perf: Improve support for uncore JSON event lists

perf: Improve support for uncore JSON event lists

[Andi Kleen]

This patch kit further improves support for Intel uncore events in
the Linux perf user tool. The basic support has been already
merged earlier, but this makes it nicer to use.

- Collapse counts from duplicated boxes to make the output
easier to read.
- Support specifying events for multiple duplicated boxes
in an abbreviated format to shorten event specifiers
- Add support for computing Metrics defined in the event lists,
so that the event lists can extend the metrics in perf stat.
This allows to represent many events in an easier to understand
format.

Available from

git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc.git perf/builtin-json-28

v1: Initial post after being split off to own patchkit
Adding MetricName support and support for more than two events
in expressions.

v2: Address review comments. Move new hunk from refactor
patchkit to patch adding new features. Improve changelogs
slightly.

[PATCH 01/10] perf, tools, stat: Factor out callback for collecting event values

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

To be used in next patch to support automatic summing of alias
events.

v2: Move check for bad results to next patch
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/builtin-stat.c | 104 ++++++++++++++++++++++++++++++++++++----------
 1 file changed, 81 insertions(+), 23 deletions(-)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index f28719178b51..c71916eaa255 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1178,11 +1178,47 @@ static void aggr_update_shadow(void)
 	}
 }
 
+static void collect_data(struct perf_evsel *counter,
+			    void (*cb)(struct perf_evsel *counter, void *data,
+				       bool first),
+			    void *data)
+{
+	cb(counter, data, true);
+}
+
+struct aggr_data {
+	u64 ena, run, val;
+	int id;
+	int nr;
+	int cpu;
+};
+
+static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
+{
+	struct aggr_data *ad = data;
+	int cpu, cpu2, s2;
+
+	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
+		struct perf_counts_values *counts;
+
+		cpu2 = perf_evsel__cpus(counter)->map[cpu];
+		s2 = aggr_get_id(evsel_list->cpus, cpu2);
+		if (s2 != ad->id)
+			continue;
+		if (first)
+			ad->nr++;
+		counts = perf_counts(counter->counts, cpu, 0);
+		ad->val += counts->val;
+		ad->ena += counts->ena;
+		ad->run += counts->run;
+	}
+}
+
 static void print_aggr(char *prefix)
 {
 	FILE *output = stat_config.output;
 	struct perf_evsel *counter;
-	int cpu, s, s2, id, nr;
+	int s, id, nr;
 	double uval;
 	u64 ena, run, val;
 	bool first;
@@ -1197,23 +1233,20 @@ static void print_aggr(char *prefix)
 	 * Without each counter has its own line.
 	 */
 	for (s = 0; s < aggr_map->nr; s++) {
+		struct aggr_data ad;
 		if (prefix && metric_only)
 			fprintf(output, "%s", prefix);
 
-		id = aggr_map->map[s];
+		ad.id = id = aggr_map->map[s];
 		first = true;
 		evlist__for_each_entry(evsel_list, counter) {
-			val = ena = run = 0;
-			nr = 0;
-			for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
-				s2 = aggr_get_id(perf_evsel__cpus(counter), cpu);
-				if (s2 != id)
-					continue;
-				val += perf_counts(counter->counts, cpu, 0)->val;
-				ena += perf_counts(counter->counts, cpu, 0)->ena;
-				run += perf_counts(counter->counts, cpu, 0)->run;
-				nr++;
-			}
+			ad.val = ad.ena = ad.run = 0;
+			ad.nr = 0;
+			collect_data(counter, aggr_cb, &ad);
+			nr = ad.nr;
+			ena = ad.ena;
+			run = ad.run;
+			val = ad.val;
 			if (first && metric_only) {
 				first = false;
 				aggr_printout(counter, id, nr);
@@ -1257,6 +1290,21 @@ static void print_aggr_thread(struct perf_evsel *counter, char *prefix)
 	}
 }
 
+struct caggr_data {
+	double avg, avg_enabled, avg_running;
+};
+
+static void counter_aggr_cb(struct perf_evsel *counter, void *data,
+			    bool first __maybe_unused)
+{
+	struct caggr_data *cd = data;
+	struct perf_stat_evsel *ps = counter->priv;
+
+	cd->avg += avg_stats(&ps->res_stats[0]);
+	cd->avg_enabled += avg_stats(&ps->res_stats[1]);
+	cd->avg_running += avg_stats(&ps->res_stats[2]);
+}
+
 /*
  * Print out the results of a single counter:
  * aggregated counts in system-wide mode
@@ -1264,23 +1312,30 @@ static void print_aggr_thread(struct perf_evsel *counter, char *prefix)
 static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
 {
 	FILE *output = stat_config.output;
-	struct perf_stat_evsel *ps = counter->priv;
-	double avg = avg_stats(&ps->res_stats[0]);
 	double uval;
-	double avg_enabled, avg_running;
+	struct caggr_data cd = { .avg = 0.0 };
 
-	avg_enabled = avg_stats(&ps->res_stats[1]);
-	avg_running = avg_stats(&ps->res_stats[2]);
+	collect_data(counter, counter_aggr_cb, &cd);
 
 	if (prefix && !metric_only)
 		fprintf(output, "%s", prefix);
 
-	uval = avg * counter->scale;
-	printout(-1, 0, counter, uval, prefix, avg_running, avg_enabled, avg);
+	uval = cd.avg * counter->scale;
+	printout(-1, 0, counter, uval, prefix, cd.avg_running, cd.avg_enabled, cd.avg);
 	if (!metric_only)
 		fprintf(output, "\n");
 }
 
+static void counter_cb(struct perf_evsel *counter, void *data,
+		       bool first __maybe_unused)
+{
+	struct aggr_data *ad = data;
+
+	ad->val += perf_counts(counter->counts, ad->cpu, 0)->val;
+	ad->ena += perf_counts(counter->counts, ad->cpu, 0)->ena;
+	ad->run += perf_counts(counter->counts, ad->cpu, 0)->run;
+}
+
 /*
  * Print out the results of a single counter:
  * does not use aggregated count in system-wide
@@ -1293,9 +1348,12 @@ static void print_counter(struct perf_evsel *counter, char *prefix)
 	int cpu;
 
 	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
-		val = perf_counts(counter->counts, cpu, 0)->val;
-		ena = perf_counts(counter->counts, cpu, 0)->ena;
-		run = perf_counts(counter->counts, cpu, 0)->run;
+		struct aggr_data ad = { .cpu = cpu };
+
+		collect_data(counter, counter_cb, &ad);
+		val = ad.val;
+		ena = ad.ena;
+		run = ad.run;
 
 		if (prefix)
 			fprintf(output, "%s", prefix);
-- 
2.9.3

Re: [PATCH 01/10] perf, tools, stat: Factor out callback for collecting event values

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:15PM -0800, Andi Kleen wrote:

SNIP

> +static void collect_data(struct perf_evsel *counter,
> +			    void (*cb)(struct perf_evsel *counter, void *data,
> +				       bool first),
> +			    void *data)
> +{
> +	cb(counter, data, true);
> +}
> +
> +struct aggr_data {
> +	u64 ena, run, val;
> +	int id;
> +	int nr;
> +	int cpu;
> +};
> +
> +static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
> +{
> +	struct aggr_data *ad = data;
> +	int cpu, cpu2, s2;
> +
> +	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
> +		struct perf_counts_values *counts;
> +
> +		cpu2 = perf_evsel__cpus(counter)->map[cpu];
> +		s2 = aggr_get_id(evsel_list->cpus, cpu2);

that does not match the removed code.. why?
                s2 = aggr_get_id(perf_evsel__cpus(counter), cpu);

jirka

Re: [PATCH 01/10] perf, tools, stat: Factor out callback for collecting event values

[Andi Kleen]

On Fri, Mar 03, 2017 at 11:33:03AM +0100, Jiri Olsa wrote:
> On Tue, Feb 28, 2017 at 10:49:15PM -0800, Andi Kleen wrote:
> 
> SNIP
> 
> > +static void collect_data(struct perf_evsel *counter,
> > +			    void (*cb)(struct perf_evsel *counter, void *data,
> > +				       bool first),
> > +			    void *data)
> > +{
> > +	cb(counter, data, true);
> > +}
> > +
> > +struct aggr_data {
> > +	u64 ena, run, val;
> > +	int id;
> > +	int nr;
> > +	int cpu;
> > +};
> > +
> > +static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
> > +{
> > +	struct aggr_data *ad = data;
> > +	int cpu, cpu2, s2;
> > +
> > +	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
> > +		struct perf_counts_values *counts;
> > +
> > +		cpu2 = perf_evsel__cpus(counter)->map[cpu];
> > +		s2 = aggr_get_id(evsel_list->cpus, cpu2);
> 
> that does not match the removed code.. why?
>                 s2 = aggr_get_id(perf_evsel__cpus(counter), cpu);

I added it at some point during debugging. I think it's
actually a nop here, but technically it's correct (we're
always supposed to remap)

I can remove it, but it's really a small code cleanup.

-Andi

Re: [PATCH 01/10] perf, tools, stat: Factor out callback for collecting event values

[Jiri Olsa]

On Fri, Mar 03, 2017 at 03:16:11PM -0800, Andi Kleen wrote:
> On Fri, Mar 03, 2017 at 11:33:03AM +0100, Jiri Olsa wrote:
> > On Tue, Feb 28, 2017 at 10:49:15PM -0800, Andi Kleen wrote:
> > 
> > SNIP
> > 
> > > +static void collect_data(struct perf_evsel *counter,
> > > +			    void (*cb)(struct perf_evsel *counter, void *data,
> > > +				       bool first),
> > > +			    void *data)
> > > +{
> > > +	cb(counter, data, true);
> > > +}
> > > +
> > > +struct aggr_data {
> > > +	u64 ena, run, val;
> > > +	int id;
> > > +	int nr;
> > > +	int cpu;
> > > +};
> > > +
> > > +static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
> > > +{
> > > +	struct aggr_data *ad = data;
> > > +	int cpu, cpu2, s2;
> > > +
> > > +	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
> > > +		struct perf_counts_values *counts;
> > > +
> > > +		cpu2 = perf_evsel__cpus(counter)->map[cpu];
> > > +		s2 = aggr_get_id(evsel_list->cpus, cpu2);
> > 
> > that does not match the removed code.. why?
> >                 s2 = aggr_get_id(perf_evsel__cpus(counter), cpu);
> 
> I added it at some point during debugging. I think it's
> actually a nop here, but technically it's correct (we're
> always supposed to remap)
> 
> I can remove it, but it's really a small code cleanup.

please do, thanks

jirka

[PATCH 02/10] perf, tools, stat: Collapse identically named events

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

The uncore PMU has a lot of duplicated PMUs for different subsystems.
When expanding an uncore alias we usually end up with a large
number of identically named aliases, which makes perf stat
output difficult to read.

Automatically sum them up in perf stat, unless --no-merge is specified.

This can be default because only the uncores generally have duplicated
aliases. Other PMUs have unique names.

Before:

% perf stat --no-merge -a  -e unc_c_llc_lookup.any sleep 1

 Performance counter stats for 'system wide':

           694,976 Bytes unc_c_llc_lookup.any
           706,304 Bytes unc_c_llc_lookup.any
           956,608 Bytes unc_c_llc_lookup.any
           782,720 Bytes unc_c_llc_lookup.any
           605,696 Bytes unc_c_llc_lookup.any
           442,816 Bytes unc_c_llc_lookup.any
           659,328 Bytes unc_c_llc_lookup.any
           509,312 Bytes unc_c_llc_lookup.any
           263,936 Bytes unc_c_llc_lookup.any
           592,448 Bytes unc_c_llc_lookup.any
           672,448 Bytes unc_c_llc_lookup.any
           608,640 Bytes unc_c_llc_lookup.any
           641,024 Bytes unc_c_llc_lookup.any
           856,896 Bytes unc_c_llc_lookup.any
           808,832 Bytes unc_c_llc_lookup.any
           684,864 Bytes unc_c_llc_lookup.any
           710,464 Bytes unc_c_llc_lookup.any
           538,304 Bytes unc_c_llc_lookup.any

       1.002577660 seconds time elapsed

After:

% perf stat  -a  -e unc_c_llc_lookup.any sleep 1

 Performance counter stats for 'system wide':

         2,685,120 Bytes unc_c_llc_lookup.any

       1.002648032 seconds time elapsed

v2: Split collect_aliases. Rename alias flag.
v3: Make sure unsupported/not counted is always printed.
v4: Factor out callback change into separate patch.
v5: Move check for bad results here
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/Documentation/perf-stat.txt |  3 +++
 tools/perf/builtin-stat.c              | 41 ++++++++++++++++++++++++++++++++++
 tools/perf/util/evsel.h                |  1 +
 3 files changed, 45 insertions(+)

diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index d96ccd4844df..320d8020bc5b 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -237,6 +237,9 @@ To interpret the results it is usually needed to know on which
 CPUs the workload runs on. If needed the CPUs can be forced using
 taskset.
 
