xref: /kernel/linux/linux-6.6/tools/perf/util/header.c (revision 62306a36)
1// SPDX-License-Identifier: GPL-2.0
2#include <errno.h>
3#include <inttypes.h>
4#include "string2.h"
5#include <sys/param.h>
6#include <sys/types.h>
7#include <byteswap.h>
8#include <unistd.h>
9#include <regex.h>
10#include <stdio.h>
11#include <stdlib.h>
12#include <linux/compiler.h>
13#include <linux/list.h>
14#include <linux/kernel.h>
15#include <linux/bitops.h>
16#include <linux/string.h>
17#include <linux/stringify.h>
18#include <linux/zalloc.h>
19#include <sys/stat.h>
20#include <sys/utsname.h>
21#include <linux/time64.h>
22#include <dirent.h>
23#ifdef HAVE_LIBBPF_SUPPORT
24#include <bpf/libbpf.h>
25#endif
26#include <perf/cpumap.h>
27#include <tools/libc_compat.h> // reallocarray
28
29#include "dso.h"
30#include "evlist.h"
31#include "evsel.h"
32#include "util/evsel_fprintf.h"
33#include "header.h"
34#include "memswap.h"
35#include "trace-event.h"
36#include "session.h"
37#include "symbol.h"
38#include "debug.h"
39#include "cpumap.h"
40#include "pmu.h"
41#include "pmus.h"
42#include "vdso.h"
43#include "strbuf.h"
44#include "build-id.h"
45#include "data.h"
46#include <api/fs/fs.h>
47#include "asm/bug.h"
48#include "tool.h"
49#include "time-utils.h"
50#include "units.h"
51#include "util/util.h" // perf_exe()
52#include "cputopo.h"
53#include "bpf-event.h"
54#include "bpf-utils.h"
55#include "clockid.h"
56
57#include <linux/ctype.h>
58#include <internal/lib.h>
59
60#ifdef HAVE_LIBTRACEEVENT
61#include <traceevent/event-parse.h>
62#endif
63
64/*
65 * magic2 = "PERFILE2"
66 * must be a numerical value to let the endianness
67 * determine the memory layout. That way we are able
68 * to detect endianness when reading the perf.data file
69 * back.
70 *
71 * we check for legacy (PERFFILE) format.
72 */
73static const char *__perf_magic1 = "PERFFILE";
74static const u64 __perf_magic2    = 0x32454c4946524550ULL;
75static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
76
77#define PERF_MAGIC	__perf_magic2
78
79const char perf_version_string[] = PERF_VERSION;
80
81struct perf_file_attr {
82	struct perf_event_attr	attr;
83	struct perf_file_section	ids;
84};
85
86void perf_header__set_feat(struct perf_header *header, int feat)
87{
88	__set_bit(feat, header->adds_features);
89}
90
91void perf_header__clear_feat(struct perf_header *header, int feat)
92{
93	__clear_bit(feat, header->adds_features);
94}
95
96bool perf_header__has_feat(const struct perf_header *header, int feat)
97{
98	return test_bit(feat, header->adds_features);
99}
100
101static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
102{
103	ssize_t ret = writen(ff->fd, buf, size);
104
105	if (ret != (ssize_t)size)
106		return ret < 0 ? (int)ret : -1;
107	return 0;
108}
109
110static int __do_write_buf(struct feat_fd *ff,  const void *buf, size_t size)
111{
112	/* struct perf_event_header::size is u16 */
113	const size_t max_size = 0xffff - sizeof(struct perf_event_header);
114	size_t new_size = ff->size;
115	void *addr;
116
117	if (size + ff->offset > max_size)
118		return -E2BIG;
119
120	while (size > (new_size - ff->offset))
121		new_size <<= 1;
122	new_size = min(max_size, new_size);
123
124	if (ff->size < new_size) {
125		addr = realloc(ff->buf, new_size);
126		if (!addr)
127			return -ENOMEM;
128		ff->buf = addr;
129		ff->size = new_size;
130	}
131
132	memcpy(ff->buf + ff->offset, buf, size);
133	ff->offset += size;
134
135	return 0;
136}
137
138/* Return: 0 if succeeded, -ERR if failed. */
139int do_write(struct feat_fd *ff, const void *buf, size_t size)
140{
141	if (!ff->buf)
142		return __do_write_fd(ff, buf, size);
143	return __do_write_buf(ff, buf, size);
144}
145
146/* Return: 0 if succeeded, -ERR if failed. */
147static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
148{
149	u64 *p = (u64 *) set;
150	int i, ret;
151
152	ret = do_write(ff, &size, sizeof(size));
153	if (ret < 0)
154		return ret;
155
156	for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
157		ret = do_write(ff, p + i, sizeof(*p));
158		if (ret < 0)
159			return ret;
160	}
161
162	return 0;
163}
164
165/* Return: 0 if succeeded, -ERR if failed. */
166int write_padded(struct feat_fd *ff, const void *bf,
167		 size_t count, size_t count_aligned)
168{
169	static const char zero_buf[NAME_ALIGN];
170	int err = do_write(ff, bf, count);
171
172	if (!err)
173		err = do_write(ff, zero_buf, count_aligned - count);
174
175	return err;
176}
177
178#define string_size(str)						\
179	(PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
180
181/* Return: 0 if succeeded, -ERR if failed. */
182static int do_write_string(struct feat_fd *ff, const char *str)
183{
184	u32 len, olen;
185	int ret;
186
187	olen = strlen(str) + 1;
188	len = PERF_ALIGN(olen, NAME_ALIGN);
189
190	/* write len, incl. \0 */
191	ret = do_write(ff, &len, sizeof(len));
192	if (ret < 0)
193		return ret;
194
195	return write_padded(ff, str, olen, len);
196}
197
198static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
199{
200	ssize_t ret = readn(ff->fd, addr, size);
201
202	if (ret != size)
203		return ret < 0 ? (int)ret : -1;
204	return 0;
205}
206
207static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
208{
209	if (size > (ssize_t)ff->size - ff->offset)
210		return -1;
211
212	memcpy(addr, ff->buf + ff->offset, size);
213	ff->offset += size;
214
215	return 0;
216
217}
218
219static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
220{
221	if (!ff->buf)
222		return __do_read_fd(ff, addr, size);
223	return __do_read_buf(ff, addr, size);
224}
225
226static int do_read_u32(struct feat_fd *ff, u32 *addr)
227{
228	int ret;
229
230	ret = __do_read(ff, addr, sizeof(*addr));
231	if (ret)
232		return ret;
233
234	if (ff->ph->needs_swap)
235		*addr = bswap_32(*addr);
236	return 0;
237}
238
239static int do_read_u64(struct feat_fd *ff, u64 *addr)
240{
241	int ret;
242
243	ret = __do_read(ff, addr, sizeof(*addr));
244	if (ret)
245		return ret;
246
247	if (ff->ph->needs_swap)
248		*addr = bswap_64(*addr);
249	return 0;
250}
251
252static char *do_read_string(struct feat_fd *ff)
253{
254	u32 len;
255	char *buf;
256
257	if (do_read_u32(ff, &len))
258		return NULL;
259
260	buf = malloc(len);
261	if (!buf)
262		return NULL;
263
264	if (!__do_read(ff, buf, len)) {
265		/*
266		 * strings are padded by zeroes
267		 * thus the actual strlen of buf
268		 * may be less than len
269		 */
270		return buf;
271	}
272
273	free(buf);
274	return NULL;
275}
276
277/* Return: 0 if succeeded, -ERR if failed. */
278static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
279{
280	unsigned long *set;
281	u64 size, *p;
282	int i, ret;
283
284	ret = do_read_u64(ff, &size);
285	if (ret)
286		return ret;
287
288	set = bitmap_zalloc(size);
289	if (!set)
290		return -ENOMEM;
291
292	p = (u64 *) set;
293
294	for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
295		ret = do_read_u64(ff, p + i);
296		if (ret < 0) {
297			free(set);
298			return ret;
299		}
300	}
301
302	*pset  = set;
303	*psize = size;
304	return 0;
305}
306
307#ifdef HAVE_LIBTRACEEVENT
308static int write_tracing_data(struct feat_fd *ff,
309			      struct evlist *evlist)
310{
311	if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
312		return -1;
313
314	return read_tracing_data(ff->fd, &evlist->core.entries);
315}
316#endif
317
318static int write_build_id(struct feat_fd *ff,
319			  struct evlist *evlist __maybe_unused)
320{
321	struct perf_session *session;
322	int err;
323
324	session = container_of(ff->ph, struct perf_session, header);
325
326	if (!perf_session__read_build_ids(session, true))
327		return -1;
328
329	if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
330		return -1;
331
332	err = perf_session__write_buildid_table(session, ff);
333	if (err < 0) {
334		pr_debug("failed to write buildid table\n");
335		return err;
336	}
337	perf_session__cache_build_ids(session);
338
339	return 0;
340}
341
342static int write_hostname(struct feat_fd *ff,
343			  struct evlist *evlist __maybe_unused)
344{
345	struct utsname uts;
346	int ret;
347
348	ret = uname(&uts);
349	if (ret < 0)
350		return -1;
351
352	return do_write_string(ff, uts.nodename);
353}
354
355static int write_osrelease(struct feat_fd *ff,
356			   struct evlist *evlist __maybe_unused)
357{
358	struct utsname uts;
359	int ret;
360
361	ret = uname(&uts);
362	if (ret < 0)
363		return -1;
364
365	return do_write_string(ff, uts.release);
366}
367
368static int write_arch(struct feat_fd *ff,
369		      struct evlist *evlist __maybe_unused)
370{
371	struct utsname uts;
372	int ret;
373
374	ret = uname(&uts);
375	if (ret < 0)
376		return -1;
377
378	return do_write_string(ff, uts.machine);
379}
380
381static int write_version(struct feat_fd *ff,
382			 struct evlist *evlist __maybe_unused)
383{
384	return do_write_string(ff, perf_version_string);
385}
386
387static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
388{
389	FILE *file;
390	char *buf = NULL;
391	char *s, *p;
392	const char *search = cpuinfo_proc;
393	size_t len = 0;
394	int ret = -1;
395
396	if (!search)
397		return -1;
398
399	file = fopen("/proc/cpuinfo", "r");
400	if (!file)
401		return -1;
402
403	while (getline(&buf, &len, file) > 0) {
404		ret = strncmp(buf, search, strlen(search));
405		if (!ret)
406			break;
407	}
408
409	if (ret) {
410		ret = -1;
411		goto done;
412	}
413
414	s = buf;
415
416	p = strchr(buf, ':');
417	if (p && *(p+1) == ' ' && *(p+2))
418		s = p + 2;
419	p = strchr(s, '\n');
420	if (p)
421		*p = '\0';
422
423	/* squash extra space characters (branding string) */
424	p = s;
425	while (*p) {
426		if (isspace(*p)) {
427			char *r = p + 1;
428			char *q = skip_spaces(r);
429			*p = ' ';
430			if (q != (p+1))
431				while ((*r++ = *q++));
432		}
433		p++;
434	}
435	ret = do_write_string(ff, s);
436done:
437	free(buf);
438	fclose(file);
439	return ret;
440}
441
442static int write_cpudesc(struct feat_fd *ff,
443		       struct evlist *evlist __maybe_unused)
444{
445#if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
446#define CPUINFO_PROC	{ "cpu", }
447#elif defined(__s390__)
448#define CPUINFO_PROC	{ "vendor_id", }
449#elif defined(__sh__)
450#define CPUINFO_PROC	{ "cpu type", }
451#elif defined(__alpha__) || defined(__mips__)
452#define CPUINFO_PROC	{ "cpu model", }
453#elif defined(__arm__)
454#define CPUINFO_PROC	{ "model name", "Processor", }
455#elif defined(__arc__)
456#define CPUINFO_PROC	{ "Processor", }
457#elif defined(__xtensa__)
458#define CPUINFO_PROC	{ "core ID", }
459#elif defined(__loongarch__)
460#define CPUINFO_PROC	{ "Model Name", }
461#else
462#define CPUINFO_PROC	{ "model name", }
463#endif
464	const char *cpuinfo_procs[] = CPUINFO_PROC;
465#undef CPUINFO_PROC
466	unsigned int i;
467
468	for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
469		int ret;
470		ret = __write_cpudesc(ff, cpuinfo_procs[i]);
471		if (ret >= 0)
472			return ret;
473	}
474	return -1;
475}
476
477
478static int write_nrcpus(struct feat_fd *ff,
479			struct evlist *evlist __maybe_unused)
480{
481	long nr;
482	u32 nrc, nra;
483	int ret;
484
485	nrc = cpu__max_present_cpu().cpu;
486
487	nr = sysconf(_SC_NPROCESSORS_ONLN);
488	if (nr < 0)
489		return -1;
490
491	nra = (u32)(nr & UINT_MAX);
492
493	ret = do_write(ff, &nrc, sizeof(nrc));
494	if (ret < 0)
495		return ret;
496
497	return do_write(ff, &nra, sizeof(nra));
498}
499
500static int write_event_desc(struct feat_fd *ff,
501			    struct evlist *evlist)
502{
503	struct evsel *evsel;
504	u32 nre, nri, sz;
505	int ret;
506
507	nre = evlist->core.nr_entries;
508
509	/*
510	 * write number of events
511	 */
512	ret = do_write(ff, &nre, sizeof(nre));
513	if (ret < 0)
514		return ret;
515
516	/*
517	 * size of perf_event_attr struct
518	 */
519	sz = (u32)sizeof(evsel->core.attr);
520	ret = do_write(ff, &sz, sizeof(sz));
521	if (ret < 0)
522		return ret;
523
524	evlist__for_each_entry(evlist, evsel) {
525		ret = do_write(ff, &evsel->core.attr, sz);
526		if (ret < 0)
527			return ret;
528		/*
529		 * write number of unique id per event
530		 * there is one id per instance of an event
531		 *
532		 * copy into an nri to be independent of the
533		 * type of ids,
534		 */
535		nri = evsel->core.ids;
536		ret = do_write(ff, &nri, sizeof(nri));
537		if (ret < 0)
538			return ret;
539
540		/*
541		 * write event string as passed on cmdline
542		 */
543		ret = do_write_string(ff, evsel__name(evsel));
544		if (ret < 0)
545			return ret;
546		/*
547		 * write unique ids for this event
548		 */
549		ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
550		if (ret < 0)
551			return ret;
552	}
553	return 0;
554}
555
556static int write_cmdline(struct feat_fd *ff,
557			 struct evlist *evlist __maybe_unused)
558{
559	char pbuf[MAXPATHLEN], *buf;
560	int i, ret, n;
561
562	/* actual path to perf binary */
563	buf = perf_exe(pbuf, MAXPATHLEN);
564
565	/* account for binary path */
566	n = perf_env.nr_cmdline + 1;
567
568	ret = do_write(ff, &n, sizeof(n));
569	if (ret < 0)
570		return ret;
571
572	ret = do_write_string(ff, buf);
573	if (ret < 0)
574		return ret;
575
576	for (i = 0 ; i < perf_env.nr_cmdline; i++) {
577		ret = do_write_string(ff, perf_env.cmdline_argv[i]);
578		if (ret < 0)
579			return ret;
580	}
581	return 0;
582}
583
584
585static int write_cpu_topology(struct feat_fd *ff,
586			      struct evlist *evlist __maybe_unused)
587{
588	struct cpu_topology *tp;
589	u32 i;
590	int ret, j;
591
592	tp = cpu_topology__new();
593	if (!tp)
594		return -1;
595
596	ret = do_write(ff, &tp->package_cpus_lists, sizeof(tp->package_cpus_lists));
597	if (ret < 0)
598		goto done;
599
600	for (i = 0; i < tp->package_cpus_lists; i++) {
601		ret = do_write_string(ff, tp->package_cpus_list[i]);
602		if (ret < 0)
603			goto done;
604	}
605	ret = do_write(ff, &tp->core_cpus_lists, sizeof(tp->core_cpus_lists));
606	if (ret < 0)
607		goto done;
608
609	for (i = 0; i < tp->core_cpus_lists; i++) {
610		ret = do_write_string(ff, tp->core_cpus_list[i]);
611		if (ret < 0)
612			break;
613	}
614
615	ret = perf_env__read_cpu_topology_map(&perf_env);
616	if (ret < 0)
617		goto done;
618
619	for (j = 0; j < perf_env.nr_cpus_avail; j++) {
620		ret = do_write(ff, &perf_env.cpu[j].core_id,
621			       sizeof(perf_env.cpu[j].core_id));
622		if (ret < 0)
623			return ret;
624		ret = do_write(ff, &perf_env.cpu[j].socket_id,
625			       sizeof(perf_env.cpu[j].socket_id));
626		if (ret < 0)
627			return ret;
628	}
629
630	if (!tp->die_cpus_lists)
631		goto done;
632
633	ret = do_write(ff, &tp->die_cpus_lists, sizeof(tp->die_cpus_lists));
634	if (ret < 0)
635		goto done;
636
637	for (i = 0; i < tp->die_cpus_lists; i++) {
638		ret = do_write_string(ff, tp->die_cpus_list[i]);
639		if (ret < 0)
640			goto done;
641	}
642
643	for (j = 0; j < perf_env.nr_cpus_avail; j++) {
644		ret = do_write(ff, &perf_env.cpu[j].die_id,
645			       sizeof(perf_env.cpu[j].die_id));
646		if (ret < 0)
647			return ret;
648	}
649
650done:
651	cpu_topology__delete(tp);
652	return ret;
653}
654
655
656
657static int write_total_mem(struct feat_fd *ff,
658			   struct evlist *evlist __maybe_unused)
659{
660	char *buf = NULL;
661	FILE *fp;
662	size_t len = 0;
663	int ret = -1, n;
664	uint64_t mem;
665
666	fp = fopen("/proc/meminfo", "r");
667	if (!fp)
668		return -1;
669
670	while (getline(&buf, &len, fp) > 0) {
671		ret = strncmp(buf, "MemTotal:", 9);
672		if (!ret)
673			break;
674	}
675	if (!ret) {
676		n = sscanf(buf, "%*s %"PRIu64, &mem);
677		if (n == 1)
678			ret = do_write(ff, &mem, sizeof(mem));
679	} else
680		ret = -1;
681	free(buf);
682	fclose(fp);
683	return ret;
684}
685
686static int write_numa_topology(struct feat_fd *ff,
687			       struct evlist *evlist __maybe_unused)
688{
689	struct numa_topology *tp;
690	int ret = -1;
691	u32 i;
692
693	tp = numa_topology__new();
694	if (!tp)
695		return -ENOMEM;
696
697	ret = do_write(ff, &tp->nr, sizeof(u32));
698	if (ret < 0)
699		goto err;
700
701	for (i = 0; i < tp->nr; i++) {
702		struct numa_topology_node *n = &tp->nodes[i];
703
704		ret = do_write(ff, &n->node, sizeof(u32));
705		if (ret < 0)
706			goto err;
707
708		ret = do_write(ff, &n->mem_total, sizeof(u64));
709		if (ret)
710			goto err;
711
712		ret = do_write(ff, &n->mem_free, sizeof(u64));
713		if (ret)
714			goto err;
715
716		ret = do_write_string(ff, n->cpus);
717		if (ret < 0)
718			goto err;
719	}
720
721	ret = 0;
722
723err:
724	numa_topology__delete(tp);
725	return ret;
726}
727
728/*
729 * File format:
730 *
731 * struct pmu_mappings {
732 *	u32	pmu_num;
733 *	struct pmu_map {
734 *		u32	type;
735 *		char	name[];
736 *	}[pmu_num];
737 * };
738 */
739
740static int write_pmu_mappings(struct feat_fd *ff,
741			      struct evlist *evlist __maybe_unused)
742{
743	struct perf_pmu *pmu = NULL;
744	u32 pmu_num = 0;
745	int ret;
746
747	/*
748	 * Do a first pass to count number of pmu to avoid lseek so this
749	 * works in pipe mode as well.
