linux/tools/perf/util/stat.c
Andi Kleen 4804e01116 perf stat: Use affinity for opening events
Restructure the event opening in perf stat to cycle through the events
by CPU after setting affinity to that CPU.

This eliminates IPI overhead in the perf API.

We have to loop through the CPU in the outter builtin-stat code instead
of leaving that to low level functions.

It has to change the weak group fallback strategy slightly.  Since we
cannot easily undo the opens for other CPUs move the weak group retry to
a separate loop.

Before with a large test case with 94 CPUs:

  % time     seconds  usecs/call     calls    errors syscall
  ------ ----------- ----------- --------- --------- ----------------
   42.75    4.050910          67     60046       110 perf_event_open

After:

   26.86    0.944396          16     58069       110 perf_event_open

(the number changes slightly because the weak group retries
work differently and the test case relies on weak groups)

Committer notes:

Added one of the hunks in a patch provided by Andi after I noticed that
the "event times" 'perf test' entry was segfaulting.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lore.kernel.org/lkml/20191121001522.180827-10-andi@firstfloor.org
Link: http://lore.kernel.org/lkml/20191127232657.GL84886@tassilo.jf.intel.com # Fix
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2019-11-29 12:20:45 -03:00

526 lines
12 KiB
C

// SPDX-License-Identifier: GPL-2.0
#include <errno.h>
#include <inttypes.h>
#include <math.h>
#include <string.h>
#include "counts.h"
#include "cpumap.h"
#include "debug.h"
#include "header.h"
#include "stat.h"
#include "session.h"
#include "target.h"
#include "evlist.h"
#include "evsel.h"
#include "thread_map.h"
#include <linux/zalloc.h>
void update_stats(struct stats *stats, u64 val)
{
double delta;
stats->n++;
delta = val - stats->mean;
stats->mean += delta / stats->n;
stats->M2 += delta*(val - stats->mean);
if (val > stats->max)
stats->max = val;
if (val < stats->min)
stats->min = val;
}
double avg_stats(struct stats *stats)
{
return stats->mean;
}
/*
* http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
*
* (\Sum n_i^2) - ((\Sum n_i)^2)/n
* s^2 = -------------------------------
* n - 1
*
* http://en.wikipedia.org/wiki/Stddev
*
* The std dev of the mean is related to the std dev by:
*
* s
* s_mean = -------
* sqrt(n)
*
*/
double stddev_stats(struct stats *stats)
{
double variance, variance_mean;
if (stats->n < 2)
return 0.0;
variance = stats->M2 / (stats->n - 1);
variance_mean = variance / stats->n;
return sqrt(variance_mean);
}
double rel_stddev_stats(double stddev, double avg)
{
double pct = 0.0;
if (avg)
pct = 100.0 * stddev/avg;
return pct;
}
bool __perf_evsel_stat__is(struct evsel *evsel,
enum perf_stat_evsel_id id)
{
struct perf_stat_evsel *ps = evsel->stats;
return ps->id == id;
}
#define ID(id, name) [PERF_STAT_EVSEL_ID__##id] = #name
static const char *id_str[PERF_STAT_EVSEL_ID__MAX] = {
ID(NONE, x),
ID(CYCLES_IN_TX, cpu/cycles-t/),
ID(TRANSACTION_START, cpu/tx-start/),
ID(ELISION_START, cpu/el-start/),
ID(CYCLES_IN_TX_CP, cpu/cycles-ct/),
ID(TOPDOWN_TOTAL_SLOTS, topdown-total-slots),
ID(TOPDOWN_SLOTS_ISSUED, topdown-slots-issued),
ID(TOPDOWN_SLOTS_RETIRED, topdown-slots-retired),
