Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'perf-tools-fixes-for-v6.0-2022-09-08' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tools fixes from Arnaldo Carvalho de Melo:
- Fix per-thread mmaps for multi-threaded targets, noticed with
'perf top --pid' with multithreaded targets
- Fix synthesis failure warnings in 'perf record'
- Fix L2 Topdown metrics disappearance for raw events in 'perf stat'
- Fix out of bound access in some CPU masks
- Fix segfault if there is no CPU PMU table and a metric is sought,
noticed when building with NO_JEVENTS=1
- Skip dummy event attr check in 'perf script' fixing nonsensical
warning about UREGS attribute not set, as 'dummy' events have no
samples
- Fix 'iregs' field handling with dummy events on hybrid systems in
'perf script'
- Prevent potential memory leak in c2c_he_zalloc() in 'perf c2c'
- Don't install data files with x permissions
- Fix types for print format in dlfilter-show-cycles
- Switch deprecated openssl MD5_* functions to new EVP API in 'genelf'
- Remove redundant word 'contention' in 'perf lock' help message
* tag 'perf-tools-fixes-for-v6.0-2022-09-08' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
perf record: Fix synthesis failure warnings
perf tools: Don't install data files with x permissions
perf script: Fix Cannot print 'iregs' field for hybrid systems
perf lock: Remove redundant word 'contention' in help message
perf dlfilter dlfilter-show-cycles: Fix types for print format
libperf evlist: Fix per-thread mmaps for multi-threaded targets
perf c2c: Prevent potential memory leak in c2c_he_zalloc()
perf genelf: Switch deprecated openssl MD5_* functions to new EVP API
tools/perf: Fix out of bound access to cpu mask array
perf affinity: Fix out of bound access to "sched_cpus" mask
perf stat: Fix L2 Topdown metrics disappear for raw events
perf script: Skip dummy event attr check
perf metric: Return early if no CPU PMU table exists
+129
-39
+50
tools/lib/perf/evlist.c
+50
tools/lib/perf/evlist.c
···
486
486
if (ops->idx)
487
487
ops->idx(evlist, evsel, mp, idx);
488
488
489
+
pr_debug("idx %d: mmapping fd %d\n", idx, *output);
489
490
if (ops->mmap(map, mp, *output, evlist_cpu) < 0)
490
491
return -1;
491
492
···
495
494
if (!idx)
496
495
perf_evlist__set_mmap_first(evlist, map, overwrite);
497
496
} else {
497
+
pr_debug("idx %d: set output fd %d -> %d\n", idx, fd, *output);
498
498
if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, *output) != 0)
499
499
return -1;
500
500
···
522
520
}
523
521
524
522
static int
523
+
mmap_per_thread(struct perf_evlist *evlist, struct perf_evlist_mmap_ops *ops,
524
+
struct perf_mmap_param *mp)
525
+
{
526
+
int nr_threads = perf_thread_map__nr(evlist->threads);
527
+
int nr_cpus = perf_cpu_map__nr(evlist->all_cpus);
528
+
int cpu, thread, idx = 0;
529
+
int nr_mmaps = 0;
530
+
531
+
pr_debug("%s: nr cpu values (may include -1) %d nr threads %d\n",
532
+
__func__, nr_cpus, nr_threads);
533
+
534
+
/* per-thread mmaps */
535
+
for (thread = 0; thread < nr_threads; thread++, idx++) {
536
+
int output = -1;
537
+
int output_overwrite = -1;
538
+
539
+
if (mmap_per_evsel(evlist, ops, idx, mp, 0, thread, &output,
540
+
&output_overwrite, &nr_mmaps))
541
+
goto out_unmap;
542
+
}
543
+
544
+
/* system-wide mmaps i.e. per-cpu */
545
+
for (cpu = 1; cpu < nr_cpus; cpu++, idx++) {