750	 */
751	while ((pmu = perf_pmus__scan(pmu)))
752		pmu_num++;
753
754	ret = do_write(ff, &pmu_num, sizeof(pmu_num));
755	if (ret < 0)
756		return ret;
757
758	while ((pmu = perf_pmus__scan(pmu))) {
759		ret = do_write(ff, &pmu->type, sizeof(pmu->type));
760		if (ret < 0)
761			return ret;
762
763		ret = do_write_string(ff, pmu->name);
764		if (ret < 0)
765			return ret;
766	}
767
768	return 0;
769}
770
771/*
772 * File format:
773 *
774 * struct group_descs {
775 *	u32	nr_groups;
776 *	struct group_desc {
777 *		char	name[];
778 *		u32	leader_idx;
779 *		u32	nr_members;
780 *	}[nr_groups];
781 * };
782 */
783static int write_group_desc(struct feat_fd *ff,
784			    struct evlist *evlist)
785{
786	u32 nr_groups = evlist__nr_groups(evlist);
787	struct evsel *evsel;
788	int ret;
789
790	ret = do_write(ff, &nr_groups, sizeof(nr_groups));
791	if (ret < 0)
792		return ret;
793
794	evlist__for_each_entry(evlist, evsel) {
795		if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
796			const char *name = evsel->group_name ?: "{anon_group}";
797			u32 leader_idx = evsel->core.idx;
798			u32 nr_members = evsel->core.nr_members;
799
800			ret = do_write_string(ff, name);
801			if (ret < 0)
802				return ret;
803
804			ret = do_write(ff, &leader_idx, sizeof(leader_idx));
805			if (ret < 0)
806				return ret;
807
808			ret = do_write(ff, &nr_members, sizeof(nr_members));
809			if (ret < 0)
810				return ret;
811		}
812	}
813	return 0;
814}
815
816/*
817 * Return the CPU id as a raw string.
818 *
819 * Each architecture should provide a more precise id string that
820 * can be use to match the architecture's "mapfile".
821 */
822char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
823{
824	return NULL;
825}
826
827/* Return zero when the cpuid from the mapfile.csv matches the
828 * cpuid string generated on this platform.
829 * Otherwise return non-zero.
830 */
831int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
832{
833	regex_t re;
834	regmatch_t pmatch[1];
835	int match;
836
837	if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
838		/* Warn unable to generate match particular string. */
839		pr_info("Invalid regular expression %s\n", mapcpuid);
840		return 1;
841	}
842
843	match = !regexec(&re, cpuid, 1, pmatch, 0);
844	regfree(&re);
845	if (match) {
846		size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
847
848		/* Verify the entire string matched. */
849		if (match_len == strlen(cpuid))
850			return 0;
851	}
852	return 1;
853}
854
855/*
856 * default get_cpuid(): nothing gets recorded
857 * actual implementation must be in arch/$(SRCARCH)/util/header.c
858 */
859int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
860{
861	return ENOSYS; /* Not implemented */
862}
863
864static int write_cpuid(struct feat_fd *ff,
865		       struct evlist *evlist __maybe_unused)
866{
867	char buffer[64];
868	int ret;
869
870	ret = get_cpuid(buffer, sizeof(buffer));
871	if (ret)
872		return -1;
873
874	return do_write_string(ff, buffer);
875}
876
877static int write_branch_stack(struct feat_fd *ff __maybe_unused,
878			      struct evlist *evlist __maybe_unused)
879{
880	return 0;
881}
882
883static int write_auxtrace(struct feat_fd *ff,
884			  struct evlist *evlist __maybe_unused)
885{
886	struct perf_session *session;
887	int err;
888
889	if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
890		return -1;
891
892	session = container_of(ff->ph, struct perf_session, header);
893
894	err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
895	if (err < 0)
896		pr_err("Failed to write auxtrace index\n");
897	return err;
898}
899
900static int write_clockid(struct feat_fd *ff,
901			 struct evlist *evlist __maybe_unused)
902{
903	return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
904			sizeof(ff->ph->env.clock.clockid_res_ns));
905}
906
907static int write_clock_data(struct feat_fd *ff,
908			    struct evlist *evlist __maybe_unused)
909{
910	u64 *data64;
911	u32 data32;
912	int ret;
913
914	/* version */
915	data32 = 1;
916
917	ret = do_write(ff, &data32, sizeof(data32));
918	if (ret < 0)
919		return ret;
920
921	/* clockid */
922	data32 = ff->ph->env.clock.clockid;
923
924	ret = do_write(ff, &data32, sizeof(data32));
925	if (ret < 0)
926		return ret;
927
928	/* TOD ref time */
929	data64 = &ff->ph->env.clock.tod_ns;
930
931	ret = do_write(ff, data64, sizeof(*data64));
932	if (ret < 0)
933		return ret;
934
935	/* clockid ref time */
936	data64 = &ff->ph->env.clock.clockid_ns;
937
938	return do_write(ff, data64, sizeof(*data64));
939}
940
941static int write_hybrid_topology(struct feat_fd *ff,
942				 struct evlist *evlist __maybe_unused)
943{
944	struct hybrid_topology *tp;
945	int ret;
946	u32 i;
947
948	tp = hybrid_topology__new();
949	if (!tp)
950		return -ENOENT;
951
952	ret = do_write(ff, &tp->nr, sizeof(u32));
953	if (ret < 0)
954		goto err;
955
956	for (i = 0; i < tp->nr; i++) {
957		struct hybrid_topology_node *n = &tp->nodes[i];
958
959		ret = do_write_string(ff, n->pmu_name);
960		if (ret < 0)
961			goto err;
962
963		ret = do_write_string(ff, n->cpus);
964		if (ret < 0)
965			goto err;
966	}
967
968	ret = 0;
969
970err:
971	hybrid_topology__delete(tp);
972	return ret;
973}
974
975static int write_dir_format(struct feat_fd *ff,
976			    struct evlist *evlist __maybe_unused)
977{
978	struct perf_session *session;
979	struct perf_data *data;
980
981	session = container_of(ff->ph, struct perf_session, header);
982	data = session->data;
983
984	if (WARN_ON(!perf_data__is_dir(data)))
985		return -1;
986
987	return do_write(ff, &data->dir.version, sizeof(data->dir.version));
988}
989
990/*
991 * Check whether a CPU is online
992 *
993 * Returns:
994 *     1 -> if CPU is online
995 *     0 -> if CPU is offline
996 *    -1 -> error case
997 */
998int is_cpu_online(unsigned int cpu)
999{
1000	char *str;
1001	size_t strlen;
1002	char buf[256];
1003	int status = -1;
1004	struct stat statbuf;
1005
1006	snprintf(buf, sizeof(buf),
1007		"/sys/devices/system/cpu/cpu%d", cpu);
1008	if (stat(buf, &statbuf) != 0)
1009		return 0;
1010
1011	/*
1012	 * Check if /sys/devices/system/cpu/cpux/online file
1013	 * exists. Some cases cpu0 won't have online file since
1014	 * it is not expected to be turned off generally.
1015	 * In kernels without CONFIG_HOTPLUG_CPU, this
1016	 * file won't exist
1017	 */
1018	snprintf(buf, sizeof(buf),
1019		"/sys/devices/system/cpu/cpu%d/online", cpu);
1020	if (stat(buf, &statbuf) != 0)
1021		return 1;
1022
1023	/*
1024	 * Read online file using sysfs__read_str.
1025	 * If read or open fails, return -1.
1026	 * If read succeeds, return value from file
1027	 * which gets stored in "str"
1028	 */
1029	snprintf(buf, sizeof(buf),
1030		"devices/system/cpu/cpu%d/online", cpu);
1031
1032	if (sysfs__read_str(buf, &str, &strlen) < 0)
1033		return status;
1034
1035	status = atoi(str);
1036
1037	free(str);
1038	return status;
1039}
1040
1041#ifdef HAVE_LIBBPF_SUPPORT
1042static int write_bpf_prog_info(struct feat_fd *ff,
1043			       struct evlist *evlist __maybe_unused)
1044{
1045	struct perf_env *env = &ff->ph->env;
1046	struct rb_root *root;
1047	struct rb_node *next;
1048	int ret;
1049
1050	down_read(&env->bpf_progs.lock);
1051
1052	ret = do_write(ff, &env->bpf_progs.infos_cnt,
1053		       sizeof(env->bpf_progs.infos_cnt));
1054	if (ret < 0)
1055		goto out;
1056
1057	root = &env->bpf_progs.infos;
1058	next = rb_first(root);
1059	while (next) {
1060		struct bpf_prog_info_node *node;
1061		size_t len;
1062
1063		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1064		next = rb_next(&node->rb_node);
1065		len = sizeof(struct perf_bpil) +
1066			node->info_linear->data_len;
1067
1068		/* before writing to file, translate address to offset */
1069		bpil_addr_to_offs(node->info_linear);
1070		ret = do_write(ff, node->info_linear, len);
1071		/*
1072		 * translate back to address even when do_write() fails,
1073		 * so that this function never changes the data.
1074		 */
1075		bpil_offs_to_addr(node->info_linear);
1076		if (ret < 0)
1077			goto out;
1078	}
1079out:
1080	up_read(&env->bpf_progs.lock);
1081	return ret;
1082}
1083
1084static int write_bpf_btf(struct feat_fd *ff,
1085			 struct evlist *evlist __maybe_unused)
1086{
1087	struct perf_env *env = &ff->ph->env;
1088	struct rb_root *root;
1089	struct rb_node *next;
1090	int ret;
1091
1092	down_read(&env->bpf_progs.lock);
1093
1094	ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1095		       sizeof(env->bpf_progs.btfs_cnt));
1096
1097	if (ret < 0)
1098		goto out;
1099
1100	root = &env->bpf_progs.btfs;
1101	next = rb_first(root);
1102	while (next) {
1103		struct btf_node *node;
1104
1105		node = rb_entry(next, struct btf_node, rb_node);
1106		next = rb_next(&node->rb_node);
1107		ret = do_write(ff, &node->id,
1108			       sizeof(u32) * 2 + node->data_size);
1109		if (ret < 0)
1110			goto out;
1111	}
1112out:
1113	up_read(&env->bpf_progs.lock);
1114	return ret;
1115}
1116#endif // HAVE_LIBBPF_SUPPORT
1117
1118static int cpu_cache_level__sort(const void *a, const void *b)
1119{
1120	struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1121	struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1122
1123	return cache_a->level - cache_b->level;
1124}
1125
1126static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1127{
1128	if (a->level != b->level)
1129		return false;
1130
1131	if (a->line_size != b->line_size)
1132		return false;
1133
1134	if (a->sets != b->sets)
1135		return false;
1136
1137	if (a->ways != b->ways)
1138		return false;
1139
1140	if (strcmp(a->type, b->type))
1141		return false;
1142
1143	if (strcmp(a->size, b->size))
1144		return false;
1145
1146	if (strcmp(a->map, b->map))
1147		return false;
1148
1149	return true;
1150}
1151
1152static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1153{
1154	char path[PATH_MAX], file[PATH_MAX];
1155	struct stat st;
1156	size_t len;
1157
1158	scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1159	scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1160
1161	if (stat(file, &st))
1162		return 1;
1163
1164	scnprintf(file, PATH_MAX, "%s/level", path);
1165	if (sysfs__read_int(file, (int *) &cache->level))
1166		return -1;
1167
1168	scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1169	if (sysfs__read_int(file, (int *) &cache->line_size))
1170		return -1;
1171
1172	scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1173	if (sysfs__read_int(file, (int *) &cache->sets))
1174		return -1;
1175
1176	scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1177	if (sysfs__read_int(file, (int *) &cache->ways))
1178		return -1;
1179
1180	scnprintf(file, PATH_MAX, "%s/type", path);
1181	if (sysfs__read_str(file, &cache->type, &len))
1182		return -1;
1183
1184	cache->type[len] = 0;
1185	cache->type = strim(cache->type);
1186
1187	scnprintf(file, PATH_MAX, "%s/size", path);
1188	if (sysfs__read_str(file, &cache->size, &len)) {
1189		zfree(&cache->type);
1190		return -1;
1191	}
1192
1193	cache->size[len] = 0;
1194	cache->size = strim(cache->size);
1195
1196	scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1197	if (sysfs__read_str(file, &cache->map, &len)) {
1198		zfree(&cache->size);
1199		zfree(&cache->type);
1200		return -1;
1201	}
1202
1203	cache->map[len] = 0;
1204	cache->map = strim(cache->map);
1205	return 0;
1206}
1207
1208static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1209{
1210	fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1211}
1212
1213/*
1214 * Build caches levels for a particular CPU from the data in
1215 * /sys/devices/system/cpu/cpu<cpu>/cache/
1216 * The cache level data is stored in caches[] from index at
1217 * *cntp.
