[PATCH v2] perf record: add dummy event during system wide synthesis

[PATCH v2] perf record: add dummy event during system wide synthesis

[Ian Rogers]

During the processing of /proc during event synthesis new processes may
start. Add a dummy event if /proc is to be processed, to capture mmaps
for starting processes. This reuses the existing logic for
initial-delay.

v2 fixes the dummy event configuration and a branch stack issue.

Suggested-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/builtin-record.c | 19 ++++++++++++++-----
 tools/perf/util/evsel.c     |  5 ++++-
 2 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 1ab349abe904..8d1e93351298 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -805,19 +805,28 @@ static int record__open(struct record *rec)
 	int rc = 0;
 
 	/*
-	 * For initial_delay we need to add a dummy event so that we can track
-	 * PERF_RECORD_MMAP while we wait for the initial delay to enable the
-	 * real events, the ones asked by the user.
+	 * For initial_delay or system wide, we need to add a dummy event so
+	 * that we can track PERF_RECORD_MMAP to cover the delay of waiting or
+	 * event synthesis.
 	 */
-	if (opts->initial_delay) {
+	if (opts->initial_delay || target__has_cpu(&opts->target)) {
 		if (perf_evlist__add_dummy(evlist))
 			return -ENOMEM;
 
+		/* Disable tracking of mmaps on lead event. */
 		pos = evlist__first(evlist);
 		pos->tracking = 0;
+		/* Set up dummy event. */
 		pos = evlist__last(evlist);
 		pos->tracking = 1;
-		pos->core.attr.enable_on_exec = 1;
+		/*
+		 * Enable the dummy event when the process is forked for
+		 * initial_delay, immediately for system wide.
+		 */
+		if (opts->initial_delay)
+			pos->core.attr.enable_on_exec = 1;
+		else
+			pos->immediate = 1;
 	}
 
 	perf_evlist__config(evlist, opts, &callchain_param);
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 6a571d322bb2..ca8f9533d8f9 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -1163,11 +1163,14 @@ void perf_evsel__config(struct evsel *evsel, struct record_opts *opts,
 	}
 
 	/*
+	 * A dummy event never triggers any actual counter and therefore
+	 * cannot be used with branch_stack.
+	 *
 	 * For initial_delay, a dummy event is added implicitly.
 	 * The software event will trigger -EOPNOTSUPP error out,
 	 * if BRANCH_STACK bit is set.
 	 */
-	if (opts->initial_delay && is_dummy_event(evsel))
+	if (is_dummy_event(evsel))
 		perf_evsel__reset_sample_bit(evsel, BRANCH_STACK);
 }
 
-- 
2.26.1.301.g55bc3eb7cb9-goog

Re: [PATCH v2] perf record: add dummy event during system wide synthesis

[Jiri Olsa]

On Mon, Apr 20, 2020 at 11:11:03PM -0700, Ian Rogers wrote:
> During the processing of /proc during event synthesis new processes may
> start. Add a dummy event if /proc is to be processed, to capture mmaps
> for starting processes. This reuses the existing logic for
> initial-delay.
> 
> v2 fixes the dummy event configuration and a branch stack issue.

heya,
it's breaking attr tests for me:

[jolsa@krava perf]$ sudo ./perf test -v 17
17: Setup struct perf_event_attr                          :
--- start ---
test child forked, pid 1046560
running './tests/attr/test-record-graph-default'
running './tests/attr/test-record-raw'
running './tests/attr/test-record-branch-filter-any'
running './tests/attr/test-record-freq'
running './tests/attr/test-record-branch-any'
running './tests/attr/test-stat-group1'
running './tests/attr/test-record-no-samples'
running './tests/attr/test-record-graph-dwarf'
running './tests/attr/test-stat-C0'
running './tests/attr/test-stat-basic'
running './tests/attr/test-record-group'
running './tests/attr/test-record-branch-filter-k'
running './tests/attr/test-stat-group'
running './tests/attr/test-record-C0'
expected config=0, got 9
expected sample_type=391, got 455
expected disabled=1, got 0
FAILED './tests/attr/test-record-C0' - match failure
test child finished with -1
---- end ----
Setup struct perf_event_attr: FAILED!