ID(TOPDOWN_FETCH_BUBBLES, topdown-fetch-bubbles),
ID(TOPDOWN_RECOVERY_BUBBLES, topdown-recovery-bubbles),
ID(SMI_NUM, msr/smi/),
ID(APERF, msr/aperf/),
};
#undef ID
static void perf_stat_evsel_id_init(struct evsel *evsel)
{
struct perf_stat_evsel *ps = evsel->stats;
int i;
/* ps->id is 0 hence PERF_STAT_EVSEL_ID__NONE by default */
for (i = 0; i < PERF_STAT_EVSEL_ID__MAX; i++) {
if (!strcmp(perf_evsel__name(evsel), id_str[i])) {
ps->id = i;
break;
}
}
}
static void perf_evsel__reset_stat_priv(struct evsel *evsel)
{
int i;
struct perf_stat_evsel *ps = evsel->stats;
for (i = 0; i < 3; i++)
init_stats(&ps->res_stats[i]);
perf_stat_evsel_id_init(evsel);
}
static int perf_evsel__alloc_stat_priv(struct evsel *evsel)
{
evsel->stats = zalloc(sizeof(struct perf_stat_evsel));
if (evsel->stats == NULL)
return -ENOMEM;
perf_evsel__reset_stat_priv(evsel);
return 0;
}
static void perf_evsel__free_stat_priv(struct evsel *evsel)
{
struct perf_stat_evsel *ps = evsel->stats;
if (ps)
zfree(&ps->group_data);
zfree(&evsel->stats);
}
static int perf_evsel__alloc_prev_raw_counts(struct evsel *evsel,
int ncpus, int nthreads)
{
struct perf_counts *counts;
counts = perf_counts__new(ncpus, nthreads);
if (counts)
evsel->prev_raw_counts = counts;
return counts ? 0 : -ENOMEM;
}
static void perf_evsel__free_prev_raw_counts(struct evsel *evsel)
{
perf_counts__delete(evsel->prev_raw_counts);
evsel->prev_raw_counts = NULL;
}
static void perf_evsel__reset_prev_raw_counts(struct evsel *evsel)
{
if (evsel->prev_raw_counts) {
evsel->prev_raw_counts->aggr.val = 0;
evsel->prev_raw_counts->aggr.ena = 0;
evsel->prev_raw_counts->aggr.run = 0;
}
}
static int perf_evsel__alloc_stats(struct evsel *evsel, bool alloc_raw)
{
int ncpus = perf_evsel__nr_cpus(evsel);
int nthreads = perf_thread_map__nr(evsel->core.threads);
if (perf_evsel__alloc_stat_priv(evsel) < 0 ||
perf_evsel__alloc_counts(evsel, ncpus, nthreads) < 0 ||
(alloc_raw && perf_evsel__alloc_prev_raw_counts(evsel, ncpus, nthreads) < 0))
return -ENOMEM;
return 0;
}
int perf_evlist__alloc_stats(struct evlist *evlist, bool alloc_raw)
{
struct evsel *evsel;
evlist__for_each_entry(evlist, evsel) {
if (perf_evsel__alloc_stats(evsel, alloc_raw))
goto out_free;
}
return 0;
out_free:
perf_evlist__free_stats(evlist);
return -1;
}
void perf_evlist__free_stats(struct evlist *evlist)
{
struct evsel *evsel;
evlist__for_each_entry(evlist, evsel) {
perf_evsel__free_stat_priv(evsel);
perf_evsel__free_counts(evsel);
perf_evsel__free_prev_raw_counts(evsel);
}
}
void perf_evlist__reset_stats(struct evlist *evlist)
{
struct evsel *evsel;
evlist__for_each_entry(evlist, evsel) {
perf_evsel__reset_stat_priv(evsel);
perf_evsel__reset_counts(evsel);
}
}
void perf_evlist__reset_prev_raw_counts(struct evlist *evlist)
{
struct evsel *evsel;
evlist__for_each_entry(evlist, evsel)
perf_evsel__reset_prev_raw_counts(evsel);
}
static void zero_per_pkg(struct evsel *counter)
{
if (counter->per_pkg_mask)
memset(counter->per_pkg_mask, 0, cpu__max_cpu());
}
static int check_per_pkg(struct evsel *counter,
struct perf_counts_values *vals, int cpu, bool *skip)
{
unsigned long *mask = counter->per_pkg_mask;
struct perf_cpu_map *cpus = evsel__cpus(counter);
int s;
*skip = false;
if (!counter->per_pkg)
