Re: [PATCH 2/2] perf test: Add basic stat and topdown group test
From: Liang, Kan
Date: Thu May 05 2022 - 16:27:22 EST
On 5/5/2022 2:35 PM, Ian Rogers wrote:
On Thu, May 5, 2022 at 11:19 AM Liang, Kan <kan.liang@xxxxxxxxxxxxxxx> wrote:
On 5/5/2022 11:22 AM, Ian Rogers wrote:
On Thu, May 5, 2022 at 5:12 AM Liang, Kan <kan.liang@xxxxxxxxxxxxxxx> wrote:
On 5/5/2022 12:38 AM, Ian Rogers wrote:
Add a basic stat test.
Add two tests of grouping behavior for topdown events. Topdown events
are special as they must be grouped with the slots event first.
Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
---
tools/perf/tests/shell/stat.sh | 65 ++++++++++++++++++++++++++++++++++
1 file changed, 65 insertions(+)
create mode 100755 tools/perf/tests/shell/stat.sh
diff --git a/tools/perf/tests/shell/stat.sh b/tools/perf/tests/shell/stat.sh
new file mode 100755
index 000000000000..80869ea6debc
--- /dev/null
+++ b/tools/perf/tests/shell/stat.sh
@@ -0,0 +1,65 @@
+#!/bin/sh
+# perf stat tests
+# SPDX-License-Identifier: GPL-2.0
+
+set -e
+
+err=0
+test_default_stat() {
+ echo "Basic stat command test"
+ if ! perf stat true 2>&1 | egrep -q "Performance counter stats for 'true':"
+ then
+ echo "Basic stat command test [Failed]"
+ err=1
+ return
+ fi
+ echo "Basic stat command test [Success]"
+}
+
+test_topdown_groups() {
+ # Topdown events must be grouped with the slots event first. Test that
+ # parse-events reorders this.
+ echo "Topdown event group test"
+ if ! perf stat -e '{slots,topdown-retiring}' true > /dev/null 2>&1
+ then
+ echo "Topdown event group test [Skipped event parsing failed]"
+ return
+ fi
+ if perf stat -e '{slots,topdown-retiring}' true 2>&1 | egrep -q "<not supported>"
+ then
+ echo "Topdown event group test [Failed events not supported]"
+ err=1
+ return
+ fi
+ if perf stat -e '{topdown-retiring,slots}' true 2>&1 | egrep -q "<not supported>"
+ then
+ echo "Topdown event group test [Failed slots not reordered first]"
+ err=1
+ return
+ fi
+ echo "Topdown event group test [Success]"
+}
+
+test_topdown_weak_groups() {
+ # Weak groups break if the perf_event_open of multiple grouped events
+ # fails. Breaking a topdown group causes the events to fail. Test a very large
+ # grouping to see that the topdown events aren't broken out.
+ echo "Topdown weak groups test"
+ if ! perf stat -e '{slots,topdown-bad-spec,topdown-be-bound,topdown-fe-bound,topdown-retiring},branch-instructions,branch-misses,bus-cycles,cache-misses,cache-references,cpu-cycles,instructions,mem-loads,mem-stores,ref-cycles,baclears.any,ARITH.DIVIDER_ACTIVE' true > /dev/null 2>&1
+ then
+ echo "Topdown weak groups test [Skipped event parsing failed]"
+ return
+ fi
+ if perf stat -e '{slots,topdown-bad-spec,topdown-be-bound,topdown-fe-bound,topdown-retiring,branch-instructions,branch-misses,bus-cycles,cache-misses,cache-references,cpu-cycles,instructions,mem-loads,mem-stores,ref-cycles,baclears.any,ARITH.DIVIDER_ACTIVE}:W' true 2>&1 | egrep -q "<not supported>"
+ then
+ echo "Topdown weak groups test [Failed events not supported]"
+ err=1
+ return
+ fi
+ echo "Topdown weak groups test [Success]"
+}
+
Should we check the existence of the slots event before the test?
