[PATCH] perf tools: set no branch type for dummy event in PT

[PATCH] perf tools: set no branch type for dummy event in PT

[kan.liang]

From: Kan Liang <kan.liang@intel.com>

An earlier kernel patch allowed enabling PT and LBR at the same
time on Goldmont. 
commit ccbebba4c6bf ("perf/x86/intel/pt: Bypass PT vs. LBR
exclusivity if the core supports it")
However, users still cannot use Intel PT and LBRs simultaneously.
    $ sudo perf record -e cycles,intel_pt//u -b  -- sleep 1
    Error:
    PMU Hardware doesn't support sampling/overflow-interrupts.

PT implicitly adds dummy event in perf tool. dummy event is
software event which doesn't support LBR.

Always setting branch_type=no for dummy event in Intel PT.

Signed-off-by: Kan Liang <kan.liang@intel.com>
---
 tools/perf/arch/x86/util/intel-pt.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/tools/perf/arch/x86/util/intel-pt.c b/tools/perf/arch/x86/util/intel-pt.c
index f630de0..651ab9e 100644
--- a/tools/perf/arch/x86/util/intel-pt.c
+++ b/tools/perf/arch/x86/util/intel-pt.c
@@ -544,6 +544,22 @@ static int intel_pt_validate_config(struct perf_pmu *intel_pt_pmu,
 					evsel->attr.config);
 }
 
+static int add_no_lbr_config_term(struct list_head *config_terms)
+{
+	struct perf_evsel_config_term *lbr_term;
+
+	lbr_term = zalloc(sizeof(*lbr_term));
+	if (!lbr_term)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&lbr_term->list);
+	lbr_term->type = PERF_EVSEL__CONFIG_TERM_BRANCH;
+	lbr_term->val.branch = strdup("no");
+	list_add_tail(&lbr_term->list, config_terms);
+
+	return 0;
+}
+
 static int intel_pt_recording_options(struct auxtrace_record *itr,
 				      struct perf_evlist *evlist,
 				      struct record_opts *opts)
@@ -701,6 +717,8 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 				perf_evsel__set_sample_bit(switch_evsel, TIME);
 				perf_evsel__set_sample_bit(switch_evsel, CPU);
 
+				add_no_lbr_config_term(&switch_evsel->config_terms);
+
 				opts->record_switch_events = false;
 				ptr->have_sched_switch = 3;
 			} else {
@@ -760,6 +778,7 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 			/* And the CPU for switch events */
 			perf_evsel__set_sample_bit(tracking_evsel, CPU);
 		}
+		add_no_lbr_config_term(&tracking_evsel->config_terms);
 	}
 
 	/*
-- 
2.9.4

Re: [PATCH] perf tools: set no branch type for dummy event in PT

[Jiri Olsa]

On Wed, Jun 28, 2017 at 10:31:53AM -0400, kan.liang@intel.com wrote:
> From: Kan Liang <kan.liang@intel.com>
> 
> An earlier kernel patch allowed enabling PT and LBR at the same
> time on Goldmont. 
> commit ccbebba4c6bf ("perf/x86/intel/pt: Bypass PT vs. LBR
> exclusivity if the core supports it")
> However, users still cannot use Intel PT and LBRs simultaneously.
>     $ sudo perf record -e cycles,intel_pt//u -b  -- sleep 1
>     Error:
>     PMU Hardware doesn't support sampling/overflow-interrupts.
> 
> PT implicitly adds dummy event in perf tool. dummy event is
> software event which doesn't support LBR.
> 
> Always setting branch_type=no for dummy event in Intel PT.
> 
> Signed-off-by: Kan Liang <kan.liang@intel.com>
> ---
>  tools/perf/arch/x86/util/intel-pt.c | 19 +++++++++++++++++++
>  1 file changed, 19 insertions(+)
> 
> diff --git a/tools/perf/arch/x86/util/intel-pt.c b/tools/perf/arch/x86/util/intel-pt.c
> index f630de0..651ab9e 100644
> --- a/tools/perf/arch/x86/util/intel-pt.c
> +++ b/tools/perf/arch/x86/util/intel-pt.c
> @@ -544,6 +544,22 @@ static int intel_pt_validate_config(struct perf_pmu *intel_pt_pmu,
>  					evsel->attr.config);
>  }
>  
> +static int add_no_lbr_config_term(struct list_head *config_terms)
> +{
> +	struct perf_evsel_config_term *lbr_term;
> +
> +	lbr_term = zalloc(sizeof(*lbr_term));
> +	if (!lbr_term)
> +		return -ENOMEM;
> +
> +	INIT_LIST_HEAD(&lbr_term->list);
> +	lbr_term->type = PERF_EVSEL__CONFIG_TERM_BRANCH;
> +	lbr_term->val.branch = strdup("no");
> +	list_add_tail(&lbr_term->list, config_terms);
> +
> +	return 0;
> +}
> +
>  static int intel_pt_recording_options(struct auxtrace_record *itr,
>  				      struct perf_evlist *evlist,
>  				      struct record_opts *opts)
> @@ -701,6 +717,8 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
>  				perf_evsel__set_sample_bit(switch_evsel, TIME);
>  				perf_evsel__set_sample_bit(switch_evsel, CPU);
>  
> +				add_no_lbr_config_term(&switch_evsel->config_terms);
> +