1218 */
1219int build_caches_for_cpu(u32 cpu, struct cpu_cache_level caches[], u32 *cntp)
1220{
1221	u16 level;
1222
1223	for (level = 0; level < MAX_CACHE_LVL; level++) {
1224		struct cpu_cache_level c;
1225		int err;
1226		u32 i;
1227
1228		err = cpu_cache_level__read(&c, cpu, level);
1229		if (err < 0)
1230			return err;
1231
1232		if (err == 1)
1233			break;
1234
1235		for (i = 0; i < *cntp; i++) {
1236			if (cpu_cache_level__cmp(&c, &caches[i]))
1237				break;
1238		}
1239
1240		if (i == *cntp) {
1241			caches[*cntp] = c;
1242			*cntp = *cntp + 1;
1243		} else
1244			cpu_cache_level__free(&c);
1245	}
1246
1247	return 0;
1248}
1249
1250static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1251{
1252	u32 nr, cpu, cnt = 0;
1253
1254	nr = cpu__max_cpu().cpu;
1255
1256	for (cpu = 0; cpu < nr; cpu++) {
1257		int ret = build_caches_for_cpu(cpu, caches, &cnt);
1258
1259		if (ret)
1260			return ret;
1261	}
1262	*cntp = cnt;
1263	return 0;
1264}
1265
1266static int write_cache(struct feat_fd *ff,
1267		       struct evlist *evlist __maybe_unused)
1268{
1269	u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL;
1270	struct cpu_cache_level caches[max_caches];
1271	u32 cnt = 0, i, version = 1;
1272	int ret;
1273
1274	ret = build_caches(caches, &cnt);
1275	if (ret)
1276		goto out;
1277
1278	qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1279
1280	ret = do_write(ff, &version, sizeof(u32));
1281	if (ret < 0)
1282		goto out;
1283
1284	ret = do_write(ff, &cnt, sizeof(u32));
1285	if (ret < 0)
1286		goto out;
1287
1288	for (i = 0; i < cnt; i++) {
1289		struct cpu_cache_level *c = &caches[i];
1290
1291		#define _W(v)					\
1292			ret = do_write(ff, &c->v, sizeof(u32));	\
1293			if (ret < 0)				\
1294				goto out;
1295
1296		_W(level)
1297		_W(line_size)
1298		_W(sets)
1299		_W(ways)
1300		#undef _W
1301
1302		#define _W(v)						\
1303			ret = do_write_string(ff, (const char *) c->v);	\
1304			if (ret < 0)					\
1305				goto out;
1306
1307		_W(type)
1308		_W(size)
1309		_W(map)
1310		#undef _W
1311	}
1312
1313out:
1314	for (i = 0; i < cnt; i++)
1315		cpu_cache_level__free(&caches[i]);
1316	return ret;
1317}
1318
1319static int write_stat(struct feat_fd *ff __maybe_unused,
1320		      struct evlist *evlist __maybe_unused)
1321{
1322	return 0;
1323}
1324
1325static int write_sample_time(struct feat_fd *ff,
1326			     struct evlist *evlist)
1327{
1328	int ret;
1329
1330	ret = do_write(ff, &evlist->first_sample_time,
1331		       sizeof(evlist->first_sample_time));
1332	if (ret < 0)
1333		return ret;
1334
1335	return do_write(ff, &evlist->last_sample_time,
1336			sizeof(evlist->last_sample_time));
1337}
1338
1339
1340static int memory_node__read(struct memory_node *n, unsigned long idx)
1341{
1342	unsigned int phys, size = 0;
1343	char path[PATH_MAX];
1344	struct dirent *ent;
1345	DIR *dir;
1346
1347#define for_each_memory(mem, dir)					\
1348	while ((ent = readdir(dir)))					\
1349		if (strcmp(ent->d_name, ".") &&				\
1350		    strcmp(ent->d_name, "..") &&			\
1351		    sscanf(ent->d_name, "memory%u", &mem) == 1)
1352
1353	scnprintf(path, PATH_MAX,
1354		  "%s/devices/system/node/node%lu",
1355		  sysfs__mountpoint(), idx);
1356
1357	dir = opendir(path);
1358	if (!dir) {
1359		pr_warning("failed: can't open memory sysfs data\n");
1360		return -1;
1361	}
1362
1363	for_each_memory(phys, dir) {
1364		size = max(phys, size);
1365	}
1366
1367	size++;
1368
1369	n->set = bitmap_zalloc(size);
1370	if (!n->set) {
1371		closedir(dir);
1372		return -ENOMEM;
1373	}
1374
1375	n->node = idx;
1376	n->size = size;
1377
1378	rewinddir(dir);
1379
1380	for_each_memory(phys, dir) {
1381		__set_bit(phys, n->set);
1382	}
1383
1384	closedir(dir);
1385	return 0;
1386}
1387
1388static void memory_node__delete_nodes(struct memory_node *nodesp, u64 cnt)
1389{
1390	for (u64 i = 0; i < cnt; i++)
1391		bitmap_free(nodesp[i].set);
1392
1393	free(nodesp);
1394}
1395
1396static int memory_node__sort(const void *a, const void *b)
1397{
1398	const struct memory_node *na = a;
1399	const struct memory_node *nb = b;
1400
1401	return na->node - nb->node;
1402}
1403
1404static int build_mem_topology(struct memory_node **nodesp, u64 *cntp)
1405{
1406	char path[PATH_MAX];
1407	struct dirent *ent;
1408	DIR *dir;
1409	int ret = 0;
1410	size_t cnt = 0, size = 0;
1411	struct memory_node *nodes = NULL;
1412
1413	scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1414		  sysfs__mountpoint());
1415
1416	dir = opendir(path);
1417	if (!dir) {
1418		pr_debug2("%s: couldn't read %s, does this arch have topology information?\n",
1419			  __func__, path);
1420		return -1;
1421	}
1422
1423	while (!ret && (ent = readdir(dir))) {
1424		unsigned int idx;
1425		int r;
1426
1427		if (!strcmp(ent->d_name, ".") ||
1428		    !strcmp(ent->d_name, ".."))
1429			continue;
1430
1431		r = sscanf(ent->d_name, "node%u", &idx);
1432		if (r != 1)
1433			continue;
1434
1435		if (cnt >= size) {
1436			struct memory_node *new_nodes =
1437				reallocarray(nodes, cnt + 4, sizeof(*nodes));
1438
1439			if (!new_nodes) {
1440				pr_err("Failed to write MEM_TOPOLOGY, size %zd nodes\n", size);
1441				ret = -ENOMEM;
1442				goto out;
1443			}
1444			nodes = new_nodes;
1445			size += 4;
1446		}
1447		ret = memory_node__read(&nodes[cnt], idx);
1448		if (!ret)
1449			cnt += 1;
1450	}
1451out:
1452	closedir(dir);
1453	if (!ret) {
1454		*cntp = cnt;
1455		*nodesp = nodes;
1456		qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1457	} else
1458		memory_node__delete_nodes(nodes, cnt);
1459
1460	return ret;
1461}
1462
1463/*
1464 * The MEM_TOPOLOGY holds physical memory map for every
1465 * node in system. The format of data is as follows:
1466 *
1467 *  0 - version          | for future changes
1468 *  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1469 * 16 - count            | number of nodes
1470 *
1471 * For each node we store map of physical indexes for
1472 * each node:
1473 *
1474 * 32 - node id          | node index
1475 * 40 - size             | size of bitmap
1476 * 48 - bitmap           | bitmap of memory indexes that belongs to node
1477 */
1478static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1479			      struct evlist *evlist __maybe_unused)
1480{
1481	struct memory_node *nodes = NULL;
1482	u64 bsize, version = 1, i, nr = 0;
1483	int ret;
1484
1485	ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1486			      (unsigned long long *) &bsize);
1487	if (ret)
1488		return ret;
1489
1490	ret = build_mem_topology(&nodes, &nr);
1491	if (ret)
1492		return ret;
1493
1494	ret = do_write(ff, &version, sizeof(version));
1495	if (ret < 0)
1496		goto out;
1497
1498	ret = do_write(ff, &bsize, sizeof(bsize));
1499	if (ret < 0)
1500		goto out;
1501
1502	ret = do_write(ff, &nr, sizeof(nr));
1503	if (ret < 0)
1504		goto out;
1505
1506	for (i = 0; i < nr; i++) {
1507		struct memory_node *n = &nodes[i];
1508
1509		#define _W(v)						\
1510			ret = do_write(ff, &n->v, sizeof(n->v));	\
1511			if (ret < 0)					\
1512				goto out;
1513
1514		_W(node)
1515		_W(size)
1516
1517		#undef _W
1518
1519		ret = do_write_bitmap(ff, n->set, n->size);
1520		if (ret < 0)
1521			goto out;
1522	}
1523
1524out:
1525	memory_node__delete_nodes(nodes, nr);
1526	return ret;
1527}
1528
1529static int write_compressed(struct feat_fd *ff __maybe_unused,
1530			    struct evlist *evlist __maybe_unused)
1531{
1532	int ret;
1533
1534	ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1535	if (ret)
1536		return ret;
1537
1538	ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1539	if (ret)
1540		return ret;
1541
1542	ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1543	if (ret)
1544		return ret;
1545
1546	ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1547	if (ret)
1548		return ret;
1549
1550	return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1551}
1552
1553static int __write_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1554			    bool write_pmu)
1555{
1556	struct perf_pmu_caps *caps = NULL;
1557	int ret;
1558
1559	ret = do_write(ff, &pmu->nr_caps, sizeof(pmu->nr_caps));
1560	if (ret < 0)
1561		return ret;
1562
1563	list_for_each_entry(caps, &pmu->caps, list) {
1564		ret = do_write_string(ff, caps->name);
1565		if (ret < 0)
1566			return ret;
1567
1568		ret = do_write_string(ff, caps->value);
1569		if (ret < 0)
1570			return ret;
1571	}
1572
1573	if (write_pmu) {
1574		ret = do_write_string(ff, pmu->name);
1575		if (ret < 0)
1576			return ret;
1577	}
1578
1579	return ret;
1580}
1581
1582static int write_cpu_pmu_caps(struct feat_fd *ff,
1583			      struct evlist *evlist __maybe_unused)
1584{
1585	struct perf_pmu *cpu_pmu = perf_pmus__find("cpu");
1586	int ret;
1587
1588	if (!cpu_pmu)
1589		return -ENOENT;
1590
1591	ret = perf_pmu__caps_parse(cpu_pmu);
1592	if (ret < 0)
1593		return ret;
1594
1595	return __write_pmu_caps(ff, cpu_pmu, false);
1596}
1597
1598static int write_pmu_caps(struct feat_fd *ff,
1599			  struct evlist *evlist __maybe_unused)
1600{
1601	struct perf_pmu *pmu = NULL;
1602	int nr_pmu = 0;
1603	int ret;
1604
1605	while ((pmu = perf_pmus__scan(pmu))) {
1606		if (!strcmp(pmu->name, "cpu")) {
1607			/*
1608			 * The "cpu" PMU is special and covered by
1609			 * HEADER_CPU_PMU_CAPS. Note, core PMUs are
1610			 * counted/written here for ARM, s390 and Intel hybrid.
1611			 */
1612			continue;
1613		}
1614		if (perf_pmu__caps_parse(pmu) <= 0)
1615			continue;
1616		nr_pmu++;
1617	}
1618
1619	ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1620	if (ret < 0)
1621		return ret;
1622
1623	if (!nr_pmu)
1624		return 0;
1625
1626	/*
1627	 * Note older perf tools assume core PMUs come first, this is a property
1628	 * of perf_pmus__scan.