+--no-merge::
+Do not merge results from same PMUs.
+
 EXAMPLES
 --------
 
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index c71916eaa255..389c8e457bf0 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -140,6 +140,7 @@ static unsigned int		unit_width			= 4; /* strlen("unit") */
 static bool			forever				= false;
 static bool			metric_only			= false;
 static bool			force_metric_only		= false;
+static bool			no_merge			= false;
 static struct timespec		ref_time;
 static struct cpu_map		*aggr_map;
 static aggr_get_id_t		aggr_get_id;
@@ -1178,12 +1179,34 @@ static void aggr_update_shadow(void)
 	}
 }
 
+static void collect_all_aliases(struct perf_evsel *counter,
+			    void (*cb)(struct perf_evsel *counter, void *data,
+				       bool first),
+			    void *data)
+{
+	struct perf_evsel *alias;
+
+	alias = list_prepare_entry(counter, &(evsel_list->entries), node);
+	list_for_each_entry_continue (alias, &evsel_list->entries, node) {
+		if (strcmp(perf_evsel__name(alias), perf_evsel__name(counter)) ||
+		    alias->scale != counter->scale ||
+		    alias->cgrp != counter->cgrp ||
+		    strcmp(alias->unit, counter->unit) ||
+		    nsec_counter(alias) != nsec_counter(counter))
+			break;
+		alias->merged_stat = true;
+		cb(alias, data, false);
+	}
+}
+
 static void collect_data(struct perf_evsel *counter,
 			    void (*cb)(struct perf_evsel *counter, void *data,
 				       bool first),
 			    void *data)
 {
 	cb(counter, data, true);
+	if (!no_merge)
+		collect_all_aliases(counter, cb, data);
 }
 
 struct aggr_data {
@@ -1208,6 +1231,16 @@ static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
 		if (first)
 			ad->nr++;
 		counts = perf_counts(counter->counts, cpu, 0);
+		/*
+		 * When any result is bad, make them all to give
+		 * consistent output in interval mode.
+		 */
+		if (counts->ena == 0 || counts->run == 0 ||
+		    counter->counts->scaled == -1) {
+			ad->ena = 0;
+			ad->run = 0;
+			break;
+		}
 		ad->val += counts->val;
 		ad->ena += counts->ena;
 		ad->run += counts->run;
@@ -1240,6 +1273,8 @@ static void print_aggr(char *prefix)
 		ad.id = id = aggr_map->map[s];
 		first = true;
 		evlist__for_each_entry(evsel_list, counter) {
+			if (counter->merged_stat)
+				continue;
 			ad.val = ad.ena = ad.run = 0;
 			ad.nr = 0;
 			collect_data(counter, aggr_cb, &ad);
@@ -1315,6 +1350,8 @@ static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
 	double uval;
 	struct caggr_data cd = { .avg = 0.0 };
 
+	if (counter->merged_stat)
+		return;
 	collect_data(counter, counter_aggr_cb, &cd);
 
 	if (prefix && !metric_only)
@@ -1347,6 +1384,9 @@ static void print_counter(struct perf_evsel *counter, char *prefix)
 	double uval;
 	int cpu;
 
+	if (counter->merged_stat)
+		return;
+
 	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
 		struct aggr_data ad = { .cpu = cpu };
 
@@ -1691,6 +1731,7 @@ static const struct option stat_options[] = {
 		    "list of cpus to monitor in system-wide"),
 	OPT_SET_UINT('A', "no-aggr", &stat_config.aggr_mode,
 		    "disable CPU count aggregation", AGGR_NONE),
+	OPT_BOOLEAN(0, "no-merge", &no_merge, "Do not merge identical named events"),
 	OPT_STRING('x', "field-separator", &csv_sep, "separator",
 		   "print counts with custom separator"),
 	OPT_CALLBACK('G', "cgroup", &evsel_list, "name",
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 06ef6f29efa1..bd2e9b112d49 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -131,6 +131,7 @@ struct perf_evsel {
 	bool			cmdline_group_boundary;
 	struct list_head	config_terms;
 	int			bpf_fd;
+	bool			merged_stat;
 };
 
 union u64_swap {
-- 
2.9.3

Re: [PATCH 02/10] perf, tools, stat: Collapse identically named events

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:16PM -0800, Andi Kleen wrote:

SNIP

>  
>  struct aggr_data {
> @@ -1208,6 +1231,16 @@ static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
>  		if (first)
>  			ad->nr++;
>  		counts = perf_counts(counter->counts, cpu, 0);
> +		/*
> +		 * When any result is bad, make them all to give
> +		 * consistent output in interval mode.
> +		 */
> +		if (counts->ena == 0 || counts->run == 0 ||
> +		    counter->counts->scaled == -1) {
> +			ad->ena = 0;
> +			ad->run = 0;
> +			break;
> +		}

I think this needs to be in separate patch, showing
the interval output before and after the change..

thanks,
jirka

Re: [PATCH 02/10] perf, tools, stat: Collapse identically named events

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:16PM -0800, Andi Kleen wrote:

SNIP

>  	}
>  }
>  
> +static void collect_all_aliases(struct perf_evsel *counter,
> +			    void (*cb)(struct perf_evsel *counter, void *data,
> +				       bool first),
> +			    void *data)
> +{
> +	struct perf_evsel *alias;
> +
> +	alias = list_prepare_entry(counter, &(evsel_list->entries), node);
> +	list_for_each_entry_continue (alias, &evsel_list->entries, node) {
> +		if (strcmp(perf_evsel__name(alias), perf_evsel__name(counter)) ||
> +		    alias->scale != counter->scale ||
> +		    alias->cgrp != counter->cgrp ||
> +		    strcmp(alias->unit, counter->unit) ||
> +		    nsec_counter(alias) != nsec_counter(counter))
> +			break;
> +		alias->merged_stat = true;
> +		cb(alias, data, false);
> +	}
> +}
> +
>  static void collect_data(struct perf_evsel *counter,
>  			    void (*cb)(struct perf_evsel *counter, void *data,
>  				       bool first),
>  			    void *data)
>  {

could we place the merged_stat condition in here and have
collect_data return value to process or not the current
counter.. looks like we could get rid of 2 merged_stat
conditions in print_counter_aggr and print_aggr

jirka

[PATCH 03/10] perf, tools: Factor out PMU matching in parser

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Factor out the PMU name matching in the event parser into a separate function,
to use the same code for other grammar rules later.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/util/parse-events.c | 46 ++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/parse-events.h |  5 +++++
 tools/perf/util/parse-events.y | 30 +--------------------------
 3 files changed, 52 insertions(+), 29 deletions(-)

diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 281e44af31e2..5563c47c92df 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1259,6 +1259,52 @@ int parse_events_add_pmu(struct parse_events_evlist *data,
 	return evsel ? 0 : -ENOMEM;
 }
 
+int parse_events_multi_pmu_add(struct parse_events_evlist *data,
+			       char *str, struct list_head **listp)
+{
+	struct list_head *head;
+	struct parse_events_term *term;
+	struct list_head *list;
+	struct perf_pmu *pmu = NULL;
+	int ok = 0;
+
+	*listp = NULL;
+	/* Add it for all PMUs that support the alias */
+	list = malloc(sizeof(struct list_head));
+	if (!list)
+		return -1;
+	INIT_LIST_HEAD(list);
+	while ((pmu = perf_pmu__scan(pmu)) != NULL) {
+		struct perf_pmu_alias *alias;
+
+		list_for_each_entry(alias, &pmu->aliases, list) {
+			if (!strcasecmp(alias->name, str)) {
+				head = malloc(sizeof(struct list_head));
+				if (!head)
+					return -1;
+				INIT_LIST_HEAD(head);
+				if (parse_events_term__num(&term, PARSE_EVENTS__TERM_TYPE_USER,
+							   str, 1, &str, NULL) < 0)
+					return -1;
+				list_add_tail(&term->list, head);
+
+				if (!parse_events_add_pmu(data, list,
+						  pmu->name, head)) {
+					pr_debug("%s -> %s/%s/\n", str,
+						 pmu->name, alias->str);
+					ok++;
+				}
+
+				parse_events_terms__delete(head);
+			}
+		}
+	}
+	if (!ok)
+		return -1;
+	*listp = list;
+	return 0;
+}
+
 int parse_events__modifier_group(struct list_head *list,
 				 char *event_mod)
 {
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index da246a3ddb69..fb755ec248ab 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -164,6 +164,11 @@ int parse_events_add_breakpoint(struct list_head *list, int *idx,
 int parse_events_add_pmu(struct parse_events_evlist *data,
 			 struct list_head *list, char *name,
 			 struct list_head *head_config);
+
+int parse_events_multi_pmu_add(struct parse_events_evlist *data,
+			       char *str,
+			       struct list_head **listp);
+
 enum perf_pmu_event_symbol_type
 perf_pmu__parse_check(const char *name);
 void parse_events__set_leader(char *name, struct list_head *list);
diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
index a14b47ab3879..6b2bc918fc94 100644
--- a/tools/perf/util/parse-events.y
+++ b/tools/perf/util/parse-events.y
@@ -236,37 +236,9 @@ PE_NAME opt_event_config
 |
 PE_KERNEL_PMU_EVENT sep_dc
 {
-	struct parse_events_evlist *data = _data;
-	struct list_head *head;
-	struct parse_events_term *term;
 	struct list_head *list;
-	struct perf_pmu *pmu = NULL;
-	int ok = 0;
 
-	/* Add it for all PMUs that support the alias */
-	ALLOC_LIST(list);
-	while ((pmu = perf_pmu__scan(pmu)) != NULL) {
-		struct perf_pmu_alias *alias;
-
-		list_for_each_entry(alias, &pmu->aliases, list) {
-			if (!strcasecmp(alias->name, $1)) {
-				ALLOC_LIST(head);
-				ABORT_ON(parse_events_term__num(&term, PARSE_EVENTS__TERM_TYPE_USER,
-					$1, 1, &@1, NULL));
-				list_add_tail(&term->list, head);
-
-				if (!parse_events_add_pmu(data, list,
-						  pmu->name, head)) {
-					pr_debug("%s -> %s/%s/\n", $1,
-						 pmu->name, alias->str);
-					ok++;
-				}
-
-				parse_events_terms__delete(head);
-			}
-		}
-	}
-	if (!ok)
+	if (parse_events_multi_pmu_add(_data, $1, &list) < 0)
 		YYABORT;
 	$$ = list;
 }
-- 
2.9.3

Re: [PATCH 03/10] perf, tools: Factor out PMU matching in parser

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:17PM -0800, Andi Kleen wrote:

SNIP

> diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
> index a14b47ab3879..6b2bc918fc94 100644
> --- a/tools/perf/util/parse-events.y
> +++ b/tools/perf/util/parse-events.y
> @@ -236,37 +236,9 @@ PE_NAME opt_event_config
>  |
>  PE_KERNEL_PMU_EVENT sep_dc
>  {
> -	struct parse_events_evlist *data = _data;
> -	struct list_head *head;
> -	struct parse_events_term *term;
>  	struct list_head *list;
> -	struct perf_pmu *pmu = NULL;
> -	int ok = 0;
>  
> -	/* Add it for all PMUs that support the alias */
> -	ALLOC_LIST(list);
> -	while ((pmu = perf_pmu__scan(pmu)) != NULL) {
> -		struct perf_pmu_alias *alias;
> -
> -		list_for_each_entry(alias, &pmu->aliases, list) {
> -			if (!strcasecmp(alias->name, $1)) {
> -				ALLOC_LIST(head);
> -				ABORT_ON(parse_events_term__num(&term, PARSE_EVENTS__TERM_TYPE_USER,
> -					$1, 1, &@1, NULL));
> -				list_add_tail(&term->list, head);

http://marc.info/?l=linux-kernel&m=148818456131560&w=2

jirka

[PATCH 04/10] perf, tools: Expand PMU events by prefix match

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

When the user specifies a pmu directly, expand it automatically
with a prefix match for all available PMUs, similar as we do for
the normal aliases now.

This allows to specify attributes for duplicated boxes quickly.
For example uncore_cbox_{0,6}/.../ can be now specified as uncore_cbox/.../
and it gets automatically expanded for all boxes.

This generally makes it more concise to write uncore specifications, and
also avoids the need to know the exact topology of the system.

