Re: [PATCH v1 06/13] thermal: gov_fair_share: Rearrange get_trip_level()

[Rafael J. Wysocki]

On Wed, Sep 27, 2023 at 5:37 PM Daniel Lezcano
<daniel.lezcano@xxxxxxxxxx> wrote:
>
> On 27/09/2023 17:06, Rafael J. Wysocki wrote:
> > On Wed, Sep 27, 2023 at 5:00 PM Daniel Lezcano
> > <daniel.lezcano@xxxxxxxxxx> wrote:
> >>
> >> On 21/09/2023 19:54, Rafael J. Wysocki wrote:
> >>> From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> >>>
> >>> Make get_trip_level() access the thermal zone's trip table directly
> >>> instead of using __thermal_zone_get_trip() which adds overhead related
> >>> to the unnecessary bounds checking and copying the trip point data.
> >>>
> >>> Also rearrange the code in it to make it somewhat easier to follow.
> >>>
> >>> The general functionality is not expected to be changed.
> >>>
> >>> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> >>> ---
> >>>    drivers/thermal/gov_fair_share.c |   22 ++++++++++------------
> >>>    1 file changed, 10 insertions(+), 12 deletions(-)
> >>>
> >>> Index: linux-pm/drivers/thermal/gov_fair_share.c
> >>> ===================================================================
> >>> --- linux-pm.orig/drivers/thermal/gov_fair_share.c
> >>> +++ linux-pm/drivers/thermal/gov_fair_share.c
> >>> @@ -21,23 +21,21 @@
> >>>     */
> >>>    static int get_trip_level(struct thermal_zone_device *tz)
> >>>    {
> >>> -     struct thermal_trip trip;
> >>> -     int count;
> >>> +     const struct thermal_trip *trip = tz->trips;
> >>> +     int i;
> >>>
> >>> -     for (count = 0; count < tz->num_trips; count++) {
> >>> -             __thermal_zone_get_trip(tz, count, &trip);
> >>> -             if (tz->temperature < trip.temperature)
> >>> +     if (tz->temperature < trip->temperature)
> >>> +             return 0;
> >>> +
> >>> +     for (i = 0; i < tz->num_trips - 1; i++) {
> >>> +             trip++;
> >>> +             if (tz->temperature < trip->temperature)
> >>>                        break;
> >>>        }
> >>
> >> Is it possible to use for_each_thermal_trip() instead ? That would make
> >> the code more self-encapsulate
> >
> > It is possible in principle, but this is a governor which is regarded
> > as part of the core, isn't it?
> >
> > So is an extra overhead related to using a callback (which may be
> > subject to retpolines and such) really justified in this case?
>
>  From my POV, all trip points browsing should be replaced by
> for_each_thermal_trip() so any change in the future in how we go through
> the existing thermal trips will impact one place.
>
> If the routine needs to be optimized, that is something we can do also
> (may be an inline the callback?)

OK