[Freedreno] RFC: DSI host capabilities (was: [PATCH RFC 03/10] drm/panel: Add LGD panel driver for Sony Xperia XZ3)

[Neil Armstrong]

On 06/07/2023 09:24, Maxime Ripard wrote:
> On Wed, Jul 05, 2023 at 11:09:40PM +0300, Dmitry Baryshkov wrote:
>> On 05/07/2023 19:53, Maxime Ripard wrote:
>>> On Wed, Jul 05, 2023 at 06:20:13PM +0300, Dmitry Baryshkov wrote:
>>>> On Wed, 5 Jul 2023 at 17:24, Maxime Ripard <mripard at kernel.org> wrote:
>>>>>
>>>>> On Wed, Jul 05, 2023 at 04:37:57PM +0300, Dmitry Baryshkov wrote:
>>>>>>>>>
>>>>>>>>> Either way, I'm not really sure it's a good idea to multiply the
>>>>>>>>> capabilities flags of the DSI host, and we should just stick to the
>>>>>>>>> spec. If the spec says that we have to support DSC while video is
>>>>>>>>> output, then that's what the panels should expect.
>>>>>>>>
>>>>>>>> Except some panels supports DSC & non-DSC, Video and Command mode, and
>>>>>>>> all that is runtime configurable. How do you handle that ?
>>>>>>>
>>>>>>> In this case, most of the constraints are going to be on the encoder
>>>>>>> still so it should be the one driving it. The panel will only care about
>>>>>>> which mode has been selected, but it shouldn't be the one driving it,
>>>>>>> and thus we still don't really need to expose the host capabilities.
>>>>>>
>>>>>> This is an interesting perspective. This means that we can and actually have
>>>>>> to extend the drm_display_mode with the DSI data and compression
>>>>>> information.
>>>>>
>>>>> I wouldn't extend drm_display_mode, but extending one of the state
>>>>> structures definitely.
>>>>>
>>>>> We already have some extra variables in drm_connector_state for HDMI,
>>>>> I don't think it would be a big deal to add a few for MIPI-DSI.
>>>>>
>>>>> We also floated the idea for a while to create bus-specific states, with
>>>>> helpers to match. Maybe it would be a good occasion to start doing it?
>>>>>
>>>>>> For example, the panel that supports all four types for the 1080p should
>>>>>> export several modes:
>>>>>>
>>>>>> 1920x1080-command
>>>>>> 1920x1080-command-DSC
>>>>>> 1920x1080-video
>>>>>> 1920x1080-video-DSC
>>>>>>
>>>>>> where video/command and DSC are some kinds of flags and/or information in
>>>>>> the drm_display_mode? Ideally DSC also has several sub-flags, which denote
>>>>>> what kind of configuration is supported by the DSC sink (e.g. bpp, yuv,
>>>>>> etc).
>>>>>
>>>>> So we have two things to do, right? We need to expose what the panel can
>>>>> take (ie, EDID for HDMI), and then we need to tell it what we picked
>>>>> (infoframes).
>>>>>
>>>>> We already express the former in mipi_dsi_device, so we could extend the
>>>>> flags stored there.
>>>>>
>>>>> And then, we need to tie what the DSI host chose to a given atomic state
>>>>> so the panel knows what was picked and how it should set everything up.
>>>>
>>>> This is definitely something we need. Marijn has been stuck with the
>>>> panels that support different models ([1]).
>>>>
>>>> Would you prefer a separate API for this kind of information or
>>>> abusing atomic_enable() is fine from your point of view?
>>>>
>>>> My vote would be for going with existing operations, with the slight
>>>> fear of ending up with another DSI-specific hack (like
>>>> pre_enable_prev_first).
>>>
>>> I don't think we can get away without getting access to the atomic_state
>>> from the panel at least.
>>>
>>> Choosing one setup over another is likely going to depend on the mode,
>>> and that's only available in the state.
>>>
>>> We don't have to go the whole way though and create the sub-classes of
>>> drm_connector_state, but I think we should at least provide it to the
>>> panel.
>>>
>>> What do you think of creating a new set of atomic_* callbacks for
>>> panels, call that new set of functions from msm and start from there?
>>
>> We are (somewhat) bound by the panel_bridge, but yeah, it seems possible.
> 
> Bridges have access to the atomic state already, so it's another place
> to plumb this through but I guess it would still be doable?

It's definitely doable, but I fear we won't be able to test most of the
panel drivers, should we introduce a new atomic set of panel callbacks ?

Or shall be simply move the "new" panel driver supporting atomic to bridge
and only use panel_bridge for basic panels ?

Neil

> 
> Maxime