Message-Id: <20170320201711.14142-11-andi@firstfloor.org>
Date:   Mon, 20 Mar 2017 13:17:08 -0700
From:   Andi Kleen <andi@...stfloor.org>
To:     acme@...nel.org
Cc:     jolsa@...nel.org, linux-kernel@...r.kernel.org,
        Andi Kleen <ak@...ux.intel.com>
Subject: [PATCH 10/13] perf, tools, stat: Output JSON MetricExpr metric

From: Andi Kleen <ak@...ux.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.

There is no attempt to automatically add the MetricExpr event,
if it is missing, however we suggest it to the user, because
the user tool doesn't have enough information to reliably
construct a group that is guaranteed to schedule. So we leave
that to the user.

% perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}'
     1.000147889        800,085,181      unc_p_clockticks
     1.000147889         93,126,241      unc_p_freq_max_os_cycles  #     11.6
     2.000448381        800,218,217      unc_p_clockticks
     2.000448381        142,516,095      unc_p_freq_max_os_cycles  #     17.8
     3.000639852        800,243,057      unc_p_clockticks
     3.000639852        162,292,689      unc_p_freq_max_os_cycles  #     20.3

% perf stat -a -I 1000 -e '{unc_p_clockticks,unc_p_freq_max_os_cycles}' --metric-only
     1.000127077      0.9
     2.000301436      0.7
     3.000456379      0.0

v2: Change from DivideBy to MetricExpr
v3: Use expr__ prefix.  Support more than one other event.
v4: Update description
v5: Only print warning message once for multiple PMUs.
Signed-off-by: Andi Kleen <ak@...ux.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  | 195 +++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/stat.h         |   2 +
 8 files changed, 210 insertions(+)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index cd7dc3b648ca..01b589e3c3a6 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1145,6 +1145,7 @@ 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);
@@ -1480,6 +1481,7 @@ 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,
@@ -2498,6 +2500,7 @@ 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 175dc2305aa8..ef2a31f6dd06 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 e594c974c93e..91b8e83e307d 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->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 c0d487b3b925..0f1133aa3253 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -994,6 +994,7 @@ 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);
@@ -1009,6 +1010,7 @@ 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..c323cce97d98 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 +252,95 @@ 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) {
+				/* Deduping one is good enough to handle duplicated PMUs. */
+				static char *printed;
+
+				/*
+				 * 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.
+				 */
+				if (!printed || strcasecmp(printed, metric_names[i])) {
+					fprintf(stderr,
+						"Add %s event to groups to get metric expression for %s\n",
+						metric_names[i],
+						counter->name);
+					printed = strdup(metric_names[i]);
+				}
+				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 +783,32 @@ 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

Powered by blists - more mailing lists