Re: [PATCH v7 4/4] perf sched latency: Add histogram and time interval options

[Aaron Tomlin]

On Mon, Aug 03, 2026 at 11:06:16AM -0700, Namhyung Kim wrote:
> > @@ -1168,7 +1306,13 @@ add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
> >  		atoms->max_lat_start = atom->wake_up_time;
> >  		atoms->max_lat_end = timestamp;
> >  	}
> > +
> >  	atoms->nb_atoms++;
> > +
> > +	b = latency_bucket(sched, delta);
> > +	atoms->hist[b]++;
> > +	if (strcmp(thread__comm_str(atoms->thread), "swapper"))
> > +		sched->global_hist[b]++;
> 
> Why is the swapper thread not included in the global hist?
> 
> Also it's probably better to check thread__tid being 0.

Hi Namhyung,

The CPU-specific "swapper" thread runs when CPUs are idle. Therefore,
preemption/wakeup timings for idle threads do not represent _real_ task
scheduling latency. Throughout (e.g., in output_lat_thread()), idle threads
are explicitly ignored to avoid inflating 'sched->all_count' etc.

Indeed, testing for 'TID == 0' is superior. For example:

@@ -1168,7 +1306,13 @@ add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
                atoms->max_lat_start = atom->wake_up_time;
                atoms->max_lat_end = timestamp;
        }
+
        atoms->nb_atoms++;
+
+       b = latency_bucket(sched, delta);
+       atoms->hist[b]++;
+       if (thread__tid(atoms->thread) != 0)
+               sched->global_hist[b]++;

> > @@ -3659,6 +3831,21 @@ static int perf_sched__lat(struct perf_sched *sched)
> >  	perf_sched__merge_lat(sched);
> >  	perf_sched__sort_lat(sched);
> >  
> > +	next = rb_first_cached(&sched->sorted_atom_root);
> > +	while (next) {
> > +		struct work_atoms *work_list = rb_entry(next, struct work_atoms, node);
> > +
> > +		if (work_list->nb_atoms && strcmp(thread__comm_str(work_list->thread), "swapper"))
> 
> Ditto.  Comparing TID would be faster.

Acknowledged. Thanks!


Kind regards,
-- 
Aaron Tomlin