Re: [RFC] drm/msm/dpu: set possible_crtcs on cursor planes

From: Erik Tagirov

Date: Wed Nov 18 2026 - 16:24:18 EST


On Wed, Nov 18, 2026 at 10:22:48PM +0200, Dmitry Baryshkov wrote:
> On Wed, Nov 18, 2026 at 09:47:42PM +0100, Erik Tagirov wrote:
> > I am using a lesser known wayland compositor that uses the possible_crtcs field of a plane to detect
> > which one to use for the cursor. On the Lenovo T14s two cursor planes are created, each with possible_crtcs set
> > to 0xff. Which from my understanding means that they can be used with any crtc.
> >
> > However, the compositor picks the wrong cursor plane which results in a lag of ~17ms
> > on each DRM_IOCTL_MODE_SETPLANE call. Making the cursor movements sluggish.
> >
> > After some investigation, I found out that this is due to the legacy_cursor_update flag not being set
> > in drm_atomic_helper_update_plane() which has an impact in msm_atomic_commit_tail() where the fallback branch
> > is taken.
> >
> > The lag is solved with the patch below.
> >
> > Best regards
> > - Erik Tagirov
> >
> > The cursor plane must be used with the correct crtc to ensure an
> > asynchronous commit
> > ---
> > drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c
> > index da3556eb6ecc..4e0e9c2d8062 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c
> > @@ -877,6 +877,9 @@ static int _dpu_kms_drm_obj_init(struct dpu_kms *dpu_kms)
> > ret = PTR_ERR(crtc);
> > return ret;
> > }
> > + /* Ensure that cursor planes are commited asynchronously */
> > + if (cursor_planes[i])
> > + cursor_planes[i]->possible_crtcs = drm_crtc_mask(crtc);
>
> Here you are setting that the plane can only be used on a particular
> CRTC. This is not true for these planes (even being cursor, it still can
> be used on all the CRTCs).
>
Thank you for the quick reply, I understand that the hardware supports
universal planes, so this is not a good solution.

I will see how I can solve this from userspace instead, need to look
more closely at the DRM API maybe there is another way to determine what
plane belongs to what CRTC.
The current implementation does not work, because the possible_crtcs is set to 0xff
and the compositor picks the cursor plane that comes first in the list which is used by crtc109
instead of one used by crtc108. This why the legacy_cursor_update flag is not set in
drm_atomic_helper_update_plane()

BR
- Erik
> > }
> >
> > /* All CRTCs are compatible with all encoders */
> > --
> > 2.49.0
> >
>
> --
> With best wishes
> Dmitry