Commit 0197da7a authored by Ian Rogers's avatar Ian Rogers Committed by Namhyung Kim

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: default avatarIan Rogers <irogers@google.com>
Reviewed-by: default avatarAdrian Hunter <adrian.hunter@intel.com>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: James Clark <james.clark@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Yang Jihong <yangjihong1@huawei.com>
Cc: Will Deacon <will@kernel.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Kajol Jain <kjain@linux.ibm.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: John Garry <john.g.garry@oracle.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Link: https://lore.kernel.org/r/20231012175645.1849503-8-irogers@google.comSigned-off-by: default avatarNamhyung Kim <namhyung@kernel.org>
parent f20c15d1
......@@ -917,16 +917,9 @@ struct auxtrace_record *cs_etm_record_init(int *err)
* (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;
}
......@@ -20,12 +20,12 @@ void perf_pmu__arch_init(struct perf_pmu *pmu __maybe_unused)
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
......
......@@ -113,6 +113,25 @@ arm_spe_snapshot_resolve_auxtrace_defaults(struct record_opts *opts,
}
}
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)
......@@ -136,7 +155,7 @@ static int arm_spe_recording_options(struct auxtrace_record *itr,
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;
......@@ -495,26 +514,8 @@ struct auxtrace_record *arm_spe_recording_init(int *err,
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);
}
......@@ -60,7 +60,7 @@ struct intel_pt_recording {
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)
{
......@@ -84,7 +84,7 @@ static int intel_pt_parse_terms_with_default(struct perf_pmu *pmu,
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);
......@@ -177,7 +177,7 @@ static int intel_pt_pick_bit(int bits, int target)
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;
......@@ -256,18 +256,17 @@ static int intel_pt_parse_snapshot_options(struct auxtrace_record *itr,
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,
......
......@@ -23,7 +23,7 @@ void perf_pmu__arch_init(struct perf_pmu *pmu __maybe_unused)
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;
......
......@@ -27,5 +27,7 @@ struct auxtrace_record *arm_spe_recording_init(int *err,
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
......@@ -242,7 +242,7 @@ struct cs_etm_packet_queue {
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,
......
......@@ -42,6 +42,7 @@ struct auxtrace_record *intel_pt_recording_init(int *err);
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
......@@ -1418,11 +1418,10 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
}
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)) {
......@@ -1466,7 +1465,8 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
* 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;
}
......
......@@ -1402,7 +1402,7 @@ int perf_pmu__config(struct perf_pmu *pmu, struct perf_event_attr *attr,
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);
}
......@@ -2064,7 +2064,6 @@ void perf_pmu__delete(struct perf_pmu *pmu)
perf_cpu_map__put(pmu->cpus);
zfree(&pmu->default_config);
zfree(&pmu->name);
zfree(&pmu->alias_name);
zfree(&pmu->id);
......
......@@ -92,10 +92,11 @@ struct perf_pmu {
*/
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.
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment