[Intel-gfx] [PATCH 2/2] drm/i915: Allow PCH DPLL sharing regardles of DPLL_SDVO_HIGH_SPEED

[Ville Syrjälä]

On Thu, Dec 03, 2015 at 12:37:20PM +0200, Ville Syrjälä wrote:
> On Thu, Dec 03, 2015 at 09:02:12AM +0100, Daniel Vetter wrote:
> > On Wed, Nov 25, 2015 at 04:35:31PM +0200, ville.syrjala at linux.intel.com wrote:
> > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > 
> > > DPLL_SDVO_HIGH_SPEED must be set for SDVO/HDMI/DP, but nowhere is it
> > > forbidden to set it for LVDS/CRT as well. So let's move it out of the
> > > ironlake_compute_dpll() and just do it on demand in the pll enable hook.
> > > This allows the PLL to be shared in more cases when dealing with
> > > different output types.
> > > 
> > > Note that we must now call the pll enable hook regarless of the current
> > > pll state, so that DPLL_SDVO_HIGH_SPEED gets updated appropriately.
> > > 
> > > FIXME: maybe better to add a separate hook for the "pll already enabled"
> > > case?
> > > 
> > > Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/intel_display.c | 29 +++++++++++++++++++++--------
> > >  1 file changed, 21 insertions(+), 8 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> > > index 26cafeea2845..ae58f1105458 100644
> > > --- a/drivers/gpu/drm/i915/intel_display.c
> > > +++ b/drivers/gpu/drm/i915/intel_display.c
> > > @@ -1908,6 +1908,8 @@ static void intel_enable_shared_dpll(struct intel_crtc *crtc)
> > >  	if (pll->active++) {
> > >  		WARN_ON(!pll->on);
> > >  		assert_shared_dpll_enabled(dev_priv, pll);
> > > +		/* to update high speed IO clock state */
> > > +		pll->enable(dev_priv, pll);
> > >  		return;
> > >  	}
> > >  	WARN_ON(pll->on);
> > > @@ -8942,11 +8944,6 @@ static uint32_t ironlake_compute_dpll(struct intel_crtc *intel_crtc,
> > >  	dpll |= (crtc_state->pixel_multiplier - 1)
> > >  		<< PLL_REF_SDVO_HDMI_MULTIPLIER_SHIFT;
> > >  
> > > -	if (is_sdvo)
> > > -		dpll |= DPLL_SDVO_HIGH_SPEED;
> > > -	if (crtc_state->has_dp_encoder)
> > > -		dpll |= DPLL_SDVO_HIGH_SPEED;
> > > -
> > >  	/* compute bitmask from p1 value */
> > >  	dpll |= (1 << (crtc_state->dpll.p1 - 1)) << DPLL_FPA01_P1_POST_DIV_SHIFT;
> > >  	/* also FPA1 */
> > > @@ -13626,7 +13623,7 @@ static bool ibx_pch_dpll_get_hw_state(struct drm_i915_private *dev_priv,
> > >  		return false;
> > >  
> > >  	val = I915_READ(PCH_DPLL(pll->id));
> > > -	hw_state->dpll = val;
> > > +	hw_state->dpll = val & ~DPLL_SDVO_HIGH_SPEED;
> > >  	hw_state->fp0 = I915_READ(PCH_FP0(pll->id));
> > >  	hw_state->fp1 = I915_READ(PCH_FP1(pll->id));
> > >  
> > > @@ -13643,10 +13640,26 @@ static void ibx_pch_dpll_mode_set(struct drm_i915_private *dev_priv,
> > >  static void ibx_pch_dpll_enable(struct drm_i915_private *dev_priv,
> > >  				struct intel_shared_dpll *pll)
> > >  {
> > > +	struct drm_device *dev = dev_priv->dev;
> > > +	struct intel_crtc *crtc;
> > > +	u32 dpll = pll->config.hw_state.dpll;
> > > +
> > > +	/* Configure high speed IO clock as needed */
> > > +	for_each_intel_crtc(dev, crtc) {
> > > +		if (intel_crtc_to_shared_dpll(crtc) == pll &&
> > > +		    crtc->config->base.active &&
> > > +		    (crtc->config->has_dp_encoder ||
> > > +		     intel_pipe_has_type(crtc, INTEL_OUTPUT_SDVO) ||
> > > +		     intel_pipe_has_type(crtc, INTEL_OUTPUT_HDMI))) {
> > > +			dpll |= DPLL_SDVO_HIGH_SPEED;
> > > +			break;
> > > +		}
> > > +	}
> > 
> > This is a bit ugly. What about adding a needs_5x_clock to struct
> > intel_shared_dpll_config (i.e. outside of the hw pll state to avoid
> > unsharing and pipe checker warnings)? ilk_compute_dpll would set that, and
> > this function here obeys.
> > 
> > Clearing would only happen when we completely disable a pll after the last
> > user is gone.
> 
> I don't really like potentially wasting power like that. I suppose it's
> a miniscule amount most likely, but still feels a bit wrong.

Hmm. I wonder if I could make it a bitmask of some sort...

-- 
Ville Syrjälä
Intel OTC