Before

% perf stat -a -e uncore_cbox_0/event=0x35,umask=0x1,filter_opc=0x19C/,\
uncore_cbox_1/event=0x35,umask=0x1,filter_opc=0x19C/,\
uncore_cbox_2/event=0x35,umask=0x1,filter_opc=0x19C/,\
uncore_cbox_3/event=0x35,umask=0x1,filter_opc=0x19C/,\
uncore_cbox_4/event=0x35,umask=0x1,filter_opc=0x19C/,\
uncore_cbox_5/event=0x35,umask=0x1,filter_opc=0x19C/ sleep 1

After

% perf stat -a -e uncore_cbox/event=0x35,umask=0x1,filter_opc=0x19C/ sleep 1

v2: Handle all bison rules. Move multi add code to separate function.
Handle uncore_ prefix correctly.
v3: Move parse_events_multi_pmu_add to separate patch. Move uncore
prefix check to separate patch.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/util/parse-events.c | 25 +++++++++++++++++++++++++
 tools/perf/util/parse-events.h |  3 +++
 tools/perf/util/parse-events.y | 40 ++++++++++++++++++++++++++--------------
 3 files changed, 54 insertions(+), 14 deletions(-)

diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 5563c47c92df..d693e2acbda7 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -2450,6 +2450,31 @@ int parse_events_term__clone(struct parse_events_term **new,
 			term->err_term, term->err_val);
 }
 
+int parse_events_copy_term_list(struct list_head *old,
+				 struct list_head **new)
+{
+	struct parse_events_term *term, *n;
+	int ret;
+
+	if (!old) {
+		*new = NULL;
+		return 0;
+	}
+
+	*new = malloc(sizeof(struct list_head));
+	if (!*new)
+		return -ENOMEM;
+	INIT_LIST_HEAD(*new);
+
+	list_for_each_entry (term, old, list) {
+		ret = parse_events_term__clone(&n, term);
+		if (ret)
+			return ret;
+		list_add_tail(&n->list, *new);
+	}
+	return 0;
+}
+
 void parse_events_terms__purge(struct list_head *terms)
 {
 	struct parse_events_term *term, *h;
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index fb755ec248ab..33b3e52cd9b8 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -169,6 +169,9 @@ int parse_events_multi_pmu_add(struct parse_events_evlist *data,
 			       char *str,
 			       struct list_head **listp);
 
+int parse_events_copy_term_list(struct list_head *old,
+				 struct list_head **new);
+
 enum perf_pmu_event_symbol_type
 perf_pmu__parse_check(const char *name);
 void parse_events__set_leader(char *name, struct list_head *list);
diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
index 6b2bc918fc94..3209f18cd5d8 100644
--- a/tools/perf/util/parse-events.y
+++ b/tools/perf/util/parse-events.y
@@ -226,11 +226,32 @@ event_pmu:
 PE_NAME opt_event_config
 {
 	struct parse_events_evlist *data = _data;
-	struct list_head *list;
+	struct list_head *list, *orig_terms, *terms;
+
+	if (parse_events_copy_term_list($2, &orig_terms))
+		YYABORT;
 
 	ALLOC_LIST(list);
-	ABORT_ON(parse_events_add_pmu(data, list, $1, $2));
+	if (parse_events_add_pmu(data, list, $1, $2)) {
+		struct perf_pmu *pmu = NULL;
+		int ok = 0;
+
+		while ((pmu = perf_pmu__scan(pmu)) != NULL) {
+			char *name = pmu->name;
+
+			if (!strncmp($1, name, strlen($1))) {
+				if (parse_events_copy_term_list(orig_terms, &terms))
+					YYABORT;
+				if (!parse_events_add_pmu(data, list, pmu->name, terms))
+					ok++;
+				parse_events_terms__delete(terms);
+			}
+		}
+		if (!ok)
+			YYABORT;
+	}
 	parse_events_terms__delete($2);
+	parse_events_terms__delete(orig_terms);
 	$$ = list;
 }
 |
@@ -245,21 +266,12 @@ PE_KERNEL_PMU_EVENT sep_dc
 |
 PE_PMU_EVENT_PRE '-' PE_PMU_EVENT_SUF sep_dc
 {
-	struct parse_events_evlist *data = _data;
-	struct list_head *head;
-	struct parse_events_term *term;
 	struct list_head *list;
 	char pmu_name[128];
-	snprintf(&pmu_name, 128, "%s-%s", $1, $3);
 
-	ALLOC_LIST(head);
-	ABORT_ON(parse_events_term__num(&term, PARSE_EVENTS__TERM_TYPE_USER,
-					&pmu_name, 1, &@1, NULL));
-	list_add_tail(&term->list, head);
-
-	ALLOC_LIST(list);
-	ABORT_ON(parse_events_add_pmu(data, list, "cpu", head));
-	parse_events_terms__delete(head);
+	snprintf(&pmu_name, 128, "%s-%s", $1, $3);
+	if (parse_events_multi_pmu_add(_data, pmu_name, &list) < 0)
+		YYABORT;
 	$$ = list;
 }
 
-- 
2.9.3

[PATCH 05/10] perf, tools: Special case uncore_ prefix

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Special case uncore_ prefix in PMU match, to allow for shorter event
uncore specifications.

Before

perf stat -a -e uncore_cbox/event=0x35,umask=0x1,filter_opc=0x19C/ sleep 1

After

perf stat -a -e cbox/event=0x35,umask=0x1,filter_opc=0x19C/ sleep 1

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/util/parse-events.y | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
index 3209f18cd5d8..eed371b971a8 100644
--- a/tools/perf/util/parse-events.y
+++ b/tools/perf/util/parse-events.y
@@ -239,6 +239,9 @@ PE_NAME opt_event_config
 		while ((pmu = perf_pmu__scan(pmu)) != NULL) {
 			char *name = pmu->name;
 
+			if (!strncmp(name, "uncore_", 7) &&
+			    strncmp($1, "uncore_", 7))
+				name += 7;
 			if (!strncmp($1, name, strlen($1))) {
 				if (parse_events_copy_term_list(orig_terms, &terms))
 					YYABORT;
-- 
2.9.3

[PATCH 06/10] perf, tools: Add a simple expression parser for JSON

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Add a simple expression parser good enough to parse JSON relation
expressions. The parser is implemented using bison.

This is just intended as an simple parser for internal usage
in the event lists, not the beginning of a "perf scripting language"

v2: Use expr__ prefix instead of expr_
Support multiple free variables for parser
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/tests/Build          |   1 +
 tools/perf/tests/builtin-test.c |   4 +
 tools/perf/tests/expr.c         |  55 +++++++++++++
 tools/perf/tests/tests.h        |   1 +
 tools/perf/util/Build           |   5 ++
 tools/perf/util/expr.h          |  25 ++++++
 tools/perf/util/expr.y          | 173 ++++++++++++++++++++++++++++++++++++++++
 7 files changed, 264 insertions(+)
 create mode 100644 tools/perf/tests/expr.c
 create mode 100644 tools/perf/util/expr.h
 create mode 100644 tools/perf/util/expr.y

diff --git a/tools/perf/tests/Build b/tools/perf/tests/Build
index 1cb3d9b540e9..af58ebc243ef 100644
--- a/tools/perf/tests/Build
+++ b/tools/perf/tests/Build
@@ -38,6 +38,7 @@ perf-y += cpumap.o
 perf-y += stat.o
 perf-y += event_update.o
 perf-y += event-times.o
+perf-y += expr.o
 perf-y += backward-ring-buffer.o
 perf-y += sdt.o
 perf-y += is_printable_array.o
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c
index 37e326bfd2dc..db03e853ba7f 100644
--- a/tools/perf/tests/builtin-test.c
+++ b/tools/perf/tests/builtin-test.c
@@ -44,6 +44,10 @@ static struct test generic_tests[] = {
 		.func = test__parse_events,
 	},
 	{
+		.desc = "Simple expression parser",
+		.func = test__expr,
+	},
+	{
 		.desc = "PERF_RECORD_* events & perf_sample fields",
 		.func = test__PERF_RECORD,
 	},
diff --git a/tools/perf/tests/expr.c b/tools/perf/tests/expr.c
new file mode 100644
index 000000000000..554695c06c5b
--- /dev/null
+++ b/tools/perf/tests/expr.c
@@ -0,0 +1,55 @@
+#include "util/debug.h"
+#include "util/expr.h"
+#include "tests.h"
+
+static int test(struct parse_ctx *ctx, const char *e, double val2)
+{
+	double val;
+
+	if (expr__parse(&val, ctx, &e))
+		TEST_ASSERT_VAL("parse test failed", 0);
+	TEST_ASSERT_VAL("unexpected value", val == val2);
+	return 0;
+}
+
+int test__expr(int subtest __maybe_unused)
+{
+	const char *p;
+	const char **other;
+	double val;
+	int ret;
+	struct parse_ctx ctx;
+	int num_other;
+
+	expr__ctx_init(&ctx);
+	expr__add_id(&ctx, "FOO", 1);
+	expr__add_id(&ctx, "BAR", 2);
+
+	ret = test(&ctx, "1+1", 2);
+	ret |= test(&ctx, "FOO+BAR", 3);
+	ret |= test(&ctx, "(BAR/2)%2", 1);
+	ret |= test(&ctx, "1 - -4",  5);
+	ret |= test(&ctx, "(FOO-1)*2 + (BAR/2)%2 - -4",  5);
+
+	if (ret)
+		return ret;
+
+	p = "FOO/0";
+	ret = expr__parse(&val, &ctx, &p);
+	TEST_ASSERT_VAL("division by zero", ret == 1);
+
+	p = "BAR/";
+	ret = expr__parse(&val, &ctx, &p);
+	TEST_ASSERT_VAL("missing operand", ret == 1);
+
+	TEST_ASSERT_VAL("find other",
+			expr__find_other("FOO + BAR + BAZ + BOZO", "FOO", &other, &num_other) == 0);
+	TEST_ASSERT_VAL("find other", num_other == 3);
+	TEST_ASSERT_VAL("find other", !strcmp(other[0], "BAR"));
+	TEST_ASSERT_VAL("find other", !strcmp(other[1], "BAZ"));
+	TEST_ASSERT_VAL("find other", !strcmp(other[2], "BOZO"));
+	TEST_ASSERT_VAL("find other", other[3] == NULL);
+	free((void *)other);
+
+	return 0;
+}
diff --git a/tools/perf/tests/tests.h b/tools/perf/tests/tests.h
index 1fa9b9d83aa5..631859629403 100644
--- a/tools/perf/tests/tests.h
+++ b/tools/perf/tests/tests.h
@@ -62,6 +62,7 @@ int test__sample_parsing(int subtest);
 int test__keep_tracking(int subtest);
 int test__parse_no_sample_id_all(int subtest);
 int test__dwarf_unwind(int subtest);
+int test__expr(int subtest);
 int test__hists_filter(int subtest);
 int test__mmap_thread_lookup(int subtest);
 int test__thread_mg_share(int subtest);
diff --git a/tools/perf/util/Build b/tools/perf/util/Build
index 5da376bc1afc..3dff5f5960d9 100644
--- a/tools/perf/util/Build
+++ b/tools/perf/util/Build
@@ -88,6 +88,7 @@ libperf-y += mem-events.o
 libperf-y += vsprintf.o
 libperf-y += drv_configs.o
 libperf-y += time-utils.o
+libperf-y += expr-bison.o
 
 libperf-$(CONFIG_LIBBPF) += bpf-loader.o
 libperf-$(CONFIG_BPF_PROLOGUE) += bpf-prologue.o
@@ -140,6 +141,10 @@ $(OUTPUT)util/parse-events-bison.c: util/parse-events.y
 	$(call rule_mkdir)
 	$(Q)$(call echo-cmd,bison)$(BISON) -v util/parse-events.y -d $(PARSER_DEBUG_BISON) -o $@ -p parse_events_
 
