Re: [PATCH v1 2/2] phy: phy-cadence-torrent: Use PHY kernel APIs to set PHY attributes

[Kishon Vijay Abraham I <kishon@ti.com>]

Hi,

On 5/18/2020 12:24 PM, Swapnil Kashinath Jakhade wrote:
> Hi Kishon,
> 
>> -----Original Message-----
>> From: Kishon Vijay Abraham I <kishon@ti.com>
>> Sent: Wednesday, May 13, 2020 8:08 AM
>> To: Tomi Valkeinen <tomi.valkeinen@ti.com>; Maxime Ripard
>> <maxime@cerno.tech>; Swapnil Kashinath Jakhade
>> <sjakhade@cadence.com>
>> Cc: Yuti Suresh Amonkar <yamonkar@cadence.com>; linux-
>> kernel@vger.kernel.org; mark.rutland@arm.com; jsarha@ti.com;
>> praneeth@ti.com; Milind Parab <mparab@cadence.com>; Vinod Koul
>> <vkoul@kernel.org>
>> Subject: Re: [PATCH v1 2/2] phy: phy-cadence-torrent: Use PHY kernel APIs to
>> set PHY attributes
>>
>> EXTERNAL MAIL
>>
>>
>> Hi,
>>
>> On 5/8/2020 1:20 PM, Tomi Valkeinen wrote:
>>> On 07/05/2020 20:17, Maxime Ripard wrote:
>>>
>>>>> Actually, for this particular case, consumer driver will be the
>>>>> Cadence MHDP bridge driver for DisplayPort which is also under
>>>>> review process for upstreaming [1]. So this DRM bridge driver will
>>>>> make use of the PHY APIs
>>>>> phy_get_bus_width() and phy_get_max_link_rate() during execution of
>>>>> probe function to get the number of lanes and maximum link rate
>>>>> supported by Cadence Torrent PHY. This information is required to
>>>>> set the host capabilities in the DRM bridge driver, based on which
>>>>> initial values for DisplayPort link training will be determined.
>>>>>
>>>>> The changes in this PHY patch series are based on suggestions in the
>>>>> review comments in [1] which asks to use kernel PHY APIs to read
>>>>> these properties instead of directly accessing PHY device node. The
>>>>> complete driver and actual use of these APIs can be found in [2].
>>>>> This is how we are planning to use these APIs.
>>>>
>>>> I haven't really looked into the displayport spec, but I'd assume
>>>> that there's a lot more parameters that would need to be negociated
>>>> between the phy and the DP block? If so, then it would make more
>>>> sense to follow the path we did for MIPI-DSI where the parameters can
>>>> be negociated through the phy_configure / phy_validate interface.
>>>
>>> I don't think this is negotiation, but just exposing the (max)
>>> capabilities of PHY, inside which the configure can work. Maybe all
>>> the capabilities could handled with a struct (struct phy_attrs),
>>> instead of adding separate functions for each, though.
>>
>> yeah, that makes sense. Just that users should take care not to over-write all
>> the phy attributes with partial information.
> 
> It would be really helpful if you could clarify a bit regarding how to handle this
> exactly. What I could understand from Tomi' suggestion is that all PHY attributes
> in struct phy_attrs should have single pair of functions to get and set all the PHY
> attributes (e.g. phy_get_attrs / phy_set_attrs), instead of separate get/set pair of
> functions for individual attribute (bus_width, mode, max_link_rate etc). Is this
> understanding correct? If so, how should the existing functions for bus_width and
> mode be used?

Yes, your understanding is correct. There are already existing users of
bus_width, mode, so let's not disturb that. That could maybe deprecated later.

Thanks
Kishon