Re: [PATCH v3 3/4] thermal/core: Build ascending ordered indexes for the trip points

[Rafael J. Wysocki]

On Fri, Jul 22, 2022 at 9:16 AM Zhang, Rui <rui.zhang@xxxxxxxxx> wrote:
>
>
>
> > -----Original Message-----
> > From: Daniel Lezcano <daniel.lezcano@xxxxxxxxxx>
> > Sent: Thursday, July 21, 2022 5:35 PM
> > To: Zhang, Rui <rui.zhang@xxxxxxxxx>; rafael@xxxxxxxxxx
> > Cc: quic_manafm@xxxxxxxxxxx; amitk@xxxxxxxxxx; lukasz.luba@xxxxxxx;
> > linux-pm@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> > Subject: Re: [PATCH v3 3/4] thermal/core: Build ascending ordered indexes
> > for the trip points
> > Importance: High
> >
> > On 19/07/2022 16:17, Zhang Rui wrote:
> > > On Tue, 2022-07-19 at 09:22 +0200, Daniel Lezcano wrote:
> > >> On 19/07/2022 03:14, Zhang Rui wrote:
> > >>> On Mon, 2022-07-18 at 15:21 +0200, Daniel Lezcano wrote:
> > >>>>
> > >>>> Hi Zhang,
> > >>>>
> > >>>> thanks for the review
> > >>>>
> > >>>> On 18/07/2022 07:28, Zhang Rui wrote:
> > >>>>> On Fri, 2022-07-15 at 23:09 +0200, Daniel Lezcano wrote:
> > >>>>
> > >>>> [ ... ]
> > >>>>
> > >>>>>> Instead of taking the risk of breaking the existing platforms,
> > >>>>>> use an array of temperature ordered trip identifiers and make it
> > >>>>>> available for the code needing to browse the trip points in an
> > >>>>>> ordered way.
> > >>>>>>
> > >>>>>> Signed-off-by: Daniel Lezcano <daniel.lezcano@xxxxxxxxxx>
> > >>>>>> ---
> > >>>>
> > >>>> [ ... ]
> > >>>>
> > >>>>>> +static void sort_trips_indexes(struct thermal_zone_device
> > >>>>>> *tz)
> > >>>>>> +{
> > >>>>>> +       int i, j;
> > >>>>>> +
> > >>>>>> +       for (i = 0; i < tz->trips; i++)
> > >>>>>> +               tz->trips_indexes[i] = i;
> > >>>>>> +
> > >>>>>> +       for (i = 0; i < tz->trips; i++) {
> > >>>>>> +               for (j = i + 1; j < tz->trips; j++) {
> > >>>>>> +                       int t1, t2;
> > >>>>>> +
> > >>>>>> +                       tz->ops->get_trip_temp(tz, tz-
> > >>>>>>> trips_indexes[i], &t1);
> > >>>>>
> > >>>>> This line can be moved to the upper loop.
> > >>>>
> > >>>> Right, thanks!
> > >>>>
> > >>>>>> +                       tz->ops->get_trip_temp(tz, tz-
> > >>>>>>> trips_indexes[j], &t2);
> > >>>>>> +
> > >>>>>
> > >>>>> what about the disabled trip points?
> > >>>>>
> > >>>>> we should ignore those trip points and check the return value to
> > >>>>> make sure we're comparing the valid trip_temp values.
> > >>>>
> > >>>> We don't have to care about, whatever the position, the
> > >>>> corresponding trip id will be disabled by the trip init function
> > >>>> before calling this one and ignored in the handle_thermal_trip()
> > >>>> function
> > >>>
> > >>> hah, I missed this one and replied to your latest reply directly.
> > >>>
> > >>> The thing I'm concerning is that if we don't check the return value,
> > >>> for a disabled trip point, the trip_temp (t1/t2) returned is some
> > >>> random value, it all depends on the previous value set by last
> > >>> successful .get_trip_temp(), and this may screw up the sorting.
> > >>
> > >> The indexes array is the same size as the trip array, that makes the
> > >> code much less prone to errors.
> > >>
> > >> To have the same number of trip points, the index of the disabled
> > >> trip must be inserted also in the array. We don't care about its
> > >> position in the indexes array because it is discarded in the
> > >> handle_trip_point() function anyway. For this reason, the random
> > >> temperature of the disabled trip point and the resulting position in
> > >> the sorting is harmless.
> > >>
> > >> It is made on purpose to ignore the return value, so we have a
> > >> simpler code.
> > >>
> > > Let's take below case for example,
> > > say, we have three trip points 0, 1, 2, and trip point 1 is broken and
> > > disabled.
> > >
> > > trip temp for trip point 0 is 10 and for trip point 2 is 20.
> > > .get_trip_temp(tz, 1, &t) fails, and t is an uninitialized random
> > > value
> > >
> > >
> > > Initial:
> > >     trip_indexes[0]=0,trip_indexes[1]=1,trip_indexes[2]=2
> > > step1:
> > >     i=0,j=1
> > >     get trip temp for trip point trip_indexes[0]=0 and trip_indexes[1]=1
> > >     trip point 1 returns trip temp 5, and it swaps with trip point 0
> > >     so
> > >     trip_indexes[0]=1,trip_indexes[1]=0,trip_indexes[2]=2
> > > step2:
> > >     i=0,j=2
> > >     get trip temp for trip point trip_indexes[0]=1 and trip_indexes[2]=2
> > >     trip point 1 returns trip temp 25, and it swaps with trip point 2
> > >     so
> > >     trip_indexes[0]=2,trip_indexes[1]=0,trip_indexes[2]=1
> > >
> > > And the sorting is broken now.
> > >
> > > please correct me if I'm missing anything.
> >
> > Oh, nice! Thanks for the detailed explanation.
> >
> > We can initialize t1 and t2 to INT_MAX, so if the get_trip_temp() fails, they
> > will be set to the maximum temperature and it will be at the end of the array.
> >
> > Alternatively, we check the disabled bit and set the temperature to INT_MAX.
>
> IMO, we can
> 1. get the trip temp for each trip point and cache them
> 2. set the trips_disabled bit
> 3. do the sorting using the cached trip temp values
> in thermal_zone_device_trip_init() altogether.

What about the case when the trip temperature can be set from user
space?  I guess we replace the cached value then, but it may be
necessary to sort them again in theory?