+$(OUTPUT)util/expr-bison.c: util/expr.y
+	$(call rule_mkdir)
+	$(Q)$(call echo-cmd,bison)$(BISON) -v util/expr.y -d $(PARSER_DEBUG_BISON) -o $@ -p expr__
+
 $(OUTPUT)util/pmu-flex.c: util/pmu.l $(OUTPUT)util/pmu-bison.c
 	$(call rule_mkdir)
 	$(Q)$(call echo-cmd,flex)$(FLEX) -o $@ --header-file=$(OUTPUT)util/pmu-flex.h util/pmu.l
diff --git a/tools/perf/util/expr.h b/tools/perf/util/expr.h
new file mode 100644
index 000000000000..9c2760a1a96e
--- /dev/null
+++ b/tools/perf/util/expr.h
@@ -0,0 +1,25 @@
+#ifndef PARSE_CTX_H
+#define PARSE_CTX_H 1
+
+#define EXPR_MAX_OTHER 8
+#define MAX_PARSE_ID EXPR_MAX_OTHER
+
+struct parse_id {
+	const char *name;
+	double val;
+};
+
+struct parse_ctx {
+	int num_ids;
+	struct parse_id ids[MAX_PARSE_ID];
+};
+
+void expr__ctx_init(struct parse_ctx *ctx);
+void expr__add_id(struct parse_ctx *ctx, const char *id, double val);
+#ifndef IN_EXPR_Y
+int expr__parse(double *final_val, struct parse_ctx *ctx, const char **pp);
+#endif
+int expr__find_other(const char *p, const char *one, const char ***other,
+		int *num_other);
+
+#endif
diff --git a/tools/perf/util/expr.y b/tools/perf/util/expr.y
new file mode 100644
index 000000000000..a8daa7a93605
--- /dev/null
+++ b/tools/perf/util/expr.y
@@ -0,0 +1,173 @@
+/* Simple expression parser */
+%{
+#include "util.h"
+#include "util/debug.h"
+#define IN_EXPR_Y 1
+#include "expr.h"
+#include <string.h>
+
+#define MAXIDLEN 256
+%}
+
+%define api.pure full
+%parse-param { double *final_val }
+%parse-param { struct parse_ctx *ctx }
+%parse-param { const char **pp }
+%lex-param { const char **pp }
+
+%union {
+	double num;
+	char id[MAXIDLEN+1];
+}
+
+%token <num> NUMBER
+%token <id> ID
+%left '|'
+%left '^'
+%left '&'
+%left '-' '+'
+%left '*' '/' '%'
+%left NEG NOT
+%type <num> expr
+
+%{
+static int expr__lex(YYSTYPE *res, const char **pp);
+
+static void expr__error(double *final_val __maybe_unused,
+		       struct parse_ctx *ctx __maybe_unused,
+		       const char **pp __maybe_unused,
+		       const char *s)
+{
+	pr_debug("%s\n", s);
+}
+
+static int lookup_id(struct parse_ctx *ctx, char *id, double *val)
+{
+	int i;
+
+	for (i = 0; i < ctx->num_ids; i++) {
+		if (!strcasecmp(ctx->ids[i].name, id)) {
+			*val = ctx->ids[i].val;
+			return 0;
+		}
+	}
+	return -1;
+}
+
+%}
+%%
+
+all_expr: expr			{ *final_val = $1; }
+	;
+
+expr:	  NUMBER
+	| ID			{ if (lookup_id(ctx, $1, &$$) < 0) {
+					pr_debug("%s not found", $1);
+					YYABORT;
+				  }
+				}
+	| expr '+' expr		{ $$ = $1 + $3; }
+	| expr '-' expr		{ $$ = $1 - $3; }
+	| expr '*' expr		{ $$ = $1 * $3; }
+	| expr '/' expr		{ if ($3 == 0) YYABORT; $$ = $1 / $3; }
+	| expr '%' expr		{ if ((long)$3 == 0) YYABORT; $$ = (long)$1 % (long)$3; }
+	| '-' expr %prec NEG	{ $$ = -$2; }
+	| '(' expr ')'		{ $$ = $2; }
+	;
+
+%%
+
+static int expr__symbol(YYSTYPE *res, const char *p, const char **pp)
+{
+	char *dst = res->id;
+	const char *s = p;
+
+	while (isalnum(*p) || *p == '_' || *p == '.') {
+		if (p - s >= MAXIDLEN)
+			return -1;
+		*dst++ = *p++;
+	}
+	*dst = 0;
+	*pp = p;
+	return ID;
+}
+
+static int expr__lex(YYSTYPE *res, const char **pp)
+{
+	int tok;
+	const char *s;
+	const char *p = *pp;
+
+	while (isspace(*p))
+		p++;
+	s = p;
+	switch (*p++) {
+	case 'a' ... 'z':
+	case 'A' ... 'Z':
+		return expr__symbol(res, p - 1, pp);
+	case '0' ... '9': case '.':
+		res->num = strtod(s, (char **)&p);
+		tok = NUMBER;
+		break;
+	default:
+		tok = *s;
+		break;
+	}
+	*pp = p;
+	return tok;
+}
+
+/* Caller must make sure id is allocated */
+void expr__add_id(struct parse_ctx *ctx, const char *name, double val)
+{
+	int idx;
+	assert(ctx->num_ids < MAX_PARSE_ID);
+	idx = ctx->num_ids++;
+	ctx->ids[idx].name = name;
+	ctx->ids[idx].val = val;
+}
+
+void expr__ctx_init(struct parse_ctx *ctx)
+{
+	ctx->num_ids = 0;
+}
+
+int expr__find_other(const char *p, const char *one, const char ***other,
+		     int *num_otherp)
+{
+	const char *orig = p;
+	int err = -1;
+	int num_other;
+
+	*other = malloc((EXPR_MAX_OTHER + 1) * sizeof(char *));
+	if (!*other)
+		return -1;
+
+	num_other = 0;
+	for (;;) {
+		YYSTYPE val;
+		int tok = expr__lex(&val, &p);
+		if (tok == 0) {
+			err = 0;
+			break;
+		}
+		if (tok == ID && strcasecmp(one, val.id)) {
+			if (num_other >= EXPR_MAX_OTHER - 1) {
+				pr_debug("Too many extra events in %s\n", orig);
+				break;
+			}
+			(*other)[num_other] = strdup(val.id);
+			if (!(*other)[num_other])
+				return -1;
+			num_other++;
+		}
+	}
+	(*other)[num_other] = NULL;
+	*num_otherp = num_other;
+	if (err) {
+		*num_otherp = 0;
+		free(*other);
+		*other = NULL;
+	}
+	return err;
+}
-- 
2.9.3

[PATCH 07/10] perf, tools: Support MetricExpr header in JSON event list

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Add support for parsing the MetricExpr header in the JSON event lists and
storing them in the alias structure.

Used in the next patch.

v2: Change DividedBy to MetricExpr
v3: Really catch all uses of DividedBy
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/pmu-events/jevents.c    | 18 ++++++++++++++----
 tools/perf/pmu-events/jevents.h    |  2 +-
 tools/perf/pmu-events/pmu-events.h |  1 +
 tools/perf/util/pmu.c              |  9 ++++++---
 tools/perf/util/pmu.h              |  1 +
 5 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/tools/perf/pmu-events/jevents.c b/tools/perf/pmu-events/jevents.c
index eed09346a72a..0735dc2a167a 100644
--- a/tools/perf/pmu-events/jevents.c
+++ b/tools/perf/pmu-events/jevents.c
@@ -291,7 +291,8 @@ static void print_events_table_prefix(FILE *fp, const char *tblname)
 
 static int print_events_table_entry(void *data, char *name, char *event,
 				    char *desc, char *long_desc,
-				    char *pmu, char *unit, char *perpkg)
+				    char *pmu, char *unit, char *perpkg,
+				    char *metric_expr)
 {
 	struct perf_entry_data *pd = data;
 	FILE *outfp = pd->outfp;
@@ -315,6 +316,8 @@ static int print_events_table_entry(void *data, char *name, char *event,
 		fprintf(outfp, "\t.unit = \"%s\",\n", unit);
 	if (perpkg)
 		fprintf(outfp, "\t.perpkg = \"%s\",\n", perpkg);
+	if (metric_expr)
+		fprintf(outfp, "\t.metric_expr = \"%s\",\n", metric_expr);
 	fprintf(outfp, "},\n");
 
 	return 0;
@@ -362,7 +365,8 @@ static char *real_event(const char *name, char *event)
 int json_events(const char *fn,
 	  int (*func)(void *data, char *name, char *event, char *desc,
 		      char *long_desc,
-		      char *pmu, char *unit, char *perpkg),
+		      char *pmu, char *unit, char *perpkg,
+		      char *metric_expr),
 	  void *data)
 {
 	int err = -EIO;
@@ -388,6 +392,7 @@ int json_events(const char *fn,
 		char *filter = NULL;
 		char *perpkg = NULL;
 		char *unit = NULL;
+		char *metric_expr = NULL;
 		unsigned long long eventcode = 0;
 		struct msrmap *msr = NULL;
 		jsmntok_t *msrval = NULL;
@@ -398,6 +403,7 @@ int json_events(const char *fn,
 		for (j = 0; j < obj->size; j += 2) {
 			jsmntok_t *field, *val;
 			int nz;
+			char *s;
 
 			field = tok + j;
 			EXPECT(field->type == JSMN_STRING, tok + j,
@@ -444,7 +450,6 @@ int json_events(const char *fn,
 					NULL);
 			} else if (json_streq(map, field, "Unit")) {
 				const char *ppmu;
-				char *s;
 
 				ppmu = field_to_perf(unit_to_pmu, map, val);
 				if (ppmu) {
@@ -464,6 +469,10 @@ int json_events(const char *fn,
 				addfield(map, &unit, "", "", val);
 			} else if (json_streq(map, field, "PerPkg")) {
 				addfield(map, &perpkg, "", "", val);
+			} else if (json_streq(map, field, "MetricExpr")) {
+				addfield(map, &metric_expr, "", "", val);
+				for (s = metric_expr; *s; s++)
+					*s = tolower(*s);
 			}
 			/* ignore unknown fields */
 		}
@@ -488,7 +497,7 @@ int json_events(const char *fn,
 		fixname(name);
 
 		err = func(data, name, real_event(name, event), desc, long_desc,
-				pmu, unit, perpkg);
+				pmu, unit, perpkg, metric_expr);
 		free(event);
 		free(desc);
 		free(name);
@@ -498,6 +507,7 @@ int json_events(const char *fn,
 		free(filter);
 		free(perpkg);
 		free(unit);
+		free(metric_expr);
 		if (err)
 			break;
 		tok += j;
diff --git a/tools/perf/pmu-events/jevents.h b/tools/perf/pmu-events/jevents.h
index 71e13de31092..57e111bf2168 100644
--- a/tools/perf/pmu-events/jevents.h
+++ b/tools/perf/pmu-events/jevents.h
@@ -5,7 +5,7 @@ int json_events(const char *fn,
 		int (*func)(void *data, char *name, char *event, char *desc,
 				char *long_desc,
 				char *pmu,
-				char *unit, char *perpkg),
+				char *unit, char *perpkg, char *metric_expr),
 		void *data);
 char *get_cpu_str(void);
 
diff --git a/tools/perf/pmu-events/pmu-events.h b/tools/perf/pmu-events/pmu-events.h
index c669a3cdb9f0..d046e3a4ce46 100644
--- a/tools/perf/pmu-events/pmu-events.h
+++ b/tools/perf/pmu-events/pmu-events.h
@@ -13,6 +13,7 @@ struct pmu_event {
 	const char *pmu;
 	const char *unit;
 	const char *perpkg;
+	const char *metric_expr;
 };
 
 /*
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 49bfee0e3d9e..0583e1bdc762 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -231,7 +231,8 @@ static int perf_pmu__parse_snapshot(struct perf_pmu_alias *alias,
 static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
 				 char *desc, char *val,
 				 char *long_desc, char *topic,
-				 char *unit, char *perpkg)
+				 char *unit, char *perpkg,
+				 char *metric_expr)
 {
 	struct perf_pmu_alias *alias;
 	int ret;
@@ -265,6 +266,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
 		perf_pmu__parse_snapshot(alias, dir, name);
 	}
 
+	alias->metric_expr = metric_expr ? strdup(metric_expr) : NULL;
 	alias->desc = desc ? strdup(desc) : NULL;
 	alias->long_desc = long_desc ? strdup(long_desc) :
 				desc ? strdup(desc) : NULL;
@@ -294,7 +296,7 @@ static int perf_pmu__new_alias(struct list_head *list, char *dir, char *name, FI
 	buf[ret] = 0;
 
 	return __perf_pmu__new_alias(list, dir, name, NULL, buf, NULL, NULL, NULL,
-				     NULL);
+				     NULL, NULL);
 }
 
 static inline bool pmu_alias_info_file(char *name)
@@ -564,7 +566,8 @@ static void pmu_add_cpu_aliases(struct list_head *head, const char *name)
 		__perf_pmu__new_alias(head, NULL, (char *)pe->name,
 				(char *)pe->desc, (char *)pe->event,
 				(char *)pe->long_desc, (char *)pe->topic,
-				(char *)pe->unit, (char *)pe->perpkg);
+				(char *)pe->unit, (char *)pe->perpkg,
+				(char *)pe->metric_expr);
 	}
 
 out:
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 00852ddc7741..3dccb15f29e9 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -50,6 +50,7 @@ struct perf_pmu_alias {
 	double scale;
 	bool per_pkg;
 	bool snapshot;
+	char *metric_expr;
 };
 
 struct perf_pmu *perf_pmu__find(const char *name);
-- 
2.9.3

[PATCH 08/10] perf, tools, stat: Output JSON MetricExpr metric

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Add generic infrastructure to perf stat to output ratios for "MetricExpr"
entries in the event lists. Many events are more useful as ratios
than in raw form, typically some count in relation to total ticks.

