linux-stable/tools/perf/util/stat-shadow.c
Kan Liang 63e39aa6ae perf stat: Support L2 Topdown events
The TMA method level 2 metrics is supported from the Intel Sapphire
Rapids server, which expose four L2 Topdown metrics events to user
space. There are eight L2 events in total. The other four L2 Topdown
metrics events are calculated from the corresponding L1 and the exposed
L2 events.

Now, the --topdown prints the complete top-down metrics that supported
by the CPU. For the Intel Sapphire Rapids server, there are 4 L1 events
and 8 L2 events displyed in one line.

Add a new option, --td-level, to display the top-down statistics that
equal to or lower than the input level.

The L2 event is marked only when both its L1 parent event and itself
crosse the threshold.

Here is an example:

  $ perf stat --topdown --td-level=2 --no-metric-only sleep 1
  Topdown accuracy may decrease when measuring long periods.
  Please print the result regularly, e.g. -I1000

  Performance counter stats for 'sleep 1':

     16,734,390   slots
      2,100,001   topdown-retiring       # 12.6% retiring
      2,034,376   topdown-bad-spec       # 12.3% bad speculation
      4,003,128   topdown-fe-bound       # 24.1% frontend bound
        328,125   topdown-heavy-ops      #  2.0% heavy operations    #  10.6% light operations
      1,968,751   topdown-br-mispredict  # 11.9% branch mispredict   #  0.4% machine clears
      2,953,127   topdown-fetch-lat      # 17.8% fetch latency       #  6.3% fetch bandwidth
      5,906,255   topdown-mem-bound      # 35.6% memory bound        #  15.4% core bound

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/1612296553-21962-9-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2021-02-08 16:25:00 -03:00

1305 lines
37 KiB
C

// SPDX-License-Identifier: GPL-2.0
#include <stdio.h>
#include "evsel.h"
#include "stat.h"
#include "color.h"
#include "pmu.h"
#include "rblist.h"
#include "evlist.h"
#include "expr.h"
#include "metricgroup.h"
#include "cgroup.h"
#include <linux/zalloc.h>
/*
* AGGR_GLOBAL: Use CPU 0
* AGGR_SOCKET: Use first CPU of socket
* AGGR_DIE: Use first CPU of die
* AGGR_CORE: Use first CPU of core
* AGGR_NONE: Use matching CPU
* AGGR_THREAD: Not supported?
*/
struct runtime_stat rt_stat;
struct stats walltime_nsecs_stats;
struct saved_value {
struct rb_node rb_node;
struct evsel *evsel;
enum stat_type type;
int ctx;
int cpu;
struct cgroup *cgrp;
struct runtime_stat *stat;
struct stats stats;
u64 metric_total;
int metric_other;
};
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->cpu != b->cpu)
return a->cpu - b->cpu;
/*
* Previously the rbtree was used to link generic metrics.
* The keys were evsel/cpu. Now the rbtree is extended to support
* per-thread shadow stats. For shadow stats case, the keys
* are cpu/type/ctx/stat (evsel is NULL). For generic metrics
* case, the keys are still evsel/cpu (type/ctx/stat are 0 or NULL).
*/
if (a->type != b->type)
return a->type - b->type;
if (a->ctx != b->ctx)
return a->ctx - b->ctx;
if (a->cgrp != b->cgrp)
return (char *)a->cgrp < (char *)b->cgrp ? -1 : +1;
if (a->evsel == NULL && b->evsel == NULL) {
if (a->stat == b->stat)
return 0;
if ((char *)a->stat < (char *)b->stat)
return -1;
return 1;
}
if (a->evsel == b->evsel)
return 0;
if ((char *)a->evsel < (char *)b->evsel)
return -1;
return +1;
}
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 void saved_value_delete(struct rblist *rblist __maybe_unused,
struct rb_node *rb_node)
{
struct saved_value *v;
BUG_ON(!rb_node);
v = container_of(rb_node, struct saved_value, rb_node);
free(v);
}
static struct saved_value *saved_value_lookup(struct evsel *evsel,
int cpu,
bool create,
enum stat_type type,
int ctx,
struct runtime_stat *st,
struct cgroup *cgrp)
{
struct rblist *rblist;
struct rb_node *nd;
struct saved_value dm = {
.cpu = cpu,
.evsel = evsel,
.type = type,
.ctx = ctx,
.stat = st,
.cgrp = cgrp,
};
rblist = &st->value_list;
/* don't use context info for clock events */
if (type == STAT_NSECS)
dm.ctx = 0;
nd = rblist__find(rblist, &dm);
if (nd)
return container_of(nd, struct saved_value, rb_node);
if (create) {
