On Wed, Jun 17, 2020 at 11:41 AM Anson Huang <anson.hu...@nxp.com> wrote:
> > > > > > I'm ok with the change. But I'm curious how can this module be > > > autoloaded without MODULE_DEVICE_TABLE. > > > Have you tested if it can work? > > > > > > > I ONLY tested the manual insmod, if want to support auto load, may need > > some more change, will try it later and send out a V2 if needed. > > The further check shows that, if want to support auto load, the platform > device > register needs to be done in somewhere else which is built-in (in my test, I > move it > to clk-imx8qxp.c's probe), and also need to add below module alias in this > driver, > because it has no device node in DT and no device table in driver. > > +MODULE_ALIAS("platform:imx-scu-soc"); > > Since this driver has no device node in DT, and the target is to build all > SoC specific > drivers as module, so the best way is to add a virtual device node in DT in > order to support > auto load? I see that there is indeed a driver for the device node in drivers/firmware/imx/imx-scu.c, the only reason for this module using device_initcall() with a manual platform_device_register_simple() seems to be that we cannot have two platform drivers bind to the same device node. I think a cleaner way to handle this would be to just move the entire soc driver into the firmware driver and then remove the duplication. Arnd