Re: [PATCH v3 3/3] perf tools: Enable on a list of CPUs for hybrid

[Jin, Yao]

Hi Jiri,

On 7/22/2021 6:19 PM, Jiri Olsa wrote:
> On Wed, Jul 21, 2021 at 12:30:11PM +0800, Jin, Yao wrote:
> > Hi Jiri,
> >
> > On 7/20/2021 5:16 PM, Jiri Olsa wrote:
> > > On Tue, Jul 20, 2021 at 03:07:02PM +0800, Jin, Yao wrote:
> > >
> > > SNIP
> > >
> > > > OK, evlist__fix_cpus() is better, use this name in v4.
> > > >
> > > > > > +{
> > > > > > +	struct perf_cpu_map *cpus;
> > > > > > +	struct evsel *evsel, *tmp;
> > > > > > +	struct perf_pmu *pmu;
> > > > > > +	int ret, unmatched_count = 0, events_nr = 0;
> > > > > > +
> > > > > > +	if (!perf_pmu__has_hybrid() || !cpu_list)
> > > > > > +		return 0;
> > > > > > +
> > > > > > +	cpus = perf_cpu_map__new(cpu_list);
> > > > > > +	if (!cpus)
> > > > > > +		return -1;
> > > > > > +
> > > > > > +	evlist__for_each_entry_safe(evlist, tmp, evsel) {
> > > > > > +		struct perf_cpu_map *matched_cpus, *unmatched_cpus;
> > > > > > +		char buf1[128], buf2[128];
> > > > > > +
> > > > > > +		pmu = perf_pmu__find_hybrid_pmu(evsel->pmu_name);
> > > > > > +		if (!pmu)
> > > > > > +			continue;
> > > > > > +
> > > > > > +		ret = perf_pmu__cpus_match(pmu, cpus, &matched_cpus,
> > > > > > +					   &unmatched_cpus);
> > > > > > +		if (ret)
> > > > > > +			goto out;
> > > > > > +
> > > > > > +		events_nr++;
> > > > > > +
> > > > > > +		if (matched_cpus->nr > 0 && (unmatched_cpus->nr > 0 ||
> > > > > > +		    matched_cpus->nr < cpus->nr ||
> > > > > > +		    matched_cpus->nr < pmu->cpus->nr)) {
> > > > > > +			perf_cpu_map__put(evsel->core.cpus);
> > > > > > +			perf_cpu_map__put(evsel->core.own_cpus);
> > > > > > +			evsel->core.cpus = perf_cpu_map__get(matched_cpus);
> > > > > > +			evsel->core.own_cpus = perf_cpu_map__get(matched_cpus);
> > > > >
> > > > > I'm bit confused in here.. AFAIUI there's 2 evsel objects create
> > > > > for hybrid 'cycles' ... should they have already proper cpus set?
> > > >
> > > > For 'cycles', yes two evsels are created automatically. One is for atom CPU
> > > > (e.g. 8-11), the other is for core CPU (e.g. 0-7). In this example, these 2
> > > > evsels have already the cpus set.
> > >
> > > hum, so those evsels are created with pmu's cpus, right?
> >
> > Yes, that's right. But we also check and adjust the evsel->cpus by using
> > user's cpu list on hybrid (what the evlist__use_cpu_list() does).
> >
> > > > While the 'cpus' here is just the user specified cpu list.
> > > > cpus = perf_cpu_map__new(cpu_list);
> > >
> > > then I think they will be changed by evlist__create_maps
> > > with whatever user wants?
> >
> > No, it will not be changed by evlist__create_maps.
> >
> > In evlist__create_maps(),
> > evlist->core.has_user_cpus = !!target->cpu_list && !target->hybrid;
> >
> > It disables has_user_cpus for hybrid.
> >
> > So in __perf_evlist__propagate_maps, they will not be changed by evlist->cpus.
> >
> > if (!evsel->own_cpus || evlist->has_user_cpus) {
> > 	perf_cpu_map__put(evsel->cpus);
> > 	evsel->cpus = perf_cpu_map__get(evlist->cpus);
> > 	
> > > could we just change __perf_evlist__propagate_maps to follow
> > > pmu's cpus?
> >
> > In __perf_evlist__propagate_maps, it has already followed pmu's cpus because
> > the evlist->has_user_cpus is false for hybrid.
>
> sorry for delay

Never mind. :)

> ok, so we first fix the cpus on hybrid events and then
> propagate maps.. I guess it's ok, because it's in libperf
> and that has no notion of hybrid so far

Yes. If we want the libperf to be hybrid aware, the interface has to be modified but actually we 
need to avoid modifying the libperf interface. So I finally decide to adjust the evsel->cpus first 
and then propatate maps.

> could you please rename that function so it's also obvious
> it's for hybrid only
>
>    evlist__fix_hybrid_cpus ? not sure ;-)

Sure, I will rename the funciton in v4.

> and add some comment with example to explain what the
> function is doing

Got it!

Thanks
Jin Yao

> thanks,
> jirka