[RFC PATCH v1 1/5] perf stat: Parse and find tpebs events when parsing metrics to prepare for perf record sampling
From: weilin . wang
Date: Wed Feb 21 2024 - 02:22:08 EST
From: Weilin Wang <weilin.wang@xxxxxxxxx>
Metrics that use tpebs values would use the :retire_latency keyword in
formulas. We put all these events into a list and pass the list to perf
record to collect their retire latency value.
Signed-off-by: Weilin Wang <weilin.wang@xxxxxxxxx>
---
tools/perf/builtin-stat.c | 38 +++++++++++++++++++++----
tools/perf/util/metricgroup.c | 52 +++++++++++++++++++++++++++--------
tools/perf/util/metricgroup.h | 10 ++++++-
tools/perf/util/stat.h | 2 ++
4 files changed, 84 insertions(+), 18 deletions(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 07b48f6df48e..93a7cba30f12 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -162,6 +162,7 @@ static struct perf_stat_config stat_config = {
.ctl_fd = -1,
.ctl_fd_ack = -1,
.iostat_run = false,
+ .tpebs_event_size = 0,
};
static bool cpus_map_matched(struct evsel *a, struct evsel *b)
@@ -686,6 +687,12 @@ static enum counter_recovery stat_handle_error(struct evsel *counter)
return COUNTER_FATAL;
}
+static int __run_perf_record(void)
+{
+ pr_debug("Prepare perf record for retire_latency\n");
+ return 0;
+}
+
static int __run_perf_stat(int argc, const char **argv, int run_idx)
{
int interval = stat_config.interval;
@@ -703,6 +710,15 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
int err;
bool second_pass = false;
+ //Prepare perf record for sampling event retire_latency before fork and prepare workload
+ if (stat_config.tpebs_event_size > 0) {
+ int ret;
+
+ ret = __run_perf_record();
+ if (ret)
+ return ret;
+ }
+
if (forks) {
if (evlist__prepare_workload(evsel_list, &target, argv, is_pipe, workload_exec_failed_signal) < 0) {
perror("failed to prepare workload");
@@ -2061,7 +2077,9 @@ static int add_default_attributes(void)
stat_config.metric_no_threshold,
stat_config.user_requested_cpu_list,
stat_config.system_wide,
- &stat_config.metric_events);
+ &stat_config.metric_events,
+ &stat_config.tpebs_events,
+ &stat_config.tpebs_event_size);
}
if (smi_cost) {
@@ -2094,7 +2112,9 @@ static int add_default_attributes(void)
stat_config.metric_no_threshold,
stat_config.user_requested_cpu_list,
stat_config.system_wide,
- &stat_config.metric_events);
+ &stat_config.metric_events,
+ &stat_config.tpebs_events,
+ &stat_config.tpebs_event_size);
}
if (topdown_run) {
@@ -2128,7 +2148,9 @@ static int add_default_attributes(void)
/*metric_no_threshold=*/true,
stat_config.user_requested_cpu_list,
stat_config.system_wide,
- &stat_config.metric_events) < 0)
+ &stat_config.metric_events,
+ &stat_config.tpebs_events,
+ &stat_config.tpebs_event_size) < 0)
return -1;
}
@@ -2169,7 +2191,9 @@ static int add_default_attributes(void)
/*metric_no_threshold=*/true,
stat_config.user_requested_cpu_list,
stat_config.system_wide,
- &stat_config.metric_events) < 0)
+ &stat_config.metric_events,
+ /*&stat_config.tpebs_events=*/NULL,
+ /*stat_config.tpebs_event_size=*/0) < 0)
return -1;
evlist__for_each_entry(metric_evlist, metric_evsel) {
@@ -2689,6 +2713,8 @@ int cmd_stat(int argc, const char **argv)
}
}
+ INIT_LIST_HEAD(&stat_config.tpebs_events);
+
/*
* Metric parsing needs to be delayed as metrics may optimize events
* knowing the target is system-wide.