Transfer the MetricExpr information from the alias to the evsel.

We mark the events that need to be collected for MetricExpr, and also
link the events using them with a pointer. The code is careful
to always prefer the right event in the same group to minimize
multiplexing errors. At the moment only a single relation is supported.

Then add a rblist to the stat shadow code that remembers stats based
on the cpu and context.

Then finally update and retrieve and print these values similarly to the
existing hardcoded perf metrics. We use the simple expression parser
added earlier to evaluate the expression.

Normally we just output the result without further commentary,
but for --metric-only this would lead to empty columns. So for this
case use the original event as description.

So far there is no attempt to automatically add the MetricExpr event,
if it is missing, however we suggest it to the user.

$ perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}'
     1.000228813        800,139,950      unc_p_clockticks
     1.000228813        789,833,783      unc_p_freq_max_os_cycles  #     98.7
     2.000654229        800,308,990      unc_p_clockticks
     2.000654229        396,214,238      unc_p_freq_max_os_cycles  #     49.5

$ perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}' --metric-only
     1.000206740     48.0
     2.000451543     48.1

v2: Change from DivideBy to MetricExpr
v3: Use expr__ prefix.  Support more than one other event.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/builtin-stat.c      |   3 +
 tools/perf/util/evsel.c        |   3 +
 tools/perf/util/evsel.h        |   3 +
 tools/perf/util/parse-events.c |   1 +
 tools/perf/util/pmu.c          |   2 +
 tools/perf/util/pmu.h          |   1 +
 tools/perf/util/stat-shadow.c  | 188 +++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/stat.h         |   2 +
 8 files changed, 203 insertions(+)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 389c8e457bf0..a6cc5ad479b9 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1141,6 +1141,7 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 	out.print_metric = pm;
 	out.new_line = nl;
 	out.ctx = &os;
+	out.force_header = false;
 
 	if (csv_output && !metric_only) {
 		print_noise(counter, noise);
@@ -1478,6 +1479,7 @@ static void print_metric_headers(const char *prefix, bool no_indent)
 		out.ctx = &os;
 		out.print_metric = print_metric_header;
 		out.new_line = new_line_metric;
+		out.force_header = true;
 		os.evsel = counter;
 		perf_stat__print_shadow_stats(counter, 0,
 					      0,
@@ -2460,6 +2462,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	argc = parse_options_subcommand(argc, argv, stat_options, stat_subcommands,
 					(const char **) stat_usage,
 					PARSE_OPT_STOP_AT_NON_OPTION);
+	perf_stat__collect_metric_expr(evsel_list);
 	perf_stat__init_shadow_stats();
 
 	if (csv_sep) {
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index ac59710b79e0..a25dc688ab74 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -236,6 +236,9 @@ void perf_evsel__init(struct perf_evsel *evsel,
 	evsel->sample_size = __perf_evsel__sample_size(attr->sample_type);
 	perf_evsel__calc_id_pos(evsel);
 	evsel->cmdline_group_boundary = false;
+	evsel->metric_expr   = NULL;
+	evsel->metric_events = NULL;
+	evsel->collect_stat  = false;
 }
 
 struct perf_evsel *perf_evsel__new_idx(struct perf_event_attr *attr, int idx)
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index bd2e9b112d49..8f1f61826fdf 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -132,6 +132,9 @@ struct perf_evsel {
 	struct list_head	config_terms;
 	int			bpf_fd;
 	bool			merged_stat;
+	const char *		metric_expr;
+	struct perf_evsel	**metric_events;
+	bool			collect_stat;
 };
 
 union u64_swap {
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index d693e2acbda7..fcb0b6854642 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1254,6 +1254,7 @@ int parse_events_add_pmu(struct parse_events_evlist *data,
 		evsel->scale = info.scale;
 		evsel->per_pkg = info.per_pkg;
 		evsel->snapshot = info.snapshot;
+		evsel->metric_expr = info.metric_expr;
 	}
 
 	return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 0583e1bdc762..4d9632390ab1 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -985,6 +985,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,
 	info->unit     = NULL;
 	info->scale    = 0.0;
 	info->snapshot = false;
+	info->metric_expr = NULL;
 
 	list_for_each_entry_safe(term, h, head_terms, list) {
 		alias = pmu_find_alias(pmu, term);
@@ -1000,6 +1001,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,
 
 		if (alias->per_pkg)
 			info->per_pkg = true;
+		info->metric_expr = alias->metric_expr;
 
 		list_del(&term->list);
 		free(term);
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 3dccb15f29e9..27f078ccc594 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -31,6 +31,7 @@ struct perf_pmu {
 
 struct perf_pmu_info {
 	const char *unit;
+	const char *metric_expr;
 	double scale;
 	bool per_pkg;
 	bool snapshot;
diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
index 8a2bbd2a4d82..e8e76744fa32 100644
--- a/tools/perf/util/stat-shadow.c
+++ b/tools/perf/util/stat-shadow.c
@@ -3,6 +3,9 @@
 #include "stat.h"
 #include "color.h"
 #include "pmu.h"
+#include "rblist.h"
+#include "evlist.h"
+#include "expr.h"
 
 enum {
 	CTX_BIT_USER	= 1 << 0,
@@ -41,13 +44,73 @@ static struct stats runtime_topdown_slots_issued[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_topdown_slots_retired[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_topdown_fetch_bubbles[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_topdown_recovery_bubbles[NUM_CTX][MAX_NR_CPUS];
+static struct rblist runtime_saved_values;
 static bool have_frontend_stalled;
 
 struct stats walltime_nsecs_stats;
 
+struct saved_value {
+	struct rb_node rb_node;
+	struct perf_evsel *evsel;
+	int cpu;
+	int ctx;
+	struct stats stats;
+};
+
+static int saved_value_cmp(struct rb_node *rb_node, const void *entry)
+{
+	struct saved_value *a = container_of(rb_node,
+					     struct saved_value,
+					     rb_node);
+	const struct saved_value *b = entry;
+
+	if (a->ctx != b->ctx)
+		return a->ctx - b->ctx;
+	if (a->cpu != b->cpu)
+		return a->cpu - b->cpu;
+	return a->evsel - b->evsel;
+}
+
+static struct rb_node *saved_value_new(struct rblist *rblist __maybe_unused,
+				     const void *entry)
+{
+	struct saved_value *nd = malloc(sizeof(struct saved_value));
+
+	if (!nd)
+		return NULL;
+	memcpy(nd, entry, sizeof(struct saved_value));
+	return &nd->rb_node;
+}
+
+static struct saved_value *saved_value_lookup(struct perf_evsel *evsel,
+					      int cpu, int ctx,
+					      bool create)
+{
+	struct rb_node *nd;
+	struct saved_value dm = {
+		.cpu = cpu,
+		.ctx = ctx,
+		.evsel = evsel,
+	};
+	nd = rblist__find(&runtime_saved_values, &dm);
+	if (nd)
+		return container_of(nd, struct saved_value, rb_node);
+	if (create) {
+		rblist__add_node(&runtime_saved_values, &dm);
+		nd = rblist__find(&runtime_saved_values, &dm);
+		if (nd)
+			return container_of(nd, struct saved_value, rb_node);
+	}
+	return NULL;
+}
+
 void perf_stat__init_shadow_stats(void)
 {
 	have_frontend_stalled = pmu_have_event("cpu", "stalled-cycles-frontend");
+	rblist__init(&runtime_saved_values);
+	runtime_saved_values.node_cmp = saved_value_cmp;
+	runtime_saved_values.node_new = saved_value_new;
+	/* No delete for now */
 }
 
 static int evsel_context(struct perf_evsel *evsel)
@@ -70,6 +133,8 @@ static int evsel_context(struct perf_evsel *evsel)
 
 void perf_stat__reset_shadow_stats(void)
 {
+	struct rb_node *pos, *next;
+
 	memset(runtime_nsecs_stats, 0, sizeof(runtime_nsecs_stats));
 	memset(runtime_cycles_stats, 0, sizeof(runtime_cycles_stats));
 	memset(runtime_stalled_cycles_front_stats, 0, sizeof(runtime_stalled_cycles_front_stats));
@@ -92,6 +157,15 @@ void perf_stat__reset_shadow_stats(void)
 	memset(runtime_topdown_slots_issued, 0, sizeof(runtime_topdown_slots_issued));
 	memset(runtime_topdown_fetch_bubbles, 0, sizeof(runtime_topdown_fetch_bubbles));
 	memset(runtime_topdown_recovery_bubbles, 0, sizeof(runtime_topdown_recovery_bubbles));
+
+	next = rb_first(&runtime_saved_values.entries);
+	while (next) {
+		pos = next;
+		next = rb_next(pos);
+		memset(&container_of(pos, struct saved_value, rb_node)->stats,
+		       0,
+		       sizeof(struct stats));
+	}
 }
 
 /*
@@ -143,6 +217,12 @@ void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
 		update_stats(&runtime_dtlb_cache_stats[ctx][cpu], count[0]);
 	else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
 		update_stats(&runtime_itlb_cache_stats[ctx][cpu], count[0]);
+
+	if (counter->collect_stat) {
+		struct saved_value *v = saved_value_lookup(counter, cpu, ctx,
+							   true);
+		update_stats(&v->stats, count[0]);
+	}
 }
 
 /* used for get_ratio_color() */
@@ -172,6 +252,88 @@ static const char *get_ratio_color(enum grc_type type, double ratio)
 	return color;
 }
 
+static struct perf_evsel *perf_stat__find_event(struct perf_evlist *evsel_list,
+						const char *name)
+{
+	struct perf_evsel *c2;
+
+	evlist__for_each_entry (evsel_list, c2) {
+		if (!strcasecmp(c2->name, name))
+			return c2;
+	}
+	return NULL;
+}
+
+/* Mark MetricExpr target events and link events using them to them. */
+void perf_stat__collect_metric_expr(struct perf_evlist *evsel_list)
+{
+	struct perf_evsel *counter, *leader, **metric_events, *oc;
+	bool found;
+	const char **metric_names;
+	int i;
+	int num_metric_names;
+
+	evlist__for_each_entry(evsel_list, counter) {
+		bool invalid = false;
+
+		leader = counter->leader;
+		if (!counter->metric_expr)
+			continue;
+		metric_events = counter->metric_events;
+		if (!metric_events) {
+			if (expr__find_other(counter->metric_expr, counter->name,
+						&metric_names, &num_metric_names) < 0)
+				continue;
+
+			metric_events = calloc(sizeof(struct perf_evsel *),
+					       num_metric_names + 1);
+			if (!metric_events)
+				return;
+			counter->metric_events = metric_events;
+		}
+
+		for (i = 0; i < num_metric_names; i++) {
+			found = false;
+			if (leader) {
+				/* Search in group */
+				for_each_group_member (oc, leader) {
+					if (!strcasecmp(oc->name, metric_names[i])) {
+						found = true;
+						break;
+					}
+				}
+			}
+			if (!found) {
+				/* Search ignoring groups */
+				oc = perf_stat__find_event(evsel_list, metric_names[i]);
+			}
+			if (!oc) {
+				/*
+				 * Adding events automatically would be difficult, because
+				 * it would risk creating groups that are not schedulable.
+				 * perf stat doesn't understand all the scheduling constraints
+				 * of events. So we ask the user instead to add the missing
+				 * events.
+				 */
+				fprintf(stderr, "Add %s event to groups to get metric expression for %s\n",
+						metric_names[i],
+						counter->name);
+				invalid = true;
+				continue;
+			}
+			metric_events[i] = oc;
+			oc->collect_stat = true;
+		}
+		metric_events[i] = NULL;
+		free(metric_names);
+		if (invalid) {
+			free(metric_events);
+			counter->metric_events = NULL;
+			counter->metric_expr = NULL;
+		}
+	}
+}
+
 static void print_stalled_cycles_frontend(int cpu,
 					  struct perf_evsel *evsel, double avg,
 					  struct perf_stat_output_ctx *out)
@@ -614,6 +776,32 @@ void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 					be_bound * 100.);
 		else
 			print_metric(ctxp, NULL, NULL, name, 0);
+	} else if (evsel->metric_expr) {
+		struct parse_ctx pctx;
+		int i;
+
+		expr__ctx_init(&pctx);
+		expr__add_id(&pctx, evsel->name, avg);
+		for (i = 0; evsel->metric_events[i]; i++) {
+			struct saved_value *v;
+
+			v = saved_value_lookup(evsel->metric_events[i], cpu, ctx, false);
+			if (!v)
+				break;
+			expr__add_id(&pctx, evsel->metric_events[i]->name,
+					     avg_stats(&v->stats));
+		}
+		if (!evsel->metric_events[i]) {
+			const char *p = evsel->metric_expr;
+
+			if (expr__parse(&ratio, &pctx, &p) == 0)
+				print_metric(ctxp, NULL, "%8.1f",
+					out->force_header ? evsel->name : "",
+					ratio);
+			else
+				print_metric(ctxp, NULL, NULL, "", 0);
+		} else
+			print_metric(ctxp, NULL, NULL, "", 0);
 	} else if (runtime_nsecs_stats[cpu].n != 0) {
 		char unit = 'M';
 		char unit_buf[10];
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index c29bb94c48a4..0a65ae23f495 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -85,11 +85,13 @@ struct perf_stat_output_ctx {
 	void *ctx;
 	print_metric_t print_metric;
 	new_line_t new_line;
+	bool force_header;
 };
 
 void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 				   double avg, int cpu,
 				   struct perf_stat_output_ctx *out);
+void perf_stat__collect_metric_expr(struct perf_evlist *);
 
 int perf_evlist__alloc_stats(struct perf_evlist *evlist, bool alloc_raw);
 void perf_evlist__free_stats(struct perf_evlist *evlist);
-- 
2.9.3

