1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <inttypes.h>
4 #include "string2.h"
5 #include <sys/param.h>
6 #include <sys/types.h>
7 #include <byteswap.h>
8 #include <unistd.h>
9 #include <regex.h>
10 #include <stdio.h>
11 #include <stdlib.h>
12 #include <linux/compiler.h>
13 #include <linux/list.h>
14 #include <linux/kernel.h>
15 #include <linux/bitops.h>
16 #include <linux/string.h>
17 #include <linux/stringify.h>
18 #include <linux/zalloc.h>
19 #include <sys/stat.h>
20 #include <sys/utsname.h>
21 #include <linux/time64.h>
22 #include <dirent.h>
23 #ifdef HAVE_LIBBPF_SUPPORT
24 #include <bpf/libbpf.h>
25 #endif
26 #include <perf/cpumap.h>
27 #include <tools/libc_compat.h> // reallocarray
28
29 #include "dso.h"
30 #include "evlist.h"
31 #include "evsel.h"
32 #include "util/evsel_fprintf.h"
33 #include "header.h"
34 #include "memswap.h"
35 #include "trace-event.h"
36 #include "session.h"
37 #include "symbol.h"
38 #include "debug.h"
39 #include "cpumap.h"
40 #include "pmu.h"
41 #include "pmus.h"
42 #include "vdso.h"
43 #include "strbuf.h"
44 #include "build-id.h"
45 #include "data.h"
46 #include <api/fs/fs.h>
47 #include "asm/bug.h"
48 #include "tool.h"
49 #include "time-utils.h"
50 #include "units.h"
51 #include "util/util.h" // perf_exe()
52 #include "cputopo.h"
53 #include "bpf-event.h"
54 #include "bpf-utils.h"
55 #include "clockid.h"
56
57 #include <linux/ctype.h>
58 #include <internal/lib.h>
59
60 #ifdef HAVE_LIBTRACEEVENT
61 #include <traceevent/event-parse.h>
62 #endif
63
64 /*
65 * magic2 = "PERFILE2"
66 * must be a numerical value to let the endianness
67 * determine the memory layout. That way we are able
68 * to detect endianness when reading the perf.data file
69 * back.
70 *
71 * we check for legacy (PERFFILE) format.
72 */
73 static const char *__perf_magic1 = "PERFFILE";
74 static const u64 __perf_magic2 = 0x32454c4946524550ULL;
75 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
76
77 #define PERF_MAGIC __perf_magic2
78
79 const char perf_version_string[] = PERF_VERSION;
80
81 struct perf_file_attr {
82 struct perf_event_attr attr;
83 struct perf_file_section ids;
84 };
85
86 void perf_header__set_feat(struct perf_header *header, int feat)
87 {
88 __set_bit(feat, header->adds_features);
89 }
90
91 void perf_header__clear_feat(struct perf_header *header, int feat)
92 {
93 __clear_bit(feat, header->adds_features);
94 }
95
96 bool perf_header__has_feat(const struct perf_header *header, int feat)
97 {
98 return test_bit(feat, header->adds_features);
99 }
100
101 static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
102 {
103 ssize_t ret = writen(ff->fd, buf, size);
104
105 if (ret != (ssize_t)size)
106 return ret < 0 ? (int)ret : -1;
107 return 0;
108 }
109
110 static int __do_write_buf(struct feat_fd *ff, const void *buf, size_t size)
111 {
112 /* struct perf_event_header::size is u16 */
113 const size_t max_size = 0xffff - sizeof(struct perf_event_header);
114 size_t new_size = ff->size;
115 void *addr;
116
117 if (size + ff->offset > max_size)
118 return -E2BIG;
119
120 while (size > (new_size - ff->offset))
121 new_size <<= 1;
122 new_size = min(max_size, new_size);
123
124 if (ff->size < new_size) {
125 addr = realloc(ff->buf, new_size);
126 if (!addr)
127 return -ENOMEM;
128 ff->buf = addr;
129 ff->size = new_size;
130 }
131
132 memcpy(ff->buf + ff->offset, buf, size);
133 ff->offset += size;
134
135 return 0;
136 }
137
138 /* Return: 0 if succeeded, -ERR if failed. */
139 int do_write(struct feat_fd *ff, const void *buf, size_t size)
140 {
141 if (!ff->buf)
142 return __do_write_fd(ff, buf, size);
143 return __do_write_buf(ff, buf, size);
144 }
145
146 /* Return: 0 if succeeded, -ERR if failed. */
147 static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
148 {
149 u64 *p = (u64 *) set;
150 int i, ret;
151
152 ret = do_write(ff, &size, sizeof(size));
153 if (ret < 0)
154 return ret;
155
156 for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
157 ret = do_write(ff, p + i, sizeof(*p));
158 if (ret < 0)
159 return ret;
160 }
161
162 return 0;
163 }
164
165 /* Return: 0 if succeeded, -ERR if failed. */
166 int write_padded(struct feat_fd *ff, const void *bf,
167 size_t count, size_t count_aligned)
168 {
169 static const char zero_buf[NAME_ALIGN];
170 int err = do_write(ff, bf, count);
171
172 if (!err)
173 err = do_write(ff, zero_buf, count_aligned - count);
174
175 return err;
176 }
177
178 #define string_size(str) \
179 (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
180
181 /* Return: 0 if succeeded, -ERR if failed. */
182 static int do_write_string(struct feat_fd *ff, const char *str)
183 {
184 u32 len, olen;
185 int ret;
186
187 olen = strlen(str) + 1;
188 len = PERF_ALIGN(olen, NAME_ALIGN);
189
190 /* write len, incl. \0 */
191 ret = do_write(ff, &len, sizeof(len));
192 if (ret < 0)
193 return ret;
194
195 return write_padded(ff, str, olen, len);
196 }
197
198 static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
199 {
200 ssize_t ret = readn(ff->fd, addr, size);
201
202 if (ret != size)
203 return ret < 0 ? (int)ret : -1;
204 return 0;
205 }
206
207 static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
208 {
209 if (size > (ssize_t)ff->size - ff->offset)
210 return -1;
211
212 memcpy(addr, ff->buf + ff->offset, size);
213 ff->offset += size;
214
215 return 0;
216
217 }
218
219 static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
220 {
221 if (!ff->buf)
222 return __do_read_fd(ff, addr, size);
223 return __do_read_buf(ff, addr, size);
224 }
225
226 static int do_read_u32(struct feat_fd *ff, u32 *addr)
227 {
228 int ret;
229
230 ret = __do_read(ff, addr, sizeof(*addr));
231 if (ret)
232 return ret;
233
234 if (ff->ph->needs_swap)
235 *addr = bswap_32(*addr);
236 return 0;
237 }
238
239 static int do_read_u64(struct feat_fd *ff, u64 *addr)
240 {
241 int ret;
242
243 ret = __do_read(ff, addr, sizeof(*addr));
244 if (ret)
245 return ret;
246
247 if (ff->ph->needs_swap)
248 *addr = bswap_64(*addr);
249 return 0;
250 }
251
252 static char *do_read_string(struct feat_fd *ff)
253 {
254 u32 len;
255 char *buf;
256
257 if (do_read_u32(ff, &len))
258 return NULL;
259
260 buf = malloc(len);
261 if (!buf)
262 return NULL;
263
264 if (!__do_read(ff, buf, len)) {
265 /*
266 * strings are padded by zeroes
267 * thus the actual strlen of buf
268 * may be less than len
269 */
270 return buf;
271 }
272
273 free(buf);
274 return NULL;
275 }
276
277 /* Return: 0 if succeeded, -ERR if failed. */
278 static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
279 {
280 unsigned long *set;
281 u64 size, *p;
282 int i, ret;
283
284 ret = do_read_u64(ff, &size);
285 if (ret)
286 return ret;
287
288 set = bitmap_zalloc(size);
289 if (!set)
290 return -ENOMEM;
291
292 p = (u64 *) set;
293
294 for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
295 ret = do_read_u64(ff, p + i);
296 if (ret < 0) {
297 free(set);
298 return ret;
299 }
300 }
301
302 *pset = set;
303 *psize = size;
304 return 0;
305 }
306
307 #ifdef HAVE_LIBTRACEEVENT
308 static int write_tracing_data(struct feat_fd *ff,
309 struct evlist *evlist)
310 {
311 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
312 return -1;
313
314 return read_tracing_data(ff->fd, &evlist->core.entries);
315 }
316 #endif
317
318 static int write_build_id(struct feat_fd *ff,
319 struct evlist *evlist __maybe_unused)
320 {
321 struct perf_session *session;
322 int err;
323
324 session = container_of(ff->ph, struct perf_session, header);
325
326 if (!perf_session__read_build_ids(session, true))
327 return -1;
328
329 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
330 return -1;
331
332 err = perf_session__write_buildid_table(session, ff);
333 if (err < 0) {
334 pr_debug("failed to write buildid table\n");
335 return err;
336 }
337 perf_session__cache_build_ids(session);
338
339 return 0;
340 }
341
342 static int write_hostname(struct feat_fd *ff,
343 struct evlist *evlist __maybe_unused)
344 {
345 struct utsname uts;
346 int ret;
347
348 ret = uname(&uts);
349 if (ret < 0)
350 return -1;
351
352 return do_write_string(ff, uts.nodename);
353 }
354
355 static int write_osrelease(struct feat_fd *ff,
356 struct evlist *evlist __maybe_unused)
357 {
358 struct utsname uts;
359 int ret;
360
361 ret = uname(&uts);
362 if (ret < 0)
363 return -1;
364
365 return do_write_string(ff, uts.release);
366 }
367
368 static int write_arch(struct feat_fd *ff,
369 struct evlist *evlist __maybe_unused)
370 {
371 struct utsname uts;
372 int ret;
373
374 ret = uname(&uts);
375 if (ret < 0)
376 return -1;
377
378 return do_write_string(ff, uts.machine);
379 }
380
381 static int write_version(struct feat_fd *ff,
382 struct evlist *evlist __maybe_unused)
383 {
384 return do_write_string(ff, perf_version_string);
385 }
386
387 static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
388 {
389 FILE *file;
390 char *buf = NULL;
391 char *s, *p;
392 const char *search = cpuinfo_proc;
393 size_t len = 0;
394 int ret = -1;
395
396 if (!search)
397 return -1;
398
399 file = fopen("/proc/cpuinfo", "r");
400 if (!file)
401 return -1;
402
403 while (getline(&buf, &len, file) > 0) {
404 ret = strncmp(buf, search, strlen(search));
405 if (!ret)
406 break;
407 }
408
409 if (ret) {
410 ret = -1;
411 goto done;
412 }
413
414 s = buf;
415
416 p = strchr(buf, ':');
417 if (p && *(p+1) == ' ' && *(p+2))
418 s = p + 2;
419 p = strchr(s, '\n');
420 if (p)
421 *p = '\0';
422
423 /* squash extra space characters (branding string) */
424 p = s;
425 while (*p) {
426 if (isspace(*p)) {
427 char *r = p + 1;
428 char *q = skip_spaces(r);
429 *p = ' ';
430 if (q != (p+1))
431 while ((*r++ = *q++));
432 }
433 p++;
434 }
435 ret = do_write_string(ff, s);
436 done:
437 free(buf);
438 fclose(file);
439 return ret;
440 }
441
442 static int write_cpudesc(struct feat_fd *ff,
443 struct evlist *evlist __maybe_unused)
444 {
445 #if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
446 #define CPUINFO_PROC { "cpu", }
447 #elif defined(__s390__)
448 #define CPUINFO_PROC { "vendor_id", }
449 #elif defined(__sh__)
450 #define CPUINFO_PROC { "cpu type", }
451 #elif defined(__alpha__) || defined(__mips__)
452 #define CPUINFO_PROC { "cpu model", }
453 #elif defined(__arm__)
454 #define CPUINFO_PROC { "model name", "Processor", }
455 #elif defined(__arc__)
456 #define CPUINFO_PROC { "Processor", }
457 #elif defined(__xtensa__)
458 #define CPUINFO_PROC { "core ID", }
459 #elif defined(__loongarch__)
460 #define CPUINFO_PROC { "Model Name", }
461 #else
462 #define CPUINFO_PROC { "model name", }
463 #endif
464 const char *cpuinfo_procs[] = CPUINFO_PROC;
465 #undef CPUINFO_PROC
466 unsigned int i;
467
468 for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
469 int ret;
470 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
471 if (ret >= 0)
472 return ret;
473 }
474 return -1;
475 }
476
477
478 static int write_nrcpus(struct feat_fd *ff,
479 struct evlist *evlist __maybe_unused)
480 {
481 long nr;
482 u32 nrc, nra;
483 int ret;
484
485 nrc = cpu__max_present_cpu().cpu;
486
487 nr = sysconf(_SC_NPROCESSORS_ONLN);
488 if (nr < 0)
489 return -1;
490
491 nra = (u32)(nr & UINT_MAX);
492
493 ret = do_write(ff, &nrc, sizeof(nrc));
494 if (ret < 0)
495 return ret;
496
497 return do_write(ff, &nra, sizeof(nra));
498 }
499
500 static int write_event_desc(struct feat_fd *ff,
501 struct evlist *evlist)
502 {
503 struct evsel *evsel;
504 u32 nre, nri, sz;
505 int ret;
506
507 nre = evlist->core.nr_entries;
508
509 /*
510 * write number of events
511 */
512 ret = do_write(ff, &nre, sizeof(nre));
513 if (ret < 0)
514 return ret;
515
516 /*
517 * size of perf_event_attr struct
518 */
519 sz = (u32)sizeof(evsel->core.attr);
520 ret = do_write(ff, &sz, sizeof(sz));
521 if (ret < 0)
522 return ret;
523
524 evlist__for_each_entry(evlist, evsel) {
525 ret = do_write(ff, &evsel->core.attr, sz);
526 if (ret < 0)
527 return ret;
528 /*
529 * write number of unique id per event
530 * there is one id per instance of an event
531 *
532 * copy into an nri to be independent of the
533 * type of ids,
534 */
535 nri = evsel->core.ids;
536 ret = do_write(ff, &nri, sizeof(nri));
537 if (ret < 0)
538 return ret;
539
540 /*
541 * write event string as passed on cmdline
542 */
543 ret = do_write_string(ff, evsel__name(evsel));
544 if (ret < 0)
545 return ret;
546 /*
547 * write unique ids for this event
548 */
549 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
550 if (ret < 0)
551 return ret;
552 }
553 return 0;
554 }
555
556 static int write_cmdline(struct feat_fd *ff,
557 struct evlist *evlist __maybe_unused)
558 {
559 char pbuf[MAXPATHLEN], *buf;
560 int i, ret, n;
561
562 /* actual path to perf binary */
563 buf = perf_exe(pbuf, MAXPATHLEN);
564
565 /* account for binary path */
566 n = perf_env.nr_cmdline + 1;
567
568 ret = do_write(ff, &n, sizeof(n));
569 if (ret < 0)
570 return ret;
571
572 ret = do_write_string(ff, buf);
573 if (ret < 0)
574 return ret;
575
576 for (i = 0 ; i < perf_env.nr_cmdline; i++) {
577 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
578 if (ret < 0)
579 return ret;
580 }
581 return 0;
582 }
583
584
585 static int write_cpu_topology(struct feat_fd *ff,
586 struct evlist *evlist __maybe_unused)
587 {
588 struct cpu_topology *tp;
589 u32 i;
590 int ret, j;
591
592 tp = cpu_topology__new();
593 if (!tp)
594 return -1;
595
596 ret = do_write(ff, &tp->package_cpus_lists, sizeof(tp->package_cpus_lists));
597 if (ret < 0)
598 goto done;
599
600 for (i = 0; i < tp->package_cpus_lists; i++) {
601 ret = do_write_string(ff, tp->package_cpus_list[i]);
602 if (ret < 0)
603 goto done;
604 }
605 ret = do_write(ff, &tp->core_cpus_lists, sizeof(tp->core_cpus_lists));
606 if (ret < 0)
607 goto done;
608
609 for (i = 0; i < tp->core_cpus_lists; i++) {
610 ret = do_write_string(ff, tp->core_cpus_list[i]);
611 if (ret < 0)
612 break;
613 }
614
615 ret = perf_env__read_cpu_topology_map(&perf_env);
616 if (ret < 0)
617 goto done;
618
619 for (j = 0; j < perf_env.nr_cpus_avail; j++) {
620 ret = do_write(ff, &perf_env.cpu[j].core_id,
621 sizeof(perf_env.cpu[j].core_id));
622 if (ret < 0)
623 return ret;
624 ret = do_write(ff, &perf_env.cpu[j].socket_id,
625 sizeof(perf_env.cpu[j].socket_id));
626 if (ret < 0)
627 return ret;
628 }
629
630 if (!tp->die_cpus_lists)
631 goto done;
632
633 ret = do_write(ff, &tp->die_cpus_lists, sizeof(tp->die_cpus_lists));
634 if (ret < 0)
635 goto done;
636
637 for (i = 0; i < tp->die_cpus_lists; i++) {
638 ret = do_write_string(ff, tp->die_cpus_list[i]);
639 if (ret < 0)
640 goto done;
641 }
642
643 for (j = 0; j < perf_env.nr_cpus_avail; j++) {
644 ret = do_write(ff, &perf_env.cpu[j].die_id,
645 sizeof(perf_env.cpu[j].die_id));
646 if (ret < 0)
647 return ret;
648 }
649
650 done:
651 cpu_topology__delete(tp);
652 return ret;
653 }
654
655
656
657 static int write_total_mem(struct feat_fd *ff,
658 struct evlist *evlist __maybe_unused)
659 {
660 char *buf = NULL;
661 FILE *fp;
662 size_t len = 0;
663 int ret = -1, n;
664 uint64_t mem;
665
666 fp = fopen("/proc/meminfo", "r");
667 if (!fp)
668 return -1;
669
670 while (getline(&buf, &len, fp) > 0) {
671 ret = strncmp(buf, "MemTotal:", 9);
672 if (!ret)
673 break;
674 }
675 if (!ret) {
676 n = sscanf(buf, "%*s %"PRIu64, &mem);
677 if (n == 1)
678 ret = do_write(ff, &mem, sizeof(mem));
679 } else
680 ret = -1;
681 free(buf);
682 fclose(fp);
683 return ret;
684 }
685
686 static int write_numa_topology(struct feat_fd *ff,
687 struct evlist *evlist __maybe_unused)
688 {
689 struct numa_topology *tp;
690 int ret = -1;
691 u32 i;
692
693 tp = numa_topology__new();
694 if (!tp)
695 return -ENOMEM;
696
697 ret = do_write(ff, &tp->nr, sizeof(u32));
698 if (ret < 0)
699 goto err;
700
701 for (i = 0; i < tp->nr; i++) {
702 struct numa_topology_node *n = &tp->nodes[i];
703
704 ret = do_write(ff, &n->node, sizeof(u32));
705 if (ret < 0)
706 goto err;
707
708 ret = do_write(ff, &n->mem_total, sizeof(u64));
709 if (ret)
710 goto err;
711
712 ret = do_write(ff, &n->mem_free, sizeof(u64));
713 if (ret)
714 goto err;
715
716 ret = do_write_string(ff, n->cpus);
717 if (ret < 0)
718 goto err;
719 }
720
721 ret = 0;
722
723 err:
724 numa_topology__delete(tp);
725 return ret;
726 }
727
728 /*
729 * File format:
730 *
731 * struct pmu_mappings {
732 * u32 pmu_num;
733 * struct pmu_map {
734 * u32 type;
735 * char name[];
736 * }[pmu_num];
737 * };
738 */
739
740 static int write_pmu_mappings(struct feat_fd *ff,
741 struct evlist *evlist __maybe_unused)
742 {
743 struct perf_pmu *pmu = NULL;
744 u32 pmu_num = 0;
745 int ret;
746
747 /*
748 * Do a first pass to count number of pmu to avoid lseek so this
749 * works in pipe mode as well.
