Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <errno.h>
3#include <inttypes.h>
4#include "util.h"
5#include "string2.h"
6#include <sys/param.h>
7#include <sys/types.h>
8#include <byteswap.h>
9#include <unistd.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/stringify.h>
17#include <sys/stat.h>
18#include <sys/utsname.h>
19#include <linux/time64.h>
20#include <dirent.h>
21
22#include "evlist.h"
23#include "evsel.h"
24#include "header.h"
25#include "memswap.h"
26#include "../perf.h"
27#include "trace-event.h"
28#include "session.h"
29#include "symbol.h"
30#include "debug.h"
31#include "cpumap.h"
32#include "pmu.h"
33#include "vdso.h"
34#include "strbuf.h"
35#include "build-id.h"
36#include "data.h"
37#include <api/fs/fs.h>
38#include "asm/bug.h"
39#include "tool.h"
40#include "time-utils.h"
41#include "units.h"
42
43#include "sane_ctype.h"
44
45/*
46 * magic2 = "PERFILE2"
47 * must be a numerical value to let the endianness
48 * determine the memory layout. That way we are able
49 * to detect endianness when reading the perf.data file
50 * back.
51 *
52 * we check for legacy (PERFFILE) format.
53 */
54static const char *__perf_magic1 = "PERFFILE";
55static const u64 __perf_magic2 = 0x32454c4946524550ULL;
56static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
57
58#define PERF_MAGIC __perf_magic2
59
60const char perf_version_string[] = PERF_VERSION;
61
62struct perf_file_attr {
63 struct perf_event_attr attr;
64 struct perf_file_section ids;
65};
66
67struct feat_fd {
68 struct perf_header *ph;
69 int fd;
70 void *buf; /* Either buf != NULL or fd >= 0 */
71 ssize_t offset;
72 size_t size;
73 struct perf_evsel *events;
74};
75
76void perf_header__set_feat(struct perf_header *header, int feat)
77{
78 set_bit(feat, header->adds_features);
79}
80
81void perf_header__clear_feat(struct perf_header *header, int feat)
82{
83 clear_bit(feat, header->adds_features);
84}
85
86bool perf_header__has_feat(const struct perf_header *header, int feat)
87{
88 return test_bit(feat, header->adds_features);
89}
90
91static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
92{
93 ssize_t ret = writen(ff->fd, buf, size);
94
95 if (ret != (ssize_t)size)
96 return ret < 0 ? (int)ret : -1;
97 return 0;
98}
99
100static int __do_write_buf(struct feat_fd *ff, const void *buf, size_t size)
101{
102 /* struct perf_event_header::size is u16 */
103 const size_t max_size = 0xffff - sizeof(struct perf_event_header);
104 size_t new_size = ff->size;
105 void *addr;
106
107 if (size + ff->offset > max_size)
108 return -E2BIG;
109
110 while (size > (new_size - ff->offset))
111 new_size <<= 1;
112 new_size = min(max_size, new_size);
113
114 if (ff->size < new_size) {
115 addr = realloc(ff->buf, new_size);
116 if (!addr)
117 return -ENOMEM;
118 ff->buf = addr;
119 ff->size = new_size;
120 }
121
122 memcpy(ff->buf + ff->offset, buf, size);
123 ff->offset += size;
124
125 return 0;
126}
127
128/* Return: 0 if succeded, -ERR if failed. */
129int do_write(struct feat_fd *ff, const void *buf, size_t size)
130{
131 if (!ff->buf)
132 return __do_write_fd(ff, buf, size);
133 return __do_write_buf(ff, buf, size);
134}
135
136/* Return: 0 if succeded, -ERR if failed. */
137static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
138{
139 u64 *p = (u64 *) set;
140 int i, ret;
141
142 ret = do_write(ff, &size, sizeof(size));
143 if (ret < 0)
144 return ret;
145
146 for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
147 ret = do_write(ff, p + i, sizeof(*p));
148 if (ret < 0)
149 return ret;
150 }
151
152 return 0;
153}
154
155/* Return: 0 if succeded, -ERR if failed. */
156int write_padded(struct feat_fd *ff, const void *bf,
157 size_t count, size_t count_aligned)
158{
159 static const char zero_buf[NAME_ALIGN];
160 int err = do_write(ff, bf, count);
161
162 if (!err)
163 err = do_write(ff, zero_buf, count_aligned - count);
164
165 return err;
166}
167
168#define string_size(str) \
169 (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
170
171/* Return: 0 if succeded, -ERR if failed. */
172static int do_write_string(struct feat_fd *ff, const char *str)
173{
174 u32 len, olen;
175 int ret;
176
177 olen = strlen(str) + 1;
178 len = PERF_ALIGN(olen, NAME_ALIGN);
179
180 /* write len, incl. \0 */
181 ret = do_write(ff, &len, sizeof(len));
182 if (ret < 0)
183 return ret;
184
185 return write_padded(ff, str, olen, len);
186}
187
188static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
189{
190 ssize_t ret = readn(ff->fd, addr, size);
191
192 if (ret != size)
193 return ret < 0 ? (int)ret : -1;
194 return 0;
195}
196
197static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
198{
199 if (size > (ssize_t)ff->size - ff->offset)
200 return -1;
201
202 memcpy(addr, ff->buf + ff->offset, size);
203 ff->offset += size;
204
205 return 0;
206
207}
208
209static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
210{
211 if (!ff->buf)
212 return __do_read_fd(ff, addr, size);
213 return __do_read_buf(ff, addr, size);
214}
215
216static int do_read_u32(struct feat_fd *ff, u32 *addr)
217{
218 int ret;
219
220 ret = __do_read(ff, addr, sizeof(*addr));
221 if (ret)
222 return ret;
223
224 if (ff->ph->needs_swap)
225 *addr = bswap_32(*addr);
226 return 0;
227}
228
229static int do_read_u64(struct feat_fd *ff, u64 *addr)
230{
231 int ret;
232
233 ret = __do_read(ff, addr, sizeof(*addr));
234 if (ret)
235 return ret;
236
237 if (ff->ph->needs_swap)
238 *addr = bswap_64(*addr);
239 return 0;
240}
241
242static char *do_read_string(struct feat_fd *ff)
243{
244 u32 len;
245 char *buf;
246
247 if (do_read_u32(ff, &len))
248 return NULL;
249
250 buf = malloc(len);
251 if (!buf)
252 return NULL;
253
254 if (!__do_read(ff, buf, len)) {
255 /*
256 * strings are padded by zeroes
257 * thus the actual strlen of buf
258 * may be less than len
259 */
260 return buf;
261 }
262
263 free(buf);
264 return NULL;
265}
266
267/* Return: 0 if succeded, -ERR if failed. */
268static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
269{
270 unsigned long *set;
271 u64 size, *p;
272 int i, ret;
273
274 ret = do_read_u64(ff, &size);
275 if (ret)
276 return ret;
277
278 set = bitmap_alloc(size);
279 if (!set)
280 return -ENOMEM;
281
282 bitmap_zero(set, size);
283
284 p = (u64 *) set;
285
286 for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
287 ret = do_read_u64(ff, p + i);
288 if (ret < 0) {
289 free(set);
290 return ret;
291 }
292 }
293
294 *pset = set;
295 *psize = size;
296 return 0;
297}
298
299static int write_tracing_data(struct feat_fd *ff,
300 struct perf_evlist *evlist)
301{
302 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
303 return -1;
304
305 return read_tracing_data(ff->fd, &evlist->entries);
306}
307
308static int write_build_id(struct feat_fd *ff,
309 struct perf_evlist *evlist __maybe_unused)
310{
311 struct perf_session *session;
312 int err;
313
314 session = container_of(ff->ph, struct perf_session, header);
315
316 if (!perf_session__read_build_ids(session, true))
317 return -1;
318
319 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
320 return -1;
321
322 err = perf_session__write_buildid_table(session, ff);
323 if (err < 0) {
324 pr_debug("failed to write buildid table\n");
325 return err;
326 }
327 perf_session__cache_build_ids(session);
328
329 return 0;
330}
331
332static int write_hostname(struct feat_fd *ff,
333 struct perf_evlist *evlist __maybe_unused)
334{
335 struct utsname uts;
336 int ret;
337
338 ret = uname(&uts);
339 if (ret < 0)
340 return -1;
341
342 return do_write_string(ff, uts.nodename);
343}
344
345static int write_osrelease(struct feat_fd *ff,
346 struct perf_evlist *evlist __maybe_unused)
347{
348 struct utsname uts;
349 int ret;
350
351 ret = uname(&uts);
352 if (ret < 0)
353 return -1;
354
355 return do_write_string(ff, uts.release);
356}
357
358static int write_arch(struct feat_fd *ff,
359 struct perf_evlist *evlist __maybe_unused)
360{
361 struct utsname uts;
362 int ret;
363
364 ret = uname(&uts);
365 if (ret < 0)
366 return -1;
367
368 return do_write_string(ff, uts.machine);
369}
370
371static int write_version(struct feat_fd *ff,
372 struct perf_evlist *evlist __maybe_unused)
373{
374 return do_write_string(ff, perf_version_string);
375}
376
377static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
378{
379 FILE *file;
380 char *buf = NULL;
381 char *s, *p;
382 const char *search = cpuinfo_proc;
383 size_t len = 0;
384 int ret = -1;
385
386 if (!search)
387 return -1;
388
389 file = fopen("/proc/cpuinfo", "r");
390 if (!file)
391 return -1;
392
393 while (getline(&buf, &len, file) > 0) {
394 ret = strncmp(buf, search, strlen(search));
395 if (!ret)
396 break;
397 }
398
399 if (ret) {
400 ret = -1;
401 goto done;
402 }
403
404 s = buf;
405
406 p = strchr(buf, ':');
407 if (p && *(p+1) == ' ' && *(p+2))
408 s = p + 2;
409 p = strchr(s, '\n');
410 if (p)
411 *p = '\0';
412
413 /* squash extra space characters (branding string) */
414 p = s;
415 while (*p) {
416 if (isspace(*p)) {
417 char *r = p + 1;
418 char *q = r;
419 *p = ' ';
420 while (*q && isspace(*q))
421 q++;
422 if (q != (p+1))
423 while ((*r++ = *q++));
424 }
425 p++;
426 }
427 ret = do_write_string(ff, s);
428done:
429 free(buf);
430 fclose(file);
431 return ret;
432}
433
434static int write_cpudesc(struct feat_fd *ff,
435 struct perf_evlist *evlist __maybe_unused)
436{
437 const char *cpuinfo_procs[] = CPUINFO_PROC;
438 unsigned int i;
439
440 for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
441 int ret;
442 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
443 if (ret >= 0)
444 return ret;
445 }
446 return -1;
447}
448
449
450static int write_nrcpus(struct feat_fd *ff,
451 struct perf_evlist *evlist __maybe_unused)
452{
453 long nr;
454 u32 nrc, nra;
455 int ret;
456
457 nrc = cpu__max_present_cpu();
458
459 nr = sysconf(_SC_NPROCESSORS_ONLN);
460 if (nr < 0)
461 return -1;
462
463 nra = (u32)(nr & UINT_MAX);
464
465 ret = do_write(ff, &nrc, sizeof(nrc));
466 if (ret < 0)
467 return ret;
468
469 return do_write(ff, &nra, sizeof(nra));
470}
471
472static int write_event_desc(struct feat_fd *ff,
473 struct perf_evlist *evlist)
474{
475 struct perf_evsel *evsel;
476 u32 nre, nri, sz;
477 int ret;
478
479 nre = evlist->nr_entries;
480
481 /*
482 * write number of events
483 */
484 ret = do_write(ff, &nre, sizeof(nre));
485 if (ret < 0)
486 return ret;
487
488 /*
489 * size of perf_event_attr struct
490 */
491 sz = (u32)sizeof(evsel->attr);
492 ret = do_write(ff, &sz, sizeof(sz));
493 if (ret < 0)
494 return ret;
495
496 evlist__for_each_entry(evlist, evsel) {
497 ret = do_write(ff, &evsel->attr, sz);
498 if (ret < 0)
499 return ret;
500 /*
501 * write number of unique id per event
502 * there is one id per instance of an event
503 *
504 * copy into an nri to be independent of the
505 * type of ids,
506 */
507 nri = evsel->ids;
508 ret = do_write(ff, &nri, sizeof(nri));
509 if (ret < 0)
510 return ret;
511
512 /*
513 * write event string as passed on cmdline
514 */
515 ret = do_write_string(ff, perf_evsel__name(evsel));
516 if (ret < 0)
517 return ret;
518 /*
519 * write unique ids for this event
520 */
521 ret = do_write(ff, evsel->id, evsel->ids * sizeof(u64));
522 if (ret < 0)
523 return ret;
524 }
525 return 0;
526}
527
528static int write_cmdline(struct feat_fd *ff,
529 struct perf_evlist *evlist __maybe_unused)
530{
531 char buf[MAXPATHLEN];
532 u32 n;
533 int i, ret;
534
535 /* actual path to perf binary */
536 ret = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
537 if (ret <= 0)
538 return -1;
539
540 /* readlink() does not add null termination */
541 buf[ret] = '\0';
542
543 /* account for binary path */
544 n = perf_env.nr_cmdline + 1;
545
546 ret = do_write(ff, &n, sizeof(n));
547 if (ret < 0)
548 return ret;
549
550 ret = do_write_string(ff, buf);
551 if (ret < 0)
552 return ret;
553
554 for (i = 0 ; i < perf_env.nr_cmdline; i++) {
555 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
556 if (ret < 0)
557 return ret;
558 }
559 return 0;
560}
561
562#define CORE_SIB_FMT \
563 "/sys/devices/system/cpu/cpu%d/topology/core_siblings_list"
564#define THRD_SIB_FMT \
565 "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list"
566
567struct cpu_topo {
568 u32 cpu_nr;
569 u32 core_sib;
570 u32 thread_sib;
571 char **core_siblings;
572 char **thread_siblings;
573};
574
575static int build_cpu_topo(struct cpu_topo *tp, int cpu)
576{
577 FILE *fp;
578 char filename[MAXPATHLEN];
579 char *buf = NULL, *p;
580 size_t len = 0;
581 ssize_t sret;
582 u32 i = 0;
583 int ret = -1;
584
585 sprintf(filename, CORE_SIB_FMT, cpu);
586 fp = fopen(filename, "r");
587 if (!fp)
588 goto try_threads;
589
590 sret = getline(&buf, &len, fp);
591 fclose(fp);
592 if (sret <= 0)
593 goto try_threads;
594
595 p = strchr(buf, '\n');
596 if (p)
597 *p = '\0';
598
599 for (i = 0; i < tp->core_sib; i++) {
600 if (!strcmp(buf, tp->core_siblings[i]))
601 break;
602 }
603 if (i == tp->core_sib) {
604 tp->core_siblings[i] = buf;
605 tp->core_sib++;
606 buf = NULL;
607 len = 0;
608 }
609 ret = 0;
610
611try_threads:
612 sprintf(filename, THRD_SIB_FMT, cpu);
613 fp = fopen(filename, "r");
614 if (!fp)
615 goto done;
616
617 if (getline(&buf, &len, fp) <= 0)
618 goto done;
619
620 p = strchr(buf, '\n');
621 if (p)
622 *p = '\0';
623
624 for (i = 0; i < tp->thread_sib; i++) {
625 if (!strcmp(buf, tp->thread_siblings[i]))
626 break;
627 }
628 if (i == tp->thread_sib) {
629 tp->thread_siblings[i] = buf;
630 tp->thread_sib++;
631 buf = NULL;
632 }
633 ret = 0;
634done:
635 if(fp)
636 fclose(fp);
637 free(buf);
638 return ret;
639}
640
641static void free_cpu_topo(struct cpu_topo *tp)
642{
643 u32 i;
644
645 if (!tp)
646 return;
647
648 for (i = 0 ; i < tp->core_sib; i++)
649 zfree(&tp->core_siblings[i]);
650
651 for (i = 0 ; i < tp->thread_sib; i++)
652 zfree(&tp->thread_siblings[i]);
653
654 free(tp);
655}
656
657static struct cpu_topo *build_cpu_topology(void)
658{
659 struct cpu_topo *tp = NULL;
660 void *addr;
661 u32 nr, i;
662 size_t sz;
663 long ncpus;
664 int ret = -1;
665 struct cpu_map *map;
666
667 ncpus = cpu__max_present_cpu();
668
669 /* build online CPU map */
670 map = cpu_map__new(NULL);
671 if (map == NULL) {
672 pr_debug("failed to get system cpumap\n");
673 return NULL;
674 }
675
676 nr = (u32)(ncpus & UINT_MAX);
677
678 sz = nr * sizeof(char *);
679 addr = calloc(1, sizeof(*tp) + 2 * sz);
680 if (!addr)
681 goto out_free;
682
683 tp = addr;
684 tp->cpu_nr = nr;
685 addr += sizeof(*tp);
686 tp->core_siblings = addr;
687 addr += sz;
688 tp->thread_siblings = addr;
689
690 for (i = 0; i < nr; i++) {
691 if (!cpu_map__has(map, i))
692 continue;
693
694 ret = build_cpu_topo(tp, i);
695 if (ret < 0)
696 break;
697 }
698
699out_free:
700 cpu_map__put(map);
701 if (ret) {
702 free_cpu_topo(tp);
703 tp = NULL;
704 }
705 return tp;
706}
707
708static int write_cpu_topology(struct feat_fd *ff,
709 struct perf_evlist *evlist __maybe_unused)
710{
711 struct cpu_topo *tp;
712 u32 i;
713 int ret, j;
714
715 tp = build_cpu_topology();
716 if (!tp)
717 return -1;
718
719 ret = do_write(ff, &tp->core_sib, sizeof(tp->core_sib));
720 if (ret < 0)
721 goto done;
722
723 for (i = 0; i < tp->core_sib; i++) {
724 ret = do_write_string(ff, tp->core_siblings[i]);
725 if (ret < 0)
726 goto done;
727 }
728 ret = do_write(ff, &tp->thread_sib, sizeof(tp->thread_sib));
729 if (ret < 0)
730 goto done;
731
732 for (i = 0; i < tp->thread_sib; i++) {
733 ret = do_write_string(ff, tp->thread_siblings[i]);
734 if (ret < 0)
735 break;
736 }
737
738 ret = perf_env__read_cpu_topology_map(&perf_env);
739 if (ret < 0)
740 goto done;
741
742 for (j = 0; j < perf_env.nr_cpus_avail; j++) {
743 ret = do_write(ff, &perf_env.cpu[j].core_id,
744 sizeof(perf_env.cpu[j].core_id));
745 if (ret < 0)
746 return ret;
747 ret = do_write(ff, &perf_env.cpu[j].socket_id,
748 sizeof(perf_env.cpu[j].socket_id));
749 if (ret < 0)
750 return ret;
751 }
752done:
753 free_cpu_topo(tp);
754 return ret;
755}
756
757
758
759static int write_total_mem(struct feat_fd *ff,
760 struct perf_evlist *evlist __maybe_unused)
761{
762 char *buf = NULL;
763 FILE *fp;
764 size_t len = 0;
765 int ret = -1, n;
766 uint64_t mem;
767
768 fp = fopen("/proc/meminfo", "r");
769 if (!fp)
770 return -1;
771
772 while (getline(&buf, &len, fp) > 0) {
773 ret = strncmp(buf, "MemTotal:", 9);
774 if (!ret)
775 break;
776 }
777 if (!ret) {
778 n = sscanf(buf, "%*s %"PRIu64, &mem);
779 if (n == 1)
780 ret = do_write(ff, &mem, sizeof(mem));
781 } else
782 ret = -1;
783 free(buf);
784 fclose(fp);
785 return ret;
786}
787
788static int write_topo_node(struct feat_fd *ff, int node)
789{
790 char str[MAXPATHLEN];
791 char field[32];
792 char *buf = NULL, *p;
793 size_t len = 0;
794 FILE *fp;
795 u64 mem_total, mem_free, mem;
796 int ret = -1;
797
798 sprintf(str, "/sys/devices/system/node/node%d/meminfo", node);
799 fp = fopen(str, "r");
800 if (!fp)
801 return -1;
802
803 while (getline(&buf, &len, fp) > 0) {
804 /* skip over invalid lines */
805 if (!strchr(buf, ':'))
806 continue;
807 if (sscanf(buf, "%*s %*d %31s %"PRIu64, field, &mem) != 2)
808 goto done;
809 if (!strcmp(field, "MemTotal:"))
810 mem_total = mem;
811 if (!strcmp(field, "MemFree:"))
812 mem_free = mem;
813 }
814
815 fclose(fp);
816 fp = NULL;
817
818 ret = do_write(ff, &mem_total, sizeof(u64));
819 if (ret)
820 goto done;
821
822 ret = do_write(ff, &mem_free, sizeof(u64));
823 if (ret)
824 goto done;
825
826 ret = -1;
827 sprintf(str, "/sys/devices/system/node/node%d/cpulist", node);
828
829 fp = fopen(str, "r");
830 if (!fp)
831 goto done;
832
833 if (getline(&buf, &len, fp) <= 0)
834 goto done;
835
836 p = strchr(buf, '\n');
837 if (p)
838 *p = '\0';
839
840 ret = do_write_string(ff, buf);
841done:
842 free(buf);
843 if (fp)
844 fclose(fp);
845 return ret;
846}
847
848static int write_numa_topology(struct feat_fd *ff,
849 struct perf_evlist *evlist __maybe_unused)
850{
851 char *buf = NULL;
852 size_t len = 0;
853 FILE *fp;
854 struct cpu_map *node_map = NULL;
855 char *c;
856 u32 nr, i, j;
857 int ret = -1;
858
859 fp = fopen("/sys/devices/system/node/online", "r");
860 if (!fp)
861 return -1;
862
863 if (getline(&buf, &len, fp) <= 0)
864 goto done;
865
866 c = strchr(buf, '\n');
867 if (c)
868 *c = '\0';
869
870 node_map = cpu_map__new(buf);
871 if (!node_map)
872 goto done;
873
874 nr = (u32)node_map->nr;
875
876 ret = do_write(ff, &nr, sizeof(nr));
877 if (ret < 0)
878 goto done;
879
880 for (i = 0; i < nr; i++) {
881 j = (u32)node_map->map[i];
882 ret = do_write(ff, &j, sizeof(j));
883 if (ret < 0)
884 break;
885
886 ret = write_topo_node(ff, i);
887 if (ret < 0)
888 break;
889 }
890done:
891 free(buf);
892 fclose(fp);
893 cpu_map__put(node_map);
894 return ret;
895}
896
897/*
898 * File format:
899 *
900 * struct pmu_mappings {
901 * u32 pmu_num;
902 * struct pmu_map {
903 * u32 type;
904 * char name[];
905 * }[pmu_num];
906 * };
907 */
908
909static int write_pmu_mappings(struct feat_fd *ff,
910 struct perf_evlist *evlist __maybe_unused)
911{
912 struct perf_pmu *pmu = NULL;
913 u32 pmu_num = 0;
914 int ret;
915
916 /*
917 * Do a first pass to count number of pmu to avoid lseek so this
918 * works in pipe mode as well.
919 */
920 while ((pmu = perf_pmu__scan(pmu))) {
921 if (!pmu->name)
922 continue;
923 pmu_num++;
924 }
925
926 ret = do_write(ff, &pmu_num, sizeof(pmu_num));
927 if (ret < 0)
928 return ret;
929
930 while ((pmu = perf_pmu__scan(pmu))) {
931 if (!pmu->name)
932 continue;
933
934 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
935 if (ret < 0)
936 return ret;
937
938 ret = do_write_string(ff, pmu->name);
939 if (ret < 0)
940 return ret;
941 }
942
943 return 0;
944}
945
946/*
947 * File format:
948 *
949 * struct group_descs {
950 * u32 nr_groups;
951 * struct group_desc {
952 * char name[];
953 * u32 leader_idx;
954 * u32 nr_members;
955 * }[nr_groups];
956 * };
957 */
958static int write_group_desc(struct feat_fd *ff,
959 struct perf_evlist *evlist)
960{
961 u32 nr_groups = evlist->nr_groups;
962 struct perf_evsel *evsel;
963 int ret;
964
965 ret = do_write(ff, &nr_groups, sizeof(nr_groups));
966 if (ret < 0)
967 return ret;
968
969 evlist__for_each_entry(evlist, evsel) {
970 if (perf_evsel__is_group_leader(evsel) &&
971 evsel->nr_members > 1) {
972 const char *name = evsel->group_name ?: "{anon_group}";
973 u32 leader_idx = evsel->idx;
974 u32 nr_members = evsel->nr_members;
975
976 ret = do_write_string(ff, name);
977 if (ret < 0)
978 return ret;
979
980 ret = do_write(ff, &leader_idx, sizeof(leader_idx));
981 if (ret < 0)
982 return ret;
983
984 ret = do_write(ff, &nr_members, sizeof(nr_members));
985 if (ret < 0)
986 return ret;
987 }
988 }
989 return 0;
990}
991
992/*
993 * default get_cpuid(): nothing gets recorded
994 * actual implementation must be in arch/$(SRCARCH)/util/header.c
995 */
996int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
997{
998 return -1;
999}
1000
1001static int write_cpuid(struct feat_fd *ff,
1002 struct perf_evlist *evlist __maybe_unused)
1003{
1004 char buffer[64];
1005 int ret;
1006
1007 ret = get_cpuid(buffer, sizeof(buffer));
1008 if (!ret)
1009 goto write_it;
1010
1011 return -1;
1012write_it:
1013 return do_write_string(ff, buffer);
1014}
1015
1016static int write_branch_stack(struct feat_fd *ff __maybe_unused,
1017 struct perf_evlist *evlist __maybe_unused)
1018{
1019 return 0;
1020}
1021
1022static int write_auxtrace(struct feat_fd *ff,
1023 struct perf_evlist *evlist __maybe_unused)
1024{
1025 struct perf_session *session;
1026 int err;
1027
1028 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
1029 return -1;
1030
1031 session = container_of(ff->ph, struct perf_session, header);
1032
1033 err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
1034 if (err < 0)
1035 pr_err("Failed to write auxtrace index\n");
1036 return err;
1037}
1038
1039static int cpu_cache_level__sort(const void *a, const void *b)
1040{
1041 struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1042 struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1043
1044 return cache_a->level - cache_b->level;
1045}
1046
1047static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1048{
1049 if (a->level != b->level)
1050 return false;
1051
1052 if (a->line_size != b->line_size)
1053 return false;
1054
1055 if (a->sets != b->sets)
1056 return false;
1057
1058 if (a->ways != b->ways)
1059 return false;
1060
1061 if (strcmp(a->type, b->type))
1062 return false;
1063
1064 if (strcmp(a->size, b->size))
1065 return false;
1066
1067 if (strcmp(a->map, b->map))
1068 return false;
1069
1070 return true;
1071}
1072
1073static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1074{
1075 char path[PATH_MAX], file[PATH_MAX];
1076 struct stat st;
1077 size_t len;
1078
1079 scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1080 scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1081
1082 if (stat(file, &st))
1083 return 1;
1084
1085 scnprintf(file, PATH_MAX, "%s/level", path);
1086 if (sysfs__read_int(file, (int *) &cache->level))
1087 return -1;
1088
1089 scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1090 if (sysfs__read_int(file, (int *) &cache->line_size))
1091 return -1;
1092
1093 scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1094 if (sysfs__read_int(file, (int *) &cache->sets))
1095 return -1;
1096
1097 scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1098 if (sysfs__read_int(file, (int *) &cache->ways))
1099 return -1;
1100
1101 scnprintf(file, PATH_MAX, "%s/type", path);
1102 if (sysfs__read_str(file, &cache->type, &len))
1103 return -1;
1104
1105 cache->type[len] = 0;
1106 cache->type = rtrim(cache->type);
1107
1108 scnprintf(file, PATH_MAX, "%s/size", path);
1109 if (sysfs__read_str(file, &cache->size, &len)) {
1110 free(cache->type);
1111 return -1;
1112 }
1113
1114 cache->size[len] = 0;
1115 cache->size = rtrim(cache->size);
1116
1117 scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1118 if (sysfs__read_str(file, &cache->map, &len)) {
1119 free(cache->map);
1120 free(cache->type);
1121 return -1;
1122 }
1123
1124 cache->map[len] = 0;
1125 cache->map = rtrim(cache->map);
1126 return 0;
1127}
1128
1129static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1130{
1131 fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1132}
1133
1134static int build_caches(struct cpu_cache_level caches[], u32 size, u32 *cntp)
1135{
1136 u32 i, cnt = 0;
1137 long ncpus;
1138 u32 nr, cpu;
1139 u16 level;
1140
1141 ncpus = sysconf(_SC_NPROCESSORS_CONF);
1142 if (ncpus < 0)
1143 return -1;
1144
1145 nr = (u32)(ncpus & UINT_MAX);
1146
1147 for (cpu = 0; cpu < nr; cpu++) {
1148 for (level = 0; level < 10; level++) {
1149 struct cpu_cache_level c;
1150 int err;
1151
1152 err = cpu_cache_level__read(&c, cpu, level);
1153 if (err < 0)
1154 return err;
1155
1156 if (err == 1)
1157 break;
1158
1159 for (i = 0; i < cnt; i++) {
1160 if (cpu_cache_level__cmp(&c, &caches[i]))
1161 break;
1162 }
1163
1164 if (i == cnt)
1165 caches[cnt++] = c;
1166 else
1167 cpu_cache_level__free(&c);
1168
1169 if (WARN_ONCE(cnt == size, "way too many cpu caches.."))
