mbox series

[0/3] phy: qcom: edp: Add support for DT phy mode configuration

Message ID 20231219-x1e80100-phy-edp-compatible-refactor-v1-0-f9e77752953d@linaro.org (mailing list archive)
Headers show
Series phy: qcom: edp: Add support for DT phy mode configuration | expand

Message

Abel Vesa Dec. 19, 2023, 8:55 p.m. UTC
Until now, all platform that supported both eDP and DP had different
compatibles for each mode. Using different compatibles for basically
the same IP block but for a different configuration is bad way all
around. There is a new compute platform from Qualcomm that supports
both eDP and DP with the same PHY. So instead of following the old
method, we should allow the mode to be configured from devicetree.

There has been an off-list discussion on what would be the right way
to pass on the PHY mode information to the driver and it has been
concluded that phy-cells is the way to go. This means that basically
the controller will pass another value (that is, the PHY type) to
its 'phys' DT property.

For this, we need both the bindings value and the PHY mode value to be
added as well.

The controller part will follow shortly. But for now, lets see where
this is going.

There has been another attempt at this here:
https://lore.kernel.org/all/20231122-phy-qualcomm-edp-x1e80100-v3-3-576fc4e9559d@linaro.org/

Compared to that version, this one uses the phy-cells method and drops
the X1E80100 support. The X1E80100 support will be a separate patchset.

Signed-off-by: Abel Vesa <abel.vesa@linaro.org>
---
Abel Vesa (3):
      dt-bindings: phy: Add PHY_TYPE_EDP definition
      phy: Add PHY Embedded DisplayPort mode
      phy: qcom: edp: Allow PHY mode configuration via devicetree

 drivers/phy/qualcomm/phy-qcom-edp.c | 89 ++++++++++++++++++++++++++++---------
 include/dt-bindings/phy/phy.h       |  1 +
 include/linux/phy/phy.h             |  3 +-
 3 files changed, 70 insertions(+), 23 deletions(-)
---
base-commit: 0e182d9523f6c0af49357fcd812eaa702bd4b403
change-id: 20231219-x1e80100-phy-edp-compatible-refactor-8733eca7ccda

Best regards,

Comments

Dmitry Baryshkov Dec. 21, 2023, 4:27 p.m. UTC | #1
On Tue, 19 Dec 2023 at 22:55, Abel Vesa <abel.vesa@linaro.org> wrote:
>
> Until now, all platform that supported both eDP and DP had different
> compatibles for each mode. Using different compatibles for basically
> the same IP block but for a different configuration is bad way all
> around. There is a new compute platform from Qualcomm that supports
> both eDP and DP with the same PHY. So instead of following the old
> method, we should allow the mode to be configured from devicetree.
>
> There has been an off-list discussion on what would be the right way
> to pass on the PHY mode information to the driver and it has been
> concluded that phy-cells is the way to go. This means that basically
> the controller will pass another value (that is, the PHY type) to
> its 'phys' DT property.
>
> For this, we need both the bindings value and the PHY mode value to be
> added as well.
>
> The controller part will follow shortly. But for now, lets see where
> this is going.
>
> There has been another attempt at this here:
> https://lore.kernel.org/all/20231122-phy-qualcomm-edp-x1e80100-v3-3-576fc4e9559d@linaro.org/
>
> Compared to that version, this one uses the phy-cells method and drops
> the X1E80100 support. The X1E80100 support will be a separate patchset.

After several back and forth discussions, I think that this approach
is not correct and not that easy to extend. Instead I'd like to
suggest adding a property to the DP controller, which enables eDP
behaviour (and thus makes DP driver call phy_set_mode()). Something
like this:
dp: displayport-controller@ae0000 {
    compatible = "qcom,sm8000-dp";
    /* reg, interrupts, etc */
   edp-interface;
   /* or simpler */
   is-edp;
};

What do you think?
Konrad Dybcio Jan. 3, 2024, 1:42 p.m. UTC | #2
On 21.12.2023 17:27, Dmitry Baryshkov wrote:
> On Tue, 19 Dec 2023 at 22:55, Abel Vesa <abel.vesa@linaro.org> wrote:
>>
>> Until now, all platform that supported both eDP and DP had different
>> compatibles for each mode. Using different compatibles for basically
>> the same IP block but for a different configuration is bad way all
>> around. There is a new compute platform from Qualcomm that supports
>> both eDP and DP with the same PHY. So instead of following the old
>> method, we should allow the mode to be configured from devicetree.
>>
>> There has been an off-list discussion on what would be the right way
>> to pass on the PHY mode information to the driver and it has been
>> concluded that phy-cells is the way to go. This means that basically
>> the controller will pass another value (that is, the PHY type) to
>> its 'phys' DT property.
>>
>> For this, we need both the bindings value and the PHY mode value to be
>> added as well.
>>
>> The controller part will follow shortly. But for now, lets see where
>> this is going.
>>
>> There has been another attempt at this here:
>> https://lore.kernel.org/all/20231122-phy-qualcomm-edp-x1e80100-v3-3-576fc4e9559d@linaro.org/
>>
>> Compared to that version, this one uses the phy-cells method and drops
>> the X1E80100 support. The X1E80100 support will be a separate patchset.
> 
> After several back and forth discussions, I think that this approach
> is not correct and not that easy to extend. Instead I'd like to
> suggest adding a property to the DP controller, which enables eDP
> behaviour (and thus makes DP driver call phy_set_mode()). Something
> like this:
> dp: displayport-controller@ae0000 {
>     compatible = "qcom,sm8000-dp";
>     /* reg, interrupts, etc */
>    edp-interface;
>    /* or simpler */
>    is-edp;
> };
> 
> What do you think?