jirka

> 
> Suggested-by: Stephane Eranian <eranian@google.com>
> Signed-off-by: Ian Rogers <irogers@google.com>
> ---
>  tools/perf/builtin-record.c | 19 ++++++++++++++-----
>  tools/perf/util/evsel.c     |  5 ++++-
>  2 files changed, 18 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 1ab349abe904..8d1e93351298 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -805,19 +805,28 @@ static int record__open(struct record *rec)
>  	int rc = 0;
>  
>  	/*
> -	 * For initial_delay we need to add a dummy event so that we can track
> -	 * PERF_RECORD_MMAP while we wait for the initial delay to enable the
> -	 * real events, the ones asked by the user.
> +	 * For initial_delay or system wide, we need to add a dummy event so
> +	 * that we can track PERF_RECORD_MMAP to cover the delay of waiting or
> +	 * event synthesis.
>  	 */
> -	if (opts->initial_delay) {
> +	if (opts->initial_delay || target__has_cpu(&opts->target)) {
>  		if (perf_evlist__add_dummy(evlist))
>  			return -ENOMEM;
>  
> +		/* Disable tracking of mmaps on lead event. */
>  		pos = evlist__first(evlist);
>  		pos->tracking = 0;
> +		/* Set up dummy event. */
>  		pos = evlist__last(evlist);
>  		pos->tracking = 1;
> -		pos->core.attr.enable_on_exec = 1;
> +		/*
> +		 * Enable the dummy event when the process is forked for
> +		 * initial_delay, immediately for system wide.
> +		 */
> +		if (opts->initial_delay)
> +			pos->core.attr.enable_on_exec = 1;
> +		else
> +			pos->immediate = 1;
>  	}
>  
>  	perf_evlist__config(evlist, opts, &callchain_param);
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 6a571d322bb2..ca8f9533d8f9 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -1163,11 +1163,14 @@ void perf_evsel__config(struct evsel *evsel, struct record_opts *opts,
>  	}
>  
>  	/*
> +	 * A dummy event never triggers any actual counter and therefore
> +	 * cannot be used with branch_stack.
> +	 *
>  	 * For initial_delay, a dummy event is added implicitly.
>  	 * The software event will trigger -EOPNOTSUPP error out,
>  	 * if BRANCH_STACK bit is set.
>  	 */
> -	if (opts->initial_delay && is_dummy_event(evsel))
> +	if (is_dummy_event(evsel))
>  		perf_evsel__reset_sample_bit(evsel, BRANCH_STACK);
>  }
>  
> -- 
> 2.26.1.301.g55bc3eb7cb9-goog
> 

Re: [PATCH v2] perf record: add dummy event during system wide synthesis

[Ian Rogers]

On Wed, Apr 22, 2020 at 1:46 AM Jiri Olsa <jolsa@redhat.com> wrote:
>
> On Mon, Apr 20, 2020 at 11:11:03PM -0700, Ian Rogers wrote:
> > During the processing of /proc during event synthesis new processes may
> > start. Add a dummy event if /proc is to be processed, to capture mmaps
> > for starting processes. This reuses the existing logic for
> > initial-delay.
> >
> > v2 fixes the dummy event configuration and a branch stack issue.
>
> heya,
> it's breaking attr tests for me:
>
> [jolsa@krava perf]$ sudo ./perf test -v 17
> 17: Setup struct perf_event_attr                          :
> --- start ---
> test child forked, pid 1046560
> running './tests/attr/test-record-graph-default'
> running './tests/attr/test-record-raw'
> running './tests/attr/test-record-branch-filter-any'
> running './tests/attr/test-record-freq'
> running './tests/attr/test-record-branch-any'
> running './tests/attr/test-stat-group1'
> running './tests/attr/test-record-no-samples'
> running './tests/attr/test-record-graph-dwarf'
> running './tests/attr/test-stat-C0'
> running './tests/attr/test-stat-basic'
> running './tests/attr/test-record-group'
> running './tests/attr/test-record-branch-filter-k'
> running './tests/attr/test-stat-group'
> running './tests/attr/test-record-C0'
> expected config=0, got 9
> expected sample_type=391, got 455
> expected disabled=1, got 0
> FAILED './tests/attr/test-record-C0' - match failure
> test child finished with -1
> ---- end ----
> Setup struct perf_event_attr: FAILED!
>
> jirka

