Re: [PATCH 0/6 v1] perf test attr: Fix the re-enabled test

[Namhyung Kim <namhyung@kernel.org>]

On Tue, Oct 06, 2026 at 06:13:03PM +0200, Arnaldo Carvalho de Melo wrote:
> From: Arnaldo Carvalho de Melo <acme@redhat.com>
> 
> Hello all,
> 
> This series fixes the "Perf attribute expectations test" (perf test 82,
> formerly "Setup struct perf_event_attr").
> 
> Patches 1-2 are Michael Petlan's prerequisite fixes that make the shell
> wrapper propagate attr.py's return value and correct the
> perf_event_attr size expectation after struct config4 was added.  They
> are included at the start of the series so that it applies cleanly on
> the current perf-tools-next (and for the review bots to verify the new
> patches); the new fixes are patches 3-6.
> 
> With the test reporting failures again, several stale expectations
> surfaced and are addressed here:
> 
> - test-record-graph-dwarf still expected PERF_SAMPLE_ADDR and
>   PERF_SAMPLE_DATA_SRC for --call-graph dwarf.  Those were dropped by
>   bdd051e249141c79 ("perf record: Split --data-mmap option"), which
>   decoupled data MMAP recording from the sample address bits.
> 
> - The record and stat group tests expected cache-misses and
>   instructions to be the legacy PERF_TYPE_HARDWARE events, but since
>   b1c5efbfd92eb84d ("perf parse-events: Remove hard coded legacy
>   hardware and cache parsing") they resolve to the core PMU sysfs/JSON
>   events, so request the legacy encodings explicitly with the
>   legacy-hardware-config term.
> 
> - The perf stat default metrics open their events in groups and the
>   raw encodings of instructions, branch-misses, cpu-cycles and
>   stalled-cycles-frontend/backend differ between Intel and AMD, so
>   base-stat now accepts the standalone and group variants and
>   test-stat-default/detailed-{1,2,3} gained the per-vendor raw
>   encodings.
> 
> - attr.py no longer requires a recorded event to be in a group when
>   the matching expectation does not declare one, allowing the stat
>   tests to verify event attributes without encoding the metric
>   grouping.  Expectations that declare a group are still verified in
>   both directions.
> 
> The test passes with 'perf test -vv attribute' on x86_64.  The AMD
> encodings come from the amd_pmu_event_map tables in
> arch/x86/events/amd/core.c.

On my machine, the current tmp.perf-tools-next which contains these
changes fails like this.

  $ tools/perf/perf test -v 82
  ---- start ----
  test child forked, pid 931585
  /home/namhyung/project/linux/tools/perf/tests/shell/attr.sh: line 20: python: command not found
  ---- end(-1) ----
   82: Perf attribute expectations test                                                 : FAILED!
  
  === Test Summary ===
  Passed main tests : 0
  Passed subtests   : 0
  Skipped tests     : 0
  Failed tests      : 1
  List of failed tests:
     82: Perf attribute expectations test

The below patch fixes the problem.

Thanks,
Namhyung


diff --git a/tools/perf/tests/shell/attr.sh b/tools/perf/tests/shell/attr.sh
index 307e1535795a34dc..ca3ca4a193971ec1 100755
--- a/tools/perf/tests/shell/attr.sh
+++ b/tools/perf/tests/shell/attr.sh
@@ -4,6 +4,10 @@
 
 err=0
 
+shelldir=$(dirname "$0")
+# shellcheck source=lib/setup_python.sh
+. "${shelldir}"/lib/setup_python.sh
+
 cleanup() {
   trap - EXIT TERM INT
 }
@@ -17,7 +21,7 @@ trap trap_cleanup EXIT TERM INT
 
 shelldir=$(dirname "$0")
 perf_path=$(which perf)
-python "${shelldir}"/lib/attr.py -d "${shelldir}"/attr -v -p "$perf_path"
+$PYTHON "${shelldir}"/lib/attr.py -d "${shelldir}"/attr -v -p "$perf_path"
 err=$?
 cleanup
 exit $err