[Intel-gfx] [PATCH] drm/i915/dp: Fix the t11_t12 delay for non GEN 9 LP platforms at HW readout and HW write

[Manasi Navare]

On Wed, Jun 21, 2017 at 11:03:58PM +0300, Ville Syrjälä wrote:
> On Wed, Jun 21, 2017 at 12:37:43PM -0700, Manasi Navare wrote:
> > According to the eDP spec the minimum value for panel power cycle delay
> > (t11_t12) is 500ms and as per the Bspec, PP_DIVISOR panel power cycle
> > delay field should be programmed to "+1" value. Eg: To have a delay
> > of 500ms this should be programmed to 6. This patch fixes the write
> > by adding +1 to the pp_div value so it programs the correct min
> > required panel power cycle delay.
> > Since we program it to +1 value, when we perform HW readout, this
> > value should subtract 1 before verifying pps state. This patch makes
> > this correction as well to avoid "PPS state mismatch" error.
> > This patch also adds a case where if the readout is 0 for the first readout,
> > then read it as 0, dont subtract.
> > 
> > Signed-off-by: Manasi Navare <manasi.d.navare at intel.com>
> > Cc: Ville Syrjala <ville.syrjala at linux.intel.com>
> > Cc: Clint Taylor <Clinton.A.Taylor at intel.com>
> > ---
> >  drivers/gpu/drm/i915/intel_dp.c | 22 +++++++++++-----------
> >  1 file changed, 11 insertions(+), 11 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
> > index bca4ac1..089e373 100644
> > --- a/drivers/gpu/drm/i915/intel_dp.c
> > +++ b/drivers/gpu/drm/i915/intel_dp.c
> > @@ -5149,6 +5149,7 @@ intel_pps_readout_hw_state(struct drm_i915_private *dev_priv,
> >  			   struct intel_dp *intel_dp, struct edp_power_seq *seq)
> >  {
> >  	u32 pp_on, pp_off, pp_div = 0, pp_ctl = 0;
> > +	u16 pp_cycle_delay = 0;
> >  	struct pps_registers regs;
> >  
> >  	intel_pps_get_registers(dev_priv, intel_dp, &regs);
> > @@ -5177,17 +5178,16 @@ intel_pps_readout_hw_state(struct drm_i915_private *dev_priv,
> >  	seq->t10 = (pp_off & PANEL_POWER_DOWN_DELAY_MASK) >>
> >  		   PANEL_POWER_DOWN_DELAY_SHIFT;
> >  
> > -	if (IS_GEN9_LP(dev_priv) || HAS_PCH_CNP(dev_priv)) {
> > -		u16 tmp = (pp_ctl & BXT_POWER_CYCLE_DELAY_MASK) >>
> > +	if (IS_GEN9_LP(dev_priv) || HAS_PCH_CNP(dev_priv))
> > +		pp_cycle_delay =  (pp_ctl & BXT_POWER_CYCLE_DELAY_MASK) >>
> >  			BXT_POWER_CYCLE_DELAY_SHIFT;
> > -		if (tmp > 0)
> > -			seq->t11_t12 = (tmp - 1) * 1000;
> > -		else
> > -			seq->t11_t12 = 0;
> > -	} else {
> > -		seq->t11_t12 = ((pp_div & PANEL_POWER_CYCLE_DELAY_MASK) >>
> > -		       PANEL_POWER_CYCLE_DELAY_SHIFT) * 1000;
> > -	}
> > +	else
> > +		pp_cycle_delay = (pp_div & PANEL_POWER_CYCLE_DELAY_MASK) >>
> > +			PANEL_POWER_CYCLE_DELAY_SHIFT;
> > +	if (pp_cycle_delay > 0)
> > +		seq->t11_t12 = (pp_cycle_delay - 1) * 1000;
> > +	else
> > +		seq->t11_t12 = 0;
> 
> I think it's probably easier to go the other way and just add the +100
> msec to the vbt delay, and nuke the BXT/CNP special casing in the code.
>

The reason I am doing the -1 here is that this hw_readout gets called
in intel_dp_pps_verify each time during edp_panel_on and it reads the
values written into the register so lets say we wrote 6 into the register for
500ms then it will read 6 so we need to subtract 1 and multiply by 1000 to actually
get 5000 that gets written into intel->pps_delays. 

Also keeping it at 5000 makes more sense because thats the number in edp spec.
rather than (adding 100ms to 500) * 1000 so storing 6000.

Manasi

> >  }
> >  
> >  static void
> > @@ -5341,7 +5341,7 @@ intel_dp_init_panel_power_sequencer_registers(struct drm_device *dev,
> >  				<< BXT_POWER_CYCLE_DELAY_SHIFT);
> >  	} else {
> >  		pp_div = ((100 * div)/2 - 1) << PP_REFERENCE_DIVIDER_SHIFT;
> > -		pp_div |= (DIV_ROUND_UP(seq->t11_t12, 1000)
> > +		pp_div |= (DIV_ROUND_UP(seq->t11_t12 + 1, 1000)
> >  				<< PANEL_POWER_CYCLE_DELAY_SHIFT);
> >  	}
> >  
> > -- 
> > 2.1.4
> 
> -- 
> Ville Syrjälä
> Intel OTC