[PATCH v2 0/2] drm/colorop: Keep colorop state consistent across atomic commits

[PATCH v2 0/2] drm/colorop: Keep colorop state consistent across atomic commits

[Chaitanya Kumar Borah]

This series aims to keep colorop state consistent across atomic
transactions by ensuring it accurately reflects committed hardware
state and remains part of the atomic update whenever its associated
plane is involved.

It contains two changes:
- Preserves the bypass value in duplicated colorop state.

_drm_atomic_helper_colorop_duplicate_state() unconditionally reset
bypass to true, which means the duplicated state no longer reflects the
committed hardware state. Since bypass directly controls whether the
colorop is active in hardware, this can lead to an unintended disable
during subsequent commits.

This could potentially be a problem also for colorops where bypass value
is immutably false.

Conceptually, I consider 'bypass' to behave similar to 'visible' in plane 
state - it represents current HW state and should therefore be preserved
across duplication.

- Add affected colorops with affected plane

Colorops are unique in the DRM model. While they are DRM objects with their
own states, they are logically attached to a plane and exposed through
a plane property. In some sense, they share the same hierarchy as CRTC and
planes while following a different 'ownership' model.

Given that enabling a CRTC pulls in all its affected planes into the atomic
state, it follows that when a plane is added, its associated colorops are
also included. Otherwise, during modesets or internal commits, colorop state
may be missing from the transaction, resulting in inconsistent or incomplete
state updates.

That said, I do have a concern about potentially inflating the atomic
state by automatically pulling in colorops from the core. It is not
entirely clear to me whether inclusion of affected colorops should be
handled in core, or left to individual drivers.

My understanding of the atomic framework is still evolving, so
I would appreciate feedback from those more familiar with the intended
design direction.

==
Chaitanya

P.S/Background/TL;DR:

I discovered inconsistency with the colorop state while analysing CRC mismatches
in kms_color_pipeline test cases[1]. Visual inspection reveals that while CRC is
being collected degamma block has been reset. This was traced back to the internal
commit that the driver does to disable PSR2 and selective fetch for CRC collection.

crtc_crc_open
    -> intel_crtc_set_crc_source
        -> intel_crtc_crc_setup_workarounds
            -> drm_atomic_commit

During this flow colorop states are never added to the atomic state which in turn
makes intel_plane_color_copy_uapi_to_hw_state() disable the colorops.

If we add the colorops, to the atomic state, the problem still persisted because
while duplicating the colorop state, 'bypass' was getting reset to true.

The two changes made in this series fixes the issue.

[1] https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_18001/shard-mtlp-6/igt@kms_color_pipeline@plane-lut1d.html

v2:
  - Add affected colorops only when a pipeline is enabled

Cc: Simon Ser <contact@emersion.fr>
Cc: Alex Hung <alex.hung@amd.com>
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Daniel Stone <daniels@collabora.com>
Cc: Melissa Wen <mwen@igalia.com>
Cc: Sebastian Wick <sebastian.wick@redhat.com>
Cc: Alex Hung <alex.hung@amd.com>
Cc: Uma Shankar <uma.shankar@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Louis Chauvet <louis.chauvet@bootlin.com>
Cc: <stable@vger.kernel.org> #v6.19+

Chaitanya Kumar Borah (2):
  drm/colorop: Preserve bypass value in duplicate_state()
  drm/atomic: Add affected colorops with affected planes

 drivers/gpu/drm/drm_atomic.c  | 7 +++++++
 drivers/gpu/drm/drm_colorop.c | 2 --
 2 files changed, 7 insertions(+), 2 deletions(-)

-- 
2.25.1

[PATCH v2 1/2] drm/colorop: Preserve bypass value in duplicate_state()

[Chaitanya Kumar Borah]

__drm_atomic_helper_colorop_duplicate_state() unconditionally
sets state->bypass = true after copying the existing state.

This override causes the new atomic state to no longer reflect
the currently committed hardware state. Since the bypass property
directly controls whether the colorop is active in hardware,
resetting it to true can inadvertently disable an active colorop
during a subsequent commit, particularly for internal driver commits
where userspace does not touch the property.

Drop the unconditional assignment and preserve the duplicated
bypass value.

Fixes: 8c5ea1745f4c ("drm/colorop: Add BYPASS property")
Cc: <stable@vger.kernel.org> #v6.19+
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Signed-off-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
---
 drivers/gpu/drm/drm_colorop.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/gpu/drm/drm_colorop.c b/drivers/gpu/drm/drm_colorop.c
index f421c623b3f0..e44a738c4c14 100644
--- a/drivers/gpu/drm/drm_colorop.c
+++ b/drivers/gpu/drm/drm_colorop.c
@@ -466,8 +466,6 @@ static void __drm_atomic_helper_colorop_duplicate_state(struct drm_colorop *colo
 
 	if (state->data)
 		drm_property_blob_get(state->data);
-
-	state->bypass = true;
 }
 
 struct drm_colorop_state *
-- 
2.25.1

Re: [PATCH v2 1/2] drm/colorop: Preserve bypass value in duplicate_state()

