perf test: Add debug line to diagnose broken metrics
Ian Rogers
reviewerPATCH 1/2 perf test: Add debug line to diagnose broken metrics
Printing out the metric name and architecture makes finding the source of a failure easier.
Signed-off-by Ian Rogers <[email protected]>
--- tools/perf/tests/pmu-events.c | 1 + 1 file changed, 1 insertion(+)
tools/perf/tests/pmu-events.c +1 -0
@@ -1115,6 +1115,7 @@ static int test__parsing_fake(struct test_suite *test __maybe_unused,break ;if (!pe->metric_expr)continue ; +pr_debug ("Found metric '%s' for '%s'\n" , pe->metric_name, map->cpuid); err =metric_parse_fake (pe->metric_expr);if (err)return err;
Signature · 1 line
2.37.0.rc0.161.g10f37bed90-goog
Ian Rogers
reviewerPATCH 2/2 perf test: Make all metrics test more tolerant
Metric names are truncated so don't try to match all of one. Allow AMX metrics to skip as floating point ones do. Metrics for optane memory can also skip rather than fail. Add a system wide check for uncore metrics. Restructure code to avoid extensive nesting.
Some impetus for this in:
https://lore.kernel.org/lkml/[email protected]/
Suggested-by John Garry <[email protected]>
Signed-off-by Ian Rogers <[email protected]>
--- tools/perf/tests/shell/stat_all_metrics.sh | 47 ++++++++++++++-------- 1 file changed, 31 insertions(+), 16 deletions(-)
tools/perf/tests/shell/stat_all_metrics.sh +31 -16
@@ -1,26 +1,41 @@ -#!/bin/sh +#!/bin/bash # perf all metrics test # SPDX-License-Identifier: GPL-2.0 -set -e - err=0 for m in $(perf list --raw-dump metrics); do echo "Testing $m" result=$(perf stat -M "$m" true 2>&1) - if [[ ! "$result" =~ "$m" ]] && [[ ! "$result" =~ "<not supported>" ]]; then - # We failed to see the metric and the events are support. Possibly the - # workload was too small so retry with something longer. - result=$(perf stat -M "$m" perf bench internals synthesize 2>&1) - if [[ ! "$result" =~ "$m" ]]; then - echo "Metric '$m' not printed in:" - echo "$result" - if [[ "$result" =~ "FP_ARITH" && "$err" != "1" ]]; then - echo "Skip, not fail, for FP issues" - err=2 - else - err=1 - fi + if [[ "$result" =~ "${m:0:50}" ]] || [[ "$result" =~ "<not supported>" ]] + then + continue + fi + # Failed so try system wide. + result=$(perf stat -M "$m" -a true 2>&1) + if [[ "$result" =~ "${m:0:50}" ]] + then + continue + fi + # Failed again, possibly the workload was too small so retry with something + # longer. + result=$(perf stat -M "$m" perf bench internals synthesize 2>&1) + if [[ "$result" =~ "${m:0:50}" ]] + then + continue + fi + echo "Metric '$m' not printed in:" + echo "$result" + if [[ "$err" != "1" ]] + then + err=2 + if [[ "$result" =~ "FP_ARITH" || "$result" =~ "AMX" ]] + then + echo "Skip, not fail, for FP issues" + elif [[ "$result" =~ "PMM" ]] + then + echo "Skip, not fail, for Optane memory issues" + else + err=1 fi fi done
Signature · 1 line
2.37.0.rc0.161.g10f37bed90-goog
Em Thu, Jul 07, 2022 at 08:34:48AM -0700, Ian Rogers escreveu:
Quoted 2 lines
Printing out the metric name and architecture makes finding the source of a failure easier.
Thanks, applied both patches.
- Arnaldo
Quoted 19 lines
Signed-off-by Ian Rogers <[email protected]>--- tools/perf/tests/pmu-events.c | 1 + 1 file changed, 1 insertion(+)tools/perf/tests/pmu-events.c +1
-0@@ -1115,6 +1115,7 @@ static int test__parsing_fake(struct test_suite *test __maybe_unused,break ;if (!pe->metric_expr)continue ; +pr_debug ("Found metric '%s' for '%s'\n" , pe->metric_name, map->cpuid); err =metric_parse_fake (pe->metric_expr);if (err)return err;Signature · 1 line
2.37.0.rc0.161.g10f37bed90-goog
Signature · 1 line
- Arnaldo