On Fri, Jan 25, 2019 at 01:06:45PM +0100, Jonas Bonn wrote: > On 25/01/2019 12:53, Baolin Wang wrote: > > On Fri, 25 Jan 2019 at 19:44, Jonas Bonn wrote: > > Can we configure it at runtime by the device rather than at DT time by > > the controller? If yes, we already have a patch for this, please > > check: > It's a characteristic of the SPI slave, in the same sense as CPOL/CPHA are, > and therefore it makes sense to specify it in the device tree. No, that doesn't follow at all. There's two bits here - where the configuration gets done and the mechanism by which it gets done. If something in DT is completely orthogonal to which device it is a property of. > Having this as device property rather than a transfer property allows this > to be configured one time in setup() rather than having to fiddle with the > configuration register for every transfer. That doesn't mean that the coniguration should be done in DT though, and given that this presumably is a property of the device there seems to be no reason why we'd have it in DT - if every instance of the device is going to need to set the property we should just figure it out from the compatble string instead.