Re: [PATCH] drm/i915/display: C20 clock state verification

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

On Fri, Dec 15, 2023 at 11:10:52AM +0200, Ville Syrjälä wrote:
> On Fri, Dec 15, 2023 at 10:53:31AM +0200, Imre Deak wrote:
> > On Fri, Dec 15, 2023 at 10:00:57AM +0200, Mika Kahola wrote:
> > > Add clock state verification for C20. Since we
> > > are usign either A or B contexts, which are
> > > selected based on clock rate, we first need to
> > > calculate hw clock and use that clock to select
> > > which context we are using.
> > 
> > Could the description be clearer that the patch _fixes_ the context
> > selection? (Also the subject line should always say _what_ the patch
> > does.)
> > 
> > > 
> > > Signed-off-by: Mika Kahola <mika.kahola@intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/display/intel_cx0_phy.c | 8 +++++++-
> > >  1 file changed, 7 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/display/intel_cx0_phy.c b/drivers/gpu/drm/i915/display/intel_cx0_phy.c
> > > index 775c1c4a8978..6757e9f941e4 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_cx0_phy.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_cx0_phy.c
> > > @@ -3079,8 +3079,9 @@ static void intel_c20pll_state_verify(const struct intel_crtc_state *state,
> > >  	const struct intel_c20pll_state *mpll_sw_state = &state->cx0pll_state.c20;
> > >  	bool use_mplla;
> > >  	int i;
> > > +	int hw_clock = intel_c20pll_calc_port_clock(encoder, mpll_hw_state);
> > >  
> > > -	use_mplla = intel_c20_use_mplla(mpll_hw_state->clock);
> > > +	use_mplla = intel_c20_use_mplla(hw_clock);
> > 
> > It's mpll_hw_state->tx[0] C20_PHY_USE_MPLLB which tells the HW which
> > context to use, so I think it's better to use the same condition here.
> 
> Yes, one should never assume anything about how the register values
> were calculated/etc. That would pretty much defeat the whole purpose
> of doing readout and state check.
> 
> BTW I don't see even being set anywhere C20_PHY_USE_MPLLB.
> Do we not use it or is it encoded in that ugly hex soup in
> intel_c20_compute_hdmi_tmds_pll()? 

Yes, the pll_state->tx[0] line there. For DP, it's in the
intel_c20pll_state predefined tables.

> Instead of repeating the mistakes of the VLV PHY code someone should
> convert that into human readable form...
> 
> > 
> > >  	if (use_mplla) {
> > >  		for (i = 0; i < ARRAY_SIZE(mpll_sw_state->mplla); i++) {
> > >  			I915_STATE_WARN(i915, mpll_hw_state->mplla[i] != mpll_sw_state->mplla[i],
> > > @@ -3110,6 +3111,11 @@ static void intel_c20pll_state_verify(const struct intel_crtc_state *state,
> > >  				crtc->base.base.id, crtc->base.name, i,
> > >  				mpll_sw_state->cmn[i], mpll_hw_state->cmn[i]);
> > >  	}
> > > +
> > > +	I915_STATE_WARN(i915, hw_clock != mpll_sw_state->clock,
> > > +			"[CRTC:%d:%s] mismatch in C20: Register CLOCK (expected %d, found %d)",
> > > +			crtc->base.base.id, crtc->base.name,
> > > +			mpll_sw_state->clock, hw_clock);
> > 
> > I think the intel_crtc_state::port_clock SW/HW state verification is
> > equivalent to the above, but it's good to verify it here as well. I
> > would store hw_clock to mpll_hw_state->clock in
> > intel_c20pll_readout_hw_state() though.
> > 
> > >  }
> > >  
> > >  void intel_cx0pll_state_verify(struct intel_atomic_state *state,
> > > -- 
> > > 2.34.1
> > > 
> 
> -- 
> Ville Syrjälä
> Intel