750 */
751 while ((pmu = perf_pmus__scan(pmu)))
752 pmu_num++;
753
754 ret = do_write(ff, &pmu_num, sizeof(pmu_num));
755 if (ret < 0)
756 return ret;
757
758 while ((pmu = perf_pmus__scan(pmu))) {
759 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
760 if (ret < 0)
761 return ret;
762
763 ret = do_write_string(ff, pmu->name);
764 if (ret < 0)
765 return ret;
766 }
767
768 return 0;
769 }
770
771 /*
772 * File format:
773 *
774 * struct group_descs {
775 * u32 nr_groups;
776 * struct group_desc {
777 * char name[];
778 * u32 leader_idx;
779 * u32 nr_members;
780 * }[nr_groups];
781 * };
782 */
783 static int write_group_desc(struct feat_fd *ff,
784 struct evlist *evlist)
785 {
786 u32 nr_groups = evlist__nr_groups(evlist);
787 struct evsel *evsel;
788 int ret;
789
790 ret = do_write(ff, &nr_groups, sizeof(nr_groups));
791 if (ret < 0)
792 return ret;
793
794 evlist__for_each_entry(evlist, evsel) {
795 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
796 const char *name = evsel->group_name ?: "{anon_group}";
797 u32 leader_idx = evsel->core.idx;
798 u32 nr_members = evsel->core.nr_members;
799
800 ret = do_write_string(ff, name);
801 if (ret < 0)
802 return ret;
803
804 ret = do_write(ff, &leader_idx, sizeof(leader_idx));
805 if (ret < 0)
806 return ret;
807
808 ret = do_write(ff, &nr_members, sizeof(nr_members));
809 if (ret < 0)
810 return ret;
811 }
812 }
813 return 0;
814 }
815
816 /*
817 * Return the CPU id as a raw string.
818 *
819 * Each architecture should provide a more precise id string that
820 * can be use to match the architecture's "mapfile".
821 */
822 char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
823 {
824 return NULL;
825 }
826
827 /* Return zero when the cpuid from the mapfile.csv matches the
828 * cpuid string generated on this platform.
829 * Otherwise return non-zero.
830 */
831 int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
832 {
833 regex_t re;
834 regmatch_t pmatch[1];
835 int match;
836
837 if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
838 /* Warn unable to generate match particular string. */
839 pr_info("Invalid regular expression %s\n", mapcpuid);
840 return 1;
841 }
842
843 match = !regexec(&re, cpuid, 1, pmatch, 0);
844 regfree(&re);
845 if (match) {
846 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
847
848 /* Verify the entire string matched. */
849 if (match_len == strlen(cpuid))
850 return 0;
851 }
852 return 1;
853 }
854
855 /*
856 * default get_cpuid(): nothing gets recorded
857 * actual implementation must be in arch/$(SRCARCH)/util/header.c
858 */
859 int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
860 {
861 return ENOSYS; /* Not implemented */
862 }
863
864 static int write_cpuid(struct feat_fd *ff,
865 struct evlist *evlist __maybe_unused)
866 {
867 char buffer[64];
868 int ret;
869
870 ret = get_cpuid(buffer, sizeof(buffer));
871 if (ret)
872 return -1;
873
874 return do_write_string(ff, buffer);
875 }
876
877 static int write_branch_stack(struct feat_fd *ff __maybe_unused,
878 struct evlist *evlist __maybe_unused)
879 {
880 return 0;
881 }
882
883 static int write_auxtrace(struct feat_fd *ff,
884 struct evlist *evlist __maybe_unused)
885 {
886 struct perf_session *session;
887 int err;
888
889 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
890 return -1;
891
892 session = container_of(ff->ph, struct perf_session, header);
893
894 err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
895 if (err < 0)
896 pr_err("Failed to write auxtrace index\n");
897 return err;
898 }
899
900 static int write_clockid(struct feat_fd *ff,
901 struct evlist *evlist __maybe_unused)
902 {
903 return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
904 sizeof(ff->ph->env.clock.clockid_res_ns));
905 }
906
907 static int write_clock_data(struct feat_fd *ff,
908 struct evlist *evlist __maybe_unused)
909 {
910 u64 *data64;
911 u32 data32;
912 int ret;
913
914 /* version */
915 data32 = 1;
916
917 ret = do_write(ff, &data32, sizeof(data32));
918 if (ret < 0)
919 return ret;
920
921 /* clockid */
922 data32 = ff->ph->env.clock.clockid;
923
924 ret = do_write(ff, &data32, sizeof(data32));
925 if (ret < 0)
926 return ret;
927
928 /* TOD ref time */
929 data64 = &ff->ph->env.clock.tod_ns;
930
931 ret = do_write(ff, data64, sizeof(*data64));
932 if (ret < 0)
933 return ret;
934
935 /* clockid ref time */
936 data64 = &ff->ph->env.clock.clockid_ns;
937
938 return do_write(ff, data64, sizeof(*data64));
939 }
940
941 static int write_hybrid_topology(struct feat_fd *ff,
942 struct evlist *evlist __maybe_unused)
943 {
944 struct hybrid_topology *tp;
945 int ret;
946 u32 i;
947
948 tp = hybrid_topology__new();
949 if (!tp)
950 return -ENOENT;
951
952 ret = do_write(ff, &tp->nr, sizeof(u32));
953 if (ret < 0)
954 goto err;
955
956 for (i = 0; i < tp->nr; i++) {
957 struct hybrid_topology_node *n = &tp->nodes[i];
958
959 ret = do_write_string(ff, n->pmu_name);
960 if (ret < 0)
961 goto err;
962
963 ret = do_write_string(ff, n->cpus);
964 if (ret < 0)
965 goto err;
966 }
967
968 ret = 0;
969
970 err:
971 hybrid_topology__delete(tp);
972 return ret;
973 }
974
975 static int write_dir_format(struct feat_fd *ff,
976 struct evlist *evlist __maybe_unused)
977 {
978 struct perf_session *session;
979 struct perf_data *data;
980
981 session = container_of(ff->ph, struct perf_session, header);
982 data = session->data;
983
984 if (WARN_ON(!perf_data__is_dir(data)))
985 return -1;
986
987 return do_write(ff, &data->dir.version, sizeof(data->dir.version));
988 }
989
990 /*
991 * Check whether a CPU is online
992 *
993 * Returns:
994 * 1 -> if CPU is online
995 * 0 -> if CPU is offline
996 * -1 -> error case
997 */
998 int is_cpu_online(unsigned int cpu)
999 {
1000 char *str;
1001 size_t strlen;
1002 char buf[256];
1003 int status = -1;
1004 struct stat statbuf;
1005
1006 snprintf(buf, sizeof(buf),
1007 "/sys/devices/system/cpu/cpu%d", cpu);
1008 if (stat(buf, &statbuf) != 0)
1009 return 0;
1010
1011 /*
1012 * Check if /sys/devices/system/cpu/cpux/online file
1013 * exists. Some cases cpu0 won't have online file since
1014 * it is not expected to be turned off generally.
1015 * In kernels without CONFIG_HOTPLUG_CPU, this
1016 * file won't exist
1017 */
1018 snprintf(buf, sizeof(buf),
1019 "/sys/devices/system/cpu/cpu%d/online", cpu);
1020 if (stat(buf, &statbuf) != 0)
1021 return 1;
1022
1023 /*
1024 * Read online file using sysfs__read_str.
1025 * If read or open fails, return -1.
1026 * If read succeeds, return value from file
1027 * which gets stored in "str"
1028 */
1029 snprintf(buf, sizeof(buf),
1030 "devices/system/cpu/cpu%d/online", cpu);
1031
1032 if (sysfs__read_str(buf, &str, &strlen) < 0)
1033 return status;
1034
1035 status = atoi(str);
1036
1037 free(str);
1038 return status;
1039 }
1040
1041 #ifdef HAVE_LIBBPF_SUPPORT
1042 static int write_bpf_prog_info(struct feat_fd *ff,
1043 struct evlist *evlist __maybe_unused)
1044 {
1045 struct perf_env *env = &ff->ph->env;
1046 struct rb_root *root;
1047 struct rb_node *next;
1048 int ret;
1049
1050 down_read(&env->bpf_progs.lock);
1051
1052 ret = do_write(ff, &env->bpf_progs.infos_cnt,
1053 sizeof(env->bpf_progs.infos_cnt));
1054 if (ret < 0)
1055 goto out;
1056
1057 root = &env->bpf_progs.infos;
1058 next = rb_first(root);
1059 while (next) {
1060 struct bpf_prog_info_node *node;
1061 size_t len;
1062
1063 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1064 next = rb_next(&node->rb_node);
1065 len = sizeof(struct perf_bpil) +
1066 node->info_linear->data_len;
1067
1068 /* before writing to file, translate address to offset */
1069 bpil_addr_to_offs(node->info_linear);
1070 ret = do_write(ff, node->info_linear, len);
1071 /*
1072 * translate back to address even when do_write() fails,
1073 * so that this function never changes the data.
1074 */
1075 bpil_offs_to_addr(node->info_linear);
1076 if (ret < 0)
1077 goto out;
1078 }
1079 out:
1080 up_read(&env->bpf_progs.lock);
1081 return ret;
1082 }
1083
1084 static int write_bpf_btf(struct feat_fd *ff,
1085 struct evlist *evlist __maybe_unused)
1086 {
1087 struct perf_env *env = &ff->ph->env;
1088 struct rb_root *root;
1089 struct rb_node *next;
1090 int ret;
1091
1092 down_read(&env->bpf_progs.lock);
1093
1094 ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1095 sizeof(env->bpf_progs.btfs_cnt));
1096
1097 if (ret < 0)
1098 goto out;
1099
1100 root = &env->bpf_progs.btfs;
1101 next = rb_first(root);
1102 while (next) {
1103 struct btf_node *node;
1104
1105 node = rb_entry(next, struct btf_node, rb_node);
1106 next = rb_next(&node->rb_node);
1107 ret = do_write(ff, &node->id,
1108 sizeof(u32) * 2 + node->data_size);
1109 if (ret < 0)
1110 goto out;
1111 }
1112 out:
1113 up_read(&env->bpf_progs.lock);
1114 return ret;
1115 }
1116 #endif // HAVE_LIBBPF_SUPPORT
1117
1118 static int cpu_cache_level__sort(const void *a, const void *b)
1119 {
1120 struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1121 struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1122
1123 return cache_a->level - cache_b->level;
1124 }
1125
1126 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1127 {
1128 if (a->level != b->level)
1129 return false;
1130
1131 if (a->line_size != b->line_size)
1132 return false;
1133
1134 if (a->sets != b->sets)
1135 return false;
1136
1137 if (a->ways != b->ways)
1138 return false;
1139
1140 if (strcmp(a->type, b->type))
1141 return false;
1142
1143 if (strcmp(a->size, b->size))
1144 return false;
1145
1146 if (strcmp(a->map, b->map))
1147 return false;
1148
1149 return true;
1150 }
1151
1152 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1153 {
1154 char path[PATH_MAX], file[PATH_MAX];
1155 struct stat st;
1156 size_t len;
1157
1158 scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1159 scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1160
1161 if (stat(file, &st))
1162 return 1;
1163
1164 scnprintf(file, PATH_MAX, "%s/level", path);
1165 if (sysfs__read_int(file, (int *) &cache->level))
1166 return -1;
1167
1168 scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1169 if (sysfs__read_int(file, (int *) &cache->line_size))
1170 return -1;
1171
1172 scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1173 if (sysfs__read_int(file, (int *) &cache->sets))
1174 return -1;
1175
1176 scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1177 if (sysfs__read_int(file, (int *) &cache->ways))
1178 return -1;
1179
1180 scnprintf(file, PATH_MAX, "%s/type", path);
1181 if (sysfs__read_str(file, &cache->type, &len))
1182 return -1;
1183
1184 cache->type[len] = 0;
1185 cache->type = strim(cache->type);
1186
1187 scnprintf(file, PATH_MAX, "%s/size", path);
1188 if (sysfs__read_str(file, &cache->size, &len)) {
1189 zfree(&cache->type);
1190 return -1;
1191 }
1192
1193 cache->size[len] = 0;
1194 cache->size = strim(cache->size);
1195
1196 scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1197 if (sysfs__read_str(file, &cache->map, &len)) {
1198 zfree(&cache->size);
1199 zfree(&cache->type);
1200 return -1;
1201 }
1202
1203 cache->map[len] = 0;
1204 cache->map = strim(cache->map);
1205 return 0;
1206 }
1207
1208 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1209 {
1210 fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1211 }
1212
1213 /*
1214 * Build caches levels for a particular CPU from the data in
1215 * /sys/devices/system/cpu/cpu<cpu>/cache/
1216 * The cache level data is stored in caches[] from index at
1217 * *cntp.