hum, why can't you change the sample bit directly? with:

			perf_evsel__reset_sample_bit(switch_evsel, BRANCH_STACK);

jirka

RE: [PATCH] perf tools: set no branch type for dummy event in PT

[Liang, Kan]

> On Wed, Jun 28, 2017 at 10:31:53AM -0400, kan.liang@intel.com wrote:
> > From: Kan Liang <kan.liang@intel.com>
> >
> > An earlier kernel patch allowed enabling PT and LBR at the same time
> > on Goldmont.
> > commit ccbebba4c6bf ("perf/x86/intel/pt: Bypass PT vs. LBR exclusivity
> > if the core supports it") However, users still cannot use Intel PT and
> > LBRs simultaneously.
> >     $ sudo perf record -e cycles,intel_pt//u -b  -- sleep 1
> >     Error:
> >     PMU Hardware doesn't support sampling/overflow-interrupts.
> >
> > PT implicitly adds dummy event in perf tool. dummy event is software
> > event which doesn't support LBR.
> >
> > Always setting branch_type=no for dummy event in Intel PT.
> >
> > Signed-off-by: Kan Liang <kan.liang@intel.com>
> > ---
> >  tools/perf/arch/x86/util/intel-pt.c | 19 +++++++++++++++++++
> >  1 file changed, 19 insertions(+)
> >
> > diff --git a/tools/perf/arch/x86/util/intel-pt.c
> > b/tools/perf/arch/x86/util/intel-pt.c
> > index f630de0..651ab9e 100644
> > --- a/tools/perf/arch/x86/util/intel-pt.c
> > +++ b/tools/perf/arch/x86/util/intel-pt.c
> > @@ -544,6 +544,22 @@ static int intel_pt_validate_config(struct perf_pmu
> *intel_pt_pmu,
> >  					evsel->attr.config);
> >  }
> >
> > +static int add_no_lbr_config_term(struct list_head *config_terms) {
> > +	struct perf_evsel_config_term *lbr_term;
> > +
> > +	lbr_term = zalloc(sizeof(*lbr_term));
> > +	if (!lbr_term)
> > +		return -ENOMEM;
> > +
> > +	INIT_LIST_HEAD(&lbr_term->list);
> > +	lbr_term->type = PERF_EVSEL__CONFIG_TERM_BRANCH;
> > +	lbr_term->val.branch = strdup("no");
> > +	list_add_tail(&lbr_term->list, config_terms);
> > +
> > +	return 0;
> > +}
> > +
> >  static int intel_pt_recording_options(struct auxtrace_record *itr,
> >  				      struct perf_evlist *evlist,
> >  				      struct record_opts *opts)
> > @@ -701,6 +717,8 @@ static int intel_pt_recording_options(struct
> auxtrace_record *itr,
> >  				perf_evsel__set_sample_bit(switch_evsel,
> TIME);
> >  				perf_evsel__set_sample_bit(switch_evsel,
> CPU);
> >
> > +				add_no_lbr_config_term(&switch_evsel-
> >config_terms);
> > +
> 
> hum, why can't you change the sample bit directly? with:
> 
> 			perf_evsel__reset_sample_bit(switch_evsel,
> BRANCH_STACK);

It will be overwrite in perf_evsel__config.

Thanks,
Kan

Re: [PATCH] perf tools: set no branch type for dummy event in PT

[Jiri Olsa]

On Thu, Jun 29, 2017 at 03:31:45PM +0000, Liang, Kan wrote:

SNIP

> > >  static int intel_pt_recording_options(struct auxtrace_record *itr,
> > >  				      struct perf_evlist *evlist,
> > >  				      struct record_opts *opts)
> > > @@ -701,6 +717,8 @@ static int intel_pt_recording_options(struct
> > auxtrace_record *itr,
> > >  				perf_evsel__set_sample_bit(switch_evsel,
> > TIME);
> > >  				perf_evsel__set_sample_bit(switch_evsel,
> > CPU);
> > >
> > > +				add_no_lbr_config_term(&switch_evsel-
> > >config_terms);
> > > +
> > 
> > hum, why can't you change the sample bit directly? with:
> > 
> > 			perf_evsel__reset_sample_bit(switch_evsel,
> > BRANCH_STACK);
> 
> It will be overwrite in perf_evsel__config.
> 

where? you set the evsel->no_aux_samples 

jirka

RE: [PATCH] perf tools: set no branch type for dummy event in PT

[Liang, Kan]

