Re: [PATCH v3 3/5] perf stat: Remove evlist__add_default_attrs use strings

[Ian Rogers]

On Wed, May 29, 2024 at 10:39 AM Ian Rogers <irogers@xxxxxxxxxx> wrote:
>
> On Wed, May 29, 2024 at 8:39 AM James Clark <james.clark@xxxxxxx> wrote:
> >
> >
> >
> > On 10/05/2024 06:37, Ian Rogers wrote:
> > > add_default_atttributes would add evsels by having pre-created
> > > perf_event_attr, however, this needed fixing for hybrid as the
> > > extended PMU type was necessary for each core PMU. The logic for this
> > > was in an arch specific x86 function and wasn't present for ARM,
> > > meaning that default events weren't being opened on all PMUs on
> > > ARM. Change the creation of the default events to use parse_events and
> > > strings as that will open the events on all PMUs.
> > >
> > > Rather than try to detect events on PMUs before parsing, parse the
> > > event but skip its output in stat-display.
> > >
> > > The previous order of hardware events was: cycles,
> > > stalled-cycles-frontend, stalled-cycles-backend, instructions. As
> > > instructions is a more fundamental concept the order is changed to:
> > > instructions, cycles, stalled-cycles-frontend, stalled-cycles-backend.
> > >
> > > Closes: https://lore.kernel.org/lkml/CAP-5=fVABSBZnsmtRn1uF-k-G1GWM-L5SgiinhPTfHbQsKXb_g@xxxxxxxxxxxxxx/
> >
> > Taking a look at this one now. I think some example commands and outputs
> > in the commit message would be helpful because there are quite a few
> > different things mentioned in the closes link.
> >
> > But I'm assuming this is just for the command without specifying an event:
> >
> >   $ perf stat
> >
> > I didn't realise that wasn't working properly and I'd missed that Mark
> > spotted it in that link.
>
> Hi James,
>
> Do you want to take on owning fixing this? My concern is that by using
> event parsing:
>
> +                               "context-switches,"
> +                               "cpu-migrations,"
> +                               "page-faults,"
> +                               "instructions,"
> +                               "cycles,"
> +                               "stalled-cycles-frontend,"
> +                               "stalled-cycles-backend,"
> +                               "branches,"
> +                               "branch-misses",
>
> any of the names could conflict with something advertised on an ARM
> PMU somewhere. Clearly cycles has already proved to be controversial
> and broken for perf record on Neoverse.

Oh and these events are also broken on ARM M? PMUs with the revert
that's in v6.10, so I don't know how to go about fixing this.

Thanks,
Ian