Please excuse my alzheimer, but why did we not go with phy-type after
the last discussion?

Konrad
Abel Vesa Jan. 15, 2024, 9:52 a.m. UTC | #3
On 24-01-03 14:42:49, Konrad Dybcio wrote:
> On 21.12.2023 17:27, Dmitry Baryshkov wrote:
> > On Tue, 19 Dec 2023 at 22:55, Abel Vesa <abel.vesa@linaro.org> wrote:
> >>
> >> Until now, all platform that supported both eDP and DP had different
> >> compatibles for each mode. Using different compatibles for basically
> >> the same IP block but for a different configuration is bad way all
> >> around. There is a new compute platform from Qualcomm that supports
> >> both eDP and DP with the same PHY. So instead of following the old
> >> method, we should allow the mode to be configured from devicetree.
> >>
> >> There has been an off-list discussion on what would be the right way
> >> to pass on the PHY mode information to the driver and it has been
> >> concluded that phy-cells is the way to go. This means that basically
> >> the controller will pass another value (that is, the PHY type) to
> >> its 'phys' DT property.
> >>
> >> For this, we need both the bindings value and the PHY mode value to be
> >> added as well.
> >>
> >> The controller part will follow shortly. But for now, lets see where
> >> this is going.
> >>
> >> There has been another attempt at this here:
> >> https://lore.kernel.org/all/20231122-phy-qualcomm-edp-x1e80100-v3-3-576fc4e9559d@linaro.org/
> >>
> >> Compared to that version, this one uses the phy-cells method and drops
> >> the X1E80100 support. The X1E80100 support will be a separate patchset.
> > 
> > After several back and forth discussions, I think that this approach
> > is not correct and not that easy to extend. Instead I'd like to
> > suggest adding a property to the DP controller, which enables eDP
> > behaviour (and thus makes DP driver call phy_set_mode()). Something
> > like this:
> > dp: displayport-controller@ae0000 {
> >     compatible = "qcom,sm8000-dp";
> >     /* reg, interrupts, etc */
> >    edp-interface;
> >    /* or simpler */
> >    is-edp;
> > };
> > 
> > What do you think?
> 
> Please excuse my alzheimer, but why did we not go with phy-type after
> the last discussion?

phy-type would be a property of the phy. That way we would need pass
the mode to the controller. So it was concluded that passing that
information from the controller via phy_set_mode is more straightforward.

> 
> Konrad
Konrad Dybcio Jan. 25, 2024, 4:36 p.m. UTC | #4
On 1/15/24 10:52, Abel Vesa wrote:
> On 24-01-03 14:42:49, Konrad Dybcio wrote:
>> On 21.12.2023 17:27, Dmitry Baryshkov wrote:
>>> On Tue, 19 Dec 2023 at 22:55, Abel Vesa <abel.vesa@linaro.org> wrote:
>>>>
>>>> Until now, all platform that supported both eDP and DP had different
>>>> compatibles for each mode. Using different compatibles for basically
>>>> the same IP block but for a different configuration is bad way all
>>>> around. There is a new compute platform from Qualcomm that supports
>>>> both eDP and DP with the same PHY. So instead of following the old
>>>> method, we should allow the mode to be configured from devicetree.
>>>>
>>>> There has been an off-list discussion on what would be the right way
>>>> to pass on the PHY mode information to the driver and it has been
>>>> concluded that phy-cells is the way to go. This means that basically
>>>> the controller will pass another value (that is, the PHY type) to
>>>> its 'phys' DT property.
>>>>
>>>> For this, we need both the bindings value and the PHY mode value to be
>>>> added as well.
>>>>
>>>> The controller part will follow shortly. But for now, lets see where
>>>> this is going.
>>>>
>>>> There has been another attempt at this here:
>>>> https://lore.kernel.org/all/20231122-phy-qualcomm-edp-x1e80100-v3-3-576fc4e9559d@linaro.org/
>>>>
>>>> Compared to that version, this one uses the phy-cells method and drops
>>>> the X1E80100 support. The X1E80100 support will be a separate patchset.
>>>
>>> After several back and forth discussions, I think that this approach
>>> is not correct and not that easy to extend. Instead I'd like to
>>> suggest adding a property to the DP controller, which enables eDP
>>> behaviour (and thus makes DP driver call phy_set_mode()). Something
>>> like this:
>>> dp: displayport-controller@ae0000 {
>>>      compatible = "qcom,sm8000-dp";
>>>      /* reg, interrupts, etc */
>>>     edp-interface;
>>>     /* or simpler */
>>>     is-edp;
>>> };
>>>
>>> What do you think?
>>
>> Please excuse my alzheimer, but why did we not go with phy-type after
>> the last discussion?
> 
> phy-type would be a property of the phy. That way we would need pass
> the mode to the controller. So it was concluded that passing that
> information from the controller via phy_set_mode is more straightforward.

Eeh, reluctantly, I'm gonna say yes. It's not the prettiest solution,
but none of the ones I can think of seem much better.

Konrad