> 
> On Thu, Jun 29, 2017 at 03:31:45PM +0000, Liang, Kan wrote:
> 
> SNIP
> 
> > > >  static int intel_pt_recording_options(struct auxtrace_record *itr,
> > > >  				      struct perf_evlist *evlist,
> > > >  				      struct record_opts *opts) @@ -701,6
> +717,8 @@ static
> > > > int intel_pt_recording_options(struct
> > > auxtrace_record *itr,
> > > >  				perf_evsel__set_sample_bit(switch_evsel,
> > > TIME);
> > > >  				perf_evsel__set_sample_bit(switch_evsel,
> > > CPU);
> > > >
> > > > +				add_no_lbr_config_term(&switch_evsel-
> > > >config_terms);
> > > > +
> > >
> > > hum, why can't you change the sample bit directly? with:
> > >
> > > 			perf_evsel__reset_sample_bit(switch_evsel,
> > > BRANCH_STACK);
> >
> > It will be overwrite in perf_evsel__config.
> >
> 
> where? you set the evsel->no_aux_samples

Yes for switch_evsel, but no for tracking_evsel.

If it's only for switch_evsel, yes, we can change the sample bit directly.
But I think we should use the same method for both of them.

Thanks,
Kan

Re: [PATCH] perf tools: set no branch type for dummy event in PT

[Jiri Olsa]

On Thu, Jun 29, 2017 at 03:50:29PM +0000, Liang, Kan wrote:
> 
> 
> > 
> > On Thu, Jun 29, 2017 at 03:31:45PM +0000, Liang, Kan wrote:
> > 
> > SNIP
> > 
> > > > >  static int intel_pt_recording_options(struct auxtrace_record *itr,
> > > > >  				      struct perf_evlist *evlist,
> > > > >  				      struct record_opts *opts) @@ -701,6
> > +717,8 @@ static
> > > > > int intel_pt_recording_options(struct
> > > > auxtrace_record *itr,
> > > > >  				perf_evsel__set_sample_bit(switch_evsel,
> > > > TIME);
> > > > >  				perf_evsel__set_sample_bit(switch_evsel,
> > > > CPU);
> > > > >
> > > > > +				add_no_lbr_config_term(&switch_evsel-
> > > > >config_terms);
> > > > > +
> > > >
> > > > hum, why can't you change the sample bit directly? with:
> > > >
> > > > 			perf_evsel__reset_sample_bit(switch_evsel,
> > > > BRANCH_STACK);
> > >
> > > It will be overwrite in perf_evsel__config.
> > >
> > 
> > where? you set the evsel->no_aux_samples
> 
> Yes for switch_evsel, but no for tracking_evsel.
> 
> If it's only for switch_evsel, yes, we can change the sample bit directly.
> But I think we should use the same method for both of them.


we could set the no_aux_samples for the tracking_evsel as well

jirka

RE: [PATCH] perf tools: set no branch type for dummy event in PT

[Liang, Kan]

> On Thu, Jun 29, 2017 at 03:50:29PM +0000, Liang, Kan wrote:
> >
> >
> > >
> > > On Thu, Jun 29, 2017 at 03:31:45PM +0000, Liang, Kan wrote:
> > >
> > > SNIP
> > >
> > > > > >  static int intel_pt_recording_options(struct auxtrace_record *itr,
> > > > > >  				      struct perf_evlist *evlist,
> > > > > >  				      struct record_opts *opts) @@ -
> 701,6
> > > +717,8 @@ static
> > > > > > int intel_pt_recording_options(struct
> > > > > auxtrace_record *itr,
> > > > > >
> 	perf_evsel__set_sample_bit(switch_evsel,
> > > > > TIME);
> > > > > >
> 	perf_evsel__set_sample_bit(switch_evsel,
> > > > > CPU);
> > > > > >
> > > > > > +
> 	add_no_lbr_config_term(&switch_evsel-
> > > > > >config_terms);
> > > > > > +
> > > > >
> > > > > hum, why can't you change the sample bit directly? with:
> > > > >
> > > > > 			perf_evsel__reset_sample_bit(switch_evsel,
> > > > > BRANCH_STACK);
> > > >
> > > > It will be overwrite in perf_evsel__config.
> > > >
> > >
> > > where? you set the evsel->no_aux_samples
> >
> > Yes for switch_evsel, but no for tracking_evsel.
> >
> > If it's only for switch_evsel, yes, we can change the sample bit directly.
> > But I think we should use the same method for both of them.
> 
> 
> we could set the no_aux_samples for the tracking_evsel as well
> 

Yes, I will do it in next version.

Thanks,
Kan