1170 goto out;
1171 }
1172 }
1173 out:
1174 *cntp = cnt;
1175 return 0;
1176}
1177
1178#define MAX_CACHES 2000
1179
1180static int write_cache(struct feat_fd *ff,
1181 struct perf_evlist *evlist __maybe_unused)
1182{
1183 struct cpu_cache_level caches[MAX_CACHES];
1184 u32 cnt = 0, i, version = 1;
1185 int ret;
1186
1187 ret = build_caches(caches, MAX_CACHES, &cnt);
1188 if (ret)
1189 goto out;
1190
1191 qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1192
1193 ret = do_write(ff, &version, sizeof(u32));
1194 if (ret < 0)
1195 goto out;
1196
1197 ret = do_write(ff, &cnt, sizeof(u32));
1198 if (ret < 0)
1199 goto out;
1200
1201 for (i = 0; i < cnt; i++) {
1202 struct cpu_cache_level *c = &caches[i];
1203
1204 #define _W(v) \
1205 ret = do_write(ff, &c->v, sizeof(u32)); \
1206 if (ret < 0) \
1207 goto out;
1208
1209 _W(level)
1210 _W(line_size)
1211 _W(sets)
1212 _W(ways)
1213 #undef _W
1214
1215 #define _W(v) \
1216 ret = do_write_string(ff, (const char *) c->v); \
1217 if (ret < 0) \
1218 goto out;
1219
1220 _W(type)
1221 _W(size)
1222 _W(map)
1223 #undef _W
1224 }
1225
1226out:
1227 for (i = 0; i < cnt; i++)
1228 cpu_cache_level__free(&caches[i]);
1229 return ret;
1230}
1231
1232static int write_stat(struct feat_fd *ff __maybe_unused,
1233 struct perf_evlist *evlist __maybe_unused)
1234{
1235 return 0;
1236}
1237
1238static int write_sample_time(struct feat_fd *ff,
1239 struct perf_evlist *evlist)
1240{
1241 int ret;
1242
1243 ret = do_write(ff, &evlist->first_sample_time,
1244 sizeof(evlist->first_sample_time));
1245 if (ret < 0)
1246 return ret;
1247
1248 return do_write(ff, &evlist->last_sample_time,
1249 sizeof(evlist->last_sample_time));
1250}
1251
1252
1253static int memory_node__read(struct memory_node *n, unsigned long idx)
1254{
1255 unsigned int phys, size = 0;
1256 char path[PATH_MAX];
1257 struct dirent *ent;
1258 DIR *dir;
1259
1260#define for_each_memory(mem, dir) \
1261 while ((ent = readdir(dir))) \
1262 if (strcmp(ent->d_name, ".") && \
1263 strcmp(ent->d_name, "..") && \
1264 sscanf(ent->d_name, "memory%u", &mem) == 1)
1265
1266 scnprintf(path, PATH_MAX,
1267 "%s/devices/system/node/node%lu",
1268 sysfs__mountpoint(), idx);
1269
1270 dir = opendir(path);
1271 if (!dir) {
1272 pr_warning("failed: cant' open memory sysfs data\n");
1273 return -1;
1274 }
1275
1276 for_each_memory(phys, dir) {
1277 size = max(phys, size);
1278 }
1279
1280 size++;
1281
1282 n->set = bitmap_alloc(size);
1283 if (!n->set) {
1284 closedir(dir);
1285 return -ENOMEM;
1286 }
1287
1288 bitmap_zero(n->set, size);
1289 n->node = idx;
1290 n->size = size;
1291
1292 rewinddir(dir);
1293
1294 for_each_memory(phys, dir) {
1295 set_bit(phys, n->set);
1296 }
1297
1298 closedir(dir);
1299 return 0;
1300}
1301
1302static int memory_node__sort(const void *a, const void *b)
1303{
1304 const struct memory_node *na = a;
1305 const struct memory_node *nb = b;
1306
1307 return na->node - nb->node;
1308}
1309
1310static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
1311{
1312 char path[PATH_MAX];
1313 struct dirent *ent;
1314 DIR *dir;
1315 u64 cnt = 0;
1316 int ret = 0;
1317
1318 scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1319 sysfs__mountpoint());
1320
1321 dir = opendir(path);
1322 if (!dir) {
1323 pr_debug2("%s: could't read %s, does this arch have topology information?\n",
1324 __func__, path);
1325 return -1;
1326 }
1327
1328 while (!ret && (ent = readdir(dir))) {
1329 unsigned int idx;
1330 int r;
1331
1332 if (!strcmp(ent->d_name, ".") ||
1333 !strcmp(ent->d_name, ".."))
1334 continue;
1335
1336 r = sscanf(ent->d_name, "node%u", &idx);
1337 if (r != 1)
1338 continue;
1339
1340 if (WARN_ONCE(cnt >= size,
1341 "failed to write MEM_TOPOLOGY, way too many nodes\n"))
1342 return -1;
1343
1344 ret = memory_node__read(&nodes[cnt++], idx);
1345 }
1346
1347 *cntp = cnt;
1348 closedir(dir);
1349
1350 if (!ret)
1351 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1352
1353 return ret;
1354}
1355
1356#define MAX_MEMORY_NODES 2000
1357
1358/*
1359 * The MEM_TOPOLOGY holds physical memory map for every
1360 * node in system. The format of data is as follows:
1361 *
1362 * 0 - version | for future changes
1363 * 8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1364 * 16 - count | number of nodes
1365 *
1366 * For each node we store map of physical indexes for
1367 * each node:
1368 *
1369 * 32 - node id | node index
1370 * 40 - size | size of bitmap
1371 * 48 - bitmap | bitmap of memory indexes that belongs to node
1372 */
1373static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1374 struct perf_evlist *evlist __maybe_unused)
1375{
1376 static struct memory_node nodes[MAX_MEMORY_NODES];
1377 u64 bsize, version = 1, i, nr;
1378 int ret;
1379
1380 ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1381 (unsigned long long *) &bsize);
1382 if (ret)
1383 return ret;
1384
1385 ret = build_mem_topology(&nodes[0], MAX_MEMORY_NODES, &nr);
1386 if (ret)
1387 return ret;
1388
1389 ret = do_write(ff, &version, sizeof(version));
1390 if (ret < 0)
1391 goto out;
1392
1393 ret = do_write(ff, &bsize, sizeof(bsize));
1394 if (ret < 0)
1395 goto out;
1396
1397 ret = do_write(ff, &nr, sizeof(nr));
1398 if (ret < 0)
1399 goto out;
1400
1401 for (i = 0; i < nr; i++) {
1402 struct memory_node *n = &nodes[i];
1403
1404 #define _W(v) \
1405 ret = do_write(ff, &n->v, sizeof(n->v)); \
1406 if (ret < 0) \
1407 goto out;
1408
1409 _W(node)
1410 _W(size)
1411
1412 #undef _W
1413
1414 ret = do_write_bitmap(ff, n->set, n->size);
1415 if (ret < 0)
1416 goto out;
1417 }
1418
1419out:
1420 return ret;
1421}
1422
1423static void print_hostname(struct feat_fd *ff, FILE *fp)
1424{
1425 fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1426}
1427
1428static void print_osrelease(struct feat_fd *ff, FILE *fp)
1429{
1430 fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1431}
1432
1433static void print_arch(struct feat_fd *ff, FILE *fp)
1434{
1435 fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1436}
1437
1438static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1439{
1440 fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1441}
1442
1443static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1444{
1445 fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1446 fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1447}
1448
1449static void print_version(struct feat_fd *ff, FILE *fp)
1450{
1451 fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1452}
1453
1454static void print_cmdline(struct feat_fd *ff, FILE *fp)
1455{
1456 int nr, i;
1457
1458 nr = ff->ph->env.nr_cmdline;
1459
1460 fprintf(fp, "# cmdline : ");
1461
1462 for (i = 0; i < nr; i++)
1463 fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1464 fputc('\n', fp);
1465}
1466
1467static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1468{
1469 struct perf_header *ph = ff->ph;
1470 int cpu_nr = ph->env.nr_cpus_avail;
1471 int nr, i;
1472 char *str;
1473
1474 nr = ph->env.nr_sibling_cores;
1475 str = ph->env.sibling_cores;
1476
1477 for (i = 0; i < nr; i++) {
1478 fprintf(fp, "# sibling cores : %s\n", str);
1479 str += strlen(str) + 1;
1480 }
1481
1482 nr = ph->env.nr_sibling_threads;
1483 str = ph->env.sibling_threads;
1484
1485 for (i = 0; i < nr; i++) {
1486 fprintf(fp, "# sibling threads : %s\n", str);
1487 str += strlen(str) + 1;
1488 }
1489
1490 if (ph->env.cpu != NULL) {
1491 for (i = 0; i < cpu_nr; i++)
1492 fprintf(fp, "# CPU %d: Core ID %d, Socket ID %d\n", i,
1493 ph->env.cpu[i].core_id, ph->env.cpu[i].socket_id);
1494 } else
1495 fprintf(fp, "# Core ID and Socket ID information is not available\n");
1496}
1497
1498static void free_event_desc(struct perf_evsel *events)
1499{
1500 struct perf_evsel *evsel;
1501
1502 if (!events)
1503 return;
1504
1505 for (evsel = events; evsel->attr.size; evsel++) {
1506 zfree(&evsel->name);
1507 zfree(&evsel->id);
1508 }
1509
1510 free(events);
1511}
1512
1513static struct perf_evsel *read_event_desc(struct feat_fd *ff)
1514{
1515 struct perf_evsel *evsel, *events = NULL;
1516 u64 *id;
1517 void *buf = NULL;
1518 u32 nre, sz, nr, i, j;
1519 size_t msz;
1520
1521 /* number of events */
1522 if (do_read_u32(ff, &nre))
1523 goto error;
1524
1525 if (do_read_u32(ff, &sz))
1526 goto error;
1527
1528 /* buffer to hold on file attr struct */
1529 buf = malloc(sz);
1530 if (!buf)
1531 goto error;
1532
1533 /* the last event terminates with evsel->attr.size == 0: */
1534 events = calloc(nre + 1, sizeof(*events));
1535 if (!events)
1536 goto error;
1537
1538 msz = sizeof(evsel->attr);
1539 if (sz < msz)
1540 msz = sz;
1541
1542 for (i = 0, evsel = events; i < nre; evsel++, i++) {
1543 evsel->idx = i;
1544
1545 /*
1546 * must read entire on-file attr struct to
1547 * sync up with layout.
1548 */
1549 if (__do_read(ff, buf, sz))
1550 goto error;
1551
1552 if (ff->ph->needs_swap)
1553 perf_event__attr_swap(buf);
1554
1555 memcpy(&evsel->attr, buf, msz);
1556
1557 if (do_read_u32(ff, &nr))
1558 goto error;
1559
1560 if (ff->ph->needs_swap)
1561 evsel->needs_swap = true;
1562
1563 evsel->name = do_read_string(ff);
1564 if (!evsel->name)
1565 goto error;
1566
1567 if (!nr)
1568 continue;
1569
1570 id = calloc(nr, sizeof(*id));
1571 if (!id)
1572 goto error;
1573 evsel->ids = nr;
1574 evsel->id = id;
1575
1576 for (j = 0 ; j < nr; j++) {
1577 if (do_read_u64(ff, id))
1578 goto error;
1579 id++;
1580 }
1581 }
1582out:
1583 free(buf);
1584 return events;
1585error:
1586 free_event_desc(events);
1587 events = NULL;
1588 goto out;
1589}
1590
1591static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
1592 void *priv __maybe_unused)
1593{
1594 return fprintf(fp, ", %s = %s", name, val);
1595}
1596
1597static void print_event_desc(struct feat_fd *ff, FILE *fp)
1598{
1599 struct perf_evsel *evsel, *events;
1600 u32 j;
1601 u64 *id;
1602
1603 if (ff->events)
1604 events = ff->events;
1605 else
1606 events = read_event_desc(ff);
1607
1608 if (!events) {
1609 fprintf(fp, "# event desc: not available or unable to read\n");
1610 return;
1611 }
1612
1613 for (evsel = events; evsel->attr.size; evsel++) {
1614 fprintf(fp, "# event : name = %s, ", evsel->name);
1615
1616 if (evsel->ids) {
1617 fprintf(fp, ", id = {");
1618 for (j = 0, id = evsel->id; j < evsel->ids; j++, id++) {
1619 if (j)
1620 fputc(',', fp);
1621 fprintf(fp, " %"PRIu64, *id);
1622 }
1623 fprintf(fp, " }");
1624 }
1625
1626 perf_event_attr__fprintf(fp, &evsel->attr, __desc_attr__fprintf, NULL);
1627
1628 fputc('\n', fp);
1629 }
1630
1631 free_event_desc(events);
1632 ff->events = NULL;
1633}
1634
1635static void print_total_mem(struct feat_fd *ff, FILE *fp)
1636{
1637 fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
1638}
1639
1640static void print_numa_topology(struct feat_fd *ff, FILE *fp)
1641{
1642 int i;
1643 struct numa_node *n;
1644
1645 for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
1646 n = &ff->ph->env.numa_nodes[i];
1647
1648 fprintf(fp, "# node%u meminfo : total = %"PRIu64" kB,"
1649 " free = %"PRIu64" kB\n",
1650 n->node, n->mem_total, n->mem_free);
1651
1652 fprintf(fp, "# node%u cpu list : ", n->node);
1653 cpu_map__fprintf(n->map, fp);
1654 }
1655}
1656
1657static void print_cpuid(struct feat_fd *ff, FILE *fp)
1658{
1659 fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
1660}
1661
1662static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
1663{
1664 fprintf(fp, "# contains samples with branch stack\n");
1665}
1666
1667static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
1668{
1669 fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
1670}
1671
1672static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
1673{
1674 fprintf(fp, "# contains stat data\n");
1675}
1676
1677static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
1678{
1679 int i;
1680
1681 fprintf(fp, "# CPU cache info:\n");
1682 for (i = 0; i < ff->ph->env.caches_cnt; i++) {
1683 fprintf(fp, "# ");
1684 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
1685 }
1686}
1687
1688static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
1689{
1690 const char *delimiter = "# pmu mappings: ";
1691 char *str, *tmp;
1692 u32 pmu_num;
1693 u32 type;
1694
1695 pmu_num = ff->ph->env.nr_pmu_mappings;
1696 if (!pmu_num) {
1697 fprintf(fp, "# pmu mappings: not available\n");
1698 return;
1699 }
1700
1701 str = ff->ph->env.pmu_mappings;
1702
1703 while (pmu_num) {
1704 type = strtoul(str, &tmp, 0);
1705 if (*tmp != ':')
1706 goto error;
1707
1708 str = tmp + 1;
1709 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
1710
1711 delimiter = ", ";
1712 str += strlen(str) + 1;
1713 pmu_num--;
1714 }
1715
1716 fprintf(fp, "\n");
1717
1718 if (!pmu_num)
1719 return;
1720error:
1721 fprintf(fp, "# pmu mappings: unable to read\n");
1722}
1723
1724static void print_group_desc(struct feat_fd *ff, FILE *fp)
1725{
1726 struct perf_session *session;
1727 struct perf_evsel *evsel;
1728 u32 nr = 0;
1729
1730 session = container_of(ff->ph, struct perf_session, header);
1731
1732 evlist__for_each_entry(session->evlist, evsel) {
1733 if (perf_evsel__is_group_leader(evsel) &&
1734 evsel->nr_members > 1) {
1735 fprintf(fp, "# group: %s{%s", evsel->group_name ?: "",
1736 perf_evsel__name(evsel));
1737
1738 nr = evsel->nr_members - 1;
1739 } else if (nr) {
1740 fprintf(fp, ",%s", perf_evsel__name(evsel));
1741
1742 if (--nr == 0)
1743 fprintf(fp, "}\n");
1744 }
1745 }
1746}
1747
1748static void print_sample_time(struct feat_fd *ff, FILE *fp)
1749{
1750 struct perf_session *session;
1751 char time_buf[32];
1752 double d;
1753
1754 session = container_of(ff->ph, struct perf_session, header);
1755
1756 timestamp__scnprintf_usec(session->evlist->first_sample_time,
1757 time_buf, sizeof(time_buf));
1758 fprintf(fp, "# time of first sample : %s\n", time_buf);
1759
1760 timestamp__scnprintf_usec(session->evlist->last_sample_time,
1761 time_buf, sizeof(time_buf));
1762 fprintf(fp, "# time of last sample : %s\n", time_buf);
1763
1764 d = (double)(session->evlist->last_sample_time -
1765 session->evlist->first_sample_time) / NSEC_PER_MSEC;
1766
1767 fprintf(fp, "# sample duration : %10.3f ms\n", d);
1768}
1769
1770static void memory_node__fprintf(struct memory_node *n,
1771 unsigned long long bsize, FILE *fp)
1772{
1773 char buf_map[100], buf_size[50];
1774 unsigned long long size;
1775
1776 size = bsize * bitmap_weight(n->set, n->size);
1777 unit_number__scnprintf(buf_size, 50, size);
1778
1779 bitmap_scnprintf(n->set, n->size, buf_map, 100);
1780 fprintf(fp, "# %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
1781}
1782
1783static void print_mem_topology(struct feat_fd *ff, FILE *fp)
1784{
1785 struct memory_node *nodes;
1786 int i, nr;
1787
1788 nodes = ff->ph->env.memory_nodes;
1789 nr = ff->ph->env.nr_memory_nodes;
1790
1791 fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
1792 nr, ff->ph->env.memory_bsize);
1793
1794 for (i = 0; i < nr; i++) {
1795 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
1796 }
1797}
1798
1799static int __event_process_build_id(struct build_id_event *bev,
1800 char *filename,
1801 struct perf_session *session)
1802{
1803 int err = -1;
1804 struct machine *machine;
1805 u16 cpumode;
1806 struct dso *dso;
1807 enum dso_kernel_type dso_type;
1808
1809 machine = perf_session__findnew_machine(session, bev->pid);
1810 if (!machine)
1811 goto out;
1812
1813 cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1814
1815 switch (cpumode) {
1816 case PERF_RECORD_MISC_KERNEL:
1817 dso_type = DSO_TYPE_KERNEL;
1818 break;
1819 case PERF_RECORD_MISC_GUEST_KERNEL:
1820 dso_type = DSO_TYPE_GUEST_KERNEL;
1821 break;
1822 case PERF_RECORD_MISC_USER:
1823 case PERF_RECORD_MISC_GUEST_USER:
1824 dso_type = DSO_TYPE_USER;
1825 break;
1826 default:
1827 goto out;
1828 }
1829
1830 dso = machine__findnew_dso(machine, filename);
1831 if (dso != NULL) {
1832 char sbuild_id[SBUILD_ID_SIZE];
1833
1834 dso__set_build_id(dso, &bev->build_id);
1835
1836 if (dso_type != DSO_TYPE_USER) {
1837 struct kmod_path m = { .name = NULL, };
1838
1839 if (!kmod_path__parse_name(&m, filename) && m.kmod)
1840 dso__set_module_info(dso, &m, machine);
1841 else
1842 dso->kernel = dso_type;
1843
1844 free(m.name);
1845 }
1846
1847 build_id__sprintf(dso->build_id, sizeof(dso->build_id),
1848 sbuild_id);
1849 pr_debug("build id event received for %s: %s\n",
1850 dso->long_name, sbuild_id);
1851 dso__put(dso);
1852 }
1853
1854 err = 0;
1855out:
1856 return err;
1857}
1858
1859static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
1860 int input, u64 offset, u64 size)
1861{
1862 struct perf_session *session = container_of(header, struct perf_session, header);
1863 struct {
1864 struct perf_event_header header;
1865 u8 build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
1866 char filename[0];
1867 } old_bev;
1868 struct build_id_event bev;
1869 char filename[PATH_MAX];
1870 u64 limit = offset + size;
1871
1872 while (offset < limit) {
1873 ssize_t len;
1874
1875 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
1876 return -1;
1877
1878 if (header->needs_swap)
1879 perf_event_header__bswap(&old_bev.header);
1880
1881 len = old_bev.header.size - sizeof(old_bev);
1882 if (readn(input, filename, len) != len)
1883 return -1;
1884
1885 bev.header = old_bev.header;
1886
1887 /*
1888 * As the pid is the missing value, we need to fill
1889 * it properly. The header.misc value give us nice hint.
1890 */
1891 bev.pid = HOST_KERNEL_ID;
1892 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
1893 bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
1894 bev.pid = DEFAULT_GUEST_KERNEL_ID;
1895
1896 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
1897 __event_process_build_id(&bev, filename, session);
1898
1899 offset += bev.header.size;
1900 }
1901
1902 return 0;
1903}
1904
1905static int perf_header__read_build_ids(struct perf_header *header,
1906 int input, u64 offset, u64 size)
1907{
1908 struct perf_session *session = container_of(header, struct perf_session, header);
1909 struct build_id_event bev;
1910 char filename[PATH_MAX];
1911 u64 limit = offset + size, orig_offset = offset;
1912 int err = -1;
1913
1914 while (offset < limit) {
1915 ssize_t len;
1916
1917 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
1918 goto out;
1919
1920 if (header->needs_swap)
1921 perf_event_header__bswap(&bev.header);
1922
1923 len = bev.header.size - sizeof(bev);
1924 if (readn(input, filename, len) != len)
1925 goto out;
1926 /*
1927 * The a1645ce1 changeset:
1928 *
1929 * "perf: 'perf kvm' tool for monitoring guest performance from host"
1930 *
1931 * Added a field to struct build_id_event that broke the file
1932 * format.
1933 *
1934 * Since the kernel build-id is the first entry, process the
1935 * table using the old format if the well known
1936 * '[kernel.kallsyms]' string for the kernel build-id has the
1937 * first 4 characters chopped off (where the pid_t sits).
1938 */
1939 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
1940 if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
1941 return -1;
1942 return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
1943 }
1944
1945 __event_process_build_id(&bev, filename, session);
1946
1947 offset += bev.header.size;
1948 }
1949 err = 0;
1950out:
1951 return err;
1952}
1953
1954/* Macro for features that simply need to read and store a string. */
1955#define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
1956static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
1957{\
1958 ff->ph->env.__feat_env = do_read_string(ff); \
1959 return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
1960}
1961
1962FEAT_PROCESS_STR_FUN(hostname, hostname);
1963FEAT_PROCESS_STR_FUN(osrelease, os_release);
1964FEAT_PROCESS_STR_FUN(version, version);
1965FEAT_PROCESS_STR_FUN(arch, arch);
1966FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
1967FEAT_PROCESS_STR_FUN(cpuid, cpuid);
1968
1969static int process_tracing_data(struct feat_fd *ff, void *data)
1970{
1971 ssize_t ret = trace_report(ff->fd, data, false);
1972
1973 return ret < 0 ? -1 : 0;
1974}
1975
1976static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
1977{
1978 if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
1979 pr_debug("Failed to read buildids, continuing...\n");
1980 return 0;
1981}
1982
1983static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
1984{
1985 int ret;
1986 u32 nr_cpus_avail, nr_cpus_online;
1987
1988 ret = do_read_u32(ff, &nr_cpus_avail);
1989 if (ret)
1990 return ret;
1991
1992 ret = do_read_u32(ff, &nr_cpus_online);
1993 if (ret)
1994 return ret;
1995 ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
1996 ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
1997 return 0;
1998}
1999
2000static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2001{
2002 u64 total_mem;
2003 int ret;
2004
2005 ret = do_read_u64(ff, &total_mem);
2006 if (ret)
2007 return -1;
2008 ff->ph->env.total_mem = (unsigned long long)total_mem;
2009 return 0;
2010}
2011
2012static struct perf_evsel *
2013perf_evlist__find_by_index(struct perf_evlist *evlist, int idx)
2014{
2015 struct perf_evsel *evsel;
2016
2017 evlist__for_each_entry(evlist, evsel) {
2018 if (evsel->idx == idx)
2019 return evsel;
2020 }
2021
2022 return NULL;
2023}
2024
2025static void
2026perf_evlist__set_event_name(struct perf_evlist *evlist,
2027 struct perf_evsel *event)
2028{
2029 struct perf_evsel *evsel;
2030
2031 if (!event->name)
2032 return;
2033
2034 evsel = perf_evlist__find_by_index(evlist, event->idx);
2035 if (!evsel)
2036 return;
2037
2038 if (evsel->name)
2039 return;
2040
2041 evsel->name = strdup(event->name);
2042}
2043
2044static int
2045process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2046{
2047 struct perf_session *session;
2048 struct perf_evsel *evsel, *events = read_event_desc(ff);
2049
2050 if (!events)
2051 return 0;
2052
2053 session = container_of(ff->ph, struct perf_session, header);
2054
2055 if (session->data->is_pipe) {
2056 /* Save events for reading later by print_event_desc,
2057 * since they can't be read again in pipe mode. */
2058 ff->events = events;
2059 }
2060
2061 for (evsel = events; evsel->attr.size; evsel++)
2062 perf_evlist__set_event_name(session->evlist, evsel);
2063
2064 if (!session->data->is_pipe)
2065 free_event_desc(events);
2066
2067 return 0;
2068}
2069
2070static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2071{
2072 char *str, *cmdline = NULL, **argv = NULL;
2073 u32 nr, i, len = 0;
2074
2075 if (do_read_u32(ff, &nr))
2076 return -1;
2077
2078 ff->ph->env.nr_cmdline = nr;
2079
2080 cmdline = zalloc(ff->size + nr + 1);
2081 if (!cmdline)
2082 return -1;
2083
2084 argv = zalloc(sizeof(char *) * (nr + 1));
2085 if (!argv)
2086 goto error;
2087
2088 for (i = 0; i < nr; i++) {
2089 str = do_read_string(ff);
2090 if (!str)
2091 goto error;
2092
2093 argv[i] = cmdline + len;
2094 memcpy(argv[i], str, strlen(str) + 1);
2095 len += strlen(str) + 1;
2096 free(str);
2097 }
2098 ff->ph->env.cmdline = cmdline;
2099 ff->ph->env.cmdline_argv = (const char **) argv;
2100 return 0;
2101
2102error:
2103 free(argv);
2104 free(cmdline);
2105 return -1;
2106}
2107
2108static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2109{
2110 u32 nr, i;
2111 char *str;
2112 struct strbuf sb;
2113 int cpu_nr = ff->ph->env.nr_cpus_avail;
2114 u64 size = 0;
2115 struct perf_header *ph = ff->ph;
2116
2117 ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2118 if (!ph->env.cpu)
2119 return -1;
2120
2121 if (do_read_u32(ff, &nr))
2122 goto free_cpu;
2123
2124 ph->env.nr_sibling_cores = nr;
2125 size += sizeof(u32);
2126 if (strbuf_init(&sb, 128) < 0)
2127 goto free_cpu;
2128
2129 for (i = 0; i < nr; i++) {
2130 str = do_read_string(ff);
2131 if (!str)
2132 goto error;
2133
2134 /* include a NULL character at the end */
2135 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2136 goto error;
2137 size += string_size(str);
2138 free(str);
2139 }
2140 ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2141
2142 if (do_read_u32(ff, &nr))
2143 return -1;
2144
2145 ph->env.nr_sibling_threads = nr;
2146 size += sizeof(u32);
2147
2148 for (i = 0; i < nr; i++) {
2149 str = do_read_string(ff);
2150 if (!str)
2151 goto error;
2152
2153 /* include a NULL character at the end */
2154 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2155 goto error;
2156 size += string_size(str);
2157 free(str);
2158 }
2159 ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2160
2161 /*
2162 * The header may be from old perf,
2163 * which doesn't include core id and socket id information.
2164 */
2165 if (ff->size <= size) {
2166 zfree(&ph->env.cpu);
2167 return 0;
2168 }
2169
2170 for (i = 0; i < (u32)cpu_nr; i++) {
2171 if (do_read_u32(ff, &nr))
2172 goto free_cpu;
2173
2174 ph->env.cpu[i].core_id = nr;
2175
2176 if (do_read_u32(ff, &nr))
2177 goto free_cpu;
2178
2179 if (nr != (u32)-1 && nr > (u32)cpu_nr) {
2180 pr_debug("socket_id number is too big."