546
+
int output = -1;
547
+
int output_overwrite = -1;
548
+
549
+
if (mmap_per_evsel(evlist, ops, idx, mp, cpu, 0, &output,
550
+
&output_overwrite, &nr_mmaps))
551
+
goto out_unmap;
552
+
}
553
+
554
+
if (nr_mmaps != evlist->nr_mmaps)
555
+
pr_err("Miscounted nr_mmaps %d vs %d\n", nr_mmaps, evlist->nr_mmaps);
556
+
557
+
return 0;
558
+
559
+
out_unmap:
560
+
perf_evlist__munmap(evlist);
561
+
return -1;
562
+
}
563
+
564
+
static int
525
565
mmap_per_cpu(struct perf_evlist *evlist, struct perf_evlist_mmap_ops *ops,
526
566
struct perf_mmap_param *mp)
527
567
{
···
571
527
int nr_cpus = perf_cpu_map__nr(evlist->all_cpus);
572
528
int nr_mmaps = 0;
573
529
int cpu, thread;
530
+
531
+
pr_debug("%s: nr cpu values %d nr threads %d\n", __func__, nr_cpus, nr_threads);
574
532
575
533
for (cpu = 0; cpu < nr_cpus; cpu++) {
576
534
int output = -1;
···
615
569
struct perf_evlist_mmap_ops *ops,
616
570
struct perf_mmap_param *mp)
617
571
{
572
+
const struct perf_cpu_map *cpus = evlist->all_cpus;
618
573
struct perf_evsel *evsel;
619
574
620
575
if (!ops || !ops->get || !ops->mmap)
···
634
587
635
588
if (evlist->pollfd.entries == NULL && perf_evlist__alloc_pollfd(evlist) < 0)
636
589
return -ENOMEM;
590
+
591
+
if (perf_cpu_map__empty(cpus))
592
+
return mmap_per_thread(evlist, ops, mp);
637
593
638
594
return mmap_per_cpu(evlist, ops, mp);
639
595
}
+12
-12
tools/perf/Makefile.perf
+12
-12
tools/perf/Makefile.perf
···
954
954
$(call QUIET_INSTALL, bpf-headers) \
955
955
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perf_include_instdir_SQ)/bpf'; \
956
956
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perf_include_instdir_SQ)/bpf/linux'; \
957
-
$(INSTALL) include/bpf/*.h -t '$(DESTDIR_SQ)$(perf_include_instdir_SQ)/bpf'; \
958
-
$(INSTALL) include/bpf/linux/*.h -t '$(DESTDIR_SQ)$(perf_include_instdir_SQ)/bpf/linux'
957
+
$(INSTALL) include/bpf/*.h -m 644 -t '$(DESTDIR_SQ)$(perf_include_instdir_SQ)/bpf'; \
958
+
$(INSTALL) include/bpf/linux/*.h -m 644 -t '$(DESTDIR_SQ)$(perf_include_instdir_SQ)/bpf/linux'
959
959
$(call QUIET_INSTALL, bpf-examples) \
960
960
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perf_examples_instdir_SQ)/bpf'; \
961
-
$(INSTALL) examples/bpf/*.c -t '$(DESTDIR_SQ)$(perf_examples_instdir_SQ)/bpf'
961
+
$(INSTALL) examples/bpf/*.c -m 644 -t '$(DESTDIR_SQ)$(perf_examples_instdir_SQ)/bpf'
962
962
endif
963
963
$(call QUIET_INSTALL, perf-archive) \
964
964
$(INSTALL) $(OUTPUT)perf-archive -t '$(DESTDIR_SQ)$(perfexec_instdir_SQ)'
···
967
967
ifndef NO_LIBAUDIT
968
968
$(call QUIET_INSTALL, strace/groups) \
969
969
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(STRACE_GROUPS_INSTDIR_SQ)'; \
970
-
$(INSTALL) trace/strace/groups/* -t '$(DESTDIR_SQ)$(STRACE_GROUPS_INSTDIR_SQ)'
970
+
$(INSTALL) trace/strace/groups/* -m 644 -t '$(DESTDIR_SQ)$(STRACE_GROUPS_INSTDIR_SQ)'
971
971
endif
972
972
ifndef NO_LIBPERL
973
973
$(call QUIET_INSTALL, perl-scripts) \
974
974
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl/Perf-Trace-Util/lib/Perf/Trace'; \
975
-
$(INSTALL) scripts/perl/Perf-Trace-Util/lib/Perf/Trace/* -t '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl/Perf-Trace-Util/lib/Perf/Trace'; \
976
-
$(INSTALL) scripts/perl/*.pl -t '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl'; \
975
+