1218 */
1219 int build_caches_for_cpu(u32 cpu, struct cpu_cache_level caches[], u32 *cntp)
1220 {
1221 u16 level;
1222
1223 for (level = 0; level < MAX_CACHE_LVL; level++) {
1224 struct cpu_cache_level c;
1225 int err;
1226 u32 i;
1227
1228 err = cpu_cache_level__read(&c, cpu, level);
1229 if (err < 0)
1230 return err;
1231
1232 if (err == 1)
1233 break;
1234
1235 for (i = 0; i < *cntp; i++) {
1236 if (cpu_cache_level__cmp(&c, &caches[i]))
1237 break;
1238 }
1239
1240 if (i == *cntp) {
1241 caches[*cntp] = c;
1242 *cntp = *cntp + 1;
1243 } else
1244 cpu_cache_level__free(&c);
1245 }
1246
1247 return 0;
1248 }
1249
1250 static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1251 {
1252 u32 nr, cpu, cnt = 0;
1253
1254 nr = cpu__max_cpu().cpu;
1255
1256 for (cpu = 0; cpu < nr; cpu++) {
1257 int ret = build_caches_for_cpu(cpu, caches, &cnt);
1258
1259 if (ret)
1260 return ret;
1261 }
1262 *cntp = cnt;
1263 return 0;
1264 }
1265
1266 static int write_cache(struct feat_fd *ff,
1267 struct evlist *evlist __maybe_unused)
1268 {
1269 u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL;
1270 struct cpu_cache_level caches[max_caches];
1271 u32 cnt = 0, i, version = 1;
1272 int ret;
1273
1274 ret = build_caches(caches, &cnt);
1275 if (ret)
1276 goto out;
1277
1278 qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1279
1280 ret = do_write(ff, &version, sizeof(u32));
1281 if (ret < 0)
1282 goto out;
1283
1284 ret = do_write(ff, &cnt, sizeof(u32));
1285 if (ret < 0)
1286 goto out;
1287
1288 for (i = 0; i < cnt; i++) {
1289 struct cpu_cache_level *c = &caches[i];
1290
1291 #define _W(v) \
1292 ret = do_write(ff, &c->v, sizeof(u32)); \
1293 if (ret < 0) \
1294 goto out;
1295
1296 _W(level)
1297 _W(line_size)
1298 _W(sets)
1299 _W(ways)
1300 #undef _W
1301
1302 #define _W(v) \
1303 ret = do_write_string(ff, (const char *) c->v); \
1304 if (ret < 0) \
1305 goto out;
1306
1307 _W(type)
1308 _W(size)
1309 _W(map)
1310 #undef _W
1311 }
1312
1313 out:
1314 for (i = 0; i < cnt; i++)
1315 cpu_cache_level__free(&caches[i]);
1316 return ret;
1317 }
1318
1319 static int write_stat(struct feat_fd *ff __maybe_unused,
1320 struct evlist *evlist __maybe_unused)
1321 {
1322 return 0;
1323 }
1324
1325 static int write_sample_time(struct feat_fd *ff,
1326 struct evlist *evlist)
1327 {
1328 int ret;
1329
1330 ret = do_write(ff, &evlist->first_sample_time,
1331 sizeof(evlist->first_sample_time));
1332 if (ret < 0)
1333 return ret;
1334
1335 return do_write(ff, &evlist->last_sample_time,
1336 sizeof(evlist->last_sample_time));
1337 }
1338
1339
1340 static int memory_node__read(struct memory_node *n, unsigned long idx)
1341 {
1342 unsigned int phys, size = 0;
1343 char path[PATH_MAX];
1344 struct dirent *ent;
1345 DIR *dir;
1346
1347 #define for_each_memory(mem, dir) \
1348 while ((ent = readdir(dir))) \
1349 if (strcmp(ent->d_name, ".") && \
1350 strcmp(ent->d_name, "..") && \
1351 sscanf(ent->d_name, "memory%u", &mem) == 1)
1352
1353 scnprintf(path, PATH_MAX,
1354 "%s/devices/system/node/node%lu",
1355 sysfs__mountpoint(), idx);
1356
1357 dir = opendir(path);
1358 if (!dir) {
1359 pr_warning("failed: can't open memory sysfs data\n");
1360 return -1;
1361 }
1362
1363 for_each_memory(phys, dir) {
1364 size = max(phys, size);
1365 }
1366
1367 size++;
1368
1369 n->set = bitmap_zalloc(size);
1370 if (!n->set) {
1371 closedir(dir);
1372 return -ENOMEM;
1373 }
1374
1375 n->node = idx;
1376 n->size = size;
1377
1378 rewinddir(dir);
1379
1380 for_each_memory(phys, dir) {
1381 __set_bit(phys, n->set);
1382 }
1383
1384 closedir(dir);
1385 return 0;
1386 }
1387
1388 static void memory_node__delete_nodes(struct memory_node *nodesp, u64 cnt)
1389 {
1390 for (u64 i = 0; i < cnt; i++)
1391 bitmap_free(nodesp[i].set);
1392
1393 free(nodesp);
1394 }
1395
1396 static int memory_node__sort(const void *a, const void *b)
1397 {
1398 const struct memory_node *na = a;
1399 const struct memory_node *nb = b;
1400
1401 return na->node - nb->node;
1402 }
1403
1404 static int build_mem_topology(struct memory_node **nodesp, u64 *cntp)
1405 {
1406 char path[PATH_MAX];
1407 struct dirent *ent;
1408 DIR *dir;
1409 int ret = 0;
1410 size_t cnt = 0, size = 0;
1411 struct memory_node *nodes = NULL;
1412
1413 scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1414 sysfs__mountpoint());
1415
1416 dir = opendir(path);
1417 if (!dir) {
1418 pr_debug2("%s: couldn't read %s, does this arch have topology information?\n",
1419 __func__, path);
1420 return -1;
1421 }
1422
1423 while (!ret && (ent = readdir(dir))) {
1424 unsigned int idx;
1425 int r;
1426
1427 if (!strcmp(ent->d_name, ".") ||
1428 !strcmp(ent->d_name, ".."))
1429 continue;
1430
1431 r = sscanf(ent->d_name, "node%u", &idx);
1432 if (r != 1)
1433 continue;
1434
1435 if (cnt >= size) {
1436 struct memory_node *new_nodes =
1437 reallocarray(nodes, cnt + 4, sizeof(*nodes));
1438
1439 if (!new_nodes) {
1440 pr_err("Failed to write MEM_TOPOLOGY, size %zd nodes\n", size);
1441 ret = -ENOMEM;
1442 goto out;
1443 }
1444 nodes = new_nodes;
1445 size += 4;
1446 }
1447 ret = memory_node__read(&nodes[cnt], idx);
1448 if (!ret)
1449 cnt += 1;
1450 }
1451 out:
1452 closedir(dir);
1453 if (!ret) {
1454 *cntp = cnt;
1455 *nodesp = nodes;
1456 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1457 } else
1458 memory_node__delete_nodes(nodes, cnt);
1459
1460 return ret;
1461 }
1462
1463 /*
1464 * The MEM_TOPOLOGY holds physical memory map for every
1465 * node in system. The format of data is as follows:
1466 *
1467 * 0 - version | for future changes
1468 * 8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1469 * 16 - count | number of nodes
1470 *
1471 * For each node we store map of physical indexes for
1472 * each node:
1473 *
1474 * 32 - node id | node index
1475 * 40 - size | size of bitmap
1476 * 48 - bitmap | bitmap of memory indexes that belongs to node
1477 */
1478 static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1479 struct evlist *evlist __maybe_unused)
1480 {
1481 struct memory_node *nodes = NULL;
1482 u64 bsize, version = 1, i, nr = 0;
1483 int ret;
1484
1485 ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1486 (unsigned long long *) &bsize);
1487 if (ret)
1488 return ret;
1489
1490 ret = build_mem_topology(&nodes, &nr);
1491 if (ret)
1492 return ret;
1493
1494 ret = do_write(ff, &version, sizeof(version));
1495 if (ret < 0)
1496 goto out;
1497
1498 ret = do_write(ff, &bsize, sizeof(bsize));
1499 if (ret < 0)
1500 goto out;
1501
1502 ret = do_write(ff, &nr, sizeof(nr));
1503 if (ret < 0)
1504 goto out;
1505
1506 for (i = 0; i < nr; i++) {
1507 struct memory_node *n = &nodes[i];
1508
1509 #define _W(v) \
1510 ret = do_write(ff, &n->v, sizeof(n->v)); \
1511 if (ret < 0) \
1512 goto out;
1513
1514 _W(node)
1515 _W(size)
1516
1517 #undef _W
1518
1519 ret = do_write_bitmap(ff, n->set, n->size);
1520 if (ret < 0)
1521 goto out;
1522 }
1523
1524 out:
1525 memory_node__delete_nodes(nodes, nr);
1526 return ret;
1527 }
1528
1529 static int write_compressed(struct feat_fd *ff __maybe_unused,
1530 struct evlist *evlist __maybe_unused)
1531 {
1532 int ret;
1533
1534 ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1535 if (ret)
1536 return ret;
1537
1538 ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1539 if (ret)
1540 return ret;
1541
1542 ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1543 if (ret)
1544 return ret;
1545
1546 ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1547 if (ret)
1548 return ret;
1549
1550 return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1551 }
1552
1553 static int __write_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1554 bool write_pmu)
1555 {
1556 struct perf_pmu_caps *caps = NULL;
1557 int ret;
1558
1559 ret = do_write(ff, &pmu->nr_caps, sizeof(pmu->nr_caps));
1560 if (ret < 0)
1561 return ret;
1562
1563 list_for_each_entry(caps, &pmu->caps, list) {
1564 ret = do_write_string(ff, caps->name);
1565 if (ret < 0)
1566 return ret;
1567
1568 ret = do_write_string(ff, caps->value);
1569 if (ret < 0)
1570 return ret;
1571 }
1572
1573 if (write_pmu) {
1574 ret = do_write_string(ff, pmu->name);
1575 if (ret < 0)
1576 return ret;
1577 }
1578
1579 return ret;
1580 }
1581
1582 static int write_cpu_pmu_caps(struct feat_fd *ff,
1583 struct evlist *evlist __maybe_unused)
1584 {
1585 struct perf_pmu *cpu_pmu = perf_pmus__find("cpu");
1586 int ret;
1587
1588 if (!cpu_pmu)
1589 return -ENOENT;
1590
1591 ret = perf_pmu__caps_parse(cpu_pmu);
1592 if (ret < 0)
1593 return ret;
1594
1595 return __write_pmu_caps(ff, cpu_pmu, false);
1596 }
1597
1598 static int write_pmu_caps(struct feat_fd *ff,
1599 struct evlist *evlist __maybe_unused)
1600 {
1601 struct perf_pmu *pmu = NULL;
1602 int nr_pmu = 0;
1603 int ret;
1604
1605 while ((pmu = perf_pmus__scan(pmu))) {
1606 if (!strcmp(pmu->name, "cpu")) {
1607 /*
1608 * The "cpu" PMU is special and covered by
1609 * HEADER_CPU_PMU_CAPS. Note, core PMUs are
1610 * counted/written here for ARM, s390 and Intel hybrid.
1611 */
1612 continue;
1613 }
1614 if (perf_pmu__caps_parse(pmu) <= 0)
1615 continue;
1616 nr_pmu++;
1617 }
1618
1619 ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1620 if (ret < 0)
1621 return ret;
1622
1623 if (!nr_pmu)
1624 return 0;
1625
1626 /*
1627 * Note older perf tools assume core PMUs come first, this is a property
1628 * of perf_pmus__scan.