2181 "You may need to upgrade the perf tool.\n");
2182 goto free_cpu;
2183 }
2184
2185 ph->env.cpu[i].socket_id = nr;
2186 }
2187
2188 return 0;
2189
2190error:
2191 strbuf_release(&sb);
2192free_cpu:
2193 zfree(&ph->env.cpu);
2194 return -1;
2195}
2196
2197static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2198{
2199 struct numa_node *nodes, *n;
2200 u32 nr, i;
2201 char *str;
2202
2203 /* nr nodes */
2204 if (do_read_u32(ff, &nr))
2205 return -1;
2206
2207 nodes = zalloc(sizeof(*nodes) * nr);
2208 if (!nodes)
2209 return -ENOMEM;
2210
2211 for (i = 0; i < nr; i++) {
2212 n = &nodes[i];
2213
2214 /* node number */
2215 if (do_read_u32(ff, &n->node))
2216 goto error;
2217
2218 if (do_read_u64(ff, &n->mem_total))
2219 goto error;
2220
2221 if (do_read_u64(ff, &n->mem_free))
2222 goto error;
2223
2224 str = do_read_string(ff);
2225 if (!str)
2226 goto error;
2227
2228 n->map = cpu_map__new(str);
2229 if (!n->map)
2230 goto error;
2231
2232 free(str);
2233 }
2234 ff->ph->env.nr_numa_nodes = nr;
2235 ff->ph->env.numa_nodes = nodes;
2236 return 0;
2237
2238error:
2239 free(nodes);
2240 return -1;
2241}
2242
2243static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2244{
2245 char *name;
2246 u32 pmu_num;
2247 u32 type;
2248 struct strbuf sb;
2249
2250 if (do_read_u32(ff, &pmu_num))
2251 return -1;
2252
2253 if (!pmu_num) {
2254 pr_debug("pmu mappings not available\n");
2255 return 0;
2256 }
2257
2258 ff->ph->env.nr_pmu_mappings = pmu_num;
2259 if (strbuf_init(&sb, 128) < 0)
2260 return -1;
2261
2262 while (pmu_num) {
2263 if (do_read_u32(ff, &type))
2264 goto error;
2265
2266 name = do_read_string(ff);
2267 if (!name)
2268 goto error;
2269
2270 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2271 goto error;
2272 /* include a NULL character at the end */
2273 if (strbuf_add(&sb, "", 1) < 0)
2274 goto error;
2275
2276 if (!strcmp(name, "msr"))
2277 ff->ph->env.msr_pmu_type = type;
2278
2279 free(name);
2280 pmu_num--;
2281 }
2282 ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2283 return 0;
2284
2285error:
2286 strbuf_release(&sb);
2287 return -1;
2288}
2289
2290static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2291{
2292 size_t ret = -1;
2293 u32 i, nr, nr_groups;
2294 struct perf_session *session;
2295 struct perf_evsel *evsel, *leader = NULL;
2296 struct group_desc {
2297 char *name;
2298 u32 leader_idx;
2299 u32 nr_members;
2300 } *desc;
2301
2302 if (do_read_u32(ff, &nr_groups))
2303 return -1;
2304
2305 ff->ph->env.nr_groups = nr_groups;
2306 if (!nr_groups) {
2307 pr_debug("group desc not available\n");
2308 return 0;
2309 }
2310
2311 desc = calloc(nr_groups, sizeof(*desc));
2312 if (!desc)
2313 return -1;
2314
2315 for (i = 0; i < nr_groups; i++) {
2316 desc[i].name = do_read_string(ff);
2317 if (!desc[i].name)
2318 goto out_free;
2319
2320 if (do_read_u32(ff, &desc[i].leader_idx))
2321 goto out_free;
2322
2323 if (do_read_u32(ff, &desc[i].nr_members))
2324 goto out_free;
2325 }
2326
2327 /*
2328 * Rebuild group relationship based on the group_desc
2329 */
2330 session = container_of(ff->ph, struct perf_session, header);
2331 session->evlist->nr_groups = nr_groups;
2332
2333 i = nr = 0;
2334 evlist__for_each_entry(session->evlist, evsel) {
2335 if (evsel->idx == (int) desc[i].leader_idx) {
2336 evsel->leader = evsel;
2337 /* {anon_group} is a dummy name */
2338 if (strcmp(desc[i].name, "{anon_group}")) {
2339 evsel->group_name = desc[i].name;
2340 desc[i].name = NULL;
2341 }
2342 evsel->nr_members = desc[i].nr_members;
2343
2344 if (i >= nr_groups || nr > 0) {
2345 pr_debug("invalid group desc\n");
2346 goto out_free;
2347 }
2348
2349 leader = evsel;
2350 nr = evsel->nr_members - 1;
2351 i++;
2352 } else if (nr) {
2353 /* This is a group member */
2354 evsel->leader = leader;
2355
2356 nr--;
2357 }
2358 }
2359
2360 if (i != nr_groups || nr != 0) {
2361 pr_debug("invalid group desc\n");
2362 goto out_free;
2363 }
2364
2365 ret = 0;
2366out_free:
2367 for (i = 0; i < nr_groups; i++)
2368 zfree(&desc[i].name);
2369 free(desc);
2370
2371 return ret;
2372}
2373
2374static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2375{
2376 struct perf_session *session;
2377 int err;
2378
2379 session = container_of(ff->ph, struct perf_session, header);
2380
2381 err = auxtrace_index__process(ff->fd, ff->size, session,
2382 ff->ph->needs_swap);
2383 if (err < 0)
2384 pr_err("Failed to process auxtrace index\n");
2385 return err;
2386}
2387
2388static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2389{
2390 struct cpu_cache_level *caches;
2391 u32 cnt, i, version;
2392
2393 if (do_read_u32(ff, &version))
2394 return -1;
2395
2396 if (version != 1)
2397 return -1;
2398
2399 if (do_read_u32(ff, &cnt))
2400 return -1;
2401
2402 caches = zalloc(sizeof(*caches) * cnt);
2403 if (!caches)
2404 return -1;
2405
2406 for (i = 0; i < cnt; i++) {
2407 struct cpu_cache_level c;
2408
2409 #define _R(v) \
2410 if (do_read_u32(ff, &c.v))\
2411 goto out_free_caches; \
2412
2413 _R(level)
2414 _R(line_size)
2415 _R(sets)
2416 _R(ways)
2417 #undef _R
2418
2419 #define _R(v) \
2420 c.v = do_read_string(ff); \
2421 if (!c.v) \
2422 goto out_free_caches;
2423
2424 _R(type)
2425 _R(size)
2426 _R(map)
2427 #undef _R
2428
2429 caches[i] = c;
2430 }
2431
2432 ff->ph->env.caches = caches;
2433 ff->ph->env.caches_cnt = cnt;
2434 return 0;
2435out_free_caches:
2436 free(caches);
2437 return -1;
2438}
2439
2440static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2441{
2442 struct perf_session *session;
2443 u64 first_sample_time, last_sample_time;
2444 int ret;
2445
2446 session = container_of(ff->ph, struct perf_session, header);
2447
2448 ret = do_read_u64(ff, &first_sample_time);
2449 if (ret)
2450 return -1;
2451
2452 ret = do_read_u64(ff, &last_sample_time);
2453 if (ret)
2454 return -1;
2455
2456 session->evlist->first_sample_time = first_sample_time;
2457 session->evlist->last_sample_time = last_sample_time;
2458 return 0;
2459}
2460
2461static int process_mem_topology(struct feat_fd *ff,
2462 void *data __maybe_unused)
2463{
2464 struct memory_node *nodes;
2465 u64 version, i, nr, bsize;
2466 int ret = -1;
2467
2468 if (do_read_u64(ff, &version))
2469 return -1;
2470
2471 if (version != 1)
2472 return -1;
2473
2474 if (do_read_u64(ff, &bsize))
2475 return -1;
2476
2477 if (do_read_u64(ff, &nr))
2478 return -1;
2479
2480 nodes = zalloc(sizeof(*nodes) * nr);
2481 if (!nodes)
2482 return -1;
2483
2484 for (i = 0; i < nr; i++) {
2485 struct memory_node n;
2486
2487 #define _R(v) \
2488 if (do_read_u64(ff, &n.v)) \
2489 goto out; \
2490
2491 _R(node)
2492 _R(size)
2493
2494 #undef _R
2495
2496 if (do_read_bitmap(ff, &n.set, &n.size))
2497 goto out;
2498
2499 nodes[i] = n;
2500 }
2501
2502 ff->ph->env.memory_bsize = bsize;
2503 ff->ph->env.memory_nodes = nodes;
2504 ff->ph->env.nr_memory_nodes = nr;
2505 ret = 0;
2506
2507out:
2508 if (ret)
2509 free(nodes);
2510 return ret;
2511}
2512
2513struct feature_ops {
2514 int (*write)(struct feat_fd *ff, struct perf_evlist *evlist);
2515 void (*print)(struct feat_fd *ff, FILE *fp);
2516 int (*process)(struct feat_fd *ff, void *data);
2517 const char *name;
2518 bool full_only;
2519 bool synthesize;
2520};
2521
2522#define FEAT_OPR(n, func, __full_only) \
2523 [HEADER_##n] = { \
2524 .name = __stringify(n), \
2525 .write = write_##func, \
2526 .print = print_##func, \
2527 .full_only = __full_only, \
2528 .process = process_##func, \
2529 .synthesize = true \
2530 }
2531
2532#define FEAT_OPN(n, func, __full_only) \
2533 [HEADER_##n] = { \
2534 .name = __stringify(n), \
2535 .write = write_##func, \
2536 .print = print_##func, \
2537 .full_only = __full_only, \
2538 .process = process_##func \
2539 }
2540
2541/* feature_ops not implemented: */
2542#define print_tracing_data NULL
2543#define print_build_id NULL
2544
2545#define process_branch_stack NULL
2546#define process_stat NULL
2547
2548
2549static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
2550 FEAT_OPN(TRACING_DATA, tracing_data, false),
2551 FEAT_OPN(BUILD_ID, build_id, false),
2552 FEAT_OPR(HOSTNAME, hostname, false),
2553 FEAT_OPR(OSRELEASE, osrelease, false),
2554 FEAT_OPR(VERSION, version, false),
2555 FEAT_OPR(ARCH, arch, false),
2556 FEAT_OPR(NRCPUS, nrcpus, false),
2557 FEAT_OPR(CPUDESC, cpudesc, false),
2558 FEAT_OPR(CPUID, cpuid, false),
2559 FEAT_OPR(TOTAL_MEM, total_mem, false),
2560 FEAT_OPR(EVENT_DESC, event_desc, false),
2561 FEAT_OPR(CMDLINE, cmdline, false),
2562 FEAT_OPR(CPU_TOPOLOGY, cpu_topology, true),
2563 FEAT_OPR(NUMA_TOPOLOGY, numa_topology, true),
2564 FEAT_OPN(BRANCH_STACK, branch_stack, false),
2565 FEAT_OPR(PMU_MAPPINGS, pmu_mappings, false),
2566 FEAT_OPN(GROUP_DESC, group_desc, false),
2567 FEAT_OPN(AUXTRACE, auxtrace, false),
2568 FEAT_OPN(STAT, stat, false),
2569 FEAT_OPN(CACHE, cache, true),
2570 FEAT_OPR(SAMPLE_TIME, sample_time, false),
2571 FEAT_OPR(MEM_TOPOLOGY, mem_topology, true),
2572};
2573
2574struct header_print_data {
2575 FILE *fp;
2576 bool full; /* extended list of headers */
2577};
2578
2579static int perf_file_section__fprintf_info(struct perf_file_section *section,
2580 struct perf_header *ph,
2581 int feat, int fd, void *data)
2582{
2583 struct header_print_data *hd = data;
2584 struct feat_fd ff;
2585
2586 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
2587 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
2588 "%d, continuing...\n", section->offset, feat);
2589 return 0;
2590 }
2591 if (feat >= HEADER_LAST_FEATURE) {
2592 pr_warning("unknown feature %d\n", feat);
2593 return 0;
2594 }
2595 if (!feat_ops[feat].print)
2596 return 0;
2597
2598 ff = (struct feat_fd) {
2599 .fd = fd,
2600 .ph = ph,
2601 };
2602
2603 if (!feat_ops[feat].full_only || hd->full)
2604 feat_ops[feat].print(&ff, hd->fp);
2605 else
2606 fprintf(hd->fp, "# %s info available, use -I to display\n",
2607 feat_ops[feat].name);
2608
2609 return 0;
2610}
2611
2612int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
2613{
2614 struct header_print_data hd;
2615 struct perf_header *header = &session->header;
2616 int fd = perf_data__fd(session->data);
2617 struct stat st;
2618 int ret, bit;
2619
2620 hd.fp = fp;
2621 hd.full = full;
2622
2623 ret = fstat(fd, &st);
2624 if (ret == -1)
2625 return -1;
2626
2627 fprintf(fp, "# captured on : %s", ctime(&st.st_ctime));
2628
2629 fprintf(fp, "# header version : %u\n", header->version);
2630 fprintf(fp, "# data offset : %" PRIu64 "\n", header->data_offset);
2631 fprintf(fp, "# data size : %" PRIu64 "\n", header->data_size);
2632 fprintf(fp, "# feat offset : %" PRIu64 "\n", header->feat_offset);
2633
2634 perf_header__process_sections(header, fd, &hd,
2635 perf_file_section__fprintf_info);
2636
2637 if (session->data->is_pipe)
2638 return 0;
2639
2640 fprintf(fp, "# missing features: ");
2641 for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
2642 if (bit)
2643 fprintf(fp, "%s ", feat_ops[bit].name);
2644 }
2645
2646 fprintf(fp, "\n");
2647 return 0;
2648}
2649
2650static int do_write_feat(struct feat_fd *ff, int type,
2651 struct perf_file_section **p,
2652 struct perf_evlist *evlist)
2653{
2654 int err;
2655 int ret = 0;
2656
2657 if (perf_header__has_feat(ff->ph, type)) {
2658 if (!feat_ops[type].write)
2659 return -1;
2660
2661 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
2662 return -1;
2663
2664 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
2665
2666 err = feat_ops[type].write(ff, evlist);
2667 if (err < 0) {
2668 pr_debug("failed to write feature %s\n", feat_ops[type].name);
2669
2670 /* undo anything written */
2671 lseek(ff->fd, (*p)->offset, SEEK_SET);
2672
2673 return -1;
2674 }
2675 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
2676 (*p)++;
2677 }
2678 return ret;
2679}
2680
2681static int perf_header__adds_write(struct perf_header *header,
2682 struct perf_evlist *evlist, int fd)
2683{
2684 int nr_sections;
2685 struct feat_fd ff;
2686 struct perf_file_section *feat_sec, *p;
2687 int sec_size;
2688 u64 sec_start;
2689 int feat;
2690 int err;
2691
2692 ff = (struct feat_fd){
2693 .fd = fd,
2694 .ph = header,
2695 };
2696
2697 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
2698 if (!nr_sections)
2699 return 0;
2700
2701 feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
2702 if (feat_sec == NULL)
2703 return -ENOMEM;
2704
2705 sec_size = sizeof(*feat_sec) * nr_sections;
2706
2707 sec_start = header->feat_offset;
2708 lseek(fd, sec_start + sec_size, SEEK_SET);
2709
2710 for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
2711 if (do_write_feat(&ff, feat, &p, evlist))
2712 perf_header__clear_feat(header, feat);
2713 }
2714
2715 lseek(fd, sec_start, SEEK_SET);
2716 /*
2717 * may write more than needed due to dropped feature, but
2718 * this is okay, reader will skip the mising entries
2719 */
2720 err = do_write(&ff, feat_sec, sec_size);
2721 if (err < 0)
2722 pr_debug("failed to write feature section\n");
2723 free(feat_sec);
2724 return err;
2725}
2726
2727int perf_header__write_pipe(int fd)
2728{
2729 struct perf_pipe_file_header f_header;
2730 struct feat_fd ff;
2731 int err;
2732
2733 ff = (struct feat_fd){ .fd = fd };
2734
2735 f_header = (struct perf_pipe_file_header){
2736 .magic = PERF_MAGIC,
2737 .size = sizeof(f_header),
2738 };
2739
2740 err = do_write(&ff, &f_header, sizeof(f_header));
2741 if (err < 0) {
2742 pr_debug("failed to write perf pipe header\n");
2743 return err;
2744 }
2745
2746 return 0;
2747}
2748
2749int perf_session__write_header(struct perf_session *session,
2750 struct perf_evlist *evlist,
2751 int fd, bool at_exit)
2752{
2753 struct perf_file_header f_header;
2754 struct perf_file_attr f_attr;
2755 struct perf_header *header = &session->header;
2756 struct perf_evsel *evsel;
2757 struct feat_fd ff;
2758 u64 attr_offset;
2759 int err;
2760
2761 ff = (struct feat_fd){ .fd = fd};
2762 lseek(fd, sizeof(f_header), SEEK_SET);
2763
2764 evlist__for_each_entry(session->evlist, evsel) {
2765 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
2766 err = do_write(&ff, evsel->id, evsel->ids * sizeof(u64));
2767 if (err < 0) {
2768 pr_debug("failed to write perf header\n");
2769 return err;
2770 }
2771 }
2772
2773 attr_offset = lseek(ff.fd, 0, SEEK_CUR);
2774
2775 evlist__for_each_entry(evlist, evsel) {
2776 f_attr = (struct perf_file_attr){
2777 .attr = evsel->attr,
2778 .ids = {
2779 .offset = evsel->id_offset,
2780 .size = evsel->ids * sizeof(u64),
2781 }
2782 };
2783 err = do_write(&ff, &f_attr, sizeof(f_attr));
2784 if (err < 0) {
2785 pr_debug("failed to write perf header attribute\n");
2786 return err;
2787 }
2788 }
2789
2790 if (!header->data_offset)
2791 header->data_offset = lseek(fd, 0, SEEK_CUR);
2792 header->feat_offset = header->data_offset + header->data_size;
2793
2794 if (at_exit) {
2795 err = perf_header__adds_write(header, evlist, fd);
2796 if (err < 0)
2797 return err;
2798 }
2799
2800 f_header = (struct perf_file_header){
2801 .magic = PERF_MAGIC,
2802 .size = sizeof(f_header),
2803 .attr_size = sizeof(f_attr),
2804 .attrs = {
2805 .offset = attr_offset,
2806 .size = evlist->nr_entries * sizeof(f_attr),
2807 },
2808 .data = {
2809 .offset = header->data_offset,
2810 .size = header->data_size,
2811 },
2812 /* event_types is ignored, store zeros */
2813 };
2814
2815 memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
2816
2817 lseek(fd, 0, SEEK_SET);
2818 err = do_write(&ff, &f_header, sizeof(f_header));
2819 if (err < 0) {
2820 pr_debug("failed to write perf header\n");
2821 return err;
2822 }
2823 lseek(fd, header->data_offset + header->data_size, SEEK_SET);
2824
2825 return 0;
2826}
2827
2828static int perf_header__getbuffer64(struct perf_header *header,
2829 int fd, void *buf, size_t size)
2830{
2831 if (readn(fd, buf, size) <= 0)
2832 return -1;
2833
2834 if (header->needs_swap)
2835 mem_bswap_64(buf, size);
2836
2837 return 0;
2838}
2839
2840int perf_header__process_sections(struct perf_header *header, int fd,
2841 void *data,
2842 int (*process)(struct perf_file_section *section,
2843 struct perf_header *ph,
2844 int feat, int fd, void *data))
2845{
2846 struct perf_file_section *feat_sec, *sec;
2847 int nr_sections;
2848 int sec_size;
2849 int feat;
2850 int err;
2851
2852 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
2853 if (!nr_sections)
2854 return 0;
2855
2856 feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
2857 if (!feat_sec)
2858 return -1;
2859
2860 sec_size = sizeof(*feat_sec) * nr_sections;
2861
2862 lseek(fd, header->feat_offset, SEEK_SET);
2863
2864 err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
2865 if (err < 0)
2866 goto out_free;
2867
2868 for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
2869 err = process(sec++, header, feat, fd, data);
2870 if (err < 0)
2871 goto out_free;
2872 }
2873 err = 0;
2874out_free:
2875 free(feat_sec);
2876 return err;
2877}
2878
2879static const int attr_file_abi_sizes[] = {
2880 [0] = PERF_ATTR_SIZE_VER0,
2881 [1] = PERF_ATTR_SIZE_VER1,
2882 [2] = PERF_ATTR_SIZE_VER2,
2883 [3] = PERF_ATTR_SIZE_VER3,
2884 [4] = PERF_ATTR_SIZE_VER4,
2885 0,
2886};
2887
2888/*
2889 * In the legacy file format, the magic number is not used to encode endianness.
2890 * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
2891 * on ABI revisions, we need to try all combinations for all endianness to
2892 * detect the endianness.
2893 */
2894static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
2895{
2896 uint64_t ref_size, attr_size;
2897 int i;
2898
2899 for (i = 0 ; attr_file_abi_sizes[i]; i++) {
2900 ref_size = attr_file_abi_sizes[i]
2901 + sizeof(struct perf_file_section);
2902 if (hdr_sz != ref_size) {
2903 attr_size = bswap_64(hdr_sz);
2904 if (attr_size != ref_size)
2905 continue;
2906
2907 ph->needs_swap = true;
2908 }
2909 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
2910 i,
2911 ph->needs_swap);
2912 return 0;
2913 }
2914 /* could not determine endianness */
2915 return -1;
2916}
2917
2918#define PERF_PIPE_HDR_VER0 16
2919
2920static const size_t attr_pipe_abi_sizes[] = {
2921 [0] = PERF_PIPE_HDR_VER0,
2922 0,
2923};
2924
2925/*
2926 * In the legacy pipe format, there is an implicit assumption that endiannesss
2927 * between host recording the samples, and host parsing the samples is the
2928 * same. This is not always the case given that the pipe output may always be
2929 * redirected into a file and analyzed on a different machine with possibly a
2930 * different endianness and perf_event ABI revsions in the perf tool itself.
2931 */
2932static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
2933{
2934 u64 attr_size;
2935 int i;
2936
2937 for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
2938 if (hdr_sz != attr_pipe_abi_sizes[i]) {
2939 attr_size = bswap_64(hdr_sz);
2940 if (attr_size != hdr_sz)
2941 continue;
2942
2943 ph->needs_swap = true;
2944 }
2945 pr_debug("Pipe ABI%d perf.data file detected\n", i);
2946 return 0;
2947 }
2948 return -1;
2949}
2950
2951bool is_perf_magic(u64 magic)
2952{
2953 if (!memcmp(&magic, __perf_magic1, sizeof(magic))
2954 || magic == __perf_magic2
2955 || magic == __perf_magic2_sw)
2956 return true;
2957
2958 return false;
2959}
2960
2961static int check_magic_endian(u64 magic, uint64_t hdr_sz,
2962 bool is_pipe, struct perf_header *ph)
2963{
2964 int ret;
2965
2966 /* check for legacy format */
2967 ret = memcmp(&magic, __perf_magic1, sizeof(magic));
2968 if (ret == 0) {
2969 ph->version = PERF_HEADER_VERSION_1;
2970 pr_debug("legacy perf.data format\n");
2971 if (is_pipe)
2972 return try_all_pipe_abis(hdr_sz, ph);
2973
2974 return try_all_file_abis(hdr_sz, ph);
2975 }
2976 /*
2977 * the new magic number serves two purposes:
2978 * - unique number to identify actual perf.data files
2979 * - encode endianness of file
2980 */
2981 ph->version = PERF_HEADER_VERSION_2;
2982
2983 /* check magic number with one endianness */
2984 if (magic == __perf_magic2)
2985 return 0;
2986
2987 /* check magic number with opposite endianness */
2988 if (magic != __perf_magic2_sw)
2989 return -1;
2990
2991 ph->needs_swap = true;
2992
2993 return 0;
2994}
2995
2996int perf_file_header__read(struct perf_file_header *header,
2997 struct perf_header *ph, int fd)
2998{
2999 ssize_t ret;
3000
3001 lseek(fd, 0, SEEK_SET);
3002
3003 ret = readn(fd, header, sizeof(*header));
3004 if (ret <= 0)
3005 return -1;
3006
3007 if (check_magic_endian(header->magic,
3008 header->attr_size, false, ph) < 0) {
3009 pr_debug("magic/endian check failed\n");
3010 return -1;
3011 }
3012
3013 if (ph->needs_swap) {
3014 mem_bswap_64(header, offsetof(struct perf_file_header,
3015 adds_features));
3016 }
3017
3018 if (header->size != sizeof(*header)) {
3019 /* Support the previous format */
3020 if (header->size == offsetof(typeof(*header), adds_features))
3021 bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3022 else
3023 return -1;
3024 } else if (ph->needs_swap) {
3025 /*
3026 * feature bitmap is declared as an array of unsigned longs --
3027 * not good since its size can differ between the host that
3028 * generated the data file and the host analyzing the file.
3029 *
3030 * We need to handle endianness, but we don't know the size of
3031 * the unsigned long where the file was generated. Take a best
3032 * guess at determining it: try 64-bit swap first (ie., file
3033 * created on a 64-bit host), and check if the hostname feature
3034 * bit is set (this feature bit is forced on as of fbe96f2).
3035 * If the bit is not, undo the 64-bit swap and try a 32-bit
3036 * swap. If the hostname bit is still not set (e.g., older data
3037 * file), punt and fallback to the original behavior --
3038 * clearing all feature bits and setting buildid.
3039 */
3040 mem_bswap_64(&header->adds_features,
3041 BITS_TO_U64(HEADER_FEAT_BITS));
3042
3043 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3044 /* unswap as u64 */
3045 mem_bswap_64(&header->adds_features,
3046 BITS_TO_U64(HEADER_FEAT_BITS));
3047
3048 /* unswap as u32 */
3049 mem_bswap_32(&header->adds_features,
3050 BITS_TO_U32(HEADER_FEAT_BITS));
3051 }
3052
3053 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3054 bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3055 set_bit(HEADER_BUILD_ID, header->adds_features);
3056 }
3057 }
3058
3059 memcpy(&ph->adds_features, &header->adds_features,
3060 sizeof(ph->adds_features));
3061
3062 ph->data_offset = header->data.offset;
3063 ph->data_size = header->data.size;
3064 ph->feat_offset = header->data.offset + header->data.size;
3065 return 0;
3066}
3067
3068static int perf_file_section__process(struct perf_file_section *section,
3069 struct perf_header *ph,
3070 int feat, int fd, void *data)
3071{
3072 struct feat_fd fdd = {
3073 .fd = fd,
3074 .ph = ph,
3075 .size = section->size,
3076 .offset = section->offset,
3077 };
3078
3079 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3080 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3081 "%d, continuing...\n", section->offset, feat);
3082 return 0;
3083 }
3084
3085 if (feat >= HEADER_LAST_FEATURE) {
3086 pr_debug("unknown feature %d, continuing...\n", feat);
3087 return 0;
3088 }
3089
3090 if (!feat_ops[feat].process)
3091 return 0;
3092
3093 return feat_ops[feat].process(&fdd, data);
3094}
3095
3096static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
3097 struct perf_header *ph, int fd,
3098 bool repipe)
3099{
3100 struct feat_fd ff = {
3101 .fd = STDOUT_FILENO,
3102 .ph = ph,
3103 };
3104 ssize_t ret;
3105
3106 ret = readn(fd, header, sizeof(*header));
3107 if (ret <= 0)
3108 return -1;
3109
3110 if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
3111 pr_debug("endian/magic failed\n");
3112 return -1;
3113 }
3114
3115 if (ph->needs_swap)
3116 header->size = bswap_64(header->size);
3117
3118 if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
3119 return -1;
3120
3121 return 0;
3122}
3123
3124static int perf_header__read_pipe(struct perf_session *session)
3125{
3126 struct perf_header *header = &session->header;
3127 struct perf_pipe_file_header f_header;
3128
3129 if (perf_file_header__read_pipe(&f_header, header,
3130 perf_data__fd(session->data),
3131 session->repipe) < 0) {
3132 pr_debug("incompatible file format\n");
3133 return -EINVAL;
3134 }
3135
3136 return 0;
3137}
3138
3139static int read_attr(int fd, struct perf_header *ph,
3140 struct perf_file_attr *f_attr)
3141{
3142 struct perf_event_attr *attr = &f_attr->attr;
3143 size_t sz, left;
3144 size_t our_sz = sizeof(f_attr->attr);
3145 ssize_t ret;
3146
3147 memset(f_attr, 0, sizeof(*f_attr));
3148
3149 /* read minimal guaranteed structure */
3150 ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
3151 if (ret <= 0) {
3152 pr_debug("cannot read %d bytes of header attr\n",
3153 PERF_ATTR_SIZE_VER0);
3154 return -1;
3155 }
3156
3157 /* on file perf_event_attr size */
3158 sz = attr->size;
3159
3160 if (ph->needs_swap)
3161 sz = bswap_32(sz);
3162
3163 if (sz == 0) {
3164 /* assume ABI0 */
3165 sz = PERF_ATTR_SIZE_VER0;
3166 } else if (sz > our_sz) {
3167 pr_debug("file uses a more recent and unsupported ABI"
3168 " (%zu bytes extra)\n", sz - our_sz);
3169 return -1;
3170 }
3171 /* what we have not yet read and that we know about */
3172 left = sz - PERF_ATTR_SIZE_VER0;
3173 if (left) {
3174 void *ptr = attr;
3175 ptr += PERF_ATTR_SIZE_VER0;
3176
3177 ret = readn(fd, ptr, left);
3178 }
3179 /* read perf_file_section, ids are read in caller */
3180 ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
3181
3182 return ret <= 0 ? -1 : 0;
3183}
3184
3185static int perf_evsel__prepare_tracepoint_event(struct perf_evsel *evsel,
3186 struct pevent *pevent)
3187{
3188 struct event_format *event;
3189 char bf[128];
3190
3191 /* already prepared */
3192 if (evsel->tp_format)
3193 return 0;
3194
3195 if (pevent == NULL) {
3196 pr_debug("broken or missing trace data\n");
3197 return -1;
3198 }
3199
3200 event = pevent_find_event(pevent, evsel->attr.config);
3201 if (event == NULL) {
3202 pr_debug("cannot find event format for %d\n", (int)evsel->attr.config);
3203 return -1;
3204 }
3205
3206 if (!evsel->name) {
3207 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
3208 evsel->name = strdup(bf);
3209 if (evsel->name == NULL)
3210 return -1;
3211 }
3212
3213 evsel->tp_format = event;
3214 return 0;
3215}
3216
3217static int perf_evlist__prepare_tracepoint_events(struct perf_evlist *evlist,
3218 struct pevent *pevent)
3219{
3220 struct perf_evsel *pos;
3221
3222 evlist__for_each_entry(evlist, pos) {
3223 if (pos->attr.type == PERF_TYPE_TRACEPOINT &&
3224 perf_evsel__prepare_tracepoint_event(pos, pevent))
3225 return -1;
3226 }
3227
3228 return 0;
3229}
3230
3231int perf_session__read_header(struct perf_session *session)
3232{
3233 struct perf_data *data = session->data;
3234 struct perf_header *header = &session->header;
3235 struct perf_file_header f_header;
3236 struct perf_file_attr f_attr;
3237 u64 f_id;
3238 int nr_attrs, nr_ids, i, j;
3239 int fd = perf_data__fd(data);
3240
3241 session->evlist = perf_evlist__new();
3242 if (session->evlist == NULL)
3243 return -ENOMEM;
3244
3245 session->evlist->env = &header->env;
3246 session->machines.host.env = &header->env;
3247 if (perf_data__is_pipe(data))
3248 return perf_header__read_pipe(session);
3249
3250 if (perf_file_header__read(&f_header, header, fd) < 0)
3251 return -EINVAL;
3252
3253 /*
3254 * Sanity check that perf.data was written cleanly; data size is
3255 * initialized to 0 and updated only if the on_exit function is run.
