Re: [PATCH v5 0/7] can: refactoring of can-dev module and of Kbuild

[Oliver Hartkopp <socketcan@hartkopp.net>]

Hi Vincent,

On 07.06.22 04:49, Vincent MAILHOL wrote:
> On Tue. 7 Jun. 2022 at 04:43, Oliver Hartkopp <socketcan@hartkopp.net> wrote:
>>

>>>             |
>>>             +-> All other CAN devices not relying on RX offload
>>>             |
>>>             +-> CAN rx offload
>>>                 symbol: CONFIG_CAN_RX_OFFLOAD
>>
>> Is this still true in patch series 5?
>>
>> If I understood it correctly CONFIG_CAN_BITTIMING and
>> CONFIG_CAN_RX_OFFLOAD can be enabled by the user and
>> (alternatively/additionally) the selection of "flexcan, m_can, mcp251xfd
>> and ti_hecc" enables CONFIG_CAN_RX_OFFLOAD too.
>>
>> Right?
> 
> Yes, this is correct. Maybe what troubles you is the meaning of the
> "x --> y" arrow in the graph. I said it denotes that "y depends on x".
> Here "depends on" has a loose meaning. It translates to either:
>    * Feature Y is encapsulated in Kbuild by some "if X/endif" and won't
> show up unless X is selected.
>    * Feature Y has a "selects X" tag and will forcibly enable X if selected.
> 
> CONFIG_CAN_*CALC*_BITTIMING is on the left side of an arrow starting
> from CONFIG_CAN_NETLINK so it "depends" on CONFIG_CAN_NETLINK. On the
> other hand, CONFIG_CAN_*CALC*_BITTIMING does not have any arrow
> starting from it so indeed, it can be enabled by the user
> independently of the other features as long as CONFIG_CAN_NETLINK is
> selected.

Ok.

> CONFIG_CAN_RX_OFFLOAD is also on the left side of an arrow starting
> from CONFIG_CAN_NETLINK. Furthermore, there is an arrow starting from
> CONFIG_CAN_RX_OFFLOAD going to the "rx offload drivers". So those
> drivers need CONFIG_CAN_RX_OFFLOAD (which is implemented using the
> "selects CONFIG_CAN_RX_OFFLOAD"). However, CONFIG_CAN_RX_OFFLOAD can
> be selected independently of the "rx offload drivers" as long as its
> CONFIG_CAN_NETLINK dependency is met.
> 
> So I think that the diagram is correct. Maybe rephrasing the cover
> letter as below would address your concerns?
I applied your series and played with the options and it works like 
charm - and as expected.

But the point remains that from your figure I would still assume that 
the M_CAN driver would only show up when CONFIG_CAN_RX_OFFLOAD was 
selected by the user.

But the current (good) implementation shows *all* drivers and selects 
CONFIG_CAN_RX_OFFLOAD when e.g. M_CAN is selected.

So what about:

   symbol: CONFIG_NETDEVICES
   |
   +-> CAN Device Drivers
       symbol: CONFIG_CAN_DEV
       |
       +-> software/virtual CAN device drivers
       |   (at time of writing: slcan, vcan, vxcan)
       |
       +-> hardware CAN device drivers with Netlink support
           symbol: CONFIG_CAN_NETLINK (matches previous CONFIG_CAN_DEV)
           |
           +-> CAN bit-timing calculation (optional for all drivers)
           |   symbol: CONFIG_CAN_BITTIMING
           |
           +-> CAN rx offload (optional but selected by some drivers)
           |   symbol: CONFIG_CAN_RX_OFFLOAD
           |
           +-> CAN devices drivers
               (some may select CONFIG_CAN_RX_OFFLOAD)

(I also added 'hardware' to CAN device drivers with Netlink support) to 
have a distinction to 'software/virtual' CAN device drivers)

At least this would help me to understand the new configuration setup.

Best regards,
Oliver