1629 */
1630 pmu = NULL;
1631 while ((pmu = perf_pmus__scan(pmu))) {
1632 if (!strcmp(pmu->name, "cpu")) {
1633 /* Skip as above. */
1634 continue;
1635 }
1636 if (perf_pmu__caps_parse(pmu) <= 0)
1637 continue;
1638 ret = __write_pmu_caps(ff, pmu, true);
1639 if (ret < 0)
1640 return ret;
1641 }
1642 return 0;
1643 }
1644
1645 static void print_hostname(struct feat_fd *ff, FILE *fp)
1646 {
1647 fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1648 }
1649
1650 static void print_osrelease(struct feat_fd *ff, FILE *fp)
1651 {
1652 fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1653 }
1654
1655 static void print_arch(struct feat_fd *ff, FILE *fp)
1656 {
1657 fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1658 }
1659
1660 static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1661 {
1662 fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1663 }
1664
1665 static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1666 {
1667 fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1668 fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1669 }
1670
1671 static void print_version(struct feat_fd *ff, FILE *fp)
1672 {
1673 fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1674 }
1675
1676 static void print_cmdline(struct feat_fd *ff, FILE *fp)
1677 {
1678 int nr, i;
1679
1680 nr = ff->ph->env.nr_cmdline;
1681
1682 fprintf(fp, "# cmdline : ");
1683
1684 for (i = 0; i < nr; i++) {
1685 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1686 if (!argv_i) {
1687 fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1688 } else {
1689 char *mem = argv_i;
1690 do {
1691 char *quote = strchr(argv_i, '\'');
1692 if (!quote)
1693 break;
1694 *quote++ = '\0';
1695 fprintf(fp, "%s\\\'", argv_i);
1696 argv_i = quote;
1697 } while (1);
1698 fprintf(fp, "%s ", argv_i);
1699 free(mem);
1700 }
1701 }
1702 fputc('\n', fp);
1703 }
1704
1705 static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1706 {
1707 struct perf_header *ph = ff->ph;
1708 int cpu_nr = ph->env.nr_cpus_avail;
1709 int nr, i;
1710 char *str;
1711
1712 nr = ph->env.nr_sibling_cores;
1713 str = ph->env.sibling_cores;
1714
1715 for (i = 0; i < nr; i++) {
1716 fprintf(fp, "# sibling sockets : %s\n", str);
1717 str += strlen(str) + 1;
1718 }
1719
1720 if (ph->env.nr_sibling_dies) {
1721 nr = ph->env.nr_sibling_dies;
1722 str = ph->env.sibling_dies;
1723
1724 for (i = 0; i < nr; i++) {
1725 fprintf(fp, "# sibling dies : %s\n", str);
1726 str += strlen(str) + 1;
1727 }
1728 }
1729
1730 nr = ph->env.nr_sibling_threads;
1731 str = ph->env.sibling_threads;
1732
1733 for (i = 0; i < nr; i++) {
1734 fprintf(fp, "# sibling threads : %s\n", str);
1735 str += strlen(str) + 1;
1736 }
1737
1738 if (ph->env.nr_sibling_dies) {
1739 if (ph->env.cpu != NULL) {
1740 for (i = 0; i < cpu_nr; i++)
1741 fprintf(fp, "# CPU %d: Core ID %d, "
1742 "Die ID %d, Socket ID %d\n",
1743 i, ph->env.cpu[i].core_id,
1744 ph->env.cpu[i].die_id,
1745 ph->env.cpu[i].socket_id);
1746 } else
1747 fprintf(fp, "# Core ID, Die ID and Socket ID "
1748 "information is not available\n");
1749 } else {
1750 if (ph->env.cpu != NULL) {
1751 for (i = 0; i < cpu_nr; i++)
1752 fprintf(fp, "# CPU %d: Core ID %d, "
1753 "Socket ID %d\n",
1754 i, ph->env.cpu[i].core_id,
1755 ph->env.cpu[i].socket_id);
1756 } else
1757 fprintf(fp, "# Core ID and Socket ID "
1758 "information is not available\n");
1759 }
1760 }
1761
1762 static void print_clockid(struct feat_fd *ff, FILE *fp)
1763 {
1764 fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1765 ff->ph->env.clock.clockid_res_ns * 1000);
1766 }
1767
1768 static void print_clock_data(struct feat_fd *ff, FILE *fp)
1769 {
1770 struct timespec clockid_ns;
1771 char tstr[64], date[64];
1772 struct timeval tod_ns;
1773 clockid_t clockid;
1774 struct tm ltime;
1775 u64 ref;
1776
1777 if (!ff->ph->env.clock.enabled) {
1778 fprintf(fp, "# reference time disabled\n");
1779 return;
1780 }
1781
1782 /* Compute TOD time. */
1783 ref = ff->ph->env.clock.tod_ns;
1784 tod_ns.tv_sec = ref / NSEC_PER_SEC;
1785 ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1786 tod_ns.tv_usec = ref / NSEC_PER_USEC;
1787
1788 /* Compute clockid time. */
1789 ref = ff->ph->env.clock.clockid_ns;
1790 clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1791 ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1792 clockid_ns.tv_nsec = ref;
1793
1794 clockid = ff->ph->env.clock.clockid;
1795
1796 if (localtime_r(&tod_ns.tv_sec, <ime) == NULL)
1797 snprintf(tstr, sizeof(tstr), "<error>");
1798 else {
1799 strftime(date, sizeof(date), "%F %T", <ime);
1800 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1801 date, (int) tod_ns.tv_usec);
1802 }
1803
1804 fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1805 fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1806 tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1807 (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1808 clockid_name(clockid));
1809 }
1810
1811 static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1812 {
1813 int i;
1814 struct hybrid_node *n;
1815
1816 fprintf(fp, "# hybrid cpu system:\n");
1817 for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1818 n = &ff->ph->env.hybrid_nodes[i];
1819 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1820 }
1821 }
1822
1823 static void print_dir_format(struct feat_fd *ff, FILE *fp)
1824 {
1825 struct perf_session *session;
1826 struct perf_data *data;
1827
1828 session = container_of(ff->ph, struct perf_session, header);
1829 data = session->data;
1830
1831 fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1832 }
1833
1834 #ifdef HAVE_LIBBPF_SUPPORT
1835 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1836 {
1837 struct perf_env *env = &ff->ph->env;
1838 struct rb_root *root;
1839 struct rb_node *next;
1840
1841 down_read(&env->bpf_progs.lock);
1842
1843 root = &env->bpf_progs.infos;
1844 next = rb_first(root);
1845
1846 while (next) {
1847 struct bpf_prog_info_node *node;
1848
1849 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1850 next = rb_next(&node->rb_node);
1851
1852 __bpf_event__print_bpf_prog_info(&node->info_linear->info,
1853 env, fp);
1854 }
1855
1856 up_read(&env->bpf_progs.lock);
1857 }
1858
1859 static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1860 {
1861 struct perf_env *env = &ff->ph->env;
1862 struct rb_root *root;
1863 struct rb_node *next;
1864
1865 down_read(&env->bpf_progs.lock);
1866
1867 root = &env->bpf_progs.btfs;
1868 next = rb_first(root);
1869
1870 while (next) {
1871 struct btf_node *node;
1872
1873 node = rb_entry(next, struct btf_node, rb_node);
1874 next = rb_next(&node->rb_node);
1875 fprintf(fp, "# btf info of id %u\n", node->id);
1876 }
1877
1878 up_read(&env->bpf_progs.lock);
1879 }
1880 #endif // HAVE_LIBBPF_SUPPORT
1881
1882 static void free_event_desc(struct evsel *events)
1883 {
1884 struct evsel *evsel;
1885
1886 if (!events)
1887 return;
1888
1889 for (evsel = events; evsel->core.attr.size; evsel++) {
1890 zfree(&evsel->name);
1891 zfree(&evsel->core.id);
1892 }
1893
1894 free(events);
1895 }
1896
1897 static bool perf_attr_check(struct perf_event_attr *attr)
1898 {
1899 if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1900 pr_warning("Reserved bits are set unexpectedly. "
1901 "Please update perf tool.\n");
1902 return false;
1903 }
1904
1905 if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1906 pr_warning("Unknown sample type (0x%llx) is detected. "
1907 "Please update perf tool.\n",
1908 attr->sample_type);
1909 return false;
1910 }
1911
1912 if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1913 pr_warning("Unknown read format (0x%llx) is detected. "
1914 "Please update perf tool.\n",
1915 attr->read_format);
1916 return false;
1917 }
1918
1919 if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1920 (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1921 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1922 "Please update perf tool.\n",
1923 attr->branch_sample_type);
1924
1925 return false;
1926 }
1927
1928 return true;
1929 }
1930
1931 static struct evsel *read_event_desc(struct feat_fd *ff)
1932 {
1933 struct evsel *evsel, *events = NULL;
1934 u64 *id;
1935 void *buf = NULL;
1936 u32 nre, sz, nr, i, j;
1937 size_t msz;
1938
1939 /* number of events */
1940 if (do_read_u32(ff, &nre))
1941 goto error;
1942
1943 if (do_read_u32(ff, &sz))
1944 goto error;
1945
1946 /* buffer to hold on file attr struct */
1947 buf = malloc(sz);
1948 if (!buf)
1949 goto error;
1950
1951 /* the last event terminates with evsel->core.attr.size == 0: */
1952 events = calloc(nre + 1, sizeof(*events));
1953 if (!events)
1954 goto error;
1955
1956 msz = sizeof(evsel->core.attr);
1957 if (sz < msz)
1958 msz = sz;
1959
1960 for (i = 0, evsel = events; i < nre; evsel++, i++) {
1961 evsel->core.idx = i;
1962
1963 /*
1964 * must read entire on-file attr struct to
1965 * sync up with layout.
1966 */
1967 if (__do_read(ff, buf, sz))
1968 goto error;
1969
1970 if (ff->ph->needs_swap)
1971 perf_event__attr_swap(buf);
1972
1973 memcpy(&evsel->core.attr, buf, msz);
1974
1975 if (!perf_attr_check(&evsel->core.attr))
1976 goto error;
1977
1978 if (do_read_u32(ff, &nr))
1979 goto error;
1980
1981 if (ff->ph->needs_swap)
1982 evsel->needs_swap = true;
1983
1984 evsel->name = do_read_string(ff);
1985 if (!evsel->name)
1986 goto error;
1987
1988 if (!nr)
1989 continue;
1990
1991 id = calloc(nr, sizeof(*id));
1992 if (!id)
1993 goto error;
1994 evsel->core.ids = nr;
1995 evsel->core.id = id;
1996
1997 for (j = 0 ; j < nr; j++) {
1998 if (do_read_u64(ff, id))
1999 goto error;
2000 id++;
2001 }
2002 }
2003 out:
2004 free(buf);
2005 return events;
2006 error:
2007 free_event_desc(events);
2008 events = NULL;
2009 goto out;
2010 }
2011
2012 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
2013 void *priv __maybe_unused)
2014 {
2015 return fprintf(fp, ", %s = %s", name, val);
2016 }
2017
2018 static void print_event_desc(struct feat_fd *ff, FILE *fp)
2019 {
2020 struct evsel *evsel, *events;
2021 u32 j;
2022 u64 *id;
2023
2024 if (ff->events)
2025 events = ff->events;
2026 else
2027 events = read_event_desc(ff);
2028
2029 if (!events) {
2030 fprintf(fp, "# event desc: not available or unable to read\n");
2031 return;
2032 }
2033
2034 for (evsel = events; evsel->core.attr.size; evsel++) {
2035 fprintf(fp, "# event : name = %s, ", evsel->name);
2036
2037 if (evsel->core.ids) {
2038 fprintf(fp, ", id = {");
2039 for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
2040 if (j)
2041 fputc(',', fp);
2042 fprintf(fp, " %"PRIu64, *id);
2043 }
2044 fprintf(fp, " }");
2045 }
2046
2047 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
2048
2049 fputc('\n', fp);
2050 }
2051
2052 free_event_desc(events);
2053 ff->events = NULL;
2054 }
2055
2056 static void print_total_mem(struct feat_fd *ff, FILE *fp)
2057 {
2058 fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
2059 }
2060
2061 static void print_numa_topology(struct feat_fd *ff, FILE *fp)
2062 {
2063 int i;
2064 struct numa_node *n;
2065
2066 for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
2067 n = &ff->ph->env.numa_nodes[i];
2068
2069 fprintf(fp, "# node%u meminfo : total = %"PRIu64" kB,"
2070 " free = %"PRIu64" kB\n",
2071 n->node, n->mem_total, n->mem_free);
2072
2073 fprintf(fp, "# node%u cpu list : ", n->node);
2074 cpu_map__fprintf(n->map, fp);
2075 }
2076 }
2077
2078 static void print_cpuid(struct feat_fd *ff, FILE *fp)
2079 {
2080 fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
2081 }
2082
2083 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
2084 {
2085 fprintf(fp, "# contains samples with branch stack\n");
2086 }
2087
2088 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
2089 {
2090 fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
2091 }
2092
2093 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
2094 {
2095 fprintf(fp, "# contains stat data\n");
2096 }
2097
2098 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
2099 {
2100 int i;
2101
2102 fprintf(fp, "# CPU cache info:\n");
2103 for (i = 0; i < ff->ph->env.caches_cnt; i++) {
2104 fprintf(fp, "# ");
2105 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
2106 }
2107 }
2108
2109 static void print_compressed(struct feat_fd *ff, FILE *fp)
2110 {
2111 fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
2112 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
2113 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2114 }
2115
2116 static void __print_pmu_caps(FILE *fp, int nr_caps, char **caps, char *pmu_name)
2117 {
2118 const char *delimiter = "";
2119 int i;
2120
2121 if (!nr_caps) {
2122 fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2123 return;
2124 }
2125
2126 fprintf(fp, "# %s pmu capabilities: ", pmu_name);
2127 for (i = 0; i < nr_caps; i++) {
2128 fprintf(fp, "%s%s", delimiter, caps[i]);
2129 delimiter = ", ";
2130 }
2131
2132 fprintf(fp, "\n");
2133 }
2134
2135 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2136 {
2137 __print_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2138 ff->ph->env.cpu_pmu_caps, (char *)"cpu");
2139 }
2140
2141 static void print_pmu_caps(struct feat_fd *ff, FILE *fp)
2142 {
2143 struct pmu_caps *pmu_caps;
2144
2145 for (int i = 0; i < ff->ph->env.nr_pmus_with_caps; i++) {
2146 pmu_caps = &ff->ph->env.pmu_caps[i];
2147 __print_pmu_caps(fp, pmu_caps->nr_caps, pmu_caps->caps,
2148 pmu_caps->pmu_name);
2149 }
2150
2151 if (strcmp(perf_env__arch(&ff->ph->env), "x86") == 0 &&
2152 perf_env__has_pmu_mapping(&ff->ph->env, "ibs_op")) {
2153 char *max_precise = perf_env__find_pmu_cap(&ff->ph->env, "cpu", "max_precise");
2154
2155 if (max_precise != NULL && atoi(max_precise) == 0)
2156 fprintf(fp, "# AMD systems uses ibs_op// PMU for some precise events, e.g.: cycles:p, see the 'perf list' man page for further details.\n");
2157 }
2158 }
2159
2160 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2161 {
2162 const char *delimiter = "# pmu mappings: ";
2163 char *str, *tmp;
2164 u32 pmu_num;
2165 u32 type;
2166
2167 pmu_num = ff->ph->env.nr_pmu_mappings;
2168 if (!pmu_num) {
2169 fprintf(fp, "# pmu mappings: not available\n");
2170 return;
2171 }
2172
2173 str = ff->ph->env.pmu_mappings;
2174
2175 while (pmu_num) {
2176 type = strtoul(str, &tmp, 0);
2177 if (*tmp != ':')
2178 goto error;
2179
2180 str = tmp + 1;
2181 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2182
2183 delimiter = ", ";
2184 str += strlen(str) + 1;
2185 pmu_num--;
2186 }
2187
2188 fprintf(fp, "\n");
2189
2190 if (!pmu_num)
2191 return;
2192 error:
2193 fprintf(fp, "# pmu mappings: unable to read\n");
2194 }
2195
2196 static void print_group_desc(struct feat_fd *ff, FILE *fp)
2197 {
2198 struct perf_session *session;
2199 struct evsel *evsel;
2200 u32 nr = 0;
2201
2202 session = container_of(ff->ph, struct perf_session, header);
2203
2204 evlist__for_each_entry(session->evlist, evsel) {
2205 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2206 fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2207
2208 nr = evsel->core.nr_members - 1;
2209 } else if (nr) {
2210 fprintf(fp, ",%s", evsel__name(evsel));
2211
2212 if (--nr == 0)
2213 fprintf(fp, "}\n");
2214 }
2215 }
2216 }
2217
2218 static void print_sample_time(struct feat_fd *ff, FILE *fp)
2219 {
2220 struct perf_session *session;
2221 char time_buf[32];
2222 double d;
2223
2224 session = container_of(ff->ph, struct perf_session, header);
2225
2226 timestamp__scnprintf_usec(session->evlist->first_sample_time,
2227 time_buf, sizeof(time_buf));
2228 fprintf(fp, "# time of first sample : %s\n", time_buf);
2229
2230 timestamp__scnprintf_usec(session->evlist->last_sample_time,
2231 time_buf, sizeof(time_buf));
2232 fprintf(fp, "# time of last sample : %s\n", time_buf);
2233
2234 d = (double)(session->evlist->last_sample_time -
2235 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2236
2237 fprintf(fp, "# sample duration : %10.3f ms\n", d);
2238 }
2239
2240 static void memory_node__fprintf(struct memory_node *n,
2241 unsigned long long bsize, FILE *fp)
2242 {
2243 char buf_map[100], buf_size[50];
2244 unsigned long long size;
2245
2246 size = bsize * bitmap_weight(n->set, n->size);
2247 unit_number__scnprintf(buf_size, 50, size);
2248
2249 bitmap_scnprintf(n->set, n->size, buf_map, 100);
2250 fprintf(fp, "# %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2251 }
2252
2253 static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2254 {
2255 struct memory_node *nodes;
2256 int i, nr;
2257
2258 nodes = ff->ph->env.memory_nodes;
2259 nr = ff->ph->env.nr_memory_nodes;
2260
2261 fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2262 nr, ff->ph->env.memory_bsize);
2263
2264 for (i = 0; i < nr; i++) {
2265 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2266 }
2267 }
2268
2269 static int __event_process_build_id(struct perf_record_header_build_id *bev,
2270 char *filename,
2271 struct perf_session *session)
2272 {
2273 int err = -1;
2274 struct machine *machine;
2275 u16 cpumode;
2276 struct dso *dso;
2277 enum dso_space_type dso_space;
2278
2279 machine = perf_session__findnew_machine(session, bev->pid);
2280 if (!machine)
2281 goto out;
2282
2283 cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2284
2285 switch (cpumode) {
2286 case PERF_RECORD_MISC_KERNEL:
2287 dso_space = DSO_SPACE__KERNEL;
2288 break;
2289 case PERF_RECORD_MISC_GUEST_KERNEL:
2290 dso_space = DSO_SPACE__KERNEL_GUEST;
2291 break;
2292 case PERF_RECORD_MISC_USER:
2293 case PERF_RECORD_MISC_GUEST_USER:
2294 dso_space = DSO_SPACE__USER;
2295 break;
2296 default:
2297 goto out;
2298 }
2299
2300 dso = machine__findnew_dso(machine, filename);
2301 if (dso != NULL) {
2302 char sbuild_id[SBUILD_ID_SIZE];
2303 struct build_id bid;
2304 size_t size = BUILD_ID_SIZE;
2305
2306 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2307 size = bev->size;
2308
2309 build_id__init(&bid, bev->data, size);
2310 dso__set_build_id(dso, &bid);
2311 dso__set_header_build_id(dso, true);
2312
2313 if (dso_space != DSO_SPACE__USER) {
2314 struct kmod_path m = { .name = NULL, };
2315
2316 if (!kmod_path__parse_name(&m, filename) && m.kmod)
2317 dso__set_module_info(dso, &m, machine);
2318
2319 dso__set_kernel(dso, dso_space);
2320 free(m.name);
2321 }
2322
2323 build_id__sprintf(dso__bid(dso), sbuild_id);
2324 pr_debug("build id event received for %s: %s [%zu]\n",
2325 dso__long_name(dso), sbuild_id, size);
2326 dso__put(dso);
2327 }
2328
2329 err = 0;
2330 out:
2331 return err;
2332 }
2333
2334 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2335 int input, u64 offset, u64 size)
2336 {
2337 struct perf_session *session = container_of(header, struct perf_session, header);
2338 struct {
2339 struct perf_event_header header;
2340 u8 build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2341 char filename[0];
2342 } old_bev;
2343 struct perf_record_header_build_id bev;
2344 char filename[PATH_MAX];
2345 u64 limit = offset + size;
2346
2347 while (offset < limit) {
2348 ssize_t len;
2349
2350 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2351 return -1;
2352
2353 if (header->needs_swap)
2354 perf_event_header__bswap(&old_bev.header);
2355
2356 len = old_bev.header.size - sizeof(old_bev);
2357 if (readn(input, filename, len) != len)
2358 return -1;
2359
2360 bev.header = old_bev.header;
2361
2362 /*
2363 * As the pid is the missing value, we need to fill
2364 * it properly. The header.misc value give us nice hint.