3256 * If data size is still 0 then the file contains only partial
3257 * information. Just warn user and process it as much as it can.
3258 */
3259 if (f_header.data.size == 0) {
3260 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
3261 "Was the 'perf record' command properly terminated?\n",
3262 data->file.path);
3263 }
3264
3265 nr_attrs = f_header.attrs.size / f_header.attr_size;
3266 lseek(fd, f_header.attrs.offset, SEEK_SET);
3267
3268 for (i = 0; i < nr_attrs; i++) {
3269 struct perf_evsel *evsel;
3270 off_t tmp;
3271
3272 if (read_attr(fd, header, &f_attr) < 0)
3273 goto out_errno;
3274
3275 if (header->needs_swap) {
3276 f_attr.ids.size = bswap_64(f_attr.ids.size);
3277 f_attr.ids.offset = bswap_64(f_attr.ids.offset);
3278 perf_event__attr_swap(&f_attr.attr);
3279 }
3280
3281 tmp = lseek(fd, 0, SEEK_CUR);
3282 evsel = perf_evsel__new(&f_attr.attr);
3283
3284 if (evsel == NULL)
3285 goto out_delete_evlist;
3286
3287 evsel->needs_swap = header->needs_swap;
3288 /*
3289 * Do it before so that if perf_evsel__alloc_id fails, this
3290 * entry gets purged too at perf_evlist__delete().
3291 */
3292 perf_evlist__add(session->evlist, evsel);
3293
3294 nr_ids = f_attr.ids.size / sizeof(u64);
3295 /*
3296 * We don't have the cpu and thread maps on the header, so
3297 * for allocating the perf_sample_id table we fake 1 cpu and
3298 * hattr->ids threads.
3299 */
3300 if (perf_evsel__alloc_id(evsel, 1, nr_ids))
3301 goto out_delete_evlist;
3302
3303 lseek(fd, f_attr.ids.offset, SEEK_SET);
3304
3305 for (j = 0; j < nr_ids; j++) {
3306 if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
3307 goto out_errno;
3308
3309 perf_evlist__id_add(session->evlist, evsel, 0, j, f_id);
3310 }
3311
3312 lseek(fd, tmp, SEEK_SET);
3313 }
3314
3315 symbol_conf.nr_events = nr_attrs;
3316
3317 perf_header__process_sections(header, fd, &session->tevent,
3318 perf_file_section__process);
3319
3320 if (perf_evlist__prepare_tracepoint_events(session->evlist,
3321 session->tevent.pevent))
3322 goto out_delete_evlist;
3323
3324 return 0;
3325out_errno:
3326 return -errno;
3327
3328out_delete_evlist:
3329 perf_evlist__delete(session->evlist);
3330 session->evlist = NULL;
3331 return -ENOMEM;
3332}
3333
3334int perf_event__synthesize_attr(struct perf_tool *tool,
3335 struct perf_event_attr *attr, u32 ids, u64 *id,
3336 perf_event__handler_t process)
3337{
3338 union perf_event *ev;
3339 size_t size;
3340 int err;
3341
3342 size = sizeof(struct perf_event_attr);
3343 size = PERF_ALIGN(size, sizeof(u64));
3344 size += sizeof(struct perf_event_header);
3345 size += ids * sizeof(u64);
3346
3347 ev = malloc(size);
3348
3349 if (ev == NULL)
3350 return -ENOMEM;
3351
3352 ev->attr.attr = *attr;
3353 memcpy(ev->attr.id, id, ids * sizeof(u64));
3354
3355 ev->attr.header.type = PERF_RECORD_HEADER_ATTR;
3356 ev->attr.header.size = (u16)size;
3357
3358 if (ev->attr.header.size == size)
3359 err = process(tool, ev, NULL, NULL);
3360 else
3361 err = -E2BIG;
3362
3363 free(ev);
3364
3365 return err;
3366}
3367
3368int perf_event__synthesize_features(struct perf_tool *tool,
3369 struct perf_session *session,
3370 struct perf_evlist *evlist,
3371 perf_event__handler_t process)
3372{
3373 struct perf_header *header = &session->header;
3374 struct feat_fd ff;
3375 struct feature_event *fe;
3376 size_t sz, sz_hdr;
3377 int feat, ret;
3378
3379 sz_hdr = sizeof(fe->header);
3380 sz = sizeof(union perf_event);
3381 /* get a nice alignment */
3382 sz = PERF_ALIGN(sz, page_size);
3383
3384 memset(&ff, 0, sizeof(ff));
3385
3386 ff.buf = malloc(sz);
3387 if (!ff.buf)
3388 return -ENOMEM;
3389
3390 ff.size = sz - sz_hdr;
3391
3392 for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3393 if (!feat_ops[feat].synthesize) {
3394 pr_debug("No record header feature for header :%d\n", feat);
3395 continue;
3396 }
3397
3398 ff.offset = sizeof(*fe);
3399
3400 ret = feat_ops[feat].write(&ff, evlist);
3401 if (ret || ff.offset <= (ssize_t)sizeof(*fe)) {
3402 pr_debug("Error writing feature\n");
3403 continue;
3404 }
3405 /* ff.buf may have changed due to realloc in do_write() */
3406 fe = ff.buf;
3407 memset(fe, 0, sizeof(*fe));
3408
3409 fe->feat_id = feat;
3410 fe->header.type = PERF_RECORD_HEADER_FEATURE;
3411 fe->header.size = ff.offset;
3412
3413 ret = process(tool, ff.buf, NULL, NULL);
3414 if (ret) {
3415 free(ff.buf);
3416 return ret;
3417 }
3418 }
3419
3420 /* Send HEADER_LAST_FEATURE mark. */
3421 fe = ff.buf;
3422 fe->feat_id = HEADER_LAST_FEATURE;
3423 fe->header.type = PERF_RECORD_HEADER_FEATURE;
3424 fe->header.size = sizeof(*fe);
3425
3426 ret = process(tool, ff.buf, NULL, NULL);
3427
3428 free(ff.buf);
3429 return ret;
3430}
3431
3432int perf_event__process_feature(struct perf_tool *tool,
3433 union perf_event *event,
3434 struct perf_session *session __maybe_unused)
3435{
3436 struct feat_fd ff = { .fd = 0 };
3437 struct feature_event *fe = (struct feature_event *)event;
3438 int type = fe->header.type;
3439 u64 feat = fe->feat_id;
3440
3441 if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
3442 pr_warning("invalid record type %d in pipe-mode\n", type);
3443 return 0;
3444 }
3445 if (feat == HEADER_RESERVED || feat > HEADER_LAST_FEATURE) {
3446 pr_warning("invalid record type %d in pipe-mode\n", type);
3447 return -1;
3448 }
3449
3450 if (!feat_ops[feat].process)
3451 return 0;
3452
3453 ff.buf = (void *)fe->data;
3454 ff.size = event->header.size - sizeof(event->header);
3455 ff.ph = &session->header;
3456
3457 if (feat_ops[feat].process(&ff, NULL))
3458 return -1;
3459
3460 if (!feat_ops[feat].print || !tool->show_feat_hdr)
3461 return 0;
3462
3463 if (!feat_ops[feat].full_only ||
3464 tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
3465 feat_ops[feat].print(&ff, stdout);
3466 } else {
3467 fprintf(stdout, "# %s info available, use -I to display\n",
3468 feat_ops[feat].name);
3469 }
3470
3471 return 0;
3472}
3473
3474static struct event_update_event *
3475event_update_event__new(size_t size, u64 type, u64 id)
3476{
3477 struct event_update_event *ev;
3478
3479 size += sizeof(*ev);
3480 size = PERF_ALIGN(size, sizeof(u64));
3481
3482 ev = zalloc(size);
3483 if (ev) {
3484 ev->header.type = PERF_RECORD_EVENT_UPDATE;
3485 ev->header.size = (u16)size;
3486 ev->type = type;
3487 ev->id = id;
3488 }
3489 return ev;
3490}
3491
3492int
3493perf_event__synthesize_event_update_unit(struct perf_tool *tool,
3494 struct perf_evsel *evsel,
3495 perf_event__handler_t process)
3496{
3497 struct event_update_event *ev;
3498 size_t size = strlen(evsel->unit);
3499 int err;
3500
3501 ev = event_update_event__new(size + 1, PERF_EVENT_UPDATE__UNIT, evsel->id[0]);
3502 if (ev == NULL)
3503 return -ENOMEM;
3504
3505 strncpy(ev->data, evsel->unit, size);
3506 err = process(tool, (union perf_event *)ev, NULL, NULL);
3507 free(ev);
3508 return err;
3509}
3510
3511int
3512perf_event__synthesize_event_update_scale(struct perf_tool *tool,
3513 struct perf_evsel *evsel,
3514 perf_event__handler_t process)
3515{
3516 struct event_update_event *ev;
3517 struct event_update_event_scale *ev_data;
3518 int err;
3519
3520 ev = event_update_event__new(sizeof(*ev_data), PERF_EVENT_UPDATE__SCALE, evsel->id[0]);
3521 if (ev == NULL)
3522 return -ENOMEM;
3523
3524 ev_data = (struct event_update_event_scale *) ev->data;
3525 ev_data->scale = evsel->scale;
3526 err = process(tool, (union perf_event*) ev, NULL, NULL);
3527 free(ev);
3528 return err;
3529}
3530
3531int
3532perf_event__synthesize_event_update_name(struct perf_tool *tool,
3533 struct perf_evsel *evsel,
3534 perf_event__handler_t process)
3535{
3536 struct event_update_event *ev;
3537 size_t len = strlen(evsel->name);
3538 int err;
3539
3540 ev = event_update_event__new(len + 1, PERF_EVENT_UPDATE__NAME, evsel->id[0]);
3541 if (ev == NULL)
3542 return -ENOMEM;
3543
3544 strncpy(ev->data, evsel->name, len);
3545 err = process(tool, (union perf_event*) ev, NULL, NULL);
3546 free(ev);
3547 return err;
3548}
3549
3550int
3551perf_event__synthesize_event_update_cpus(struct perf_tool *tool,
3552 struct perf_evsel *evsel,
3553 perf_event__handler_t process)
3554{
3555 size_t size = sizeof(struct event_update_event);
3556 struct event_update_event *ev;
3557 int max, err;
3558 u16 type;
3559
3560 if (!evsel->own_cpus)
3561 return 0;
3562
3563 ev = cpu_map_data__alloc(evsel->own_cpus, &size, &type, &max);
3564 if (!ev)
3565 return -ENOMEM;
3566
3567 ev->header.type = PERF_RECORD_EVENT_UPDATE;
3568 ev->header.size = (u16)size;
3569 ev->type = PERF_EVENT_UPDATE__CPUS;
3570 ev->id = evsel->id[0];
3571
3572 cpu_map_data__synthesize((struct cpu_map_data *) ev->data,
3573 evsel->own_cpus,
3574 type, max);
3575
3576 err = process(tool, (union perf_event*) ev, NULL, NULL);
3577 free(ev);
3578 return err;
3579}
3580
3581size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
3582{
3583 struct event_update_event *ev = &event->event_update;
3584 struct event_update_event_scale *ev_scale;
3585 struct event_update_event_cpus *ev_cpus;
3586 struct cpu_map *map;
3587 size_t ret;
3588
3589 ret = fprintf(fp, "\n... id: %" PRIu64 "\n", ev->id);
3590
3591 switch (ev->type) {
3592 case PERF_EVENT_UPDATE__SCALE:
3593 ev_scale = (struct event_update_event_scale *) ev->data;
3594 ret += fprintf(fp, "... scale: %f\n", ev_scale->scale);
3595 break;
3596 case PERF_EVENT_UPDATE__UNIT:
3597 ret += fprintf(fp, "... unit: %s\n", ev->data);
3598 break;
3599 case PERF_EVENT_UPDATE__NAME:
3600 ret += fprintf(fp, "... name: %s\n", ev->data);
3601 break;
3602 case PERF_EVENT_UPDATE__CPUS:
3603 ev_cpus = (struct event_update_event_cpus *) ev->data;
3604 ret += fprintf(fp, "... ");
3605
3606 map = cpu_map__new_data(&ev_cpus->cpus);
3607 if (map)
3608 ret += cpu_map__fprintf(map, fp);
3609 else
3610 ret += fprintf(fp, "failed to get cpus\n");
3611 break;
3612 default:
3613 ret += fprintf(fp, "... unknown type\n");
3614 break;
3615 }
3616
3617 return ret;
3618}
3619
3620int perf_event__synthesize_attrs(struct perf_tool *tool,
3621 struct perf_session *session,
3622 perf_event__handler_t process)
3623{
3624 struct perf_evsel *evsel;
3625 int err = 0;
3626
3627 evlist__for_each_entry(session->evlist, evsel) {
3628 err = perf_event__synthesize_attr(tool, &evsel->attr, evsel->ids,
3629 evsel->id, process);
3630 if (err) {
3631 pr_debug("failed to create perf header attribute\n");
3632 return err;
3633 }
3634 }
3635
3636 return err;
3637}
3638
3639static bool has_unit(struct perf_evsel *counter)
3640{
3641 return counter->unit && *counter->unit;
3642}
3643
3644static bool has_scale(struct perf_evsel *counter)
3645{
3646 return counter->scale != 1;
3647}
3648
3649int perf_event__synthesize_extra_attr(struct perf_tool *tool,
3650 struct perf_evlist *evsel_list,
3651 perf_event__handler_t process,
3652 bool is_pipe)
3653{
3654 struct perf_evsel *counter;
3655 int err;
3656
3657 /*
3658 * Synthesize other events stuff not carried within
3659 * attr event - unit, scale, name
3660 */
3661 evlist__for_each_entry(evsel_list, counter) {
3662 if (!counter->supported)
3663 continue;
3664
3665 /*
3666 * Synthesize unit and scale only if it's defined.
3667 */
3668 if (has_unit(counter)) {
3669 err = perf_event__synthesize_event_update_unit(tool, counter, process);
3670 if (err < 0) {
3671 pr_err("Couldn't synthesize evsel unit.\n");
3672 return err;
3673 }
3674 }
3675
3676 if (has_scale(counter)) {
3677 err = perf_event__synthesize_event_update_scale(tool, counter, process);
3678 if (err < 0) {
3679 pr_err("Couldn't synthesize evsel counter.\n");
3680 return err;
3681 }
3682 }
3683
3684 if (counter->own_cpus) {
3685 err = perf_event__synthesize_event_update_cpus(tool, counter, process);
3686 if (err < 0) {
3687 pr_err("Couldn't synthesize evsel cpus.\n");
3688 return err;
3689 }
3690 }
3691
3692 /*
3693 * Name is needed only for pipe output,
3694 * perf.data carries event names.
3695 */
3696 if (is_pipe) {
3697 err = perf_event__synthesize_event_update_name(tool, counter, process);
3698 if (err < 0) {
3699 pr_err("Couldn't synthesize evsel name.\n");
3700 return err;
3701 }
3702 }
3703 }
3704 return 0;
3705}
3706
3707int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
3708 union perf_event *event,
3709 struct perf_evlist **pevlist)
3710{
3711 u32 i, ids, n_ids;
3712 struct perf_evsel *evsel;
3713 struct perf_evlist *evlist = *pevlist;
3714
3715 if (evlist == NULL) {
3716 *pevlist = evlist = perf_evlist__new();
3717 if (evlist == NULL)
3718 return -ENOMEM;
3719 }
3720
3721 evsel = perf_evsel__new(&event->attr.attr);
3722 if (evsel == NULL)
3723 return -ENOMEM;
3724
3725 perf_evlist__add(evlist, evsel);
3726
3727 ids = event->header.size;
3728 ids -= (void *)&event->attr.id - (void *)event;
3729 n_ids = ids / sizeof(u64);
3730 /*
3731 * We don't have the cpu and thread maps on the header, so
3732 * for allocating the perf_sample_id table we fake 1 cpu and
3733 * hattr->ids threads.
3734 */
3735 if (perf_evsel__alloc_id(evsel, 1, n_ids))
3736 return -ENOMEM;
3737
3738 for (i = 0; i < n_ids; i++) {
3739 perf_evlist__id_add(evlist, evsel, 0, i, event->attr.id[i]);
3740 }
3741
3742 symbol_conf.nr_events = evlist->nr_entries;
3743
3744 return 0;
3745}
3746
3747int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
3748 union perf_event *event,
3749 struct perf_evlist **pevlist)
3750{
3751 struct event_update_event *ev = &event->event_update;
3752 struct event_update_event_scale *ev_scale;
3753 struct event_update_event_cpus *ev_cpus;
3754 struct perf_evlist *evlist;
3755 struct perf_evsel *evsel;
3756 struct cpu_map *map;
3757
3758 if (!pevlist || *pevlist == NULL)
3759 return -EINVAL;
3760
3761 evlist = *pevlist;
3762
3763 evsel = perf_evlist__id2evsel(evlist, ev->id);
3764 if (evsel == NULL)
3765 return -EINVAL;
3766
3767 switch (ev->type) {
3768 case PERF_EVENT_UPDATE__UNIT:
3769 evsel->unit = strdup(ev->data);
3770 break;
3771 case PERF_EVENT_UPDATE__NAME:
3772 evsel->name = strdup(ev->data);
3773 break;
3774 case PERF_EVENT_UPDATE__SCALE:
3775 ev_scale = (struct event_update_event_scale *) ev->data;
3776 evsel->scale = ev_scale->scale;
3777 break;
3778 case PERF_EVENT_UPDATE__CPUS:
3779 ev_cpus = (struct event_update_event_cpus *) ev->data;
3780
3781 map = cpu_map__new_data(&ev_cpus->cpus);
3782 if (map)
3783 evsel->own_cpus = map;
3784 else
3785 pr_err("failed to get event_update cpus\n");
3786 default:
3787 break;
3788 }
3789
3790 return 0;
3791}
3792
3793int perf_event__synthesize_tracing_data(struct perf_tool *tool, int fd,
3794 struct perf_evlist *evlist,
3795 perf_event__handler_t process)
3796{
3797 union perf_event ev;
3798 struct tracing_data *tdata;
3799 ssize_t size = 0, aligned_size = 0, padding;
3800 struct feat_fd ff;
3801 int err __maybe_unused = 0;
3802
3803 /*
3804 * We are going to store the size of the data followed
3805 * by the data contents. Since the fd descriptor is a pipe,
3806 * we cannot seek back to store the size of the data once
3807 * we know it. Instead we:
3808 *
3809 * - write the tracing data to the temp file
3810 * - get/write the data size to pipe
3811 * - write the tracing data from the temp file
3812 * to the pipe
3813 */
3814 tdata = tracing_data_get(&evlist->entries, fd, true);
3815 if (!tdata)
3816 return -1;
3817
3818 memset(&ev, 0, sizeof(ev));
3819
3820 ev.tracing_data.header.type = PERF_RECORD_HEADER_TRACING_DATA;
3821 size = tdata->size;
3822 aligned_size = PERF_ALIGN(size, sizeof(u64));
3823 padding = aligned_size - size;
3824 ev.tracing_data.header.size = sizeof(ev.tracing_data);
3825 ev.tracing_data.size = aligned_size;
3826
3827 process(tool, &ev, NULL, NULL);
3828
3829 /*
3830 * The put function will copy all the tracing data
3831 * stored in temp file to the pipe.
3832 */
3833 tracing_data_put(tdata);
3834
3835 ff = (struct feat_fd){ .fd = fd };
3836 if (write_padded(&ff, NULL, 0, padding))
3837 return -1;
3838
3839 return aligned_size;
3840}
3841
3842int perf_event__process_tracing_data(struct perf_tool *tool __maybe_unused,
3843 union perf_event *event,
3844 struct perf_session *session)
3845{
3846 ssize_t size_read, padding, size = event->tracing_data.size;
3847 int fd = perf_data__fd(session->data);
3848 off_t offset = lseek(fd, 0, SEEK_CUR);
3849 char buf[BUFSIZ];
3850
3851 /* setup for reading amidst mmap */
3852 lseek(fd, offset + sizeof(struct tracing_data_event),
3853 SEEK_SET);
3854
3855 size_read = trace_report(fd, &session->tevent,
3856 session->repipe);
3857 padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
3858
3859 if (readn(fd, buf, padding) < 0) {
3860 pr_err("%s: reading input file", __func__);
3861 return -1;
3862 }
3863 if (session->repipe) {
3864 int retw = write(STDOUT_FILENO, buf, padding);
3865 if (retw <= 0 || retw != padding) {
3866 pr_err("%s: repiping tracing data padding", __func__);
3867 return -1;
3868 }
3869 }
3870
3871 if (size_read + padding != size) {
3872 pr_err("%s: tracing data size mismatch", __func__);
3873 return -1;
3874 }
3875
3876 perf_evlist__prepare_tracepoint_events(session->evlist,
3877 session->tevent.pevent);
3878
3879 return size_read + padding;
3880}
3881
3882int perf_event__synthesize_build_id(struct perf_tool *tool,
3883 struct dso *pos, u16 misc,
3884 perf_event__handler_t process,
3885 struct machine *machine)
3886{
3887 union perf_event ev;
3888 size_t len;
3889 int err = 0;
3890
3891 if (!pos->hit)
3892 return err;
3893
3894 memset(&ev, 0, sizeof(ev));
3895
3896 len = pos->long_name_len + 1;
3897 len = PERF_ALIGN(len, NAME_ALIGN);
3898 memcpy(&ev.build_id.build_id, pos->build_id, sizeof(pos->build_id));
3899 ev.build_id.header.type = PERF_RECORD_HEADER_BUILD_ID;
3900 ev.build_id.header.misc = misc;
3901 ev.build_id.pid = machine->pid;
3902 ev.build_id.header.size = sizeof(ev.build_id) + len;
3903 memcpy(&ev.build_id.filename, pos->long_name, pos->long_name_len);
3904
3905 err = process(tool, &ev, NULL, machine);
3906
3907 return err;
3908}
3909
3910int perf_event__process_build_id(struct perf_tool *tool __maybe_unused,
3911 union perf_event *event,
3912 struct perf_session *session)
3913{
3914 __event_process_build_id(&event->build_id,
3915 event->build_id.filename,
3916 session);
3917 return 0;
3918}
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
28#include "dso.h"
29#include "evlist.h"
30#include "evsel.h"
31#include "util/evsel_fprintf.h"
32#include "header.h"
33#include "memswap.h"
34#include "trace-event.h"
35#include "session.h"
36#include "symbol.h"
37#include "debug.h"
38#include "cpumap.h"
39#include "pmu.h"
40#include "vdso.h"
41#include "strbuf.h"
42#include "build-id.h"
43#include "data.h"
44#include <api/fs/fs.h>
45#include "asm/bug.h"
46#include "tool.h"
47#include "time-utils.h"
48#include "units.h"
49#include "util/util.h" // perf_exe()
50#include "cputopo.h"
51#include "bpf-event.h"
52#include "bpf-utils.h"
53#include "clockid.h"
54#include "pmu-hybrid.h"
55
56#include <linux/ctype.h>
57#include <internal/lib.h>
58
59#ifdef HAVE_LIBTRACEEVENT
60#include <traceevent/event-parse.h>
61#endif
62
63/*
64 * magic2 = "PERFILE2"
65 * must be a numerical value to let the endianness
66 * determine the memory layout. That way we are able
67 * to detect endianness when reading the perf.data file
68 * back.
69 *
70 * we check for legacy (PERFFILE) format.