return 0;
if (perf_cpu_map__empty(cpus))
return 0;
if (!mask) {
mask = zalloc(cpu__max_cpu());
if (!mask)
return -ENOMEM;
counter->per_pkg_mask = mask;
}
/*
* we do not consider an event that has not run as a good
* instance to mark a package as used (skip=1). Otherwise
* we may run into a situation where the first CPU in a package
* is not running anything, yet the second is, and this function
* would mark the package as used after the first CPU and would
* not read the values from the second CPU.
*/
if (!(vals->run && vals->ena))
return 0;
s = cpu_map__get_socket(cpus, cpu, NULL);
if (s < 0)
return -1;
*skip = test_and_set_bit(s, mask) == 1;
return 0;
}
static int
process_counter_values(struct perf_stat_config *config, struct evsel *evsel,
int cpu, int thread,
struct perf_counts_values *count)
{
struct perf_counts_values *aggr = &evsel->counts->aggr;
static struct perf_counts_values zero;
bool skip = false;
if (check_per_pkg(evsel, count, cpu, &skip)) {
pr_err("failed to read per-pkg counter\n");
return -1;
}
if (skip)
count = &zero;
switch (config->aggr_mode) {
case AGGR_THREAD:
case AGGR_CORE:
case AGGR_DIE:
case AGGR_SOCKET:
case AGGR_NODE:
case AGGR_NONE:
if (!evsel->snapshot)
perf_evsel__compute_deltas(evsel, cpu, thread, count);
perf_counts_values__scale(count, config->scale, NULL);
if ((config->aggr_mode == AGGR_NONE) && (!evsel->percore)) {
perf_stat__update_shadow_stats(evsel, count->val,
cpu, &rt_stat);
}
if (config->aggr_mode == AGGR_THREAD) {
if (config->stats)
perf_stat__update_shadow_stats(evsel,
count->val, 0, &config->stats[thread]);
else
perf_stat__update_shadow_stats(evsel,
count->val, 0, &rt_stat);
}
break;
case AGGR_GLOBAL:
aggr->val += count->val;
aggr->ena += count->ena;
aggr->run += count->run;
case AGGR_UNSET:
default:
break;
}
return 0;
}
static int process_counter_maps(struct perf_stat_config *config,
struct evsel *counter)
{
int nthreads = perf_thread_map__nr(counter->core.threads);
int ncpus = perf_evsel__nr_cpus(counter);
int cpu, thread;
if (counter->core.system_wide)
nthreads = 1;
for (thread = 0; thread < nthreads; thread++) {
for (cpu = 0; cpu < ncpus; cpu++) {
if (process_counter_values(config, counter, cpu, thread,
perf_counts(counter->counts, cpu, thread)))
return -1;
}
}
return 0;
}
int perf_stat_process_counter(struct perf_stat_config *config,
struct evsel *counter)
{
struct perf_counts_values *aggr = &counter->counts->aggr;
struct perf_stat_evsel *ps = counter->stats;
u64 *count = counter->counts->aggr.values;
int i, ret;
aggr->val = aggr->ena = aggr->run = 0;
/*
* We calculate counter's data every interval,
* and the display code shows ps->res_stats
* avg value. We need to zero the stats for
* interval mode, otherwise overall avg running
* averages will be shown for each interval.
*/
if (config->interval)
init_stats(ps->res_stats);
if (counter->per_pkg)
zero_per_pkg(counter);
ret = process_counter_maps(config, counter);
if (ret)
return ret;
if (config->aggr_mode != AGGR_GLOBAL)
return 0;
if (!counter->snapshot)
perf_evsel__compute_deltas(counter, -1, -1, aggr);
perf_counts_values__scale(aggr, config->scale, &counter->counts->scaled);
for (i = 0; i < 3; i++)
update_stats(&ps->res_stats[i], count[i]);
if (verbose > 0) {