rblist__add_node(rblist, &dm);
nd = rblist__find(rblist, &dm);
if (nd)
return container_of(nd, struct saved_value, rb_node);
}
return NULL;
}
void runtime_stat__init(struct runtime_stat *st)
{
struct rblist *rblist = &st->value_list;
rblist__init(rblist);
rblist->node_cmp = saved_value_cmp;
rblist->node_new = saved_value_new;
rblist->node_delete = saved_value_delete;
}
void runtime_stat__exit(struct runtime_stat *st)
{
rblist__exit(&st->value_list);
}
void perf_stat__init_shadow_stats(void)
{
runtime_stat__init(&rt_stat);
}
static int evsel_context(struct evsel *evsel)
{
int ctx = 0;
if (evsel->core.attr.exclude_kernel)
ctx |= CTX_BIT_KERNEL;
if (evsel->core.attr.exclude_user)
ctx |= CTX_BIT_USER;
if (evsel->core.attr.exclude_hv)
ctx |= CTX_BIT_HV;
if (evsel->core.attr.exclude_host)
ctx |= CTX_BIT_HOST;
if (evsel->core.attr.exclude_idle)
ctx |= CTX_BIT_IDLE;
return ctx;
}
static void reset_stat(struct runtime_stat *st)
{
struct rblist *rblist;
struct rb_node *pos, *next;
rblist = &st->value_list;
next = rb_first_cached(&rblist->entries);
while (next) {
pos = next;
next = rb_next(pos);
memset(&container_of(pos, struct saved_value, rb_node)->stats,
0,
sizeof(struct stats));
}
}
void perf_stat__reset_shadow_stats(void)
{
reset_stat(&rt_stat);
memset(&walltime_nsecs_stats, 0, sizeof(walltime_nsecs_stats));
}
void perf_stat__reset_shadow_per_stat(struct runtime_stat *st)
{
reset_stat(st);
}
struct runtime_stat_data {
int ctx;
struct cgroup *cgrp;
};
static void update_runtime_stat(struct runtime_stat *st,
enum stat_type type,
int cpu, u64 count,
struct runtime_stat_data *rsd)
{
struct saved_value *v = saved_value_lookup(NULL, cpu, true, type,
rsd->ctx, st, rsd->cgrp);
if (v)
update_stats(&v->stats, count);
}
/*
* Update various tracking values we maintain to print
* more semantic information such as miss/hit ratios,
* instruction rates, etc:
*/
void perf_stat__update_shadow_stats(struct evsel *counter, u64 count,
int cpu, struct runtime_stat *st)
{
u64 count_ns = count;
struct saved_value *v;
struct runtime_stat_data rsd = {
.ctx = evsel_context(counter),
.cgrp = counter->cgrp,
};
count *= counter->scale;
if (evsel__is_clock(counter))
update_runtime_stat(st, STAT_NSECS, cpu, count_ns, &rsd);
else if (evsel__match(counter, HARDWARE, HW_CPU_CYCLES))
update_runtime_stat(st, STAT_CYCLES, cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, CYCLES_IN_TX))
update_runtime_stat(st, STAT_CYCLES_IN_TX, cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TRANSACTION_START))
update_runtime_stat(st, STAT_TRANSACTION, cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, ELISION_START))
update_runtime_stat(st, STAT_ELISION, cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_TOTAL_SLOTS))
update_runtime_stat(st, STAT_TOPDOWN_TOTAL_SLOTS,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_SLOTS_ISSUED))
update_runtime_stat(st, STAT_TOPDOWN_SLOTS_ISSUED,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_SLOTS_RETIRED))
update_runtime_stat(st, STAT_TOPDOWN_SLOTS_RETIRED,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_FETCH_BUBBLES))
update_runtime_stat(st, STAT_TOPDOWN_FETCH_BUBBLES,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_RECOVERY_BUBBLES))
update_runtime_stat(st, STAT_TOPDOWN_RECOVERY_BUBBLES,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_RETIRING))
update_runtime_stat(st, STAT_TOPDOWN_RETIRING,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_BAD_SPEC))
update_runtime_stat(st, STAT_TOPDOWN_BAD_SPEC,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_FE_BOUND))
update_runtime_stat(st, STAT_TOPDOWN_FE_BOUND,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_BE_BOUND))
update_runtime_stat(st, STAT_TOPDOWN_BE_BOUND,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_HEAVY_OPS))
update_runtime_stat(st, STAT_TOPDOWN_HEAVY_OPS,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_BR_MISPREDICT))
update_runtime_stat(st, STAT_TOPDOWN_BR_MISPREDICT,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_FETCH_LAT))
update_runtime_stat(st, STAT_TOPDOWN_FETCH_LAT,
cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, TOPDOWN_MEM_BOUND))
