[PATCH v2] thermal/of: support thermal zones w/o trips subnode

[Dragan Simic]

Hello Chen-Yu,

On 2024-08-15 06:45, Chen-Yu Tsai wrote:
> On Mon, Aug 12, 2024 at 12:46 PM Chen-Yu Tsai <wenst at chromium.org> 
> wrote:
>> 
>> On Mon, Aug 12, 2024 at 9:22 AM Dragan Simic <dsimic at manjaro.org> 
>> wrote:
>> >
>> > Hello Chen-Yu,
>> >
>> > Thanks for the patch.  Please see one comment below.
>> >
>> > On 2024-08-09 09:08, Chen-Yu Tsai wrote:
>> > > From: Icenowy Zheng <uwu at icenowy.me>
>> > >
>> > > Although the current device tree binding of thermal zones require the
>> > > trips subnode, the binding in kernel v5.15 does not require it, and
>> > > many
>> > > device trees shipped with the kernel, for example,
>> > > allwinner/sun50i-a64.dtsi and mediatek/mt8183-kukui.dtsi in ARM64,
>> > > still
>> > > comply to the old binding and contain no trips subnode.
>> > >
>> > > Allow the code to successfully register thermal zones w/o trips subnode
>> > > for DT binding compatibility now.
>> > >
>> > > Furtherly, the inconsistency between DTs and bindings should be
>> > > resolved
>> > > by either adding empty trips subnode or dropping the trips subnode
>> > > requirement.
>> > >
>> > > Fixes: d0c75fa2c17f ("thermal/of: Initialize trip points separately")
>> > > Signed-off-by: Icenowy Zheng <uwu at icenowy.me>
>> > > Reviewed-by: Mark Brown <broonie at kernel.org>
>> > > Signed-off-by: Chen-Yu Tsai <wenst at chromium.org>
>> > > ---
>> > > Resurrecting this patch specifically for MediaTek MT8183 Kukui devices.
>> > >
>> > > Changes since v1:
>> > > - set *ntrips at beginning of thermal_of_trips_init()
>> > > - Keep goto out_of_node_put in of_get_child_count(trips) == 0 branch
>> > > - Check return value of thermal_of_trips_init(), if it is -ENXIO, print
>> > >   warning and clear |trips| pointer
>> > > - Drop |mask| change, as the variable was removed
>> > >
>> > > I kept Mark's reviewed-by since the changes are more stylish than
>> > > functional.
>> > > ---
>> > >  drivers/thermal/thermal_of.c | 19 ++++++++++++-------
>> > >  1 file changed, 12 insertions(+), 7 deletions(-)
>> > >
>> > > diff --git a/drivers/thermal/thermal_of.c
>> > > b/drivers/thermal/thermal_of.c
>> > > index aa34b6e82e26..f237e74c92fc 100644
>> > > --- a/drivers/thermal/thermal_of.c
>> > > +++ b/drivers/thermal/thermal_of.c
>> > > @@ -128,16 +128,17 @@ static struct thermal_trip
>> > > *thermal_of_trips_init(struct device_node *np, int *n
>> > >       struct device_node *trips, *trip;
>> > >       int ret, count;
>> > >
>> > > +     *ntrips = 0;
>> > >       trips = of_get_child_by_name(np, "trips");
>> > >       if (!trips) {
>> > > -             pr_err("Failed to find 'trips' node\n");
>> > > -             return ERR_PTR(-EINVAL);
>> > > +             pr_debug("Failed to find 'trips' node\n");
>> > > +             return ERR_PTR(-ENXIO);
>> > >       }
>> > >
>> > >       count = of_get_child_count(trips);
>> > >       if (!count) {
>> > > -             pr_err("No trip point defined\n");
>> > > -             ret = -EINVAL;
>> > > +             pr_debug("No trip point defined\n");
>> > > +             ret = -ENXIO;
>> > >               goto out_of_node_put;
>> > >       }
>> > >
>> > > @@ -162,7 +163,6 @@ static struct thermal_trip
>> > > *thermal_of_trips_init(struct device_node *np, int *n
>> > >
>> > >  out_kfree:
>> > >       kfree(tt);
>> > > -     *ntrips = 0;
>> > >  out_of_node_put:
>> > >       of_node_put(trips);
>> >
>> > It might be a bit cleaner to keep the "*ntrips = 0" assignment
>> > in the error handling path(s) only, with the positions of the goto
>> > labels adjusted a bit, and then assign -ENXIO to "ret" and jump
>> > to the right label when of_get_child_by_name(np, "trips") fails,
>> > instead of returning from there.
>> >
>> > If it's unclear what I'm talking about, please let me know and
>> > I'll send back the proposed hunk.
>> 
>> I think I understand: move "*ntrips = 0" to after of_node_put() in the
>> error path, and have the "!trips" branch jump to "out_of_node_put" as
>> well. That works since of_node_put() checks the pointer first.
>> 
>> I'll wait a bit and see if there are any more comments.
> 
> Actually, Krzysztof (+CC) is cleaning up this function using scoped
> variables. So it might actually make more sense to move "*ntrips = 0"
> to the top once the error path is completely removed.

I see, it would make sense to move "*ntrips = 0" to the top, but what
bugs me with that approach a bit is that we's still have another 
instance
of "*ntrips = 0" in the error paths.  Thus, it might be cleaner to have
only one instance of "*ntrips = 0", in the error paths, and use "ret = 
..."
plus "goto ..." pairs instead of single "return ..." statements.

That way, we'd keep "*ntrips = 0" in the error pathso only, which would
clearly show that's part of the error handling only.  Though, I'd be 
also
fine with moving "*ntrips = 0" to the top, if you find that cleaner.

>> > > @@ -490,8 +490,13 @@ static struct thermal_zone_device
>> > > *thermal_of_zone_register(struct device_node *
>> > >
>> > >       trips = thermal_of_trips_init(np, &ntrips);
>> > >       if (IS_ERR(trips)) {
>> > > -             pr_err("Failed to find trip points for %pOFn id=%d\n", sensor, id);
>> > > -             return ERR_CAST(trips);
>> > > +             if (PTR_ERR(trips) != -ENXIO) {
>> > > +                     pr_err("Failed to find trip points for %pOFn id=%d\n", sensor, id);
>> > > +                     return ERR_CAST(trips);
>> > > +             }
>> > > +
>> > > +             pr_warn("Failed to find trip points for %pOFn id=%d\n", sensor, id);
>> > > +             trips = NULL;
>> > >       }
>> > >
>> > >       ret = thermal_of_monitor_init(np, &delay, &pdelay);