Re: [PATCH v3 1/3] perf: Skip and warn on unknown format 'configN' attrs

[Rob Herring]

On Thu, Sep 29, 2022 at 1:48 AM Leo Yan <leo.yan@xxxxxxxxxx> wrote:
>
> Hi Rob,
>
> On Wed, Sep 14, 2022 at 03:08:34PM -0500, Rob Herring wrote:
>
> [...]
>
> > +void perf_pmu__warn_invalid_formats(struct perf_pmu *pmu)
> > +{
> > +     struct perf_pmu_format *format;
> > +
> > +     list_for_each_entry(format, &pmu->format, list)
> > +             if (format->value >= PERF_PMU_FORMAT_VALUE_CONFIG_END) {
> > +                     pr_warning("WARNING: '%s' format '%s' requires 'perf_event_attr::config%d'"
> > +                                "which is not supported by this version of perf!\n",
> > +                                pmu->name, format->name, format->value);
> > +                     return;
> > +             }
> > +}
>
> Though I saw you and Namhyung have discussion in underway, this patch
> set is fine for me.  I validated the patches at my side (with a bit
> hacking in Arm SPE driver for faking invert filter).  You could add my
> tested tag for this patch set:
>
> Tested-by: Leo Yan <leo.yan@xxxxxxxxxx>
>
> But I want to remind two things after I used "perf test" to validate
> this patch set:
>
>   $ ./perf test list
>   6: Parse event definition strings
>   6:1: Test event parsing
>   6:2: Test parsing of "hybrid" CPU events
>   6:3: Parsing of all PMU events from sysfs
>   6:4: Parsing of given PMU events from sysfs
>   6:5: Parsing of aliased events from sysfs
>   6:6: Parsing of aliased events
>   6:7: Parsing of terms (event modifiers)
>   $ ./perf test -v 6
>
> The first one is this patch set introduces segmentation fault for the
> case "Parsing of aliased events" (See tests/parse-events.c).  But the
> issue is caused by the test case itself; we need to add below line into
> test_event_fake_pmu() for initialisation list header.

Thanks.

>     INIT_LIST_HEAD(&perf_pmu__fake.format);

I ended up fixing this in perf_pmu__warn_invalid_formats() instead as
the test dealing with internal stuct pmu details didn't seem right:

+       /* fake pmu doesn't have format list */
+       if (pmu == &perf_pmu__fake)
+               return;
+

>
> The second question is for testing config3 in "perf test".  You could
> see the file tests/parse-events.c has included several test cases for
> config/config1/config2.  It's good to add the same testing for config3
> as well, please see test__checkevent_pmu() and test__checkterms_simple()
> for relevant code.

Okay, will add in the next version.

Rob