Re: [PATCH v3 0/5] phy: core: rework phy_set_mode to accept phy mode and submode

From: Maxime Ripard
Date: Mon Dec 17 2018 - 09:06:39 EST


Hi Grygorii, Kishon,

On Mon, Nov 19, 2018 at 07:24:19PM -0600, Grygorii Strashko wrote:
> Thank you for your review.
> I've not added "Tested-by"/"Acked-by" tags due to code changes in v3.
>
> As was discussed in [1] I'm posting series which introduces rework of
> phy_set_mode to accept phy mode and submode. I've dropped TI specific patches as
> this change is pretty big by itself.
>
> Patch 1 is cumulative change which refactors PHY framework code to
> support dual level PHYs mode configuration - PHY mode and PHY submode. It
> extends .set_mode() callback to support additional parameter "int submode"
> and converts all corresponding PHY drivers to support new .set_mode()
> callback declaration.
> The new extended PHY API
> int phy_set_mode_ext(struct phy *phy, enum phy_mode mode, int submode)
> is introduced to support dual level PHYs mode configuration and existing
> phy_set_mode() API is converted to macros, so PHY framework consumers do
> not need to be changed (~21 matches).
>
> Patches 2-4: Add new PHY's mode to be used by Ethernet PHY interface drivers or
> multipurpose PHYs like serdes and convert ocelot-serdes and mvebu-cp110-comphy
> PHY drivers to use recently introduced PHY_MODE_ETHERNET and phy_set_mode_ext().
>
> Patch 5 - removes unused, ethernet specific phy modes from enum phy_mode.
>
> Testing:
> - series tested on TI am335x/am437x/am5(dra7) paltforms.
> - other driver build tested.

I realise I'm a bit late to the party, but while working on the D-PHY
support, I noticed a few things that could be improved.

I guess the main issue is that the sub-mode is completely opaque to
the generic phy framework now. This might not be a big issue, and I
assume that it has been done that way because the net framework
already has a define for the submode it wants.

However, this creates a bunch of drawbacks at the phy framework level:

- phy_set_mode will now pass a submode of 0, all the time. This means
that the behaviour is undefined for all the modes not using the
submodes at the moment, and phy_interface_t seems to have the value
PHY_INTERFACE_MODE_NA matching 0, but I guess this could change in
the future (or the guarantee is not documented anywhere).
- on a similar note, there's no documentation for which value to pass
to phy_set_mode_ext when used with something else than a
PHY_MODE_ETHERNET.
- at the provider level, if you're supporting a phy that isn't using
the submodes, you have no way to filter out or reject any subnode,
since you have no idea what the "no submode" value is.

I guess this can be addressed by:

A) defining a generic phy framework wide unused / invalid phy submode,
that wouldn't collide with the subnode values (such as -1?), and
making phy_set_mode_ext use that.

B) moving the phy submodes definition to the generic phy headers. This
would allow to have a documented, obvious link between a mode and its
subnodes, for all the actors involved (consumer, provider, and
framework) without prior knowledge.

C) Document what the submodes expectations are

What do you think?
Maxime

--
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Attachment: signature.asc
Description: PGP signature