Re: [PATCH] drm/i915/dp_mst: Fix MST state after a sink reset

[Imre Deak <imre.deak@intel.com>]

On Thu, Aug 22, 2024 at 05:53:19PM +0300, Jani Nikula wrote:
> On Thu, 22 Aug 2024, Imre Deak <imre.deak@intel.com> wrote:
> > On Thu, Aug 22, 2024 at 01:00:03PM +0300, Jani Nikula wrote:
> >> On Wed, 21 Aug 2024, Imre Deak <imre.deak@intel.com> wrote:
> >> > On Wed, Aug 21, 2024 at 05:19:11PM +0300, Jani Nikula wrote:
> >> >> On Wed, 24 Jul 2024, Imre Deak <imre.deak@intel.com> wrote:
> >> >> > In some cases the sink can reset itself after it was configured into MST
> >> >> > mode, without the driver noticing the disconnected state. For instance
> >> >> > the reset may happen in the middle of a modeset, or the (long) HPD pulse
> >> >> > generated may be not long enough for the encoder detect handler to
> >> >> > observe the HPD's deasserted state. In this case the sink's DPCD
> >> >> > register programmed to enable MST will be reset, while the driver still
> >> >> > assumes MST is still enabled. Detect this condition, which will tear
> >> >> > down and recreate/re-enable the MST topology.
> >> >> >
> >> >> > Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/11195
> >> >> > Signed-off-by: Imre Deak <imre.deak@intel.com>
> >> >> > ---
> >> >> >  drivers/gpu/drm/i915/display/intel_dp.c     |  3 +++
> >> >> >  drivers/gpu/drm/i915/display/intel_dp_mst.c | 30 +++++++++++++++++++++
> >> >> >  drivers/gpu/drm/i915/display/intel_dp_mst.h |  1 +
> >> >> >  3 files changed, 34 insertions(+)
> >> >> >
> >> >> > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> >> >> > index 1e43e32e05199..c621f6daf8235 100644
> >> >> > --- a/drivers/gpu/drm/i915/display/intel_dp.c
> >> >> > +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> >> >> > @@ -5878,6 +5878,9 @@ intel_dp_detect(struct drm_connector *connector,
> >> >> >  	else
> >> >> >  		status = connector_status_disconnected;
> >> >> >  
> >> >> > +	if (!intel_dp_mst_verify_dpcd_state(intel_dp))
> >> >> > +		status = connector_status_disconnected;
> >> >> > +
> >> >> 
> >> >> So I still don't understand. We've got a detect which we've determined
> >> >> is connected. But then we disconnect because the sink state has changed.
> >> >> 
> >> >> How do we get another connect?
> >> >
> >> > The detect in this case will be retried with a 1 sec delay, since the
> >> > mode changes back to SST and the connector state stays disconnected.
> >> 
> >> What exactly triggers the retry? I don't see this triggering the
> >> INTEL_HOTPLUG_RETRY case.
> >
> > drm_connector::status is connector_status_disconnected before the detect
> > (as always for an MST root connector)
> 
> This was one of the missing pieces for me. I just thought we'd get here
> with connector_status_connected && is_mst == true, which would lead to a
> change.

Yes, what the connector status means for an MST root connector is not
very clear. How the detect will get retried also needs a comment here I
think, I'll add that.

> Very well then. One final nitpick below.
> 
> > and it stays
> > connector_status_disconnected. The connector's mode will also change to
> > SST (intel_dp::is_mst from true to false). This condition will lead to
> > the detect getting retried.
> >
> > A detect will actually happen in any case, since removing the MST
> > connectors results in a (global) hotplug uevent being sent, to which
> > userspace should respond with re-probing all connectors.
> >
> >> I'm just trying to understand, please bear with me.
> >> [...]
> >> >> > diff --git a/drivers/gpu/drm/i915/display/intel_dp_mst.c b/drivers/gpu/drm/i915/display/intel_dp_mst.c
> >> >> > index 27ce5c3f5951e..89b147e37b400 100644
> >> >> > --- a/drivers/gpu/drm/i915/display/intel_dp_mst.c
> >> >> > +++ b/drivers/gpu/drm/i915/display/intel_dp_mst.c
> >> >> > @@ -1998,3 +1998,33 @@ bool intel_dp_mst_crtc_needs_modeset(struct intel_atomic_state *state,
> >> >> >  
> >> >> >  	return false;
> >> >> >  }
> >> >> > +
> >> >> > +/**
> >> >> > + * intel_dp_mst_verify_dpcd_state - verify the MST SW enabled state wrt. the DPCD
> >> >> > + * @intel_dp: DP port object
> >> >> > + *
> >> >> > + * Verify if @intel_dp's MST enabled SW state matches the corresponding DPCD
> >> >> > + * state. A long HPD pulse -not long enough to be detected as a disconnected
> >> >> > + * state - could've reset the DPCD state, which requires tearing
> >> >> > + * down/recreating the MST topology.
> >> >> > + *
> >> >> > + * Returns %true if the SW MST enabled and DPCD states match, %false
> >> >> > + * otherwise.
> >> >> > + */
> >> >> > +bool intel_dp_mst_verify_dpcd_state(struct intel_dp *intel_dp)
> >> >> > +{
> >> >> > +	int ret;
> >> >> > +	u8 val;
> >> >> > +
> >> >> > +	if (!intel_dp->is_mst)
> >> >> > +		return true;
> >> >> > +
> >> >> > +	ret = drm_dp_dpcd_readb(intel_dp->mst_mgr.aux, DP_MSTM_CTRL, &val);
> >> >> > +	if (ret < 0)
> >> >> > +		return false;
> >> >> > +
> >> >> > +	if (val != (DP_MST_EN | DP_UP_REQ_EN | DP_UPSTREAM_IS_SRC))
> 
> Maybe let's add drm_dbg_kms() here so we know what's going on?

Ok, can add it.

> 
> BR,
> Jani.
> 
> 
> >> >> > +		return false;
> >> >> > +
> >> >> > +	return true;
> >> >> > +}
> >> >> > diff --git a/drivers/gpu/drm/i915/display/intel_dp_mst.h b/drivers/gpu/drm/i915/display/intel_dp_mst.h
> >> >> > index 8ca1d599091c6..9e4c7679f1c3a 100644
> >> >> > --- a/drivers/gpu/drm/i915/display/intel_dp_mst.h
> >> >> > +++ b/drivers/gpu/drm/i915/display/intel_dp_mst.h
> >> >> > @@ -27,5 +27,6 @@ int intel_dp_mst_atomic_check_link(struct intel_atomic_state *state,
> >> >> >  				   struct intel_link_bw_limits *limits);
> >> >> >  bool intel_dp_mst_crtc_needs_modeset(struct intel_atomic_state *state,
> >> >> >  				     struct intel_crtc *crtc);
> >> >> > +bool intel_dp_mst_verify_dpcd_state(struct intel_dp *intel_dp);
> >> >> >  
> >> >> >  #endif /* __INTEL_DP_MST_H__ */
> >> >> 
> >> >> -- 
> >> >> Jani Nikula, Intel
> >> 
> >> -- 
> >> Jani Nikula, Intel
> 
> -- 
> Jani Nikula, Intel