update_runtime_stat(st, STAT_TOPDOWN_MEM_BOUND,
cpu, count, &rsd);
else if (evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND))
update_runtime_stat(st, STAT_STALLED_CYCLES_FRONT,
cpu, count, &rsd);
else if (evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND))
update_runtime_stat(st, STAT_STALLED_CYCLES_BACK,
cpu, count, &rsd);
else if (evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS))
update_runtime_stat(st, STAT_BRANCHES, cpu, count, &rsd);
else if (evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES))
update_runtime_stat(st, STAT_CACHEREFS, cpu, count, &rsd);
else if (evsel__match(counter, HW_CACHE, HW_CACHE_L1D))
update_runtime_stat(st, STAT_L1_DCACHE, cpu, count, &rsd);
else if (evsel__match(counter, HW_CACHE, HW_CACHE_L1I))
update_runtime_stat(st, STAT_L1_ICACHE, cpu, count, &rsd);
else if (evsel__match(counter, HW_CACHE, HW_CACHE_LL))
update_runtime_stat(st, STAT_LL_CACHE, cpu, count, &rsd);
else if (evsel__match(counter, HW_CACHE, HW_CACHE_DTLB))
update_runtime_stat(st, STAT_DTLB_CACHE, cpu, count, &rsd);
else if (evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
update_runtime_stat(st, STAT_ITLB_CACHE, cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, SMI_NUM))
update_runtime_stat(st, STAT_SMI_NUM, cpu, count, &rsd);
else if (perf_stat_evsel__is(counter, APERF))
update_runtime_stat(st, STAT_APERF, cpu, count, &rsd);
if (counter->collect_stat) {
v = saved_value_lookup(counter, cpu, true, STAT_NONE, 0, st,
rsd.cgrp);
update_stats(&v->stats, count);
if (counter->metric_leader)
v->metric_total += count;
} else if (counter->metric_leader) {
v = saved_value_lookup(counter->metric_leader,
cpu, true, STAT_NONE, 0, st, rsd.cgrp);
v->metric_total += count;
v->metric_other++;
}
}
/* used for get_ratio_color() */
enum grc_type {
GRC_STALLED_CYCLES_FE,
GRC_STALLED_CYCLES_BE,
GRC_CACHE_MISSES,
GRC_MAX_NR
};
static const char *get_ratio_color(enum grc_type type, double ratio)
{
static const double grc_table[GRC_MAX_NR][3] = {
[GRC_STALLED_CYCLES_FE] = { 50.0, 30.0, 10.0 },
[GRC_STALLED_CYCLES_BE] = { 75.0, 50.0, 20.0 },
[GRC_CACHE_MISSES] = { 20.0, 10.0, 5.0 },
};
const char *color = PERF_COLOR_NORMAL;
if (ratio > grc_table[type][0])
color = PERF_COLOR_RED;
else if (ratio > grc_table[type][1])
color = PERF_COLOR_MAGENTA;
else if (ratio > grc_table[type][2])
color = PERF_COLOR_YELLOW;
return color;
}
static struct evsel *perf_stat__find_event(struct evlist *evsel_list,
const char *name)
{
struct evsel *c2;
evlist__for_each_entry (evsel_list, c2) {
if (!strcasecmp(c2->name, name) && !c2->collect_stat)
return c2;
}
return NULL;
}
/* Mark MetricExpr target events and link events using them to them. */
void perf_stat__collect_metric_expr(struct evlist *evsel_list)
{
struct evsel *counter, *leader, **metric_events, *oc;
bool found;
struct expr_parse_ctx ctx;
struct hashmap_entry *cur;
size_t bkt;
int i;
expr__ctx_init(&ctx);
evlist__for_each_entry(evsel_list, counter) {
bool invalid = false;
leader = counter->leader;
if (!counter->metric_expr)
continue;
expr__ctx_clear(&ctx);
metric_events = counter->metric_events;
if (!metric_events) {
if (expr__find_other(counter->metric_expr,
counter->name,
&ctx, 1) < 0)
continue;
metric_events = calloc(sizeof(struct evsel *),
hashmap__size(&ctx.ids) + 1);
if (!metric_events) {
expr__ctx_clear(&ctx);
return;
}
counter->metric_events = metric_events;
}
i = 0;
hashmap__for_each_entry((&ctx.ids), cur, bkt) {
const char *metric_name = (const char *)cur->key;
found = false;
if (leader) {
/* Search in group */
for_each_group_member (oc, leader) {
if (!strcasecmp(oc->name,
metric_name) &&
!oc->collect_stat) {
found = true;
break;
}
}
}
if (!found) {
/* Search ignoring groups */
oc = perf_stat__find_event(evsel_list,
metric_name);
}
if (!oc) {
/* Deduping one is good enough to handle duplicated PMUs. */
static char *printed;
/*
* Adding events automatically would be difficult, because
* it would risk creating groups that are not schedulable.
* perf stat doesn't understand all the scheduling constraints
* of events. So we ask the user instead to add the missing
* events.
