On Thu, Oct 13, 2022 at 12:14:12PM +0200, Sascha Hauer wrote: > On Wed, Oct 12, 2022 at 11:08:06AM -0500, Rob Herring wrote: > > On Tue, Oct 11, 2022 at 10:29:24AM +0200, Sascha Hauer wrote: > > > Following the example of samsung,picophy-dc-vol-level-adjust more > > > phy tuning properties are added for configuring the remaining bitfields > > > in the USBNC_n_PHY_CFG1 register. > > > > All these properties really doesn't scale. These properties should go > > in the phy node as they are properties or the phy. There's no rule that > > you can only read properties from the driver's device node. > > I understand and agree. > > On i.MX8M we currently use the usb-nop-xceiv. I guess it's not an option > to just add these properties there, so we'll need a phy node with a new > compatible like fsl,imx8mm-usbphy. The driver would basically just > register a usb-nop-xceiv and the node would be a container for the new > property. Does this sound sane? I think it would be fine if you do: compatible = "fsl,imx8mm-usbphy", "usb-nop-xceiv"; You'll have to rework the usb-nop-xceiv schema 'select' like we have to do for any compatible appearing in multiple schema files. Or don't have the fallback and add "fsl,imx8mm-usbphy" to the phy-generic.c driver. But that should be marked for stable for at least some compatibility with old kernels. For existing kernel binaries to work, you have to go with the former option. Rob