@@ -2702,7 +2728,9 @@ int cmd_stat(int argc, const char **argv)
stat_config.metric_no_threshold,
stat_config.user_requested_cpu_list,
stat_config.system_wide,
- &stat_config.metric_events);
+ &stat_config.metric_events,
+ &stat_config.tpebs_events,
+ &stat_config.tpebs_event_size);
zfree(&metrics);
}
diff --git a/tools/perf/util/metricgroup.c b/tools/perf/util/metricgroup.c
index 6231044a491e..6c16e5a0b1fc 100644
--- a/tools/perf/util/metricgroup.c
+++ b/tools/perf/util/metricgroup.c
@@ -275,7 +275,8 @@ static bool contains_metric_id(struct evsel **metric_events, int num_events,
*/
static int setup_metric_events(const char *pmu, struct hashmap *ids,
struct evlist *metric_evlist,
- struct evsel ***out_metric_events)
+ struct evsel ***out_metric_events,
+ size_t tpebs_event_size)
{
struct evsel **metric_events;
const char *metric_id;
@@ -284,7 +285,7 @@ static int setup_metric_events(const char *pmu, struct hashmap *ids,
bool all_pmus = !strcmp(pmu, "all") || perf_pmus__num_core_pmus() == 1 || !is_pmu_core(pmu);
*out_metric_events = NULL;
- ids_size = hashmap__size(ids);
+ ids_size = hashmap__size(ids) - tpebs_event_size;
metric_events = calloc(sizeof(void *), ids_size + 1);
if (!metric_events)
@@ -321,6 +322,7 @@ static int setup_metric_events(const char *pmu, struct hashmap *ids,
}
}
if (matched_events < ids_size) {
+ pr_debug("Error: matched_events = %lu, ids_size = %lu\n", matched_events, ids_size);
free(metric_events);
return -EINVAL;
}
@@ -668,7 +670,9 @@ static int decode_all_metric_ids(struct evlist *perf_evlist, const char *modifie
static int metricgroup__build_event_string(struct strbuf *events,
const struct expr_parse_ctx *ctx,
const char *modifier,
- bool group_events)
+ bool group_events,
+ struct list_head *tpebs_events __maybe_unused,
+ size_t *tpebs_event_size)
{
struct hashmap_entry *cur;
size_t bkt;
@@ -681,9 +685,21 @@ static int metricgroup__build_event_string(struct strbuf *events,
hashmap__for_each_entry(ctx->ids, cur, bkt) {
const char *sep, *rsep, *id = cur->pkey;
enum perf_tool_event ev;
+ char *p = strstr(id, ":retire_latency");
pr_debug("found event %s\n", id);
+ if (p) {
+ struct tpebs_event *new_event = malloc(sizeof(struct tpebs_event));
+ *p = '\0';
+ new_event->tpebs_name = strdup(id);
+ *tpebs_event_size += 1;
+ pr_debug("retire_latency required, tpebs_event_size=%lu, new_event=%s\n",
+ *tpebs_event_size, new_event->name);
+ list_add_tail(&new_event->nd, tpebs_events);
+ continue;
+ }
+
/* Always move tool events outside of the group. */
ev = perf_tool_event__from_str(id);
if (ev != PERF_TOOL_NONE) {
@@ -1447,7 +1463,8 @@ static int build_combined_expr_ctx(const struct list_head *metric_list,
static int parse_ids(bool metric_no_merge, struct perf_pmu *fake_pmu,
struct expr_parse_ctx *ids, const char *modifier,
bool group_events, const bool tool_events[PERF_TOOL_MAX],
- struct evlist **out_evlist)
+ struct evlist **out_evlist, struct list_head *tpebs_events,
+ size_t *tpebs_event_size)
{
struct parse_events_error parse_error;
struct evlist *parsed_evlist;
@@ -1490,7 +1507,7 @@ static int parse_ids(bool metric_no_merge, struct perf_pmu *fake_pmu,
}
}
ret = metricgroup__build_event_string(&events, ids, modifier,
- group_events);
+ group_events, tpebs_events, tpebs_event_size);
if (ret)
return ret;
@@ -1529,7 +1546,9 @@ static int parse_groups(struct evlist *perf_evlist,
bool system_wide,
