[PATCH RFC 4/9] perf jevents: Add sys_events_find_events_table()

[Ian Rogers]

On Wed, Jul 19, 2023 at 8:37 AM John Garry <john.g.garry at oracle.com> wrote:
>
> On 19/07/2023 16:25, Ian Rogers wrote:
> >> I was thinking about this a little further. So you suggest that the
> >> metric expression contains PMU name per term, like
> >> "cpu_atom at instructions@ / cpu_atom at cycles@" - how would/could this work
> >> for PMUs with more complex naming, like the form hisi_siclXXX_cpa_YYY?
> >> Would we use the "Unit" expression for the metric name, like
> >> "@hisi_sicl,cpa at event_foo"?
> > How does this work for events? The "@hisi_sicl,cpa at event_foo" looks
> > strange, shouldn't it be "hisi_sicl,cpa at event_foo@" but then hisi_sicl
> > looks like an event name.
>
> Yeah, that was a typo from me - like you say, it would be
> hisi_sicl,cpa at event_foo@
>
> So is that what you would be suggesting then, such that we specify the
> PMU in the metric terms? It does look a bit odd :)

I guess there could be some kind of regular expression syntax:
(hisi_scl|cpa)@event_foo

We have the "has_event" function now in the metric expressions:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/expr.c?h=perf-tools-next#n480

So it could be:
hisi_scl at event_foo@ if has_event(hisi_scl at event_foo@) else cpa at event_foo@

Which has the advantage of working today, but the disadvantage of being verbose.

Thanks,
Ian


> >
> >>>> BTW, which git repo/branch do you guys use for dev? I thought that it
> >>>> would be acme git, but Namhyung says "We moved to new repos from acme to
> >>>> perf/perf-tools and perf/perf-tools-next" - where is repo "perf"?
> >>> Current development is here now:
> >>> https://urldefense.com/v3/__https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/log/?h=perf-tools-next__;!!ACWV5N9M2RV99hQ!OQDHOClSjd6nVZhmgzrK3RwzXuQpP54QhqyIKpITa_MFD4PLdS7yPYSnvInFja9nrFx9Sd-UnlsJ6XUqAh4$
> >> Can that be added to the MAINTAINERS file? I suppose it is ok under
> >> "PERFORMANCE EVENTS SUBSYTEM", since the two would-be git repos listed
> >> under that same entry would be pretty obvious in purpose.
> > Arnaldo could you take a look at doing this?
>
> Thanks,
> John