On Tue, Feb 27, 2024 at 11:50 PM Pali Rohár <pali@xxxxxxxxxx> wrote: > On Tuesday 27 February 2024 23:19:19 Andy Shevchenko wrote: > > On Tue, Feb 27, 2024 at 11:04 PM Pali Rohár <pali@xxxxxxxxxx> wrote: ... > > I'm wondering why we need all this. We have notifiers when a device is > > added / removed. We can provide a board_info for the device and attach > > it to the proper adapter, no? > > I do not know how flexible are notifiers. Can notifier call our callback > when new "struct i2c_adapter *adapter" was instanced? You can follow notifications of *an* I2C adapter being added / removed. With that, you can filter which one is that. Based on that you may attach a saved (at __init as you talked about in the reply to Hans) board_info with all necessary information. Something like this (combined) https://elixir.bootlin.com/linux/latest/source/drivers/ptp/ptp_ocp.c#L4515 https://elixir.bootlin.com/linux/latest/source/drivers/input/mouse/psmouse-smbus.c#L194 > > > With this simple change all dell smo8800 code would be in its subdir > > > drivers/platform/x86/dell/ and i2c-i801.c would get rid of smo code. > > > > > > This approach does not change any functionality, so should be absolutely > > > safe. > > > > > > Future changes will be done only in drivers/platform/x86/dell/ subdir, > > > touching i801 would not be needed at all. > > > > Still these exported functions are not the best solution we can do, > > right? We should be able to decouple them without need for the custom > > APIs. > > Well, what I described here is a simple change which get rid of the one > problem: i2c-i801.c contains SMO88xx related code and changing SMO88xx > logic (like adding a new device id) requires touching unrelated > i2c-i801.c source file. `get rid of one problem` --> `replace one by another (but maybe less critical, dunno) problem`. The new one is the spread of custom APIs for a single user, which also requires an additional, shared header file and all hell with the Kconfig dependencies. > I like small changes which can be easily reviewed and address one > problem. Step by step. That is why I proposed it here. > > For decoupling it is needed to get newly instanced adapter (if the > mentioned notifier is able to tell this information) and also it is > needed to check if the adapter is the i801. Yes. -- With Best Regards, Andy Shevchenko