1629	 */
1630	pmu = NULL;
1631	while ((pmu = perf_pmus__scan(pmu))) {
1632		if (!strcmp(pmu->name, "cpu")) {
1633			/* Skip as above. */
1634			continue;
1635		}
1636		if (perf_pmu__caps_parse(pmu) <= 0)
1637			continue;
1638		ret = __write_pmu_caps(ff, pmu, true);
1639		if (ret < 0)
1640			return ret;
1641	}
1642	return 0;
1643}
1644
1645static void print_hostname(struct feat_fd *ff, FILE *fp)
1646{
1647	fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1648}
1649
1650static void print_osrelease(struct feat_fd *ff, FILE *fp)
1651{
1652	fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1653}
1654
1655static void print_arch(struct feat_fd *ff, FILE *fp)
1656{
1657	fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1658}
1659
1660static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1661{
1662	fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1663}
1664
1665static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1666{
1667	fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1668	fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1669}
1670
1671static void print_version(struct feat_fd *ff, FILE *fp)
1672{
1673	fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1674}
1675
1676static void print_cmdline(struct feat_fd *ff, FILE *fp)
1677{
1678	int nr, i;
1679
1680	nr = ff->ph->env.nr_cmdline;
1681
1682	fprintf(fp, "# cmdline : ");
1683
1684	for (i = 0; i < nr; i++) {
1685		char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1686		if (!argv_i) {
1687			fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1688		} else {
1689			char *mem = argv_i;
1690			do {
1691				char *quote = strchr(argv_i, '\'');
1692				if (!quote)
1693					break;
1694				*quote++ = '\0';
1695				fprintf(fp, "%s\\\'", argv_i);
1696				argv_i = quote;
1697			} while (1);
1698			fprintf(fp, "%s ", argv_i);
1699			free(mem);
1700		}
1701	}
1702	fputc('\n', fp);
1703}
1704
1705static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1706{
1707	struct perf_header *ph = ff->ph;
1708	int cpu_nr = ph->env.nr_cpus_avail;
1709	int nr, i;
1710	char *str;
1711
1712	nr = ph->env.nr_sibling_cores;
1713	str = ph->env.sibling_cores;
1714
1715	for (i = 0; i < nr; i++) {
1716		fprintf(fp, "# sibling sockets : %s\n", str);
1717		str += strlen(str) + 1;
1718	}
1719
1720	if (ph->env.nr_sibling_dies) {
1721		nr = ph->env.nr_sibling_dies;
1722		str = ph->env.sibling_dies;
1723
1724		for (i = 0; i < nr; i++) {
1725			fprintf(fp, "# sibling dies    : %s\n", str);
1726			str += strlen(str) + 1;
1727		}
1728	}
1729
1730	nr = ph->env.nr_sibling_threads;
1731	str = ph->env.sibling_threads;
1732
1733	for (i = 0; i < nr; i++) {
1734		fprintf(fp, "# sibling threads : %s\n", str);
1735		str += strlen(str) + 1;
1736	}
1737
1738	if (ph->env.nr_sibling_dies) {
1739		if (ph->env.cpu != NULL) {
1740			for (i = 0; i < cpu_nr; i++)
1741				fprintf(fp, "# CPU %d: Core ID %d, "
1742					    "Die ID %d, Socket ID %d\n",
1743					    i, ph->env.cpu[i].core_id,
1744					    ph->env.cpu[i].die_id,
1745					    ph->env.cpu[i].socket_id);
1746		} else
1747			fprintf(fp, "# Core ID, Die ID and Socket ID "
1748				    "information is not available\n");
1749	} else {
1750		if (ph->env.cpu != NULL) {
1751			for (i = 0; i < cpu_nr; i++)
1752				fprintf(fp, "# CPU %d: Core ID %d, "
1753					    "Socket ID %d\n",
1754					    i, ph->env.cpu[i].core_id,
1755					    ph->env.cpu[i].socket_id);
1756		} else
1757			fprintf(fp, "# Core ID and Socket ID "
1758				    "information is not available\n");
1759	}
1760}
1761
1762static void print_clockid(struct feat_fd *ff, FILE *fp)
1763{
1764	fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1765		ff->ph->env.clock.clockid_res_ns * 1000);
1766}
1767
1768static void print_clock_data(struct feat_fd *ff, FILE *fp)
1769{
1770	struct timespec clockid_ns;
1771	char tstr[64], date[64];
1772	struct timeval tod_ns;
1773	clockid_t clockid;
1774	struct tm ltime;
1775	u64 ref;
1776
1777	if (!ff->ph->env.clock.enabled) {
1778		fprintf(fp, "# reference time disabled\n");
1779		return;
1780	}
1781
1782	/* Compute TOD time. */
1783	ref = ff->ph->env.clock.tod_ns;
1784	tod_ns.tv_sec = ref / NSEC_PER_SEC;
1785	ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1786	tod_ns.tv_usec = ref / NSEC_PER_USEC;
1787
1788	/* Compute clockid time. */
1789	ref = ff->ph->env.clock.clockid_ns;
1790	clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1791	ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1792	clockid_ns.tv_nsec = ref;
1793
1794	clockid = ff->ph->env.clock.clockid;
1795
1796	if (localtime_r(&tod_ns.tv_sec, &ltime) == NULL)
1797		snprintf(tstr, sizeof(tstr), "<error>");
1798	else {
1799		strftime(date, sizeof(date), "%F %T", &ltime);
1800		scnprintf(tstr, sizeof(tstr), "%s.%06d",
1801			  date, (int) tod_ns.tv_usec);
1802	}
1803
1804	fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1805	fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1806		    tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1807		    (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1808		    clockid_name(clockid));
1809}
1810
1811static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1812{
1813	int i;
1814	struct hybrid_node *n;
1815
1816	fprintf(fp, "# hybrid cpu system:\n");
1817	for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1818		n = &ff->ph->env.hybrid_nodes[i];
1819		fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1820	}
1821}
1822
1823static void print_dir_format(struct feat_fd *ff, FILE *fp)
1824{
1825	struct perf_session *session;
1826	struct perf_data *data;
1827
1828	session = container_of(ff->ph, struct perf_session, header);
1829	data = session->data;
1830
1831	fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1832}
1833
1834#ifdef HAVE_LIBBPF_SUPPORT
1835static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1836{
1837	struct perf_env *env = &ff->ph->env;
1838	struct rb_root *root;
1839	struct rb_node *next;
1840
1841	down_read(&env->bpf_progs.lock);
1842
1843	root = &env->bpf_progs.infos;
1844	next = rb_first(root);
1845
1846	while (next) {
1847		struct bpf_prog_info_node *node;
1848
1849		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1850		next = rb_next(&node->rb_node);
1851
1852		__bpf_event__print_bpf_prog_info(&node->info_linear->info,
1853						 env, fp);
1854	}
1855
1856	up_read(&env->bpf_progs.lock);
1857}
1858
1859static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1860{
1861	struct perf_env *env = &ff->ph->env;
1862	struct rb_root *root;
1863	struct rb_node *next;
1864
1865	down_read(&env->bpf_progs.lock);
1866
1867	root = &env->bpf_progs.btfs;
1868	next = rb_first(root);
1869
1870	while (next) {
1871		struct btf_node *node;
1872
1873		node = rb_entry(next, struct btf_node, rb_node);
1874		next = rb_next(&node->rb_node);
1875		fprintf(fp, "# btf info of id %u\n", node->id);
1876	}
1877
1878	up_read(&env->bpf_progs.lock);
1879}
1880#endif // HAVE_LIBBPF_SUPPORT
1881
1882static void free_event_desc(struct evsel *events)
1883{
1884	struct evsel *evsel;
1885
1886	if (!events)
1887		return;
1888
1889	for (evsel = events; evsel->core.attr.size; evsel++) {
1890		zfree(&evsel->name);
1891		zfree(&evsel->core.id);
1892	}
1893
1894	free(events);
1895}
1896
1897static bool perf_attr_check(struct perf_event_attr *attr)
1898{
1899	if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1900		pr_warning("Reserved bits are set unexpectedly. "
1901			   "Please update perf tool.\n");
1902		return false;
1903	}
1904
1905	if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1906		pr_warning("Unknown sample type (0x%llx) is detected. "
1907			   "Please update perf tool.\n",
1908			   attr->sample_type);
1909		return false;
1910	}
1911
1912	if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1913		pr_warning("Unknown read format (0x%llx) is detected. "
1914			   "Please update perf tool.\n",
1915			   attr->read_format);
1916		return false;
1917	}
1918
1919	if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1920	    (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1921		pr_warning("Unknown branch sample type (0x%llx) is detected. "
1922			   "Please update perf tool.\n",
1923			   attr->branch_sample_type);
1924
1925		return false;
1926	}
1927
1928	return true;
1929}
1930
1931static struct evsel *read_event_desc(struct feat_fd *ff)
1932{
1933	struct evsel *evsel, *events = NULL;
1934	u64 *id;
1935	void *buf = NULL;
1936	u32 nre, sz, nr, i, j;
1937	size_t msz;
1938
1939	/* number of events */
1940	if (do_read_u32(ff, &nre))
1941		goto error;
1942
1943	if (do_read_u32(ff, &sz))
1944		goto error;
1945
1946	/* buffer to hold on file attr struct */
1947	buf = malloc(sz);
1948	if (!buf)
1949		goto error;
1950
1951	/* the last event terminates with evsel->core.attr.size == 0: */
1952	events = calloc(nre + 1, sizeof(*events));
1953	if (!events)
1954		goto error;
1955
1956	msz = sizeof(evsel->core.attr);
1957	if (sz < msz)
1958		msz = sz;
1959
1960	for (i = 0, evsel = events; i < nre; evsel++, i++) {
1961		evsel->core.idx = i;
1962
1963		/*
1964		 * must read entire on-file attr struct to
1965		 * sync up with layout.
1966		 */
1967		if (__do_read(ff, buf, sz))
1968			goto error;
1969
1970		if (ff->ph->needs_swap)
1971			perf_event__attr_swap(buf);
1972
1973		memcpy(&evsel->core.attr, buf, msz);
1974
1975		if (!perf_attr_check(&evsel->core.attr))
1976			goto error;
1977
1978		if (do_read_u32(ff, &nr))
1979			goto error;
1980
1981		if (ff->ph->needs_swap)
1982			evsel->needs_swap = true;
1983
1984		evsel->name = do_read_string(ff);
1985		if (!evsel->name)
1986			goto error;
1987
1988		if (!nr)
1989			continue;
1990
1991		id = calloc(nr, sizeof(*id));
1992		if (!id)
1993			goto error;
1994		evsel->core.ids = nr;
1995		evsel->core.id = id;
1996
1997		for (j = 0 ; j < nr; j++) {
1998			if (do_read_u64(ff, id))
1999				goto error;
2000			id++;
2001		}
2002	}
2003out:
2004	free(buf);
2005	return events;
2006error:
2007	free_event_desc(events);
2008	events = NULL;
2009	goto out;
2010}
2011
2012static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
2013				void *priv __maybe_unused)
2014{
2015	return fprintf(fp, ", %s = %s", name, val);
2016}
2017
2018static void print_event_desc(struct feat_fd *ff, FILE *fp)
2019{
2020	struct evsel *evsel, *events;
2021	u32 j;
2022	u64 *id;
2023
2024	if (ff->events)
2025		events = ff->events;
2026	else
2027		events = read_event_desc(ff);
2028
2029	if (!events) {
2030		fprintf(fp, "# event desc: not available or unable to read\n");
2031		return;
2032	}
2033
2034	for (evsel = events; evsel->core.attr.size; evsel++) {
2035		fprintf(fp, "# event : name = %s, ", evsel->name);
2036
2037		if (evsel->core.ids) {
2038			fprintf(fp, ", id = {");
2039			for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
2040				if (j)
2041					fputc(',', fp);
2042				fprintf(fp, " %"PRIu64, *id);
2043			}
2044			fprintf(fp, " }");
2045		}
2046
2047		perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
2048
2049		fputc('\n', fp);
2050	}
2051
2052	free_event_desc(events);
2053	ff->events = NULL;
2054}
2055
2056static void print_total_mem(struct feat_fd *ff, FILE *fp)
2057{
2058	fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
2059}
2060
2061static void print_numa_topology(struct feat_fd *ff, FILE *fp)
2062{
2063	int i;
2064	struct numa_node *n;
2065
2066	for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
2067		n = &ff->ph->env.numa_nodes[i];
2068
2069		fprintf(fp, "# node%u meminfo  : total = %"PRIu64" kB,"
2070			    " free = %"PRIu64" kB\n",
2071			n->node, n->mem_total, n->mem_free);
2072
2073		fprintf(fp, "# node%u cpu list : ", n->node);
2074		cpu_map__fprintf(n->map, fp);
2075	}
2076}
2077
2078static void print_cpuid(struct feat_fd *ff, FILE *fp)
2079{
2080	fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
2081}
2082
2083static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
2084{
2085	fprintf(fp, "# contains samples with branch stack\n");
2086}
2087
2088static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
2089{
2090	fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
2091}
2092
2093static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
2094{
2095	fprintf(fp, "# contains stat data\n");
2096}
2097
2098static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
2099{
2100	int i;
2101
2102	fprintf(fp, "# CPU cache info:\n");
2103	for (i = 0; i < ff->ph->env.caches_cnt; i++) {
2104		fprintf(fp, "#  ");
2105		cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
2106	}
2107}
2108
2109static void print_compressed(struct feat_fd *ff, FILE *fp)
2110{
2111	fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
2112		ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
2113		ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2114}
2115
2116static void __print_pmu_caps(FILE *fp, int nr_caps, char **caps, char *pmu_name)
2117{
2118	const char *delimiter = "";
2119	int i;
2120
2121	if (!nr_caps) {
2122		fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2123		return;
2124	}
2125
2126	fprintf(fp, "# %s pmu capabilities: ", pmu_name);
2127	for (i = 0; i < nr_caps; i++) {
2128		fprintf(fp, "%s%s", delimiter, caps[i]);
2129		delimiter = ", ";
2130	}
2131
2132	fprintf(fp, "\n");
2133}
2134
2135static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2136{
2137	__print_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2138			 ff->ph->env.cpu_pmu_caps, (char *)"cpu");
2139}
2140
2141static void print_pmu_caps(struct feat_fd *ff, FILE *fp)
2142{
2143	struct pmu_caps *pmu_caps;
2144
2145	for (int i = 0; i < ff->ph->env.nr_pmus_with_caps; i++) {
2146		pmu_caps = &ff->ph->env.pmu_caps[i];
2147		__print_pmu_caps(fp, pmu_caps->nr_caps, pmu_caps->caps,
2148				 pmu_caps->pmu_name);
2149	}
2150}
2151
2152static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2153{
2154	const char *delimiter = "# pmu mappings: ";
2155	char *str, *tmp;
2156	u32 pmu_num;
2157	u32 type;
2158
2159	pmu_num = ff->ph->env.nr_pmu_mappings;
2160	if (!pmu_num) {
2161		fprintf(fp, "# pmu mappings: not available\n");
2162		return;
2163	}
2164
2165	str = ff->ph->env.pmu_mappings;
2166
2167	while (pmu_num) {
2168		type = strtoul(str, &tmp, 0);
2169		if (*tmp != ':')
2170			goto error;
2171
2172		str = tmp + 1;
2173		fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2174
2175		delimiter = ", ";
2176		str += strlen(str) + 1;
2177		pmu_num--;
2178	}
2179
2180	fprintf(fp, "\n");
2181
2182	if (!pmu_num)
2183		return;
2184error:
2185	fprintf(fp, "# pmu mappings: unable to read\n");
2186}
2187
2188static void print_group_desc(struct feat_fd *ff, FILE *fp)
2189{
2190	struct perf_session *session;
2191	struct evsel *evsel;
2192	u32 nr = 0;
2193
2194	session = container_of(ff->ph, struct perf_session, header);
2195
2196	evlist__for_each_entry(session->evlist, evsel) {
2197		if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2198			fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2199
2200			nr = evsel->core.nr_members - 1;
2201		} else if (nr) {
2202			fprintf(fp, ",%s", evsel__name(evsel));
2203
2204			if (--nr == 0)
2205				fprintf(fp, "}\n");
2206		}
2207	}
2208}
2209
2210static void print_sample_time(struct feat_fd *ff, FILE *fp)
2211{
2212	struct perf_session *session;
2213	char time_buf[32];
2214	double d;
2215
2216	session = container_of(ff->ph, struct perf_session, header);
2217
2218	timestamp__scnprintf_usec(session->evlist->first_sample_time,
2219				  time_buf, sizeof(time_buf));
2220	fprintf(fp, "# time of first sample : %s\n", time_buf);
2221
2222	timestamp__scnprintf_usec(session->evlist->last_sample_time,
2223				  time_buf, sizeof(time_buf));
2224	fprintf(fp, "# time of last sample : %s\n", time_buf);
2225
2226	d = (double)(session->evlist->last_sample_time -
2227		session->evlist->first_sample_time) / NSEC_PER_MSEC;
2228
2229	fprintf(fp, "# sample duration : %10.3f ms\n", d);
2230}
2231
2232static void memory_node__fprintf(struct memory_node *n,
2233				 unsigned long long bsize, FILE *fp)
2234{
2235	char buf_map[100], buf_size[50];
2236	unsigned long long size;
2237
2238	size = bsize * bitmap_weight(n->set, n->size);
2239	unit_number__scnprintf(buf_size, 50, size);
2240
2241	bitmap_scnprintf(n->set, n->size, buf_map, 100);
2242	fprintf(fp, "#  %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2243}
2244
2245static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2246{
2247	struct memory_node *nodes;
2248	int i, nr;
2249
2250	nodes = ff->ph->env.memory_nodes;
2251	nr    = ff->ph->env.nr_memory_nodes;
2252
2253	fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2254		nr, ff->ph->env.memory_bsize);
2255
2256	for (i = 0; i < nr; i++) {
2257		memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2258	}
2259}
2260
2261static int __event_process_build_id(struct perf_record_header_build_id *bev,
2262				    char *filename,
2263				    struct perf_session *session)
2264{
2265	int err = -1;
2266	struct machine *machine;
2267	u16 cpumode;
2268	struct dso *dso;
2269	enum dso_space_type dso_space;
2270
2271	machine = perf_session__findnew_machine(session, bev->pid);
2272	if (!machine)
2273		goto out;
2274
2275	cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2276
2277	switch (cpumode) {
2278	case PERF_RECORD_MISC_KERNEL:
2279		dso_space = DSO_SPACE__KERNEL;
2280		break;
2281	case PERF_RECORD_MISC_GUEST_KERNEL:
2282		dso_space = DSO_SPACE__KERNEL_GUEST;
2283		break;
2284	case PERF_RECORD_MISC_USER:
2285	case PERF_RECORD_MISC_GUEST_USER:
2286		dso_space = DSO_SPACE__USER;
2287		break;
2288	default:
2289		goto out;
2290	}
2291
2292	dso = machine__findnew_dso(machine, filename);
2293	if (dso != NULL) {
2294		char sbuild_id[SBUILD_ID_SIZE];
2295		struct build_id bid;
2296		size_t size = BUILD_ID_SIZE;
2297
2298		if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2299			size = bev->size;
2300
2301		build_id__init(&bid, bev->data, size);
2302		dso__set_build_id(dso, &bid);
2303		dso->header_build_id = 1;
2304
2305		if (dso_space != DSO_SPACE__USER) {
2306			struct kmod_path m = { .name = NULL, };
2307
2308			if (!kmod_path__parse_name(&m, filename) && m.kmod)
2309				dso__set_module_info(dso, &m, machine);
2310
2311			dso->kernel = dso_space;
2312			free(m.name);
2313		}
2314
2315		build_id__sprintf(&dso->bid, sbuild_id);
2316		pr_debug("build id event received for %s: %s [%zu]\n",
2317			 dso->long_name, sbuild_id, size);
2318		dso__put(dso);
2319	}
2320
2321	err = 0;
2322out:
2323	return err;
2324}
2325
2326static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2327						 int input, u64 offset, u64 size)
2328{
2329	struct perf_session *session = container_of(header, struct perf_session, header);
2330	struct {
2331		struct perf_event_header   header;
2332		u8			   build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2333		char			   filename[0];
2334	} old_bev;
2335	struct perf_record_header_build_id bev;
2336	char filename[PATH_MAX];
2337	u64 limit = offset + size;
2338
2339	while (offset < limit) {
2340		ssize_t len;
2341
2342		if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2343			return -1;
2344
2345		if (header->needs_swap)
2346			perf_event_header__bswap(&old_bev.header);
2347
2348		len = old_bev.header.size - sizeof(old_bev);
2349		if (readn(input, filename, len) != len)
2350			return -1;
2351
2352		bev.header = old_bev.header;
2353
2354		/*
2355		 * As the pid is the missing value, we need to fill
2356		 * it properly. The header.misc value give us nice hint.