*/
if (!printed ||
strcasecmp(printed, metric_name)) {
fprintf(stderr,
"Add %s event to groups to get metric expression for %s\n",
metric_name,
counter->name);
printed = strdup(metric_name);
}
invalid = true;
continue;
}
metric_events[i++] = oc;
oc->collect_stat = true;
}
metric_events[i] = NULL;
if (invalid) {
free(metric_events);
counter->metric_events = NULL;
counter->metric_expr = NULL;
}
}
expr__ctx_clear(&ctx);
}
static double runtime_stat_avg(struct runtime_stat *st,
enum stat_type type, int cpu,
struct runtime_stat_data *rsd)
{
struct saved_value *v;
v = saved_value_lookup(NULL, cpu, false, type, rsd->ctx, st, rsd->cgrp);
if (!v)
return 0.0;
return avg_stats(&v->stats);
}
static double runtime_stat_n(struct runtime_stat *st,
enum stat_type type, int cpu,
struct runtime_stat_data *rsd)
{
struct saved_value *v;
v = saved_value_lookup(NULL, cpu, false, type, rsd->ctx, st, rsd->cgrp);
if (!v)
return 0.0;
return v->stats.n;
}
static void print_stalled_cycles_frontend(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_CYCLES, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_STALLED_CYCLES_FE, ratio);
if (ratio)
out->print_metric(config, out->ctx, color, "%7.2f%%", "frontend cycles idle",
ratio);
else
out->print_metric(config, out->ctx, NULL, NULL, "frontend cycles idle", 0);
}
static void print_stalled_cycles_backend(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_CYCLES, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_STALLED_CYCLES_BE, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "backend cycles idle", ratio);
}
static void print_branch_misses(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_BRANCHES, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_CACHE_MISSES, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "of all branches", ratio);
}
static void print_l1_dcache_misses(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_L1_DCACHE, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_CACHE_MISSES, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "of all L1-dcache accesses", ratio);
}
static void print_l1_icache_misses(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_L1_ICACHE, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_CACHE_MISSES, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "of all L1-icache accesses", ratio);
}
static void print_dtlb_cache_misses(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_DTLB_CACHE, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_CACHE_MISSES, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "of all dTLB cache accesses", ratio);
}
static void print_itlb_cache_misses(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_ITLB_CACHE, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_CACHE_MISSES, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "of all iTLB cache accesses", ratio);
}
static void print_ll_cache_misses(struct perf_stat_config *config,
int cpu, double avg,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double total, ratio = 0.0;
const char *color;
total = runtime_stat_avg(st, STAT_LL_CACHE, cpu, rsd);
if (total)
ratio = avg / total * 100.0;
color = get_ratio_color(GRC_CACHE_MISSES, ratio);
out->print_metric(config, out->ctx, color, "%7.2f%%", "of all LL-cache accesses", ratio);
}
/*
* High level "TopDown" CPU core pipe line bottleneck break down.
*
* Basic concept following
* Yasin, A Top Down Method for Performance analysis and Counter architecture
* ISPASS14
*
* The CPU pipeline is divided into 4 areas that can be bottlenecks:
*
* Frontend -> Backend -> Retiring
* BadSpeculation in addition means out of order execution that is thrown away
* (for example branch mispredictions)
* Frontend is instruction decoding.
* Backend is execution, like computation and accessing data in memory
* Retiring is good execution that is not directly bottlenecked
*
* The formulas are computed in slots.
* A slot is an entry in the pipeline each for the pipeline width
* (for example a 4-wide pipeline has 4 slots for each cycle)
*
* Formulas:
* BadSpeculation = ((SlotsIssued - SlotsRetired) + RecoveryBubbles) /
* TotalSlots
* Retiring = SlotsRetired / TotalSlots
* FrontendBound = FetchBubbles / TotalSlots
* BackendBound = 1.0 - BadSpeculation - Retiring - FrontendBound
*
* The kernel provides the mapping to the low level CPU events and any scaling
* needed for the CPU pipeline width, for example:
*
* TotalSlots = Cycles * 4
*
* The scaling factor is communicated in the sysfs unit.
*
* In some cases the CPU may not be able to measure all the formulas due to
* missing events. In this case multiple formulas are combined, as possible.
*
* Full TopDown supports more levels to sub-divide each area: for example
* BackendBound into computing bound and memory bound. For now we only
* support Level 1 TopDown.
*/
static double sanitize_val(double x)
{
if (x < 0 && x >= -0.02)
return 0.0;
return x;
}
static double td_total_slots(int cpu, struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
return runtime_stat_avg(st, STAT_TOPDOWN_TOTAL_SLOTS, cpu, rsd);
}
static double td_bad_spec(int cpu, struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double bad_spec = 0;
double total_slots;
double total;
total = runtime_stat_avg(st, STAT_TOPDOWN_SLOTS_ISSUED, cpu, rsd) -
runtime_stat_avg(st, STAT_TOPDOWN_SLOTS_RETIRED, cpu, rsd) +
runtime_stat_avg(st, STAT_TOPDOWN_RECOVERY_BUBBLES, cpu, rsd);
total_slots = td_total_slots(cpu, st, rsd);
if (total_slots)
bad_spec = total / total_slots;
return sanitize_val(bad_spec);
}
static double td_retiring(int cpu, struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double retiring = 0;
double total_slots = td_total_slots(cpu, st, rsd);
double ret_slots = runtime_stat_avg(st, STAT_TOPDOWN_SLOTS_RETIRED,
cpu, rsd);
if (total_slots)
retiring = ret_slots / total_slots;
return retiring;
}
static double td_fe_bound(int cpu, struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double fe_bound = 0;
double total_slots = td_total_slots(cpu, st, rsd);
double fetch_bub = runtime_stat_avg(st, STAT_TOPDOWN_FETCH_BUBBLES,
cpu, rsd);
if (total_slots)
fe_bound = fetch_bub / total_slots;
return fe_bound;
}
static double td_be_bound(int cpu, struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double sum = (td_fe_bound(cpu, st, rsd) +
td_bad_spec(cpu, st, rsd) +
td_retiring(cpu, st, rsd));
if (sum == 0)
return 0;
return sanitize_val(1.0 - sum);
}
/*
* Kernel reports metrics multiplied with slots. To get back
* the ratios we need to recreate the sum.
