Re: [RFC PATCH v4 4/4] ptp_ocp: implement DPLL ops

[Jiri Pirko <jiri@resnulli.us>]

Wed, Dec 07, 2022 at 06:05:24PM CET, kuba@kernel.org wrote:
>On Wed, 7 Dec 2022 14:19:22 +0100 Jiri Pirko wrote:
>>> FWIW auxdev makes absolutely no sense to me for DPLL :/
>>> So Jiri, please say why.
>>
>> Why not? It's a subdev of a device. In mlx5, we have separate auxdevs
>> for eth, rdma, vnet, representors. DPLL is also a separate entity which
>> could be instantiated independently (as it is not really dependent on
>> eth/rdma/etc)). Auxdev looks like an awesome fit. Why do you think it is
>> not?
>>
>> Also, what's good about auxdev is that you can maintain them quite
>> independetly. So there is going to be driver/dpll/ directory which would
>> contain all dpll drivers.
>
>To what practical benefit? Where do we draw the line? Do you want
>PTP clocks to also be auxdevs? DPLL lives in netdev, we don't have
>to complicate things. auxdev is a Conway's law solution.

Auxdev infra is quite simple to implement, I'm not sure what do you mean
by complicating thing here.


>
>mlx5 already looks like sausage meat, it's already minced so you can
>fit it there quite easily, but don't force it on non-enterprise devices.

Not forcing, just suggesting. It's a low-hanging fruit, why not reach
it?

>
>There is non 1:1 relationship with a bus device and subsystem in Linux,
>LMK when you convinced Greg otherwise.

Sure there is not. But maybe that is due to the simple fact that auxdev
was introduces, what, 2 years back? My point is, we are introducing new
subsystem, wouldn't it be nice to start it clean?