fprintf(config->output, "%s: %" PRIu64 " %" PRIu64 " %" PRIu64 "\n",
perf_evsel__name(counter), count[0], count[1], count[2]);
}
/*
* Save the full runtime - to allow normalization during printout:
*/
perf_stat__update_shadow_stats(counter, *count, 0, &rt_stat);
return 0;
}
int perf_event__process_stat_event(struct perf_session *session,
union perf_event *event)
{
struct perf_counts_values count;
struct perf_record_stat *st = &event->stat;
struct evsel *counter;
count.val = st->val;
count.ena = st->ena;
count.run = st->run;
counter = perf_evlist__id2evsel(session->evlist, st->id);
if (!counter) {
pr_err("Failed to resolve counter for stat event.\n");
return -EINVAL;
}
*perf_counts(counter->counts, st->cpu, st->thread) = count;
counter->supported = true;
return 0;
}
size_t perf_event__fprintf_stat(union perf_event *event, FILE *fp)
{
struct perf_record_stat *st = (struct perf_record_stat *)event;
size_t ret;
ret = fprintf(fp, "\n... id %" PRI_lu64 ", cpu %d, thread %d\n",
st->id, st->cpu, st->thread);
ret += fprintf(fp, "... value %" PRI_lu64 ", enabled %" PRI_lu64 ", running %" PRI_lu64 "\n",
st->val, st->ena, st->run);
return ret;
}
size_t perf_event__fprintf_stat_round(union perf_event *event, FILE *fp)
{
struct perf_record_stat_round *rd = (struct perf_record_stat_round *)event;
size_t ret;
ret = fprintf(fp, "\n... time %" PRI_lu64 ", type %s\n", rd->time,
rd->type == PERF_STAT_ROUND_TYPE__FINAL ? "FINAL" : "INTERVAL");
return ret;
}
size_t perf_event__fprintf_stat_config(union perf_event *event, FILE *fp)
{
struct perf_stat_config sc;
size_t ret;
perf_event__read_stat_config(&sc, &event->stat_config);
ret = fprintf(fp, "\n");
ret += fprintf(fp, "... aggr_mode %d\n", sc.aggr_mode);
ret += fprintf(fp, "... scale %d\n", sc.scale);
ret += fprintf(fp, "... interval %u\n", sc.interval);
return ret;
}
int create_perf_stat_counter(struct evsel *evsel,
struct perf_stat_config *config,
struct target *target,
int cpu)
{
struct perf_event_attr *attr = &evsel->core.attr;
struct evsel *leader = evsel->leader;
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING;
/*
* The event is part of non trivial group, let's enable
* the group read (for leader) and ID retrieval for all
* members.
*/
if (leader->core.nr_members > 1)
attr->read_format |= PERF_FORMAT_ID|PERF_FORMAT_GROUP;
attr->inherit = !config->no_inherit;
/*
* Some events get initialized with sample_(period/type) set,
* like tracepoints. Clear it up for counting.
*/
attr->sample_period = 0;
if (config->identifier)
attr->sample_type = PERF_SAMPLE_IDENTIFIER;
if (config->all_user) {
attr->exclude_kernel = 1;
attr->exclude_user = 0;
}
if (config->all_kernel) {
attr->exclude_kernel = 0;
attr->exclude_user = 1;
}
/*
* Disabling all counters initially, they will be enabled
* either manually by us or by kernel via enable_on_exec
* set later.
*/
if (perf_evsel__is_group_leader(evsel)) {
attr->disabled = 1;
/*
* In case of initial_delay we enable tracee
* events manually.
*/
if (target__none(target) && !config->initial_delay)
attr->enable_on_exec = 1;
}
if (target__has_cpu(target) && !target__has_per_thread(target))
return perf_evsel__open_per_cpu(evsel, evsel__cpus(evsel), cpu);
return perf_evsel__open_per_thread(evsel, evsel->core.threads);
}