2357		 */
2358		bev.pid	= HOST_KERNEL_ID;
2359		if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2360		    bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2361			bev.pid	= DEFAULT_GUEST_KERNEL_ID;
2362
2363		memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2364		__event_process_build_id(&bev, filename, session);
2365
2366		offset += bev.header.size;
2367	}
2368
2369	return 0;
2370}
2371
2372static int perf_header__read_build_ids(struct perf_header *header,
2373				       int input, u64 offset, u64 size)
2374{
2375	struct perf_session *session = container_of(header, struct perf_session, header);
2376	struct perf_record_header_build_id bev;
2377	char filename[PATH_MAX];
2378	u64 limit = offset + size, orig_offset = offset;
2379	int err = -1;
2380
2381	while (offset < limit) {
2382		ssize_t len;
2383
2384		if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2385			goto out;
2386
2387		if (header->needs_swap)
2388			perf_event_header__bswap(&bev.header);
2389
2390		len = bev.header.size - sizeof(bev);
2391		if (readn(input, filename, len) != len)
2392			goto out;
2393		/*
2394		 * The a1645ce1 changeset:
2395		 *
2396		 * "perf: 'perf kvm' tool for monitoring guest performance from host"
2397		 *
2398		 * Added a field to struct perf_record_header_build_id that broke the file
2399		 * format.
2400		 *
2401		 * Since the kernel build-id is the first entry, process the
2402		 * table using the old format if the well known
2403		 * '[kernel.kallsyms]' string for the kernel build-id has the
2404		 * first 4 characters chopped off (where the pid_t sits).
2405		 */
2406		if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2407			if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2408				return -1;
2409			return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2410		}
2411
2412		__event_process_build_id(&bev, filename, session);
2413
2414		offset += bev.header.size;
2415	}
2416	err = 0;
2417out:
2418	return err;
2419}
2420
2421/* Macro for features that simply need to read and store a string. */
2422#define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2423static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2424{\
2425	free(ff->ph->env.__feat_env);		     \
2426	ff->ph->env.__feat_env = do_read_string(ff); \
2427	return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2428}
2429
2430FEAT_PROCESS_STR_FUN(hostname, hostname);
2431FEAT_PROCESS_STR_FUN(osrelease, os_release);
2432FEAT_PROCESS_STR_FUN(version, version);
2433FEAT_PROCESS_STR_FUN(arch, arch);
2434FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2435FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2436
2437#ifdef HAVE_LIBTRACEEVENT
2438static int process_tracing_data(struct feat_fd *ff, void *data)
2439{
2440	ssize_t ret = trace_report(ff->fd, data, false);
2441
2442	return ret < 0 ? -1 : 0;
2443}
2444#endif
2445
2446static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2447{
2448	if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2449		pr_debug("Failed to read buildids, continuing...\n");
2450	return 0;
2451}
2452
2453static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2454{
2455	int ret;
2456	u32 nr_cpus_avail, nr_cpus_online;
2457
2458	ret = do_read_u32(ff, &nr_cpus_avail);
2459	if (ret)
2460		return ret;
2461
2462	ret = do_read_u32(ff, &nr_cpus_online);
2463	if (ret)
2464		return ret;
2465	ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2466	ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2467	return 0;
2468}
2469
2470static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2471{
2472	u64 total_mem;
2473	int ret;
2474
2475	ret = do_read_u64(ff, &total_mem);
2476	if (ret)
2477		return -1;
2478	ff->ph->env.total_mem = (unsigned long long)total_mem;
2479	return 0;
2480}
2481
2482static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2483{
2484	struct evsel *evsel;
2485
2486	evlist__for_each_entry(evlist, evsel) {
2487		if (evsel->core.idx == idx)
2488			return evsel;
2489	}
2490
2491	return NULL;
2492}
2493
2494static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2495{
2496	struct evsel *evsel;
2497
2498	if (!event->name)
2499		return;
2500
2501	evsel = evlist__find_by_index(evlist, event->core.idx);
2502	if (!evsel)
2503		return;
2504
2505	if (evsel->name)
2506		return;
2507
2508	evsel->name = strdup(event->name);
2509}
2510
2511static int
2512process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2513{
2514	struct perf_session *session;
2515	struct evsel *evsel, *events = read_event_desc(ff);
2516
2517	if (!events)
2518		return 0;
2519
2520	session = container_of(ff->ph, struct perf_session, header);
2521
2522	if (session->data->is_pipe) {
2523		/* Save events for reading later by print_event_desc,
2524		 * since they can't be read again in pipe mode. */
2525		ff->events = events;
2526	}
2527
2528	for (evsel = events; evsel->core.attr.size; evsel++)
2529		evlist__set_event_name(session->evlist, evsel);
2530
2531	if (!session->data->is_pipe)
2532		free_event_desc(events);
2533
2534	return 0;
2535}
2536
2537static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2538{
2539	char *str, *cmdline = NULL, **argv = NULL;
2540	u32 nr, i, len = 0;
2541
2542	if (do_read_u32(ff, &nr))
2543		return -1;
2544
2545	ff->ph->env.nr_cmdline = nr;
2546
2547	cmdline = zalloc(ff->size + nr + 1);
2548	if (!cmdline)
2549		return -1;
2550
2551	argv = zalloc(sizeof(char *) * (nr + 1));
2552	if (!argv)
2553		goto error;
2554
2555	for (i = 0; i < nr; i++) {
2556		str = do_read_string(ff);
2557		if (!str)
2558			goto error;
2559
2560		argv[i] = cmdline + len;
2561		memcpy(argv[i], str, strlen(str) + 1);
2562		len += strlen(str) + 1;
2563		free(str);
2564	}
2565	ff->ph->env.cmdline = cmdline;
2566	ff->ph->env.cmdline_argv = (const char **) argv;
2567	return 0;
2568
2569error:
2570	free(argv);
2571	free(cmdline);
2572	return -1;
2573}
2574
2575static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2576{
2577	u32 nr, i;
2578	char *str;
2579	struct strbuf sb;
2580	int cpu_nr = ff->ph->env.nr_cpus_avail;
2581	u64 size = 0;
2582	struct perf_header *ph = ff->ph;
2583	bool do_core_id_test = true;
2584
2585	ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2586	if (!ph->env.cpu)
2587		return -1;
2588
2589	if (do_read_u32(ff, &nr))
2590		goto free_cpu;
2591
2592	ph->env.nr_sibling_cores = nr;
2593	size += sizeof(u32);
2594	if (strbuf_init(&sb, 128) < 0)
2595		goto free_cpu;
2596
2597	for (i = 0; i < nr; i++) {
2598		str = do_read_string(ff);
2599		if (!str)
2600			goto error;
2601
2602		/* include a NULL character at the end */
2603		if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2604			goto error;
2605		size += string_size(str);
2606		free(str);
2607	}
2608	ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2609
2610	if (do_read_u32(ff, &nr))
2611		return -1;
2612
2613	ph->env.nr_sibling_threads = nr;
2614	size += sizeof(u32);
2615
2616	for (i = 0; i < nr; i++) {
2617		str = do_read_string(ff);
2618		if (!str)
2619			goto error;
2620
2621		/* include a NULL character at the end */
2622		if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2623			goto error;
2624		size += string_size(str);
2625		free(str);
2626	}
2627	ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2628
2629	/*
2630	 * The header may be from old perf,
2631	 * which doesn't include core id and socket id information.
2632	 */
2633	if (ff->size <= size) {
2634		zfree(&ph->env.cpu);
2635		return 0;
2636	}
2637
2638	/* On s390 the socket_id number is not related to the numbers of cpus.
2639	 * The socket_id number might be higher than the numbers of cpus.
2640	 * This depends on the configuration.
2641	 * AArch64 is the same.
2642	 */
2643	if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2644			  || !strncmp(ph->env.arch, "aarch64", 7)))
2645		do_core_id_test = false;
2646
2647	for (i = 0; i < (u32)cpu_nr; i++) {
2648		if (do_read_u32(ff, &nr))
2649			goto free_cpu;
2650
2651		ph->env.cpu[i].core_id = nr;
2652		size += sizeof(u32);
2653
2654		if (do_read_u32(ff, &nr))
2655			goto free_cpu;
2656
2657		if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2658			pr_debug("socket_id number is too big."
2659				 "You may need to upgrade the perf tool.\n");
2660			goto free_cpu;
2661		}
2662
2663		ph->env.cpu[i].socket_id = nr;
2664		size += sizeof(u32);
2665	}
2666
2667	/*
2668	 * The header may be from old perf,
2669	 * which doesn't include die information.
2670	 */
2671	if (ff->size <= size)
2672		return 0;
2673
2674	if (do_read_u32(ff, &nr))
2675		return -1;
2676
2677	ph->env.nr_sibling_dies = nr;
2678	size += sizeof(u32);
2679
2680	for (i = 0; i < nr; i++) {
2681		str = do_read_string(ff);
2682		if (!str)
2683			goto error;
2684
2685		/* include a NULL character at the end */
2686		if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2687			goto error;
2688		size += string_size(str);
2689		free(str);
2690	}
2691	ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2692
2693	for (i = 0; i < (u32)cpu_nr; i++) {
2694		if (do_read_u32(ff, &nr))
2695			goto free_cpu;
2696
2697		ph->env.cpu[i].die_id = nr;
2698	}
2699
2700	return 0;
2701
2702error:
2703	strbuf_release(&sb);
2704free_cpu:
2705	zfree(&ph->env.cpu);
2706	return -1;
2707}
2708
2709static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2710{
2711	struct numa_node *nodes, *n;
2712	u32 nr, i;
2713	char *str;
2714
2715	/* nr nodes */
2716	if (do_read_u32(ff, &nr))
2717		return -1;
2718
2719	nodes = zalloc(sizeof(*nodes) * nr);
2720	if (!nodes)
2721		return -ENOMEM;
2722
2723	for (i = 0; i < nr; i++) {
2724		n = &nodes[i];
2725
2726		/* node number */
2727		if (do_read_u32(ff, &n->node))
2728			goto error;
2729
2730		if (do_read_u64(ff, &n->mem_total))
2731			goto error;
2732
2733		if (do_read_u64(ff, &n->mem_free))
2734			goto error;
2735
2736		str = do_read_string(ff);
2737		if (!str)
2738			goto error;
2739
2740		n->map = perf_cpu_map__new(str);
2741		if (!n->map)
2742			goto error;
2743
2744		free(str);
2745	}
2746	ff->ph->env.nr_numa_nodes = nr;
2747	ff->ph->env.numa_nodes = nodes;
2748	return 0;
2749
2750error:
2751	free(nodes);
2752	return -1;
2753}
2754
2755static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2756{
2757	char *name;
2758	u32 pmu_num;
2759	u32 type;
2760	struct strbuf sb;
2761
2762	if (do_read_u32(ff, &pmu_num))
2763		return -1;
2764
2765	if (!pmu_num) {
2766		pr_debug("pmu mappings not available\n");
2767		return 0;
2768	}
2769
2770	ff->ph->env.nr_pmu_mappings = pmu_num;
2771	if (strbuf_init(&sb, 128) < 0)
2772		return -1;
2773
2774	while (pmu_num) {
2775		if (do_read_u32(ff, &type))
2776			goto error;
2777
2778		name = do_read_string(ff);
2779		if (!name)
2780			goto error;
2781
2782		if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2783			goto error;
2784		/* include a NULL character at the end */
2785		if (strbuf_add(&sb, "", 1) < 0)
2786			goto error;
2787
2788		if (!strcmp(name, "msr"))
2789			ff->ph->env.msr_pmu_type = type;
2790
2791		free(name);
2792		pmu_num--;
2793	}
2794	ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2795	return 0;
2796
2797error:
2798	strbuf_release(&sb);
2799	return -1;
2800}
2801
2802static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2803{
2804	size_t ret = -1;
2805	u32 i, nr, nr_groups;
2806	struct perf_session *session;
2807	struct evsel *evsel, *leader = NULL;
2808	struct group_desc {
2809		char *name;
2810		u32 leader_idx;
2811		u32 nr_members;
2812	} *desc;
2813
2814	if (do_read_u32(ff, &nr_groups))
2815		return -1;
2816
2817	ff->ph->env.nr_groups = nr_groups;
2818	if (!nr_groups) {
2819		pr_debug("group desc not available\n");
2820		return 0;
2821	}
2822
2823	desc = calloc(nr_groups, sizeof(*desc));
2824	if (!desc)
2825		return -1;
2826
2827	for (i = 0; i < nr_groups; i++) {
2828		desc[i].name = do_read_string(ff);
2829		if (!desc[i].name)
2830			goto out_free;
2831
2832		if (do_read_u32(ff, &desc[i].leader_idx))
2833			goto out_free;
2834
2835		if (do_read_u32(ff, &desc[i].nr_members))
2836			goto out_free;
2837	}
2838
2839	/*
2840	 * Rebuild group relationship based on the group_desc
2841	 */
2842	session = container_of(ff->ph, struct perf_session, header);
2843
2844	i = nr = 0;
2845	evlist__for_each_entry(session->evlist, evsel) {
2846		if (i < nr_groups && evsel->core.idx == (int) desc[i].leader_idx) {
2847			evsel__set_leader(evsel, evsel);
2848			/* {anon_group} is a dummy name */
2849			if (strcmp(desc[i].name, "{anon_group}")) {
2850				evsel->group_name = desc[i].name;
2851				desc[i].name = NULL;
2852			}
2853			evsel->core.nr_members = desc[i].nr_members;
2854
2855			if (i >= nr_groups || nr > 0) {
2856				pr_debug("invalid group desc\n");
2857				goto out_free;
2858			}
2859
2860			leader = evsel;
2861			nr = evsel->core.nr_members - 1;
2862			i++;
2863		} else if (nr) {
2864			/* This is a group member */
2865			evsel__set_leader(evsel, leader);
2866
2867			nr--;
2868		}
2869	}
2870
2871	if (i != nr_groups || nr != 0) {
2872		pr_debug("invalid group desc\n");
2873		goto out_free;
2874	}
2875
2876	ret = 0;
2877out_free:
2878	for (i = 0; i < nr_groups; i++)
2879		zfree(&desc[i].name);
2880	free(desc);
2881
2882	return ret;
2883}
2884
2885static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2886{
2887	struct perf_session *session;
2888	int err;
2889
2890	session = container_of(ff->ph, struct perf_session, header);
2891
2892	err = auxtrace_index__process(ff->fd, ff->size, session,
2893				      ff->ph->needs_swap);
2894	if (err < 0)
2895		pr_err("Failed to process auxtrace index\n");
2896	return err;
2897}
2898
2899static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2900{
2901	struct cpu_cache_level *caches;
2902	u32 cnt, i, version;
2903
2904	if (do_read_u32(ff, &version))
2905		return -1;
2906
2907	if (version != 1)
2908		return -1;
2909
2910	if (do_read_u32(ff, &cnt))
2911		return -1;
2912
2913	caches = zalloc(sizeof(*caches) * cnt);
2914	if (!caches)
2915		return -1;
2916
2917	for (i = 0; i < cnt; i++) {
2918		struct cpu_cache_level c;
2919
2920		#define _R(v)						\
2921			if (do_read_u32(ff, &c.v))\
2922				goto out_free_caches;			\
2923
2924		_R(level)
2925		_R(line_size)
2926		_R(sets)
2927		_R(ways)
2928		#undef _R
2929
2930		#define _R(v)					\
2931			c.v = do_read_string(ff);		\
2932			if (!c.v)				\
2933				goto out_free_caches;
2934
2935		_R(type)
2936		_R(size)
2937		_R(map)
2938		#undef _R
2939
2940		caches[i] = c;
2941	}
2942
2943	ff->ph->env.caches = caches;
2944	ff->ph->env.caches_cnt = cnt;
2945	return 0;
2946out_free_caches:
2947	free(caches);
2948	return -1;
2949}
2950
2951static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2952{
2953	struct perf_session *session;
2954	u64 first_sample_time, last_sample_time;
2955	int ret;
2956
2957	session = container_of(ff->ph, struct perf_session, header);
2958
2959	ret = do_read_u64(ff, &first_sample_time);
2960	if (ret)
2961		return -1;
2962
2963	ret = do_read_u64(ff, &last_sample_time);
2964	if (ret)
2965		return -1;