2365 */
2366 bev.pid = HOST_KERNEL_ID;
2367 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2368 bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2369 bev.pid = DEFAULT_GUEST_KERNEL_ID;
2370
2371 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2372 __event_process_build_id(&bev, filename, session);
2373
2374 offset += bev.header.size;
2375 }
2376
2377 return 0;
2378 }
2379
2380 static int perf_header__read_build_ids(struct perf_header *header,
2381 int input, u64 offset, u64 size)
2382 {
2383 struct perf_session *session = container_of(header, struct perf_session, header);
2384 struct perf_record_header_build_id bev;
2385 char filename[PATH_MAX];
2386 u64 limit = offset + size, orig_offset = offset;
2387 int err = -1;
2388
2389 while (offset < limit) {
2390 ssize_t len;
2391
2392 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2393 goto out;
2394
2395 if (header->needs_swap)
2396 perf_event_header__bswap(&bev.header);
2397
2398 len = bev.header.size - sizeof(bev);
2399 if (readn(input, filename, len) != len)
2400 goto out;
2401 /*
2402 * The a1645ce1 changeset:
2403 *
2404 * "perf: 'perf kvm' tool for monitoring guest performance from host"
2405 *
2406 * Added a field to struct perf_record_header_build_id that broke the file
2407 * format.
2408 *
2409 * Since the kernel build-id is the first entry, process the
2410 * table using the old format if the well known
2411 * '[kernel.kallsyms]' string for the kernel build-id has the
2412 * first 4 characters chopped off (where the pid_t sits).
2413 */
2414 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2415 if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2416 return -1;
2417 return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2418 }
2419
2420 __event_process_build_id(&bev, filename, session);
2421
2422 offset += bev.header.size;
2423 }
2424 err = 0;
2425 out:
2426 return err;
2427 }
2428
2429 /* Macro for features that simply need to read and store a string. */
2430 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2431 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2432 {\
2433 free(ff->ph->env.__feat_env); \
2434 ff->ph->env.__feat_env = do_read_string(ff); \
2435 return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2436 }
2437
2438 FEAT_PROCESS_STR_FUN(hostname, hostname);
2439 FEAT_PROCESS_STR_FUN(osrelease, os_release);
2440 FEAT_PROCESS_STR_FUN(version, version);
2441 FEAT_PROCESS_STR_FUN(arch, arch);
2442 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2443 FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2444
2445 #ifdef HAVE_LIBTRACEEVENT
2446 static int process_tracing_data(struct feat_fd *ff, void *data)
2447 {
2448 ssize_t ret = trace_report(ff->fd, data, false);
2449
2450 return ret < 0 ? -1 : 0;
2451 }
2452 #endif
2453
2454 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2455 {
2456 if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2457 pr_debug("Failed to read buildids, continuing...\n");
2458 return 0;
2459 }
2460
2461 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2462 {
2463 int ret;
2464 u32 nr_cpus_avail, nr_cpus_online;
2465
2466 ret = do_read_u32(ff, &nr_cpus_avail);
2467 if (ret)
2468 return ret;
2469
2470 ret = do_read_u32(ff, &nr_cpus_online);
2471 if (ret)
2472 return ret;
2473 ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2474 ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2475 return 0;
2476 }
2477
2478 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2479 {
2480 u64 total_mem;
2481 int ret;
2482
2483 ret = do_read_u64(ff, &total_mem);
2484 if (ret)
2485 return -1;
2486 ff->ph->env.total_mem = (unsigned long long)total_mem;
2487 return 0;
2488 }
2489
2490 static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2491 {
2492 struct evsel *evsel;
2493
2494 evlist__for_each_entry(evlist, evsel) {
2495 if (evsel->core.idx == idx)
2496 return evsel;
2497 }
2498
2499 return NULL;
2500 }
2501
2502 static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2503 {
2504 struct evsel *evsel;
2505
2506 if (!event->name)
2507 return;
2508
2509 evsel = evlist__find_by_index(evlist, event->core.idx);
2510 if (!evsel)
2511 return;
2512
2513 if (evsel->name)
2514 return;
2515
2516 evsel->name = strdup(event->name);
2517 }
2518
2519 static int
2520 process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2521 {
2522 struct perf_session *session;
2523 struct evsel *evsel, *events = read_event_desc(ff);
2524
2525 if (!events)
2526 return 0;
2527
2528 session = container_of(ff->ph, struct perf_session, header);
2529
2530 if (session->data->is_pipe) {
2531 /* Save events for reading later by print_event_desc,
2532 * since they can't be read again in pipe mode. */
2533 ff->events = events;
2534 }
2535
2536 for (evsel = events; evsel->core.attr.size; evsel++)
2537 evlist__set_event_name(session->evlist, evsel);
2538
2539 if (!session->data->is_pipe)
2540 free_event_desc(events);
2541
2542 return 0;
2543 }
2544
2545 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2546 {
2547 char *str, *cmdline = NULL, **argv = NULL;
2548 u32 nr, i, len = 0;
2549
2550 if (do_read_u32(ff, &nr))
2551 return -1;
2552
2553 ff->ph->env.nr_cmdline = nr;
2554
2555 cmdline = zalloc(ff->size + nr + 1);
2556 if (!cmdline)
2557 return -1;
2558
2559 argv = zalloc(sizeof(char *) * (nr + 1));
2560 if (!argv)
2561 goto error;
2562
2563 for (i = 0; i < nr; i++) {
2564 str = do_read_string(ff);
2565 if (!str)
2566 goto error;
2567
2568 argv[i] = cmdline + len;
2569 memcpy(argv[i], str, strlen(str) + 1);
2570 len += strlen(str) + 1;
2571 free(str);
2572 }
2573 ff->ph->env.cmdline = cmdline;
2574 ff->ph->env.cmdline_argv = (const char **) argv;
2575 return 0;
2576
2577 error:
2578 free(argv);
2579 free(cmdline);
2580 return -1;
2581 }
2582
2583 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2584 {
2585 u32 nr, i;
2586 char *str = NULL;
2587 struct strbuf sb;
2588 int cpu_nr = ff->ph->env.nr_cpus_avail;
2589 u64 size = 0;
2590 struct perf_header *ph = ff->ph;
2591 bool do_core_id_test = true;
2592
2593 ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2594 if (!ph->env.cpu)
2595 return -1;
2596
2597 if (do_read_u32(ff, &nr))
2598 goto free_cpu;
2599
2600 ph->env.nr_sibling_cores = nr;
2601 size += sizeof(u32);
2602 if (strbuf_init(&sb, 128) < 0)
2603 goto free_cpu;
2604
2605 for (i = 0; i < nr; i++) {
2606 str = do_read_string(ff);
2607 if (!str)
2608 goto error;
2609
2610 /* include a NULL character at the end */
2611 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2612 goto error;
2613 size += string_size(str);
2614 zfree(&str);
2615 }
2616 ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2617
2618 if (do_read_u32(ff, &nr))
2619 return -1;
2620
2621 ph->env.nr_sibling_threads = nr;
2622 size += sizeof(u32);
2623
2624 for (i = 0; i < nr; i++) {
2625 str = do_read_string(ff);
2626 if (!str)
2627 goto error;
2628
2629 /* include a NULL character at the end */
2630 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2631 goto error;
2632 size += string_size(str);
2633 zfree(&str);
2634 }
2635 ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2636
2637 /*
2638 * The header may be from old perf,
2639 * which doesn't include core id and socket id information.
2640 */
2641 if (ff->size <= size) {
2642 zfree(&ph->env.cpu);
2643 return 0;
2644 }
2645
2646 /* On s390 the socket_id number is not related to the numbers of cpus.
2647 * The socket_id number might be higher than the numbers of cpus.
2648 * This depends on the configuration.
2649 * AArch64 is the same.
2650 */
2651 if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2652 || !strncmp(ph->env.arch, "aarch64", 7)))
2653 do_core_id_test = false;
2654
2655 for (i = 0; i < (u32)cpu_nr; i++) {
2656 if (do_read_u32(ff, &nr))
2657 goto free_cpu;
2658
2659 ph->env.cpu[i].core_id = nr;
2660 size += sizeof(u32);
2661
2662 if (do_read_u32(ff, &nr))
2663 goto free_cpu;
2664
2665 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2666 pr_debug("socket_id number is too big."
2667 "You may need to upgrade the perf tool.\n");
2668 goto free_cpu;
2669 }
2670
2671 ph->env.cpu[i].socket_id = nr;
2672 size += sizeof(u32);
2673 }
2674
2675 /*
2676 * The header may be from old perf,
2677 * which doesn't include die information.
2678 */
2679 if (ff->size <= size)
2680 return 0;
2681
2682 if (do_read_u32(ff, &nr))
2683 return -1;
2684
2685 ph->env.nr_sibling_dies = nr;
2686 size += sizeof(u32);
2687
2688 for (i = 0; i < nr; i++) {
2689 str = do_read_string(ff);
2690 if (!str)
2691 goto error;
2692
2693 /* include a NULL character at the end */
2694 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2695 goto error;
2696 size += string_size(str);
2697 zfree(&str);
2698 }
2699 ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2700
2701 for (i = 0; i < (u32)cpu_nr; i++) {
2702 if (do_read_u32(ff, &nr))
2703 goto free_cpu;
2704
2705 ph->env.cpu[i].die_id = nr;
2706 }
2707
2708 return 0;
2709
2710 error:
2711 strbuf_release(&sb);
2712 zfree(&str);
2713 free_cpu:
2714 zfree(&ph->env.cpu);
2715 return -1;
2716 }
2717
2718 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2719 {
2720 struct numa_node *nodes, *n;
2721 u32 nr, i;
2722 char *str;
2723
2724 /* nr nodes */
2725 if (do_read_u32(ff, &nr))
2726 return -1;
2727
2728 nodes = zalloc(sizeof(*nodes) * nr);
2729 if (!nodes)
2730 return -ENOMEM;
2731
2732 for (i = 0; i < nr; i++) {
2733 n = &nodes[i];
2734
2735 /* node number */
2736 if (do_read_u32(ff, &n->node))
2737 goto error;
2738
2739 if (do_read_u64(ff, &n->mem_total))
2740 goto error;
2741
2742 if (do_read_u64(ff, &n->mem_free))
2743 goto error;
2744
2745 str = do_read_string(ff);
2746 if (!str)
2747 goto error;
2748
2749 n->map = perf_cpu_map__new(str);
2750 free(str);
2751 if (!n->map)
2752 goto error;
2753 }
2754 ff->ph->env.nr_numa_nodes = nr;
2755 ff->ph->env.numa_nodes = nodes;
2756 return 0;
2757
2758 error:
2759 free(nodes);
2760 return -1;
2761 }
2762
2763 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2764 {
2765 char *name;
2766 u32 pmu_num;
2767 u32 type;
2768 struct strbuf sb;
2769
2770 if (do_read_u32(ff, &pmu_num))
2771 return -1;
2772
2773 if (!pmu_num) {
2774 pr_debug("pmu mappings not available\n");
2775 return 0;
2776 }
2777
2778 ff->ph->env.nr_pmu_mappings = pmu_num;
2779 if (strbuf_init(&sb, 128) < 0)
2780 return -1;
2781
2782 while (pmu_num) {
2783 if (do_read_u32(ff, &type))
2784 goto error;
2785
2786 name = do_read_string(ff);
2787 if (!name)
2788 goto error;
2789
2790 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2791 goto error;
2792 /* include a NULL character at the end */
2793 if (strbuf_add(&sb, "", 1) < 0)
2794 goto error;
2795
2796 if (!strcmp(name, "msr"))
2797 ff->ph->env.msr_pmu_type = type;
2798
2799 free(name);
2800 pmu_num--;
2801 }
2802 ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2803 return 0;
2804
2805 error:
2806 strbuf_release(&sb);
2807 return -1;
2808 }
2809
2810 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2811 {
2812 size_t ret = -1;
2813 u32 i, nr, nr_groups;
2814 struct perf_session *session;
2815 struct evsel *evsel, *leader = NULL;
2816 struct group_desc {
2817 char *name;
2818 u32 leader_idx;
2819 u32 nr_members;
2820 } *desc;
2821
2822 if (do_read_u32(ff, &nr_groups))
2823 return -1;
2824
2825 ff->ph->env.nr_groups = nr_groups;
2826 if (!nr_groups) {
2827 pr_debug("group desc not available\n");
2828 return 0;
2829 }
2830
2831 desc = calloc(nr_groups, sizeof(*desc));
2832 if (!desc)
2833 return -1;
2834
2835 for (i = 0; i < nr_groups; i++) {
2836 desc[i].name = do_read_string(ff);
2837 if (!desc[i].name)
2838 goto out_free;
2839
2840 if (do_read_u32(ff, &desc[i].leader_idx))
2841 goto out_free;
2842
2843 if (do_read_u32(ff, &desc[i].nr_members))
2844 goto out_free;
2845 }
2846
2847 /*
2848 * Rebuild group relationship based on the group_desc
2849 */
2850 session = container_of(ff->ph, struct perf_session, header);
2851
2852 i = nr = 0;
2853 evlist__for_each_entry(session->evlist, evsel) {
2854 if (i < nr_groups && evsel->core.idx == (int) desc[i].leader_idx) {
2855 evsel__set_leader(evsel, evsel);
2856 /* {anon_group} is a dummy name */
2857 if (strcmp(desc[i].name, "{anon_group}")) {
2858 evsel->group_name = desc[i].name;
2859 desc[i].name = NULL;
2860 }
2861 evsel->core.nr_members = desc[i].nr_members;
2862
2863 if (i >= nr_groups || nr > 0) {
2864 pr_debug("invalid group desc\n");
2865 goto out_free;
2866 }
2867
2868 leader = evsel;
2869 nr = evsel->core.nr_members - 1;
2870 i++;
2871 } else if (nr) {
2872 /* This is a group member */
2873 evsel__set_leader(evsel, leader);
2874
2875 nr--;
2876 }
2877 }
2878
2879 if (i != nr_groups || nr != 0) {
2880 pr_debug("invalid group desc\n");
2881 goto out_free;
2882 }
2883
2884 ret = 0;
2885 out_free:
2886 for (i = 0; i < nr_groups; i++)
2887 zfree(&desc[i].name);
2888 free(desc);
2889
2890 return ret;
2891 }
2892
2893 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2894 {
2895 struct perf_session *session;
2896 int err;
2897
2898 session = container_of(ff->ph, struct perf_session, header);
2899
2900 err = auxtrace_index__process(ff->fd, ff->size, session,
2901 ff->ph->needs_swap);
2902 if (err < 0)
2903 pr_err("Failed to process auxtrace index\n");
2904 return err;
2905 }
2906
2907 static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2908 {
2909 struct cpu_cache_level *caches;
2910 u32 cnt, i, version;
2911
2912 if (do_read_u32(ff, &version))
2913 return -1;
2914
2915 if (version != 1)
2916 return -1;
2917
2918 if (do_read_u32(ff, &cnt))
2919 return -1;
2920
2921 caches = zalloc(sizeof(*caches) * cnt);
2922 if (!caches)
2923 return -1;
2924
2925 for (i = 0; i < cnt; i++) {
2926 struct cpu_cache_level *c = &caches[i];
2927
2928 #define _R(v) \
2929 if (do_read_u32(ff, &c->v)) \
2930 goto out_free_caches; \
2931
2932 _R(level)
2933 _R(line_size)
2934 _R(sets)
2935 _R(ways)
2936 #undef _R
2937
2938 #define _R(v) \
2939 c->v = do_read_string(ff); \
2940 if (!c->v) \
2941 goto out_free_caches; \
2942
2943 _R(type)
2944 _R(size)
2945 _R(map)
2946 #undef _R
2947 }
2948
2949 ff->ph->env.caches = caches;
2950 ff->ph->env.caches_cnt = cnt;
2951 return 0;
2952 out_free_caches:
2953 for (i = 0; i < cnt; i++) {
2954 free(caches[i].type);
2955 free(caches[i].size);
2956 free(caches[i].map);
2957 }
2958 free(caches);
2959 return -1;
2960 }
2961
2962 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2963 {
2964 struct perf_session *session;
2965 u64 first_sample_time, last_sample_time;
2966 int ret;
2967
2968 session = container_of(ff->ph, struct perf_session, header);
2969
2970 ret = do_read_u64(ff, &first_sample_time);
2971 if (ret)
2972 return -1;
2973
2974 ret = do_read_u64(ff, &last_sample_time);
2975 if (ret)
2976 return -1;
2977
2978 session->evlist->first_sample_time = first_sample_time;