*/
static double td_metric_ratio(int cpu, enum stat_type type,
struct runtime_stat *stat,
struct runtime_stat_data *rsd)
{
double sum = runtime_stat_avg(stat, STAT_TOPDOWN_RETIRING, cpu, rsd) +
runtime_stat_avg(stat, STAT_TOPDOWN_FE_BOUND, cpu, rsd) +
runtime_stat_avg(stat, STAT_TOPDOWN_BE_BOUND, cpu, rsd) +
runtime_stat_avg(stat, STAT_TOPDOWN_BAD_SPEC, cpu, rsd);
double d = runtime_stat_avg(stat, type, cpu, rsd);
if (sum)
return d / sum;
return 0;
}
/*
* ... but only if most of the values are actually available.
* We allow two missing.
*/
static bool full_td(int cpu, struct runtime_stat *stat,
struct runtime_stat_data *rsd)
{
int c = 0;
if (runtime_stat_avg(stat, STAT_TOPDOWN_RETIRING, cpu, rsd) > 0)
c++;
if (runtime_stat_avg(stat, STAT_TOPDOWN_BE_BOUND, cpu, rsd) > 0)
c++;
if (runtime_stat_avg(stat, STAT_TOPDOWN_FE_BOUND, cpu, rsd) > 0)
c++;
if (runtime_stat_avg(stat, STAT_TOPDOWN_BAD_SPEC, cpu, rsd) > 0)
c++;
return c >= 2;
}
static void print_smi_cost(struct perf_stat_config *config, int cpu,
struct perf_stat_output_ctx *out,
struct runtime_stat *st,
struct runtime_stat_data *rsd)
{
double smi_num, aperf, cycles, cost = 0.0;
const char *color = NULL;
smi_num = runtime_stat_avg(st, STAT_SMI_NUM, cpu, rsd);
aperf = runtime_stat_avg(st, STAT_APERF, cpu, rsd);
cycles = runtime_stat_avg(st, STAT_CYCLES, cpu, rsd);
if ((cycles == 0) || (aperf == 0))
return;
if (smi_num)
cost = (aperf - cycles) / aperf * 100.00;
if (cost > 10)
color = PERF_COLOR_RED;
out->print_metric(config, out->ctx, color, "%8.1f%%", "SMI cycles%", cost);
out->print_metric(config, out->ctx, NULL, "%4.0f", "SMI#", smi_num);
}
static int prepare_metric(struct evsel **metric_events,
struct metric_ref *metric_refs,
struct expr_parse_ctx *pctx,
int cpu,
struct runtime_stat *st)
{
double scale;
char *n, *pn;
int i, j, ret;
expr__ctx_init(pctx);
for (i = 0; metric_events[i]; i++) {
struct saved_value *v;
struct stats *stats;
u64 metric_total = 0;
if (!strcmp(metric_events[i]->name, "duration_time")) {
stats = &walltime_nsecs_stats;
scale = 1e-9;
} else {
v = saved_value_lookup(metric_events[i], cpu, false,
STAT_NONE, 0, st,
metric_events[i]->cgrp);
if (!v)
break;
stats = &v->stats;
scale = 1.0;
if (v->metric_other)
metric_total = v->metric_total;
}
n = strdup(metric_events[i]->name);
if (!n)
return -ENOMEM;
/*
* This display code with --no-merge adds [cpu] postfixes.
* These are not supported by the parser. Remove everything
* after the space.
*/
pn = strchr(n, ' ');
if (pn)
*pn = 0;
if (metric_total)
expr__add_id_val(pctx, n, metric_total);
else
expr__add_id_val(pctx, n, avg_stats(stats)*scale);
}
for (j = 0; metric_refs && metric_refs[j].metric_name; j++) {
ret = expr__add_ref(pctx, &metric_refs[j]);
if (ret)
return ret;
}
return i;
}
static void generic_metric(struct perf_stat_config *config,
const char *metric_expr,
struct evsel **metric_events,
struct metric_ref *metric_refs,
char *name,
const char *metric_name,
const char *metric_unit,
int runtime,
int cpu,
struct perf_stat_output_ctx *out,
struct runtime_stat *st)
{
print_metric_t print_metric = out->print_metric;
struct expr_parse_ctx pctx;
double ratio, scale;
int i;
void *ctxp = out->ctx;
i = prepare_metric(metric_events, metric_refs, &pctx, cpu, st);
if (i < 0)
return;
if (!metric_events[i]) {
if (expr__parse(&ratio, &pctx, metric_expr, runtime) == 0) {
char *unit;
char metric_bf[64];
if (metric_unit && metric_name) {
if (perf_pmu__convert_scale(metric_unit,
&unit, &scale) >= 0) {
ratio *= scale;
}
if (strstr(metric_expr, "?"))
scnprintf(metric_bf, sizeof(metric_bf),
"%s %s_%d", unit, metric_name, runtime);
else
scnprintf(metric_bf, sizeof(metric_bf),
"%s %s", unit, metric_name);
print_metric(config, ctxp, NULL, "%8.1f",
metric_bf, ratio);
} else {
print_metric(config, ctxp, NULL, "%8.2f",
metric_name ?
metric_name :
out->force_header ? name : "",
ratio);
}
} else {
print_metric(config, ctxp, NULL, NULL,
out->force_header ?
