Re: [PATCH v5 0/3] Thermal ACPI APIs for generic trip points

[srinivas pandruvada]

On Wed, 2023-01-18 at 22:01 +0100, Daniel Lezcano wrote:
> On 18/01/2023 21:53, srinivas pandruvada wrote:
> > On Wed, 2023-01-18 at 21:00 +0100, Daniel Lezcano wrote:
> > > On 18/01/2023 20:16, srinivas pandruvada wrote:
> > > 
> > > [ ... ]
> > > 
> > > > > > But we'd better wait for the thermald test result from
> > > > > > Srinvias.
> > > > > 
> > > > > A quick test show that things still work with thermald and
> > > > > these
> > > > > changes.
> > > > 
> > > > But I have a question. In some devices trip point temperature
> > > > is
> > > > not
> > > > static. When hardware changes, we get notification. For example
> > > > INT3403_PERF_TRIP_POINT_CHANGED for INT3403 drivers.
> > > > Currently get_trip can get the latest changed value. But if we
> > > > preregister, we need some mechanism to update them.
> > > 
> > > When the notification INT3403_PERF_TRIP_POINT_CHANGED happens, we
> > > call
> > > int340x_thermal_read_trips() which in turn updates the trip
> > > points.
> > > 
> > 
> > Not sure how we handle concurrency here when driver can freely
> > update
> > trips while thermal core is using trips.
> 
> Don't we have the same race without this patch ? The thermal core can
> call get_trip_temp() while there is an update, no ?
Yes it is. But I can add a mutex locally here to solve.
But not any longer.

I think you need some thermal_zone_read_lock/unlock() in core, which
can use rcu. Even mutex is fine as there will be no contention as
updates to trips will be rare.

Thanks,
Srinivas

>