On Mon, Feb 17, 2020 at 09:48:36AM +0100, Michael Walle wrote: > > My opinion is that all compatibles should be defined explicitly in > > bindings doc. In above example, the possible values of <processor> > > should be given. This must be done anyway, as we are moving to > > json-schema bindings. > > But if they are listed in the document, they also have to be in the > of_device_id table, correct? I do not think so. Documenting compatibles used in DTS now doesn't necessarily mean we need to use it in kernel driver right away. Bindings doc is a specification for device tree, not kernel. With the compatible in DTS and bindings, kernel can start using it at any time when there is a need, like dealing with SoC quirks or bugs found later. Shawn > Which somehow contradicts the talk Pankaj > mentioned [1,2]. Eg. > > compatible = "fsl,ls1028ar1-flexcan","fsl,lx2160ar1-flexcan"; > > Doesn't make any sense, because the "fsl,ls1028ar1-flexcan" is alreay > in the driver and the fallback "fsl,lx2160ar1-flexcan" isn't needed. > > OTOH the talk is already 2 to 3 years old and things might have changed > since then. > > -michael > > [1] https://elinux.org/images/0/0e/OSELAS.Presentation-ELCE2017-DT.pdf > [2] https://www.youtube.com/watch?v=6iguKSJJfxo
