On 20/01/2021 16:40, Ian Rogers wrote:
> On Wed, Jan 20, 2021 at 8:23 AM John Garry <john.garry@huawei.com
> <mailto:john.garry@huawei.com>> wrote:
>
>     Metrics containing duration_time cause a segfault:
>
>     $./perf stat -v -M L1D_Cache_Fill_BW sleep 1
>     Using CPUID GenuineIntel-6-3D-4
>     metric expr 64 * l1d.replacement / 1000000000 / duration_time for
>     L1D_Cache_Fill_BW
>     found event duration_time
>     found event l1d.replacement
>     adding {l1d.replacement}:W,duration_time
>     l1d.replacement -> cpu/umask=0x1,(null)=0x1e8483,event=0x51/
>     Segmentation fault
>
>     In commit c2337d67199a ("perf metricgroup: Fix metrics using aliases
>     covering multiple PMUs"), the logic in find_evsel_group() when iter'ing
>     events was changed to not only select events in same group, but also for
>     aliased PMUs.
>
>     Checking whether events were for aliased PMUs was done by comparing the
>     event PMU name. This was not safe for duration_time event, which has no
>     associated PMU (and no PMU name), so fix by checking if the event
>     PMU name
>     is set also.
>
>
> Thanks for this, it should be fairly easy to add a test. Could we do this?

I don't mind following up with that.

>
>     Fixes: c2337d67199a ("perf metricgroup: Fix metrics using aliases
>     covering multiple PMUs")
>     Reported-by: Joakim Zhang <qiangqing.zhang@nxp.com
>     <mailto:qiangqing.zhang@nxp.com>>
>     Signed-off-by: John Garry <john.garry@huawei.com
>     <mailto:john.garry@huawei.com>>
>
>     diff --git a/tools/perf/util/metricgroup.c
>     b/tools/perf/util/metricgroup.c
>     index 2e60ee170abc..e6d3452031e5 100644
>     --- a/tools/perf/util/metricgroup.c
>     +++ b/tools/perf/util/metricgroup.c
>     @@ -162,6 +162,14 @@ static bool contains_event(struct evsel
>     **metric_events, int num_events,
>              return false;
>       }
>
>     +static bool evsel_same_pmu(struct evsel *ev1, struct evsel *ev2)
>     +{
>     +       if (!ev1->pmu_name || !ev2->pmu_name)
>     +               return false;
>
>
> What about the case of "!ev1->pmu_name && !ev2->pmu_name" ?

As far as I know, it should not happen, since duration_time is a special
event. More below.

>
> Thanks,
> Ian
>
>     +
>     +       return !strcmp(ev1->pmu_name, ev2->pmu_name);
>     +}
>     +
>       /**
>        * Find a group of events in perf_evlist that correspond to those
>     from a parsed
>        * metric expression. Note, as find_evsel_group is called in the
>     same order as
>     @@ -280,8 +288,7 @@ static struct evsel *find_evsel_group(struct
>     evlist *perf_evlist,
>                               */
>                              if (!has_constraint &&
>                                  ev->leader != metric_events[i]->leader &&
>     -                           !strcmp(ev->leader->pmu_name,
>     -                                   metric_events[i]->leader->pmu_name))
>     +                           evsel_same_pmu(ev->leader,
>     metric_events[i]->leader))

ev->leader->pmu_name == NULL for only duration_time event. And we don't
get here for ev == metric_events[i] == duration_time event (as we use
evlist__for_each_entry_continue() and duration_time is always last in
metric_events[]), so both event arguments should not have pmu_name ==
NULL. Indeed, I could just check metric_events[i]->leader->pmu_name !=
NULL, but thought it better to check both for safety.

Cheers,
John

>                                      break;
>                              if (!strcmp(metric_events[i]->name,
>     ev->name)) {
>                                      set_bit(ev->idx, evlist_used);
>     --
>     2.26.2
>