$(INSTALL) scripts/perl/Perf-Trace-Util/lib/Perf/Trace/* -m 644 -t '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl/Perf-Trace-Util/lib/Perf/Trace'; \
976
+
$(INSTALL) scripts/perl/*.pl -m 644 -t '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl'; \
977
977
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl/bin'; \
978
978
$(INSTALL) scripts/perl/bin/* -t '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/scripts/perl/bin'
979
979
endif
···
990
990
$(INSTALL) $(DLFILTERS) '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/dlfilters';
991
991
$(call QUIET_INSTALL, perf_completion-script) \
992
992
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(sysconfdir_SQ)/bash_completion.d'; \
993
-
$(INSTALL) perf-completion.sh '$(DESTDIR_SQ)$(sysconfdir_SQ)/bash_completion.d/perf'
993
+
$(INSTALL) perf-completion.sh -m 644 '$(DESTDIR_SQ)$(sysconfdir_SQ)/bash_completion.d/perf'
994
994
$(call QUIET_INSTALL, perf-tip) \
995
995
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(tip_instdir_SQ)'; \
996
-
$(INSTALL) Documentation/tips.txt -t '$(DESTDIR_SQ)$(tip_instdir_SQ)'
996
+
$(INSTALL) Documentation/tips.txt -m 644 -t '$(DESTDIR_SQ)$(tip_instdir_SQ)'
997
997
998
998
install-tests: all install-gtk
999
999
$(call QUIET_INSTALL, tests) \
1000
1000
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests'; \
1001
-
$(INSTALL) tests/attr.py '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests'; \
1001
+
$(INSTALL) tests/attr.py -m 644 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests'; \
1002
1002
$(INSTALL) tests/pe-file.exe* '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests'; \
1003
1003
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/attr'; \
1004
-
$(INSTALL) tests/attr/* '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/attr'; \
1004
+
$(INSTALL) tests/attr/* -m 644 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/attr'; \
1005
1005
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \
1006
1006
$(INSTALL) tests/shell/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell'; \
1007
1007
$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \
1008
-
$(INSTALL) tests/shell/lib/*.sh '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \
1009
-
$(INSTALL) tests/shell/lib/*.py '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
1008
+
$(INSTALL) tests/shell/lib/*.sh -m 644 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'; \
1009
+
$(INSTALL) tests/shell/lib/*.py -m 644 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/shell/lib'
1010
1010
1011
1011
install-bin: install-tools install-tests install-traceevent-plugins
1012
1012
+9
-3
tools/perf/builtin-c2c.c
+9
-3
tools/perf/builtin-c2c.c
···
146
146
147
147
c2c_he->cpuset = bitmap_zalloc(c2c.cpus_cnt);
148
148
if (!c2c_he->cpuset)
149
-
return NULL;
149
+
goto out_free;
150
150
151
151
c2c_he->nodeset = bitmap_zalloc(c2c.nodes_cnt);
152
152
if (!c2c_he->nodeset)
153
-
return NULL;
153
+
goto out_free;
154
154
155
155
c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
156
156
if (!c2c_he->node_stats)
157
-
return NULL;
157
+
goto out_free;
158
158
159
159
init_stats(&c2c_he->cstats.lcl_hitm);
160
160
init_stats(&c2c_he->cstats.rmt_hitm);
···
163
163
init_stats(&c2c_he->cstats.load);
164
164
165
165
return &c2c_he->he;
166
+
167
+
out_free:
168
+
free(c2c_he->nodeset);
169
+
free(c2c_he->cpuset);
