[Intel-gfx] [PATCH v3 3/7] drm/i915: protect access to DP_TP_* on non-dp

[Ville Syrjälä]

On Wed, Sep 04, 2019 at 12:45:35AM +0000, Souza, Jose wrote:
> On Tue, 2019-09-03 at 10:16 -0700, Matt Roper wrote:
> > On Thu, Aug 29, 2019 at 01:37:55PM +0300, Ville Syrjälä wrote:
> > > On Thu, Aug 29, 2019 at 02:25:50AM -0700, Lucas De Marchi wrote:
> > > > DP_TP_{CTL,STATUS} should only be programmed when the encoder is
> > > > intel_dp.
> > > > Checking its current usages intel_disable_ddi_buf() is the only
> > > > offender, with other places being protected by checks like
> > > > pipe_config->fec_enable that is only set by intel_dp.
> > > > 
> > > > Cc: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > Signed-off-by: Lucas De Marchi <lucas.demarchi at intel.com>
> > > > ---
> > > >  drivers/gpu/drm/i915/display/intel_ddi.c | 10 ++++++----
> > > >  1 file changed, 6 insertions(+), 4 deletions(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > b/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > index 3180dacb5be4..df3e4fe7e3e9 100644
> > > > --- a/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > +++ b/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > @@ -3462,10 +3462,12 @@ static void intel_disable_ddi_buf(struct
> > > > intel_encoder *encoder,
> > > >  		wait = true;
> > > >  	}
> > > >  
> > > > -	val = I915_READ(DP_TP_CTL(port));
> > > > -	val &= ~(DP_TP_CTL_ENABLE | DP_TP_CTL_LINK_TRAIN_MASK);
> > > > -	val |= DP_TP_CTL_LINK_TRAIN_PAT1;
> > > > -	I915_WRITE(DP_TP_CTL(port), val);
> > > > +	if (intel_encoder_is_dp(encoder)) {
> > > 
> > > Doesn't really make sense to me. Either we just do it (because a
> > > DDI is
> > > just a DDI so DP_TP_CTL does exist always), or we only do it when
> > > driving
> > > DP and not when driving HDMI.
> > 
> > I agree; I don't think there's a need to avoid program programming
> > the
> > register just because we weren't previously in DP mode.
> 
> The problem of always programing DP_TP_CTL comes with TGL, when
> DP_TP_CTL() moves to transcoder, see next patch: drm/i915/tgl: move
> DP_TP_* to transcoder.
> 
> We are adding intel_dp->regs.dp_tp_ctl and initializing(this is
> necessary for MST for SST we could keep the current approach) it in DP
> paths, we could move it to intel_encoder or intel_digital_port and
> initialized it for HDMI too but it would not make any sense for someone
> reading HDMI sequences.

I'm not 100% sold on intel_dp->regs thing. Eventually I'd like to
remove all that crud from intel_dp and instead just plumb the crtc
state everywhere. But let's go with it for now.

Hmm. I just glanced at the modeset sequence in the spec, and it doesn't
touch DP_TP_CTL for HDMI, so I guess we might as well follow that same
sequence here and just go with intel_crtc_has_dp_encoder() instead
of intel_encoder_is_dp().

> 
> And to move this to a DP specific function would force us to create
> another function to execute the last "wait DDI_BUF_CTL to idle".
> 
> BSpec: 53339 and 22243
> 
> Personally I prefer this patch solution but let me know your thoughts
> after this explanation.
> 
> > 
> > But I do question whether a RMW is necessary; it seems like just
> > writing
> > a constant 0 to this register would be sufficient for the disable
> > sequence.
> > 
> > 
> > Matt
> > 
> > > For the latter I would perhaps suggest moving all this extra junk
> > > out
> > > from intel_disable_ddi_buf() into the DP specific code paths,
> > > leaving
> > > just the actual DDI_BUF_CTL disable here.
> > > 
> > > > +		val = I915_READ(DP_TP_CTL(port));
> > > > +		val &= ~(DP_TP_CTL_ENABLE | DP_TP_CTL_LINK_TRAIN_MASK);
> > > > +		val |= DP_TP_CTL_LINK_TRAIN_PAT1;
> > > > +		I915_WRITE(DP_TP_CTL(port), val);
> > > > +	}
> > > >  
> > > >  	/* Disable FEC in DP Sink */
> > > >  	intel_ddi_disable_fec_state(encoder, crtc_state);
> > > > -- 
> > > > 2.23.0
> > > 
> > > -- 
> > > Ville Syrjälä
> > > Intel

-- 
Ville Syrjälä
Intel