The perf metrics feature only be available on the new platform after
ICL. It doesn't work on Atom.
Also, I think the test may fails on the hybrid platform, since big core
and small core have different formula for the topdown. I think we should
avoid the test for the hybrid platform for now.
+Zhengjun, who is fixing the topdown gap for the hybrid platform. I
think he may take care of the hybrid support later.
Thanks Kan, the test filters out systems that don't support the events
and silently skips the test. The main purpose of the test is to make
sure the somewhat complicated grouping operations for Icelake have
some coverage. Adding more coverage for hybrid would be great, but not
something I think gates this change.
Sure, we can add the coverage for hybrid later. But please make sure the
test can filter out both the systems which doesn't support perf metircs
and the hybrid system.
Thanks,
Kan
If the test fails on hybrid then that feels like value add :-)
Indeed. The test case is valid.
Could you please use architecture events to replace the
baclears.any,ARITH.DIVIDER_ACTIVE?
It's not guaranteed that the two events are supported for the future
platforms.
I use the duplicated cache-misses,cache-references events to replace
baclears.any,ARITH.DIVIDER_ACTIVE and tested on the ADL. The results
look good.
./perf stat -e
'{slots,topdown-bad-spec,topdown-be-bound,topdown-fe-bound,topdown-retiring,branch-instructions,branch-misses,bus-cycles,cache-misses,cache-references,cpu-cycles,instructions,mem-loads,mem-stores,ref-cycles,cache-misses,cache-references}:W'
true
WARNING: events in group from different hybrid PMUs!
Performance counter stats for 'true':
3,603,798 cpu_core/slots/
480,506 cpu_core/topdown-bad-spec/ # 13.3% bad
speculation
<not counted> cpu_atom/topdown-bad-spec/
(0.00%)
833,819 cpu_core/topdown-be-bound/ # 23.0% backend
bound
<not counted> cpu_atom/topdown-be-bound/
(0.00%)
1,483,916 cpu_core/topdown-fe-bound/ # 41.0% frontend
bound
<not counted> cpu_atom/topdown-fe-bound/
(0.00%)
819,687 cpu_core/topdown-retiring/ # 22.7% retiring
<not counted> cpu_atom/topdown-retiring/
(0.00%)
151,010 cpu_core/branch-instructions/
<not counted> cpu_atom/branch-instructions/
(0.00%)
4,402 cpu_core/branch-misses/
<not counted> cpu_atom/branch-misses/
(0.00%)
1,456,308 cpu_core/bus-cycles/
<not counted> cpu_atom/bus-cycles/
(0.00%)
1,132 cpu_core/cache-misses/
<not counted> cpu_atom/cache-misses/
(0.00%)
14,226 cpu_core/cache-references/
<not counted> cpu_atom/cache-references/
(0.00%)
600,633 cpu_core/cpu-cycles/
<not counted> cpu_atom/cpu-cycles/
(0.00%)
737,806 cpu_core/instructions/
<not counted> cpu_atom/instructions/
(0.00%)
0 cpu_core/mem-loads/
<not counted> cpu_atom/mem-loads/
(0.00%)
120,151 cpu_core/mem-stores/
<not counted> cpu_atom/mem-stores/
(0.00%)
1,456,308 cpu_core/ref-cycles/
<not counted> cpu_atom/ref-cycles/
(0.00%)
<not counted> cpu_core/cache-misses/
(0.00%)
<not counted> cpu_atom/cache-misses/
(0.00%)
<not counted> cpu_core/cache-references/
(0.00%)
<not counted> cpu_atom/cache-references/
(0.00%)
0.001518955 seconds time elapsed
0.001531000 seconds user
0.000000000 seconds sys
Thanks,
Kan
We
genuinely have broken grouping functions. We could just add to the
test a skip if /sys/devices/cpu_core and /sys/devices/cpu_atom
directories exist (making assumptions on where sysfs is mounted). I'm
not yet able to test on Alderlake hence not wanting to have a lot of
untested code.
Thanks,
Ian
Thanks,
Ian
Thanks,
Kan
+test_default_stat
+test_topdown_groups
+test_topdown_weak_groups
+exit $err