Re: [PATCH] USB: typec: tps6598x: use device 'type' field to identify devices

[Roger Quadros <rogerq@kernel.org>]

+Rob & Krzysztof

On 01/12/2023 10:10, Heikki Krogerus wrote:
> Hi Roger,
> 
> On Thu, Nov 30, 2023 at 03:30:54PM +0200, Roger Quadros wrote:
>> Hi Heikki,
>>
>> On 30/11/2023 12:54, Heikki Krogerus wrote:
>>> Hi Roger,
>>>
>>>>> Why not just match against the structures themselves?
>>>>>
>>>>>         if (tps->data == &tps25750_data)
>>>>>                 ...
>>>>
>>>> Then you need to declare tps25750_data and friends at the top of the file?
>>>>
>>>> A better approach might be to have type agnostic quirk flags for the special
>>>> behavior required for different types. This way, multiple devices can share
>>>> the same quirk if needed.
>>>>
>>>> e.g.
>>>> NEEDS_POWER_UP instead of TIPD_TYPE_APPLE_CD321X
>>>> SKIP_VID_READ instead of TIPD_TYPE_TI_TPS25750X
>>>> INIT_ON_RESUME instead of TIPD_TYPE_TI_TPS25750X
>>>>
>>>> Also rename cd321x_switch_power_state() to tps6598x_switch_power_state().
>>>
>>> No. Functions like that isolate cd321x specific functionality into an
>>> actual "function" just like they should.
>>>
>>> Quirk flags mean that if something breaks, it will almost always break
>>> for everybody (there is no real isolation with quirk flags), and when
>>> things are fixed and when features are added, we are forced to always
>>> "dance" around those quirk flags - you always have to consider them.
>>>
>>> Platform/device type checks are just as bad IMO, but in one way they
>>> are better than quirk flags. There is no question about what a
>>> platform check is checking, but quirk flags can so easily become
>>> incomprehensible (just what exactly does it mean when you say
>>> NEEDS_POWER_UP, SKIP_VID_READ and so on (you would need to document
>>> those quirks, which is waste of effort, and in reality nobody will do).
>>>
>>> In case of tipd/code.c, it should be converted into a library that
>>> only has the common/shared functionality. CD321, TPS2579x, TPS6598x
>>> and what ever there is, then will have a glue driver that handles
>>> everything that specific for their controller type.
>>
>> Do you mean that you want to treat the 3 devices as different incompatible devices
>> so each one has a separate driver which warrants for a different DT binding
>> for each and also Kconfig symbol?
> 
> I did not consider that, I was thinking that we would still continue
> with just one probe driver for all of these, but now that you
> mentioned this, maybe it would actually make sense to have separate
> full fledged probing drivers for all of these. Do you think it would
> be better like that? Would it be a problem to split the bindings?

I'm no DT expert but looks like an overkill to me.

-- 
cheers,
-roger