Re: [PATCH v5 1/3] thermal: qcom-spmi: Use PMIC thermal stage 2 for critical trip points

[Matthias Kaehlcke <mka@chromium.org>]

Hi Doug,

On Wed, Jul 25, 2018 at 04:19:56PM -0700, Doug Anderson wrote:

> On Tue, Jul 24, 2018 at 4:46 PM, Matthias Kaehlcke <mka@chromium.org> wrote:
> > +static int qpnp_tm_update_critical_trip_temp(struct qpnp_tm_chip *chip,
> > +                                            int temp)
> > +{
> > +       u8 reg;
> > +       bool disable_s2_shutdown = false;
> > +       int ret;
> > +
> > +       WARN_ON(!mutex_is_locked(&chip->lock));
> > +
> > +       /*
> > +        * Default: S2 and S3 shutdown enabled, thresholds at
> > +        * 105C/125C/145C, monitoring at 25Hz
> > +        */
> > +       reg = SHUTDOWN_CTRL1_RATE_25HZ;
> > +
> > +       if ((temp == THERMAL_TEMP_INVALID) ||
> > +           (temp < STAGE2_THRESHOLD_MIN)) {
> > +               chip->thresh = THRESH_MIN;
> > +               goto skip;
> > +       }
> > +
> > +       if (temp <= STAGE2_THRESHOLD_MAX) {
> > +               chip->thresh = THRESH_MAX -
> > +                       ((STAGE2_THRESHOLD_MAX - temp) /
> > +                        TEMP_THRESH_STEP);
> > +               disable_s2_shutdown = true;
> > +       } else {
> > +               chip->thresh = THRESH_MAX;
> > +
> > +               if (!IS_ERR(chip->adc))
> > +                       disable_s2_shutdown = true;
> > +               else
> > +                       dev_warn(chip->dev,
> > +                                "No ADC is configured and critical temperature is above the maximum stage 2 threshold of 140°C! Configuring stage 2 shutdown at 140°C.\n");
> 
> Putting a non-ASCII character (the degree symbol) in your commit
> message is one thing, but are you sure it's wise to put it in the
> kernel logs?

A few other drivers also do this
(drivers/gpu/drm/nouveau/nvkm/subdev/clk/base.c,
drivers/macintosh/windfarm_pm121.c), however that doesn't mean it's a
good idea. Will change to degC or C.

> > +       }
> > +
> > +skip:
> > +       reg |= chip->thresh;
> > +       if (disable_s2_shutdown)
> > +               reg |= SHUTDOWN_CTRL1_OVERRIDE_S2;
> > +
> > +       ret = qpnp_tm_write(chip, QPNP_TM_REG_SHUTDOWN_CTRL1, reg);
> > +       if (ret < 0)
> > +               return ret;
> > +
> > +       return ret;
> 
> Simplify the above lines to:
> 
> return qpnp_tm_write(chip, QPNP_TM_REG_SHUTDOWN_CTRL1, reg);

Ouch, my code is indeed dumb ...

> > @@ -313,12 +441,7 @@ static int qpnp_tm_probe(struct platform_device *pdev)
> >         if (ret < 0)
> >                 return ret;
> >
> > -       chip->tz_dev = devm_thermal_zone_of_sensor_register(&pdev->dev, 0, chip,
> > -                                                       &qpnp_tm_sensor_ops);
> > -       if (IS_ERR(chip->tz_dev)) {
> > -               dev_err(&pdev->dev, "failed to register sensor\n");
> > -               return PTR_ERR(chip->tz_dev);
> > -       }
> > +       chip->initialized = true;
> 
> Should we add "thermal_zone_device_update(chip->tz_dev,
> THERMAL_EVENT_UNSPECIFIED);" here

Seems reasonable, will do.

> ...also: do we care about any type of locking for chip->initialized?
> Technically we can be running on weakly ordered memory so if
> qpnp_tm_update_temp_no_adc() is running on a different processor then
> possibly it could still keep returning the default temperature for a
> little while.  We could try to analyze whether there's some sort of
> implicit barrier or we could add manual memory barriers, but generally
> I try to avoid that and just do the simple locking...  What about just
> setting chip-Initialized = true at the end of qpnp_tm_init() while the
> mutex is still held?

Thanks for pointing that out. I agree that we should keep things
simple, chip->initialized to true at the end of qpnp_tm_init() sounds
good to me.

> I'd also love to hear from someone with more thermal framework
> experience to make sure it's legit to return a default value if
> someone calls us while we're initting.  It seems sane to me but nice
> to confirm it's OK.

An alternative could be to return THERMAL_TEMP_INVALID, however I
don't see this handled outside of thermal_core.c, not sure if it could
throw some other code off.

Comments from thermal folks on either approach (or alternatives) are
definitely welcome :)

> Overall I like the idea of this patch so hopefully others do too.
> Thanks for sending it out!

Thanks for the review!

Matthias