Re: [PATCH 3/7] drm/amd/display: Add handling for new "active color format" property

[Werner Sembach <wse@tuxedocomputers.com>]

Hi,

Am 10.01.24 um 14:09 schrieb Daniel Vetter:
> On Wed, 10 Jan 2024 at 13:53, Andri Yngvason <andri@yngvason.is> wrote:
>> mið., 10. jan. 2024 kl. 11:10 skrifaði Daniel Vetter <daniel@ffwll.ch>:
>>> On Tue, Jan 09, 2024 at 06:11:00PM +0000, Andri Yngvason wrote:
>>>> +     /* Extract information from crtc to communicate it to userspace as connector properties */
>>>> +     for_each_new_connector_in_state(state, connector, new_con_state, i) {
>>>> +             struct drm_crtc *crtc = new_con_state->crtc;
>>>> +             struct dc_stream_state *stream;
>>>> +
>>>> +             if (crtc) {
>>>> +                     new_crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
>>>> +                     dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
>>>> +                     stream = dm_new_crtc_state->stream;
>>>> +
>>>> +                     if (stream) {
>>>> +                             drm_connector_set_active_color_format_property(connector,
>>>> +                                     convert_dc_pixel_encoding_into_drm_color_format(
>>>> +                                             dm_new_crtc_state->stream->timing.pixel_encoding));
>>>> +                     }
>>>> +             } else {
>>>> +                     drm_connector_set_active_color_format_property(connector, 0);
>>> Just realized an even bigger reason why your current design doesn't work:
>>> You don't have locking here.
>>>
>>> And you cannot grab the required lock, which is
>>> drm_dev->mode_config.mutex, because that would result in deadlocks. So
>>> this really needs to use the atomic state based design I've described.
>>>
>> Maybe we should just drop "actual color format" and instead fail the
>> modeset if the "preferred color format" property cannot be satisfied?
>> It seems like the simplest thing to do here, though it is perhaps less
>> convenient for userspace. In that case, the "preferred color format"
>> property should just be called "color format".
> Yeah that's more in line with how other atomic properties work. This
> way userspace can figure out what works with a TEST_ONLY commit too.
> And for this to work you probably want to have an "automatic" setting
> too.
> -Sima

The problem with TEST_ONLY probing is that color format settings are 
interdependent: https://gitlab.freedesktop.org/drm/amd/-/issues/476#note_966634

So changing any other setting may require every color format to be TEST_ONLY 
probed again.

Greetings

Werner