71 */
72static const char *__perf_magic1 = "PERFFILE";
73static const u64 __perf_magic2 = 0x32454c4946524550ULL;
74static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
75
76#define PERF_MAGIC __perf_magic2
77
78const char perf_version_string[] = PERF_VERSION;
79
80struct perf_file_attr {
81 struct perf_event_attr attr;
82 struct perf_file_section ids;
83};
84
85void perf_header__set_feat(struct perf_header *header, int feat)
86{
87 __set_bit(feat, header->adds_features);
88}
89
90void perf_header__clear_feat(struct perf_header *header, int feat)
91{
92 __clear_bit(feat, header->adds_features);
93}
94
95bool perf_header__has_feat(const struct perf_header *header, int feat)
96{
97 return test_bit(feat, header->adds_features);
98}
99
100static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
101{
102 ssize_t ret = writen(ff->fd, buf, size);
103
104 if (ret != (ssize_t)size)
105 return ret < 0 ? (int)ret : -1;
106 return 0;
107}
108
109static int __do_write_buf(struct feat_fd *ff, const void *buf, size_t size)
110{
111 /* struct perf_event_header::size is u16 */
112 const size_t max_size = 0xffff - sizeof(struct perf_event_header);
113 size_t new_size = ff->size;
114 void *addr;
115
116 if (size + ff->offset > max_size)
117 return -E2BIG;
118
119 while (size > (new_size - ff->offset))
120 new_size <<= 1;
121 new_size = min(max_size, new_size);
122
123 if (ff->size < new_size) {
124 addr = realloc(ff->buf, new_size);
125 if (!addr)
126 return -ENOMEM;
127 ff->buf = addr;
128 ff->size = new_size;
129 }
130
131 memcpy(ff->buf + ff->offset, buf, size);
132 ff->offset += size;
133
134 return 0;
135}
136
137/* Return: 0 if succeeded, -ERR if failed. */
138int do_write(struct feat_fd *ff, const void *buf, size_t size)
139{
140 if (!ff->buf)
141 return __do_write_fd(ff, buf, size);
142 return __do_write_buf(ff, buf, size);
143}
144
145/* Return: 0 if succeeded, -ERR if failed. */
146static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
147{
148 u64 *p = (u64 *) set;
149 int i, ret;
150
151 ret = do_write(ff, &size, sizeof(size));
152 if (ret < 0)
153 return ret;
154
155 for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
156 ret = do_write(ff, p + i, sizeof(*p));
157 if (ret < 0)
158 return ret;
159 }
160
161 return 0;
162}
163
164/* Return: 0 if succeeded, -ERR if failed. */
165int write_padded(struct feat_fd *ff, const void *bf,
166 size_t count, size_t count_aligned)
167{
168 static const char zero_buf[NAME_ALIGN];
169 int err = do_write(ff, bf, count);
170
171 if (!err)
172 err = do_write(ff, zero_buf, count_aligned - count);
173
174 return err;
175}
176
177#define string_size(str) \
178 (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
179
180/* Return: 0 if succeeded, -ERR if failed. */
181static int do_write_string(struct feat_fd *ff, const char *str)
182{
183 u32 len, olen;
184 int ret;
185
186 olen = strlen(str) + 1;
187 len = PERF_ALIGN(olen, NAME_ALIGN);
188
189 /* write len, incl. \0 */
190 ret = do_write(ff, &len, sizeof(len));
191 if (ret < 0)
192 return ret;
193
194 return write_padded(ff, str, olen, len);
195}
196
197static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
198{
199 ssize_t ret = readn(ff->fd, addr, size);
200
201 if (ret != size)
202 return ret < 0 ? (int)ret : -1;
203 return 0;
204}
205
206static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
207{
208 if (size > (ssize_t)ff->size - ff->offset)
209 return -1;
210
211 memcpy(addr, ff->buf + ff->offset, size);
212 ff->offset += size;
213
214 return 0;
215
216}
217
218static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
219{
220 if (!ff->buf)
221 return __do_read_fd(ff, addr, size);
222 return __do_read_buf(ff, addr, size);
223}
224
225static int do_read_u32(struct feat_fd *ff, u32 *addr)
226{
227 int ret;
228
229 ret = __do_read(ff, addr, sizeof(*addr));
230 if (ret)
231 return ret;
232
233 if (ff->ph->needs_swap)
234 *addr = bswap_32(*addr);
235 return 0;
236}
237
238static int do_read_u64(struct feat_fd *ff, u64 *addr)
239{
240 int ret;
241
242 ret = __do_read(ff, addr, sizeof(*addr));
243 if (ret)
244 return ret;
245
246 if (ff->ph->needs_swap)
247 *addr = bswap_64(*addr);
248 return 0;
249}
250
251static char *do_read_string(struct feat_fd *ff)
252{
253 u32 len;
254 char *buf;
255
256 if (do_read_u32(ff, &len))
257 return NULL;
258
259 buf = malloc(len);
260 if (!buf)
261 return NULL;
262
263 if (!__do_read(ff, buf, len)) {
264 /*
265 * strings are padded by zeroes
266 * thus the actual strlen of buf
267 * may be less than len
268 */
269 return buf;
270 }
271
272 free(buf);
273 return NULL;
274}
275
276/* Return: 0 if succeeded, -ERR if failed. */
277static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
278{
279 unsigned long *set;
280 u64 size, *p;
281 int i, ret;
282
283 ret = do_read_u64(ff, &size);
284 if (ret)
285 return ret;
286
287 set = bitmap_zalloc(size);
288 if (!set)
289 return -ENOMEM;
290
291 p = (u64 *) set;
292
293 for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
294 ret = do_read_u64(ff, p + i);
295 if (ret < 0) {
296 free(set);
297 return ret;
298 }
299 }
300
301 *pset = set;
302 *psize = size;
303 return 0;
304}
305
306#ifdef HAVE_LIBTRACEEVENT
307static int write_tracing_data(struct feat_fd *ff,
308 struct evlist *evlist)
309{
310 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
311 return -1;
312
313 return read_tracing_data(ff->fd, &evlist->core.entries);
314}
315#endif
316
317static int write_build_id(struct feat_fd *ff,
318 struct evlist *evlist __maybe_unused)
319{
320 struct perf_session *session;
321 int err;
322
323 session = container_of(ff->ph, struct perf_session, header);
324
325 if (!perf_session__read_build_ids(session, true))
326 return -1;
327
328 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
329 return -1;
330
331 err = perf_session__write_buildid_table(session, ff);
332 if (err < 0) {
333 pr_debug("failed to write buildid table\n");
334 return err;
335 }
336 perf_session__cache_build_ids(session);
337
338 return 0;
339}
340
341static int write_hostname(struct feat_fd *ff,
342 struct evlist *evlist __maybe_unused)
343{
344 struct utsname uts;
345 int ret;
346
347 ret = uname(&uts);
348 if (ret < 0)
349 return -1;
350
351 return do_write_string(ff, uts.nodename);
352}
353
354static int write_osrelease(struct feat_fd *ff,
355 struct evlist *evlist __maybe_unused)
356{
357 struct utsname uts;
358 int ret;
359
360 ret = uname(&uts);
361 if (ret < 0)
362 return -1;
363
364 return do_write_string(ff, uts.release);
365}
366
367static int write_arch(struct feat_fd *ff,
368 struct evlist *evlist __maybe_unused)
369{
370 struct utsname uts;
371 int ret;
372
373 ret = uname(&uts);
374 if (ret < 0)
375 return -1;
376
377 return do_write_string(ff, uts.machine);
378}
379
380static int write_version(struct feat_fd *ff,
381 struct evlist *evlist __maybe_unused)
382{
383 return do_write_string(ff, perf_version_string);
384}
385
386static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
387{
388 FILE *file;
389 char *buf = NULL;
390 char *s, *p;
391 const char *search = cpuinfo_proc;
392 size_t len = 0;
393 int ret = -1;
394
395 if (!search)
396 return -1;
397
398 file = fopen("/proc/cpuinfo", "r");
399 if (!file)
400 return -1;
401
402 while (getline(&buf, &len, file) > 0) {
403 ret = strncmp(buf, search, strlen(search));
404 if (!ret)
405 break;
406 }
407
408 if (ret) {
409 ret = -1;
410 goto done;
411 }
412
413 s = buf;
414
415 p = strchr(buf, ':');
416 if (p && *(p+1) == ' ' && *(p+2))
417 s = p + 2;
418 p = strchr(s, '\n');
419 if (p)
420 *p = '\0';
421
422 /* squash extra space characters (branding string) */
423 p = s;
424 while (*p) {
425 if (isspace(*p)) {
426 char *r = p + 1;
427 char *q = skip_spaces(r);
428 *p = ' ';
429 if (q != (p+1))
430 while ((*r++ = *q++));
431 }
432 p++;
433 }
434 ret = do_write_string(ff, s);
435done:
436 free(buf);
437 fclose(file);
438 return ret;
439}
440
441static int write_cpudesc(struct feat_fd *ff,
442 struct evlist *evlist __maybe_unused)
443{
444#if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
445#define CPUINFO_PROC { "cpu", }
446#elif defined(__s390__)
447#define CPUINFO_PROC { "vendor_id", }
448#elif defined(__sh__)
449#define CPUINFO_PROC { "cpu type", }
450#elif defined(__alpha__) || defined(__mips__)
451#define CPUINFO_PROC { "cpu model", }
452#elif defined(__arm__)
453#define CPUINFO_PROC { "model name", "Processor", }
454#elif defined(__arc__)
455#define CPUINFO_PROC { "Processor", }
456#elif defined(__xtensa__)
457#define CPUINFO_PROC { "core ID", }
458#else
459#define CPUINFO_PROC { "model name", }
460#endif
461 const char *cpuinfo_procs[] = CPUINFO_PROC;
462#undef CPUINFO_PROC
463 unsigned int i;
464
465 for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
466 int ret;
467 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
468 if (ret >= 0)
469 return ret;
470 }
471 return -1;
472}
473
474
475static int write_nrcpus(struct feat_fd *ff,
476 struct evlist *evlist __maybe_unused)
477{
478 long nr;
479 u32 nrc, nra;
480 int ret;
481
482 nrc = cpu__max_present_cpu().cpu;
483
484 nr = sysconf(_SC_NPROCESSORS_ONLN);
485 if (nr < 0)
486 return -1;
487
488 nra = (u32)(nr & UINT_MAX);
489
490 ret = do_write(ff, &nrc, sizeof(nrc));
491 if (ret < 0)
492 return ret;
493
494 return do_write(ff, &nra, sizeof(nra));
495}
496
497static int write_event_desc(struct feat_fd *ff,
498 struct evlist *evlist)
499{
500 struct evsel *evsel;
501 u32 nre, nri, sz;
502 int ret;
503
504 nre = evlist->core.nr_entries;
505
506 /*
507 * write number of events
508 */
509 ret = do_write(ff, &nre, sizeof(nre));
510 if (ret < 0)
511 return ret;
512
513 /*
514 * size of perf_event_attr struct
515 */
516 sz = (u32)sizeof(evsel->core.attr);
517 ret = do_write(ff, &sz, sizeof(sz));
518 if (ret < 0)
519 return ret;
520
521 evlist__for_each_entry(evlist, evsel) {
522 ret = do_write(ff, &evsel->core.attr, sz);
523 if (ret < 0)
524 return ret;
525 /*
526 * write number of unique id per event
527 * there is one id per instance of an event
528 *
529 * copy into an nri to be independent of the
530 * type of ids,
531 */
532 nri = evsel->core.ids;
533 ret = do_write(ff, &nri, sizeof(nri));
534 if (ret < 0)
535 return ret;
536
537 /*
538 * write event string as passed on cmdline
539 */
540 ret = do_write_string(ff, evsel__name(evsel));
541 if (ret < 0)
542 return ret;
543 /*
544 * write unique ids for this event
545 */
546 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
547 if (ret < 0)
548 return ret;
549 }
550 return 0;
551}
552
553static int write_cmdline(struct feat_fd *ff,
554 struct evlist *evlist __maybe_unused)
555{
556 char pbuf[MAXPATHLEN], *buf;
557 int i, ret, n;
558
559 /* actual path to perf binary */
560 buf = perf_exe(pbuf, MAXPATHLEN);
561
562 /* account for binary path */
563 n = perf_env.nr_cmdline + 1;
564
565 ret = do_write(ff, &n, sizeof(n));
566 if (ret < 0)
567 return ret;
568
569 ret = do_write_string(ff, buf);
570 if (ret < 0)
571 return ret;
572
573 for (i = 0 ; i < perf_env.nr_cmdline; i++) {
574 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
575 if (ret < 0)
576 return ret;
577 }
578 return 0;
579}
580
581
582static int write_cpu_topology(struct feat_fd *ff,
583 struct evlist *evlist __maybe_unused)
584{
585 struct cpu_topology *tp;
586 u32 i;
587 int ret, j;
588
589 tp = cpu_topology__new();
590 if (!tp)
591 return -1;
592
593 ret = do_write(ff, &tp->package_cpus_lists, sizeof(tp->package_cpus_lists));
594 if (ret < 0)
595 goto done;
596
597 for (i = 0; i < tp->package_cpus_lists; i++) {
598 ret = do_write_string(ff, tp->package_cpus_list[i]);
599 if (ret < 0)
600 goto done;
601 }
602 ret = do_write(ff, &tp->core_cpus_lists, sizeof(tp->core_cpus_lists));
603 if (ret < 0)
604 goto done;
605
606 for (i = 0; i < tp->core_cpus_lists; i++) {
607 ret = do_write_string(ff, tp->core_cpus_list[i]);
608 if (ret < 0)
609 break;
610 }
611
612 ret = perf_env__read_cpu_topology_map(&perf_env);
613 if (ret < 0)
614 goto done;
615
616 for (j = 0; j < perf_env.nr_cpus_avail; j++) {
617 ret = do_write(ff, &perf_env.cpu[j].core_id,
618 sizeof(perf_env.cpu[j].core_id));
619 if (ret < 0)
620 return ret;
621 ret = do_write(ff, &perf_env.cpu[j].socket_id,
622 sizeof(perf_env.cpu[j].socket_id));
623 if (ret < 0)
624 return ret;
625 }
626
627 if (!tp->die_cpus_lists)
628 goto done;
629
630 ret = do_write(ff, &tp->die_cpus_lists, sizeof(tp->die_cpus_lists));
631 if (ret < 0)
632 goto done;
633
634 for (i = 0; i < tp->die_cpus_lists; i++) {
635 ret = do_write_string(ff, tp->die_cpus_list[i]);
636 if (ret < 0)
637 goto done;
638 }
639
640 for (j = 0; j < perf_env.nr_cpus_avail; j++) {
641 ret = do_write(ff, &perf_env.cpu[j].die_id,
642 sizeof(perf_env.cpu[j].die_id));
643 if (ret < 0)
644 return ret;
645 }
646
647done:
648 cpu_topology__delete(tp);
649 return ret;
650}
651
652
653
654static int write_total_mem(struct feat_fd *ff,
655 struct evlist *evlist __maybe_unused)
656{
657 char *buf = NULL;
658 FILE *fp;
659 size_t len = 0;
660 int ret = -1, n;
661 uint64_t mem;
662
663 fp = fopen("/proc/meminfo", "r");
664 if (!fp)
665 return -1;
666
667 while (getline(&buf, &len, fp) > 0) {
668 ret = strncmp(buf, "MemTotal:", 9);
669 if (!ret)
670 break;
671 }
672 if (!ret) {
673 n = sscanf(buf, "%*s %"PRIu64, &mem);
674 if (n == 1)
675 ret = do_write(ff, &mem, sizeof(mem));
676 } else
677 ret = -1;
678 free(buf);
679 fclose(fp);
680 return ret;
681}
682
683static int write_numa_topology(struct feat_fd *ff,
684 struct evlist *evlist __maybe_unused)
685{
686 struct numa_topology *tp;
687 int ret = -1;
688 u32 i;
689
690 tp = numa_topology__new();
691 if (!tp)
692 return -ENOMEM;
693
694 ret = do_write(ff, &tp->nr, sizeof(u32));
695 if (ret < 0)
696 goto err;
697
698 for (i = 0; i < tp->nr; i++) {
699 struct numa_topology_node *n = &tp->nodes[i];
700
701 ret = do_write(ff, &n->node, sizeof(u32));
702 if (ret < 0)
703 goto err;
704
705 ret = do_write(ff, &n->mem_total, sizeof(u64));
706 if (ret)
707 goto err;
708
709 ret = do_write(ff, &n->mem_free, sizeof(u64));
710 if (ret)
711 goto err;
712
713 ret = do_write_string(ff, n->cpus);
714 if (ret < 0)
715 goto err;
716 }
717
718 ret = 0;
719
720err:
721 numa_topology__delete(tp);
722 return ret;
723}
724
725/*
726 * File format:
727 *
728 * struct pmu_mappings {
729 * u32 pmu_num;
730 * struct pmu_map {
731 * u32 type;
732 * char name[];
733 * }[pmu_num];
734 * };
735 */
736
737static int write_pmu_mappings(struct feat_fd *ff,
738 struct evlist *evlist __maybe_unused)
739{
740 struct perf_pmu *pmu = NULL;
741 u32 pmu_num = 0;
742 int ret;
743
744 /*
745 * Do a first pass to count number of pmu to avoid lseek so this
746 * works in pipe mode as well.
747 */
748 while ((pmu = perf_pmu__scan(pmu))) {
749 if (!pmu->name)
750 continue;
751 pmu_num++;
752 }
753
754 ret = do_write(ff, &pmu_num, sizeof(pmu_num));
755 if (ret < 0)
756 return ret;
757
758 while ((pmu = perf_pmu__scan(pmu))) {
759 if (!pmu->name)
760 continue;
761
762 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
763 if (ret < 0)
764 return ret;
765
766 ret = do_write_string(ff, pmu->name);
767 if (ret < 0)
768 return ret;
769 }
770
771 return 0;
772}
773
774/*
775 * File format:
776 *
777 * struct group_descs {
778 * u32 nr_groups;
779 * struct group_desc {
780 * char name[];
781 * u32 leader_idx;
782 * u32 nr_members;
783 * }[nr_groups];
784 * };
785 */
786static int write_group_desc(struct feat_fd *ff,
787 struct evlist *evlist)
788{
789 u32 nr_groups = evlist->core.nr_groups;
790 struct evsel *evsel;
791 int ret;
792
793 ret = do_write(ff, &nr_groups, sizeof(nr_groups));
794 if (ret < 0)
795 return ret;
796
797 evlist__for_each_entry(evlist, evsel) {
798 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
799 const char *name = evsel->group_name ?: "{anon_group}";
800 u32 leader_idx = evsel->core.idx;
801 u32 nr_members = evsel->core.nr_members;
802
803 ret = do_write_string(ff, name);
804 if (ret < 0)
805 return ret;
806
807 ret = do_write(ff, &leader_idx, sizeof(leader_idx));
808 if (ret < 0)
809 return ret;
810
811 ret = do_write(ff, &nr_members, sizeof(nr_members));
812 if (ret < 0)
813 return ret;
814 }
815 }
816 return 0;
817}
818
819/*
820 * Return the CPU id as a raw string.
821 *
822 * Each architecture should provide a more precise id string that
823 * can be use to match the architecture's "mapfile".
824 */
825char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
826{
827 return NULL;
828}
829
830/* Return zero when the cpuid from the mapfile.csv matches the
831 * cpuid string generated on this platform.
832 * Otherwise return non-zero.
833 */
834int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
835{
836 regex_t re;
837 regmatch_t pmatch[1];
838 int match;
839
840 if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
841 /* Warn unable to generate match particular string. */
842 pr_info("Invalid regular expression %s\n", mapcpuid);
843 return 1;
844 }
845
846 match = !regexec(&re, cpuid, 1, pmatch, 0);
847 regfree(&re);
848 if (match) {
849 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
850
851 /* Verify the entire string matched. */
852 if (match_len == strlen(cpuid))
853 return 0;
854 }
855 return 1;
856}
857
858/*
859 * default get_cpuid(): nothing gets recorded
860 * actual implementation must be in arch/$(SRCARCH)/util/header.c
861 */
862int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
863{
864 return ENOSYS; /* Not implemented */
865}
866
867static int write_cpuid(struct feat_fd *ff,
868 struct evlist *evlist __maybe_unused)
869{
870 char buffer[64];
871 int ret;
872
873 ret = get_cpuid(buffer, sizeof(buffer));
874 if (ret)
875 return -1;
876
877 return do_write_string(ff, buffer);
878}
879
880static int write_branch_stack(struct feat_fd *ff __maybe_unused,
881 struct evlist *evlist __maybe_unused)
882{
883 return 0;
884}
885
886static int write_auxtrace(struct feat_fd *ff,
887 struct evlist *evlist __maybe_unused)
888{
889 struct perf_session *session;
890 int err;
891
892 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
893 return -1;
894
895 session = container_of(ff->ph, struct perf_session, header);
896
897 err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
898 if (err < 0)
899 pr_err("Failed to write auxtrace index\n");
900 return err;
901}
902
903static int write_clockid(struct feat_fd *ff,
904 struct evlist *evlist __maybe_unused)
905{
906 return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
907 sizeof(ff->ph->env.clock.clockid_res_ns));
908}
909
910static int write_clock_data(struct feat_fd *ff,
911 struct evlist *evlist __maybe_unused)
912{
913 u64 *data64;
914 u32 data32;
915 int ret;
916
917 /* version */
918 data32 = 1;
919
920 ret = do_write(ff, &data32, sizeof(data32));
921 if (ret < 0)
922 return ret;
923
924 /* clockid */
925 data32 = ff->ph->env.clock.clockid;
926
927 ret = do_write(ff, &data32, sizeof(data32));
928 if (ret < 0)
929 return ret;
930
931 /* TOD ref time */
932 data64 = &ff->ph->env.clock.tod_ns;
933
934 ret = do_write(ff, data64, sizeof(*data64));
935 if (ret < 0)
936 return ret;
937
938 /* clockid ref time */
939 data64 = &ff->ph->env.clock.clockid_ns;
940
941 return do_write(ff, data64, sizeof(*data64));
942}
943
944static int write_hybrid_topology(struct feat_fd *ff,
945 struct evlist *evlist __maybe_unused)
946{
947 struct hybrid_topology *tp;
948 int ret;
949 u32 i;
950
951 tp = hybrid_topology__new();
952 if (!tp)
953 return -ENOENT;
954
955 ret = do_write(ff, &tp->nr, sizeof(u32));
956 if (ret < 0)
957 goto err;
958
959 for (i = 0; i < tp->nr; i++) {
960 struct hybrid_topology_node *n = &tp->nodes[i];
961
962 ret = do_write_string(ff, n->pmu_name);
963 if (ret < 0)
964 goto err;
965
966 ret = do_write_string(ff, n->cpus);
967 if (ret < 0)
968 goto err;
969 }
970
971 ret = 0;
972
973err:
974 hybrid_topology__delete(tp);
975 return ret;
976}
977
978static int write_dir_format(struct feat_fd *ff,
979 struct evlist *evlist __maybe_unused)
980{
981 struct perf_session *session;
982 struct perf_data *data;
983
984 session = container_of(ff->ph, struct perf_session, header);
985 data = session->data;
986
987 if (WARN_ON(!perf_data__is_dir(data)))
988 return -1;
989
990 return do_write(ff, &data->dir.version, sizeof(data->dir.version));
991}
992
993/*
994 * Check whether a CPU is online
995 *
996 * Returns:
997 * 1 -> if CPU is online
998 * 0 -> if CPU is offline
999 * -1 -> error case
1000 */
1001int is_cpu_online(unsigned int cpu)
1002{
1003 char *str;
1004 size_t strlen;
1005 char buf[256];
1006 int status = -1;
1007 struct stat statbuf;
1008
1009 snprintf(buf, sizeof(buf),
1010 "/sys/devices/system/cpu/cpu%d", cpu);
1011 if (stat(buf, &statbuf) != 0)
1012 return 0;
1013
1014 /*
1015 * Check if /sys/devices/system/cpu/cpux/online file
1016 * exists. Some cases cpu0 won't have online file since
1017 * it is not expected to be turned off generally.
1018 * In kernels without CONFIG_HOTPLUG_CPU, this
1019 * file won't exist
1020 */
1021 snprintf(buf, sizeof(buf),
1022 "/sys/devices/system/cpu/cpu%d/online", cpu);
1023 if (stat(buf, &statbuf) != 0)
1024 return 1;
1025
1026 /*
1027 * Read online file using sysfs__read_str.
1028 * If read or open fails, return -1.
1029 * If read succeeds, return value from file
1030 * which gets stored in "str"
1031 */
1032 snprintf(buf, sizeof(buf),
1033 "devices/system/cpu/cpu%d/online", cpu);
1034
1035 if (sysfs__read_str(buf, &str, &strlen) < 0)
1036 return status;
1037
1038 status = atoi(str);
1039
1040 free(str);
1041 return status;
1042}
1043
1044#ifdef HAVE_LIBBPF_SUPPORT
1045static int write_bpf_prog_info(struct feat_fd *ff,
1046 struct evlist *evlist __maybe_unused)
1047{
1048 struct perf_env *env = &ff->ph->env;
1049 struct rb_root *root;
1050 struct rb_node *next;
1051 int ret;
1052
1053 down_read(&env->bpf_progs.lock);
1054
1055 ret = do_write(ff, &env->bpf_progs.infos_cnt,
1056 sizeof(env->bpf_progs.infos_cnt));
1057 if (ret < 0)
1058 goto out;
1059
1060 root = &env->bpf_progs.infos;
1061 next = rb_first(root);
1062 while (next) {
1063 struct bpf_prog_info_node *node;
1064 size_t len;
1065
1066 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1067 next = rb_next(&node->rb_node);
1068 len = sizeof(struct perf_bpil) +
1069 node->info_linear->data_len;
1070
1071 /* before writing to file, translate address to offset */
1072 bpil_addr_to_offs(node->info_linear);
1073 ret = do_write(ff, node->info_linear, len);
1074 /*
1075 * translate back to address even when do_write() fails,
1076 * so that this function never changes the data.
1077 */
1078 bpil_offs_to_addr(node->info_linear);
1079 if (ret < 0)
1080 goto out;
1081 }
1082out:
1083 up_read(&env->bpf_progs.lock);
1084 return ret;
1085}
1086
1087static int write_bpf_btf(struct feat_fd *ff,
1088 struct evlist *evlist __maybe_unused)
1089{
1090 struct perf_env *env = &ff->ph->env;
1091 struct rb_root *root;
1092 struct rb_node *next;
1093 int ret;
1094
1095 down_read(&env->bpf_progs.lock);
1096
1097 ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1098 sizeof(env->bpf_progs.btfs_cnt));
1099
1100 if (ret < 0)
1101 goto out;
1102
1103 root = &env->bpf_progs.btfs;
1104 next = rb_first(root);
1105 while (next) {
1106 struct btf_node *node;
1107
1108 node = rb_entry(next, struct btf_node, rb_node);
1109 next = rb_next(&node->rb_node);
1110 ret = do_write(ff, &node->id,
1111 sizeof(u32) * 2 + node->data_size);
1112 if (ret < 0)
1113 goto out;
1114 }
1115out:
1116 up_read(&env->bpf_progs.lock);
1117 return ret;
1118}
1119#endif // HAVE_LIBBPF_SUPPORT
1120
1121static int cpu_cache_level__sort(const void *a, const void *b)
1122{
1123 struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1124 struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1125
1126 return cache_a->level - cache_b->level;
1127}
1128
1129static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1130{
1131 if (a->level != b->level)
1132 return false;
1133
1134 if (a->line_size != b->line_size)
1135 return false;
1136
1137 if (a->sets != b->sets)
1138 return false;
1139
1140 if (a->ways != b->ways)
1141 return false;
1142
1143 if (strcmp(a->type, b->type))
1144 return false;
1145
1146 if (strcmp(a->size, b->size))
1147 return false;
1148
1149 if (strcmp(a->map, b->map))
1150 return false;
1151
1152 return true;
1153}
1154
1155static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1156{
1157 char path[PATH_MAX], file[PATH_MAX];
1158 struct stat st;
1159 size_t len;
1160
1161 scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1162 scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1163
1164 if (stat(file, &st))
1165 return 1;
1166
1167 scnprintf(file, PATH_MAX, "%s/level", path);
1168 if (sysfs__read_int(file, (int *) &cache->level))
1169 return -1;
1170
1171 scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1172 if (sysfs__read_int(file, (int *) &cache->line_size))
1173 return -1;
1174
1175 scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1176 if (sysfs__read_int(file, (int *) &cache->sets))
1177 return -1;
1178
1179 scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1180 if (sysfs__read_int(file, (int *) &cache->ways))
1181 return -1;
1182
1183 scnprintf(file, PATH_MAX, "%s/type", path);
1184 if (sysfs__read_str(file, &cache->type, &len))
1185 return -1;
1186
1187 cache->type[len] = 0;
1188 cache->type = strim(cache->type);
1189
1190 scnprintf(file, PATH_MAX, "%s/size", path);
1191 if (sysfs__read_str(file, &cache->size, &len)) {
1192 zfree(&cache->type);
1193 return -1;
1194 }
1195
1196 cache->size[len] = 0;
1197 cache->size = strim(cache->size);
1198
1199 scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1200 if (sysfs__read_str(file, &cache->map, &len)) {
1201 zfree(&cache->size);
1202 zfree(&cache->type);
1203 return -1;
1204 }
1205
1206 cache->map[len] = 0;
1207 cache->map = strim(cache->map);
1208 return 0;
1209}
1210
1211static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1212{
1213 fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1214}
1215
1216#define MAX_CACHE_LVL 4
1217
1218static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1219{
1220 u32 i, cnt = 0;
1221 u32 nr, cpu;
1222 u16 level;
1223
1224 nr = cpu__max_cpu().cpu;
1225
1226 for (cpu = 0; cpu < nr; cpu++) {
1227 for (level = 0; level < MAX_CACHE_LVL; level++) {
1228 struct cpu_cache_level c;
1229 int err;
1230
1231 err = cpu_cache_level__read(&c, cpu, level);
1232 if (err < 0)
1233 return err;
1234
1235 if (err == 1)
1236 break;
1237
1238 for (i = 0; i < cnt; i++) {
1239 if (cpu_cache_level__cmp(&c, &caches[i]))
1240 break;
1241 }
1242
1243 if (i == cnt)
1244 caches[cnt++] = c;
1245 else
1246 cpu_cache_level__free(&c);
1247 }
1248 }
1249 *cntp = cnt;
1250 return 0;
1251}
1252
1253static int write_cache(struct feat_fd *ff,
1254 struct evlist *evlist __maybe_unused)
1255{
1256 u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL;
1257 struct cpu_cache_level caches[max_caches];
1258 u32 cnt = 0, i, version = 1;
1259 int ret;
1260
1261 ret = build_caches(caches, &cnt);
1262 if (ret)
1263 goto out;
1264
1265 qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1266
1267 ret = do_write(ff, &version, sizeof(u32));
1268 if (ret < 0)
1269 goto out;
1270
1271 ret = do_write(ff, &cnt, sizeof(u32));
1272 if (ret < 0)
1273 goto out;
1274
1275 for (i = 0; i < cnt; i++) {
1276 struct cpu_cache_level *c = &caches[i];
1277
1278 #define _W(v) \
1279 ret = do_write(ff, &c->v, sizeof(u32)); \
1280 if (ret < 0) \
1281 goto out;
1282
1283 _W(level)
1284 _W(line_size)
1285 _W(sets)
1286 _W(ways)
1287 #undef _W
1288
1289 #define _W(v) \
1290 ret = do_write_string(ff, (const char *) c->v); \
1291 if (ret < 0) \
1292 goto out;
1293
1294 _W(type)
1295 _W(size)
1296 _W(map)
1297 #undef _W
1298 }
1299
1300out:
1301 for (i = 0; i < cnt; i++)
1302 cpu_cache_level__free(&caches[i]);
1303 return ret;
1304}
1305
1306static int write_stat(struct feat_fd *ff __maybe_unused,
1307 struct evlist *evlist __maybe_unused)
1308{
1309 return 0;
1310}
1311
1312static int write_sample_time(struct feat_fd *ff,
1313 struct evlist *evlist)
1314{
1315 int ret;
1316
1317 ret = do_write(ff, &evlist->first_sample_time,
1318 sizeof(evlist->first_sample_time));
1319 if (ret < 0)
1320 return ret;
1321
1322 return do_write(ff, &evlist->last_sample_time,
1323 sizeof(evlist->last_sample_time));
1324}
1325
1326
1327static int memory_node__read(struct memory_node *n, unsigned long idx)
1328{
1329 unsigned int phys, size = 0;
1330 char path[PATH_MAX];
1331 struct dirent *ent;
1332 DIR *dir;
1333
1334#define for_each_memory(mem, dir) \
1335 while ((ent = readdir(dir))) \
1336 if (strcmp(ent->d_name, ".") && \
1337 strcmp(ent->d_name, "..") && \
1338 sscanf(ent->d_name, "memory%u", &mem) == 1)
1339
1340 scnprintf(path, PATH_MAX,
1341 "%s/devices/system/node/node%lu",
1342 sysfs__mountpoint(), idx);
1343
1344 dir = opendir(path);
1345 if (!dir) {
1346 pr_warning("failed: can't open memory sysfs data\n");
1347 return -1;
1348 }
1349
1350 for_each_memory(phys, dir) {
1351 size = max(phys, size);
1352 }
1353
1354 size++;
1355
1356 n->set = bitmap_zalloc(size);
1357 if (!n->set) {
1358 closedir(dir);
1359 return -ENOMEM;
1360 }
1361
1362 n->node = idx;
1363 n->size = size;
1364
1365 rewinddir(dir);
1366
1367 for_each_memory(phys, dir) {
1368 __set_bit(phys, n->set);
1369 }
1370
1371 closedir(dir);
1372 return 0;
1373}
1374
1375static int memory_node__sort(const void *a, const void *b)
1376{
1377 const struct memory_node *na = a;
1378 const struct memory_node *nb = b;
1379
1380 return na->node - nb->node;
1381}
1382
1383static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
1384{
1385 char path[PATH_MAX];
1386 struct dirent *ent;
1387 DIR *dir;
1388 u64 cnt = 0;
1389 int ret = 0;
1390
1391 scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1392 sysfs__mountpoint());
1393
1394 dir = opendir(path);
1395 if (!dir) {
1396 pr_debug2("%s: couldn't read %s, does this arch have topology information?\n",
1397 __func__, path);
1398 return -1;
1399 }
1400
1401 while (!ret && (ent = readdir(dir))) {
1402 unsigned int idx;
1403 int r;
1404
1405 if (!strcmp(ent->d_name, ".") ||
1406 !strcmp(ent->d_name, ".."))