2966
2967	session->evlist->first_sample_time = first_sample_time;
2968	session->evlist->last_sample_time = last_sample_time;
2969	return 0;
2970}
2971
2972static int process_mem_topology(struct feat_fd *ff,
2973				void *data __maybe_unused)
2974{
2975	struct memory_node *nodes;
2976	u64 version, i, nr, bsize;
2977	int ret = -1;
2978
2979	if (do_read_u64(ff, &version))
2980		return -1;
2981
2982	if (version != 1)
2983		return -1;
2984
2985	if (do_read_u64(ff, &bsize))
2986		return -1;
2987
2988	if (do_read_u64(ff, &nr))
2989		return -1;
2990
2991	nodes = zalloc(sizeof(*nodes) * nr);
2992	if (!nodes)
2993		return -1;
2994
2995	for (i = 0; i < nr; i++) {
2996		struct memory_node n;
2997
2998		#define _R(v)				\
2999			if (do_read_u64(ff, &n.v))	\
3000				goto out;		\
3001
3002		_R(node)
3003		_R(size)
3004
3005		#undef _R
3006
3007		if (do_read_bitmap(ff, &n.set, &n.size))
3008			goto out;
3009
3010		nodes[i] = n;
3011	}
3012
3013	ff->ph->env.memory_bsize    = bsize;
3014	ff->ph->env.memory_nodes    = nodes;
3015	ff->ph->env.nr_memory_nodes = nr;
3016	ret = 0;
3017
3018out:
3019	if (ret)
3020		free(nodes);
3021	return ret;
3022}
3023
3024static int process_clockid(struct feat_fd *ff,
3025			   void *data __maybe_unused)
3026{
3027	if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
3028		return -1;
3029
3030	return 0;
3031}
3032
3033static int process_clock_data(struct feat_fd *ff,
3034			      void *_data __maybe_unused)
3035{
3036	u32 data32;
3037	u64 data64;
3038
3039	/* version */
3040	if (do_read_u32(ff, &data32))
3041		return -1;
3042
3043	if (data32 != 1)
3044		return -1;
3045
3046	/* clockid */
3047	if (do_read_u32(ff, &data32))
3048		return -1;
3049
3050	ff->ph->env.clock.clockid = data32;
3051
3052	/* TOD ref time */
3053	if (do_read_u64(ff, &data64))
3054		return -1;
3055
3056	ff->ph->env.clock.tod_ns = data64;
3057
3058	/* clockid ref time */
3059	if (do_read_u64(ff, &data64))
3060		return -1;
3061
3062	ff->ph->env.clock.clockid_ns = data64;
3063	ff->ph->env.clock.enabled = true;
3064	return 0;
3065}
3066
3067static int process_hybrid_topology(struct feat_fd *ff,
3068				   void *data __maybe_unused)
3069{
3070	struct hybrid_node *nodes, *n;
3071	u32 nr, i;
3072
3073	/* nr nodes */
3074	if (do_read_u32(ff, &nr))
3075		return -1;
3076
3077	nodes = zalloc(sizeof(*nodes) * nr);
3078	if (!nodes)
3079		return -ENOMEM;
3080
3081	for (i = 0; i < nr; i++) {
3082		n = &nodes[i];
3083
3084		n->pmu_name = do_read_string(ff);
3085		if (!n->pmu_name)
3086			goto error;
3087
3088		n->cpus = do_read_string(ff);
3089		if (!n->cpus)
3090			goto error;
3091	}
3092
3093	ff->ph->env.nr_hybrid_nodes = nr;
3094	ff->ph->env.hybrid_nodes = nodes;
3095	return 0;
3096
3097error:
3098	for (i = 0; i < nr; i++) {
3099		free(nodes[i].pmu_name);
3100		free(nodes[i].cpus);
3101	}
3102
3103	free(nodes);
3104	return -1;
3105}
3106
3107static int process_dir_format(struct feat_fd *ff,
3108			      void *_data __maybe_unused)
3109{
3110	struct perf_session *session;
3111	struct perf_data *data;
3112
3113	session = container_of(ff->ph, struct perf_session, header);
3114	data = session->data;
3115
3116	if (WARN_ON(!perf_data__is_dir(data)))
3117		return -1;
3118
3119	return do_read_u64(ff, &data->dir.version);
3120}
3121
3122#ifdef HAVE_LIBBPF_SUPPORT
3123static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3124{
3125	struct bpf_prog_info_node *info_node;
3126	struct perf_env *env = &ff->ph->env;
3127	struct perf_bpil *info_linear;
3128	u32 count, i;
3129	int err = -1;
3130
3131	if (ff->ph->needs_swap) {
3132		pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3133		return 0;
3134	}
3135
3136	if (do_read_u32(ff, &count))
3137		return -1;
3138
3139	down_write(&env->bpf_progs.lock);
3140
3141	for (i = 0; i < count; ++i) {
3142		u32 info_len, data_len;
3143
3144		info_linear = NULL;
3145		info_node = NULL;
3146		if (do_read_u32(ff, &info_len))
3147			goto out;
3148		if (do_read_u32(ff, &data_len))
3149			goto out;
3150
3151		if (info_len > sizeof(struct bpf_prog_info)) {
3152			pr_warning("detected invalid bpf_prog_info\n");
3153			goto out;
3154		}
3155
3156		info_linear = malloc(sizeof(struct perf_bpil) +
3157				     data_len);
3158		if (!info_linear)
3159			goto out;
3160		info_linear->info_len = sizeof(struct bpf_prog_info);
3161		info_linear->data_len = data_len;
3162		if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3163			goto out;
3164		if (__do_read(ff, &info_linear->info, info_len))
3165			goto out;
3166		if (info_len < sizeof(struct bpf_prog_info))
3167			memset(((void *)(&info_linear->info)) + info_len, 0,
3168			       sizeof(struct bpf_prog_info) - info_len);
3169
3170		if (__do_read(ff, info_linear->data, data_len))
3171			goto out;
3172
3173		info_node = malloc(sizeof(struct bpf_prog_info_node));
3174		if (!info_node)
3175			goto out;
3176
3177		/* after reading from file, translate offset to address */
3178		bpil_offs_to_addr(info_linear);
3179		info_node->info_linear = info_linear;
3180		__perf_env__insert_bpf_prog_info(env, info_node);
3181	}
3182
3183	up_write(&env->bpf_progs.lock);
3184	return 0;
3185out:
3186	free(info_linear);
3187	free(info_node);
3188	up_write(&env->bpf_progs.lock);
3189	return err;
3190}
3191
3192static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3193{
3194	struct perf_env *env = &ff->ph->env;
3195	struct btf_node *node = NULL;
3196	u32 count, i;
3197	int err = -1;
3198
3199	if (ff->ph->needs_swap) {
3200		pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3201		return 0;
3202	}
3203
3204	if (do_read_u32(ff, &count))
3205		return -1;
3206
3207	down_write(&env->bpf_progs.lock);
3208
3209	for (i = 0; i < count; ++i) {
3210		u32 id, data_size;
3211
3212		if (do_read_u32(ff, &id))
3213			goto out;
3214		if (do_read_u32(ff, &data_size))
3215			goto out;
3216
3217		node = malloc(sizeof(struct btf_node) + data_size);
3218		if (!node)
3219			goto out;
3220
3221		node->id = id;
3222		node->data_size = data_size;
3223
3224		if (__do_read(ff, node->data, data_size))
3225			goto out;
3226
3227		__perf_env__insert_btf(env, node);
3228		node = NULL;
3229	}
3230
3231	err = 0;
3232out:
3233	up_write(&env->bpf_progs.lock);
3234	free(node);
3235	return err;
3236}
3237#endif // HAVE_LIBBPF_SUPPORT
3238
3239static int process_compressed(struct feat_fd *ff,
3240			      void *data __maybe_unused)
3241{
3242	if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3243		return -1;
3244
3245	if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3246		return -1;
3247
3248	if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3249		return -1;
3250
3251	if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3252		return -1;
3253
3254	if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3255		return -1;
3256
3257	return 0;
3258}
3259
3260static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
3261			      char ***caps, unsigned int *max_branches)
3262{
3263	char *name, *value, *ptr;
3264	u32 nr_pmu_caps, i;
3265
3266	*nr_caps = 0;
3267	*caps = NULL;
3268
3269	if (do_read_u32(ff, &nr_pmu_caps))
3270		return -1;
3271
3272	if (!nr_pmu_caps)
3273		return 0;
3274
3275	*caps = zalloc(sizeof(char *) * nr_pmu_caps);
3276	if (!*caps)
3277		return -1;
3278
3279	for (i = 0; i < nr_pmu_caps; i++) {
3280		name = do_read_string(ff);
3281		if (!name)
3282			goto error;
3283
3284		value = do_read_string(ff);
3285		if (!value)
3286			goto free_name;
3287
3288		if (asprintf(&ptr, "%s=%s", name, value) < 0)
3289			goto free_value;
3290
3291		(*caps)[i] = ptr;
3292
3293		if (!strcmp(name, "branches"))
3294			*max_branches = atoi(value);
3295
3296		free(value);
3297		free(name);
3298	}
3299	*nr_caps = nr_pmu_caps;
3300	return 0;
3301
3302free_value:
3303	free(value);
3304free_name:
3305	free(name);
3306error:
3307	for (; i > 0; i--)
3308		free((*caps)[i - 1]);
3309	free(*caps);
3310	*caps = NULL;
3311	*nr_caps = 0;
3312	return -1;
3313}
3314
3315static int process_cpu_pmu_caps(struct feat_fd *ff,
3316				void *data __maybe_unused)
3317{
3318	int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3319				     &ff->ph->env.cpu_pmu_caps,
3320				     &ff->ph->env.max_branches);
3321
3322	if (!ret && !ff->ph->env.cpu_pmu_caps)
3323		pr_debug("cpu pmu capabilities not available\n");
3324	return ret;
3325}
3326
3327static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused)
3328{
3329	struct pmu_caps *pmu_caps;
3330	u32 nr_pmu, i;
3331	int ret;
3332	int j;
3333
3334	if (do_read_u32(ff, &nr_pmu))
3335		return -1;
3336
3337	if (!nr_pmu) {
3338		pr_debug("pmu capabilities not available\n");
3339		return 0;
3340	}
3341
3342	pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu);
3343	if (!pmu_caps)
3344		return -ENOMEM;
3345
3346	for (i = 0; i < nr_pmu; i++) {
3347		ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps,
3348					 &pmu_caps[i].caps,
3349					 &pmu_caps[i].max_branches);
3350		if (ret)
3351			goto err;
3352
3353		pmu_caps[i].pmu_name = do_read_string(ff);
3354		if (!pmu_caps[i].pmu_name) {
3355			ret = -1;
3356			goto err;
3357		}
3358		if (!pmu_caps[i].nr_caps) {
3359			pr_debug("%s pmu capabilities not available\n",
3360				 pmu_caps[i].pmu_name);
3361		}
3362	}
3363
3364	ff->ph->env.nr_pmus_with_caps = nr_pmu;
3365	ff->ph->env.pmu_caps = pmu_caps;
3366	return 0;
3367
3368err:
3369	for (i = 0; i < nr_pmu; i++) {
3370		for (j = 0; j < pmu_caps[i].nr_caps; j++)
3371			free(pmu_caps[i].caps[j]);
3372		free(pmu_caps[i].caps);
3373		free(pmu_caps[i].pmu_name);
3374	}
3375
3376	free(pmu_caps);
3377	return ret;
3378}
3379
3380#define FEAT_OPR(n, func, __full_only) \
3381	[HEADER_##n] = {					\
3382		.name	    = __stringify(n),			\
3383		.write	    = write_##func,			\
3384		.print	    = print_##func,			\
3385		.full_only  = __full_only,			\
3386		.process    = process_##func,			\
3387		.synthesize = true				\
3388	}
3389
3390#define FEAT_OPN(n, func, __full_only) \
3391	[HEADER_##n] = {					\
3392		.name	    = __stringify(n),			\
3393		.write	    = write_##func,			\
3394		.print	    = print_##func,			\
3395		.full_only  = __full_only,			\
3396		.process    = process_##func			\
3397	}
3398
3399/* feature_ops not implemented: */
3400#define print_tracing_data	NULL
3401#define print_build_id		NULL
3402
3403#define process_branch_stack	NULL
3404#define process_stat		NULL
3405
3406// Only used in util/synthetic-events.c
3407const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3408
3409const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3410#ifdef HAVE_LIBTRACEEVENT
3411	FEAT_OPN(TRACING_DATA,	tracing_data,	false),
3412#endif
3413	FEAT_OPN(BUILD_ID,	build_id,	false),
3414	FEAT_OPR(HOSTNAME,	hostname,	false),
3415	FEAT_OPR(OSRELEASE,	osrelease,	false),
3416	FEAT_OPR(VERSION,	version,	false),
3417	FEAT_OPR(ARCH,		arch,		false),
3418	FEAT_OPR(NRCPUS,	nrcpus,		false),
3419	FEAT_OPR(CPUDESC,	cpudesc,	false),
3420	FEAT_OPR(CPUID,		cpuid,		false),
3421	FEAT_OPR(TOTAL_MEM,	total_mem,	false),
3422	FEAT_OPR(EVENT_DESC,	event_desc,	false),
3423	FEAT_OPR(CMDLINE,	cmdline,	false),
3424	FEAT_OPR(CPU_TOPOLOGY,	cpu_topology,	true),
3425	FEAT_OPR(NUMA_TOPOLOGY,	numa_topology,	true),
3426	FEAT_OPN(BRANCH_STACK,	branch_stack,	false),
3427	FEAT_OPR(PMU_MAPPINGS,	pmu_mappings,	false),
3428	FEAT_OPR(GROUP_DESC,	group_desc,	false),
3429	FEAT_OPN(AUXTRACE,	auxtrace,	false),
3430	FEAT_OPN(STAT,		stat,		false),
3431	FEAT_OPN(CACHE,		cache,		true),
3432	FEAT_OPR(SAMPLE_TIME,	sample_time,	false),
3433	FEAT_OPR(MEM_TOPOLOGY,	mem_topology,	true),
3434	FEAT_OPR(CLOCKID,	clockid,	false),
3435	FEAT_OPN(DIR_FORMAT,	dir_format,	false),
3436#ifdef HAVE_LIBBPF_SUPPORT
3437	FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
3438	FEAT_OPR(BPF_BTF,       bpf_btf,        false),
3439#endif
3440	FEAT_OPR(COMPRESSED,	compressed,	false),
3441	FEAT_OPR(CPU_PMU_CAPS,	cpu_pmu_caps,	false),
3442	FEAT_OPR(CLOCK_DATA,	clock_data,	false),
3443	FEAT_OPN(HYBRID_TOPOLOGY,	hybrid_topology,	true),
3444	FEAT_OPR(PMU_CAPS,	pmu_caps,	false),
3445};
3446
3447struct header_print_data {
3448	FILE *fp;
3449	bool full; /* extended list of headers */
3450};
3451
3452static int perf_file_section__fprintf_info(struct perf_file_section *section,
3453					   struct perf_header *ph,
3454					   int feat, int fd, void *data)
3455{
3456	struct header_print_data *hd = data;
3457	struct feat_fd ff;
3458
3459	if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3460		pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3461				"%d, continuing...\n", section->offset, feat);
3462		return 0;
3463	}
3464	if (feat >= HEADER_LAST_FEATURE) {
3465		pr_warning("unknown feature %d\n", feat);
3466		return 0;
3467	}
3468	if (!feat_ops[feat].print)
3469		return 0;
3470
3471	ff = (struct  feat_fd) {
3472		.fd = fd,
3473		.ph = ph,
3474	};
3475
3476	if (!feat_ops[feat].full_only || hd->full)
3477		feat_ops[feat].print(&ff, hd->fp);
3478	else
3479		fprintf(hd->fp, "# %s info available, use -I to display\n",
3480			feat_ops[feat].name);
3481
3482	return 0;
3483}
3484
3485int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3486{
3487	struct header_print_data hd;
3488	struct perf_header *header = &session->header;
3489	int fd = perf_data__fd(session->data);
3490	struct stat st;
3491	time_t stctime;
3492	int ret, bit;
3493
3494	hd.fp = fp;
3495	hd.full = full;
3496
3497	ret = fstat(fd, &st);
3498	if (ret == -1)
3499		return -1;
3500
3501	stctime = st.st_mtime;
3502	fprintf(fp, "# captured on    : %s", ctime(&stctime));
3503
3504	fprintf(fp, "# header version : %u\n", header->version);
3505	fprintf(fp, "# data offset    : %" PRIu64 "\n", header->data_offset);
3506	fprintf(fp, "# data size      : %" PRIu64 "\n", header->data_size);
3507	fprintf(fp, "# feat offset    : %" PRIu64 "\n", header->feat_offset);
3508
3509	perf_header__process_sections(header, fd, &hd,
3510				      perf_file_section__fprintf_info);
3511
3512	if (session->data->is_pipe)
3513		return 0;
3514
3515	fprintf(fp, "# missing features: ");
3516	for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3517		if (bit)
3518			fprintf(fp, "%s ", feat_ops[bit].name);
3519	}
3520
3521	fprintf(fp, "\n");
3522	return 0;
3523}
3524
3525struct header_fw {
3526	struct feat_writer	fw;
3527	struct feat_fd		*ff;
3528};
3529
3530static int feat_writer_cb(struct feat_writer *fw, void *buf, size_t sz)
3531{
3532	struct header_fw *h = container_of(fw, struct header_fw, fw);
3533
3534	return do_write(h->ff, buf, sz);
3535}
3536
3537static int do_write_feat(struct feat_fd *ff, int type,
3538			 struct perf_file_section **p,
3539			 struct evlist *evlist,
3540			 struct feat_copier *fc)
3541{
3542	int err;
3543	int ret = 0;
3544
3545	if (perf_header__has_feat(ff->ph, type)) {
3546		if (!feat_ops[type].write)
3547			return -1;
3548
3549		if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3550			return -1;
3551
3552		(*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3553
3554		/*
3555		 * Hook to let perf inject copy features sections from the input
3556		 * file.