[Harry Wentland]

On 2026-03-10 07:32, Chaitanya Kumar Borah wrote:
> __drm_atomic_helper_colorop_duplicate_state() unconditionally
> sets state->bypass = true after copying the existing state.
> 
> This override causes the new atomic state to no longer reflect
> the currently committed hardware state. Since the bypass property
> directly controls whether the colorop is active in hardware,
> resetting it to true can inadvertently disable an active colorop
> during a subsequent commit, particularly for internal driver commits
> where userspace does not touch the property.
> 
> Drop the unconditional assignment and preserve the duplicated
> bypass value.
> 
> Fixes: 8c5ea1745f4c ("drm/colorop: Add BYPASS property")
> Cc: <stable@vger.kernel.org> #v6.19+
> Reviewed-by: Uma Shankar <uma.shankar@intel.com>

Reviewed-by: Harry Wentland <harry.wentland@amd.com>

Harry

> Signed-off-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
> ---
>  drivers/gpu/drm/drm_colorop.c | 2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_colorop.c b/drivers/gpu/drm/drm_colorop.c
> index f421c623b3f0..e44a738c4c14 100644
> --- a/drivers/gpu/drm/drm_colorop.c
> +++ b/drivers/gpu/drm/drm_colorop.c
> @@ -466,8 +466,6 @@ static void __drm_atomic_helper_colorop_duplicate_state(struct drm_colorop *colo
>  
>  	if (state->data)
>  		drm_property_blob_get(state->data);
> -
> -	state->bypass = true;
>  }
>  
>  struct drm_colorop_state *

[PATCH v2 2/2] drm/atomic: Add affected colorops with affected planes

[Chaitanya Kumar Borah]

When drm_atomic_add_affected_planes() adds a plane to the atomic
state, the associated colorops are not guaranteed to be included.
This can leave colorop state out of the transaction when planes
are pulled in implicitly (eg. during modeset or internal commits).

Also add affected colorops when adding affected planes to keep
plane and color pipeline state consistent within the atomic
transaction.

v2: Add affected colorops only when a pipeline is enabled

Fixes: 2afc3184f3b3 ("drm/plane: Add COLOR PIPELINE property")
Cc: <stable@vger.kernel.org> #v6.19+
Reviewed-by: Uma Shankar <uma.shankar@intel.com> #v1
Signed-off-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
---
 drivers/gpu/drm/drm_atomic.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
index 04925166df98..dd9f27cfe991 100644
--- a/drivers/gpu/drm/drm_atomic.c
+++ b/drivers/gpu/drm/drm_atomic.c
@@ -1587,6 +1587,7 @@ drm_atomic_add_affected_planes(struct drm_atomic_state *state,
 	const struct drm_crtc_state *old_crtc_state =
 		drm_atomic_get_old_crtc_state(state, crtc);
 	struct drm_plane *plane;
+	int ret;
 
 	WARN_ON(!drm_atomic_get_new_crtc_state(state, crtc));
 
@@ -1600,6 +1601,12 @@ drm_atomic_add_affected_planes(struct drm_atomic_state *state,
 
 		if (IS_ERR(plane_state))
 			return PTR_ERR(plane_state);
+
+		if (plane_state->color_pipeline) {
+			ret = drm_atomic_add_affected_colorops(state, plane);
+			if (ret)
+				return ret;
+		}
 	}
 	return 0;
 }
-- 
2.25.1

Re: [PATCH v2 2/2] drm/atomic: Add affected colorops with affected planes

[Harry Wentland]

On 2026-03-10 07:32, Chaitanya Kumar Borah wrote:
> When drm_atomic_add_affected_planes() adds a plane to the atomic
> state, the associated colorops are not guaranteed to be included.
> This can leave colorop state out of the transaction when planes
> are pulled in implicitly (eg. during modeset or internal commits).
> 
> Also add affected colorops when adding affected planes to keep
> plane and color pipeline state consistent within the atomic
> transaction.
> 
> v2: Add affected colorops only when a pipeline is enabled
> 
> Fixes: 2afc3184f3b3 ("drm/plane: Add COLOR PIPELINE property")
> Cc: <stable@vger.kernel.org> #v6.19+
> Reviewed-by: Uma Shankar <uma.shankar@intel.com> #v1

Reviewed-by: Harry Wentland <harry.wentland@amd.com>

Harry

> Signed-off-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
> ---
>  drivers/gpu/drm/drm_atomic.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/drivers/gpu/drm/drm_atomic.c b/drivers/gpu/drm/drm_atomic.c
> index 04925166df98..dd9f27cfe991 100644
> --- a/drivers/gpu/drm/drm_atomic.c
> +++ b/drivers/gpu/drm/drm_atomic.c
> @@ -1587,6 +1587,7 @@ drm_atomic_add_affected_planes(struct drm_atomic_state *state,
>  	const struct drm_crtc_state *old_crtc_state =
>  		drm_atomic_get_old_crtc_state(state, crtc);
>  	struct drm_plane *plane;
> +	int ret;
>  
>  	WARN_ON(!drm_atomic_get_new_crtc_state(state, crtc));
>  
> @@ -1600,6 +1601,12 @@ drm_atomic_add_affected_planes(struct drm_atomic_state *state,
>  
>  		if (IS_ERR(plane_state))
>  			return PTR_ERR(plane_state);
> +
> +		if (plane_state->color_pipeline) {
> +			ret = drm_atomic_add_affected_colorops(state, plane);
> +			if (ret)
> +				return ret;
> +		}
>  	}
>  	return 0;
>  }

Re: [PATCH v2 0/2] drm/colorop: Keep colorop state consistent across atomic commits

[Harry Wentland]

Would you like me to pull these into drm-misc-next? I'd love
the changes as base for me next patchset.

