perf pmu: Lazily compute default config

The default config is computed during creation of the PMU and may do
things like scanning sysfs, when the PMU may just be used as part of
scanning. Change default_config to perf_event_attr_init_default, a
callback that is used when a default config needs initializing. This
avoids holding onto the memory for a perf_event_attr and copying.

On a tigerlake laptop running the pmu-scan benchmark:

Before:
Running 'internals/pmu-scan' benchmark:
Computing performance of sysfs PMU event scan for 100 times
Average core PMU scanning took: 28.780 usec (+- 0.503 usec)
Average PMU scanning took: 283.480 usec (+- 18.471 usec)
Number of openat syscalls: 30,227

After:
Running 'internals/pmu-scan' benchmark:
Computing performance of sysfs PMU event scan for 100 times
Average core PMU scanning took: 27.880 usec (+- 0.169 usec)
Average PMU scanning took: 245.260 usec (+- 15.758 usec)
Number of openat syscalls: 28,914

Over 3 runs it is a nearly 12% reduction in execution time and a 4.3%
of openat calls.

Signed-off-by: Ian Rogers <[e-mail]>
Reviewed-by: Adrian Hunter <[e-mail]>
Cc: Ravi Bangoria <[e-mail]>
Cc: James Clark <[e-mail]>
Cc: Suzuki K Poulose <[e-mail]>
Cc: Yang Jihong <[e-mail]>
Cc: Will Deacon <[e-mail]>
Cc: Leo Yan <[e-mail]>
Cc: Mike Leach <[e-mail]>
Cc: Jing Zhang <[e-mail]>
Cc: Kajol Jain <[e-mail]>
Cc: Thomas Richter <[e-mail]>
Cc: Kan Liang <[e-mail]>
Cc: John Garry <[e-mail]>
Cc: [e-mail]
Cc: [e-mail]
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Namhyung Kim <[e-mail]>

tools/perf/arch/arm/util/cs-etm.c

···
  * (CFG_CHG and evsel__set_config_if_unset()). If no default is set then user
  * changes aren't tracked.
  */
-struct perf_event_attr *
-cs_etm_get_default_config(struct perf_pmu *pmu __maybe_unused)
+void
+cs_etm_get_default_config(const struct perf_pmu *pmu __maybe_unused,
+			  struct perf_event_attr *attr)
 {
-	struct perf_event_attr *attr;
-
-	attr = zalloc(sizeof(struct perf_event_attr));
-	if (!attr)
-		return NULL;
-
 	attr->sample_period = 1;
-
-	return attr;
 }

tools/perf/arch/arm/util/pmu.c

···
 	if (!strcmp(pmu->name, CORESIGHT_ETM_PMU_NAME)) {
 		/* add ETM default config here */
 		pmu->selectable = true;
-		pmu->default_config = cs_etm_get_default_config(pmu);
+		pmu->perf_event_attr_init_default = cs_etm_get_default_config;
 #if defined(__aarch64__)
 	} else if (strstarts(pmu->name, ARM_SPE_PMU_NAME)) {
 		pmu->selectable = true;
 		pmu->is_uncore = false;
-		pmu->default_config = arm_spe_pmu_default_config(pmu);
+		pmu->perf_event_attr_init_default = arm_spe_pmu_default_config;
 	} else if (strstarts(pmu->name, HISI_PTT_PMU_NAME)) {
 		pmu->selectable = true;
 #endif

tools/perf/arch/arm64/util/arm-spe.c

···
 	}
 }
 
+static __u64 arm_spe_pmu__sample_period(const struct perf_pmu *arm_spe_pmu)
+{
+	static __u64 sample_period;
+
+	if (sample_period)
+		return sample_period;
+
+	/*
+	 * If kernel driver doesn't advertise a minimum,
+	 * use max allowable by PMSIDR_EL1.INTERVAL
+	 */
+	if (perf_pmu__scan_file(arm_spe_pmu, "caps/min_interval", "%llu",
+				&sample_period) != 1) {
+		pr_debug("arm_spe driver doesn't advertise a min. interval. Using 4096\n");
+		sample_period = 4096;
+	}
+	return sample_period;
+}
+
 static int arm_spe_recording_options(struct auxtrace_record *itr,
 				     struct evlist *evlist,
 				     struct record_opts *opts)
···
 				return -EINVAL;
 			}
 			evsel->core.attr.freq = 0;
-			evsel->core.attr.sample_period = arm_spe_pmu->default_config->sample_period;
+			evsel->core.attr.sample_period = arm_spe_pmu__sample_period(arm_spe_pmu);
 			evsel->needs_auxtrace_mmap = true;
 			arm_spe_evsel = evsel;
 			opts->full_auxtrace = true;
···
 	return &sper->itr;
 }
 
-struct perf_event_attr
-*arm_spe_pmu_default_config(struct perf_pmu *arm_spe_pmu)
+void
+arm_spe_pmu_default_config(const struct perf_pmu *arm_spe_pmu, struct perf_event_attr *attr)
 {
-	struct perf_event_attr *attr;
-
-	attr = zalloc(sizeof(struct perf_event_attr));
-	if (!attr) {
-		pr_err("arm_spe default config cannot allocate a perf_event_attr\n");
-		return NULL;
-	}
-
-	/*
-	 * If kernel driver doesn't advertise a minimum,
-	 * use max allowable by PMSIDR_EL1.INTERVAL
-	 */
-	if (perf_pmu__scan_file(arm_spe_pmu, "caps/min_interval", "%llu",
-				  &attr->sample_period) != 1) {
-		pr_debug("arm_spe driver doesn't advertise a min. interval. Using 4096\n");
-		attr->sample_period = 4096;
-	}
-
-	return attr;
+	attr->sample_period = arm_spe_pmu__sample_period(arm_spe_pmu);
 }

tools/perf/arch/x86/util/intel-pt.c

···
 	size_t				priv_size;
 };
 
