lkml.org 
[lkml]   [2021]   [Oct]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v2] perf evsel: Fix missing exclude_{host,guest} setting
Hi Jiri,

On Tue, Oct 26, 2021 at 11:32 AM Jiri Olsa <jolsa@redhat.com> wrote:
>
> On Mon, Oct 25, 2021 at 09:33:51PM -0700, Namhyung Kim wrote:
> > The current logic for the perf missing feature has a bug that it can
> > wrongly clear some modifiers like G or H. Actually some PMUs don't
> > support any filtering or exclusion while others do. But we check it
> > as a global feature.
> >
> > For example, the cycles event can have 'G' modifier to enable it only
> > in the guest mode on x86. When you don't run any VMs it'll return 0.
> >
> > # perf stat -a -e cycles:G sleep 1
> >
> > Performance counter stats for 'system wide':
> >
> > 0 cycles:G
> >
> > 1.000721670 seconds time elapsed
> >
> > But when it's used with other pmu events that don't support G modifier,
> > it'll be reset and return non-zero values.
> >
> > # perf stat -a -e cycles:G,msr/tsc/ sleep 1
> >
> > Performance counter stats for 'system wide':
> >
> > 538,029,960 cycles:G
> > 16,924,010,738 msr/tsc/
> >
> > 1.001815327 seconds time elapsed
> >
> > This is because of the missing feature detection logic being global.
> > Add a hashmap to set pmu-specific exclude_host/guest features.
> >
> > Reported-by: Stephane Eranian <eranian@google.com>
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> > ---
> > v2 changes)
> > * change to enum perf_missing_pmu_features
> > * pass NULL to hashmap__find() to skip checking
> > * add a blank line after declaration
> >
> > tools/perf/util/evsel.c | 47 +++++++++++++++++++++++++++++++++++------
> > tools/perf/util/evsel.h | 7 ++++++
> > 2 files changed, 48 insertions(+), 6 deletions(-)
> >
> > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> > index dbfeceb2546c..f33801b4615d 100644
> > --- a/tools/perf/util/evsel.c
> > +++ b/tools/perf/util/evsel.c
> > @@ -1434,6 +1434,10 @@ void evsel__delete(struct evsel *evsel)
> > {
> > evsel__exit(evsel);
> > free(evsel);
> > +
> > + /* just free it for the first evsel */
> > + hashmap__free(perf_missing_features.pmu);
> > + perf_missing_features.pmu = NULL;
>
> should this just remove evsel from perf_missing_features.pmu?

Nop, it'll release all the hash nodes (pmu). Each evsel can come
from a same PMU.

>
> I'm not sure when we should actually release this.. what's your rationale
> for releasing it in here?

Yeah, that's a problem when we should release it. Maybe we can
add a function and call it in the exit path but I didn't want to change
all the callsites just for this.

As evsel__delete() is called in the exit path, I thought it's a
place where we can free the hashmap. But I'm not very happy
with that either.. Do you have a better idea?

If another evsel is open after this function, I think it should
be fine since it will create a new hash map and check the
missing feature again.


>
> > }
> >
> > void evsel__compute_deltas(struct evsel *evsel, int cpu, int thread,
> > @@ -1791,6 +1795,23 @@ static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
> > return 0;
> > }
> >
> > +#define PMU_HASH_BITS 4
> > +
> > +static size_t pmu_hash(const void *key, void *ctx __maybe_unused)
> > +{
> > + const struct evsel *evsel = key;
> > +
> > + return hash_bits(evsel->core.attr.type, PMU_HASH_BITS);
> > +}
> > +
> > +static bool pmu_equal(const void *key1, const void *key2, void *ctx __maybe_unused)
> > +{
> > + const struct evsel *a = key1;
> > + const struct evsel *b = key2;
> > +
> > + return a->core.attr.type == b->core.attr.type;
> > +}
> > +
> > static void evsel__disable_missing_features(struct evsel *evsel)
> > {
> > if (perf_missing_features.weight_struct) {
> > @@ -1807,8 +1828,13 @@ static void evsel__disable_missing_features(struct evsel *evsel)
> > evsel->open_flags &= ~(unsigned long)PERF_FLAG_FD_CLOEXEC;
> > if (perf_missing_features.mmap2)
> > evsel->core.attr.mmap2 = 0;
> > - if (perf_missing_features.exclude_guest)
> > - evsel->core.attr.exclude_guest = evsel->core.attr.exclude_host = 0;
> > + if (perf_missing_features.exclude_guest) {
> > + /* we only have EXCLUDE_GUEST bit, let's skip checking */
> > + if (hashmap__find(perf_missing_features.pmu, evsel, NULL)) {
> > + evsel->core.attr.exclude_guest = 0;
> > + evsel->core.attr.exclude_host = 0;
> > + }
> > + }
> > if (perf_missing_features.lbr_flags)
> > evsel->core.attr.branch_sample_type &= ~(PERF_SAMPLE_BRANCH_NO_FLAGS |
> > PERF_SAMPLE_BRANCH_NO_CYCLES);
> > @@ -1840,6 +1866,9 @@ int evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
> >
> > bool evsel__detect_missing_features(struct evsel *evsel)
> > {
> > + if (perf_missing_features.pmu == NULL)
> > + perf_missing_features.pmu = hashmap__new(pmu_hash, pmu_equal, NULL);
>
> missing allocation fail check

Ok, will change.

Thanks for your review,
Namhyung


>
> > +
> > /*
> > * Must probe features in the order they were added to the
> > * perf_event_attr interface.
> > @@ -1900,10 +1929,16 @@ bool evsel__detect_missing_features(struct evsel *evsel)
> > perf_missing_features.mmap2 = true;
> > pr_debug2_peo("switching off mmap2\n");
> > return true;
> > - } else if (!perf_missing_features.exclude_guest &&
> > - (evsel->core.attr.exclude_guest || evsel->core.attr.exclude_host)) {
> > - perf_missing_features.exclude_guest = true;
> > - pr_debug2_peo("switching off exclude_guest, exclude_host\n");
> > + } else if ((evsel->core.attr.exclude_guest || evsel->core.attr.exclude_host) &&
> > + !hashmap__find(perf_missing_features.pmu, evsel, NULL)) {
> > + uintptr_t pmu_features = PERF_MISSING_PMU_EXCLUDE_GUEST;
> > +
> > + hashmap__add(perf_missing_features.pmu, evsel, (void *)pmu_features);
> > +
> > + if (!perf_missing_features.exclude_guest) {
> > + perf_missing_features.exclude_guest = true;
> > + pr_debug2_peo("switching off exclude_guest, exclude_host\n");
> > + }
> > return true;
> > } else if (!perf_missing_features.sample_id_all) {
> > perf_missing_features.sample_id_all = true;
> > diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
> > index 1f7edfa8568a..11b5ece19f0e 100644
> > --- a/tools/perf/util/evsel.h
> > +++ b/tools/perf/util/evsel.h
> > @@ -172,6 +172,13 @@ struct perf_missing_features {
> > bool data_page_size;
> > bool code_page_size;
> > bool weight_struct;
> > +
> > + /* contains enum perf_missing_pmu_features below */
> > + struct hashmap *pmu;
> > +};
> > +
> > +enum perf_missing_pmu_features {
> > + PERF_MISSING_PMU_EXCLUDE_GUEST = 1UL << 0,
> > };
> >
> > extern struct perf_missing_features perf_missing_features;
> > --
> > 2.33.0.1079.g6e70778dc9-goog
> >
>

\
 
 \ /
  Last update: 2021-10-26 21:31    [W:0.042 / U:0.104 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site