On Fri, Aug 26, 2022 at 11:11:21AM +0200, Michael Rodin wrote:
> "detect" should not be called and its return value shall not be used when a
> connector is forced as hinted in the commit d50ba256b5f1 ("drm/kms: start
> adding command line interface using fb.") and the commit 6fe14acd496e
> ("drm: Document drm_connector_funcs"). One negative side effect of doing
> this is observed on the RCar3 SoCs which use the dw-hdmi driver. It
> continues executing drm_helper_hpd_irq_event even if its connector is
> forced to ON. As consequence drm_helper_hpd_irq_event calls "detect" so the
> connector status is updated to "disconnected":
> 
> [  420.201527] [drm:drm_helper_hpd_irq_event] [CONNECTOR:76:HDMI-A-1] status updated from connected to disconnected
> 
> This status is corrected by drm_helper_probe_single_connector_modes shortly
> after this because this function checks if a connector is forced:
> 
> [  420.218703] [drm:drm_helper_probe_single_connector_modes] [CONNECTOR:76:HDMI-A-1] status updated from disconnected to connected
> 
> To avoid similar issues this commit adapts functions which call "detect"
> so they check if a connector is forced and return the correct status.
> 
> Fixes: 949f08862d66 ("drm: Make the connector .detect() callback optional")
> Signed-off-by: Michael Rodin <mrodin@de.adit-jv.com>
> ---
>  drivers/gpu/drm/drm_probe_helper.c | 16 ++++++++++++++--
>  1 file changed, 14 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_probe_helper.c b/drivers/gpu/drm/drm_probe_helper.c
> index bb427c5a4f1f..1691047d0472 100644
> --- a/drivers/gpu/drm/drm_probe_helper.c
> +++ b/drivers/gpu/drm/drm_probe_helper.c
> @@ -289,7 +289,12 @@ drm_helper_probe_detect_ctx(struct drm_connector *connector, bool force)
>  retry:
>  	ret = drm_modeset_lock(&connector->dev->mode_config.connection_mutex, &ctx);
>  	if (!ret) {
> -		if (funcs->detect_ctx)
> +		if (connector->force == DRM_FORCE_ON ||
> +		    connector->force == DRM_FORCE_ON_DIGITAL)
> +			ret = connector_status_connected;
> +		else if (connector->force == DRM_FORCE_OFF)
> +			ret = connector_status_disconnected;
> +		else if (funcs->detect_ctx)
>  			ret = funcs->detect_ctx(connector, &ctx, force);
>  		else if (connector->funcs->detect)
>  			ret = connector->funcs->detect(connector, force);

Actually, I think this creates a regression on vc4 at least (and
possibly i915?), when the HDMI scrambling is on.

When a monitor is disconnected and then connected again, the mode won't
change by itself, but it will lose its scrambling status. The way we
dealt with it (and documented here:
https://lore.kernel.org/all/20220829134731.213478-9-maxime@cerno.tech/)
is to have detect enabling the scrambler again if it detects a
disconnection/connection transition.

If we force the status on all the time, then we will never get called
and we won't have the opportunity to enable the scrambler again.

Maxime