On 06/20/2012 02:40 PM, Arnd Bergmann wrote: > On Wednesday 20 June 2012, Mark Brown wrote: >> On Wed, Jun 20, 2012 at 07:24:09PM +0000, Arnd Bergmann wrote: >> >>> 1. What is it that the new property can express that cannot already >>> be expressed by using the phandle. >> >> As well as being able to refer to the object from within the device tree >> we also need to be able to tell what the object represents - we have a >> bunch of regulators in an array under a node for a PMIC and we want to >> know which regulator on the physical device each array entry corresponds >> to. Previously this was being done by parsing the phandle name but that >> means we can't have more than one device with the same set of names. > > I see. > > But that sounds more like what the "reg" property does than what > the "compatible" property does for other devices. In other words, you > want to know which instance you are describing, not what kind of > object it is. Yes. > For a "compatible" property, I would expect two objects that have > the same register-level interface but are responsible for different > physical objects to have the same "compatible" values, but here > you specifically need distinct "regulator-compatible" values. I'd originally suggested "regulator-id" instead of "regulator-compatible" in order to avoid any association with the pre-existing "compatible" property. Mark didn't like that, although I don't think I quite understood why. >>> 2. Why is it called "regulator-compatible"? If it's similar to the >>> "compatible" property, don't you have to have a binding for each >>> possible string? If it's not related to the "compatible" property, >>> why is it named in a similar way? >> >> Yes, there are bindings defined already for all the relevant devices. >> Elsewhere in the binding document you'll see a list of all the >> regulators on the PMIC and the names by which the device tree binding >> knows them. > > Ah, so we just change the way how the binding refers to those names, > rather than the individual strings. > > It makes much more sense now, so please let's make sure that explanation > ends up in the changelog. > > I still have two more questions: > > * In case of tps65910, the individual regulators are all numbered, so > the driver can find them by their "reg" property. Should we still > mandate the use of the "regulator-compatible" property? It sounds > to me that we could better make it an optional property in those > cases. The problem is that dtc has no named constants. Using raw integers instead of names would make the .dts file rather unreadable. The issue is much more accute for regulators than say GPIOs or IRQs because there's likely no relative order to the set of regulators defined by the documentation, unlike for GPIOs/IRQs where the integer (often) is the object's primary ID. > * What is the difference between "regulator-name" and > "regulator-compatible"? Those are the same most of the time > in the patches, so can't we just make sure they are always the > same, and drop the confusingly named "regulator-compatible" one? regulator-compatible is the name of the regulator within the chip; the HW sub-module say. regulator-name is the board-/system-level name of the regulator. Perhaps a signal name from the schematic, or maybe function name like "SD card supply". -- To unsubscribe from this list: send the line "unsubscribe linux-doc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
