Re: [PATCH] perf pmu: Fix event list for uncore PMUs

[Arnaldo Carvalho de Melo]

Em Tue, Dec 21, 2021 at 10:14:42AM +0000, John Garry escreveu:
> On 21/12/2021 09:35, Jiri Olsa wrote:
> > On Tue, Dec 21, 2021 at 09:10:37AM +0000, John Garry wrote:
> > > On 21/12/2021 07:58, Jiri Olsa wrote:
> > > > > +	/* Different names -> never duplicates */
> > > > > +	if (strcmp(alias_a->name, alias_b->name))
> > > > > +		return false;
> > > > > +	if (!alias_a->pmu)
> > > > > +		return true;
> > > > > +	if (!alias_b->pmu)
> > > > > +		return true;
> > > > nit could be:
> > > > 
> > > > 	if (!alias_a->pmu || !alias_b->pmu)
> > > > 		return true;
> > > > 
> > > > would be great to have more comments explaining the check
> > > > 
> > > 
> > > This is just a sanity check that both strings are non-NULL as we do a
> > > strcmp() next. So would this be better:
> > > 
> > > if (!alias_a->pmu || !alias_b->pmu || !strcmp(alias_a->pmu, alias_b->pmu))
> > > 	return true
> > > 
> > > ?
> > > 
> > > It will spill a line.
> > 
> > sure, it cought my eye because the is_cpu check later is done on
> > the same line, so I started wondering what's the difference ;-)
> > 
> 
> Now thinking a bit more I am not confident that this patch is a full fix.
> 
> arm have heterogeneous CPU systems as well - which are not "hybrid" - and I
> need to ensure that aliasing is still working properly there, as I think
> that this following check would stop removing duplicates there:
> 
> +	/* uncore PMUs */
> +	if (!alias_a->is_cpu && !alias_b->is_cpu)
> +		return true;
> +	return false;

I was about to process this, do you think its better to revert the
original patch while this gets fixed?

- Arnaldo