170
+
free(c2c_he);
171
+
return NULL;
166
172
}
167
173
168
174
static void c2c_he_free(void *he)
+1
-2
tools/perf/builtin-lock.c
+1
-2
tools/perf/builtin-lock.c
+26
-8
tools/perf/builtin-record.c
+26
-8
tools/perf/builtin-record.c
···
1906
1906
1907
1907
err = perf_event__synthesize_bpf_events(session, process_synthesized_event,
1908
1908
machine, opts);
1909
-
if (err < 0)
1909
+
if (err < 0) {
1910
1910
pr_warning("Couldn't synthesize bpf events.\n");
1911
+
err = 0;
1912
+
}
1911
1913
1912
1914
if (rec->opts.synth & PERF_SYNTH_CGROUP) {
1913
1915
err = perf_event__synthesize_cgroups(tool, process_synthesized_event,
1914
1916
machine);
1915
-
if (err < 0)
1917
+
if (err < 0) {
1916
1918
pr_warning("Couldn't synthesize cgroup events.\n");
1919
+
err = 0;
1920
+
}
1917
1921
}
1918
1922
1919
1923
if (rec->opts.nr_threads_synthesize > 1) {
···
3362
3358
3363
3359
struct option *record_options = __record_options;
3364
3360
3365
-
static void record__mmap_cpu_mask_init(struct mmap_cpu_mask *mask, struct perf_cpu_map *cpus)
3361
+
static int record__mmap_cpu_mask_init(struct mmap_cpu_mask *mask, struct perf_cpu_map *cpus)
3366
3362
{
3367
3363
struct perf_cpu cpu;
3368
3364
int idx;
3369
3365
3370
3366
if (cpu_map__is_dummy(cpus))
3371
-
return;
3367
+
return 0;
3372
3368
3373
-
perf_cpu_map__for_each_cpu(cpu, idx, cpus)
3369
+
perf_cpu_map__for_each_cpu(cpu, idx, cpus) {
3370
+
/* Return ENODEV is input cpu is greater than max cpu */
3371
+
if ((unsigned long)cpu.cpu > mask->nbits)
3372
+
return -ENODEV;
3374
3373
set_bit(cpu.cpu, mask->bits);
3374
+
}
3375
+
3376
+
return 0;
3375
3377
}
3376
3378
3377
3379
static int record__mmap_cpu_mask_init_spec(struct mmap_cpu_mask *mask, const char *mask_spec)
···
3389
3379
return -ENOMEM;
3390
3380
3391
3381
bitmap_zero(mask->bits, mask->nbits);
3392
-
record__mmap_cpu_mask_init(mask, cpus);
3382
+
if (record__mmap_cpu_mask_init(mask, cpus))
3383
+
return -ENODEV;
3384
+
3393
3385
perf_cpu_map__put(cpus);
3394
3386
3395
3387
return 0;
···
3473
3461
pr_err("Failed to allocate CPUs mask\n");
3474
3462
return ret;
3475
3463
}
3476
-
record__mmap_cpu_mask_init(&cpus_mask, cpus);
3464
+
3465
+
ret = record__mmap_cpu_mask_init(&cpus_mask, cpus);
3466
+
if (ret) {
3467
+
pr_err("Failed to init cpu mask\n");
3468
+
goto out_free_cpu_mask;
3469
+
}
3477
3470
3478
3471
ret = record__thread_mask_alloc(&full_mask, cpu__max_cpu().cpu);
3479
3472
if (ret) {
···
3719
3702
if (ret)
3720
3703
return ret;
3721
3704
3722
-
record__mmap_cpu_mask_init(&rec->thread_masks->maps, cpus);
3705
+
if (record__mmap_cpu_mask_init(&rec->thread_masks->maps, cpus))
3706
+
return -ENODEV;
3723
3707
3724
3708
rec->nr_threads = 1;
3725
3709
+5
tools/perf/builtin-script.c
+5
tools/perf/builtin-script.c
···
445
445
struct perf_event_attr *attr = &evsel->core.attr;
446
446
bool allow_user_set;
447
447
448
+
if (evsel__is_dummy_event(evsel))
449
+
return 0;
450
+
448
451
if (perf_header__has_feat(&session->header, HEADER_STAT))
449
452
return 0;
450
453
···
569
566
struct evsel *evsel;
570
567
571
568
evlist__for_each_entry(evlist, evsel) {
569
+
if (evsel__is_dummy_event(evsel))
570
+
continue;
572
571
if (output_type(evsel->core.attr.type) == (int)type)
573
572
return evsel;
574
573
}
+3
-2
tools/perf/builtin-stat.c
+3
-2
tools/perf/builtin-stat.c
···
1932
1932
