Skip to contentSkip to conversation

perf record: Cache build-ID of hit DSOs only

PATCH Applied Namhyung Kim 2 participants 5 messages Jul 31, 2025 – Aug 1, 2025 1 message not in the archive perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

PATCH 1/1 perf record: Cache build-ID of hit DSOs only

It post-processes samples to find which DSO has samples. Based on that info, it can save used DSOs in the build-ID cache directory. But for some reason, it saves all DSOs without checking the hit mark. Skipping unused DSOs can give some speedup especially with --buildid-mmap being default.

On my idle machine, time perf record -a sleep 1 goes down from 3 sec to 1.5 sec with this change.

---
 tools/perf/util/build-id.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/build-id.c +1 -1
index e2b295fe4d2fe552..a7018a3b0437a699 100644
@@ -872,7 +872,7 @@ static int dso__cache_build_id(struct dso *dso, struct machine *machine,
 	char *allocated_name = NULL;
 	int ret = 0;
 
-	if (!dso__has_build_id(dso))
+	if (!dso__has_build_id(dso) || !dso__hit(dso))
 		return 0;
 
 	if (dso__is_kcore(dso)) {
Signature · 1 line
2.50.1
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Namhyung Kim

It post-processes samples to find which DSO has samples. Based on that info, it can save used DSOs in the build-ID cache directory. But for some reason, it saves all DSOs without checking the hit mark. Skipping unused DSOs can give some speedup especially with --buildid-mmap being default.

On my idle machine, time perf record -a sleep 1 goes down from 3 sec to 1.5 sec with this change.

Good stuff, and this is in line with the original intent, don't cache uninteresting things.

But now I have do some digging, as this should've been the case since the start, why would we go to the trouble of traversing perf.data, processing all samples, yadda, yadda to then not look at it when caching files?

The whole process of reading the build ids at the end is done here:

bool dsos__read_build_ids(struct dsos *dsos, bool with_hits)

{
        struct dsos__read_build_ids_cb_args args = {
                .with_hits = with_hits,
                .have_build_id = false,
        };

        dsos__for_each_dso(dsos, dsos__read_build_ids_cb, &args);
        return args.have_build_id;
}

And that dsos__read_build_ids_cb thing specifically looks at:

static int dsos__read_build_ids_cb(struct dso *dso, void *data)

{
        struct dsos__read_build_ids_cb_args *args = data;
        struct nscookie nsc;
        struct build_id bid = { .size = 0, };
                                
        if (args->with_hits && !dso__hit(dso) && !dso__is_vdso(dso))
                return 0;

<SNIP>

So it will not try to read the build id if that DSO has no samples.

But, that was written before PERF_RECORD_MMAP* came with a build-id, so it _will_ have a build-id and thus checking if it has hits is needed.

In the past DSOs without hits wouldn't have a build-id because dsos__read_build_ids_cb() would not read that build-id from the ELF file.

Ok, now that makes sense:

This could have a Fixes attached to it, one that doesn't fixes something that is not working, but speeds up processing by overcoming an oversight when adding build-ids to MMAP records, so I think a:

Fixes e29386c8f7d71fa5 ("perf record: Add --buildid-mmap option to enable PERF_RECORD_MMAP2's build id")

Is worth.

- Arnaldo

Quoted 20 lines
---
 tools/perf/util/build-id.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/build-id.c +1 -1
index e2b295fe4d2fe552..a7018a3b0437a699 100644
@@ -872,7 +872,7 @@ static int dso__cache_build_id(struct dso *dso, struct machine *machine,
 	char *allocated_name = NULL;
 	int ret = 0;
 
-	if (!dso__has_build_id(dso))
+	if (!dso__has_build_id(dso) || !dso__hit(dso))
 		return 0;
 
 	if (dso__is_kcore(dso)) {
Signature · 1 line
2.50.1
author maintainer
#2 lore raw

Hi Arnaldo,

Quoted 43 lines · Arnaldo Carvalho de Melo
Quoted 8 lines · Namhyung Kim

It post-processes samples to find which DSO has samples. Based on that info, it can save used DSOs in the build-ID cache directory. But for some reason, it saves all DSOs without checking the hit mark. Skipping unused DSOs can give some speedup especially with --buildid-mmap being default.

On my idle machine, time perf record -a sleep 1 goes down from 3 sec to 1.5 sec with this change.

Good stuff, and this is in line with the original intent, don't cache uninteresting things.

But now I have do some digging, as this should've been the case since the start, why would we go to the trouble of traversing perf.data, processing all samples, yadda, yadda to then not look at it when caching files?