(metric_name ? metric_name : name) : "", 0);
}
} else {
print_metric(config, ctxp, NULL, NULL,
out->force_header ?
(metric_name ? metric_name : name) : "", 0);
}
expr__ctx_clear(&pctx);
}
double test_generic_metric(struct metric_expr *mexp, int cpu, struct runtime_stat *st)
{
struct expr_parse_ctx pctx;
double ratio = 0.0;
if (prepare_metric(mexp->metric_events, mexp->metric_refs, &pctx, cpu, st) < 0)
goto out;
if (expr__parse(&ratio, &pctx, mexp->metric_expr, 1))
ratio = 0.0;
out:
expr__ctx_clear(&pctx);
return ratio;
}
void perf_stat__print_shadow_stats(struct perf_stat_config *config,
struct evsel *evsel,
double avg, int cpu,
struct perf_stat_output_ctx *out,
struct rblist *metric_events,
struct runtime_stat *st)
{
void *ctxp = out->ctx;
print_metric_t print_metric = out->print_metric;
double total, ratio = 0.0, total2;
const char *color = NULL;
struct runtime_stat_data rsd = {
.ctx = evsel_context(evsel),
.cgrp = evsel->cgrp,
};
struct metric_event *me;
int num = 1;
if (evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) {
total = runtime_stat_avg(st, STAT_CYCLES, cpu, &rsd);
if (total) {
ratio = avg / total;
print_metric(config, ctxp, NULL, "%7.2f ",
"insn per cycle", ratio);
} else {
print_metric(config, ctxp, NULL, NULL, "insn per cycle", 0);
}
total = runtime_stat_avg(st, STAT_STALLED_CYCLES_FRONT, cpu, &rsd);
total = max(total, runtime_stat_avg(st,
STAT_STALLED_CYCLES_BACK,
cpu, &rsd));
if (total && avg) {
out->new_line(config, ctxp);
ratio = total / avg;
print_metric(config, ctxp, NULL, "%7.2f ",
"stalled cycles per insn",
ratio);
}
} else if (evsel__match(evsel, HARDWARE, HW_BRANCH_MISSES)) {
if (runtime_stat_n(st, STAT_BRANCHES, cpu, &rsd) != 0)
print_branch_misses(config, cpu, avg, out, st, &rsd);
else
print_metric(config, ctxp, NULL, NULL, "of all branches", 0);
} else if (
evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_L1D |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16))) {
if (runtime_stat_n(st, STAT_L1_DCACHE, cpu, &rsd) != 0)
print_l1_dcache_misses(config, cpu, avg, out, st, &rsd);
else
print_metric(config, ctxp, NULL, NULL, "of all L1-dcache accesses", 0);
} else if (
evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_L1I |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16))) {
if (runtime_stat_n(st, STAT_L1_ICACHE, cpu, &rsd) != 0)
print_l1_icache_misses(config, cpu, avg, out, st, &rsd);
else
print_metric(config, ctxp, NULL, NULL, "of all L1-icache accesses", 0);
} else if (
evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_DTLB |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16))) {
if (runtime_stat_n(st, STAT_DTLB_CACHE, cpu, &rsd) != 0)
print_dtlb_cache_misses(config, cpu, avg, out, st, &rsd);
else
print_metric(config, ctxp, NULL, NULL, "of all dTLB cache accesses", 0);
} else if (
evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_ITLB |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16))) {
if (runtime_stat_n(st, STAT_ITLB_CACHE, cpu, &rsd) != 0)
print_itlb_cache_misses(config, cpu, avg, out, st, &rsd);
else
print_metric(config, ctxp, NULL, NULL, "of all iTLB cache accesses", 0);
} else if (
evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_LL |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16))) {
if (runtime_stat_n(st, STAT_LL_CACHE, cpu, &rsd) != 0)
print_ll_cache_misses(config, cpu, avg, out, st, &rsd);
else
print_metric(config, ctxp, NULL, NULL, "of all LL-cache accesses", 0);
} else if (evsel__match(evsel, HARDWARE, HW_CACHE_MISSES)) {
total = runtime_stat_avg(st, STAT_CACHEREFS, cpu, &rsd);
if (total)
ratio = avg * 100 / total;
if (runtime_stat_n(st, STAT_CACHEREFS, cpu, &rsd) != 0)
print_metric(config, ctxp, NULL, "%8.3f %%",
"of all cache refs", ratio);
else
print_metric(config, ctxp, NULL, NULL, "of all cache refs", 0);
} else if (evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) {