Re: [PATCH 08/10] perf, tools, stat: Output JSON MetricExpr metric

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:22PM -0800, Andi Kleen wrote:
> From: Andi Kleen <ak@linux.intel.com>
> 
> Add generic infrastructure to perf stat to output ratios for "MetricExpr"
> entries in the event lists. Many events are more useful as ratios
> than in raw form, typically some count in relation to total ticks.
> 
> Transfer the MetricExpr information from the alias to the evsel.
> 
> We mark the events that need to be collected for MetricExpr, and also
> link the events using them with a pointer. The code is careful
> to always prefer the right event in the same group to minimize
> multiplexing errors. At the moment only a single relation is supported.
> 
> Then add a rblist to the stat shadow code that remembers stats based
> on the cpu and context.
> 
> Then finally update and retrieve and print these values similarly to the
> existing hardcoded perf metrics. We use the simple expression parser
> added earlier to evaluate the expression.
> 
> Normally we just output the result without further commentary,
> but for --metric-only this would lead to empty columns. So for this
> case use the original event as description.
> 
> So far there is no attempt to automatically add the MetricExpr event,
> if it is missing, however we suggest it to the user.
> 
> $ perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}'
>      1.000228813        800,139,950      unc_p_clockticks
>      1.000228813        789,833,783      unc_p_freq_max_os_cycles  #     98.7
>      2.000654229        800,308,990      unc_p_clockticks
>      2.000654229        396,214,238      unc_p_freq_max_os_cycles  #     49.5
> 
> $ perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}' --metric-only
>      1.000206740     48.0
>      2.000451543     48.1

The comment above suggest I'll see 'original event as description' in this example??

However, could you just make those metric names working in here?
Displaying original event for metric value is misleading.

jirka

[PATCH 09/10] perf, tools, list: Support printing MetricExpr with -v

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Output the metric expr in perf list when -v is specified, so that the user
can check the formula.

Before:

% perf list -v
  ...
  unc_m_power_channel_ppd
       [Cycles where DRAM ranks are in power down (CKE) mode. Derived from unc_m_power_channel_ppd. Unit:
        uncore_imc]
        uncore_imc_2/event=0x85/

After:

% perf list -v
  ...
  unc_m_power_channel_ppd
       [Cycles where DRAM ranks are in power down (CKE) mode. Derived from unc_m_power_channel_ppd. Unit:
        uncore_imc]
        Perf: uncore_imc_2/event=0x85/ MetricExpr: (unc_m_power_channel_ppd / unc_m_clockticks) * 100.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/util/pmu.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 4d9632390ab1..2cb463ababb8 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -1096,6 +1096,7 @@ struct sevent {
 	char *topic;
 	char *str;
 	char *pmu;
+	char *metric_expr;
 };
 
 static int cmp_sevent(const void *a, const void *b)
@@ -1194,6 +1195,7 @@ void print_pmu_events(const char *event_glob, bool name_only, bool quiet_flag,
 			aliases[j].topic = alias->topic;
 			aliases[j].str = alias->str;
 			aliases[j].pmu = pmu->name;
+			aliases[j].metric_expr = alias->metric_expr;
 			j++;
 		}
 		if (pmu->selectable &&
@@ -1228,8 +1230,12 @@ void print_pmu_events(const char *event_glob, bool name_only, bool quiet_flag,
 			printf("%*s", 8, "[");
 			wordwrap(aliases[j].desc, 8, columns, 0);
 			printf("]\n");
-			if (verbose)
-				printf("%*s%s/%s/\n", 8, "", aliases[j].pmu, aliases[j].str);
+			if (verbose) {
+				printf("%*sPerf: %s/%s/", 8, "", aliases[j].pmu, aliases[j].str);
+				if (aliases[j].metric_expr)
+					printf(" MetricExpr: %s", aliases[j].metric_expr);
+				putchar('\n');
+			}
 		} else
 			printf("  %-50s [Kernel PMU event]\n", aliases[j].name);
 		printed++;
-- 
2.9.3

[PATCH 10/10] perf, tools: Add support for MetricName JSON attribute

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Add support for a new JSON event attribute to name MetricExpr for better output
in perf stat.
If the event has no MetricName it uses the normal event name instead to describe
the metric.

So far no names are added.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/pmu-events/jevents.c    | 14 +++++++++++---
 tools/perf/pmu-events/jevents.h    |  3 ++-
 tools/perf/pmu-events/pmu-events.h |  1 +
 tools/perf/util/evsel.c            |  1 +
 tools/perf/util/evsel.h            |  1 +
 tools/perf/util/parse-events.c     |  1 +
 tools/perf/util/pmu.c              | 15 ++++++++++++---
 tools/perf/util/pmu.h              |  2 ++
 tools/perf/util/stat-shadow.c      |  4 +++-
 9 files changed, 34 insertions(+), 8 deletions(-)

diff --git a/tools/perf/pmu-events/jevents.c b/tools/perf/pmu-events/jevents.c
index 0735dc2a167a..81f2ef3b15cf 100644
--- a/tools/perf/pmu-events/jevents.c
+++ b/tools/perf/pmu-events/jevents.c
@@ -292,7 +292,8 @@ static void print_events_table_prefix(FILE *fp, const char *tblname)
 static int print_events_table_entry(void *data, char *name, char *event,
 				    char *desc, char *long_desc,
 				    char *pmu, char *unit, char *perpkg,
-				    char *metric_expr)
+				    char *metric_expr,
+				    char *metric_name)
 {
 	struct perf_entry_data *pd = data;
 	FILE *outfp = pd->outfp;
@@ -318,6 +319,8 @@ static int print_events_table_entry(void *data, char *name, char *event,
 		fprintf(outfp, "\t.perpkg = \"%s\",\n", perpkg);
 	if (metric_expr)
 		fprintf(outfp, "\t.metric_expr = \"%s\",\n", metric_expr);
+	if (metric_name)
+		fprintf(outfp, "\t.metric_name = \"%s\",\n", metric_name);
 	fprintf(outfp, "},\n");
 
 	return 0;
@@ -366,7 +369,8 @@ int json_events(const char *fn,
 	  int (*func)(void *data, char *name, char *event, char *desc,
 		      char *long_desc,
 		      char *pmu, char *unit, char *perpkg,
-		      char *metric_expr),
+		      char *metric_expr,
+		      char *metric_name),
 	  void *data)
 {
 	int err = -EIO;
@@ -393,6 +397,7 @@ int json_events(const char *fn,
 		char *perpkg = NULL;
 		char *unit = NULL;
 		char *metric_expr = NULL;
+		char *metric_name = NULL;
 		unsigned long long eventcode = 0;
 		struct msrmap *msr = NULL;
 		jsmntok_t *msrval = NULL;
@@ -469,6 +474,8 @@ int json_events(const char *fn,
 				addfield(map, &unit, "", "", val);
 			} else if (json_streq(map, field, "PerPkg")) {
 				addfield(map, &perpkg, "", "", val);
+			} else if (json_streq(map, field, "MetricName")) {
+				addfield(map, &metric_name, "", "", val);
 			} else if (json_streq(map, field, "MetricExpr")) {
 				addfield(map, &metric_expr, "", "", val);
 				for (s = metric_expr; *s; s++)
@@ -497,7 +504,7 @@ int json_events(const char *fn,
 		fixname(name);
 
 		err = func(data, name, real_event(name, event), desc, long_desc,
-				pmu, unit, perpkg, metric_expr);
+				pmu, unit, perpkg, metric_expr, metric_name);
 		free(event);
 		free(desc);
 		free(name);
@@ -508,6 +515,7 @@ int json_events(const char *fn,
 		free(perpkg);
 		free(unit);
 		free(metric_expr);
+		free(metric_name);
 		if (err)
 			break;
 		tok += j;
diff --git a/tools/perf/pmu-events/jevents.h b/tools/perf/pmu-events/jevents.h
index 57e111bf2168..611fac01913d 100644
--- a/tools/perf/pmu-events/jevents.h
+++ b/tools/perf/pmu-events/jevents.h
@@ -5,7 +5,8 @@ int json_events(const char *fn,
 		int (*func)(void *data, char *name, char *event, char *desc,
 				char *long_desc,
 				char *pmu,
-				char *unit, char *perpkg, char *metric_expr),
+				char *unit, char *perpkg, char *metric_expr,
+				char *metric_name),
 		void *data);
 char *get_cpu_str(void);
 
diff --git a/tools/perf/pmu-events/pmu-events.h b/tools/perf/pmu-events/pmu-events.h
index d046e3a4ce46..569eab3688dd 100644
--- a/tools/perf/pmu-events/pmu-events.h
+++ b/tools/perf/pmu-events/pmu-events.h
@@ -14,6 +14,7 @@ struct pmu_event {
 	const char *unit;
 	const char *perpkg;
 	const char *metric_expr;
+	const char *metric_name;
 };
 
 /*
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index a25dc688ab74..c16a1e1253a2 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -237,6 +237,7 @@ void perf_evsel__init(struct perf_evsel *evsel,
 	perf_evsel__calc_id_pos(evsel);
 	evsel->cmdline_group_boundary = false;
 	evsel->metric_expr   = NULL;
+	evsel->metric_name   = NULL;
 	evsel->metric_events = NULL;
 	evsel->collect_stat  = false;
 }
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 8f1f61826fdf..d101695c482c 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -133,6 +133,7 @@ struct perf_evsel {
 	int			bpf_fd;
 	bool			merged_stat;
 	const char *		metric_expr;
+	const char *		metric_name;
 	struct perf_evsel	**metric_events;
 	bool			collect_stat;
 };
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index fcb0b6854642..4b063ff01410 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1255,6 +1255,7 @@ int parse_events_add_pmu(struct parse_events_evlist *data,
 		evsel->per_pkg = info.per_pkg;
 		evsel->snapshot = info.snapshot;
 		evsel->metric_expr = info.metric_expr;
+		evsel->metric_name = info.metric_name;
 	}
 
 	return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 2cb463ababb8..633530fbfe1e 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -232,7 +232,8 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
 				 char *desc, char *val,
 				 char *long_desc, char *topic,
 				 char *unit, char *perpkg,
-				 char *metric_expr)
+				 char *metric_expr,
+				 char *metric_name)
 {
 	struct perf_pmu_alias *alias;
 	int ret;
@@ -267,6 +268,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
 	}
 
 	alias->metric_expr = metric_expr ? strdup(metric_expr) : NULL;
+	alias->metric_name = metric_name ? strdup(metric_name): NULL;
 	alias->desc = desc ? strdup(desc) : NULL;
 	alias->long_desc = long_desc ? strdup(long_desc) :
 				desc ? strdup(desc) : NULL;
@@ -296,7 +298,7 @@ static int perf_pmu__new_alias(struct list_head *list, char *dir, char *name, FI
 	buf[ret] = 0;
 
 	return __perf_pmu__new_alias(list, dir, name, NULL, buf, NULL, NULL, NULL,
-				     NULL, NULL);
+				     NULL, NULL, NULL);
 }
 