2979 session->evlist->last_sample_time = last_sample_time;
2980 return 0;
2981 }
2982
2983 static int process_mem_topology(struct feat_fd *ff,
2984 void *data __maybe_unused)
2985 {
2986 struct memory_node *nodes;
2987 u64 version, i, nr, bsize;
2988 int ret = -1;
2989
2990 if (do_read_u64(ff, &version))
2991 return -1;
2992
2993 if (version != 1)
2994 return -1;
2995
2996 if (do_read_u64(ff, &bsize))
2997 return -1;
2998
2999 if (do_read_u64(ff, &nr))
3000 return -1;
3001
3002 nodes = zalloc(sizeof(*nodes) * nr);
3003 if (!nodes)
3004 return -1;
3005
3006 for (i = 0; i < nr; i++) {
3007 struct memory_node n;
3008
3009 #define _R(v) \
3010 if (do_read_u64(ff, &n.v)) \
3011 goto out; \
3012
3013 _R(node)
3014 _R(size)
3015
3016 #undef _R
3017
3018 if (do_read_bitmap(ff, &n.set, &n.size))
3019 goto out;
3020
3021 nodes[i] = n;
3022 }
3023
3024 ff->ph->env.memory_bsize = bsize;
3025 ff->ph->env.memory_nodes = nodes;
3026 ff->ph->env.nr_memory_nodes = nr;
3027 ret = 0;
3028
3029 out:
3030 if (ret)
3031 free(nodes);
3032 return ret;
3033 }
3034
3035 static int process_clockid(struct feat_fd *ff,
3036 void *data __maybe_unused)
3037 {
3038 if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
3039 return -1;
3040
3041 return 0;
3042 }
3043
3044 static int process_clock_data(struct feat_fd *ff,
3045 void *_data __maybe_unused)
3046 {
3047 u32 data32;
3048 u64 data64;
3049
3050 /* version */
3051 if (do_read_u32(ff, &data32))
3052 return -1;
3053
3054 if (data32 != 1)
3055 return -1;
3056
3057 /* clockid */
3058 if (do_read_u32(ff, &data32))
3059 return -1;
3060
3061 ff->ph->env.clock.clockid = data32;
3062
3063 /* TOD ref time */
3064 if (do_read_u64(ff, &data64))
3065 return -1;
3066
3067 ff->ph->env.clock.tod_ns = data64;
3068
3069 /* clockid ref time */
3070 if (do_read_u64(ff, &data64))
3071 return -1;
3072
3073 ff->ph->env.clock.clockid_ns = data64;
3074 ff->ph->env.clock.enabled = true;
3075 return 0;
3076 }
3077
3078 static int process_hybrid_topology(struct feat_fd *ff,
3079 void *data __maybe_unused)
3080 {
3081 struct hybrid_node *nodes, *n;
3082 u32 nr, i;
3083
3084 /* nr nodes */
3085 if (do_read_u32(ff, &nr))
3086 return -1;
3087
3088 nodes = zalloc(sizeof(*nodes) * nr);
3089 if (!nodes)
3090 return -ENOMEM;
3091
3092 for (i = 0; i < nr; i++) {
3093 n = &nodes[i];
3094
3095 n->pmu_name = do_read_string(ff);
3096 if (!n->pmu_name)
3097 goto error;
3098
3099 n->cpus = do_read_string(ff);
3100 if (!n->cpus)
3101 goto error;
3102 }
3103
3104 ff->ph->env.nr_hybrid_nodes = nr;
3105 ff->ph->env.hybrid_nodes = nodes;
3106 return 0;
3107
3108 error:
3109 for (i = 0; i < nr; i++) {
3110 free(nodes[i].pmu_name);
3111 free(nodes[i].cpus);
3112 }
3113
3114 free(nodes);
3115 return -1;
3116 }
3117
3118 static int process_dir_format(struct feat_fd *ff,
3119 void *_data __maybe_unused)
3120 {
3121 struct perf_session *session;
3122 struct perf_data *data;
3123
3124 session = container_of(ff->ph, struct perf_session, header);
3125 data = session->data;
3126
3127 if (WARN_ON(!perf_data__is_dir(data)))
3128 return -1;
3129
3130 return do_read_u64(ff, &data->dir.version);
3131 }
3132
3133 #ifdef HAVE_LIBBPF_SUPPORT
3134 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3135 {
3136 struct bpf_prog_info_node *info_node;
3137 struct perf_env *env = &ff->ph->env;
3138 struct perf_bpil *info_linear;
3139 u32 count, i;
3140 int err = -1;
3141
3142 if (ff->ph->needs_swap) {
3143 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3144 return 0;
3145 }
3146
3147 if (do_read_u32(ff, &count))
3148 return -1;
3149
3150 down_write(&env->bpf_progs.lock);
3151
3152 for (i = 0; i < count; ++i) {
3153 u32 info_len, data_len;
3154
3155 info_linear = NULL;
3156 info_node = NULL;
3157 if (do_read_u32(ff, &info_len))
3158 goto out;
3159 if (do_read_u32(ff, &data_len))
3160 goto out;
3161
3162 if (info_len > sizeof(struct bpf_prog_info)) {
3163 pr_warning("detected invalid bpf_prog_info\n");
3164 goto out;
3165 }
3166
3167 info_linear = malloc(sizeof(struct perf_bpil) +
3168 data_len);
3169 if (!info_linear)
3170 goto out;
3171 info_linear->info_len = sizeof(struct bpf_prog_info);
3172 info_linear->data_len = data_len;
3173 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3174 goto out;
3175 if (__do_read(ff, &info_linear->info, info_len))
3176 goto out;
3177 if (info_len < sizeof(struct bpf_prog_info))
3178 memset(((void *)(&info_linear->info)) + info_len, 0,
3179 sizeof(struct bpf_prog_info) - info_len);
3180
3181 if (__do_read(ff, info_linear->data, data_len))
3182 goto out;
3183
3184 info_node = malloc(sizeof(struct bpf_prog_info_node));
3185 if (!info_node)
3186 goto out;
3187
3188 /* after reading from file, translate offset to address */
3189 bpil_offs_to_addr(info_linear);
3190 info_node->info_linear = info_linear;
3191 __perf_env__insert_bpf_prog_info(env, info_node);
3192 }
3193
3194 up_write(&env->bpf_progs.lock);
3195 return 0;
3196 out:
3197 free(info_linear);
3198 free(info_node);
3199 up_write(&env->bpf_progs.lock);
3200 return err;
3201 }
3202
3203 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3204 {
3205 struct perf_env *env = &ff->ph->env;
3206 struct btf_node *node = NULL;
3207 u32 count, i;
3208 int err = -1;
3209
3210 if (ff->ph->needs_swap) {
3211 pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3212 return 0;
3213 }
3214
3215 if (do_read_u32(ff, &count))
3216 return -1;
3217
3218 down_write(&env->bpf_progs.lock);
3219
3220 for (i = 0; i < count; ++i) {
3221 u32 id, data_size;
3222
3223 if (do_read_u32(ff, &id))
3224 goto out;
3225 if (do_read_u32(ff, &data_size))
3226 goto out;
3227
3228 node = malloc(sizeof(struct btf_node) + data_size);
3229 if (!node)
3230 goto out;
3231
3232 node->id = id;
3233 node->data_size = data_size;
3234
3235 if (__do_read(ff, node->data, data_size))
3236 goto out;
3237
3238 __perf_env__insert_btf(env, node);
3239 node = NULL;
3240 }
3241
3242 err = 0;
3243 out:
3244 up_write(&env->bpf_progs.lock);
3245 free(node);
3246 return err;
3247 }
3248 #endif // HAVE_LIBBPF_SUPPORT
3249
3250 static int process_compressed(struct feat_fd *ff,
3251 void *data __maybe_unused)
3252 {
3253 if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3254 return -1;
3255
3256 if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3257 return -1;
3258
3259 if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3260 return -1;
3261
3262 if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3263 return -1;
3264
3265 if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3266 return -1;
3267
3268 return 0;
3269 }
3270
3271 static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
3272 char ***caps, unsigned int *max_branches,
3273 unsigned int *br_cntr_nr,
3274 unsigned int *br_cntr_width)
3275 {
3276 char *name, *value, *ptr;
3277 u32 nr_pmu_caps, i;
3278
3279 *nr_caps = 0;
3280 *caps = NULL;
3281
3282 if (do_read_u32(ff, &nr_pmu_caps))
3283 return -1;
3284
3285 if (!nr_pmu_caps)
3286 return 0;
3287
3288 *caps = zalloc(sizeof(char *) * nr_pmu_caps);
3289 if (!*caps)
3290 return -1;
3291
3292 for (i = 0; i < nr_pmu_caps; i++) {
3293 name = do_read_string(ff);
3294 if (!name)
3295 goto error;
3296
3297 value = do_read_string(ff);
3298 if (!value)
3299 goto free_name;
3300
3301 if (asprintf(&ptr, "%s=%s", name, value) < 0)
3302 goto free_value;
3303
3304 (*caps)[i] = ptr;
3305
3306 if (!strcmp(name, "branches"))
3307 *max_branches = atoi(value);
3308
3309 if (!strcmp(name, "branch_counter_nr"))
3310 *br_cntr_nr = atoi(value);
3311
3312 if (!strcmp(name, "branch_counter_width"))
3313 *br_cntr_width = atoi(value);
3314
3315 free(value);
3316 free(name);
3317 }
3318 *nr_caps = nr_pmu_caps;
3319 return 0;
3320
3321 free_value:
3322 free(value);
3323 free_name:
3324 free(name);
3325 error:
3326 for (; i > 0; i--)
3327 free((*caps)[i - 1]);
3328 free(*caps);
3329 *caps = NULL;
3330 *nr_caps = 0;
3331 return -1;
3332 }
3333
3334 static int process_cpu_pmu_caps(struct feat_fd *ff,
3335 void *data __maybe_unused)
3336 {
3337 int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3338 &ff->ph->env.cpu_pmu_caps,
3339 &ff->ph->env.max_branches,
3340 &ff->ph->env.br_cntr_nr,
3341 &ff->ph->env.br_cntr_width);
3342
3343 if (!ret && !ff->ph->env.cpu_pmu_caps)
3344 pr_debug("cpu pmu capabilities not available\n");
3345 return ret;
3346 }
3347
3348 static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused)
3349 {
3350 struct pmu_caps *pmu_caps;
3351 u32 nr_pmu, i;
3352 int ret;
3353 int j;
3354
3355 if (do_read_u32(ff, &nr_pmu))
3356 return -1;
3357
3358 if (!nr_pmu) {
3359 pr_debug("pmu capabilities not available\n");
3360 return 0;
3361 }
3362
3363 pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
3364 if (!pmu_caps)
3365 return -ENOMEM;
3366
3367 for (i = 0; i < nr_pmu; i++) {
3368 ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps,
3369 &pmu_caps[i].caps,
3370 &pmu_caps[i].max_branches,
3371 &pmu_caps[i].br_cntr_nr,
3372 &pmu_caps[i].br_cntr_width);
3373 if (ret)
3374 goto err;
3375
3376 pmu_caps[i].pmu_name = do_read_string(ff);
3377 if (!pmu_caps[i].pmu_name) {
3378 ret = -1;
3379 goto err;
3380 }
3381 if (!pmu_caps[i].nr_caps) {
3382 pr_debug("%s pmu capabilities not available\n",
3383 pmu_caps[i].pmu_name);
3384 }
3385 }
3386
3387 ff->ph->env.nr_pmus_with_caps = nr_pmu;
3388 ff->ph->env.pmu_caps = pmu_caps;
3389 return 0;
3390
3391 err:
3392 for (i = 0; i < nr_pmu; i++) {
3393 for (j = 0; j < pmu_caps[i].nr_caps; j++)
3394 free(pmu_caps[i].caps[j]);
3395 free(pmu_caps[i].caps);
3396 free(pmu_caps[i].pmu_name);
3397 }
3398
3399 free(pmu_caps);
3400 return ret;
3401 }
3402
3403 #define FEAT_OPR(n, func, __full_only) \
3404 [HEADER_##n] = { \
3405 .name = __stringify(n), \
3406 .write = write_##func, \
3407 .print = print_##func, \
3408 .full_only = __full_only, \
3409 .process = process_##func, \
3410 .synthesize = true \
3411 }
3412
3413 #define FEAT_OPN(n, func, __full_only) \
3414 [HEADER_##n] = { \
3415 .name = __stringify(n), \
3416 .write = write_##func, \
3417 .print = print_##func, \
3418 .full_only = __full_only, \
3419 .process = process_##func \
3420 }
3421
3422 /* feature_ops not implemented: */
3423 #define print_tracing_data NULL
3424 #define print_build_id NULL
3425
3426 #define process_branch_stack NULL
3427 #define process_stat NULL
3428
3429 // Only used in util/synthetic-events.c
3430 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3431
3432 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3433 #ifdef HAVE_LIBTRACEEVENT
3434 FEAT_OPN(TRACING_DATA, tracing_data, false),
3435 #endif
3436 FEAT_OPN(BUILD_ID, build_id, false),
3437 FEAT_OPR(HOSTNAME, hostname, false),
3438 FEAT_OPR(OSRELEASE, osrelease, false),
3439 FEAT_OPR(VERSION, version, false),
3440 FEAT_OPR(ARCH, arch, false),
3441 FEAT_OPR(NRCPUS, nrcpus, false),
3442 FEAT_OPR(CPUDESC, cpudesc, false),
3443 FEAT_OPR(CPUID, cpuid, false),
3444 FEAT_OPR(TOTAL_MEM, total_mem, false),
3445 FEAT_OPR(EVENT_DESC, event_desc, false),
3446 FEAT_OPR(CMDLINE, cmdline, false),
3447 FEAT_OPR(CPU_TOPOLOGY, cpu_topology, true),
3448 FEAT_OPR(NUMA_TOPOLOGY, numa_topology, true),
3449 FEAT_OPN(BRANCH_STACK, branch_stack, false),
3450 FEAT_OPR(PMU_MAPPINGS, pmu_mappings, false),
3451 FEAT_OPR(GROUP_DESC, group_desc, false),
3452 FEAT_OPN(AUXTRACE, auxtrace, false),
3453 FEAT_OPN(STAT, stat, false),
3454 FEAT_OPN(CACHE, cache, true),
3455 FEAT_OPR(SAMPLE_TIME, sample_time, false),
3456 FEAT_OPR(MEM_TOPOLOGY, mem_topology, true),
3457 FEAT_OPR(CLOCKID, clockid, false),
3458 FEAT_OPN(DIR_FORMAT, dir_format, false),
3459 #ifdef HAVE_LIBBPF_SUPPORT
3460 FEAT_OPR(BPF_PROG_INFO, bpf_prog_info, false),
3461 FEAT_OPR(BPF_BTF, bpf_btf, false),
3462 #endif
3463 FEAT_OPR(COMPRESSED, compressed, false),
3464 FEAT_OPR(CPU_PMU_CAPS, cpu_pmu_caps, false),
3465 FEAT_OPR(CLOCK_DATA, clock_data, false),
3466 FEAT_OPN(HYBRID_TOPOLOGY, hybrid_topology, true),
3467 FEAT_OPR(PMU_CAPS, pmu_caps, false),
3468 };
3469
3470 struct header_print_data {
3471 FILE *fp;
3472 bool full; /* extended list of headers */
3473 };
3474
3475 static int perf_file_section__fprintf_info(struct perf_file_section *section,
3476 struct perf_header *ph,
3477 int feat, int fd, void *data)
3478 {
3479 struct header_print_data *hd = data;
3480 struct feat_fd ff;
3481
3482 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3483 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3484 "%d, continuing...\n", section->offset, feat);
3485 return 0;
3486 }
3487 if (feat >= HEADER_LAST_FEATURE) {
3488 pr_warning("unknown feature %d\n", feat);
3489 return 0;
3490 }
3491 if (!feat_ops[feat].print)
3492 return 0;
3493
3494 ff = (struct feat_fd) {
3495 .fd = fd,
3496 .ph = ph,
3497 };
3498
3499 if (!feat_ops[feat].full_only || hd->full)
3500 feat_ops[feat].print(&ff, hd->fp);
3501 else
3502 fprintf(hd->fp, "# %s info available, use -I to display\n",
3503 feat_ops[feat].name);
3504
3505 return 0;
3506 }
3507
3508 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3509 {
3510 struct header_print_data hd;
3511 struct perf_header *header = &session->header;
3512 int fd = perf_data__fd(session->data);
3513 struct stat st;
3514 time_t stctime;
3515 int ret, bit;
3516
3517 hd.fp = fp;
3518 hd.full = full;
3519
3520 ret = fstat(fd, &st);
3521 if (ret == -1)
3522 return -1;
3523
3524 stctime = st.st_mtime;
3525 fprintf(fp, "# captured on : %s", ctime(&stctime));
3526
3527 fprintf(fp, "# header version : %u\n", header->version);
3528 fprintf(fp, "# data offset : %" PRIu64 "\n", header->data_offset);
3529 fprintf(fp, "# data size : %" PRIu64 "\n", header->data_size);
3530 fprintf(fp, "# feat offset : %" PRIu64 "\n", header->feat_offset);
3531
3532 perf_header__process_sections(header, fd, &hd,
3533 perf_file_section__fprintf_info);
3534
3535 if (session->data->is_pipe)
3536 return 0;
3537
3538 fprintf(fp, "# missing features: ");
3539 for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3540 if (bit)
3541 fprintf(fp, "%s ", feat_ops[bit].name);
3542 }
3543
3544 fprintf(fp, "\n");
3545 return 0;
3546 }
3547
3548 struct header_fw {
3549 struct feat_writer fw;
3550 struct feat_fd *ff;
3551 };
3552
3553 static int feat_writer_cb(struct feat_writer *fw, void *buf, size_t sz)
3554 {
3555 struct header_fw *h = container_of(fw, struct header_fw, fw);
3556
3557 return do_write(h->ff, buf, sz);
3558 }
3559
3560 static int do_write_feat(struct feat_fd *ff, int type,
3561 struct perf_file_section **p,
3562 struct evlist *evlist,
3563 struct feat_copier *fc)
3564 {
3565 int err;
3566 int ret = 0;
3567
3568 if (perf_header__has_feat(ff->ph, type)) {
3569 if (!feat_ops[type].write)
3570 return -1;
3571
3572 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3573 return -1;
3574
3575 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3576
3577 /*
3578 * Hook to let perf inject copy features sections from the input
3579 * file.