3557		 */
3558		if (fc && fc->copy) {
3559			struct header_fw h = {
3560				.fw.write = feat_writer_cb,
3561				.ff = ff,
3562			};
3563
3564			/* ->copy() returns 0 if the feature was not copied */
3565			err = fc->copy(fc, type, &h.fw);
3566		} else {
3567			err = 0;
3568		}
3569		if (!err)
3570			err = feat_ops[type].write(ff, evlist);
3571		if (err < 0) {
3572			pr_debug("failed to write feature %s\n", feat_ops[type].name);
3573
3574			/* undo anything written */
3575			lseek(ff->fd, (*p)->offset, SEEK_SET);
3576
3577			return -1;
3578		}
3579		(*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3580		(*p)++;
3581	}
3582	return ret;
3583}
3584
3585static int perf_header__adds_write(struct perf_header *header,
3586				   struct evlist *evlist, int fd,
3587				   struct feat_copier *fc)
3588{
3589	int nr_sections;
3590	struct feat_fd ff;
3591	struct perf_file_section *feat_sec, *p;
3592	int sec_size;
3593	u64 sec_start;
3594	int feat;
3595	int err;
3596
3597	ff = (struct feat_fd){
3598		.fd  = fd,
3599		.ph = header,
3600	};
3601
3602	nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3603	if (!nr_sections)
3604		return 0;
3605
3606	feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3607	if (feat_sec == NULL)
3608		return -ENOMEM;
3609
3610	sec_size = sizeof(*feat_sec) * nr_sections;
3611
3612	sec_start = header->feat_offset;
3613	lseek(fd, sec_start + sec_size, SEEK_SET);
3614
3615	for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3616		if (do_write_feat(&ff, feat, &p, evlist, fc))
3617			perf_header__clear_feat(header, feat);
3618	}
3619
3620	lseek(fd, sec_start, SEEK_SET);
3621	/*
3622	 * may write more than needed due to dropped feature, but
3623	 * this is okay, reader will skip the missing entries
3624	 */
3625	err = do_write(&ff, feat_sec, sec_size);
3626	if (err < 0)
3627		pr_debug("failed to write feature section\n");
3628	free(feat_sec);
3629	return err;
3630}
3631
3632int perf_header__write_pipe(int fd)
3633{
3634	struct perf_pipe_file_header f_header;
3635	struct feat_fd ff;
3636	int err;
3637
3638	ff = (struct feat_fd){ .fd = fd };
3639
3640	f_header = (struct perf_pipe_file_header){
3641		.magic	   = PERF_MAGIC,
3642		.size	   = sizeof(f_header),
3643	};
3644
3645	err = do_write(&ff, &f_header, sizeof(f_header));
3646	if (err < 0) {
3647		pr_debug("failed to write perf pipe header\n");
3648		return err;
3649	}
3650
3651	return 0;
3652}
3653
3654static int perf_session__do_write_header(struct perf_session *session,
3655					 struct evlist *evlist,
3656					 int fd, bool at_exit,
3657					 struct feat_copier *fc)
3658{
3659	struct perf_file_header f_header;
3660	struct perf_file_attr   f_attr;
3661	struct perf_header *header = &session->header;
3662	struct evsel *evsel;
3663	struct feat_fd ff;
3664	u64 attr_offset;
3665	int err;
3666
3667	ff = (struct feat_fd){ .fd = fd};
3668	lseek(fd, sizeof(f_header), SEEK_SET);
3669
3670	evlist__for_each_entry(session->evlist, evsel) {
3671		evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3672		err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3673		if (err < 0) {
3674			pr_debug("failed to write perf header\n");
3675			return err;
3676		}
3677	}
3678
3679	attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3680
3681	evlist__for_each_entry(evlist, evsel) {
3682		if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3683			/*
3684			 * We are likely in "perf inject" and have read
3685			 * from an older file. Update attr size so that
3686			 * reader gets the right offset to the ids.
3687			 */
3688			evsel->core.attr.size = sizeof(evsel->core.attr);
3689		}
3690		f_attr = (struct perf_file_attr){
3691			.attr = evsel->core.attr,
3692			.ids  = {
3693				.offset = evsel->id_offset,
3694				.size   = evsel->core.ids * sizeof(u64),
3695			}
3696		};
3697		err = do_write(&ff, &f_attr, sizeof(f_attr));
3698		if (err < 0) {
3699			pr_debug("failed to write perf header attribute\n");
3700			return err;
3701		}
3702	}
3703
3704	if (!header->data_offset)
3705		header->data_offset = lseek(fd, 0, SEEK_CUR);
3706	header->feat_offset = header->data_offset + header->data_size;
3707
3708	if (at_exit) {
3709		err = perf_header__adds_write(header, evlist, fd, fc);
3710		if (err < 0)
3711			return err;
3712	}
3713
3714	f_header = (struct perf_file_header){
3715		.magic	   = PERF_MAGIC,
3716		.size	   = sizeof(f_header),
3717		.attr_size = sizeof(f_attr),
3718		.attrs = {
3719			.offset = attr_offset,
3720			.size   = evlist->core.nr_entries * sizeof(f_attr),
3721		},
3722		.data = {
3723			.offset = header->data_offset,
3724			.size	= header->data_size,
3725		},
3726		/* event_types is ignored, store zeros */
3727	};
3728
3729	memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3730
3731	lseek(fd, 0, SEEK_SET);
3732	err = do_write(&ff, &f_header, sizeof(f_header));
3733	if (err < 0) {
3734		pr_debug("failed to write perf header\n");
3735		return err;
3736	}
3737	lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3738
3739	return 0;
3740}
3741
3742int perf_session__write_header(struct perf_session *session,
3743			       struct evlist *evlist,
3744			       int fd, bool at_exit)
3745{
3746	return perf_session__do_write_header(session, evlist, fd, at_exit, NULL);
3747}
3748
3749size_t perf_session__data_offset(const struct evlist *evlist)
3750{
3751	struct evsel *evsel;
3752	size_t data_offset;
3753
3754	data_offset = sizeof(struct perf_file_header);
3755	evlist__for_each_entry(evlist, evsel) {
3756		data_offset += evsel->core.ids * sizeof(u64);
3757	}
3758	data_offset += evlist->core.nr_entries * sizeof(struct perf_file_attr);
3759
3760	return data_offset;
3761}
3762
3763int perf_session__inject_header(struct perf_session *session,
3764				struct evlist *evlist,
3765				int fd,
3766				struct feat_copier *fc)
3767{
3768	return perf_session__do_write_header(session, evlist, fd, true, fc);
3769}
3770
3771static int perf_header__getbuffer64(struct perf_header *header,
3772				    int fd, void *buf, size_t size)
3773{
3774	if (readn(fd, buf, size) <= 0)
3775		return -1;
3776
3777	if (header->needs_swap)
3778		mem_bswap_64(buf, size);
3779
3780	return 0;
3781}
3782
3783int perf_header__process_sections(struct perf_header *header, int fd,
3784				  void *data,
3785				  int (*process)(struct perf_file_section *section,
3786						 struct perf_header *ph,
3787						 int feat, int fd, void *data))
3788{
3789	struct perf_file_section *feat_sec, *sec;
3790	int nr_sections;
3791	int sec_size;
3792	int feat;
3793	int err;
3794
3795	nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3796	if (!nr_sections)
3797		return 0;
3798
3799	feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3800	if (!feat_sec)
3801		return -1;
3802
3803	sec_size = sizeof(*feat_sec) * nr_sections;
3804
3805	lseek(fd, header->feat_offset, SEEK_SET);
3806
3807	err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3808	if (err < 0)
3809		goto out_free;
3810
3811	for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3812		err = process(sec++, header, feat, fd, data);
3813		if (err < 0)
3814			goto out_free;
3815	}
3816	err = 0;
3817out_free:
3818	free(feat_sec);
3819	return err;
3820}
3821
3822static const int attr_file_abi_sizes[] = {
3823	[0] = PERF_ATTR_SIZE_VER0,
3824	[1] = PERF_ATTR_SIZE_VER1,
3825	[2] = PERF_ATTR_SIZE_VER2,
3826	[3] = PERF_ATTR_SIZE_VER3,
3827	[4] = PERF_ATTR_SIZE_VER4,
3828	0,
3829};
3830
3831/*
3832 * In the legacy file format, the magic number is not used to encode endianness.
3833 * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3834 * on ABI revisions, we need to try all combinations for all endianness to
3835 * detect the endianness.
3836 */
3837static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3838{
3839	uint64_t ref_size, attr_size;
3840	int i;
3841
3842	for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3843		ref_size = attr_file_abi_sizes[i]
3844			 + sizeof(struct perf_file_section);
3845		if (hdr_sz != ref_size) {
3846			attr_size = bswap_64(hdr_sz);
3847			if (attr_size != ref_size)
3848				continue;
3849
3850			ph->needs_swap = true;
3851		}
3852		pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3853			 i,
3854			 ph->needs_swap);
3855		return 0;
3856	}
3857	/* could not determine endianness */
3858	return -1;
3859}
3860
3861#define PERF_PIPE_HDR_VER0	16
3862
3863static const size_t attr_pipe_abi_sizes[] = {
3864	[0] = PERF_PIPE_HDR_VER0,
3865	0,
3866};
3867
3868/*
3869 * In the legacy pipe format, there is an implicit assumption that endianness
3870 * between host recording the samples, and host parsing the samples is the
3871 * same. This is not always the case given that the pipe output may always be
3872 * redirected into a file and analyzed on a different machine with possibly a
3873 * different endianness and perf_event ABI revisions in the perf tool itself.
3874 */
3875static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3876{
3877	u64 attr_size;
3878	int i;
3879
3880	for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3881		if (hdr_sz != attr_pipe_abi_sizes[i]) {
3882			attr_size = bswap_64(hdr_sz);
3883			if (attr_size != hdr_sz)
3884				continue;
3885
3886			ph->needs_swap = true;
3887		}
3888		pr_debug("Pipe ABI%d perf.data file detected\n", i);
3889		return 0;
3890	}
3891	return -1;
3892}
3893
3894bool is_perf_magic(u64 magic)
3895{
3896	if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3897		|| magic == __perf_magic2
3898		|| magic == __perf_magic2_sw)
3899		return true;
3900
3901	return false;
3902}
3903
3904static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3905			      bool is_pipe, struct perf_header *ph)
3906{
3907	int ret;
3908
3909	/* check for legacy format */
3910	ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3911	if (ret == 0) {
3912		ph->version = PERF_HEADER_VERSION_1;
3913		pr_debug("legacy perf.data format\n");
3914		if (is_pipe)
3915			return try_all_pipe_abis(hdr_sz, ph);
3916
3917		return try_all_file_abis(hdr_sz, ph);
3918	}
3919	/*
3920	 * the new magic number serves two purposes:
3921	 * - unique number to identify actual perf.data files
3922	 * - encode endianness of file
3923	 */
3924	ph->version = PERF_HEADER_VERSION_2;
3925
3926	/* check magic number with one endianness */
3927	if (magic == __perf_magic2)
3928		return 0;
3929
3930	/* check magic number with opposite endianness */
3931	if (magic != __perf_magic2_sw)
3932		return -1;
3933
3934	ph->needs_swap = true;
3935
3936	return 0;
3937}
3938
3939int perf_file_header__read(struct perf_file_header *header,
3940			   struct perf_header *ph, int fd)
3941{
3942	ssize_t ret;
3943
3944	lseek(fd, 0, SEEK_SET);
3945
3946	ret = readn(fd, header, sizeof(*header));
3947	if (ret <= 0)
3948		return -1;
3949
3950	if (check_magic_endian(header->magic,
3951			       header->attr_size, false, ph) < 0) {
3952		pr_debug("magic/endian check failed\n");
3953		return -1;
3954	}
3955
3956	if (ph->needs_swap) {
3957		mem_bswap_64(header, offsetof(struct perf_file_header,
3958			     adds_features));
3959	}
3960
3961	if (header->size != sizeof(*header)) {
3962		/* Support the previous format */
3963		if (header->size == offsetof(typeof(*header), adds_features))
3964			bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3965		else
3966			return -1;
3967	} else if (ph->needs_swap) {
3968		/*
3969		 * feature bitmap is declared as an array of unsigned longs --
3970		 * not good since its size can differ between the host that
3971		 * generated the data file and the host analyzing the file.
3972		 *
3973		 * We need to handle endianness, but we don't know the size of
3974		 * the unsigned long where the file was generated. Take a best
3975		 * guess at determining it: try 64-bit swap first (ie., file
3976		 * created on a 64-bit host), and check if the hostname feature
3977		 * bit is set (this feature bit is forced on as of fbe96f2).
3978		 * If the bit is not, undo the 64-bit swap and try a 32-bit
3979		 * swap. If the hostname bit is still not set (e.g., older data
3980		 * file), punt and fallback to the original behavior --
3981		 * clearing all feature bits and setting buildid.
