Hi Andy, On Wed, 18 Jan 2023 19:43:23 +0200 Andy Shevchenko <andriy.shevchenko@xxxxxxxxx> wrote: > On Wed, Jan 18, 2023 at 07:28:20PM +0200, Tomi Valkeinen wrote: > > On 18/01/2023 18:01, Andy Shevchenko wrote: > > > On Wed, Jan 18, 2023 at 02:40:24PM +0200, Tomi Valkeinen wrote: > > > > Hi, > > > > > > > > You can find the v6 from: > > > > > > > > https://lore.kernel.org/all/20230105140307.272052-1-tomi.valkeinen@xxxxxxxxxxxxxxxx/ > > > > > > > > Main changes: > > > > > > > > * i2c-atr: Use bus notifier. This allows us to drop the patch that adds > > > > the attach_client/detach_client callbacks. On the downside, it removes > > > > the option for error handling if the translation setup fails, and also > > > > doesn't provide us the pointer to the i2c_board_info. I belive both > > > > are acceptable downsides. > > > > > > > > * Use fwnode in the fpdlink drivers instead of OF > > > > > > > > * Addressed all the review comments (I hope) > > > > > > > > * Lots of cosmetic or minor fixes which I came up while doing the fwnode > > > > change > > > > > > I believe my comments to the first driver applies to the next two, so please > > > address them whenever you are agree / it's possible / it makes sense. > > > > > > About ATR implementation. We have the i2c bus (Linux representation of > > > the driver model) and i2c_adapter and i2c_client objects there. Can't we > > > have an i2c_client_aliased in similar way and be transparent with users? > > > Can you clarify what you mean here? > > > > The i2c_clients are not aware of the i2c-atr. They are normal i2c clients. > > The FPD-Link drivers are aware of the ATR, as the FPD-Link hardware contains > > the ATR support. > > Can't that hardware be represented as I2C adapter? In such case the ATR specifics > can be hidden from the client (drivers). > > I'm worrying about code duplication and other things that leak into drivers as > ATR callbacks. Which callbacks do you refer to? i2c_atr_ops? I don't think we can do without the attach/detach_client ones, it's where the driver-specific implementation is hooked for the generic ATR infra to call it. However now I noticed the select/deselect ops are still there. IIRC they are not used by any driver and in the past the plan was to just remove them. Tomi, do you think there is a good reason to keep them? > It might be that I didn't get how hw exactly functioning on this > level and why we need those callbacks. As far as "how hw exactly works", in case you haven't seen that, the best explanation I was able to give is in my ELCE 2019 talk, at minute ~22. It's a 2-3 minute watch. The slides have pointers to other talks and discussion. -- Luca Ceresoli, Bootlin Embedded Linux and Kernel engineering https://bootlin.com