Re: [PATCH v12 01/12] drm: add connector LUMINANCE property

From: Ville Syrjälä

Date: Fri Nov 20 2026 - 12:34:02 EST


On Fri, Nov 20, 2026 at 11:19:13AM -0600, Mario Limonciello wrote:
>
>
> On 11/20/26 10:41, Ville Syrjälä wrote:
> > On Thu, Nov 19, 2026 at 05:27:28PM -0600, Mario Limonciello wrote:
> >> Panel brightness is a property of a display, and thus of a DRM
> >> connector. Add an atomic per-connector LUMINANCE range property so
> >> compositors can manage relative panel luminance together with the rest
> >> of the display state.
> >>
> >> Keep the core backend-agnostic. Drivers create the property with their
> >> hardware range and implement connector helper callbacks to set luminance
> >> or disable panel emission. The atomic helpers sequence those callbacks
> >> around pipeline enable and disable, including luminance-only commits.
> >> The panel is left lit when the pipeline enters self refresh, and the
> >> luminance is only reprogrammed when it changes or a modeset enables the
> >> pipeline.
> >>
> >> The stored property value is independent of display power. A commit made
> >> while the connector is inactive is retained and applied on the next
> >> enable; disabling the pipeline does not overwrite it.
> >>
> >> LUMINANCE values are relative, driver-defined levels rather than
> >> absolute luminance in cd/m^2. The interface supports emissive panels as
> >> well as panels with a separate backlight.
> >>
> >> Keep the interface behind a default-off DRM_LUMINANCE option. Migrated
> >> drivers retain their legacy backlight class ABI unless the new interface
> >> is explicitly selected, allowing the DRM UAPI and legacy compatibility
> >> policy to remain separate.
> >>
> >> Acked-by: Javier Martinez Canillas <javierm@xxxxxxxxxx>
> >> Signed-off-by: Mario Limonciello <mario.limonciello@xxxxxxx>
> >> ---
> >> v11:
> >> * Don't call .disable_luminance when the CRTC enters self refresh (Sashiko)
> >> * Only call .set_luminance when the luminance changes or on a modeset
> >> that isn't a self refresh exit (Leo Li, Sashiko)
> >> * Fix the .set_luminance self refresh documentation
> >> v10:
> >> * Add tag
> >> ---
> >> drivers/gpu/drm/Kconfig | 14 +++++
> >> drivers/gpu/drm/drm_atomic_helper.c | 98 +++++++++++++++++++++++++++++-
> >> drivers/gpu/drm/drm_atomic_state_helper.c | 7 +++
> >> drivers/gpu/drm/drm_atomic_uapi.c | 5 ++
> >> drivers/gpu/drm/drm_connector.c | 57 +++++++++++++++++
> >> drivers/gpu/drm/tests/drm_connector_test.c | 28 +++++++++
> >> include/drm/drm_atomic_helper.h | 3 +
> >> include/drm/drm_connector.h | 16 +++++
> >> include/drm/drm_modeset_helper_vtables.h | 28 +++++++++
> >> 9 files changed, 255 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
> >> index 61111d50b1c65..1bbe2c2d7c76f 100644
> >> --- a/drivers/gpu/drm/Kconfig
> >> +++ b/drivers/gpu/drm/Kconfig
> >> @@ -33,6 +33,20 @@ endmenu
> >>
> >> if DRM
> >>
> >> +config DRM_LUMINANCE
> >> + bool "DRM connector luminance support"
> >> + help
> >> + Expose relative luminance as an atomic DRM connector property. Values
> >> + are driver-defined levels rather than absolute luminance in cd/m^2.
> >> + Drivers provide connector helper callbacks to program their hardware;
> >> + the DRM core does not depend on a particular backend.
> >> +
> >> + Drivers migrated to this interface will use it instead of registering
> >> + a legacy backlight class device. Leave this disabled to preserve the
> >> + de-facto userspace ABI in /sys/class/backlight.
> >
> > I don't think this will fly if it breaks the backlight class interface.
> > No one can actually enable this outside of some niche embedded use
> > cases where backwards compatibility is not an issue. I think what
> > you need is a generic backlight class implementation on top of this.
> >
>
> The idea with this was:
> 1) to get the series landed
> 2) let compositors merge solutions around it.
> 3) Then add backwards compatibility to kernel so we can flip the Kconfig
> default.

With that plan, I think you need to mark it BROKEN if it will
actually break current setups when some unsuspecting user flips
it on.

--
Ville Syrjälä
Intel