[Intel-gfx] [PATCH] drm/i915/pps: added get_pps_idx() hook as part of pps_get_register() cleanup

[Ville Syrjälä]

On Wed, Aug 03, 2022 at 11:13:38AM +0300, Jani Nikula wrote:
> On Wed, 03 Aug 2022, Animesh Manna <animesh.manna at intel.com> wrote:
> > To support dual LFP two instances of pps added from display gen12 onwards.
> > Few older platform like VLV also has dual pps support but handling is
> > different. So added separate hook get_pps_idx() to formulate which pps
> > instance to used for a soecific LFP on a specific platform.
> >
> > Simplified pps_get_register() which use get_pps_idx() hook to derive the
> > pps instance and get_pps_idx() will be initialized at pps_init().
> >
> > Signed-off-by: Animesh Manna <animesh.manna at intel.com>
> > ---
> >  drivers/gpu/drm/i915/display/intel_bios.c     |  5 ++++
> >  drivers/gpu/drm/i915/display/intel_bios.h     |  1 +
> >  .../drm/i915/display/intel_display_types.h    |  2 ++
> >  drivers/gpu/drm/i915/display/intel_pps.c      | 25 ++++++++++++++-----
> >  4 files changed, 27 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/display/intel_bios.c b/drivers/gpu/drm/i915/display/intel_bios.c
> > index 51dde5bfd956..42315615a728 100644
> > --- a/drivers/gpu/drm/i915/display/intel_bios.c
> > +++ b/drivers/gpu/drm/i915/display/intel_bios.c
> > @@ -611,6 +611,11 @@ static int opregion_get_panel_type(struct drm_i915_private *i915,
> >  	return intel_opregion_get_panel_type(i915);
> >  }
> >  
> > +bool intel_bios_is_lfp2(struct intel_encoder *encoder)
> > +{
> > +	return encoder->devdata && encoder->devdata->child.handle == DEVICE_HANDLE_LFP2;
> > +}
> 
> AFAICS the encoder never really needs to know whether it's "lfp1" or
> "lfp2". It needs to know the pps controller number.
> 
> > +
> >  static int vbt_get_panel_type(struct drm_i915_private *i915,
> >  			      const struct intel_bios_encoder_data *devdata,
> >  			      const struct edid *edid)
> > diff --git a/drivers/gpu/drm/i915/display/intel_bios.h b/drivers/gpu/drm/i915/display/intel_bios.h
> > index e47582b0de0a..aea72a87ea2c 100644
> > --- a/drivers/gpu/drm/i915/display/intel_bios.h
> > +++ b/drivers/gpu/drm/i915/display/intel_bios.h
> > @@ -251,6 +251,7 @@ bool intel_bios_is_lspcon_present(const struct drm_i915_private *i915,
> >  				  enum port port);
> >  bool intel_bios_is_lane_reversal_needed(const struct drm_i915_private *i915,
> >  					enum port port);
> > +bool intel_bios_is_lfp2(struct intel_encoder *encoder);
> >  enum aux_ch intel_bios_port_aux_ch(struct drm_i915_private *dev_priv, enum port port);
> >  bool intel_bios_get_dsc_params(struct intel_encoder *encoder,
> >  			       struct intel_crtc_state *crtc_state,
> > diff --git a/drivers/gpu/drm/i915/display/intel_display_types.h b/drivers/gpu/drm/i915/display/intel_display_types.h
> > index 0da9b208d56e..95f71a572b07 100644
> > --- a/drivers/gpu/drm/i915/display/intel_display_types.h
> > +++ b/drivers/gpu/drm/i915/display/intel_display_types.h
> > @@ -1723,6 +1723,8 @@ struct intel_dp {
> >  
> >  	/* When we last wrote the OUI for eDP */
> >  	unsigned long last_oui_write;
> > +
> > +	int (*get_pps_idx)(struct intel_dp *intel_dp);
> 
> Making this a function pointer should be a separate step. Please don't
> try to do too many things at once. Rule of thumb, one change per patch.
> 
> Also, this should be placed near the other function pointer members in
> struct intel_dp.
> 
> >  };
> >  
> >  enum lspcon_vendor {
> > diff --git a/drivers/gpu/drm/i915/display/intel_pps.c b/drivers/gpu/drm/i915/display/intel_pps.c
> > index 1b21a341962f..c9cdb302d318 100644
> > --- a/drivers/gpu/drm/i915/display/intel_pps.c
> > +++ b/drivers/gpu/drm/i915/display/intel_pps.c
> > @@ -231,6 +231,17 @@ bxt_power_sequencer_idx(struct intel_dp *intel_dp)
> >  	return backlight_controller;
> >  }
> >  
> > +static int
> > +gen12_power_sequencer_idx(struct intel_dp *intel_dp)
> > +{
> > +	struct intel_encoder *encoder = &dp_to_dig_port(intel_dp)->base;
> > +
> > +	if (intel_bios_is_lfp2(encoder))
> > +		return 1;
> 
> This is actually not how this works. The bxt_power_sequencer_idx()
> matches bspec 20149: "PWM and PPS are assumed to be aligned to be from
> same block and not mismatch" i.e. the backlight controller id and the
> pps id are the same. There are no requirements to map lfp# and pps
> controller like this, even if it might be true in the common case.
> 
> The problem is we need the information *before* we call
> intel_bios_init_panel().
> 
> It's a catch-22. We need the pps id to read the panel EDID, and we need
> the panel EDID to choose the correct panel type in VBT, which we need to
> get the pps id from the VBT.
> 
> This is highlighted in [1]. The 2nd eDP (which is not even physically
> present, only in the VBT, *sigh*) screws up the PPS for the 1st during
> init.
> 
> I think Ville had some ideas here.

What a mess.

I guess for the panel_type!=255 cases we could just
initialize the panel specific stuff earlier.

The hardest case to solve would be dual panel with
panel_type==255. For that not sure we can much more than to
read out the state of each PPS from the hardware and try to
guess if one of the enabled ones corresponds to our current
panel, and then try to do the EDID read(s).

Or maybe we could just take a shortcut and reject dual
panel + panel_type=255 combos entirely. Or did we already
run into such cases?

-- 
Ville Syrjälä
Intel