free(str);
1933
1933
}
1934
1934
1935
+
if (!stat_config.topdown_level)
1936
+
stat_config.topdown_level = TOPDOWN_MAX_LEVEL;
1937
+
1935
1938
if (!evsel_list->core.nr_entries) {
1936
1939
if (target__has_cpu(&target))
1937
1940
default_attrs0[0].config = PERF_COUNT_SW_CPU_CLOCK;
···
1951
1948
}
1952
1949
if (evlist__add_default_attrs(evsel_list, default_attrs1) < 0)
1953
1950
return -1;
1954
-
1955
-
stat_config.topdown_level = TOPDOWN_MAX_LEVEL;
1956
1951
/* Platform specific attrs */
1957
1952
if (evlist__add_default_attrs(evsel_list, default_null_attrs) < 0)
1958
1953
return -1;
+2
-2
tools/perf/dlfilters/dlfilter-show-cycles.c
+2
-2
tools/perf/dlfilters/dlfilter-show-cycles.c
···
98
98
static void print_vals(__u64 cycles, __u64 delta)
99
99
{
100
100
if (delta)
101
-
printf("%10llu %10llu ", cycles, delta);
101
+
printf("%10llu %10llu ", (unsigned long long)cycles, (unsigned long long)delta);
102
102
else
103
-
printf("%10llu %10s ", cycles, "");
103
+
printf("%10llu %10s ", (unsigned long long)cycles, "");
104
104
}
105
105
106
106
int filter_event(void *data, const struct perf_dlfilter_sample *sample, void *ctx)
+7
-1
tools/perf/util/affinity.c
+7
-1
tools/perf/util/affinity.c
···
49
49
{
50
50
int cpu_set_size = get_cpu_set_size();
51
51
52
-
if (cpu == -1)
52
+
/*
53
+
* Return:
54
+
* - if cpu is -1
55
+
* - restrict out of bound access to sched_cpus
56
+
*/
57
+
if (cpu == -1 || ((cpu >= (cpu_set_size * 8))))
53
58
return;
59
+
54
60
a->changed = true;
55
61
set_bit(cpu, a->sched_cpus);
56
62
/*
+11
-9
tools/perf/util/genelf.c
+11
-9
tools/perf/util/genelf.c
···
30
30
31
31
#define BUILD_ID_URANDOM /* different uuid for each run */
32
32
33
-
// FIXME, remove this and fix the deprecation warnings before its removed and
34
-
// We'll break for good here...
35
-
#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
36
-
37
33
#ifdef HAVE_LIBCRYPTO_SUPPORT
38
34
39
35
#define BUILD_ID_MD5
···
41
45
#endif
42
46
43
47
#ifdef BUILD_ID_MD5
48
+
#include <openssl/evp.h>
44
49
#include <openssl/md5.h>
45
50
#endif
46
51
#endif
···
139
142
static void
140
143
gen_build_id(struct buildid_note *note, unsigned long load_addr, const void *code, size_t csize)
141
144
{
142
-
MD5_CTX context;
145
+
EVP_MD_CTX *mdctx;
143
146
144
147
if (sizeof(note->build_id) < 16)
145
148
errx(1, "build_id too small for MD5");
146
149
147
-
MD5_Init(&context);
148
-
MD5_Update(&context, &load_addr, sizeof(load_addr));
149
-
MD5_Update(&context, code, csize);
150
-
MD5_Final((unsigned char *)note->build_id, &context);
150
+
mdctx = EVP_MD_CTX_new();
151
+
if (!mdctx)
152
+
errx(2, "failed to create EVP_MD_CTX");
153
+
154
+
EVP_DigestInit_ex(mdctx, EVP_md5(), NULL);
155
+
EVP_DigestUpdate(mdctx, &load_addr, sizeof(load_addr));
156
+
EVP_DigestUpdate(mdctx, code, csize);
157
+
EVP_DigestFinal_ex(mdctx, (unsigned char *)note->build_id, NULL);
158
+
EVP_MD_CTX_free(mdctx);
151
159
}
152
160
#endif
153
161
+3
tools/perf/util/metricgroup.c
+3
tools/perf/util/metricgroup.c
···
1655
1655
struct evlist *perf_evlist = *(struct evlist **)opt->value;
1656
1656
const struct pmu_events_table *table = pmu_events_table__find();
1657
1657
1658
+
if (!table)
1659
+
return -EINVAL;
1660
+
1658
1661
return parse_groups(perf_evlist, str, metric_no_group,
1659
1662
metric_no_merge, NULL, metric_events, table);
1660
1663
}