Hi Pekka,

El jue, 2 oct 2025 a las 10:34, Pekka Paalanen
(<[email protected]>) escribió:
>
> On Wed, 1 Oct 2025 11:26:09 +0200
> Guillermo Rodriguez Garcia <[email protected]> wrote:
>
> > Hi Marius,
> >
> > Yes, indeed, setting transform=normal fixes the issue.
> >
> > Is this the expected behaviour?
> >
> > In my experience, a rotation=... in the device tree is a configuration
> > parameter for the display controller (LCDC/DRM driver) that tells the
> > kernel driver that the panel is mounted physically rotated, and the
> > driver should compensate by rotating the framebuffer output before
> > sending it to the panel. I would not expect userspace to apply an
> > additional rotation.
> >
> > Weston seems to be interpreting this as "the panel is physically
> > rotated, and it is Weston's job to compensate for that rotation,
> > unless a different transform is explicitly requested". I find this a
> > bit confusing.
>
> Hi,
>
> I would believe the Weston interpretation is correct, because I would
> not assume that all hardware is able to rotate on-the-fly as needed,
> especially 90-degree rotations. OTOH, if the driver automatically
> rotated the stream to make framebuffers appear in the correct
> orientation, why would the rotation be communicated to userspace at all?
>
> Reading "panel orientation" from
> https://docs.kernel.org/gpu/drm-kms.html#standard-connector-properties
> also implies that userspace is expected to act on it.
>
> Maybe the kernel fbcon code is not using and programming rotation
> properties correctly? It should have the same issue as Weston.

I will check. I am not using fbcon currently but I can give it a go
and see what it does.

>
> For hardware which can rotate the picture on-the-fly, KMS has the plane
> property "rotation" for programming it. There does not yet seem to be a
> CRTC property for rotation after composition.
>
> Therefore this seems a driver bug: if the driver automatically rotates,
> then it should tell userspace the panel orientation is normal. Sounds
> like this might break touchscreen input coordinates without further
> calibration, though.

Will check and report back.

Thank you,

Guillermo Rodriguez Garcia
[email protected]

Reply via email to