On Fri, Aug 28, 2020 at 12:08:32PM +0300, Heikki Krogerus wrote: > Hi Greg, > > On Fri, Aug 28, 2020 at 09:43:59AM +0200, Greg Kroah-Hartman wrote: > > I still find this crazy that a whole separate driver is created just to > > read a single 32bit value. > > > > Why not put this logic in the driver that wants to read that value? > > That would be much simpler, smaller, and more obvious. > > That would mean that we start maintaining something like DMI quirk > table in those drivers. Unfortunately the IOM device is not available > on every platform. Also, even on platforms that do have it, there is > no guarantee that the device is always going to be mapped to the same > address. Ok, but putting it in another driver has the same problem as you are now depending on that driver with the driver that calls this, right? The quirk table would go in here, but not somewhere else? And if this really is only for typec stuff, put it in the typec directory, not burried off in the platform driver subdir please. > Nevertheless, I was originally hoping that we could hide the handling > of IOM somehow in ACPI without the need for an actual device object, > but it now turns out that the other features of the IOM chip have > created interest. At least our i915 guys probable have some use for it > (I don't know exactly what they are planning to use it for). > > So the fact that we may later need the device for something else, on > top of the clumsiness and most importantly risks involved with using > ACPI to take care of extra tasks (ASL tends to have bugs - bugs that > may never ever get fixed), I think the IOM device object, and the > driver that binds to it, do have a valid reason for existing. Handle the situation today, if, in the future, someone else needs/wants this, _then_ work on splitting it out into separate pieces. Don't create additional complexity today, for no benefit today. It's already caused numerous review comments/complaints the way this is designed... thanks, greg k-h