 static inline bool pmu_alias_info_file(char *name)
@@ -567,7 +569,8 @@ static void pmu_add_cpu_aliases(struct list_head *head, const char *name)
 				(char *)pe->desc, (char *)pe->event,
 				(char *)pe->long_desc, (char *)pe->topic,
 				(char *)pe->unit, (char *)pe->perpkg,
-				(char *)pe->metric_expr);
+				(char *)pe->metric_expr,
+				(char *)pe->metric_name);
 	}
 
 out:
@@ -986,6 +989,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,
 	info->scale    = 0.0;
 	info->snapshot = false;
 	info->metric_expr = NULL;
+	info->metric_name = NULL;
 
 	list_for_each_entry_safe(term, h, head_terms, list) {
 		alias = pmu_find_alias(pmu, term);
@@ -1002,6 +1006,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,
 		if (alias->per_pkg)
 			info->per_pkg = true;
 		info->metric_expr = alias->metric_expr;
+		info->metric_name = alias->metric_name;
 
 		list_del(&term->list);
 		free(term);
@@ -1097,6 +1102,7 @@ struct sevent {
 	char *str;
 	char *pmu;
 	char *metric_expr;
+	char *metric_name;
 };
 
 static int cmp_sevent(const void *a, const void *b)
@@ -1196,6 +1202,7 @@ void print_pmu_events(const char *event_glob, bool name_only, bool quiet_flag,
 			aliases[j].str = alias->str;
 			aliases[j].pmu = pmu->name;
 			aliases[j].metric_expr = alias->metric_expr;
+			aliases[j].metric_name = alias->metric_name;
 			j++;
 		}
 		if (pmu->selectable &&
@@ -1232,6 +1239,8 @@ void print_pmu_events(const char *event_glob, bool name_only, bool quiet_flag,
 			printf("]\n");
 			if (verbose) {
 				printf("%*sPerf: %s/%s/", 8, "", aliases[j].pmu, aliases[j].str);
+				if (aliases[j].metric_name)
+					printf(" MetricName: %s", aliases[j].metric_name);
 				if (aliases[j].metric_expr)
 					printf(" MetricExpr: %s", aliases[j].metric_expr);
 				putchar('\n');
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 27f078ccc594..3d4b703f5d89 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -32,6 +32,7 @@ struct perf_pmu {
 struct perf_pmu_info {
 	const char *unit;
 	const char *metric_expr;
+	const char *metric_name;
 	double scale;
 	bool per_pkg;
 	bool snapshot;
@@ -52,6 +53,7 @@ struct perf_pmu_alias {
 	bool per_pkg;
 	bool snapshot;
 	char *metric_expr;
+	char *metric_name;
 };
 
 struct perf_pmu *perf_pmu__find(const char *name);
diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
index e8e76744fa32..6883f429c742 100644
--- a/tools/perf/util/stat-shadow.c
+++ b/tools/perf/util/stat-shadow.c
@@ -796,7 +796,9 @@ void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 
 			if (expr__parse(&ratio, &pctx, &p) == 0)
 				print_metric(ctxp, NULL, "%8.1f",
-					out->force_header ? evsel->name : "",
+					evsel->metric_name ?
+					evsel->metric_name :
+					out->force_header ?  evsel->name : "",
 					ratio);
 			else
 				print_metric(ctxp, NULL, NULL, "", 0);
-- 
2.9.3

Re: [PATCH 10/10] perf, tools: Add support for MetricName JSON attribute

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:24PM -0800, Andi Kleen wrote:
> From: Andi Kleen <ak@linux.intel.com>
> 
> Add support for a new JSON event attribute to name MetricExpr for better output
> in perf stat.
> If the event has no MetricName it uses the normal event name instead to describe
> the metric.
> 
> So far no names are added.

when do you plan to add them?

jirka

Re: perf: Improve support for uncore JSON event lists

[Jiri Olsa]

On Tue, Feb 28, 2017 at 10:49:14PM -0800, Andi Kleen wrote:
> This patch kit further improves support for Intel uncore events in
> the Linux perf user tool. The basic support has been already
> merged earlier, but this makes it nicer to use.
> 
> - Collapse counts from duplicated boxes to make the output
> easier to read.
> - Support specifying events for multiple duplicated boxes
> in an abbreviated format to shorten event specifiers
> - Add support for computing Metrics defined in the event lists,
> so that the event lists can extend the metrics in perf stat.
> This allows to represent many events in an easier to understand
> format.
> 
> Available from
> 
> git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc.git perf/builtin-json-28
> 
> v1: Initial post after being split off to own patchkit
> Adding MetricName support and support for more than two events
> in expressions.
> 
> v2: Address review comments. Move new hunk from refactor
> patchkit to patch adding new features. Improve changelogs
> slightly.
> 

with my comments addressed, I'll be ok to ack it in next version

thanks,
jirka

[PATCH 01/10] perf, tools, stat: Factor out callback for collecting event values

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

To be used in next patch to support automatic summing of alias
events.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/builtin-stat.c | 114 ++++++++++++++++++++++++++++++++++++----------
 1 file changed, 91 insertions(+), 23 deletions(-)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index f28719178b51..43a7ef3d71ed 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1178,11 +1178,57 @@ static void aggr_update_shadow(void)
 	}
 }
 
+static void collect_data(struct perf_evsel *counter,
+			    void (*cb)(struct perf_evsel *counter, void *data,
+				       bool first),
+			    void *data)
+{
+	cb(counter, data, true);
+}
+
+struct aggr_data {
+	u64 ena, run, val;
+	int id;
+	int nr;
+	int cpu;
+};
+
+static void aggr_cb(struct perf_evsel *counter, void *data, bool first)
+{
+	struct aggr_data *ad = data;
+	int cpu, cpu2, s2;
+
+	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
+		struct perf_counts_values *counts;
+
+		cpu2 = perf_evsel__cpus(counter)->map[cpu];
+		s2 = aggr_get_id(evsel_list->cpus, cpu2);
+		if (s2 != ad->id)
+			continue;
+		if (first)
+			ad->nr++;
+		counts = perf_counts(counter->counts, cpu, 0);
+		/*
+		 * When any result is bad, make them all to give
+		 * consistent output in interval mode.
+		 */
+		if (counts->ena == 0 || counts->run == 0 ||
+		    counter->counts->scaled == -1) {
+			ad->ena = 0;
+			ad->run = 0;
+			break;
+		}
+		ad->val += counts->val;
+		ad->ena += counts->ena;
+		ad->run += counts->run;
+	}
+}
+
 static void print_aggr(char *prefix)
 {
 	FILE *output = stat_config.output;
 	struct perf_evsel *counter;
-	int cpu, s, s2, id, nr;
+	int s, id, nr;
 	double uval;
 	u64 ena, run, val;
 	bool first;
@@ -1197,23 +1243,20 @@ static void print_aggr(char *prefix)
 	 * Without each counter has its own line.
 	 */
 	for (s = 0; s < aggr_map->nr; s++) {
+		struct aggr_data ad;
 		if (prefix && metric_only)
 			fprintf(output, "%s", prefix);
 
-		id = aggr_map->map[s];
+		ad.id = id = aggr_map->map[s];
 		first = true;
 		evlist__for_each_entry(evsel_list, counter) {
-			val = ena = run = 0;
-			nr = 0;
-			for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
-				s2 = aggr_get_id(perf_evsel__cpus(counter), cpu);
-				if (s2 != id)
-					continue;
-				val += perf_counts(counter->counts, cpu, 0)->val;
-				ena += perf_counts(counter->counts, cpu, 0)->ena;
-				run += perf_counts(counter->counts, cpu, 0)->run;
-				nr++;
-			}
+			ad.val = ad.ena = ad.run = 0;
+			ad.nr = 0;
+			collect_data(counter, aggr_cb, &ad);
+			nr = ad.nr;
+			ena = ad.ena;
+			run = ad.run;
+			val = ad.val;
 			if (first && metric_only) {
 				first = false;
 				aggr_printout(counter, id, nr);
@@ -1257,6 +1300,21 @@ static void print_aggr_thread(struct perf_evsel *counter, char *prefix)
 	}
 }
 
+struct caggr_data {
+	double avg, avg_enabled, avg_running;
+};
+
+static void counter_aggr_cb(struct perf_evsel *counter, void *data,
+			    bool first __maybe_unused)
+{
+	struct caggr_data *cd = data;
+	struct perf_stat_evsel *ps = counter->priv;
+
+	cd->avg += avg_stats(&ps->res_stats[0]);
+	cd->avg_enabled += avg_stats(&ps->res_stats[1]);
+	cd->avg_running += avg_stats(&ps->res_stats[2]);
+}
+
 /*
  * Print out the results of a single counter:
  * aggregated counts in system-wide mode
@@ -1264,23 +1322,30 @@ static void print_aggr_thread(struct perf_evsel *counter, char *prefix)
 static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
 {
 	FILE *output = stat_config.output;
-	struct perf_stat_evsel *ps = counter->priv;
-	double avg = avg_stats(&ps->res_stats[0]);
 	double uval;
-	double avg_enabled, avg_running;
+	struct caggr_data cd = { .avg = 0.0 };
 
-	avg_enabled = avg_stats(&ps->res_stats[1]);
-	avg_running = avg_stats(&ps->res_stats[2]);
+	collect_data(counter, counter_aggr_cb, &cd);
 
 	if (prefix && !metric_only)
 		fprintf(output, "%s", prefix);
 
-	uval = avg * counter->scale;
-	printout(-1, 0, counter, uval, prefix, avg_running, avg_enabled, avg);
+	uval = cd.avg * counter->scale;
+	printout(-1, 0, counter, uval, prefix, cd.avg_running, cd.avg_enabled, cd.avg);
 	if (!metric_only)
 		fprintf(output, "\n");
 }
 
+static void counter_cb(struct perf_evsel *counter, void *data,
+		       bool first __maybe_unused)
+{
+	struct aggr_data *ad = data;
+
+	ad->val += perf_counts(counter->counts, ad->cpu, 0)->val;
+	ad->ena += perf_counts(counter->counts, ad->cpu, 0)->ena;
+	ad->run += perf_counts(counter->counts, ad->cpu, 0)->run;
+}
+
 /*
  * Print out the results of a single counter:
  * does not use aggregated count in system-wide
@@ -1293,9 +1358,12 @@ static void print_counter(struct perf_evsel *counter, char *prefix)
 	int cpu;
 
 	for (cpu = 0; cpu < perf_evsel__nr_cpus(counter); cpu++) {
-		val = perf_counts(counter->counts, cpu, 0)->val;
-		ena = perf_counts(counter->counts, cpu, 0)->ena;
-		run = perf_counts(counter->counts, cpu, 0)->run;
+		struct aggr_data ad = { .cpu = cpu };
+
+		collect_data(counter, counter_cb, &ad);
+		val = ad.val;
+		ena = ad.ena;
+		run = ad.run;
 
 		if (prefix)
 			fprintf(output, "%s", prefix);
-- 
2.9.3

[PATCH 08/10] perf, tools, stat: Output JSON MetricExpr metric

[Andi Kleen]

From: Andi Kleen <ak@linux.intel.com>

Add generic infrastructure to perf stat to output ratios for "MetricExpr"
entries in the event lists. Many events are more useful as ratios
than in raw form, typically some count in relation to total ticks.

Transfer the MetricExpr information from the alias to the evsel.

We mark the events that need to be collected for MetricExpr, and also
link the events using them with a pointer. The code is careful
to always prefer the right event in the same group to minimize
multiplexing errors. At the moment only a single relation is supported.

Then add a rblist to the stat shadow code that remembers stats based
on the cpu and context.

Then finally update and retrieve and print these values similarly to the
existing hardcoded perf metrics. We use the simple expression parser
added earlier to evaluate the expression.

Normally we just output the result without further commentary,
but for --metric-only this would lead to empty columns. So for this
case use the original event as description.

So far there is no attempt to automatically add the MetricExpr event,
if it is missing, however we suggest it to the user.

$ perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}'
     1.000228813        800,139,950      unc_p_clockticks
     1.000228813        789,833,783      unc_p_freq_max_os_cycles  #     98.7
     2.000654229        800,308,990      unc_p_clockticks
     2.000654229        396,214,238      unc_p_freq_max_os_cycles  #     49.5

$ perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}' --metric-only
     1.000206740     48.0
     2.000451543     48.1

v2: Change from DivideBy to MetricExpr
v3: Use expr__ prefix.  Support more than one other event.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/builtin-stat.c      |   3 +
 tools/perf/util/evsel.c        |   3 +
 tools/perf/util/evsel.h        |   3 +
 tools/perf/util/parse-events.c |   1 +
 tools/perf/util/pmu.c          |   2 +
 tools/perf/util/pmu.h          |   1 +
 tools/perf/util/stat-shadow.c  | 188 +++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/stat.h         |   2 +
 8 files changed, 203 insertions(+)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 389c8e457bf0..a6cc5ad479b9 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1141,6 +1141,7 @@ static void printout(int id, int nr, struct perf_evsel *counter, double uval,
 	out.print_metric = pm;
 	out.new_line = nl;
 	out.ctx = &os;
+	out.force_header = false;
 
 	if (csv_output && !metric_only) {
 		print_noise(counter, noise);
@@ -1478,6 +1479,7 @@ static void print_metric_headers(const char *prefix, bool no_indent)
 		out.ctx = &os;
 		out.print_metric = print_metric_header;
 		out.new_line = new_line_metric;
+		out.force_header = true;
 		os.evsel = counter;
 		perf_stat__print_shadow_stats(counter, 0,
 					      0,
@@ -2460,6 +2462,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	argc = parse_options_subcommand(argc, argv, stat_options, stat_subcommands,
 					(const char **) stat_usage,
 					PARSE_OPT_STOP_AT_NON_OPTION);
+	perf_stat__collect_metric_expr(evsel_list);
 	perf_stat__init_shadow_stats();
 
 	if (csv_sep) {
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index ac59710b79e0..a25dc688ab74 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -236,6 +236,9 @@ void perf_evsel__init(struct perf_evsel *evsel,
 	evsel->sample_size = __perf_evsel__sample_size(attr->sample_type);
 	perf_evsel__calc_id_pos(evsel);
 	evsel->cmdline_group_boundary = false;
+	evsel->metric_expr   = NULL;
+	evsel->metric_events = NULL;
+	evsel->collect_stat  = false;
 }
 
 struct perf_evsel *perf_evsel__new_idx(struct perf_event_attr *attr, int idx)
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index bd2e9b112d49..8f1f61826fdf 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -132,6 +132,9 @@ struct perf_evsel {
 	struct list_head	config_terms;
 	int			bpf_fd;
 	bool			merged_stat;
+	const char *		metric_expr;
+	struct perf_evsel	**metric_events;
+	bool			collect_stat;
 };
 
 union u64_swap {
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index d693e2acbda7..fcb0b6854642 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1254,6 +1254,7 @@ int parse_events_add_pmu(struct parse_events_evlist *data,
 		evsel->scale = info.scale;
 		evsel->per_pkg = info.per_pkg;
 		evsel->snapshot = info.snapshot;
+		evsel->metric_expr = info.metric_expr;
 	}
 
 	return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 0583e1bdc762..4d9632390ab1 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -985,6 +985,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,
 	info->unit     = NULL;
 	info->scale    = 0.0;
 	info->snapshot = false;
+	info->metric_expr = NULL;
 
 	list_for_each_entry_safe(term, h, head_terms, list) {
 		alias = pmu_find_alias(pmu, term);
@@ -1000,6 +1001,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,
 
 		if (alias->per_pkg)
 			info->per_pkg = true;
+		info->metric_expr = alias->metric_expr;
 
 		list_del(&term->list);
 		free(term);
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 3dccb15f29e9..27f078ccc594 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -31,6 +31,7 @@ struct perf_pmu {
 
 struct perf_pmu_info {
 	const char *unit;
+	const char *metric_expr;
 	double scale;
 	bool per_pkg;
 	bool snapshot;
diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
index 8a2bbd2a4d82..e8e76744fa32 100644
--- a/tools/perf/util/stat-shadow.c
+++ b/tools/perf/util/stat-shadow.c
@@ -3,6 +3,9 @@
 #include "stat.h"
 #include "color.h"
 #include "pmu.h"
+#include "rblist.h"
+#include "evlist.h"
+#include "expr.h"
 
 enum {
 	CTX_BIT_USER	= 1 << 0,
@@ -41,13 +44,73 @@ static struct stats runtime_topdown_slots_issued[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_topdown_slots_retired[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_topdown_fetch_bubbles[NUM_CTX][MAX_NR_CPUS];
 static struct stats runtime_topdown_recovery_bubbles[NUM_CTX][MAX_NR_CPUS];
+static struct rblist runtime_saved_values;
 static bool have_frontend_stalled;
 
 struct stats walltime_nsecs_stats;
 
+struct saved_value {
+	struct rb_node rb_node;
+	struct perf_evsel *evsel;
+	int cpu;
+	int ctx;
+	struct stats stats;
+};
+
+static int saved_value_cmp(struct rb_node *rb_node, const void *entry)
+{
+	struct saved_value *a = container_of(rb_node,
+					     struct saved_value,
+					     rb_node);
+	const struct saved_value *b = entry;
+
+	if (a->ctx != b->ctx)
+		return a->ctx - b->ctx;
+	if (a->cpu != b->cpu)
+		return a->cpu - b->cpu;
+	return a->evsel - b->evsel;
+}
+
+static struct rb_node *saved_value_new(struct rblist *rblist __maybe_unused,
+				     const void *entry)
+{
+	struct saved_value *nd = malloc(sizeof(struct saved_value));
+
+	if (!nd)
+		return NULL;
+	memcpy(nd, entry, sizeof(struct saved_value));
+	return &nd->rb_node;
+}
+
+static struct saved_value *saved_value_lookup(struct perf_evsel *evsel,
+					      int cpu, int ctx,
+					      bool create)
+{
+	struct rb_node *nd;
+	struct saved_value dm = {
+		.cpu = cpu,
+		.ctx = ctx,
+		.evsel = evsel,
+	};
+	nd = rblist__find(&runtime_saved_values, &dm);
+	if (nd)
+		return container_of(nd, struct saved_value, rb_node);
+	if (create) {
+		rblist__add_node(&runtime_saved_values, &dm);
+		nd = rblist__find(&runtime_saved_values, &dm);
+		if (nd)
+			return container_of(nd, struct saved_value, rb_node);
+	}
+	return NULL;
+}
+
 void perf_stat__init_shadow_stats(void)
 {
 	have_frontend_stalled = pmu_have_event("cpu", "stalled-cycles-frontend");
+	rblist__init(&runtime_saved_values);
+	runtime_saved_values.node_cmp = saved_value_cmp;
+	runtime_saved_values.node_new = saved_value_new;
+	/* No delete for now */
 }
 
 static int evsel_context(struct perf_evsel *evsel)
@@ -70,6 +133,8 @@ static int evsel_context(struct perf_evsel *evsel)
 
 void perf_stat__reset_shadow_stats(void)
 {
+	struct rb_node *pos, *next;
+
 	memset(runtime_nsecs_stats, 0, sizeof(runtime_nsecs_stats));
 	memset(runtime_cycles_stats, 0, sizeof(runtime_cycles_stats));
 	memset(runtime_stalled_cycles_front_stats, 0, sizeof(runtime_stalled_cycles_front_stats));
@@ -92,6 +157,15 @@ void perf_stat__reset_shadow_stats(void)
 	memset(runtime_topdown_slots_issued, 0, sizeof(runtime_topdown_slots_issued));
 	memset(runtime_topdown_fetch_bubbles, 0, sizeof(runtime_topdown_fetch_bubbles));
 	memset(runtime_topdown_recovery_bubbles, 0, sizeof(runtime_topdown_recovery_bubbles));
+
+	next = rb_first(&runtime_saved_values.entries);
+	while (next) {
+		pos = next;
+		next = rb_next(pos);
+		memset(&container_of(pos, struct saved_value, rb_node)->stats,
+		       0,
+		       sizeof(struct stats));
+	}
 }
 
 /*
@@ -143,6 +217,12 @@ void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
 		update_stats(&runtime_dtlb_cache_stats[ctx][cpu], count[0]);
 	else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
 		update_stats(&runtime_itlb_cache_stats[ctx][cpu], count[0]);
+
+	if (counter->collect_stat) {
+		struct saved_value *v = saved_value_lookup(counter, cpu, ctx,
+							   true);
+		update_stats(&v->stats, count[0]);
+	}
 }
 
 /* used for get_ratio_color() */
@@ -172,6 +252,88 @@ static const char *get_ratio_color(enum grc_type type, double ratio)
 	return color;
 }
 
+static struct perf_evsel *perf_stat__find_event(struct perf_evlist *evsel_list,
+						const char *name)
+{
+	struct perf_evsel *c2;
+
+	evlist__for_each_entry (evsel_list, c2) {
+		if (!strcasecmp(c2->name, name))
+			return c2;
+	}
+	return NULL;
+}
+
+/* Mark MetricExpr target events and link events using them to them. */
+void perf_stat__collect_metric_expr(struct perf_evlist *evsel_list)
+{
+	struct perf_evsel *counter, *leader, **metric_events, *oc;
+	bool found;
+	const char **metric_names;
+	int i;
+	int num_metric_names;
+
+	evlist__for_each_entry(evsel_list, counter) {
+		bool invalid = false;
+
+		leader = counter->leader;
+		if (!counter->metric_expr)
+			continue;
+		metric_events = counter->metric_events;
+		if (!metric_events) {
+			if (expr__find_other(counter->metric_expr, counter->name,
+						&metric_names, &num_metric_names) < 0)
+				continue;
+
+			metric_events = calloc(sizeof(struct perf_evsel *),
+					       num_metric_names + 1);
+			if (!metric_events)
+				return;
+			counter->metric_events = metric_events;
+		}
+
+		for (i = 0; i < num_metric_names; i++) {
+			found = false;
+			if (leader) {
+				/* Search in group */
+				for_each_group_member (oc, leader) {
+					if (!strcasecmp(oc->name, metric_names[i])) {
+						found = true;
+						break;
+					}
+				}
+			}
+			if (!found) {
+				/* Search ignoring groups */
+				oc = perf_stat__find_event(evsel_list, metric_names[i]);
+			}
+			if (!oc) {
+				/*
+				 * Adding events automatically would be difficult, because
+				 * it would risk creating groups that are not schedulable.
+				 * perf stat doesn't understand all the scheduling constraints
+				 * of events. So we ask the user instead to add the missing
+				 * events.
+				 */
+				fprintf(stderr, "Add %s event to groups to get metric expression for %s\n",
+						metric_names[i],
+						counter->name);
+				invalid = true;
+				continue;
+			}
+			metric_events[i] = oc;
+			oc->collect_stat = true;
+		}
+		metric_events[i] = NULL;
+		free(metric_names);
+		if (invalid) {
+			free(metric_events);
+			counter->metric_events = NULL;
+			counter->metric_expr = NULL;
+		}
+	}
+}
+
 static void print_stalled_cycles_frontend(int cpu,
 					  struct perf_evsel *evsel, double avg,
 					  struct perf_stat_output_ctx *out)
@@ -614,6 +776,32 @@ void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 					be_bound * 100.);
 		else
 			print_metric(ctxp, NULL, NULL, name, 0);
+	} else if (evsel->metric_expr) {
+		struct parse_ctx pctx;
+		int i;
+
+		expr__ctx_init(&pctx);
+		expr__add_id(&pctx, evsel->name, avg);
+		for (i = 0; evsel->metric_events[i]; i++) {
+			struct saved_value *v;
+
+			v = saved_value_lookup(evsel->metric_events[i], cpu, ctx, false);
+			if (!v)
+				break;
+			expr__add_id(&pctx, evsel->metric_events[i]->name,
+					     avg_stats(&v->stats));
+		}
+		if (!evsel->metric_events[i]) {
+			const char *p = evsel->metric_expr;
+
+			if (expr__parse(&ratio, &pctx, &p) == 0)
+				print_metric(ctxp, NULL, "%8.1f",
+					out->force_header ? evsel->name : "",
+					ratio);
+			else
+				print_metric(ctxp, NULL, NULL, "", 0);
+		} else
+			print_metric(ctxp, NULL, NULL, "", 0);
 	} else if (runtime_nsecs_stats[cpu].n != 0) {
 		char unit = 'M';
 		char unit_buf[10];
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index c29bb94c48a4..0a65ae23f495 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -85,11 +85,13 @@ struct perf_stat_output_ctx {
 	void *ctx;
 	print_metric_t print_metric;
 	new_line_t new_line;
+	bool force_header;
 };
 
 void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
 				   double avg, int cpu,
 				   struct perf_stat_output_ctx *out);
+void perf_stat__collect_metric_expr(struct perf_evlist *);
 
 int perf_evlist__alloc_stats(struct perf_evlist *evlist, bool alloc_raw);
 void perf_evlist__free_stats(struct perf_evlist *evlist);
-- 
2.9.3