struct perf_pmu *fake_pmu,
struct rblist *metric_events_list,
- const struct pmu_metrics_table *table)
+ const struct pmu_metrics_table *table,
+ struct list_head *tpebs_events,
+ size_t *tpebs_event_size)
{
struct evlist *combined_evlist = NULL;
LIST_HEAD(metric_list);
@@ -1561,7 +1580,8 @@ static int parse_groups(struct evlist *perf_evlist,
/*modifier=*/NULL,
/*group_events=*/false,
tool_events,
- &combined_evlist);
+ &combined_evlist,
+ tpebs_events, tpebs_event_size);
}
if (combined)
expr__ctx_free(combined);
@@ -1616,14 +1636,15 @@ static int parse_groups(struct evlist *perf_evlist,
}
if (!metric_evlist) {
ret = parse_ids(metric_no_merge, fake_pmu, m->pctx, m->modifier,
- m->group_events, tool_events, &m->evlist);
+ m->group_events, tool_events, &m->evlist,
+ tpebs_events, tpebs_event_size);
if (ret)
goto out;
metric_evlist = m->evlist;
}
ret = setup_metric_events(fake_pmu ? "all" : m->pmu, m->pctx->ids,
- metric_evlist, &metric_events);
+ metric_evlist, &metric_events, *tpebs_event_size);
if (ret) {
pr_err("Cannot resolve IDs for %s: %s\n",
m->metric_name, m->metric_expr);
@@ -1690,16 +1711,21 @@ int metricgroup__parse_groups(struct evlist *perf_evlist,
bool metric_no_threshold,
const char *user_requested_cpu_list,
bool system_wide,
- struct rblist *metric_events)
+ struct rblist *metric_events,
+ struct list_head *tpebs_events,
+ size_t *tpebs_event_size)
{
const struct pmu_metrics_table *table = pmu_metrics_table__find();
+ pr_debug("Test debugging\n");
+
if (!table)
return -EINVAL;
return parse_groups(perf_evlist, pmu, str, metric_no_group, metric_no_merge,
metric_no_threshold, user_requested_cpu_list, system_wide,
- /*fake_pmu=*/NULL, metric_events, table);
+ /*fake_pmu=*/NULL, metric_events, table, tpebs_events,
+ tpebs_event_size);
}
int metricgroup__parse_groups_test(struct evlist *evlist,
@@ -1713,7 +1739,9 @@ int metricgroup__parse_groups_test(struct evlist *evlist,
/*metric_no_threshold=*/false,
/*user_requested_cpu_list=*/NULL,
/*system_wide=*/false,
- &perf_pmu__fake, metric_events, table);
+ &perf_pmu__fake, metric_events, table,
+ /*tpebs_events=*/NULL,
+ /*tpebs_event_size=*/0);
}
struct metricgroup__has_metric_data {
diff --git a/tools/perf/util/metricgroup.h b/tools/perf/util/metricgroup.h
index d5325c6ec8e1..7c24ed768ff3 100644
--- a/tools/perf/util/metricgroup.h
+++ b/tools/perf/util/metricgroup.h
@@ -66,6 +66,12 @@ struct metric_expr {
int runtime;
};
+struct tpebs_event {
+ struct list_head nd;
+ const char *name;
+ const char *tpebs_name;
+};
+
struct metric_event *metricgroup__lookup(struct rblist *metric_events,
struct evsel *evsel,
bool create);
@@ -77,7 +83,9 @@ int metricgroup__parse_groups(struct evlist *perf_evlist,
bool metric_no_threshold,
const char *user_requested_cpu_list,
bool system_wide,
- struct rblist *metric_events);
+ struct rblist *metric_events,
+ struct list_head *tpebs_events,
+ size_t *tpebs_event_size);
int metricgroup__parse_groups_test(struct evlist *evlist,
const struct pmu_metrics_table *table,
const char *str,
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index 325d0fad1842..9d0186316b29 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -109,6 +109,8 @@ struct perf_stat_config {
struct cpu_aggr_map *cpus_aggr_map;
u64 *walltime_run;
struct rblist metric_events;
+ struct list_head tpebs_events;
+ size_t tpebs_event_size;
int ctl_fd;
int ctl_fd_ack;
bool ctl_fd_close;
--
2.43.0