[PATCH 10/10] perf, tools, stat: Output generic dividedby metric

From: Andi Kleen
Date: Fri Nov 18 2016 - 19:36:43 EST

From: Andi Kleen <ak@xxxxxxxxxxxxxxx>

Add generic infrastructure to perf stat to output ratios for "DividedBy"
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 dividedby information from the alias to the evsel.

We mark the events that need to be collected for DividedBy, 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.

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 ratios similarly to the
existing hardcoded perf metrics.

Normally we just output the ratio as percent 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 DividedBy 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%

Signed-off-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
---
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 | 139 +++++++++++++++++++++++++++++++++++++++++
tools/perf/util/stat.h | 2 +
8 files changed, 154 insertions(+)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 501d58a4925a..a83a72aebdcb 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);
@@ -1465,6 +1466,7 @@ 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,
@@ -2447,6 +2449,7 @@ 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_dividedby(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 e58a2fbf3b16..fabe36c58d99 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->dividedby = NULL;
+ evsel->div_event = 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 343776422821..376106e623bd 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -131,6 +131,9 @@ struct perf_evsel {
struct list_head config_terms;
int bpf_fd;
bool merged_stat;
+ const char * dividedby;
+ struct perf_evsel *div_event;
+ bool collect_stat;
};

union u64_swap {
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index fba53ba22431..02f10e72cb77 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1252,6 +1252,7 @@ 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->dividedby = info.dividedby;
}

return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 2e6f803cd56d..40c735036915 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -980,6 +980,7 @@ 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->dividedby = NULL;

list_for_each_entry_safe(term, h, head_terms, list) {
alias = pmu_find_alias(pmu, term);
@@ -995,6 +996,7 @@ int perf_pmu__check_alias(struct perf_pmu *pmu, struct list_head *head_terms,

if (alias->per_pkg)
info->per_pkg = true;
+ info->dividedby = alias->dividedby;

list_del(&term->list);
free(term);
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index faf8a7f97d03..5fbdb65064db 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 *dividedby;
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..e1f9d1bc1ef2 100644
--- a/tools/perf/util/stat-shadow.c
+++ b/tools/perf/util/stat-shadow.c
@@ -3,6 +3,8 @@
#include "stat.h"
#include "color.h"
#include "pmu.h"
+#include "rblist.h"
+#include "evlist.h"

enum {
CTX_BIT_USER = 1 << 0,
@@ -41,13 +43,73 @@ 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 +132,8 @@ 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 +156,15 @@ 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 +216,12 @@ 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 +251,55 @@ 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 DividedBy target events and link events using them to them. */
+void perf_stat__collect_dividedby(struct perf_evlist *evsel_list)
+{
+ struct perf_evsel *counter, *leader, *c2;
+ bool found;
+
+ evlist__for_each_entry(evsel_list, counter) {
+ leader = counter->leader;
+ if (!counter->dividedby)
+ continue;
+ found = false;
+ if (leader) {
+ /* Search in group */
+ for_each_group_member (c2, leader) {
+ if (!strcasecmp(c2->name, counter->dividedby)) {
+ found = true;
+ break;
+ }
+ }
+ }
+ if (!found) {
+ /* Search ignoring groups */
+ c2 = perf_stat__find_event(evsel_list, counter->dividedby);
+ }
+ if (!c2) {
+ /* Could try to automatically add the event here. */
+ fprintf(stderr, "Add %s to groups to get ratios for %s\n",
+ counter->dividedby,
+ counter->name);
+ counter->dividedby = NULL;
+ continue;
+ }
+ counter->div_event = c2;
+ c2->collect_stat = true;
+ }
+}
+
static void print_stalled_cycles_frontend(int cpu,
struct perf_evsel *evsel, double avg,
struct perf_stat_output_ctx *out)
@@ -614,6 +742,17 @@ void perf_stat__print_shadow_stats(struct perf_evsel *evsel,
be_bound * 100.);
else
print_metric(ctxp, NULL, NULL, name, 0);
+ } else if (evsel->dividedby) {
+ struct saved_value *v = saved_value_lookup(evsel->div_event, cpu, ctx,
+ false);
+ if (v) {
+ total = avg_stats(&v->stats);
+ if (total)
+ ratio = avg / total;
+ print_metric(ctxp, NULL, "%8.1f%%",
+ out->force_header ? evsel->name : "",
+ ratio * 100.);
+ }
} 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..d79e03ccd644 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_dividedby(struct perf_evlist *);

int perf_evlist__alloc_stats(struct perf_evlist *evlist, bool alloc_raw);
void perf_evlist__free_stats(struct perf_evlist *evlist);
--
2.5.5