On Sat, 18 Apr 2020 19:25:15 +0200 Paul Cercueil <paul@xxxxxxxxxxxxxxx> wrote: > Hi Jonathan, > > Le sam. 18 avril 2020 à 15:22, Jonathan Cameron <jic23@xxxxxxxxxx> a > écrit : > > On Sat, 18 Apr 2020 15:24:58 +0200 > > Paul Cercueil <paul@xxxxxxxxxxxxxxx> wrote: > > > >> Le sam. 18 avril 2020 à 15:42, Andy Shevchenko > >> <andy.shevchenko@xxxxxxxxx> a écrit : > >> > On Sat, Apr 18, 2020 at 3:10 PM Paul Cercueil > >> <paul@xxxxxxxxxxxxxxx> > >> > wrote: > >> >> Le sam. 18 avril 2020 à 14:57, Andy Shevchenko > >> >> <andy.shevchenko@xxxxxxxxx> a écrit : > >> >> > On Sat, Apr 18, 2020 at 1:48 AM Paul Cercueil > >> >> <paul@xxxxxxxxxxxxxxx> > >> >> > wrote: > >> >> >> Le sam. 18 avril 2020 à 0:49, Andy Shevchenko > >> >> >> <andy.shevchenko@xxxxxxxxx> a écrit : > >> >> >> > On Sat, Apr 18, 2020 at 12:24 AM Paul Cercueil > >> >> >> <paul@xxxxxxxxxxxxxxx> > >> >> >> > wrote: > >> >> >> >> Le sam. 18 avril 2020 à 0:10, Andy Shevchenko > >> >> >> >> <andy.shevchenko@xxxxxxxxx> a écrit : > >> >> >> >> > On Fri, Apr 17, 2020 at 11:21 PM Artur Rojek > >> >> >> >> <contact@xxxxxxxxxxxxxx> > >> >> >> >> > wrote: > >> > > >> > ... > >> > > >> >> >> >> >> +#include <linux/of.h> > >> >> >> >> > > >> >> >> >> > Do you really need this? (See below as well) > >> >> >> > > >> >> >> >> >> +static const struct of_device_id > >> >> adc_joystick_of_match[] = > >> >> >> { > >> >> >> >> >> + { .compatible = "adc-joystick", }, > >> >> >> >> >> + { }, > >> >> >> >> >> +}; > >> >> >> >> >> +MODULE_DEVICE_TABLE(of, adc_joystick_of_match); > >> >> >> >> >> + > >> >> >> >> >> +static struct platform_driver adc_joystick_driver > >> = { > >> >> >> >> >> + .driver = { > >> >> >> >> >> + .name = "adc-joystick", > >> >> >> >> > > >> >> >> >> >> + .of_match_table = > >> >> >> >> >> of_match_ptr(adc_joystick_of_match), > >> >> >> >> > > >> >> >> >> > Drop this a bit harmful of_match_ptr() macro. It > >> should go > >> >> >> with > >> >> >> >> ugly > >> >> >> >> > #ifdeffery. Here you simple introduced a compiler > >> warning. > >> >> >> >> > >> >> >> >> I assume you mean #ifdef around the of_device_id + > >> module > >> >> table > >> >> >> >> macro? > >> >> >> > > >> >> >> > Yes. > >> >> >> > > >> >> >> >> > On top of that, you are using device property API, OF > >> use > >> >> in > >> >> >> this > >> >> >> >> case > >> >> >> >> > is contradictory (at lest to some extend). > >> >> >> >> > >> >> >> >> I don't see why. The fact that the driver can work when > >> >> probed > >> >> >> from > >> >> >> >> platform code > >> >> >> > > >> >> >> > Ha-ha, tell me how. I would like to be very surprised. > >> >> >> > >> >> >> iio_map_array_register(), > >> >> >> pinctrl_register_mappings(), > >> >> >> platform_add_devices(), > >> >> >> > >> >> >> you're welcome. > >> >> > > >> >> > I think above has no relation to what I'm talking about. > >> >> > >> >> Yes it does. It allows you to map the IIO channels, set the > >> pinctrl > >> >> configurations and register a device from platform code instead > >> of > >> >> devicetree. > >> > > >> > I'm not talking about other drivers, I'm talking about this > >> driver and > >> > how it will be instantiated. Above, according to the code, can't > >> be > >> > comprehensive to fulfill this. > >> > >> This is how the platform devices were instanciated on JZ4740 before > >> we > >> switched everything to devicetree. > >> > >> >> > How *this* driver can work as a platform instantiated one? > >> >> > We seems have a conceptual misunderstanding here. > >> >> > > >> >> > For example, how can probe of this driver not fail, if it is > >> not > >> >> > backed by a DT/ACPI properties? > >> >> > >> >> platform_device_add_properties(). > >> > > >> > Yes, I waited for this. And seems you don't understand the (scope > >> of) > >> > API, you are trying to insist this driver can be used as a > >> platform > >> > one. > >> > Sorry, I must to disappoint you, it can't. Above interface is > >> created > >> > solely for quirks to support (broken) DT/ACPI tables. It's not > >> > supposed to be used as a main source for the device properties. > >> > >> The fact that it was designed for something else doesn't mean it > >> can't > >> be used. > >> > >> Anyway, this discussion is pointless. I don't think anybody would > >> want > >> to do that. > >> > >> >> >> >> doesn't mean that it shouldn't have a table to probe > >> >> >> >> from devicetree. > >> >> >> > > >> >> >> > I didn't get what you are talking about here. The idea of > >> >> >> _unified_ > >> >> >> > device property API is to get rid of OF-centric code in > >> >> favour of > >> >> >> more > >> >> >> > generic approach. Mixing those two can be done only in > >> >> specific > >> >> >> cases > >> >> >> > (here is not the one). > >> >> >> > >> >> >> And how are we mixing those two here? The only OF-centric > >> thing > >> >> >> here is > >> >> >> the device table, which is required if we want the driver to > >> >> probe > >> >> >> from > >> >> >> devicetree. > >> >> > > >> >> > Table is fine(JFYI the types and sections are defined outside > >> of > >> >> OF > >> >> > stuff, though being [heavily] used by it) , API > >> (of_match_ptr() > >> >> macro > >> >> > use) is not. > >> >> > >> >> Sorry, but that's just stupid. Please have a look at how > >> >> of_match_ptr() > >> >> macro is defined in <linux/of.h>. > >> > > >> > Call it whatever you want, but above code is broken. > >> > >> of_match_ptr() is basically defined like this: > >> > >> #ifdef CONFIG_OF > >> #define of_match_ptr(x) (x) > >> #else > >> #define of_match_ptr(x) NULL > >> #endif > >> > >> So please, enlighten me, tell me what is so wrong about what's being > >> done here. > >> > >> > It needs either of: > >> > - ugly ifdeffery > >> > - dropping of_match_ptr() > >> > - explicit dependence to OF > >> > > >> > My choice is second one. Because it makes code better and allows > >> also > >> > ACPI to use this driver (usually) without changes. > >> > >> And how is unconditionally compiling the of_match_table make it > >> magically probe from ACPI, without a acpi_match_table? > >> > >> -Paul > > > > Look up PRP0001 ACPI ID. Magic trick ;) > > > > https://www.kernel.org/doc/html/latest/firmware-guide/acpi/enumeration.html?highlight=PRP0001 > > > > It allows you to define an ACPI device in DSDT that is instantiated > > from what is effectively the DT binding including the id table. > > So what you're saying, is that the OF table should be present, even > though CONFIG_OF is not set, just in case it is probed from ACPI? Exactly. Weird isn't it :) > > -Paul > >