3580 */
3581 if (fc && fc->copy) {
3582 struct header_fw h = {
3583 .fw.write = feat_writer_cb,
3584 .ff = ff,
3585 };
3586
3587 /* ->copy() returns 0 if the feature was not copied */
3588 err = fc->copy(fc, type, &h.fw);
3589 } else {
3590 err = 0;
3591 }
3592 if (!err)
3593 err = feat_ops[type].write(ff, evlist);
3594 if (err < 0) {
3595 pr_debug("failed to write feature %s\n", feat_ops[type].name);
3596
3597 /* undo anything written */
3598 lseek(ff->fd, (*p)->offset, SEEK_SET);
3599
3600 return -1;
3601 }
3602 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3603 (*p)++;
3604 }
3605 return ret;
3606 }
3607
3608 static int perf_header__adds_write(struct perf_header *header,
3609 struct evlist *evlist, int fd,
3610 struct feat_copier *fc)
3611 {
3612 int nr_sections;
3613 struct feat_fd ff = {
3614 .fd = fd,
3615 .ph = header,
3616 };
3617 struct perf_file_section *feat_sec, *p;
3618 int sec_size;
3619 u64 sec_start;
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 = p = calloc(nr_sections, sizeof(*feat_sec));
3628 if (feat_sec == NULL)
3629 return -ENOMEM;
3630
3631 sec_size = sizeof(*feat_sec) * nr_sections;
3632
3633 sec_start = header->feat_offset;
3634 lseek(fd, sec_start + sec_size, SEEK_SET);
3635
3636 for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3637 if (do_write_feat(&ff, feat, &p, evlist, fc))
3638 perf_header__clear_feat(header, feat);
3639 }
3640
3641 lseek(fd, sec_start, SEEK_SET);
3642 /*
3643 * may write more than needed due to dropped feature, but
3644 * this is okay, reader will skip the missing entries
3645 */
3646 err = do_write(&ff, feat_sec, sec_size);
3647 if (err < 0)
3648 pr_debug("failed to write feature section\n");
3649 free(ff.buf); /* TODO: added to silence clang-tidy. */
3650 free(feat_sec);
3651 return err;
3652 }
3653
3654 int perf_header__write_pipe(int fd)
3655 {
3656 struct perf_pipe_file_header f_header;
3657 struct feat_fd ff = {
3658 .fd = fd,
3659 };
3660 int err;
3661
3662 f_header = (struct perf_pipe_file_header){
3663 .magic = PERF_MAGIC,
3664 .size = sizeof(f_header),
3665 };
3666
3667 err = do_write(&ff, &f_header, sizeof(f_header));
3668 if (err < 0) {
3669 pr_debug("failed to write perf pipe header\n");
3670 return err;
3671 }
3672 free(ff.buf);
3673 return 0;
3674 }
3675
3676 static int perf_session__do_write_header(struct perf_session *session,
3677 struct evlist *evlist,
3678 int fd, bool at_exit,
3679 struct feat_copier *fc)
3680 {
3681 struct perf_file_header f_header;
3682 struct perf_file_attr f_attr;
3683 struct perf_header *header = &session->header;
3684 struct evsel *evsel;
3685 struct feat_fd ff = {
3686 .fd = fd,
3687 };
3688 u64 attr_offset;
3689 int err;
3690
3691 lseek(fd, sizeof(f_header), SEEK_SET);
3692
3693 evlist__for_each_entry(session->evlist, evsel) {
3694 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3695 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3696 if (err < 0) {
3697 pr_debug("failed to write perf header\n");
3698 free(ff.buf);
3699 return err;
3700 }
3701 }
3702
3703 attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3704
3705 evlist__for_each_entry(evlist, evsel) {
3706 if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3707 /*
3708 * We are likely in "perf inject" and have read
3709 * from an older file. Update attr size so that
3710 * reader gets the right offset to the ids.
3711 */
3712 evsel->core.attr.size = sizeof(evsel->core.attr);
3713 }
3714 f_attr = (struct perf_file_attr){
3715 .attr = evsel->core.attr,
3716 .ids = {
3717 .offset = evsel->id_offset,
3718 .size = evsel->core.ids * sizeof(u64),
3719 }
3720 };
3721 err = do_write(&ff, &f_attr, sizeof(f_attr));
3722 if (err < 0) {
3723 pr_debug("failed to write perf header attribute\n");
3724 free(ff.buf);
3725 return err;
3726 }
3727 }
3728
3729 if (!header->data_offset)
3730 header->data_offset = lseek(fd, 0, SEEK_CUR);
3731 header->feat_offset = header->data_offset + header->data_size;
3732
3733 if (at_exit) {
3734 err = perf_header__adds_write(header, evlist, fd, fc);
3735 if (err < 0) {
3736 free(ff.buf);
3737 return err;
3738 }
3739 }
3740
3741 f_header = (struct perf_file_header){
3742 .magic = PERF_MAGIC,
3743 .size = sizeof(f_header),
3744 .attr_size = sizeof(f_attr),
3745 .attrs = {
3746 .offset = attr_offset,
3747 .size = evlist->core.nr_entries * sizeof(f_attr),
3748 },
3749 .data = {
3750 .offset = header->data_offset,
3751 .size = header->data_size,
3752 },
3753 /* event_types is ignored, store zeros */
3754 };
3755
3756 memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3757
3758 lseek(fd, 0, SEEK_SET);
3759 err = do_write(&ff, &f_header, sizeof(f_header));
3760 free(ff.buf);
3761 if (err < 0) {
3762 pr_debug("failed to write perf header\n");
3763 return err;
3764 }
3765 lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3766
3767 return 0;
3768 }
3769
3770 int perf_session__write_header(struct perf_session *session,
3771 struct evlist *evlist,
3772 int fd, bool at_exit)
3773 {
3774 return perf_session__do_write_header(session, evlist, fd, at_exit, NULL);
3775 }
3776
3777 size_t perf_session__data_offset(const struct evlist *evlist)
3778 {
3779 struct evsel *evsel;
3780 size_t data_offset;
3781
3782 data_offset = sizeof(struct perf_file_header);
3783 evlist__for_each_entry(evlist, evsel) {
3784 data_offset += evsel->core.ids * sizeof(u64);
3785 }
3786 data_offset += evlist->core.nr_entries * sizeof(struct perf_file_attr);
3787
3788 return data_offset;
3789 }
3790
3791 int perf_session__inject_header(struct perf_session *session,
3792 struct evlist *evlist,
3793 int fd,
3794 struct feat_copier *fc)
3795 {
3796 return perf_session__do_write_header(session, evlist, fd, true, fc);
3797 }
3798
3799 static int perf_header__getbuffer64(struct perf_header *header,
3800 int fd, void *buf, size_t size)
3801 {
3802 if (readn(fd, buf, size) <= 0)
3803 return -1;
3804
3805 if (header->needs_swap)
3806 mem_bswap_64(buf, size);
3807
3808 return 0;
3809 }
3810
3811 int perf_header__process_sections(struct perf_header *header, int fd,
3812 void *data,
3813 int (*process)(struct perf_file_section *section,
3814 struct perf_header *ph,
3815 int feat, int fd, void *data))
3816 {
3817 struct perf_file_section *feat_sec, *sec;
3818 int nr_sections;
3819 int sec_size;
3820 int feat;
3821 int err;
3822
3823 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3824 if (!nr_sections)
3825 return 0;
3826
3827 feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3828 if (!feat_sec)
3829 return -1;
3830
3831 sec_size = sizeof(*feat_sec) * nr_sections;
3832
3833 lseek(fd, header->feat_offset, SEEK_SET);
3834
3835 err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3836 if (err < 0)
3837 goto out_free;
3838
3839 for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3840 err = process(sec++, header, feat, fd, data);
3841 if (err < 0)
3842 goto out_free;
3843 }
3844 err = 0;
3845 out_free:
3846 free(feat_sec);
3847 return err;
3848 }
3849
3850 static const int attr_file_abi_sizes[] = {
3851 [0] = PERF_ATTR_SIZE_VER0,
3852 [1] = PERF_ATTR_SIZE_VER1,
3853 [2] = PERF_ATTR_SIZE_VER2,
3854 [3] = PERF_ATTR_SIZE_VER3,
3855 [4] = PERF_ATTR_SIZE_VER4,
3856 0,
3857 };
3858
3859 /*
3860 * In the legacy file format, the magic number is not used to encode endianness.
3861 * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3862 * on ABI revisions, we need to try all combinations for all endianness to
3863 * detect the endianness.
3864 */
3865 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3866 {
3867 uint64_t ref_size, attr_size;
3868 int i;
3869
3870 for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3871 ref_size = attr_file_abi_sizes[i]
3872 + sizeof(struct perf_file_section);
3873 if (hdr_sz != ref_size) {
3874 attr_size = bswap_64(hdr_sz);
3875 if (attr_size != ref_size)
3876 continue;
3877
3878 ph->needs_swap = true;
3879 }
3880 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3881 i,
3882 ph->needs_swap);
3883 return 0;
3884 }
3885 /* could not determine endianness */
3886 return -1;
3887 }
3888
3889 #define PERF_PIPE_HDR_VER0 16
3890
3891 static const size_t attr_pipe_abi_sizes[] = {
3892 [0] = PERF_PIPE_HDR_VER0,
3893 0,
3894 };
3895
3896 /*
3897 * In the legacy pipe format, there is an implicit assumption that endianness
3898 * between host recording the samples, and host parsing the samples is the
3899 * same. This is not always the case given that the pipe output may always be
3900 * redirected into a file and analyzed on a different machine with possibly a
3901 * different endianness and perf_event ABI revisions in the perf tool itself.
3902 */
3903 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3904 {
3905 u64 attr_size;
3906 int i;
3907
3908 for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3909 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3910 attr_size = bswap_64(hdr_sz);
3911 if (attr_size != hdr_sz)
3912 continue;
3913
3914 ph->needs_swap = true;
3915 }
3916 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3917 return 0;
3918 }
3919 return -1;
3920 }
3921
3922 bool is_perf_magic(u64 magic)
3923 {
3924 if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3925 || magic == __perf_magic2
3926 || magic == __perf_magic2_sw)
3927 return true;
3928
3929 return false;
3930 }
3931
3932 static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3933 bool is_pipe, struct perf_header *ph)
3934 {
3935 int ret;
3936
3937 /* check for legacy format */
3938 ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3939 if (ret == 0) {
3940 ph->version = PERF_HEADER_VERSION_1;
3941 pr_debug("legacy perf.data format\n");
3942 if (is_pipe)
3943 return try_all_pipe_abis(hdr_sz, ph);
3944
3945 return try_all_file_abis(hdr_sz, ph);
3946 }
3947 /*
3948 * the new magic number serves two purposes:
3949 * - unique number to identify actual perf.data files
3950 * - encode endianness of file
3951 */
3952 ph->version = PERF_HEADER_VERSION_2;
3953
3954 /* check magic number with one endianness */
3955 if (magic == __perf_magic2)
3956 return 0;
3957
3958 /* check magic number with opposite endianness */
3959 if (magic != __perf_magic2_sw)
3960 return -1;
3961
3962 ph->needs_swap = true;
3963
3964 return 0;
3965 }
3966
3967 int perf_file_header__read(struct perf_file_header *header,
3968 struct perf_header *ph, int fd)
3969 {
3970 ssize_t ret;
3971
3972 lseek(fd, 0, SEEK_SET);
3973
3974 ret = readn(fd, header, sizeof(*header));
3975 if (ret <= 0)
3976 return -1;
3977
3978 if (check_magic_endian(header->magic,
3979 header->attr_size, false, ph) < 0) {
3980 pr_debug("magic/endian check failed\n");
3981 return -1;
3982 }
3983
3984 if (ph->needs_swap) {
3985 mem_bswap_64(header, offsetof(struct perf_file_header,
3986 adds_features));
3987 }
3988
3989 if (header->size != sizeof(*header)) {
3990 /* Support the previous format */
3991 if (header->size == offsetof(typeof(*header), adds_features))
3992 bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3993 else
3994 return -1;
3995 } else if (ph->needs_swap) {
3996 /*
3997 * feature bitmap is declared as an array of unsigned longs --
3998 * not good since its size can differ between the host that
3999 * generated the data file and the host analyzing the file.
4000 *
4001 * We need to handle endianness, but we don't know the size of
4002 * the unsigned long where the file was generated. Take a best
4003 * guess at determining it: try 64-bit swap first (ie., file
4004 * created on a 64-bit host), and check if the hostname feature
4005 * bit is set (this feature bit is forced on as of fbe96f2).
4006 * If the bit is not, undo the 64-bit swap and try a 32-bit
4007 * swap. If the hostname bit is still not set (e.g., older data
4008 * file), punt and fallback to the original behavior --
4009 * clearing all feature bits and setting buildid.
