perf stat: Fix the hard-coded metrics calculation on the hybrid
authorKan Liang <kan.liang@linux.intel.com>
Thu, 6 Jun 2024 18:03:16 +0000 (11:03 -0700)
committerNamhyung Kim <namhyung@kernel.org>
Fri, 21 Jun 2024 16:43:09 +0000 (09:43 -0700)
The hard-coded metrics is wrongly calculated on the hybrid machine.

$ perf stat -e cycles,instructions -a sleep 1

 Performance counter stats for 'system wide':

        18,205,487      cpu_atom/cycles/
         9,733,603      cpu_core/cycles/
         9,423,111      cpu_atom/instructions/     #  0.52  insn per cycle
         4,268,965      cpu_core/instructions/     #  0.23  insn per cycle

The insn per cycle for cpu_core should be 4,268,965 / 9,733,603 = 0.44.

When finding the metric events, the find_stat() doesn't take the PMU
type into account. The cpu_atom/cycles/ is wrongly used to calculate
the IPC of the cpu_core.

In the hard-coded metrics, the events from a different PMU are only
SW_CPU_CLOCK and SW_TASK_CLOCK. They both have the stat type,
STAT_NSECS. Except the SW CLOCK events, check the PMU type as well.

Fixes: 0a57b910807a ("perf stat: Use counts rather than saved_value")
Reported-by: Khalil, Amiri <amiri.khalil@intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: stable@vger.kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240606180316.4122904-1-kan.liang@linux.intel.com
tools/perf/util/stat-shadow.c

index 3466aa9524421d734f5e4ef4e82979bb1954308d..6bb975e46de37e957e1b40b7e0463d4c88b8cdd2 100644 (file)
@@ -176,6 +176,13 @@ static double find_stat(const struct evsel *evsel, int aggr_idx, enum stat_type
                if (type != evsel__stat_type(cur))
                        continue;
 
+               /*
+                * Except the SW CLOCK events,
+                * ignore if not the PMU we're looking for.
+                */
+               if ((type != STAT_NSECS) && (evsel->pmu != cur->pmu))
+                       continue;
+
                aggr = &cur->stats->aggr[aggr_idx];
                if (type == STAT_NSECS)
                        return aggr->counts.val;