1407 continue;
1408
1409 r = sscanf(ent->d_name, "node%u", &idx);
1410 if (r != 1)
1411 continue;
1412
1413 if (WARN_ONCE(cnt >= size,
1414 "failed to write MEM_TOPOLOGY, way too many nodes\n")) {
1415 closedir(dir);
1416 return -1;
1417 }
1418
1419 ret = memory_node__read(&nodes[cnt++], idx);
1420 }
1421
1422 *cntp = cnt;
1423 closedir(dir);
1424
1425 if (!ret)
1426 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1427
1428 return ret;
1429}
1430
1431#define MAX_MEMORY_NODES 2000
1432
1433/*
1434 * The MEM_TOPOLOGY holds physical memory map for every
1435 * node in system. The format of data is as follows:
1436 *
1437 * 0 - version | for future changes
1438 * 8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1439 * 16 - count | number of nodes
1440 *
1441 * For each node we store map of physical indexes for
1442 * each node:
1443 *
1444 * 32 - node id | node index
1445 * 40 - size | size of bitmap
1446 * 48 - bitmap | bitmap of memory indexes that belongs to node
1447 */
1448static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1449 struct evlist *evlist __maybe_unused)
1450{
1451 static struct memory_node nodes[MAX_MEMORY_NODES];
1452 u64 bsize, version = 1, i, nr;
1453 int ret;
1454
1455 ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1456 (unsigned long long *) &bsize);
1457 if (ret)
1458 return ret;
1459
1460 ret = build_mem_topology(&nodes[0], MAX_MEMORY_NODES, &nr);
1461 if (ret)
1462 return ret;
1463
1464 ret = do_write(ff, &version, sizeof(version));
1465 if (ret < 0)
1466 goto out;
1467
1468 ret = do_write(ff, &bsize, sizeof(bsize));
1469 if (ret < 0)
1470 goto out;
1471
1472 ret = do_write(ff, &nr, sizeof(nr));
1473 if (ret < 0)
1474 goto out;
1475
1476 for (i = 0; i < nr; i++) {
1477 struct memory_node *n = &nodes[i];
1478
1479 #define _W(v) \
1480 ret = do_write(ff, &n->v, sizeof(n->v)); \
1481 if (ret < 0) \
1482 goto out;
1483
1484 _W(node)
1485 _W(size)
1486
1487 #undef _W
1488
1489 ret = do_write_bitmap(ff, n->set, n->size);
1490 if (ret < 0)
1491 goto out;
1492 }
1493
1494out:
1495 return ret;
1496}
1497
1498static int write_compressed(struct feat_fd *ff __maybe_unused,
1499 struct evlist *evlist __maybe_unused)
1500{
1501 int ret;
1502
1503 ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1504 if (ret)
1505 return ret;
1506
1507 ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1508 if (ret)
1509 return ret;
1510
1511 ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1512 if (ret)
1513 return ret;
1514
1515 ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1516 if (ret)
1517 return ret;
1518
1519 return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1520}
1521
1522static int __write_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1523 bool write_pmu)
1524{
1525 struct perf_pmu_caps *caps = NULL;
1526 int ret;
1527
1528 ret = do_write(ff, &pmu->nr_caps, sizeof(pmu->nr_caps));
1529 if (ret < 0)
1530 return ret;
1531
1532 list_for_each_entry(caps, &pmu->caps, list) {
1533 ret = do_write_string(ff, caps->name);
1534 if (ret < 0)
1535 return ret;
1536
1537 ret = do_write_string(ff, caps->value);
1538 if (ret < 0)
1539 return ret;
1540 }
1541
1542 if (write_pmu) {
1543 ret = do_write_string(ff, pmu->name);
1544 if (ret < 0)
1545 return ret;
1546 }
1547
1548 return ret;
1549}
1550
1551static int write_cpu_pmu_caps(struct feat_fd *ff,
1552 struct evlist *evlist __maybe_unused)
1553{
1554 struct perf_pmu *cpu_pmu = perf_pmu__find("cpu");
1555 int ret;
1556
1557 if (!cpu_pmu)
1558 return -ENOENT;
1559
1560 ret = perf_pmu__caps_parse(cpu_pmu);
1561 if (ret < 0)
1562 return ret;
1563
1564 return __write_pmu_caps(ff, cpu_pmu, false);
1565}
1566
1567static int write_pmu_caps(struct feat_fd *ff,
1568 struct evlist *evlist __maybe_unused)
1569{
1570 struct perf_pmu *pmu = NULL;
1571 int nr_pmu = 0;
1572 int ret;
1573
1574 while ((pmu = perf_pmu__scan(pmu))) {
1575 if (!pmu->name || !strcmp(pmu->name, "cpu") ||
1576 perf_pmu__caps_parse(pmu) <= 0)
1577 continue;
1578 nr_pmu++;
1579 }
1580
1581 ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1582 if (ret < 0)
1583 return ret;
1584
1585 if (!nr_pmu)
1586 return 0;
1587
1588 /*
1589 * Write hybrid pmu caps first to maintain compatibility with
1590 * older perf tool.
1591 */
1592 pmu = NULL;
1593 perf_pmu__for_each_hybrid_pmu(pmu) {
1594 ret = __write_pmu_caps(ff, pmu, true);
1595 if (ret < 0)
1596 return ret;
1597 }
1598
1599 pmu = NULL;
1600 while ((pmu = perf_pmu__scan(pmu))) {
1601 if (!pmu->name || !strcmp(pmu->name, "cpu") ||
1602 !pmu->nr_caps || perf_pmu__is_hybrid(pmu->name))
1603 continue;
1604
1605 ret = __write_pmu_caps(ff, pmu, true);
1606 if (ret < 0)
1607 return ret;
1608 }
1609 return 0;
1610}
1611
1612static void print_hostname(struct feat_fd *ff, FILE *fp)
1613{
1614 fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1615}
1616
1617static void print_osrelease(struct feat_fd *ff, FILE *fp)
1618{
1619 fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1620}
1621
1622static void print_arch(struct feat_fd *ff, FILE *fp)
1623{
1624 fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1625}
1626
1627static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1628{
1629 fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1630}
1631
1632static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1633{
1634 fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1635 fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1636}
1637
1638static void print_version(struct feat_fd *ff, FILE *fp)
1639{
1640 fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1641}
1642
1643static void print_cmdline(struct feat_fd *ff, FILE *fp)
1644{
1645 int nr, i;
1646
1647 nr = ff->ph->env.nr_cmdline;
1648
1649 fprintf(fp, "# cmdline : ");
1650
1651 for (i = 0; i < nr; i++) {
1652 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1653 if (!argv_i) {
1654 fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1655 } else {
1656 char *mem = argv_i;
1657 do {
1658 char *quote = strchr(argv_i, '\'');
1659 if (!quote)
1660 break;
1661 *quote++ = '\0';
1662 fprintf(fp, "%s\\\'", argv_i);
1663 argv_i = quote;
1664 } while (1);
1665 fprintf(fp, "%s ", argv_i);
1666 free(mem);
1667 }
1668 }
1669 fputc('\n', fp);
1670}
1671
1672static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1673{
1674 struct perf_header *ph = ff->ph;
1675 int cpu_nr = ph->env.nr_cpus_avail;
1676 int nr, i;
1677 char *str;
1678
1679 nr = ph->env.nr_sibling_cores;
1680 str = ph->env.sibling_cores;
1681
1682 for (i = 0; i < nr; i++) {
1683 fprintf(fp, "# sibling sockets : %s\n", str);
1684 str += strlen(str) + 1;
1685 }
1686
1687 if (ph->env.nr_sibling_dies) {
1688 nr = ph->env.nr_sibling_dies;
1689 str = ph->env.sibling_dies;
1690
1691 for (i = 0; i < nr; i++) {
1692 fprintf(fp, "# sibling dies : %s\n", str);
1693 str += strlen(str) + 1;
1694 }
1695 }
1696
1697 nr = ph->env.nr_sibling_threads;
1698 str = ph->env.sibling_threads;
1699
1700 for (i = 0; i < nr; i++) {
1701 fprintf(fp, "# sibling threads : %s\n", str);
1702 str += strlen(str) + 1;
1703 }
1704
1705 if (ph->env.nr_sibling_dies) {
1706 if (ph->env.cpu != NULL) {
1707 for (i = 0; i < cpu_nr; i++)
1708 fprintf(fp, "# CPU %d: Core ID %d, "
1709 "Die ID %d, Socket ID %d\n",
1710 i, ph->env.cpu[i].core_id,
1711 ph->env.cpu[i].die_id,
1712 ph->env.cpu[i].socket_id);
1713 } else
1714 fprintf(fp, "# Core ID, Die ID and Socket ID "
1715 "information is not available\n");
1716 } else {
1717 if (ph->env.cpu != NULL) {
1718 for (i = 0; i < cpu_nr; i++)
1719 fprintf(fp, "# CPU %d: Core ID %d, "
1720 "Socket ID %d\n",
1721 i, ph->env.cpu[i].core_id,
1722 ph->env.cpu[i].socket_id);
1723 } else
1724 fprintf(fp, "# Core ID and Socket ID "
1725 "information is not available\n");
1726 }
1727}
1728
1729static void print_clockid(struct feat_fd *ff, FILE *fp)
1730{
1731 fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1732 ff->ph->env.clock.clockid_res_ns * 1000);
1733}
1734
1735static void print_clock_data(struct feat_fd *ff, FILE *fp)
1736{
1737 struct timespec clockid_ns;
1738 char tstr[64], date[64];
1739 struct timeval tod_ns;
1740 clockid_t clockid;
1741 struct tm ltime;
1742 u64 ref;
1743
1744 if (!ff->ph->env.clock.enabled) {
1745 fprintf(fp, "# reference time disabled\n");
1746 return;
1747 }
1748
1749 /* Compute TOD time. */
1750 ref = ff->ph->env.clock.tod_ns;
1751 tod_ns.tv_sec = ref / NSEC_PER_SEC;
1752 ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1753 tod_ns.tv_usec = ref / NSEC_PER_USEC;
1754
1755 /* Compute clockid time. */
1756 ref = ff->ph->env.clock.clockid_ns;
1757 clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1758 ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1759 clockid_ns.tv_nsec = ref;
1760
1761 clockid = ff->ph->env.clock.clockid;
1762
1763 if (localtime_r(&tod_ns.tv_sec, <ime) == NULL)
1764 snprintf(tstr, sizeof(tstr), "<error>");
1765 else {
1766 strftime(date, sizeof(date), "%F %T", <ime);
1767 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1768 date, (int) tod_ns.tv_usec);
1769 }
1770
1771 fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1772 fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1773 tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1774 (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1775 clockid_name(clockid));
1776}
1777
1778static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1779{
1780 int i;
1781 struct hybrid_node *n;
1782
1783 fprintf(fp, "# hybrid cpu system:\n");
1784 for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1785 n = &ff->ph->env.hybrid_nodes[i];
1786 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1787 }
1788}
1789
1790static void print_dir_format(struct feat_fd *ff, FILE *fp)
1791{
1792 struct perf_session *session;
1793 struct perf_data *data;
1794
1795 session = container_of(ff->ph, struct perf_session, header);
1796 data = session->data;
1797
1798 fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1799}
1800
1801#ifdef HAVE_LIBBPF_SUPPORT
1802static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1803{
1804 struct perf_env *env = &ff->ph->env;
1805 struct rb_root *root;
1806 struct rb_node *next;
1807
1808 down_read(&env->bpf_progs.lock);
1809
1810 root = &env->bpf_progs.infos;
1811 next = rb_first(root);
1812
1813 while (next) {
1814 struct bpf_prog_info_node *node;
1815
1816 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1817 next = rb_next(&node->rb_node);
1818
1819 bpf_event__print_bpf_prog_info(&node->info_linear->info,
1820 env, fp);
1821 }
1822
1823 up_read(&env->bpf_progs.lock);
1824}
1825
1826static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1827{
1828 struct perf_env *env = &ff->ph->env;
1829 struct rb_root *root;
1830 struct rb_node *next;
1831
1832 down_read(&env->bpf_progs.lock);
1833
1834 root = &env->bpf_progs.btfs;
1835 next = rb_first(root);
1836
1837 while (next) {
1838 struct btf_node *node;
1839
1840 node = rb_entry(next, struct btf_node, rb_node);
1841 next = rb_next(&node->rb_node);
1842 fprintf(fp, "# btf info of id %u\n", node->id);
1843 }
1844
1845 up_read(&env->bpf_progs.lock);
1846}
1847#endif // HAVE_LIBBPF_SUPPORT
1848
1849static void free_event_desc(struct evsel *events)
1850{
1851 struct evsel *evsel;
1852
1853 if (!events)
1854 return;
1855
1856 for (evsel = events; evsel->core.attr.size; evsel++) {
1857 zfree(&evsel->name);
1858 zfree(&evsel->core.id);
1859 }
1860
1861 free(events);
1862}
1863
1864static bool perf_attr_check(struct perf_event_attr *attr)
1865{
1866 if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1867 pr_warning("Reserved bits are set unexpectedly. "
1868 "Please update perf tool.\n");
1869 return false;
1870 }
1871
1872 if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1873 pr_warning("Unknown sample type (0x%llx) is detected. "
1874 "Please update perf tool.\n",
1875 attr->sample_type);
1876 return false;
1877 }
1878
1879 if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1880 pr_warning("Unknown read format (0x%llx) is detected. "
1881 "Please update perf tool.\n",
1882 attr->read_format);
1883 return false;
1884 }
1885
1886 if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1887 (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1888 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1889 "Please update perf tool.\n",
1890 attr->branch_sample_type);
1891
1892 return false;
1893 }
1894
1895 return true;
1896}
1897
1898static struct evsel *read_event_desc(struct feat_fd *ff)
1899{
1900 struct evsel *evsel, *events = NULL;
1901 u64 *id;
1902 void *buf = NULL;
1903 u32 nre, sz, nr, i, j;
1904 size_t msz;
1905
1906 /* number of events */
1907 if (do_read_u32(ff, &nre))
1908 goto error;
1909
1910 if (do_read_u32(ff, &sz))
1911 goto error;
1912
1913 /* buffer to hold on file attr struct */
1914 buf = malloc(sz);
1915 if (!buf)
1916 goto error;
1917
1918 /* the last event terminates with evsel->core.attr.size == 0: */
1919 events = calloc(nre + 1, sizeof(*events));
1920 if (!events)
1921 goto error;
1922
1923 msz = sizeof(evsel->core.attr);
1924 if (sz < msz)
1925 msz = sz;
1926
1927 for (i = 0, evsel = events; i < nre; evsel++, i++) {
1928 evsel->core.idx = i;
1929
1930 /*
1931 * must read entire on-file attr struct to
1932 * sync up with layout.
1933 */
1934 if (__do_read(ff, buf, sz))
1935 goto error;
1936
1937 if (ff->ph->needs_swap)
1938 perf_event__attr_swap(buf);
1939
1940 memcpy(&evsel->core.attr, buf, msz);
1941
1942 if (!perf_attr_check(&evsel->core.attr))
1943 goto error;
1944
1945 if (do_read_u32(ff, &nr))
1946 goto error;
1947
1948 if (ff->ph->needs_swap)
1949 evsel->needs_swap = true;
1950
1951 evsel->name = do_read_string(ff);
1952 if (!evsel->name)
1953 goto error;
1954
1955 if (!nr)
1956 continue;
1957
1958 id = calloc(nr, sizeof(*id));
1959 if (!id)
1960 goto error;
1961 evsel->core.ids = nr;
1962 evsel->core.id = id;
1963
1964 for (j = 0 ; j < nr; j++) {
1965 if (do_read_u64(ff, id))
1966 goto error;
1967 id++;
1968 }
1969 }
1970out:
1971 free(buf);
1972 return events;
1973error:
1974 free_event_desc(events);
1975 events = NULL;
1976 goto out;
1977}
1978
1979static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
1980 void *priv __maybe_unused)
1981{
1982 return fprintf(fp, ", %s = %s", name, val);
1983}
1984
1985static void print_event_desc(struct feat_fd *ff, FILE *fp)
1986{
1987 struct evsel *evsel, *events;
1988 u32 j;
1989 u64 *id;
1990
1991 if (ff->events)
1992 events = ff->events;
1993 else
1994 events = read_event_desc(ff);
1995
1996 if (!events) {
1997 fprintf(fp, "# event desc: not available or unable to read\n");
1998 return;
1999 }
2000
2001 for (evsel = events; evsel->core.attr.size; evsel++) {
2002 fprintf(fp, "# event : name = %s, ", evsel->name);
2003
2004 if (evsel->core.ids) {
2005 fprintf(fp, ", id = {");
2006 for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
2007 if (j)
2008 fputc(',', fp);
2009 fprintf(fp, " %"PRIu64, *id);
2010 }
2011 fprintf(fp, " }");
2012 }
2013
2014 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
2015
2016 fputc('\n', fp);
2017 }
2018
2019 free_event_desc(events);
2020 ff->events = NULL;
2021}
2022
2023static void print_total_mem(struct feat_fd *ff, FILE *fp)
2024{
2025 fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
2026}
2027
2028static void print_numa_topology(struct feat_fd *ff, FILE *fp)
2029{
2030 int i;
2031 struct numa_node *n;
2032
2033 for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
2034 n = &ff->ph->env.numa_nodes[i];
2035
2036 fprintf(fp, "# node%u meminfo : total = %"PRIu64" kB,"
2037 " free = %"PRIu64" kB\n",
2038 n->node, n->mem_total, n->mem_free);
2039
2040 fprintf(fp, "# node%u cpu list : ", n->node);
2041 cpu_map__fprintf(n->map, fp);
2042 }
2043}
2044
2045static void print_cpuid(struct feat_fd *ff, FILE *fp)
2046{
2047 fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
2048}
2049
2050static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
2051{
2052 fprintf(fp, "# contains samples with branch stack\n");
2053}
2054
2055static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
2056{
2057 fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
2058}
2059
2060static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
2061{
2062 fprintf(fp, "# contains stat data\n");
2063}
2064
2065static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
2066{
2067 int i;
2068
2069 fprintf(fp, "# CPU cache info:\n");
2070 for (i = 0; i < ff->ph->env.caches_cnt; i++) {
2071 fprintf(fp, "# ");
2072 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
2073 }
2074}
2075
2076static void print_compressed(struct feat_fd *ff, FILE *fp)
2077{
2078 fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
2079 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
2080 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2081}
2082
2083static void __print_pmu_caps(FILE *fp, int nr_caps, char **caps, char *pmu_name)
2084{
2085 const char *delimiter = "";
2086 int i;
2087
2088 if (!nr_caps) {
2089 fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2090 return;
2091 }
2092
2093 fprintf(fp, "# %s pmu capabilities: ", pmu_name);
2094 for (i = 0; i < nr_caps; i++) {
2095 fprintf(fp, "%s%s", delimiter, caps[i]);
2096 delimiter = ", ";
2097 }
2098
2099 fprintf(fp, "\n");
2100}
2101
2102static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2103{
2104 __print_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2105 ff->ph->env.cpu_pmu_caps, (char *)"cpu");
2106}
2107
2108static void print_pmu_caps(struct feat_fd *ff, FILE *fp)
2109{
2110 struct pmu_caps *pmu_caps;
2111
2112 for (int i = 0; i < ff->ph->env.nr_pmus_with_caps; i++) {
2113 pmu_caps = &ff->ph->env.pmu_caps[i];
2114 __print_pmu_caps(fp, pmu_caps->nr_caps, pmu_caps->caps,
2115 pmu_caps->pmu_name);
2116 }
2117}
2118
2119static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2120{
2121 const char *delimiter = "# pmu mappings: ";
2122 char *str, *tmp;
2123 u32 pmu_num;
2124 u32 type;
2125
2126 pmu_num = ff->ph->env.nr_pmu_mappings;
2127 if (!pmu_num) {
2128 fprintf(fp, "# pmu mappings: not available\n");
2129 return;
2130 }
2131
2132 str = ff->ph->env.pmu_mappings;
2133
2134 while (pmu_num) {
2135 type = strtoul(str, &tmp, 0);
2136 if (*tmp != ':')
2137 goto error;
2138
2139 str = tmp + 1;
2140 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2141
2142 delimiter = ", ";
2143 str += strlen(str) + 1;
2144 pmu_num--;
2145 }
2146
2147 fprintf(fp, "\n");
2148
2149 if (!pmu_num)
2150 return;
2151error:
2152 fprintf(fp, "# pmu mappings: unable to read\n");
2153}
2154
2155static void print_group_desc(struct feat_fd *ff, FILE *fp)
2156{
2157 struct perf_session *session;
2158 struct evsel *evsel;
2159 u32 nr = 0;
2160
2161 session = container_of(ff->ph, struct perf_session, header);
2162
2163 evlist__for_each_entry(session->evlist, evsel) {
2164 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2165 fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2166
2167 nr = evsel->core.nr_members - 1;
2168 } else if (nr) {
2169 fprintf(fp, ",%s", evsel__name(evsel));
2170
2171 if (--nr == 0)
2172 fprintf(fp, "}\n");
2173 }
2174 }
2175}
2176
2177static void print_sample_time(struct feat_fd *ff, FILE *fp)
2178{
2179 struct perf_session *session;
2180 char time_buf[32];
2181 double d;
2182
2183 session = container_of(ff->ph, struct perf_session, header);
2184
2185 timestamp__scnprintf_usec(session->evlist->first_sample_time,
2186 time_buf, sizeof(time_buf));
2187 fprintf(fp, "# time of first sample : %s\n", time_buf);
2188
2189 timestamp__scnprintf_usec(session->evlist->last_sample_time,
2190 time_buf, sizeof(time_buf));
2191 fprintf(fp, "# time of last sample : %s\n", time_buf);
2192
2193 d = (double)(session->evlist->last_sample_time -
2194 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2195
2196 fprintf(fp, "# sample duration : %10.3f ms\n", d);
2197}
2198
2199static void memory_node__fprintf(struct memory_node *n,
2200 unsigned long long bsize, FILE *fp)
2201{
2202 char buf_map[100], buf_size[50];
2203 unsigned long long size;
2204
2205 size = bsize * bitmap_weight(n->set, n->size);
2206 unit_number__scnprintf(buf_size, 50, size);
2207
2208 bitmap_scnprintf(n->set, n->size, buf_map, 100);
2209 fprintf(fp, "# %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2210}
2211
2212static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2213{
2214 struct memory_node *nodes;
2215 int i, nr;
2216
2217 nodes = ff->ph->env.memory_nodes;
2218 nr = ff->ph->env.nr_memory_nodes;
2219
2220 fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2221 nr, ff->ph->env.memory_bsize);
2222
2223 for (i = 0; i < nr; i++) {
2224 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2225 }
2226}
2227
2228static int __event_process_build_id(struct perf_record_header_build_id *bev,
2229 char *filename,
2230 struct perf_session *session)
2231{
2232 int err = -1;
2233 struct machine *machine;
2234 u16 cpumode;
2235 struct dso *dso;
2236 enum dso_space_type dso_space;
2237
2238 machine = perf_session__findnew_machine(session, bev->pid);
2239 if (!machine)
2240 goto out;
2241
2242 cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2243
2244 switch (cpumode) {
2245 case PERF_RECORD_MISC_KERNEL:
2246 dso_space = DSO_SPACE__KERNEL;
2247 break;
2248 case PERF_RECORD_MISC_GUEST_KERNEL:
2249 dso_space = DSO_SPACE__KERNEL_GUEST;
2250 break;
2251 case PERF_RECORD_MISC_USER:
2252 case PERF_RECORD_MISC_GUEST_USER:
2253 dso_space = DSO_SPACE__USER;
2254 break;
2255 default:
2256 goto out;
2257 }
2258
2259 dso = machine__findnew_dso(machine, filename);
2260 if (dso != NULL) {
2261 char sbuild_id[SBUILD_ID_SIZE];
2262 struct build_id bid;
2263 size_t size = BUILD_ID_SIZE;
2264
2265 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2266 size = bev->size;
2267
2268 build_id__init(&bid, bev->data, size);
2269 dso__set_build_id(dso, &bid);
2270 dso->header_build_id = 1;
2271
2272 if (dso_space != DSO_SPACE__USER) {
2273 struct kmod_path m = { .name = NULL, };
2274
2275 if (!kmod_path__parse_name(&m, filename) && m.kmod)
2276 dso__set_module_info(dso, &m, machine);
2277
2278 dso->kernel = dso_space;
2279 free(m.name);
2280 }
2281
2282 build_id__sprintf(&dso->bid, sbuild_id);
2283 pr_debug("build id event received for %s: %s [%zu]\n",
2284 dso->long_name, sbuild_id, size);
2285 dso__put(dso);
2286 }
2287
2288 err = 0;
2289out:
2290 return err;
2291}
2292
2293static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2294 int input, u64 offset, u64 size)
2295{
2296 struct perf_session *session = container_of(header, struct perf_session, header);
2297 struct {
2298 struct perf_event_header header;
2299 u8 build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2300 char filename[0];
2301 } old_bev;
2302 struct perf_record_header_build_id bev;
2303 char filename[PATH_MAX];
2304 u64 limit = offset + size;
2305
2306 while (offset < limit) {
2307 ssize_t len;
2308
2309 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2310 return -1;
2311
2312 if (header->needs_swap)
2313 perf_event_header__bswap(&old_bev.header);
2314
2315 len = old_bev.header.size - sizeof(old_bev);
2316 if (readn(input, filename, len) != len)
2317 return -1;
2318
2319 bev.header = old_bev.header;
2320
2321 /*
2322 * As the pid is the missing value, we need to fill
2323 * it properly. The header.misc value give us nice hint.
2324 */
2325 bev.pid = HOST_KERNEL_ID;
2326 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2327 bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2328 bev.pid = DEFAULT_GUEST_KERNEL_ID;
2329
2330 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2331 __event_process_build_id(&bev, filename, session);
2332
2333 offset += bev.header.size;
2334 }
2335
2336 return 0;
2337}
2338
2339static int perf_header__read_build_ids(struct perf_header *header,
2340 int input, u64 offset, u64 size)
2341{
2342 struct perf_session *session = container_of(header, struct perf_session, header);
2343 struct perf_record_header_build_id bev;
2344 char filename[PATH_MAX];
2345 u64 limit = offset + size, orig_offset = offset;
2346 int err = -1;
2347
2348 while (offset < limit) {
2349 ssize_t len;
2350
2351 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2352 goto out;
2353
2354 if (header->needs_swap)
2355 perf_event_header__bswap(&bev.header);
2356
2357 len = bev.header.size - sizeof(bev);
2358 if (readn(input, filename, len) != len)
2359 goto out;
2360 /*
2361 * The a1645ce1 changeset:
2362 *
2363 * "perf: 'perf kvm' tool for monitoring guest performance from host"
2364 *
2365 * Added a field to struct perf_record_header_build_id that broke the file
2366 * format.
2367 *
2368 * Since the kernel build-id is the first entry, process the
2369 * table using the old format if the well known
2370 * '[kernel.kallsyms]' string for the kernel build-id has the
2371 * first 4 characters chopped off (where the pid_t sits).
