Re: [PATCH 08/11] drm/msm/disp/dpu1: Add support for DSC in encoder

[Vinod Koul <vkoul@kernel.org>]

On 29-07-21, 23:54, Dmitry Baryshkov wrote:
> On 15/07/2021 09:52, Vinod Koul wrote:

> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
> > index 8d942052db8a..41140b781e66 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
> > @@ -21,12 +21,17 @@
> >   #include "dpu_hw_intf.h"
> >   #include "dpu_hw_ctl.h"
> >   #include "dpu_hw_dspp.h"
> > +#include "dpu_hw_dsc.h"
> >   #include "dpu_formats.h"
> >   #include "dpu_encoder_phys.h"
> >   #include "dpu_crtc.h"
> >   #include "dpu_trace.h"
> >   #include "dpu_core_irq.h"
> > +#define DSC_MODE_SPLIT_PANEL		BIT(0)
> > +#define DSC_MODE_MULTIPLEX		BIT(1)
> > +#define DSC_MODE_VIDEO			BIT(2)
> 
> This should go into dpu_hw_dsc.h. Ah. They are already defined there and
> just redefined there. Remove the defines here.

Sure, updated

> It might be cleaner to add bool flags to struct msm_display_dsc_config and
> then calculate common mode in the dpu_hw_dsc_config().

How would that be better than calculating here? I dont see much of an
advantage.

> > +static void dpu_encoder_dsc_pipe_cfg(struct dpu_hw_dsc *hw_dsc,
> > +				     struct dpu_hw_pingpong *hw_pp,
> > +				     struct msm_display_dsc_config *dsc,
> > +				     u32 common_mode)
> > +{
> > +	if (hw_dsc->ops.dsc_config)
> > +		hw_dsc->ops.dsc_config(hw_dsc, dsc, common_mode);
> > +
> > +	if (hw_dsc->ops.dsc_config_thresh)
> > +		hw_dsc->ops.dsc_config_thresh(hw_dsc, dsc);
> > +
> > +	if (hw_pp->ops.setup_dsc)
> > +		hw_pp->ops.setup_dsc(hw_pp);
> > +
> > +	if (hw_pp->ops.enable_dsc)
> > +		hw_pp->ops.enable_dsc(hw_pp);
> 
> I think, we do not need to split these operations, I'd suggest having just
> hw_dsc->ops.dsc_config() and hw_pp->ops.enable_dsc(), merging
> dsc_config_thres() and setup_dsc() into respective methods.

Merging hw_dsc->ops.dsc_config() and hw_dsc->ops.dsc_config_thresh() would make
it from L to XL size, so lets keep them split.

We could merge the small hw_pp->ops.setup_dsc() and
hw_pp->ops.enable_dsc() though.

> >   void dpu_encoder_prepare_for_kickoff(struct drm_encoder *drm_enc)
> >   {
> >   	struct dpu_encoder_virt *dpu_enc;
> >   	struct dpu_encoder_phys *phys;
> > +	struct msm_drm_private *priv;
> >   	bool needs_hw_reset = false;
> >   	unsigned int i;
> > @@ -1841,6 +1977,10 @@ void dpu_encoder_prepare_for_kickoff(struct drm_encoder *drm_enc)
> >   			dpu_encoder_helper_hw_reset(dpu_enc->phys_encs[i]);
> >   		}
> >   	}
> > +
> > +	priv = drm_enc->dev->dev_private;
> > +	if (priv->dsc)
> > +		dpu_encoder_prep_dsc(dpu_enc, priv->dsc);
> 
> Not quite. This makes dsc config global, while we can have several encoders
> enabled at once (think of DSI + DP). So the dsc should be a per-encoder
> setting rather than global.

I agree it would make sense to have per-encoder. The DP part needs to be
comprehended for DSC and would need more changes. I think updating this
for DP then and making it generic as required for DP would be better,
right? In that case I will skip moving to encoder for now.

-- 
~Vinod