On Mon, May 15, 2017 at 11:18:45AM +0200, Quentin Schulz wrote:
> Hi Maxime,
>
> On 15/05/2017 11:11, Maxime Ripard wrote:
> > On Mon, May 15, 2017 at 09:39:02AM +0200, Quentin Schulz wrote:
> >> For the sake of DT binding stability, this IIO driver is a child of an
> >> MFD driver for Allwinner A10, A13 and A31 because there already exists a
> >> DT binding for this IP. The MFD driver has a DT node but the IIO driver
> >> does not.
> >>
> >> The IIO device registers the temperature sensor in the thermal framework
> >> using the DT node of the parent, the MFD device, so the thermal
> >> framework could match the phandle to the MFD device in the DT and the
> >> struct device used to register in the thermal framework.
> >>
> >> devm_thermal_zone_of_sensor_register was previously used to register the
> >> thermal sensor with the parent struct device of the IIO device,
> >> representing the MFD device. By doing so, we registered actually the
> >> parent in the devm routine and not the actual IIO device.
> >>
> >> This lead to the devm unregister function not being called when the IIO
> >> module driver is removed. It resulted in the thermal framework still
> >> polling the get_temp function of the IIO module while the device doesn't
> >> exist anymore, thus generated a kernel panic.
> >>
> >> Use the non-devm function instead and do the unregister manually in the
> >> remove function.
> >>
> >> Fixes: d1caa9905538 ("iio: adc: add support for Allwinner SoCs ADC")
> >>
> >> Signed-off-by: Quentin Schulz <quentin.schulz@xxxxxxxxxxxxxxxxxx>
> >> Reported-by: Corentin Labbe <clabbe.montjoie@xxxxxxxxx>
> >> ---
> [...]
> >> @@ -688,6 +687,12 @@ static int sun4i_gpadc_remove(struct platform_device *pdev)
> >>
> >> pm_runtime_put(&pdev->dev);
> >> pm_runtime_disable(&pdev->dev);
> >> +
> >> + if (pdev->dev.of_node)
> >> + thermal_zone_of_sensor_unregister(&pdev->dev, info->tzd);
> >> + else
> >> + thermal_zone_of_sensor_unregister(pdev->dev.parent, info->tzd);
> >> +
> >
> > Can't we just store the device used to create the zone in the
> > structure as well, that would avoid that non-trivial logic.
> >
>
> Yes we could.
>
> I've that same condition in the probe function, if pdev->dev.of_node
> then I continue in the probe dedicated to device probed from DT, or I
> continue with the probe dedicated to device probed via MFD.
>
> The thermal_zone_of_sensor_register is different in those two functions,
> thus, I thought that it would make more sense to replicate the same
> condition in the remove to make the relation between the device used in
> thermal_zone_of_sensor_register and unregister clear.
>
> That's just a matter of taste for me, so I'm definitely ok to write a v2
> for this change.
My point was only a cosmetic one. It's true that you have teh same
condition somewhere else, but you also have a lot of comments to
explain why there, and you don't have them here.
Adding a pointer to the structure would make it trivial for the
reader, without having to duplicate the comments.
Maxime
--
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
Attachment:
signature.asc
Description: PGP signature