2372 */
2373 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2374 if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2375 return -1;
2376 return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2377 }
2378
2379 __event_process_build_id(&bev, filename, session);
2380
2381 offset += bev.header.size;
2382 }
2383 err = 0;
2384out:
2385 return err;
2386}
2387
2388/* Macro for features that simply need to read and store a string. */
2389#define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2390static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2391{\
2392 free(ff->ph->env.__feat_env); \
2393 ff->ph->env.__feat_env = do_read_string(ff); \
2394 return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2395}
2396
2397FEAT_PROCESS_STR_FUN(hostname, hostname);
2398FEAT_PROCESS_STR_FUN(osrelease, os_release);
2399FEAT_PROCESS_STR_FUN(version, version);
2400FEAT_PROCESS_STR_FUN(arch, arch);
2401FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2402FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2403
2404#ifdef HAVE_LIBTRACEEVENT
2405static int process_tracing_data(struct feat_fd *ff, void *data)
2406{
2407 ssize_t ret = trace_report(ff->fd, data, false);
2408
2409 return ret < 0 ? -1 : 0;
2410}
2411#endif
2412
2413static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2414{
2415 if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2416 pr_debug("Failed to read buildids, continuing...\n");
2417 return 0;
2418}
2419
2420static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2421{
2422 int ret;
2423 u32 nr_cpus_avail, nr_cpus_online;
2424
2425 ret = do_read_u32(ff, &nr_cpus_avail);
2426 if (ret)
2427 return ret;
2428
2429 ret = do_read_u32(ff, &nr_cpus_online);
2430 if (ret)
2431 return ret;
2432 ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2433 ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2434 return 0;
2435}
2436
2437static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2438{
2439 u64 total_mem;
2440 int ret;
2441
2442 ret = do_read_u64(ff, &total_mem);
2443 if (ret)
2444 return -1;
2445 ff->ph->env.total_mem = (unsigned long long)total_mem;
2446 return 0;
2447}
2448
2449static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2450{
2451 struct evsel *evsel;
2452
2453 evlist__for_each_entry(evlist, evsel) {
2454 if (evsel->core.idx == idx)
2455 return evsel;
2456 }
2457
2458 return NULL;
2459}
2460
2461static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2462{
2463 struct evsel *evsel;
2464
2465 if (!event->name)
2466 return;
2467
2468 evsel = evlist__find_by_index(evlist, event->core.idx);
2469 if (!evsel)
2470 return;
2471
2472 if (evsel->name)
2473 return;
2474
2475 evsel->name = strdup(event->name);
2476}
2477
2478static int
2479process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2480{
2481 struct perf_session *session;
2482 struct evsel *evsel, *events = read_event_desc(ff);
2483
2484 if (!events)
2485 return 0;
2486
2487 session = container_of(ff->ph, struct perf_session, header);
2488
2489 if (session->data->is_pipe) {
2490 /* Save events for reading later by print_event_desc,
2491 * since they can't be read again in pipe mode. */
2492 ff->events = events;
2493 }
2494
2495 for (evsel = events; evsel->core.attr.size; evsel++)
2496 evlist__set_event_name(session->evlist, evsel);
2497
2498 if (!session->data->is_pipe)
2499 free_event_desc(events);
2500
2501 return 0;
2502}
2503
2504static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2505{
2506 char *str, *cmdline = NULL, **argv = NULL;
2507 u32 nr, i, len = 0;
2508
2509 if (do_read_u32(ff, &nr))
2510 return -1;
2511
2512 ff->ph->env.nr_cmdline = nr;
2513
2514 cmdline = zalloc(ff->size + nr + 1);
2515 if (!cmdline)
2516 return -1;
2517
2518 argv = zalloc(sizeof(char *) * (nr + 1));
2519 if (!argv)
2520 goto error;
2521
2522 for (i = 0; i < nr; i++) {
2523 str = do_read_string(ff);
2524 if (!str)
2525 goto error;
2526
2527 argv[i] = cmdline + len;
2528 memcpy(argv[i], str, strlen(str) + 1);
2529 len += strlen(str) + 1;
2530 free(str);
2531 }
2532 ff->ph->env.cmdline = cmdline;
2533 ff->ph->env.cmdline_argv = (const char **) argv;
2534 return 0;
2535
2536error:
2537 free(argv);
2538 free(cmdline);
2539 return -1;
2540}
2541
2542static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2543{
2544 u32 nr, i;
2545 char *str;
2546 struct strbuf sb;
2547 int cpu_nr = ff->ph->env.nr_cpus_avail;
2548 u64 size = 0;
2549 struct perf_header *ph = ff->ph;
2550 bool do_core_id_test = true;
2551
2552 ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2553 if (!ph->env.cpu)
2554 return -1;
2555
2556 if (do_read_u32(ff, &nr))
2557 goto free_cpu;
2558
2559 ph->env.nr_sibling_cores = nr;
2560 size += sizeof(u32);
2561 if (strbuf_init(&sb, 128) < 0)
2562 goto free_cpu;
2563
2564 for (i = 0; i < nr; i++) {
2565 str = do_read_string(ff);
2566 if (!str)
2567 goto error;
2568
2569 /* include a NULL character at the end */
2570 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2571 goto error;
2572 size += string_size(str);
2573 free(str);
2574 }
2575 ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2576
2577 if (do_read_u32(ff, &nr))
2578 return -1;
2579
2580 ph->env.nr_sibling_threads = nr;
2581 size += sizeof(u32);
2582
2583 for (i = 0; i < nr; i++) {
2584 str = do_read_string(ff);
2585 if (!str)
2586 goto error;
2587
2588 /* include a NULL character at the end */
2589 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2590 goto error;
2591 size += string_size(str);
2592 free(str);
2593 }
2594 ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2595
2596 /*
2597 * The header may be from old perf,
2598 * which doesn't include core id and socket id information.
2599 */
2600 if (ff->size <= size) {
2601 zfree(&ph->env.cpu);
2602 return 0;
2603 }
2604
2605 /* On s390 the socket_id number is not related to the numbers of cpus.
2606 * The socket_id number might be higher than the numbers of cpus.
2607 * This depends on the configuration.
2608 * AArch64 is the same.
2609 */
2610 if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2611 || !strncmp(ph->env.arch, "aarch64", 7)))
2612 do_core_id_test = false;
2613
2614 for (i = 0; i < (u32)cpu_nr; i++) {
2615 if (do_read_u32(ff, &nr))
2616 goto free_cpu;
2617
2618 ph->env.cpu[i].core_id = nr;
2619 size += sizeof(u32);
2620
2621 if (do_read_u32(ff, &nr))
2622 goto free_cpu;
2623
2624 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2625 pr_debug("socket_id number is too big."
2626 "You may need to upgrade the perf tool.\n");
2627 goto free_cpu;
2628 }
2629
2630 ph->env.cpu[i].socket_id = nr;
2631 size += sizeof(u32);
2632 }
2633
2634 /*
2635 * The header may be from old perf,
2636 * which doesn't include die information.
2637 */
2638 if (ff->size <= size)
2639 return 0;
2640
2641 if (do_read_u32(ff, &nr))
2642 return -1;
2643
2644 ph->env.nr_sibling_dies = nr;
2645 size += sizeof(u32);
2646
2647 for (i = 0; i < nr; i++) {
2648 str = do_read_string(ff);
2649 if (!str)
2650 goto error;
2651
2652 /* include a NULL character at the end */
2653 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2654 goto error;
2655 size += string_size(str);
2656 free(str);
2657 }
2658 ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2659
2660 for (i = 0; i < (u32)cpu_nr; i++) {
2661 if (do_read_u32(ff, &nr))
2662 goto free_cpu;
2663
2664 ph->env.cpu[i].die_id = nr;
2665 }
2666
2667 return 0;
2668
2669error:
2670 strbuf_release(&sb);
2671free_cpu:
2672 zfree(&ph->env.cpu);
2673 return -1;
2674}
2675
2676static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2677{
2678 struct numa_node *nodes, *n;
2679 u32 nr, i;
2680 char *str;
2681
2682 /* nr nodes */
2683 if (do_read_u32(ff, &nr))
2684 return -1;
2685
2686 nodes = zalloc(sizeof(*nodes) * nr);
2687 if (!nodes)
2688 return -ENOMEM;
2689
2690 for (i = 0; i < nr; i++) {
2691 n = &nodes[i];
2692
2693 /* node number */
2694 if (do_read_u32(ff, &n->node))
2695 goto error;
2696
2697 if (do_read_u64(ff, &n->mem_total))
2698 goto error;
2699
2700 if (do_read_u64(ff, &n->mem_free))
2701 goto error;
2702
2703 str = do_read_string(ff);
2704 if (!str)
2705 goto error;
2706
2707 n->map = perf_cpu_map__new(str);
2708 if (!n->map)
2709 goto error;
2710
2711 free(str);
2712 }
2713 ff->ph->env.nr_numa_nodes = nr;
2714 ff->ph->env.numa_nodes = nodes;
2715 return 0;
2716
2717error:
2718 free(nodes);
2719 return -1;
2720}
2721
2722static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2723{
2724 char *name;
2725 u32 pmu_num;
2726 u32 type;
2727 struct strbuf sb;
2728
2729 if (do_read_u32(ff, &pmu_num))
2730 return -1;
2731
2732 if (!pmu_num) {
2733 pr_debug("pmu mappings not available\n");
2734 return 0;
2735 }
2736
2737 ff->ph->env.nr_pmu_mappings = pmu_num;
2738 if (strbuf_init(&sb, 128) < 0)
2739 return -1;
2740
2741 while (pmu_num) {
2742 if (do_read_u32(ff, &type))
2743 goto error;
2744
2745 name = do_read_string(ff);
2746 if (!name)
2747 goto error;
2748
2749 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2750 goto error;
2751 /* include a NULL character at the end */
2752 if (strbuf_add(&sb, "", 1) < 0)
2753 goto error;
2754
2755 if (!strcmp(name, "msr"))
2756 ff->ph->env.msr_pmu_type = type;
2757
2758 free(name);
2759 pmu_num--;
2760 }
2761 ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2762 return 0;
2763
2764error:
2765 strbuf_release(&sb);
2766 return -1;
2767}
2768
2769static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2770{
2771 size_t ret = -1;
2772 u32 i, nr, nr_groups;
2773 struct perf_session *session;
2774 struct evsel *evsel, *leader = NULL;
2775 struct group_desc {
2776 char *name;
2777 u32 leader_idx;
2778 u32 nr_members;
2779 } *desc;
2780
2781 if (do_read_u32(ff, &nr_groups))
2782 return -1;
2783
2784 ff->ph->env.nr_groups = nr_groups;
2785 if (!nr_groups) {
2786 pr_debug("group desc not available\n");
2787 return 0;
2788 }
2789
2790 desc = calloc(nr_groups, sizeof(*desc));
2791 if (!desc)
2792 return -1;
2793
2794 for (i = 0; i < nr_groups; i++) {
2795 desc[i].name = do_read_string(ff);
2796 if (!desc[i].name)
2797 goto out_free;
2798
2799 if (do_read_u32(ff, &desc[i].leader_idx))
2800 goto out_free;
2801
2802 if (do_read_u32(ff, &desc[i].nr_members))
2803 goto out_free;
2804 }
2805
2806 /*
2807 * Rebuild group relationship based on the group_desc
2808 */
2809 session = container_of(ff->ph, struct perf_session, header);
2810 session->evlist->core.nr_groups = nr_groups;
2811
2812 i = nr = 0;
2813 evlist__for_each_entry(session->evlist, evsel) {
2814 if (evsel->core.idx == (int) desc[i].leader_idx) {
2815 evsel__set_leader(evsel, evsel);
2816 /* {anon_group} is a dummy name */
2817 if (strcmp(desc[i].name, "{anon_group}")) {
2818 evsel->group_name = desc[i].name;
2819 desc[i].name = NULL;
2820 }
2821 evsel->core.nr_members = desc[i].nr_members;
2822
2823 if (i >= nr_groups || nr > 0) {
2824 pr_debug("invalid group desc\n");
2825 goto out_free;
2826 }
2827
2828 leader = evsel;
2829 nr = evsel->core.nr_members - 1;
2830 i++;
2831 } else if (nr) {
2832 /* This is a group member */
2833 evsel__set_leader(evsel, leader);
2834
2835 nr--;
2836 }
2837 }
2838
2839 if (i != nr_groups || nr != 0) {
2840 pr_debug("invalid group desc\n");
2841 goto out_free;
2842 }
2843
2844 ret = 0;
2845out_free:
2846 for (i = 0; i < nr_groups; i++)
2847 zfree(&desc[i].name);
2848 free(desc);
2849
2850 return ret;
2851}
2852
2853static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2854{
2855 struct perf_session *session;
2856 int err;
2857
2858 session = container_of(ff->ph, struct perf_session, header);
2859
2860 err = auxtrace_index__process(ff->fd, ff->size, session,
2861 ff->ph->needs_swap);
2862 if (err < 0)
2863 pr_err("Failed to process auxtrace index\n");
2864 return err;
2865}
2866
2867static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2868{
2869 struct cpu_cache_level *caches;
2870 u32 cnt, i, version;
2871
2872 if (do_read_u32(ff, &version))
2873 return -1;
2874
2875 if (version != 1)
2876 return -1;
2877
2878 if (do_read_u32(ff, &cnt))
2879 return -1;
2880
2881 caches = zalloc(sizeof(*caches) * cnt);
2882 if (!caches)
2883 return -1;
2884
2885 for (i = 0; i < cnt; i++) {
2886 struct cpu_cache_level c;
2887
2888 #define _R(v) \
2889 if (do_read_u32(ff, &c.v))\
2890 goto out_free_caches; \
2891
2892 _R(level)
2893 _R(line_size)
2894 _R(sets)
2895 _R(ways)
2896 #undef _R
2897
2898 #define _R(v) \
2899 c.v = do_read_string(ff); \
2900 if (!c.v) \
2901 goto out_free_caches;
2902
2903 _R(type)
2904 _R(size)
2905 _R(map)
2906 #undef _R
2907
2908 caches[i] = c;
2909 }
2910
2911 ff->ph->env.caches = caches;
2912 ff->ph->env.caches_cnt = cnt;
2913 return 0;
2914out_free_caches:
2915 free(caches);
2916 return -1;
2917}
2918
2919static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2920{
2921 struct perf_session *session;
2922 u64 first_sample_time, last_sample_time;
2923 int ret;
2924
2925 session = container_of(ff->ph, struct perf_session, header);
2926
2927 ret = do_read_u64(ff, &first_sample_time);
2928 if (ret)
2929 return -1;
2930
2931 ret = do_read_u64(ff, &last_sample_time);
2932 if (ret)
2933 return -1;
2934
2935 session->evlist->first_sample_time = first_sample_time;
2936 session->evlist->last_sample_time = last_sample_time;
2937 return 0;
2938}
2939
2940static int process_mem_topology(struct feat_fd *ff,
2941 void *data __maybe_unused)
2942{
2943 struct memory_node *nodes;
2944 u64 version, i, nr, bsize;
2945 int ret = -1;
2946
2947 if (do_read_u64(ff, &version))
2948 return -1;
2949
2950 if (version != 1)
2951 return -1;
2952
2953 if (do_read_u64(ff, &bsize))
2954 return -1;
2955
2956 if (do_read_u64(ff, &nr))
2957 return -1;
2958
2959 nodes = zalloc(sizeof(*nodes) * nr);
2960 if (!nodes)
2961 return -1;
2962
2963 for (i = 0; i < nr; i++) {
2964 struct memory_node n;
2965
2966 #define _R(v) \
2967 if (do_read_u64(ff, &n.v)) \
2968 goto out; \
2969
2970 _R(node)
2971 _R(size)
2972
2973 #undef _R
2974
2975 if (do_read_bitmap(ff, &n.set, &n.size))
2976 goto out;
2977
2978 nodes[i] = n;
2979 }
2980
2981 ff->ph->env.memory_bsize = bsize;
2982 ff->ph->env.memory_nodes = nodes;
2983 ff->ph->env.nr_memory_nodes = nr;
2984 ret = 0;
2985
2986out:
2987 if (ret)
2988 free(nodes);
2989 return ret;
2990}
2991
2992static int process_clockid(struct feat_fd *ff,
2993 void *data __maybe_unused)
2994{
2995 if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
2996 return -1;
2997
2998 return 0;
2999}
3000
3001static int process_clock_data(struct feat_fd *ff,
3002 void *_data __maybe_unused)
3003{
3004 u32 data32;
3005 u64 data64;
3006
3007 /* version */
3008 if (do_read_u32(ff, &data32))
3009 return -1;
3010
3011 if (data32 != 1)
3012 return -1;
3013
3014 /* clockid */
3015 if (do_read_u32(ff, &data32))
3016 return -1;
3017
3018 ff->ph->env.clock.clockid = data32;
3019
3020 /* TOD ref time */
3021 if (do_read_u64(ff, &data64))
3022 return -1;
3023
3024 ff->ph->env.clock.tod_ns = data64;
3025
3026 /* clockid ref time */
3027 if (do_read_u64(ff, &data64))
3028 return -1;
3029
3030 ff->ph->env.clock.clockid_ns = data64;
3031 ff->ph->env.clock.enabled = true;
3032 return 0;
3033}
3034
3035static int process_hybrid_topology(struct feat_fd *ff,
3036 void *data __maybe_unused)
3037{
3038 struct hybrid_node *nodes, *n;
3039 u32 nr, i;
3040
3041 /* nr nodes */
3042 if (do_read_u32(ff, &nr))
3043 return -1;
3044
3045 nodes = zalloc(sizeof(*nodes) * nr);
3046 if (!nodes)
3047 return -ENOMEM;
3048
3049 for (i = 0; i < nr; i++) {
3050 n = &nodes[i];
3051
3052 n->pmu_name = do_read_string(ff);
3053 if (!n->pmu_name)
3054 goto error;
3055
3056 n->cpus = do_read_string(ff);
3057 if (!n->cpus)
3058 goto error;
3059 }
3060
3061 ff->ph->env.nr_hybrid_nodes = nr;
3062 ff->ph->env.hybrid_nodes = nodes;
3063 return 0;
3064
3065error:
3066 for (i = 0; i < nr; i++) {
3067 free(nodes[i].pmu_name);
3068 free(nodes[i].cpus);
3069 }
3070
3071 free(nodes);
3072 return -1;
3073}
3074
3075static int process_dir_format(struct feat_fd *ff,
3076 void *_data __maybe_unused)
3077{
3078 struct perf_session *session;
3079 struct perf_data *data;
3080
3081 session = container_of(ff->ph, struct perf_session, header);
3082 data = session->data;
3083
3084 if (WARN_ON(!perf_data__is_dir(data)))
3085 return -1;
3086
3087 return do_read_u64(ff, &data->dir.version);
3088}
3089
3090#ifdef HAVE_LIBBPF_SUPPORT
3091static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3092{
3093 struct bpf_prog_info_node *info_node;
3094 struct perf_env *env = &ff->ph->env;
3095 struct perf_bpil *info_linear;
3096 u32 count, i;
3097 int err = -1;
3098
3099 if (ff->ph->needs_swap) {
3100 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3101 return 0;
3102 }
3103
3104 if (do_read_u32(ff, &count))
3105 return -1;
3106
3107 down_write(&env->bpf_progs.lock);
3108
3109 for (i = 0; i < count; ++i) {
3110 u32 info_len, data_len;
3111
3112 info_linear = NULL;
3113 info_node = NULL;
3114 if (do_read_u32(ff, &info_len))
3115 goto out;
3116 if (do_read_u32(ff, &data_len))
3117 goto out;
3118
3119 if (info_len > sizeof(struct bpf_prog_info)) {
3120 pr_warning("detected invalid bpf_prog_info\n");
3121 goto out;
3122 }
3123
3124 info_linear = malloc(sizeof(struct perf_bpil) +
3125 data_len);
3126 if (!info_linear)
3127 goto out;
3128 info_linear->info_len = sizeof(struct bpf_prog_info);
3129 info_linear->data_len = data_len;
3130 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3131 goto out;
3132 if (__do_read(ff, &info_linear->info, info_len))
3133 goto out;
3134 if (info_len < sizeof(struct bpf_prog_info))
3135 memset(((void *)(&info_linear->info)) + info_len, 0,
3136 sizeof(struct bpf_prog_info) - info_len);
3137
3138 if (__do_read(ff, info_linear->data, data_len))
3139 goto out;
3140
3141 info_node = malloc(sizeof(struct bpf_prog_info_node));
3142 if (!info_node)
3143 goto out;
3144
3145 /* after reading from file, translate offset to address */
3146 bpil_offs_to_addr(info_linear);
3147 info_node->info_linear = info_linear;
3148 perf_env__insert_bpf_prog_info(env, info_node);
3149 }
3150
3151 up_write(&env->bpf_progs.lock);
3152 return 0;
3153out:
3154 free(info_linear);
3155 free(info_node);
3156 up_write(&env->bpf_progs.lock);
3157 return err;
3158}
3159
3160static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3161{
3162 struct perf_env *env = &ff->ph->env;
3163 struct btf_node *node = NULL;
3164 u32 count, i;
3165 int err = -1;
3166
3167 if (ff->ph->needs_swap) {
3168 pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3169 return 0;
3170 }
3171
3172 if (do_read_u32(ff, &count))
3173 return -1;
3174
3175 down_write(&env->bpf_progs.lock);
3176
3177 for (i = 0; i < count; ++i) {
3178 u32 id, data_size;
3179
3180 if (do_read_u32(ff, &id))
3181 goto out;
3182 if (do_read_u32(ff, &data_size))
3183 goto out;
3184
3185 node = malloc(sizeof(struct btf_node) + data_size);
3186 if (!node)
3187 goto out;
3188
3189 node->id = id;
3190 node->data_size = data_size;
3191
3192 if (__do_read(ff, node->data, data_size))
3193 goto out;
3194
3195 perf_env__insert_btf(env, node);
3196 node = NULL;
3197 }
3198
3199 err = 0;
3200out:
3201 up_write(&env->bpf_progs.lock);
3202 free(node);
3203 return err;
3204}
3205#endif // HAVE_LIBBPF_SUPPORT
3206
3207static int process_compressed(struct feat_fd *ff,
3208 void *data __maybe_unused)
3209{
3210 if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3211 return -1;
3212
3213 if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3214 return -1;
3215
3216 if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3217 return -1;
3218
3219 if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3220 return -1;
3221
3222 if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3223 return -1;
3224
3225 return 0;
3226}
3227
3228static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
3229 char ***caps, unsigned int *max_branches)
3230{
3231 char *name, *value, *ptr;
3232 u32 nr_pmu_caps, i;
3233
3234 *nr_caps = 0;
3235 *caps = NULL;
3236
3237 if (do_read_u32(ff, &nr_pmu_caps))
3238 return -1;
3239
3240 if (!nr_pmu_caps)
3241 return 0;
3242
3243 *caps = zalloc(sizeof(char *) * nr_pmu_caps);
3244 if (!*caps)
3245 return -1;
3246
3247 for (i = 0; i < nr_pmu_caps; i++) {
3248 name = do_read_string(ff);
3249 if (!name)
3250 goto error;
3251
3252 value = do_read_string(ff);
3253 if (!value)
3254 goto free_name;
3255
3256 if (asprintf(&ptr, "%s=%s", name, value) < 0)
3257 goto free_value;
3258
3259 (*caps)[i] = ptr;
3260
3261 if (!strcmp(name, "branches"))
3262 *max_branches = atoi(value);
3263
3264 free(value);
3265 free(name);
3266 }
3267 *nr_caps = nr_pmu_caps;
3268 return 0;
3269
3270free_value:
3271 free(value);
3272free_name:
3273 free(name);
3274error:
3275 for (; i > 0; i--)
3276 free((*caps)[i - 1]);
3277 free(*caps);
3278 *caps = NULL;
3279 *nr_caps = 0;
3280 return -1;
3281}
3282
3283static int process_cpu_pmu_caps(struct feat_fd *ff,
3284 void *data __maybe_unused)
3285{
3286 int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3287 &ff->ph->env.cpu_pmu_caps,
3288 &ff->ph->env.max_branches);
3289
3290 if (!ret && !ff->ph->env.cpu_pmu_caps)
3291 pr_debug("cpu pmu capabilities not available\n");
3292 return ret;
3293}
3294
3295static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused)
3296{
3297 struct pmu_caps *pmu_caps;
3298 u32 nr_pmu, i;
3299 int ret;
3300 int j;
3301
3302 if (do_read_u32(ff, &nr_pmu))
3303 return -1;
3304
3305 if (!nr_pmu) {
3306 pr_debug("pmu capabilities not available\n");
3307 return 0;
3308 }
3309
3310 pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
3311 if (!pmu_caps)
3312 return -ENOMEM;
3313
3314 for (i = 0; i < nr_pmu; i++) {
3315 ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps,
3316 &pmu_caps[i].caps,
3317 &pmu_caps[i].max_branches);
3318 if (ret)
3319 goto err;
3320
3321 pmu_caps[i].pmu_name = do_read_string(ff);
3322 if (!pmu_caps[i].pmu_name) {
3323 ret = -1;
3324 goto err;
3325 }
3326 if (!pmu_caps[i].nr_caps) {
3327 pr_debug("%s pmu capabilities not available\n",
3328 pmu_caps[i].pmu_name);
3329 }
3330 }
3331
3332 ff->ph->env.nr_pmus_with_caps = nr_pmu;
3333 ff->ph->env.pmu_caps = pmu_caps;
3334 return 0;
3335
3336err:
3337 for (i = 0; i < nr_pmu; i++) {
3338 for (j = 0; j < pmu_caps[i].nr_caps; j++)
3339 free(pmu_caps[i].caps[j]);
3340 free(pmu_caps[i].caps);
3341 free(pmu_caps[i].pmu_name);
3342 }
3343
3344 free(pmu_caps);
3345 return ret;
3346}
3347
3348#define FEAT_OPR(n, func, __full_only) \
3349 [HEADER_##n] = { \
3350 .name = __stringify(n), \
3351 .write = write_##func, \
3352 .print = print_##func, \
3353 .full_only = __full_only, \
3354 .process = process_##func, \
3355 .synthesize = true \
3356 }
3357
3358#define FEAT_OPN(n, func, __full_only) \
3359 [HEADER_##n] = { \
3360 .name = __stringify(n), \
3361 .write = write_##func, \
3362 .print = print_##func, \
3363 .full_only = __full_only, \
3364 .process = process_##func \
3365 }
3366
3367/* feature_ops not implemented: */
3368#define print_tracing_data NULL
3369#define print_build_id NULL
3370
3371#define process_branch_stack NULL
3372#define process_stat NULL
3373
3374// Only used in util/synthetic-events.c
3375const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3376
3377const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3378#ifdef HAVE_LIBTRACEEVENT
3379 FEAT_OPN(TRACING_DATA, tracing_data, false),
3380#endif
3381 FEAT_OPN(BUILD_ID, build_id, false),
3382 FEAT_OPR(HOSTNAME, hostname, false),
3383 FEAT_OPR(OSRELEASE, osrelease, false),
3384 FEAT_OPR(VERSION, version, false),
3385 FEAT_OPR(ARCH, arch, false),
3386 FEAT_OPR(NRCPUS, nrcpus, false),
3387 FEAT_OPR(CPUDESC, cpudesc, false),
3388 FEAT_OPR(CPUID, cpuid, false),
3389 FEAT_OPR(TOTAL_MEM, total_mem, false),
3390 FEAT_OPR(EVENT_DESC, event_desc, false),
3391 FEAT_OPR(CMDLINE, cmdline, false),
3392 FEAT_OPR(CPU_TOPOLOGY, cpu_topology, true),
3393 FEAT_OPR(NUMA_TOPOLOGY, numa_topology, true),
3394 FEAT_OPN(BRANCH_STACK, branch_stack, false),
3395 FEAT_OPR(PMU_MAPPINGS, pmu_mappings, false),
3396 FEAT_OPR(GROUP_DESC, group_desc, false),
3397 FEAT_OPN(AUXTRACE, auxtrace, false),
3398 FEAT_OPN(STAT, stat, false),
3399 FEAT_OPN(CACHE, cache, true),
3400 FEAT_OPR(SAMPLE_TIME, sample_time, false),
3401 FEAT_OPR(MEM_TOPOLOGY, mem_topology, true),
3402 FEAT_OPR(CLOCKID, clockid, false),
3403 FEAT_OPN(DIR_FORMAT, dir_format, false),
3404#ifdef HAVE_LIBBPF_SUPPORT
3405 FEAT_OPR(BPF_PROG_INFO, bpf_prog_info, false),
3406 FEAT_OPR(BPF_BTF, bpf_btf, false),
3407#endif
3408 FEAT_OPR(COMPRESSED, compressed, false),
3409 FEAT_OPR(CPU_PMU_CAPS, cpu_pmu_caps, false),
3410 FEAT_OPR(CLOCK_DATA, clock_data, false),
3411 FEAT_OPN(HYBRID_TOPOLOGY, hybrid_topology, true),
3412 FEAT_OPR(PMU_CAPS, pmu_caps, false),
3413};
3414
3415struct header_print_data {
3416 FILE *fp;
3417 bool full; /* extended list of headers */
3418};
3419
3420static int perf_file_section__fprintf_info(struct perf_file_section *section,
3421 struct perf_header *ph,
3422 int feat, int fd, void *data)
3423{
3424 struct header_print_data *hd = data;
3425 struct feat_fd ff;
3426
3427 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3428 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3429 "%d, continuing...\n", section->offset, feat);
3430 return 0;
3431 }
3432 if (feat >= HEADER_LAST_FEATURE) {
3433 pr_warning("unknown feature %d\n", feat);
3434 return 0;
3435 }
3436 if (!feat_ops[feat].print)
3437 return 0;
3438
3439 ff = (struct feat_fd) {
3440 .fd = fd,
3441 .ph = ph,
3442 };
3443
3444 if (!feat_ops[feat].full_only || hd->full)
3445 feat_ops[feat].print(&ff, hd->fp);
3446 else
3447 fprintf(hd->fp, "# %s info available, use -I to display\n",
3448 feat_ops[feat].name);
3449
3450 return 0;
3451}
3452
3453int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3454{
3455 struct header_print_data hd;
3456 struct perf_header *header = &session->header;
3457 int fd = perf_data__fd(session->data);
3458 struct stat st;
3459 time_t stctime;
3460 int ret, bit;
3461
3462 hd.fp = fp;
3463 hd.full = full;
3464
3465 ret = fstat(fd, &st);
3466 if (ret == -1)
3467 return -1;
3468
3469 stctime = st.st_mtime;
3470 fprintf(fp, "# captured on : %s", ctime(&stctime));
3471
3472 fprintf(fp, "# header version : %u\n", header->version);
3473 fprintf(fp, "# data offset : %" PRIu64 "\n", header->data_offset);
3474 fprintf(fp, "# data size : %" PRIu64 "\n", header->data_size);
3475 fprintf(fp, "# feat offset : %" PRIu64 "\n", header->feat_offset);
3476
3477 perf_header__process_sections(header, fd, &hd,
3478 perf_file_section__fprintf_info);
3479
3480 if (session->data->is_pipe)
3481 return 0;
3482
3483 fprintf(fp, "# missing features: ");
3484 for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3485 if (bit)
3486 fprintf(fp, "%s ", feat_ops[bit].name);
3487 }
3488
3489 fprintf(fp, "\n");
3490 return 0;
3491}
3492
3493struct header_fw {
3494 struct feat_writer fw;
3495 struct feat_fd *ff;
3496};
3497
3498static int feat_writer_cb(struct feat_writer *fw, void *buf, size_t sz)
3499{
3500 struct header_fw *h = container_of(fw, struct header_fw, fw);
3501
3502 return do_write(h->ff, buf, sz);
3503}
3504
3505static int do_write_feat(struct feat_fd *ff, int type,
3506 struct perf_file_section **p,
3507 struct evlist *evlist,
3508 struct feat_copier *fc)
3509{
3510 int err;
3511 int ret = 0;
3512
3513 if (perf_header__has_feat(ff->ph, type)) {
3514 if (!feat_ops[type].write)
3515 return -1;
3516
3517 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3518 return -1;
3519
3520 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3521
3522 /*
3523 * Hook to let perf inject copy features sections from the input
3524 * file.