-static int intel_pt_parse_terms_with_default(struct perf_pmu *pmu,
+static int intel_pt_parse_terms_with_default(const struct perf_pmu *pmu,
 					     const char *str,
 					     u64 *config)
 {
···
 	return err;
 }
 
-static int intel_pt_parse_terms(struct perf_pmu *pmu, const char *str, u64 *config)
+static int intel_pt_parse_terms(const struct perf_pmu *pmu, const char *str, u64 *config)
 {
 	*config = 0;
 	return intel_pt_parse_terms_with_default(pmu, str, config);
···
 	return pick;
 }
 
-static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
+static u64 intel_pt_default_config(const struct perf_pmu *intel_pt_pmu)
 {
 	char buf[256];
 	int mtc, mtc_periods = 0, mtc_period;
···
 	return 0;
 }
 
-struct perf_event_attr *
-intel_pt_pmu_default_config(struct perf_pmu *intel_pt_pmu)
+void intel_pt_pmu_default_config(const struct perf_pmu *intel_pt_pmu,
+				 struct perf_event_attr *attr)
 {
-	struct perf_event_attr *attr;
+	static u64 config;
+	static bool initialized;
 
-	attr = zalloc(sizeof(struct perf_event_attr));
-	if (!attr)
-		return NULL;
-
-	attr->config = intel_pt_default_config(intel_pt_pmu);
-
-	return attr;
+	if (!initialized) {
+		config = intel_pt_default_config(intel_pt_pmu);
+		initialized = true;
+	}
+	attr->config = config;
 }
 
 static const char *intel_pt_find_filter(struct evlist *evlist,

tools/perf/arch/x86/util/pmu.c

···
 	if (!strcmp(pmu->name, INTEL_PT_PMU_NAME)) {
 		pmu->auxtrace = true;
 		pmu->selectable = true;
-		pmu->default_config = intel_pt_pmu_default_config(pmu);
+		pmu->perf_event_attr_init_default = intel_pt_pmu_default_config;
 	}
 	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME)) {
 		pmu->auxtrace = true;

tools/perf/util/arm-spe.h

···
 int arm_spe_process_auxtrace_info(union perf_event *event,
 				  struct perf_session *session);
 
-struct perf_event_attr *arm_spe_pmu_default_config(struct perf_pmu *arm_spe_pmu);
+void arm_spe_pmu_default_config(const struct perf_pmu *arm_spe_pmu,
+				struct perf_event_attr *attr);
+
 #endif

tools/perf/util/cs-etm.h

···
 
 int cs_etm__process_auxtrace_info(union perf_event *event,
 				  struct perf_session *session);
-struct perf_event_attr *cs_etm_get_default_config(struct perf_pmu *pmu);
+void cs_etm_get_default_config(const struct perf_pmu *pmu, struct perf_event_attr *attr);
 
 enum cs_etm_pid_fmt {
 	CS_ETM_PIDFMT_NONE,

tools/perf/util/intel-pt.h

···
 int intel_pt_process_auxtrace_info(union perf_event *event,
 				   struct perf_session *session);
 
-struct perf_event_attr *intel_pt_pmu_default_config(struct perf_pmu *pmu);
+void intel_pt_pmu_default_config(const struct perf_pmu *intel_pt_pmu,
+				 struct perf_event_attr *attr);
 
 #endif

tools/perf/util/parse-events.c

···
 	}
 	fix_raw(&parsed_terms, pmu);
 
-	if (pmu->default_config) {
-		memcpy(&attr, pmu->default_config, sizeof(struct perf_event_attr));
-	} else {
-		memset(&attr, 0, sizeof(attr));
-	}
+	memset(&attr, 0, sizeof(attr));
+	if (pmu->perf_event_attr_init_default)
+		pmu->perf_event_attr_init_default(pmu, &attr);
+
 	attr.type = pmu->type;
 
 	if (list_empty(&parsed_terms.terms)) {
···
 	 * When using default config, record which bits of attr->config were
 	 * changed by the user.
 	 */
-	if (pmu->default_config && get_config_chgs(pmu, &parsed_terms, &config_terms)) {
+	if (pmu->perf_event_attr_init_default &&
+	    get_config_chgs(pmu, &parsed_terms, &config_terms)) {
 		parse_events_terms__exit(&parsed_terms);
 		return -ENOMEM;
 	}

tools/perf/util/pmu.c

···
 		     struct parse_events_terms *head_terms,
 		     struct parse_events_error *err)
 {
-	bool zero = !!pmu->default_config;
+	bool zero = !!pmu->perf_event_attr_init_default;
 
 	return perf_pmu__config_terms(pmu, attr, head_terms, zero, err);
 }
···
 
 	perf_cpu_map__put(pmu->cpus);
 
-	zfree(&pmu->default_config);
 	zfree(&pmu->name);
 	zfree(&pmu->alias_name);
 	zfree(&pmu->id);

tools/perf/util/pmu.h

···
 	 */
 	int max_precise;
 	/**
-	 * @default_config: Optional default perf_event_attr determined in
-	 * architecture specific code.
+	 * @perf_event_attr_init_default: Optional function to default
+	 * initialize PMU specific parts of the perf_event_attr.
 	 */
-	struct perf_event_attr *default_config;
+	void (*perf_event_attr_init_default)(const struct perf_pmu *pmu,
+					     struct perf_event_attr *attr);
 	/**
 	 * @cpus: Empty or the contents of either of:
 	 * <sysfs>/bus/event_source/devices/<name>/cpumask.