perf test: Fix perf test 114 perf record test subtest precise_max for s390
PATCH v2 1/2 perf test: Fix perf test 114 perf record test subtest precise_max for s390
On s390 the event instructions can not be used for recording. This event is only supported by perf stat.
Change the event instruction to cycles for subtest test_precise_max as suggested by James Clark.
--- tools/perf/tests/shell/record.sh | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
tools/perf/tests/shell/record.sh +2 -2
@@ -287,9 +287,9 @@ test_precise_max() { return fi # On AMD, cycles and instructions events are treated differently - if ! perf record -o "${perfdata}" -e "instructions:P" true 2> /dev/null + if ! perf record -o "${perfdata}" -e "cycles:P" true 2> /dev/null then - echo "precise_max attribute [Failed instructions:P event]" + echo "precise_max attribute [Failed cycles:P event]" err=1 return fi
Signature · 1 line
2.48.1
PATCH v2 2/2 perf test: Skip perf test 114 perf record test subtest test_leader_sampling for s390
On s390 the event instructions can not be used for recording. This event is only supported by perf stat.
Omit subtest test_leader_sampling on s390.
--- tools/perf/tests/shell/record.sh | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
tools/perf/tests/shell/record.sh +2 -1
@@ -307,7 +307,8 @@ test_system_wide test_workload test_branch_counter test_cgroup -test_leader_sampling +# On s390 event instructions can not be used for recording +[ $(uname -m) != s390x ] && test_leader_sampling test_topdown_leader_sampling test_precise_max
Signature · 1 line
2.48.1
James Clark
reviewerRe: [PATCH 2/2 v2] perf test: Skip perf test 114 perf record test subtest test_leader_sampling for s390
Quoted 15 lines · Thomas Richter
On s390 the event instructions can not be used for recording. This event is only supported by perf stat.
Omit subtest
test_leader_samplingon s390.Signed-off-by Thomas Richter <[email protected]>--- tools/perf/tests/shell/record.sh | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)tools/perf/tests/shell/record.sh +0
-0
tools/perf/tests/shell/record.sh test_system_wide · line 314
Quoted 9 lines
tools/perf/tests/shell/record.sh +2
-1@@ -307,7 +307,8 @@ test_system_wide test_workload test_branch_counter test_cgroup -test_leader_sampling +# On s390 event instructions can not be used for recording +[ $(uname -m) != s390x ] && test_leader_sampling test_topdown_leader_sampling test_precise_max
I still think we should use the cycles event instead of skipping. My intention was to use the most commonly available event with 80fd0c1eac7 ("perf tests: Make leader sampling test work without branch event"), it's not important which event is used exactly
James Clark
reviewerIs this a v2 of "[PATCH] perf test: Fix perf test 114 perf record test for s390"? Or maybe the v2 is a typo?
Quoted 17 lines · Thomas Richter
On s390 the event instructions can not be used for recording. This event is only supported by perf stat.
Change the event instruction to cycles for subtest
test_precise_maxas suggested by James Clark.Signed-off-by Thomas Richter <[email protected]>Suggested-by James Clark <[email protected]>--- tools/perf/tests/shell/record.sh | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)tools/perf/tests/shell/record.sh +0
-0
tools/perf/tests/shell/record.sh test_precise_max() { · line 290
Quoted 5 lines
tools/perf/tests/shell/record.sh +1
-1@@ -287,9 +287,9 @@ test_precise_max() { return fi # On AMD, cycles and instructions events are treated differently - if ! perf record -o "${perfdata}" -e "instructions:P" true 2> /dev/null + if ! perf record -o "${perfdata}" -e "cycles:P" true 2> /dev/null
I'm not sure if this is the right fix. Doesn't this make the test test cycles:P twice now?
I think the fix is to first test if cycles can be sampled, then test cycles:P. Then test if instructions can be sampled, then test instructions:P.
At the moment it checks the events with perf stat, and it checks them both at the same time. Obviously that doesn't work if an event works for perf stat but not perf record. And checking them both at the same time doesn't allow for one existing and not the other.
Quoted 6 lines
tools/perf/tests/shell/record.sh +1
-1then - echo "precise_max attribute [Failed instructions:P event]" + echo "precise_max attribute [Failed cycles:P event]" err=1 return fi
James Clark
reviewerQuoted 3 lines · James Clark
Is this a v2 of "[PATCH] perf test: Fix perf test 114 perf record test for s390"? Or maybe the v2 is a typo?
Nevermind, the CC list is different on the two patches that's why I got confused.
Quoted 38 lines · Thomas Richter
Quoted 17 lines
On s390 the event instructions can not be used for recording. This event is only supported by perf stat.
Change the event instruction to cycles for subtest
test_precise_maxas suggested by James Clark.Signed-off-by Thomas Richter <[email protected]>Suggested-by James Clark <[email protected]>---
tools/perf/tests/shell/record.sh| 4 ++--1 file changed, 2 insertions(+), 2 deletions(-)tools/perf/tests/shell/record.sh +0
-0@@ -287,9 +287,9 @@ test_precise_max() {Quoted 6 lines
return
fi
# On AMD, cycles and instructions events are treated differently - if ! perf record -o "${perfdata}" -e "instructions:P" true 2>/dev/null
+ if ! perf record -o "${perfdata}" -e "cycles:P" true 2>/dev/nullI'm not sure if this is the right fix. Doesn't this make the test test cycles:P twice now?
I think the fix is to first test if cycles can be sampled, then test cycles:P. Then test if instructions can be sampled, then test instructions:P.
At the moment it checks the events with perf stat, and it checks them both at the same time. Obviously that doesn't work if an event works for perf stat but not perf record. And checking them both at the same time doesn't allow for one existing and not the other.
Quoted 6 lines
then
- echo "precise_maxattribute [Failed instructions:P event]" + echo "precise_maxattribute [Failed cycles:P event]" err=1
return
fi