Harry

On 2026-03-10 07:32, Chaitanya Kumar Borah wrote:
> This series aims to keep colorop state consistent across atomic
> transactions by ensuring it accurately reflects committed hardware
> state and remains part of the atomic update whenever its associated
> plane is involved.
> 
> It contains two changes:
> - Preserves the bypass value in duplicated colorop state.
> 
> _drm_atomic_helper_colorop_duplicate_state() unconditionally reset
> bypass to true, which means the duplicated state no longer reflects the
> committed hardware state. Since bypass directly controls whether the
> colorop is active in hardware, this can lead to an unintended disable
> during subsequent commits.
> 
> This could potentially be a problem also for colorops where bypass value
> is immutably false.
> 
> Conceptually, I consider 'bypass' to behave similar to 'visible' in plane 
> state - it represents current HW state and should therefore be preserved
> across duplication.
> 
> - Add affected colorops with affected plane
> 
> Colorops are unique in the DRM model. While they are DRM objects with their
> own states, they are logically attached to a plane and exposed through
> a plane property. In some sense, they share the same hierarchy as CRTC and
> planes while following a different 'ownership' model.
> 
> Given that enabling a CRTC pulls in all its affected planes into the atomic
> state, it follows that when a plane is added, its associated colorops are
> also included. Otherwise, during modesets or internal commits, colorop state
> may be missing from the transaction, resulting in inconsistent or incomplete
> state updates.
> 
> That said, I do have a concern about potentially inflating the atomic
> state by automatically pulling in colorops from the core. It is not
> entirely clear to me whether inclusion of affected colorops should be
> handled in core, or left to individual drivers.
> 
> My understanding of the atomic framework is still evolving, so
> I would appreciate feedback from those more familiar with the intended
> design direction.
> 
> ==
> Chaitanya
> 
> P.S/Background/TL;DR:
> 
> I discovered inconsistency with the colorop state while analysing CRC mismatches
> in kms_color_pipeline test cases[1]. Visual inspection reveals that while CRC is
> being collected degamma block has been reset. This was traced back to the internal
> commit that the driver does to disable PSR2 and selective fetch for CRC collection.
> 
> crtc_crc_open
>     -> intel_crtc_set_crc_source
>         -> intel_crtc_crc_setup_workarounds
>             -> drm_atomic_commit
> 
> During this flow colorop states are never added to the atomic state which in turn
> makes intel_plane_color_copy_uapi_to_hw_state() disable the colorops.
> 
> If we add the colorops, to the atomic state, the problem still persisted because
> while duplicating the colorop state, 'bypass' was getting reset to true.
> 
> The two changes made in this series fixes the issue.
> 
> [1] https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_18001/shard-mtlp-6/igt@kms_color_pipeline@plane-lut1d.html
> 
> v2:
>   - Add affected colorops only when a pipeline is enabled
> 
> Cc: Simon Ser <contact@emersion.fr>
> Cc: Alex Hung <alex.hung@amd.com>
> Cc: Harry Wentland <harry.wentland@amd.com>
> Cc: Daniel Stone <daniels@collabora.com>
> Cc: Melissa Wen <mwen@igalia.com>
> Cc: Sebastian Wick <sebastian.wick@redhat.com>
> Cc: Alex Hung <alex.hung@amd.com>
> Cc: Uma Shankar <uma.shankar@intel.com>
> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Cc: Jani Nikula <jani.nikula@intel.com>
> Cc: Louis Chauvet <louis.chauvet@bootlin.com>
> Cc: <stable@vger.kernel.org> #v6.19+
> 
> Chaitanya Kumar Borah (2):
>   drm/colorop: Preserve bypass value in duplicate_state()
>   drm/atomic: Add affected colorops with affected planes
> 
>  drivers/gpu/drm/drm_atomic.c  | 7 +++++++
>  drivers/gpu/drm/drm_colorop.c | 2 --
>  2 files changed, 7 insertions(+), 2 deletions(-)
> 

Re: [PATCH v2 0/2] drm/colorop: Keep colorop state consistent across atomic commits

[Borah, Chaitanya Kumar]

Hello Harry,

On 3/14/2026 2:15 AM, Harry Wentland wrote:
> Would you like me to pull these into drm-misc-next? I'd love
> the changes as base for me next patchset.

Thank you for the reviews. Suraj(in cc) already merged them through 
drm-misc, they should already be in.


==
Chaitanya