3982		 */
3983		mem_bswap_64(&header->adds_features,
3984			    BITS_TO_U64(HEADER_FEAT_BITS));
3985
3986		if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3987			/* unswap as u64 */
3988			mem_bswap_64(&header->adds_features,
3989				    BITS_TO_U64(HEADER_FEAT_BITS));
3990
3991			/* unswap as u32 */
3992			mem_bswap_32(&header->adds_features,
3993				    BITS_TO_U32(HEADER_FEAT_BITS));
3994		}
3995
3996		if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3997			bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3998			__set_bit(HEADER_BUILD_ID, header->adds_features);
3999		}
4000	}
4001
4002	memcpy(&ph->adds_features, &header->adds_features,
4003	       sizeof(ph->adds_features));
4004
4005	ph->data_offset  = header->data.offset;
4006	ph->data_size	 = header->data.size;
4007	ph->feat_offset  = header->data.offset + header->data.size;
4008	return 0;
4009}
4010
4011static int perf_file_section__process(struct perf_file_section *section,
4012				      struct perf_header *ph,
4013				      int feat, int fd, void *data)
4014{
4015	struct feat_fd fdd = {
4016		.fd	= fd,
4017		.ph	= ph,
4018		.size	= section->size,
4019		.offset	= section->offset,
4020	};
4021
4022	if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
4023		pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
4024			  "%d, continuing...\n", section->offset, feat);
4025		return 0;
4026	}
4027
4028	if (feat >= HEADER_LAST_FEATURE) {
4029		pr_debug("unknown feature %d, continuing...\n", feat);
4030		return 0;
4031	}
4032
4033	if (!feat_ops[feat].process)
4034		return 0;
4035
4036	return feat_ops[feat].process(&fdd, data);
4037}
4038
4039static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
4040				       struct perf_header *ph,
4041				       struct perf_data* data,
4042				       bool repipe, int repipe_fd)
4043{
4044	struct feat_fd ff = {
4045		.fd = repipe_fd,
4046		.ph = ph,
4047	};
4048	ssize_t ret;
4049
4050	ret = perf_data__read(data, header, sizeof(*header));
4051	if (ret <= 0)
4052		return -1;
4053
4054	if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
4055		pr_debug("endian/magic failed\n");
4056		return -1;
4057	}
4058
4059	if (ph->needs_swap)
4060		header->size = bswap_64(header->size);
4061
4062	if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
4063		return -1;
4064
4065	return 0;
4066}
4067
4068static int perf_header__read_pipe(struct perf_session *session, int repipe_fd)
4069{
4070	struct perf_header *header = &session->header;
4071	struct perf_pipe_file_header f_header;
4072
4073	if (perf_file_header__read_pipe(&f_header, header, session->data,
4074					session->repipe, repipe_fd) < 0) {
4075		pr_debug("incompatible file format\n");
4076		return -EINVAL;
4077	}
4078
4079	return f_header.size == sizeof(f_header) ? 0 : -1;
4080}
4081
4082static int read_attr(int fd, struct perf_header *ph,
4083		     struct perf_file_attr *f_attr)
4084{
4085	struct perf_event_attr *attr = &f_attr->attr;
4086	size_t sz, left;
4087	size_t our_sz = sizeof(f_attr->attr);
4088	ssize_t ret;
4089
4090	memset(f_attr, 0, sizeof(*f_attr));
4091
4092	/* read minimal guaranteed structure */
4093	ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
4094	if (ret <= 0) {
4095		pr_debug("cannot read %d bytes of header attr\n",
4096			 PERF_ATTR_SIZE_VER0);
4097		return -1;
4098	}
4099
4100	/* on file perf_event_attr size */
4101	sz = attr->size;
4102
4103	if (ph->needs_swap)
4104		sz = bswap_32(sz);
4105
4106	if (sz == 0) {
4107		/* assume ABI0 */
4108		sz =  PERF_ATTR_SIZE_VER0;
4109	} else if (sz > our_sz) {
4110		pr_debug("file uses a more recent and unsupported ABI"
4111			 " (%zu bytes extra)\n", sz - our_sz);
4112		return -1;
4113	}
4114	/* what we have not yet read and that we know about */
4115	left = sz - PERF_ATTR_SIZE_VER0;
4116	if (left) {
4117		void *ptr = attr;
4118		ptr += PERF_ATTR_SIZE_VER0;
4119
4120		ret = readn(fd, ptr, left);
4121	}
4122	/* read perf_file_section, ids are read in caller */
4123	ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
4124
4125	return ret <= 0 ? -1 : 0;
4126}
4127
4128#ifdef HAVE_LIBTRACEEVENT
4129static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
4130{
4131	struct tep_event *event;
4132	char bf[128];
4133
4134	/* already prepared */
4135	if (evsel->tp_format)
4136		return 0;
4137
4138	if (pevent == NULL) {
4139		pr_debug("broken or missing trace data\n");
4140		return -1;
4141	}
4142
4143	event = tep_find_event(pevent, evsel->core.attr.config);
4144	if (event == NULL) {
4145		pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
4146		return -1;
4147	}
4148
4149	if (!evsel->name) {
4150		snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
4151		evsel->name = strdup(bf);
4152		if (evsel->name == NULL)
4153			return -1;
4154	}
4155
4156	evsel->tp_format = event;
4157	return 0;
4158}
4159
4160static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
4161{
4162	struct evsel *pos;
4163
4164	evlist__for_each_entry(evlist, pos) {
4165		if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
4166		    evsel__prepare_tracepoint_event(pos, pevent))
4167			return -1;
4168	}
4169
4170	return 0;
4171}
4172#endif
4173
4174int perf_session__read_header(struct perf_session *session, int repipe_fd)
4175{
4176	struct perf_data *data = session->data;
4177	struct perf_header *header = &session->header;
4178	struct perf_file_header	f_header;
4179	struct perf_file_attr	f_attr;
4180	u64			f_id;
4181	int nr_attrs, nr_ids, i, j, err;
4182	int fd = perf_data__fd(data);
4183
4184	session->evlist = evlist__new();
4185	if (session->evlist == NULL)
4186		return -ENOMEM;
4187
4188	session->evlist->env = &header->env;
4189	session->machines.host.env = &header->env;
4190
4191	/*
4192	 * We can read 'pipe' data event from regular file,
4193	 * check for the pipe header regardless of source.
4194	 */
4195	err = perf_header__read_pipe(session, repipe_fd);
4196	if (!err || perf_data__is_pipe(data)) {
4197		data->is_pipe = true;
4198		return err;
4199	}
4200
4201	if (perf_file_header__read(&f_header, header, fd) < 0)
4202		return -EINVAL;
4203
4204	if (header->needs_swap && data->in_place_update) {
4205		pr_err("In-place update not supported when byte-swapping is required\n");
4206		return -EINVAL;
4207	}
4208
4209	/*
4210	 * Sanity check that perf.data was written cleanly; data size is
4211	 * initialized to 0 and updated only if the on_exit function is run.
4212	 * If data size is still 0 then the file contains only partial
4213	 * information.  Just warn user and process it as much as it can.
4214	 */
4215	if (f_header.data.size == 0) {
4216		pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4217			   "Was the 'perf record' command properly terminated?\n",
4218			   data->file.path);
4219	}
4220
4221	if (f_header.attr_size == 0) {
4222		pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4223		       "Was the 'perf record' command properly terminated?\n",
4224		       data->file.path);
4225		return -EINVAL;
4226	}
4227
4228	nr_attrs = f_header.attrs.size / f_header.attr_size;
4229	lseek(fd, f_header.attrs.offset, SEEK_SET);
4230
4231	for (i = 0; i < nr_attrs; i++) {
4232		struct evsel *evsel;
4233		off_t tmp;
4234
4235		if (read_attr(fd, header, &f_attr) < 0)
4236			goto out_errno;
4237
4238		if (header->needs_swap) {
4239			f_attr.ids.size   = bswap_64(f_attr.ids.size);
4240			f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4241			perf_event__attr_swap(&f_attr.attr);
4242		}
4243
4244		tmp = lseek(fd, 0, SEEK_CUR);
4245		evsel = evsel__new(&f_attr.attr);
4246
4247		if (evsel == NULL)
4248			goto out_delete_evlist;
4249
4250		evsel->needs_swap = header->needs_swap;
4251		/*
4252		 * Do it before so that if perf_evsel__alloc_id fails, this
4253		 * entry gets purged too at evlist__delete().
4254		 */
4255		evlist__add(session->evlist, evsel);
4256
4257		nr_ids = f_attr.ids.size / sizeof(u64);
4258		/*
4259		 * We don't have the cpu and thread maps on the header, so
4260		 * for allocating the perf_sample_id table we fake 1 cpu and
4261		 * hattr->ids threads.
4262		 */
4263		if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4264			goto out_delete_evlist;
4265
4266		lseek(fd, f_attr.ids.offset, SEEK_SET);
4267
4268		for (j = 0; j < nr_ids; j++) {
4269			if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4270				goto out_errno;
4271
4272			perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4273		}
4274
4275		lseek(fd, tmp, SEEK_SET);
4276	}
4277
4278#ifdef HAVE_LIBTRACEEVENT
4279	perf_header__process_sections(header, fd, &session->tevent,
4280				      perf_file_section__process);
4281
4282	if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4283		goto out_delete_evlist;
4284#else
4285	perf_header__process_sections(header, fd, NULL, perf_file_section__process);
4286#endif
4287
4288	return 0;
4289out_errno:
4290	return -errno;
4291
4292out_delete_evlist:
4293	evlist__delete(session->evlist);
4294	session->evlist = NULL;
4295	return -ENOMEM;
4296}
4297
4298int perf_event__process_feature(struct perf_session *session,
4299				union perf_event *event)
4300{
4301	struct perf_tool *tool = session->tool;
4302	struct feat_fd ff = { .fd = 0 };
4303	struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4304	int type = fe->header.type;
4305	u64 feat = fe->feat_id;
4306	int ret = 0;
4307
4308	if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4309		pr_warning("invalid record type %d in pipe-mode\n", type);
4310		return 0;
4311	}
4312	if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4313		pr_warning("invalid record type %d in pipe-mode\n", type);
4314		return -1;
4315	}
4316
4317	if (!feat_ops[feat].process)
4318		return 0;
4319
4320	ff.buf  = (void *)fe->data;
4321	ff.size = event->header.size - sizeof(*fe);
4322	ff.ph = &session->header;
4323
4324	if (feat_ops[feat].process(&ff, NULL)) {
4325		ret = -1;
4326		goto out;
4327	}
4328
4329	if (!feat_ops[feat].print || !tool->show_feat_hdr)
4330		goto out;
4331
4332	if (!feat_ops[feat].full_only ||
4333	    tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4334		feat_ops[feat].print(&ff, stdout);
4335	} else {
4336		fprintf(stdout, "# %s info available, use -I to display\n",
4337			feat_ops[feat].name);
4338	}
4339out:
4340	free_event_desc(ff.events);
4341	return ret;
4342}
4343
4344size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4345{
4346	struct perf_record_event_update *ev = &event->event_update;
4347	struct perf_cpu_map *map;
4348	size_t ret;
4349
4350	ret = fprintf(fp, "\n... id:    %" PRI_lu64 "\n", ev->id);
4351
4352	switch (ev->type) {
4353	case PERF_EVENT_UPDATE__SCALE:
4354		ret += fprintf(fp, "... scale: %f\n", ev->scale.scale);
4355		break;
4356	case PERF_EVENT_UPDATE__UNIT:
4357		ret += fprintf(fp, "... unit:  %s\n", ev->unit);
4358		break;
4359	case PERF_EVENT_UPDATE__NAME:
4360		ret += fprintf(fp, "... name:  %s\n", ev->name);
4361		break;
4362	case PERF_EVENT_UPDATE__CPUS:
4363		ret += fprintf(fp, "... ");
4364
4365		map = cpu_map__new_data(&ev->cpus.cpus);
4366		if (map) {
4367			ret += cpu_map__fprintf(map, fp);
4368			perf_cpu_map__put(map);
4369		} else
4370			ret += fprintf(fp, "failed to get cpus\n");
4371		break;
4372	default:
4373		ret += fprintf(fp, "... unknown type\n");
4374		break;
4375	}
4376
4377	return ret;
4378}
4379
4380int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4381			     union perf_event *event,
4382			     struct evlist **pevlist)
4383{
4384	u32 i, n_ids;
4385	u64 *ids;
4386	struct evsel *evsel;
4387	struct evlist *evlist = *pevlist;
4388
4389	if (evlist == NULL) {
4390		*pevlist = evlist = evlist__new();
4391		if (evlist == NULL)
4392			return -ENOMEM;
4393	}
4394
4395	evsel = evsel__new(&event->attr.attr);
4396	if (evsel == NULL)
4397		return -ENOMEM;
4398
4399	evlist__add(evlist, evsel);
4400
4401	n_ids = event->header.size - sizeof(event->header) - event->attr.attr.size;
4402	n_ids = n_ids / sizeof(u64);
4403	/*
4404	 * We don't have the cpu and thread maps on the header, so
4405	 * for allocating the perf_sample_id table we fake 1 cpu and
4406	 * hattr->ids threads.
4407	 */
4408	if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4409		return -ENOMEM;
4410
4411	ids = perf_record_header_attr_id(event);
4412	for (i = 0; i < n_ids; i++) {
4413		perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, ids[i]);
4414	}
4415
4416	return 0;
4417}
4418
4419int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4420				     union perf_event *event,
4421				     struct evlist **pevlist)
4422{
4423	struct perf_record_event_update *ev = &event->event_update;
4424	struct evlist *evlist;
4425	struct evsel *evsel;
4426	struct perf_cpu_map *map;
4427
4428	if (dump_trace)
4429		perf_event__fprintf_event_update(event, stdout);
4430
4431	if (!pevlist || *pevlist == NULL)
4432		return -EINVAL;
4433
4434	evlist = *pevlist;
4435
4436	evsel = evlist__id2evsel(evlist, ev->id);
4437	if (evsel == NULL)
4438		return -EINVAL;
4439
4440	switch (ev->type) {
4441	case PERF_EVENT_UPDATE__UNIT:
4442		free((char *)evsel->unit);
4443		evsel->unit = strdup(ev->unit);
4444		break;
4445	case PERF_EVENT_UPDATE__NAME:
4446		free(evsel->name);
4447		evsel->name = strdup(ev->name);
4448		break;
4449	case PERF_EVENT_UPDATE__SCALE:
4450		evsel->scale = ev->scale.scale;
4451		break;
4452	case PERF_EVENT_UPDATE__CPUS:
4453		map = cpu_map__new_data(&ev->cpus.cpus);
4454		if (map) {
4455			perf_cpu_map__put(evsel->core.own_cpus);
4456			evsel->core.own_cpus = map;
4457		} else
4458			pr_err("failed to get event_update cpus\n");
4459	default:
4460		break;
4461	}
4462
4463	return 0;
4464}
4465
4466#ifdef HAVE_LIBTRACEEVENT
4467int perf_event__process_tracing_data(struct perf_session *session,
4468				     union perf_event *event)
4469{
4470	ssize_t size_read, padding, size = event->tracing_data.size;
4471	int fd = perf_data__fd(session->data);
4472	char buf[BUFSIZ];
4473
4474	/*
4475	 * The pipe fd is already in proper place and in any case
4476	 * we can't move it, and we'd screw the case where we read
4477	 * 'pipe' data from regular file. The trace_report reads
4478	 * data from 'fd' so we need to set it directly behind the
4479	 * event, where the tracing data starts.
4480	 */
4481	if (!perf_data__is_pipe(session->data)) {
4482		off_t offset = lseek(fd, 0, SEEK_CUR);
4483
4484		/* setup for reading amidst mmap */
4485		lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4486		      SEEK_SET);
4487	}
4488
4489	size_read = trace_report(fd, &session->tevent,
4490				 session->repipe);
4491	padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4492
4493	if (readn(fd, buf, padding) < 0) {
4494		pr_err("%s: reading input file", __func__);
4495		return -1;
4496	}
4497	if (session->repipe) {
4498		int retw = write(STDOUT_FILENO, buf, padding);
4499		if (retw <= 0 || retw != padding) {
4500			pr_err("%s: repiping tracing data padding", __func__);
4501			return -1;
4502		}
4503	}
4504
4505	if (size_read + padding != size) {
4506		pr_err("%s: tracing data size mismatch", __func__);
4507		return -1;
4508	}
4509
4510	evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4511
4512	return size_read + padding;
4513}
4514#endif
4515
4516int perf_event__process_build_id(struct perf_session *session,
4517				 union perf_event *event)
4518{
4519	__event_process_build_id(&event->build_id,
4520				 event->build_id.filename,
4521				 session);
4522	return 0;
4523}
4524