4010 */
4011 mem_bswap_64(&header->adds_features,
4012 BITS_TO_U64(HEADER_FEAT_BITS));
4013
4014 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
4015 /* unswap as u64 */
4016 mem_bswap_64(&header->adds_features,
4017 BITS_TO_U64(HEADER_FEAT_BITS));
4018
4019 /* unswap as u32 */
4020 mem_bswap_32(&header->adds_features,
4021 BITS_TO_U32(HEADER_FEAT_BITS));
4022 }
4023
4024 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
4025 bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
4026 __set_bit(HEADER_BUILD_ID, header->adds_features);
4027 }
4028 }
4029
4030 memcpy(&ph->adds_features, &header->adds_features,
4031 sizeof(ph->adds_features));
4032
4033 ph->data_offset = header->data.offset;
4034 ph->data_size = header->data.size;
4035 ph->feat_offset = header->data.offset + header->data.size;
4036 return 0;
4037 }
4038
4039 static int perf_file_section__process(struct perf_file_section *section,
4040 struct perf_header *ph,
4041 int feat, int fd, void *data)
4042 {
4043 struct feat_fd fdd = {
4044 .fd = fd,
4045 .ph = ph,
4046 .size = section->size,
4047 .offset = section->offset,
4048 };
4049
4050 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
4051 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
4052 "%d, continuing...\n", section->offset, feat);
4053 return 0;
4054 }
4055
4056 if (feat >= HEADER_LAST_FEATURE) {
4057 pr_debug("unknown feature %d, continuing...\n", feat);
4058 return 0;
4059 }
4060
4061 if (!feat_ops[feat].process)
4062 return 0;
4063
4064 return feat_ops[feat].process(&fdd, data);
4065 }
4066
4067 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
4068 struct perf_header *ph,
4069 struct perf_data* data,
4070 bool repipe, int repipe_fd)
4071 {
4072 struct feat_fd ff = {
4073 .fd = repipe_fd,
4074 .ph = ph,
4075 };
4076 ssize_t ret;
4077
4078 ret = perf_data__read(data, header, sizeof(*header));
4079 if (ret <= 0)
4080 return -1;
4081
4082 if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
4083 pr_debug("endian/magic failed\n");
4084 return -1;
4085 }
4086
4087 if (ph->needs_swap)
4088 header->size = bswap_64(header->size);
4089
4090 if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
4091 return -1;
4092
4093 return 0;
4094 }
4095
4096 static int perf_header__read_pipe(struct perf_session *session, int repipe_fd)
4097 {
4098 struct perf_header *header = &session->header;
4099 struct perf_pipe_file_header f_header;
4100
4101 if (perf_file_header__read_pipe(&f_header, header, session->data,
4102 session->repipe, repipe_fd) < 0) {
4103 pr_debug("incompatible file format\n");
4104 return -EINVAL;
4105 }
4106
4107 return f_header.size == sizeof(f_header) ? 0 : -1;
4108 }
4109
4110 static int read_attr(int fd, struct perf_header *ph,
4111 struct perf_file_attr *f_attr)
4112 {
4113 struct perf_event_attr *attr = &f_attr->attr;
4114 size_t sz, left;
4115 size_t our_sz = sizeof(f_attr->attr);
4116 ssize_t ret;
4117
4118 memset(f_attr, 0, sizeof(*f_attr));
4119
4120 /* read minimal guaranteed structure */
4121 ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
4122 if (ret <= 0) {
4123 pr_debug("cannot read %d bytes of header attr\n",
4124 PERF_ATTR_SIZE_VER0);
4125 return -1;
4126 }
4127
4128 /* on file perf_event_attr size */
4129 sz = attr->size;
4130
4131 if (ph->needs_swap)
4132 sz = bswap_32(sz);
4133
4134 if (sz == 0) {
4135 /* assume ABI0 */
4136 sz = PERF_ATTR_SIZE_VER0;
4137 } else if (sz > our_sz) {
4138 pr_debug("file uses a more recent and unsupported ABI"
4139 " (%zu bytes extra)\n", sz - our_sz);
4140 return -1;
4141 }
4142 /* what we have not yet read and that we know about */
4143 left = sz - PERF_ATTR_SIZE_VER0;
4144 if (left) {
4145 void *ptr = attr;
4146 ptr += PERF_ATTR_SIZE_VER0;
4147
4148 ret = readn(fd, ptr, left);
4149 }
4150 /* read perf_file_section, ids are read in caller */
4151 ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
4152
4153 return ret <= 0 ? -1 : 0;
4154 }
4155
4156 #ifdef HAVE_LIBTRACEEVENT
4157 static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
4158 {
4159 struct tep_event *event;
4160 char bf[128];
4161
4162 /* already prepared */
4163 if (evsel->tp_format)
4164 return 0;
4165
4166 if (pevent == NULL) {
4167 pr_debug("broken or missing trace data\n");
4168 return -1;
4169 }
4170
4171 event = tep_find_event(pevent, evsel->core.attr.config);
4172 if (event == NULL) {
4173 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
4174 return -1;
4175 }
4176
4177 if (!evsel->name) {
4178 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
4179 evsel->name = strdup(bf);
4180 if (evsel->name == NULL)
4181 return -1;
4182 }
4183
4184 evsel->tp_format = event;
4185 return 0;
4186 }
4187
4188 static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
4189 {
4190 struct evsel *pos;
4191
4192 evlist__for_each_entry(evlist, pos) {
4193 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
4194 evsel__prepare_tracepoint_event(pos, pevent))
4195 return -1;
4196 }
4197
4198 return 0;
4199 }
4200 #endif
4201
4202 int perf_session__read_header(struct perf_session *session, int repipe_fd)
4203 {
4204 struct perf_data *data = session->data;
4205 struct perf_header *header = &session->header;
4206 struct perf_file_header f_header;
4207 struct perf_file_attr f_attr;
4208 u64 f_id;
4209 int nr_attrs, nr_ids, i, j, err;
4210 int fd = perf_data__fd(data);
4211
4212 session->evlist = evlist__new();
4213 if (session->evlist == NULL)
4214 return -ENOMEM;
4215
4216 session->evlist->env = &header->env;
4217 session->machines.host.env = &header->env;
4218
4219 /*
4220 * We can read 'pipe' data event from regular file,
4221 * check for the pipe header regardless of source.
4222 */
4223 err = perf_header__read_pipe(session, repipe_fd);
4224 if (!err || perf_data__is_pipe(data)) {
4225 data->is_pipe = true;
4226 return err;
4227 }
4228
4229 if (perf_file_header__read(&f_header, header, fd) < 0)
4230 return -EINVAL;
4231
4232 if (header->needs_swap && data->in_place_update) {
4233 pr_err("In-place update not supported when byte-swapping is required\n");
4234 return -EINVAL;
4235 }
4236
4237 /*
4238 * Sanity check that perf.data was written cleanly; data size is
4239 * initialized to 0 and updated only if the on_exit function is run.
4240 * If data size is still 0 then the file contains only partial
4241 * information. Just warn user and process it as much as it can.
4242 */
4243 if (f_header.data.size == 0) {
4244 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4245 "Was the 'perf record' command properly terminated?\n",
4246 data->file.path);
4247 }
4248
4249 if (f_header.attr_size == 0) {
4250 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4251 "Was the 'perf record' command properly terminated?\n",
4252 data->file.path);
4253 return -EINVAL;
4254 }
4255
4256 nr_attrs = f_header.attrs.size / f_header.attr_size;
4257 lseek(fd, f_header.attrs.offset, SEEK_SET);
4258
4259 for (i = 0; i < nr_attrs; i++) {
4260 struct evsel *evsel;
4261 off_t tmp;
4262
4263 if (read_attr(fd, header, &f_attr) < 0)
4264 goto out_errno;
4265
4266 if (header->needs_swap) {
4267 f_attr.ids.size = bswap_64(f_attr.ids.size);
4268 f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4269 perf_event__attr_swap(&f_attr.attr);
4270 }
4271
4272 tmp = lseek(fd, 0, SEEK_CUR);
4273 evsel = evsel__new(&f_attr.attr);
4274
4275 if (evsel == NULL)
4276 goto out_delete_evlist;
4277
4278 evsel->needs_swap = header->needs_swap;
4279 /*
4280 * Do it before so that if perf_evsel__alloc_id fails, this
4281 * entry gets purged too at evlist__delete().
4282 */
4283 evlist__add(session->evlist, evsel);
4284
4285 nr_ids = f_attr.ids.size / sizeof(u64);
4286 /*
4287 * We don't have the cpu and thread maps on the header, so
4288 * for allocating the perf_sample_id table we fake 1 cpu and
4289 * hattr->ids threads.
4290 */
4291 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4292 goto out_delete_evlist;
4293
4294 lseek(fd, f_attr.ids.offset, SEEK_SET);
4295
4296 for (j = 0; j < nr_ids; j++) {
4297 if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4298 goto out_errno;
4299
4300 perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4301 }
4302
4303 lseek(fd, tmp, SEEK_SET);
4304 }
4305
4306 #ifdef HAVE_LIBTRACEEVENT
4307 perf_header__process_sections(header, fd, &session->tevent,
4308 perf_file_section__process);
4309
4310 if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4311 goto out_delete_evlist;
4312 #else
4313 perf_header__process_sections(header, fd, NULL, perf_file_section__process);
4314 #endif
4315
4316 return 0;
4317 out_errno:
4318 return -errno;
4319
4320 out_delete_evlist:
4321 evlist__delete(session->evlist);
4322 session->evlist = NULL;
4323 return -ENOMEM;
4324 }
4325
4326 int perf_event__process_feature(struct perf_session *session,
4327 union perf_event *event)
4328 {
4329 struct perf_tool *tool = session->tool;
4330 struct feat_fd ff = { .fd = 0 };
4331 struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4332 int type = fe->header.type;
4333 u64 feat = fe->feat_id;
4334 int ret = 0;
4335
4336 if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4337 pr_warning("invalid record type %d in pipe-mode\n", type);
4338 return 0;
4339 }
4340 if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4341 pr_warning("invalid record type %d in pipe-mode\n", type);
4342 return -1;
4343 }
4344
4345 if (!feat_ops[feat].process)
4346 return 0;
4347
4348 ff.buf = (void *)fe->data;
4349 ff.size = event->header.size - sizeof(*fe);
4350 ff.ph = &session->header;
4351
4352 if (feat_ops[feat].process(&ff, NULL)) {
4353 ret = -1;
4354 goto out;
4355 }
4356
4357 if (!feat_ops[feat].print || !tool->show_feat_hdr)
4358 goto out;
4359
4360 if (!feat_ops[feat].full_only ||
4361 tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4362 feat_ops[feat].print(&ff, stdout);
4363 } else {
4364 fprintf(stdout, "# %s info available, use -I to display\n",
4365 feat_ops[feat].name);
4366 }
4367 out:
4368 free_event_desc(ff.events);
4369 return ret;
4370 }
4371
4372 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4373 {
4374 struct perf_record_event_update *ev = &event->event_update;
4375 struct perf_cpu_map *map;
4376 size_t ret;
4377
4378 ret = fprintf(fp, "\n... id: %" PRI_lu64 "\n", ev->id);
4379
4380 switch (ev->type) {
4381 case PERF_EVENT_UPDATE__SCALE:
4382 ret += fprintf(fp, "... scale: %f\n", ev->scale.scale);
4383 break;
4384 case PERF_EVENT_UPDATE__UNIT:
4385 ret += fprintf(fp, "... unit: %s\n", ev->unit);
4386 break;
4387 case PERF_EVENT_UPDATE__NAME:
4388 ret += fprintf(fp, "... name: %s\n", ev->name);
4389 break;
4390 case PERF_EVENT_UPDATE__CPUS:
4391 ret += fprintf(fp, "... ");
4392
4393 map = cpu_map__new_data(&ev->cpus.cpus);
4394 if (map) {
4395 ret += cpu_map__fprintf(map, fp);
4396 perf_cpu_map__put(map);
4397 } else
4398 ret += fprintf(fp, "failed to get cpus\n");
4399 break;
4400 default:
4401 ret += fprintf(fp, "... unknown type\n");
4402 break;
4403 }
4404
4405 return ret;
4406 }
4407
4408 int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4409 union perf_event *event,
4410 struct evlist **pevlist)
4411 {
4412 u32 i, n_ids;
4413 u64 *ids;
4414 struct evsel *evsel;
4415 struct evlist *evlist = *pevlist;
4416
4417 if (evlist == NULL) {
4418 *pevlist = evlist = evlist__new();
4419 if (evlist == NULL)
4420 return -ENOMEM;
4421 }
4422
4423 evsel = evsel__new(&event->attr.attr);
4424 if (evsel == NULL)
4425 return -ENOMEM;
4426
4427 evlist__add(evlist, evsel);
4428
4429 n_ids = event->header.size - sizeof(event->header) - event->attr.attr.size;
4430 n_ids = n_ids / sizeof(u64);
4431 /*
4432 * We don't have the cpu and thread maps on the header, so
4433 * for allocating the perf_sample_id table we fake 1 cpu and
4434 * hattr->ids threads.
4435 */
4436 if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4437 return -ENOMEM;
4438
4439 ids = perf_record_header_attr_id(event);
4440 for (i = 0; i < n_ids; i++) {
4441 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, ids[i]);
4442 }
4443
4444 return 0;
4445 }
4446
4447 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4448 union perf_event *event,
4449 struct evlist **pevlist)
4450 {
4451 struct perf_record_event_update *ev = &event->event_update;
4452 struct evlist *evlist;
4453 struct evsel *evsel;
4454 struct perf_cpu_map *map;
4455
4456 if (dump_trace)
4457 perf_event__fprintf_event_update(event, stdout);
4458
4459 if (!pevlist || *pevlist == NULL)
4460 return -EINVAL;
4461
4462 evlist = *pevlist;
4463
4464 evsel = evlist__id2evsel(evlist, ev->id);
4465 if (evsel == NULL)
4466 return -EINVAL;
4467
4468 switch (ev->type) {
4469 case PERF_EVENT_UPDATE__UNIT:
4470 free((char *)evsel->unit);
4471 evsel->unit = strdup(ev->unit);
4472 break;
4473 case PERF_EVENT_UPDATE__NAME:
4474 free(evsel->name);
4475 evsel->name = strdup(ev->name);
4476 break;
4477 case PERF_EVENT_UPDATE__SCALE:
4478 evsel->scale = ev->scale.scale;
4479 break;
4480 case PERF_EVENT_UPDATE__CPUS:
4481 map = cpu_map__new_data(&ev->cpus.cpus);
4482 if (map) {
4483 perf_cpu_map__put(evsel->core.own_cpus);
4484 evsel->core.own_cpus = map;
4485 } else
4486 pr_err("failed to get event_update cpus\n");
4487 default:
4488 break;
4489 }
4490
4491 return 0;
4492 }
4493
4494 #ifdef HAVE_LIBTRACEEVENT
4495 int perf_event__process_tracing_data(struct perf_session *session,
4496 union perf_event *event)
4497 {
4498 ssize_t size_read, padding, size = event->tracing_data.size;
4499 int fd = perf_data__fd(session->data);
4500 char buf[BUFSIZ];
4501
4502 /*
4503 * The pipe fd is already in proper place and in any case
4504 * we can't move it, and we'd screw the case where we read
4505 * 'pipe' data from regular file. The trace_report reads
4506 * data from 'fd' so we need to set it directly behind the
4507 * event, where the tracing data starts.
4508 */
4509 if (!perf_data__is_pipe(session->data)) {
4510 off_t offset = lseek(fd, 0, SEEK_CUR);
4511
4512 /* setup for reading amidst mmap */
4513 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4514 SEEK_SET);
4515 }
4516
4517 size_read = trace_report(fd, &session->tevent,
4518 session->repipe);
4519 padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4520
4521 if (readn(fd, buf, padding) < 0) {
4522 pr_err("%s: reading input file", __func__);
4523 return -1;
4524 }
4525 if (session->repipe) {
4526 int retw = write(STDOUT_FILENO, buf, padding);
4527 if (retw <= 0 || retw != padding) {
4528 pr_err("%s: repiping tracing data padding", __func__);
4529 return -1;
4530 }
4531 }
4532
4533 if (size_read + padding != size) {
4534 pr_err("%s: tracing data size mismatch", __func__);
4535 return -1;
4536 }
4537
4538 evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4539
4540 return size_read + padding;
4541 }
4542 #endif
4543
4544 int perf_event__process_build_id(struct perf_session *session,
4545 union perf_event *event)
4546 {
4547 __event_process_build_id(&event->build_id,
4548 event->build_id.filename,
4549 session);
4550 return 0;
4551 }
4552
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.