Hi Mark, > Am 01.12.2021 um 16:10 schrieb Mark Brown <broonie@xxxxxxxxxx>: > > On Wed, Dec 01, 2021 at 03:33:24PM +0100, H. Nikolaus Schaller wrote: >>> Am 01.12.2021 um 15:03 schrieb Paul Cercueil <paul@xxxxxxxxxxxxxxx>: > >>> Please make it mandatory in DTS then, and use devm_regulator_get() in the driver. > >> Well, I just wonder why the elegant devm_regulator_get_optional() exists at all >> and seems to be used in ca. 80 places. > > Frankly because half of them are broken usages like this since people > seem determined to have the most fragile error handling they can :/ I see. I had made the mistake myself to not check for NULL pointer on regulator_disable here... > There are valid use cases for it, with things like SD cards where some > supplies are genuinely optional and simply constrain what features are > available if they're omitted from the design. You also see some devices > with the ability to replace internal regulators with external ones. > >> And if it is not allowed, why some DTS should be forced to add not physically existing dummy-regulators. > > Again, if the supply can be physically absent that is a sensible use > case but that means completely absent, not just not software > controllable. We can represent fixed voltage regulators just fine. The question may be how we can know for a more generic driver that there is always a regulator. In the present case we know the schematics but it is just one example. > >> AFAIR drivers should implement functionality defined by DTS but not the other way round: enforce DTS style. >> BTW: there is no +5 mains dummy regulator defined in ci20.dts. > > It wouldn't be the first time a DTS were incomplete, and I'm sure it > won't be the last. > >> What I fear is that if we always have to define the mains +5V (which is for example not >> defined in ci20.dts), which rules stops us from asking to add a dummy-regulator from 110/230V to +5V as well. > > It is good practice to specify the full tree of supplies all the way to > the main supply rail of the board, this ensures that if we need the > information for something we've got it (even if that thing is just that > we've got to the root of the tree). There's potential applications in > battery supplied devices for managing very low power situations. Indeed. So let's modify it as you have suggested. BR and thanks, Nikolaus