print_stalled_cycles_frontend(config, cpu, avg, out, st, &rsd);
} else if (evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_BACKEND)) {
print_stalled_cycles_backend(config, cpu, avg, out, st, &rsd);
} else if (evsel__match(evsel, HARDWARE, HW_CPU_CYCLES)) {
total = runtime_stat_avg(st, STAT_NSECS, cpu, &rsd);
if (total) {
ratio = avg / total;
print_metric(config, ctxp, NULL, "%8.3f", "GHz", ratio);
} else {
print_metric(config, ctxp, NULL, NULL, "Ghz", 0);
}
} else if (perf_stat_evsel__is(evsel, CYCLES_IN_TX)) {
total = runtime_stat_avg(st, STAT_CYCLES, cpu, &rsd);
if (total)
print_metric(config, ctxp, NULL,
"%7.2f%%", "transactional cycles",
100.0 * (avg / total));
else
print_metric(config, ctxp, NULL, NULL, "transactional cycles",
0);
} else if (perf_stat_evsel__is(evsel, CYCLES_IN_TX_CP)) {
total = runtime_stat_avg(st, STAT_CYCLES, cpu, &rsd);
total2 = runtime_stat_avg(st, STAT_CYCLES_IN_TX, cpu, &rsd);
if (total2 < avg)
total2 = avg;
if (total)
print_metric(config, ctxp, NULL, "%7.2f%%", "aborted cycles",
100.0 * ((total2-avg) / total));
else
print_metric(config, ctxp, NULL, NULL, "aborted cycles", 0);
} else if (perf_stat_evsel__is(evsel, TRANSACTION_START)) {
total = runtime_stat_avg(st, STAT_CYCLES_IN_TX, cpu, &rsd);
if (avg)
ratio = total / avg;
if (runtime_stat_n(st, STAT_CYCLES_IN_TX, cpu, &rsd) != 0)
print_metric(config, ctxp, NULL, "%8.0f",
"cycles / transaction", ratio);
else
print_metric(config, ctxp, NULL, NULL, "cycles / transaction",
0);
} else if (perf_stat_evsel__is(evsel, ELISION_START)) {
total = runtime_stat_avg(st, STAT_CYCLES_IN_TX, cpu, &rsd);
if (avg)
ratio = total / avg;
print_metric(config, ctxp, NULL, "%8.0f", "cycles / elision", ratio);
} else if (evsel__is_clock(evsel)) {
if ((ratio = avg_stats(&walltime_nsecs_stats)) != 0)
print_metric(config, ctxp, NULL, "%8.3f", "CPUs utilized",
avg / (ratio * evsel->scale));
else
print_metric(config, ctxp, NULL, NULL, "CPUs utilized", 0);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_FETCH_BUBBLES)) {
double fe_bound = td_fe_bound(cpu, st, &rsd);
if (fe_bound > 0.2)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "frontend bound",
fe_bound * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_SLOTS_RETIRED)) {
double retiring = td_retiring(cpu, st, &rsd);
if (retiring > 0.7)
color = PERF_COLOR_GREEN;
print_metric(config, ctxp, color, "%8.1f%%", "retiring",
retiring * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_RECOVERY_BUBBLES)) {
double bad_spec = td_bad_spec(cpu, st, &rsd);
if (bad_spec > 0.1)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "bad speculation",
bad_spec * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_SLOTS_ISSUED)) {
double be_bound = td_be_bound(cpu, st, &rsd);
const char *name = "backend bound";
static int have_recovery_bubbles = -1;
/* In case the CPU does not support topdown-recovery-bubbles */
if (have_recovery_bubbles < 0)
have_recovery_bubbles = pmu_have_event("cpu",
"topdown-recovery-bubbles");
if (!have_recovery_bubbles)
name = "backend bound/bad spec";
if (be_bound > 0.2)
color = PERF_COLOR_RED;
if (td_total_slots(cpu, st, &rsd) > 0)
print_metric(config, ctxp, color, "%8.1f%%", name,
be_bound * 100.);
else
print_metric(config, ctxp, NULL, NULL, name, 0);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_RETIRING) &&
full_td(cpu, st, &rsd)) {
double retiring = td_metric_ratio(cpu,
STAT_TOPDOWN_RETIRING, st,
&rsd);
if (retiring > 0.7)
color = PERF_COLOR_GREEN;
print_metric(config, ctxp, color, "%8.1f%%", "retiring",
retiring * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_FE_BOUND) &&
full_td(cpu, st, &rsd)) {
double fe_bound = td_metric_ratio(cpu,
STAT_TOPDOWN_FE_BOUND, st,
&rsd);
if (fe_bound > 0.2)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "frontend bound",
fe_bound * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_BE_BOUND) &&
full_td(cpu, st, &rsd)) {
double be_bound = td_metric_ratio(cpu,