3525 */
3526 if (fc && fc->copy) {
3527 struct header_fw h = {
3528 .fw.write = feat_writer_cb,
3529 .ff = ff,
3530 };
3531
3532 /* ->copy() returns 0 if the feature was not copied */
3533 err = fc->copy(fc, type, &h.fw);
3534 } else {
3535 err = 0;
3536 }
3537 if (!err)
3538 err = feat_ops[type].write(ff, evlist);
3539 if (err < 0) {
3540 pr_debug("failed to write feature %s\n", feat_ops[type].name);
3541
3542 /* undo anything written */
3543 lseek(ff->fd, (*p)->offset, SEEK_SET);
3544
3545 return -1;
3546 }
3547 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3548 (*p)++;
3549 }
3550 return ret;
3551}
3552
3553static int perf_header__adds_write(struct perf_header *header,
3554 struct evlist *evlist, int fd,
3555 struct feat_copier *fc)
3556{
3557 int nr_sections;
3558 struct feat_fd ff;
3559 struct perf_file_section *feat_sec, *p;
3560 int sec_size;
3561 u64 sec_start;
3562 int feat;
3563 int err;
3564
3565 ff = (struct feat_fd){
3566 .fd = fd,
3567 .ph = header,
3568 };
3569
3570 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3571 if (!nr_sections)
3572 return 0;
3573
3574 feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3575 if (feat_sec == NULL)
3576 return -ENOMEM;
3577
3578 sec_size = sizeof(*feat_sec) * nr_sections;
3579
3580 sec_start = header->feat_offset;
3581 lseek(fd, sec_start + sec_size, SEEK_SET);
3582
3583 for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3584 if (do_write_feat(&ff, feat, &p, evlist, fc))
3585 perf_header__clear_feat(header, feat);
3586 }
3587
3588 lseek(fd, sec_start, SEEK_SET);
3589 /*
3590 * may write more than needed due to dropped feature, but
3591 * this is okay, reader will skip the missing entries
3592 */
3593 err = do_write(&ff, feat_sec, sec_size);
3594 if (err < 0)
3595 pr_debug("failed to write feature section\n");
3596 free(feat_sec);
3597 return err;
3598}
3599
3600int perf_header__write_pipe(int fd)
3601{
3602 struct perf_pipe_file_header f_header;
3603 struct feat_fd ff;
3604 int err;
3605
3606 ff = (struct feat_fd){ .fd = fd };
3607
3608 f_header = (struct perf_pipe_file_header){
3609 .magic = PERF_MAGIC,
3610 .size = sizeof(f_header),
3611 };
3612
3613 err = do_write(&ff, &f_header, sizeof(f_header));
3614 if (err < 0) {
3615 pr_debug("failed to write perf pipe header\n");
3616 return err;
3617 }
3618
3619 return 0;
3620}
3621
3622static int perf_session__do_write_header(struct perf_session *session,
3623 struct evlist *evlist,
3624 int fd, bool at_exit,
3625 struct feat_copier *fc)
3626{
3627 struct perf_file_header f_header;
3628 struct perf_file_attr f_attr;
3629 struct perf_header *header = &session->header;
3630 struct evsel *evsel;
3631 struct feat_fd ff;
3632 u64 attr_offset;
3633 int err;
3634
3635 ff = (struct feat_fd){ .fd = fd};
3636 lseek(fd, sizeof(f_header), SEEK_SET);
3637
3638 evlist__for_each_entry(session->evlist, evsel) {
3639 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3640 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3641 if (err < 0) {
3642 pr_debug("failed to write perf header\n");
3643 return err;
3644 }
3645 }
3646
3647 attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3648
3649 evlist__for_each_entry(evlist, evsel) {
3650 if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3651 /*
3652 * We are likely in "perf inject" and have read
3653 * from an older file. Update attr size so that
3654 * reader gets the right offset to the ids.
3655 */
3656 evsel->core.attr.size = sizeof(evsel->core.attr);
3657 }
3658 f_attr = (struct perf_file_attr){
3659 .attr = evsel->core.attr,
3660 .ids = {
3661 .offset = evsel->id_offset,
3662 .size = evsel->core.ids * sizeof(u64),
3663 }
3664 };
3665 err = do_write(&ff, &f_attr, sizeof(f_attr));
3666 if (err < 0) {
3667 pr_debug("failed to write perf header attribute\n");
3668 return err;
3669 }
3670 }
3671
3672 if (!header->data_offset)
3673 header->data_offset = lseek(fd, 0, SEEK_CUR);
3674 header->feat_offset = header->data_offset + header->data_size;
3675
3676 if (at_exit) {
3677 err = perf_header__adds_write(header, evlist, fd, fc);
3678 if (err < 0)
3679 return err;
3680 }
3681
3682 f_header = (struct perf_file_header){
3683 .magic = PERF_MAGIC,
3684 .size = sizeof(f_header),
3685 .attr_size = sizeof(f_attr),
3686 .attrs = {
3687 .offset = attr_offset,
3688 .size = evlist->core.nr_entries * sizeof(f_attr),
3689 },
3690 .data = {
3691 .offset = header->data_offset,
3692 .size = header->data_size,
3693 },
3694 /* event_types is ignored, store zeros */
3695 };
3696
3697 memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3698
3699 lseek(fd, 0, SEEK_SET);
3700 err = do_write(&ff, &f_header, sizeof(f_header));
3701 if (err < 0) {
3702 pr_debug("failed to write perf header\n");
3703 return err;
3704 }
3705 lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3706
3707 return 0;
3708}
3709
3710int perf_session__write_header(struct perf_session *session,
3711 struct evlist *evlist,
3712 int fd, bool at_exit)
3713{
3714 return perf_session__do_write_header(session, evlist, fd, at_exit, NULL);
3715}
3716
3717size_t perf_session__data_offset(const struct evlist *evlist)
3718{
3719 struct evsel *evsel;
3720 size_t data_offset;
3721
3722 data_offset = sizeof(struct perf_file_header);
3723 evlist__for_each_entry(evlist, evsel) {
3724 data_offset += evsel->core.ids * sizeof(u64);
3725 }
3726 data_offset += evlist->core.nr_entries * sizeof(struct perf_file_attr);
3727
3728 return data_offset;
3729}
3730
3731int perf_session__inject_header(struct perf_session *session,
3732 struct evlist *evlist,
3733 int fd,
3734 struct feat_copier *fc)
3735{
3736 return perf_session__do_write_header(session, evlist, fd, true, fc);
3737}
3738
3739static int perf_header__getbuffer64(struct perf_header *header,
3740 int fd, void *buf, size_t size)
3741{
3742 if (readn(fd, buf, size) <= 0)
3743 return -1;
3744
3745 if (header->needs_swap)
3746 mem_bswap_64(buf, size);
3747
3748 return 0;
3749}
3750
3751int perf_header__process_sections(struct perf_header *header, int fd,
3752 void *data,
3753 int (*process)(struct perf_file_section *section,
3754 struct perf_header *ph,
3755 int feat, int fd, void *data))
3756{
3757 struct perf_file_section *feat_sec, *sec;
3758 int nr_sections;
3759 int sec_size;
3760 int feat;
3761 int err;
3762
3763 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3764 if (!nr_sections)
3765 return 0;
3766
3767 feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3768 if (!feat_sec)
3769 return -1;
3770
3771 sec_size = sizeof(*feat_sec) * nr_sections;
3772
3773 lseek(fd, header->feat_offset, SEEK_SET);
3774
3775 err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3776 if (err < 0)
3777 goto out_free;
3778
3779 for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3780 err = process(sec++, header, feat, fd, data);
3781 if (err < 0)
3782 goto out_free;
3783 }
3784 err = 0;
3785out_free:
3786 free(feat_sec);
3787 return err;
3788}
3789
3790static const int attr_file_abi_sizes[] = {
3791 [0] = PERF_ATTR_SIZE_VER0,
3792 [1] = PERF_ATTR_SIZE_VER1,
3793 [2] = PERF_ATTR_SIZE_VER2,
3794 [3] = PERF_ATTR_SIZE_VER3,
3795 [4] = PERF_ATTR_SIZE_VER4,
3796 0,
3797};
3798
3799/*
3800 * In the legacy file format, the magic number is not used to encode endianness.
3801 * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3802 * on ABI revisions, we need to try all combinations for all endianness to
3803 * detect the endianness.
3804 */
3805static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3806{
3807 uint64_t ref_size, attr_size;
3808 int i;
3809
3810 for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3811 ref_size = attr_file_abi_sizes[i]
3812 + sizeof(struct perf_file_section);
3813 if (hdr_sz != ref_size) {
3814 attr_size = bswap_64(hdr_sz);
3815 if (attr_size != ref_size)
3816 continue;
3817
3818 ph->needs_swap = true;
3819 }
3820 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3821 i,
3822 ph->needs_swap);
3823 return 0;
3824 }
3825 /* could not determine endianness */
3826 return -1;
3827}
3828
3829#define PERF_PIPE_HDR_VER0 16
3830
3831static const size_t attr_pipe_abi_sizes[] = {
3832 [0] = PERF_PIPE_HDR_VER0,
3833 0,
3834};
3835
3836/*
3837 * In the legacy pipe format, there is an implicit assumption that endianness
3838 * between host recording the samples, and host parsing the samples is the
3839 * same. This is not always the case given that the pipe output may always be
3840 * redirected into a file and analyzed on a different machine with possibly a
3841 * different endianness and perf_event ABI revisions in the perf tool itself.
3842 */
3843static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3844{
3845 u64 attr_size;
3846 int i;
3847
3848 for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3849 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3850 attr_size = bswap_64(hdr_sz);
3851 if (attr_size != hdr_sz)
3852 continue;
3853
3854 ph->needs_swap = true;
3855 }
3856 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3857 return 0;
3858 }
3859 return -1;
3860}
3861
3862bool is_perf_magic(u64 magic)
3863{
3864 if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3865 || magic == __perf_magic2
3866 || magic == __perf_magic2_sw)
3867 return true;
3868
3869 return false;
3870}
3871
3872static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3873 bool is_pipe, struct perf_header *ph)
3874{
3875 int ret;
3876
3877 /* check for legacy format */
3878 ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3879 if (ret == 0) {
3880 ph->version = PERF_HEADER_VERSION_1;
3881 pr_debug("legacy perf.data format\n");
3882 if (is_pipe)
3883 return try_all_pipe_abis(hdr_sz, ph);
3884
3885 return try_all_file_abis(hdr_sz, ph);
3886 }
3887 /*
3888 * the new magic number serves two purposes:
3889 * - unique number to identify actual perf.data files
3890 * - encode endianness of file
3891 */
3892 ph->version = PERF_HEADER_VERSION_2;
3893
3894 /* check magic number with one endianness */
3895 if (magic == __perf_magic2)
3896 return 0;
3897
3898 /* check magic number with opposite endianness */
3899 if (magic != __perf_magic2_sw)
3900 return -1;
3901
3902 ph->needs_swap = true;
3903
3904 return 0;
3905}
3906
3907int perf_file_header__read(struct perf_file_header *header,
3908 struct perf_header *ph, int fd)
3909{
3910 ssize_t ret;
3911
3912 lseek(fd, 0, SEEK_SET);
3913
3914 ret = readn(fd, header, sizeof(*header));
3915 if (ret <= 0)
3916 return -1;
3917
3918 if (check_magic_endian(header->magic,
3919 header->attr_size, false, ph) < 0) {
3920 pr_debug("magic/endian check failed\n");
3921 return -1;
3922 }
3923
3924 if (ph->needs_swap) {
3925 mem_bswap_64(header, offsetof(struct perf_file_header,
3926 adds_features));
3927 }
3928
3929 if (header->size != sizeof(*header)) {
3930 /* Support the previous format */
3931 if (header->size == offsetof(typeof(*header), adds_features))
3932 bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3933 else
3934 return -1;
3935 } else if (ph->needs_swap) {
3936 /*
3937 * feature bitmap is declared as an array of unsigned longs --
3938 * not good since its size can differ between the host that
3939 * generated the data file and the host analyzing the file.
3940 *
3941 * We need to handle endianness, but we don't know the size of
3942 * the unsigned long where the file was generated. Take a best
3943 * guess at determining it: try 64-bit swap first (ie., file
3944 * created on a 64-bit host), and check if the hostname feature
3945 * bit is set (this feature bit is forced on as of fbe96f2).
3946 * If the bit is not, undo the 64-bit swap and try a 32-bit
3947 * swap. If the hostname bit is still not set (e.g., older data
3948 * file), punt and fallback to the original behavior --
3949 * clearing all feature bits and setting buildid.
3950 */
3951 mem_bswap_64(&header->adds_features,
3952 BITS_TO_U64(HEADER_FEAT_BITS));
3953
3954 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3955 /* unswap as u64 */
3956 mem_bswap_64(&header->adds_features,
3957 BITS_TO_U64(HEADER_FEAT_BITS));
3958
3959 /* unswap as u32 */
3960 mem_bswap_32(&header->adds_features,
3961 BITS_TO_U32(HEADER_FEAT_BITS));
3962 }
3963
3964 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3965 bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3966 __set_bit(HEADER_BUILD_ID, header->adds_features);
3967 }
3968 }
3969
3970 memcpy(&ph->adds_features, &header->adds_features,
3971 sizeof(ph->adds_features));
3972
3973 ph->data_offset = header->data.offset;
3974 ph->data_size = header->data.size;
3975 ph->feat_offset = header->data.offset + header->data.size;
3976 return 0;
3977}
3978
3979static int perf_file_section__process(struct perf_file_section *section,
3980 struct perf_header *ph,
3981 int feat, int fd, void *data)
3982{
3983 struct feat_fd fdd = {
3984 .fd = fd,
3985 .ph = ph,
3986 .size = section->size,
3987 .offset = section->offset,
3988 };
3989
3990 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3991 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3992 "%d, continuing...\n", section->offset, feat);
3993 return 0;
3994 }
3995
3996 if (feat >= HEADER_LAST_FEATURE) {
3997 pr_debug("unknown feature %d, continuing...\n", feat);
3998 return 0;
3999 }
4000
4001 if (!feat_ops[feat].process)
4002 return 0;
4003
4004 return feat_ops[feat].process(&fdd, data);
4005}
4006
4007static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
4008 struct perf_header *ph,
4009 struct perf_data* data,
4010 bool repipe, int repipe_fd)
4011{
4012 struct feat_fd ff = {
4013 .fd = repipe_fd,
4014 .ph = ph,
4015 };
4016 ssize_t ret;
4017
4018 ret = perf_data__read(data, header, sizeof(*header));
4019 if (ret <= 0)
4020 return -1;
4021
4022 if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
4023 pr_debug("endian/magic failed\n");
4024 return -1;
4025 }
4026
4027 if (ph->needs_swap)
4028 header->size = bswap_64(header->size);
4029
4030 if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
4031 return -1;
4032
4033 return 0;
4034}
4035
4036static int perf_header__read_pipe(struct perf_session *session, int repipe_fd)
4037{
4038 struct perf_header *header = &session->header;
4039 struct perf_pipe_file_header f_header;
4040
4041 if (perf_file_header__read_pipe(&f_header, header, session->data,
4042 session->repipe, repipe_fd) < 0) {
4043 pr_debug("incompatible file format\n");
4044 return -EINVAL;
4045 }
4046
4047 return f_header.size == sizeof(f_header) ? 0 : -1;
4048}
4049
4050static int read_attr(int fd, struct perf_header *ph,
4051 struct perf_file_attr *f_attr)
4052{
4053 struct perf_event_attr *attr = &f_attr->attr;
4054 size_t sz, left;
4055 size_t our_sz = sizeof(f_attr->attr);
4056 ssize_t ret;
4057
4058 memset(f_attr, 0, sizeof(*f_attr));
4059
4060 /* read minimal guaranteed structure */
4061 ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
4062 if (ret <= 0) {
4063 pr_debug("cannot read %d bytes of header attr\n",
4064 PERF_ATTR_SIZE_VER0);
4065 return -1;
4066 }
4067
4068 /* on file perf_event_attr size */
4069 sz = attr->size;
4070
4071 if (ph->needs_swap)
4072 sz = bswap_32(sz);
4073
4074 if (sz == 0) {
4075 /* assume ABI0 */
4076 sz = PERF_ATTR_SIZE_VER0;
4077 } else if (sz > our_sz) {
4078 pr_debug("file uses a more recent and unsupported ABI"
4079 " (%zu bytes extra)\n", sz - our_sz);
4080 return -1;
4081 }
4082 /* what we have not yet read and that we know about */
4083 left = sz - PERF_ATTR_SIZE_VER0;
4084 if (left) {
4085 void *ptr = attr;
4086 ptr += PERF_ATTR_SIZE_VER0;
4087
4088 ret = readn(fd, ptr, left);
4089 }
4090 /* read perf_file_section, ids are read in caller */
4091 ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
4092
4093 return ret <= 0 ? -1 : 0;
4094}
4095
4096#ifdef HAVE_LIBTRACEEVENT
4097static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
4098{
4099 struct tep_event *event;
4100 char bf[128];
4101
4102 /* already prepared */
4103 if (evsel->tp_format)
4104 return 0;
4105
4106 if (pevent == NULL) {
4107 pr_debug("broken or missing trace data\n");
4108 return -1;
4109 }
4110
4111 event = tep_find_event(pevent, evsel->core.attr.config);
4112 if (event == NULL) {
4113 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
4114 return -1;
4115 }
4116
4117 if (!evsel->name) {
4118 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
4119 evsel->name = strdup(bf);
4120 if (evsel->name == NULL)
4121 return -1;
4122 }
4123
4124 evsel->tp_format = event;
4125 return 0;
4126}
4127
4128static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
4129{
4130 struct evsel *pos;
4131
4132 evlist__for_each_entry(evlist, pos) {
4133 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
4134 evsel__prepare_tracepoint_event(pos, pevent))
4135 return -1;
4136 }
4137
4138 return 0;
4139}
4140#endif
4141
4142int perf_session__read_header(struct perf_session *session, int repipe_fd)
4143{
4144 struct perf_data *data = session->data;
4145 struct perf_header *header = &session->header;
4146 struct perf_file_header f_header;
4147 struct perf_file_attr f_attr;
4148 u64 f_id;
4149 int nr_attrs, nr_ids, i, j, err;
4150 int fd = perf_data__fd(data);
4151
4152 session->evlist = evlist__new();
4153 if (session->evlist == NULL)
4154 return -ENOMEM;
4155
4156 session->evlist->env = &header->env;
4157 session->machines.host.env = &header->env;
4158
4159 /*
4160 * We can read 'pipe' data event from regular file,
4161 * check for the pipe header regardless of source.
4162 */
4163 err = perf_header__read_pipe(session, repipe_fd);
4164 if (!err || perf_data__is_pipe(data)) {
4165 data->is_pipe = true;
4166 return err;
4167 }
4168
4169 if (perf_file_header__read(&f_header, header, fd) < 0)
4170 return -EINVAL;
4171
4172 if (header->needs_swap && data->in_place_update) {
4173 pr_err("In-place update not supported when byte-swapping is required\n");
4174 return -EINVAL;
4175 }
4176
4177 /*
4178 * Sanity check that perf.data was written cleanly; data size is
4179 * initialized to 0 and updated only if the on_exit function is run.
4180 * If data size is still 0 then the file contains only partial
4181 * information. Just warn user and process it as much as it can.
4182 */
4183 if (f_header.data.size == 0) {
4184 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4185 "Was the 'perf record' command properly terminated?\n",
4186 data->file.path);
4187 }
4188
4189 if (f_header.attr_size == 0) {
4190 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4191 "Was the 'perf record' command properly terminated?\n",
4192 data->file.path);
4193 return -EINVAL;
4194 }
4195
4196 nr_attrs = f_header.attrs.size / f_header.attr_size;
4197 lseek(fd, f_header.attrs.offset, SEEK_SET);
4198
4199 for (i = 0; i < nr_attrs; i++) {
4200 struct evsel *evsel;
4201 off_t tmp;
4202
4203 if (read_attr(fd, header, &f_attr) < 0)
4204 goto out_errno;
4205
4206 if (header->needs_swap) {
4207 f_attr.ids.size = bswap_64(f_attr.ids.size);
4208 f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4209 perf_event__attr_swap(&f_attr.attr);
4210 }
4211
4212 tmp = lseek(fd, 0, SEEK_CUR);
4213 evsel = evsel__new(&f_attr.attr);
4214
4215 if (evsel == NULL)
4216 goto out_delete_evlist;
4217
4218 evsel->needs_swap = header->needs_swap;
4219 /*
4220 * Do it before so that if perf_evsel__alloc_id fails, this
4221 * entry gets purged too at evlist__delete().
4222 */
4223 evlist__add(session->evlist, evsel);
4224
4225 nr_ids = f_attr.ids.size / sizeof(u64);
4226 /*
4227 * We don't have the cpu and thread maps on the header, so
4228 * for allocating the perf_sample_id table we fake 1 cpu and
4229 * hattr->ids threads.
4230 */
4231 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4232 goto out_delete_evlist;
4233
4234 lseek(fd, f_attr.ids.offset, SEEK_SET);
4235
4236 for (j = 0; j < nr_ids; j++) {
4237 if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4238 goto out_errno;
4239
4240 perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4241 }
4242
4243 lseek(fd, tmp, SEEK_SET);
4244 }
4245
4246#ifdef HAVE_LIBTRACEEVENT
4247 perf_header__process_sections(header, fd, &session->tevent,
4248 perf_file_section__process);
4249
4250 if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4251 goto out_delete_evlist;
4252#else
4253 perf_header__process_sections(header, fd, NULL, perf_file_section__process);
4254#endif
4255
4256 return 0;
4257out_errno:
4258 return -errno;
4259
4260out_delete_evlist:
4261 evlist__delete(session->evlist);
4262 session->evlist = NULL;
4263 return -ENOMEM;
4264}
4265
4266int perf_event__process_feature(struct perf_session *session,
4267 union perf_event *event)
4268{
4269 struct perf_tool *tool = session->tool;
4270 struct feat_fd ff = { .fd = 0 };
4271 struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4272 int type = fe->header.type;
4273 u64 feat = fe->feat_id;
4274 int ret = 0;
4275
4276 if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4277 pr_warning("invalid record type %d in pipe-mode\n", type);
4278 return 0;
4279 }
4280 if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4281 pr_warning("invalid record type %d in pipe-mode\n", type);
4282 return -1;
4283 }
4284
4285 if (!feat_ops[feat].process)
4286 return 0;
4287
4288 ff.buf = (void *)fe->data;
4289 ff.size = event->header.size - sizeof(*fe);
4290 ff.ph = &session->header;
4291
4292 if (feat_ops[feat].process(&ff, NULL)) {
4293 ret = -1;
4294 goto out;
4295 }
4296
4297 if (!feat_ops[feat].print || !tool->show_feat_hdr)
4298 goto out;
4299
4300 if (!feat_ops[feat].full_only ||
4301 tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4302 feat_ops[feat].print(&ff, stdout);
4303 } else {
4304 fprintf(stdout, "# %s info available, use -I to display\n",
4305 feat_ops[feat].name);
4306 }
4307out:
4308 free_event_desc(ff.events);
4309 return ret;
4310}
4311
4312size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4313{
4314 struct perf_record_event_update *ev = &event->event_update;
4315 struct perf_cpu_map *map;
4316 size_t ret;
4317
4318 ret = fprintf(fp, "\n... id: %" PRI_lu64 "\n", ev->id);
4319
4320 switch (ev->type) {
4321 case PERF_EVENT_UPDATE__SCALE:
4322 ret += fprintf(fp, "... scale: %f\n", ev->scale.scale);
4323 break;
4324 case PERF_EVENT_UPDATE__UNIT:
4325 ret += fprintf(fp, "... unit: %s\n", ev->unit);
4326 break;
4327 case PERF_EVENT_UPDATE__NAME:
4328 ret += fprintf(fp, "... name: %s\n", ev->name);
4329 break;
4330 case PERF_EVENT_UPDATE__CPUS:
4331 ret += fprintf(fp, "... ");
4332
4333 map = cpu_map__new_data(&ev->cpus.cpus);
4334 if (map)
4335 ret += cpu_map__fprintf(map, fp);
4336 else
4337 ret += fprintf(fp, "failed to get cpus\n");
4338 break;
4339 default:
4340 ret += fprintf(fp, "... unknown type\n");
4341 break;
4342 }
4343
4344 return ret;
4345}
4346
4347int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4348 union perf_event *event,
4349 struct evlist **pevlist)
4350{
4351 u32 i, ids, n_ids;
4352 struct evsel *evsel;
4353 struct evlist *evlist = *pevlist;
4354
4355 if (evlist == NULL) {
4356 *pevlist = evlist = evlist__new();
4357 if (evlist == NULL)
4358 return -ENOMEM;
4359 }
4360
4361 evsel = evsel__new(&event->attr.attr);
4362 if (evsel == NULL)
4363 return -ENOMEM;
4364
4365 evlist__add(evlist, evsel);
4366
4367 ids = event->header.size;
4368 ids -= (void *)&event->attr.id - (void *)event;
4369 n_ids = ids / sizeof(u64);
4370 /*
4371 * We don't have the cpu and thread maps on the header, so
4372 * for allocating the perf_sample_id table we fake 1 cpu and
4373 * hattr->ids threads.
4374 */
4375 if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4376 return -ENOMEM;
4377
4378 for (i = 0; i < n_ids; i++) {
4379 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, event->attr.id[i]);
4380 }
4381
4382 return 0;
4383}
4384
4385int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4386 union perf_event *event,
4387 struct evlist **pevlist)
4388{
4389 struct perf_record_event_update *ev = &event->event_update;
4390 struct evlist *evlist;
4391 struct evsel *evsel;
4392 struct perf_cpu_map *map;
4393
4394 if (dump_trace)
4395 perf_event__fprintf_event_update(event, stdout);
4396
4397 if (!pevlist || *pevlist == NULL)
4398 return -EINVAL;
4399
4400 evlist = *pevlist;
4401
4402 evsel = evlist__id2evsel(evlist, ev->id);
4403 if (evsel == NULL)
4404 return -EINVAL;
4405
4406 switch (ev->type) {
4407 case PERF_EVENT_UPDATE__UNIT:
4408 free((char *)evsel->unit);
4409 evsel->unit = strdup(ev->unit);
4410 break;
4411 case PERF_EVENT_UPDATE__NAME:
4412 free(evsel->name);
4413 evsel->name = strdup(ev->name);
4414 break;
4415 case PERF_EVENT_UPDATE__SCALE:
4416 evsel->scale = ev->scale.scale;
4417 break;
4418 case PERF_EVENT_UPDATE__CPUS:
4419 map = cpu_map__new_data(&ev->cpus.cpus);
4420 if (map) {
4421 perf_cpu_map__put(evsel->core.own_cpus);
4422 evsel->core.own_cpus = map;
4423 } else
4424 pr_err("failed to get event_update cpus\n");
4425 default:
4426 break;
4427 }
4428
4429 return 0;
4430}
4431
4432#ifdef HAVE_LIBTRACEEVENT
4433int perf_event__process_tracing_data(struct perf_session *session,
4434 union perf_event *event)
4435{
4436 ssize_t size_read, padding, size = event->tracing_data.size;
4437 int fd = perf_data__fd(session->data);
4438 char buf[BUFSIZ];
4439
4440 /*
4441 * The pipe fd is already in proper place and in any case
4442 * we can't move it, and we'd screw the case where we read
4443 * 'pipe' data from regular file. The trace_report reads
4444 * data from 'fd' so we need to set it directly behind the
4445 * event, where the tracing data starts.
4446 */
4447 if (!perf_data__is_pipe(session->data)) {
4448 off_t offset = lseek(fd, 0, SEEK_CUR);
4449
4450 /* setup for reading amidst mmap */
4451 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4452 SEEK_SET);
4453 }
4454
4455 size_read = trace_report(fd, &session->tevent,
4456 session->repipe);
4457 padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4458
4459 if (readn(fd, buf, padding) < 0) {
4460 pr_err("%s: reading input file", __func__);
4461 return -1;
4462 }
4463 if (session->repipe) {
4464 int retw = write(STDOUT_FILENO, buf, padding);
4465 if (retw <= 0 || retw != padding) {
4466 pr_err("%s: repiping tracing data padding", __func__);
4467 return -1;
4468 }
4469 }
4470
4471 if (size_read + padding != size) {
4472 pr_err("%s: tracing data size mismatch", __func__);
4473 return -1;
4474 }
4475
4476 evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4477
4478 return size_read + padding;
4479}
4480#endif
4481
4482int perf_event__process_build_id(struct perf_session *session,
4483 union perf_event *event)
4484{
4485 __event_process_build_id(&event->build_id,
4486 event->build_id.filename,
4487 session);
4488 return 0;
4489}