Thanks, fixed in v3:
https://lore.kernel.org/lkml/20200422173615.59436-1-irogers@google.com/

Ian

> >
> > Suggested-by: Stephane Eranian <eranian@google.com>
> > Signed-off-by: Ian Rogers <irogers@google.com>
> > ---
> >  tools/perf/builtin-record.c | 19 ++++++++++++++-----
> >  tools/perf/util/evsel.c     |  5 ++++-
> >  2 files changed, 18 insertions(+), 6 deletions(-)
> >
> > diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> > index 1ab349abe904..8d1e93351298 100644
> > --- a/tools/perf/builtin-record.c
> > +++ b/tools/perf/builtin-record.c
> > @@ -805,19 +805,28 @@ static int record__open(struct record *rec)
> >       int rc = 0;
> >
> >       /*
> > -      * For initial_delay we need to add a dummy event so that we can track
> > -      * PERF_RECORD_MMAP while we wait for the initial delay to enable the
> > -      * real events, the ones asked by the user.
> > +      * For initial_delay or system wide, we need to add a dummy event so
> > +      * that we can track PERF_RECORD_MMAP to cover the delay of waiting or
> > +      * event synthesis.
> >        */
> > -     if (opts->initial_delay) {
> > +     if (opts->initial_delay || target__has_cpu(&opts->target)) {
> >               if (perf_evlist__add_dummy(evlist))
> >                       return -ENOMEM;
> >
> > +             /* Disable tracking of mmaps on lead event. */
> >               pos = evlist__first(evlist);
> >               pos->tracking = 0;
> > +             /* Set up dummy event. */
> >               pos = evlist__last(evlist);
> >               pos->tracking = 1;
> > -             pos->core.attr.enable_on_exec = 1;
> > +             /*
> > +              * Enable the dummy event when the process is forked for
> > +              * initial_delay, immediately for system wide.
> > +              */
> > +             if (opts->initial_delay)
> > +                     pos->core.attr.enable_on_exec = 1;
> > +             else
> > +                     pos->immediate = 1;
> >       }
> >
> >       perf_evlist__config(evlist, opts, &callchain_param);
> > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> > index 6a571d322bb2..ca8f9533d8f9 100644
> > --- a/tools/perf/util/evsel.c
> > +++ b/tools/perf/util/evsel.c
> > @@ -1163,11 +1163,14 @@ void perf_evsel__config(struct evsel *evsel, struct record_opts *opts,
> >       }
> >
> >       /*
> > +      * A dummy event never triggers any actual counter and therefore
> > +      * cannot be used with branch_stack.
> > +      *
> >        * For initial_delay, a dummy event is added implicitly.
> >        * The software event will trigger -EOPNOTSUPP error out,
> >        * if BRANCH_STACK bit is set.
> >        */
> > -     if (opts->initial_delay && is_dummy_event(evsel))
> > +     if (is_dummy_event(evsel))
> >               perf_evsel__reset_sample_bit(evsel, BRANCH_STACK);
> >  }
> >
> > --
> > 2.26.1.301.g55bc3eb7cb9-goog
> >
>