STAT_TOPDOWN_BE_BOUND, st,
&rsd);
if (be_bound > 0.2)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "backend bound",
be_bound * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_BAD_SPEC) &&
full_td(cpu, st, &rsd)) {
double bad_spec = td_metric_ratio(cpu,
STAT_TOPDOWN_BAD_SPEC, st,
&rsd);
if (bad_spec > 0.1)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "bad speculation",
bad_spec * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_HEAVY_OPS) &&
full_td(cpu, st, &rsd) && (config->topdown_level > 1)) {
double retiring = td_metric_ratio(cpu,
STAT_TOPDOWN_RETIRING, st,
&rsd);
double heavy_ops = td_metric_ratio(cpu,
STAT_TOPDOWN_HEAVY_OPS, st,
&rsd);
double light_ops = retiring - heavy_ops;
if (retiring > 0.7 && heavy_ops > 0.1)
color = PERF_COLOR_GREEN;
print_metric(config, ctxp, color, "%8.1f%%", "heavy operations",
heavy_ops * 100.);
if (retiring > 0.7 && light_ops > 0.6)
color = PERF_COLOR_GREEN;
else
color = NULL;
print_metric(config, ctxp, color, "%8.1f%%", "light operations",
light_ops * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_BR_MISPREDICT) &&
full_td(cpu, st, &rsd) && (config->topdown_level > 1)) {
double bad_spec = td_metric_ratio(cpu,
STAT_TOPDOWN_BAD_SPEC, st,
&rsd);
double br_mis = td_metric_ratio(cpu,
STAT_TOPDOWN_BR_MISPREDICT, st,
&rsd);
double m_clears = bad_spec - br_mis;
if (bad_spec > 0.1 && br_mis > 0.05)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "branch mispredict",
br_mis * 100.);
if (bad_spec > 0.1 && m_clears > 0.05)
color = PERF_COLOR_RED;
else
color = NULL;
print_metric(config, ctxp, color, "%8.1f%%", "machine clears",
m_clears * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_FETCH_LAT) &&
full_td(cpu, st, &rsd) && (config->topdown_level > 1)) {
double fe_bound = td_metric_ratio(cpu,
STAT_TOPDOWN_FE_BOUND, st,
&rsd);
double fetch_lat = td_metric_ratio(cpu,
STAT_TOPDOWN_FETCH_LAT, st,
&rsd);
double fetch_bw = fe_bound - fetch_lat;
if (fe_bound > 0.2 && fetch_lat > 0.15)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "fetch latency",
fetch_lat * 100.);
if (fe_bound > 0.2 && fetch_bw > 0.1)
color = PERF_COLOR_RED;
else
color = NULL;
print_metric(config, ctxp, color, "%8.1f%%", "fetch bandwidth",
fetch_bw * 100.);
} else if (perf_stat_evsel__is(evsel, TOPDOWN_MEM_BOUND) &&
full_td(cpu, st, &rsd) && (config->topdown_level > 1)) {
double be_bound = td_metric_ratio(cpu,
STAT_TOPDOWN_BE_BOUND, st,
&rsd);
double mem_bound = td_metric_ratio(cpu,
STAT_TOPDOWN_MEM_BOUND, st,
&rsd);
double core_bound = be_bound - mem_bound;
if (be_bound > 0.2 && mem_bound > 0.2)
color = PERF_COLOR_RED;
print_metric(config, ctxp, color, "%8.1f%%", "memory bound",
mem_bound * 100.);
if (be_bound > 0.2 && core_bound > 0.1)
color = PERF_COLOR_RED;
else
color = NULL;
print_metric(config, ctxp, color, "%8.1f%%", "Core bound",
core_bound * 100.);
} else if (evsel->metric_expr) {
generic_metric(config, evsel->metric_expr, evsel->metric_events, NULL,
evsel->name, evsel->metric_name, NULL, 1, cpu, out, st);
} else if (runtime_stat_n(st, STAT_NSECS, cpu, &rsd) != 0) {
char unit = 'M';
char unit_buf[10];
total = runtime_stat_avg(st, STAT_NSECS, cpu, &rsd);
if (total)
ratio = 1000.0 * avg / total;
if (ratio < 0.001) {
ratio *= 1000;
unit = 'K';
}
snprintf(unit_buf, sizeof(unit_buf), "%c/sec", unit);
print_metric(config, ctxp, NULL, "%8.3f", unit_buf, ratio);
} else if (perf_stat_evsel__is(evsel, SMI_NUM)) {
print_smi_cost(config, cpu, out, st, &rsd);
} else {
num = 0;
}
if ((me = metricgroup__lookup(metric_events, evsel, false)) != NULL) {
struct metric_expr *mexp;
list_for_each_entry (mexp, &me->head, nd) {
if (num++ > 0)
out->new_line(config, ctxp);
generic_metric(config, mexp->metric_expr, mexp->metric_events,
mexp->metric_refs, evsel->name, mexp->metric_name,
mexp->metric_unit, mexp->runtime, cpu, out, st);
}
}
if (num == 0)
print_metric(config, ctxp, NULL, NULL, NULL, 0);
}