL)
4221                 return -ENOMEM;
4222
4223         evlist__add(evlist, evsel);
4224
4225         ids = event->header.size;
4226         ids -= (void *)&event->attr.id - (void *)event;
4227         n_ids = ids / sizeof(u64);
4228         /*
4229          * We don't have the cpu and thread maps on the header, so
4230          * for allocating the perf_sample_id table we fake 1 cpu and
4231          * hattr->ids threads.
4232          */
4233         if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4234                 return -ENOMEM;
4235
4236         for (i = 0; i < n_ids; i++) {
4237                 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, event->attr.id[i]);
4238         }
4239
4240         return 0;
4241 }
4242
4243 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4244                                      union perf_event *event,
4245                                      struct evlist **pevlist)
4246 {
4247         struct perf_record_event_update *ev = &event->event_update;
4248         struct perf_record_event_update_scale *ev_scale;
4249         struct perf_record_event_update_cpus *ev_cpus;
4250         struct evlist *evlist;
4251         struct evsel *evsel;
4252         struct perf_cpu_map *map;
4253
4254         if (!pevlist || *pevlist == NULL)
4255                 return -EINVAL;
4256
4257         evlist = *pevlist;
4258
4259         evsel = evlist__id2evsel(evlist, ev->id);
4260         if (evsel == NULL)
4261                 return -EINVAL;
4262
4263         switch (ev->type) {
4264         case PERF_EVENT_UPDATE__UNIT:
4265                 free((char *)evsel->unit);
4266                 evsel->unit = strdup(ev->data);
4267                 break;
4268         case PERF_EVENT_UPDATE__NAME:
4269                 free(evsel->name);
4270                 evsel->name = strdup(ev->data);
4271                 break;
4272         case PERF_EVENT_UPDATE__SCALE:
4273                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4274                 evsel->scale = ev_scale->scale;
4275                 break;
4276         case PERF_EVENT_UPDATE__CPUS:
4277                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4278                 map = cpu_map__new_data(&ev_cpus->cpus);
4279                 if (map) {
4280                         perf_cpu_map__put(evsel->core.own_cpus);
4281                         evsel->core.own_cpus = map;
4282                 } else
4283                         pr_err("failed to get event_update cpus\n");
4284         default:
4285                 break;
4286         }
4287
4288         return 0;
4289 }
4290
4291 int perf_event__process_tracing_data(struct perf_session *session,
4292                                      union perf_event *event)
4293 {
4294         ssize_t size_read, padding, size = event->tracing_data.size;
4295         int fd = perf_data__fd(session->data);
4296         char buf[BUFSIZ];
4297
4298         /*
4299          * The pipe fd is already in proper place and in any case
4300          * we can't move it, and we'd screw the case where we read
4301          * 'pipe' data from regular file. The trace_report reads
4302          * data from 'fd' so we need to set it directly behind the
4303          * event, where the tracing data starts.
4304          */
4305         if (!perf_data__is_pipe(session->data)) {
4306                 off_t offset = lseek(fd, 0, SEEK_CUR);
4307
4308                 /* setup for reading amidst mmap */
4309                 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4310                       SEEK_SET);
4311         }
4312
4313         size_read = trace_report(fd, &session->tevent,
4314                                  session->repipe);
4315         padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4316
4317         if (readn(fd, buf, padding) < 0) {
4318                 pr_err("%s: reading input file", __func__);
4319                 return -1;
4320         }
4321         if (session->repipe) {
4322                 int retw = write(STDOUT_FILENO, buf, padding);
4323                 if (retw <= 0 || retw != padding) {
4324                         pr_err("%s: repiping tracing data padding", __func__);
4325                         return -1;
4326                 }
4327         }
4328
4329         if (size_read + padding != size) {
4330                 pr_err("%s: tracing data size mismatch", __func__);
4331                 return -1;
4332         }
4333
4334         evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4335
4336         return size_read + padding;
4337 }
4338
4339 int perf_event__process_build_id(struct perf_session *session,
4340                                  union perf_event *event)
4341 {
4342         __event_process_build_id(&event->build_id,
4343                                  event->build_id.filename,
4344                                  session);
4345         return 0;
4346 }