The whole process of reading the build ids at the end is done here:

bool dsos__read_build_ids(struct dsos *dsos, bool with_hits)

{
        struct dsos__read_build_ids_cb_args args = {
                .with_hits = with_hits,
                .have_build_id = false,
        };

        dsos__for_each_dso(dsos, dsos__read_build_ids_cb, &args);
        return args.have_build_id;
}

And that dsos__read_build_ids_cb thing specifically looks at:

static int dsos__read_build_ids_cb(struct dso *dso, void *data)

{
        struct dsos__read_build_ids_cb_args *args = data;
        struct nscookie nsc;
        struct build_id bid = { .size = 0, };
                                
        if (args->with_hits && !dso__hit(dso) && !dso__is_vdso(dso))
                return 0;

<SNIP>

Quoted 8 lines

So it will not try to read the build id if that DSO has no samples.

But, that was written before PERF_RECORD_MMAP* came with a build-id, so it _will_ have a build-id and thus checking if it has hits is needed.

In the past DSOs without hits wouldn't have a build-id because dsos__read_build_ids_cb() would not read that build-id from the ELF file.

Yep, I think that's the reason.

Quoted 4 lines

Ok, now that makes sense:

Thanks!

Quoted 8 lines

This could have a Fixes attached to it, one that doesn't fixes something that is not working, but speeds up processing by overcoming an oversight when adding build-ids to MMAP records, so I think a:

Fixes e29386c8f7d71fa5 ("perf record: Add --buildid-mmap option to enable PERF_RECORD_MMAP2's build id")

Is worth.

Sure, will add.

Thanks,
Namhyung

A message that is not in the archive (sent off-list or not yet received): CAP-5=fVeCriJk5OSEdr0poREL3QNZOO4XDAbbuBPPtczv54d8Q@mail.gmail.com

author maintainer
#4 lore raw

Hi Ian,

Quoted 74 lines · Ian Rogers
Quoted 70 lines · Namhyung Kim

Hi Arnaldo,

Quoted 43 lines · Arnaldo Carvalho de Melo
Quoted 8 lines · Namhyung Kim

It post-processes samples to find which DSO has samples. Based on that info, it can save used DSOs in the build-ID cache directory. But for some reason, it saves all DSOs without checking the hit mark. Skipping unused DSOs can give some speedup especially with --buildid-mmap being default.

On my idle machine, time perf record -a sleep 1 goes down from 3 sec to 1.5 sec with this change.

Good stuff, and this is in line with the original intent, don't cache uninteresting things.

But now I have do some digging, as this should've been the case since the start, why would we go to the trouble of traversing perf.data, processing all samples, yadda, yadda to then not look at it when caching files?

The whole process of reading the build ids at the end is done here:

bool dsos__read_build_ids(struct dsos *dsos, bool with_hits)

{
        struct dsos__read_build_ids_cb_args args = {
                .with_hits = with_hits,
                .have_build_id = false,
        };

        dsos__for_each_dso(dsos, dsos__read_build_ids_cb, &args);
        return args.have_build_id;
}

And that dsos__read_build_ids_cb thing specifically looks at:

static int dsos__read_build_ids_cb(struct dso *dso, void *data)

{
        struct dsos__read_build_ids_cb_args *args = data;
        struct nscookie nsc;
        struct build_id bid = { .size = 0, };

        if (args->with_hits && !dso__hit(dso) && !dso__is_vdso(dso))
                return 0;

<SNIP>

Quoted 8 lines

So it will not try to read the build id if that DSO has no samples.

But, that was written before PERF_RECORD_MMAP* came with a build-id, so it _will_ have a build-id and thus checking if it has hits is needed.

In the past DSOs without hits wouldn't have a build-id because dsos__read_build_ids_cb() would not read that build-id from the ELF file.

Yep, I think that's the reason.

Quoted 4 lines

Ok, now that makes sense:

Thanks!

Quoted 5 lines

This could have a Fixes attached to it, one that doesn't fixes something that is not working, but speeds up processing by overcoming an oversight when adding build-ids to MMAP records, so I think a:

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Fixes e29386c8f7d71fa5 ("perf record: Add --buildid-mmap option to enable PERF_RECORD_MMAP2's build id")

Is worth.

Sure, will add.

Sorry for the issue, I hope the market dsos is correct. I'm not sure the fixes is correct as that patch called out not populating the .debug. we should make sure that's still the case for people already using --buildid-mmap, they shouldn't have the .debug filled in. Sorry for my lack of a plain text email.

That's guarded by rec->no_buildid so it shouldn't change the behavior.

Thanks,
Namhyung

author maintainer Applied
#5 lore raw
In reply to Namhyung Kim
Quoted 10 lines · Namhyung Kim

It post-processes samples to find which DSO has samples. Based on that info, it can save used DSOs in the build-ID cache directory. But for some reason, it saves all DSOs without checking the hit mark. Skipping unused DSOs can give some speedup especially with --buildid-mmap being default.

On my idle machine, time perf record -a sleep 1 goes down from 3 sec to 1.5 sec with this change.

[...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung