[PATCH RFC 08/10] perf,tools: store counter val in events_stats

[kan.liang@intel.com]

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

Add a new item counter_val in events_stats to store the total counter
statistics value on each CPU.
The absolate value from counter will be stored in prev_raw_counts. Then
the detla value will be caculated and added into counter_val.

Signed-off-by: Kan Liang <kan.liang@intel.com>
---
 tools/perf/builtin-report.c | 40 ++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/event.h     |  3 +++
 2 files changed, 43 insertions(+)

diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 1d8dce2..31818bc 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -38,6 +38,8 @@
 
 #include "util/auxtrace.h"
 
+#include "util/stat.h"
+
 #include <dlfcn.h>
 #include <linux/bitmap.h>
 
@@ -213,12 +215,33 @@ static int process_counter_read_event(struct perf_tool *tool __maybe_unused,
 {
 	struct perf_evlist *evlist = session->evlist;
 	struct perf_evsel *evsel;
+	int cpu_nr = session->header.env.nr_cpus_online;
+	u32 cpu = event->counter_read.cpu;
+	struct hists *evsel_hists;
 
 	evlist__for_each(evlist, evsel) {
 		if (evsel->idx == event->counter_read.idx)
 			break;
 	}
 
+	if (evsel->prev_raw_counts == NULL) {
+		if (perf_evsel__alloc_prev_raw_counts(evsel, cpu_nr, 1))
+			return -ENOMEM;
+	}
+
+	evsel_hists = evsel__hists(evsel);
+	if (evsel_hists->stats.counter_val == NULL) {
+		evsel_hists->stats.counter_val = zalloc(cpu_nr * sizeof(u64));
+		if (evsel_hists->stats.counter_val == NULL) {
+			perf_evsel__free_prev_raw_counts(evsel);
+			return -ENOMEM;
+		}
+	}
+	evsel_hists->stats.counter_val[cpu] +=
+		event->counter_read.val - perf_counts(evsel->prev_raw_counts, cpu, 0)->val;
+
+	perf_counts(evsel->prev_raw_counts, cpu, 0)->val = event->counter_read.val;
+
 	hists__inc_nr_events(evsel__hists(evsel), event->header.type);
 
 	dump_printf(": %s CPU %d: val %" PRIu64 " ena %" PRIu64 " run %" PRIu64 " \n",
@@ -632,6 +655,21 @@ parse_percent_limit(const struct option *opt, const char *str,
 	return 0;
 }
 
+static void
+free_counter_read(struct perf_evlist *evlist)
+{
+	struct perf_evsel *evsel;
+	struct hists *evsel_hists;
+
+	evlist__for_each(evlist, evsel) {
+		evsel_hists = evsel__hists(evsel);
+		if (evsel_hists->stats.counter_val != NULL)
+			zfree(&evsel_hists->stats.counter_val);
+		if (evsel->prev_raw_counts != NULL)
+			perf_evsel__free_prev_raw_counts(evsel);
+	}
+}
+
 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 {
 	struct perf_session *session;
@@ -946,12 +984,14 @@ repeat:
 
 	ret = __cmd_report(&report);
 	if (ret == K_SWITCH_INPUT_DATA) {
+		free_counter_read(session->evlist);
 		perf_session__delete(session);
 		goto repeat;
 	} else
 		ret = 0;
 
 error:
+	free_counter_read(session->evlist);
 	perf_session__delete(session);
 	return ret;
 }
diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index 7f125ce..c8a663a 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -252,6 +252,8 @@ enum auxtrace_error_type {
  * multipling nr_events[PERF_EVENT_SAMPLE] by a frequency isn't possible to get
  * the total number of low level events, it is necessary to to sum all struct
  * sample_event.period and stash the result in total_period.
+ *
+ * counter_val is used to store the counter statistics value on each CPU
  */
 struct events_stats {
 	u64 total_period;
@@ -268,6 +270,7 @@ struct events_stats {
 	u32 nr_unprocessable_samples;
 	u32 nr_auxtrace_errors[PERF_AUXTRACE_ERROR_MAX];
 	u32 nr_proc_map_timeout;
+	u64 *counter_val;
 };
 
 struct attr_event {
-- 
1.8.3.1