[PATCH 2/6] drm/vc4: Move LBM creation out of vc4_plane_mode_set()

[Boris Brezillon]

On Thu, 15 Nov 2018 12:39:42 -0800
Eric Anholt <eric at anholt.net> wrote:

> Boris Brezillon <boris.brezillon at bootlin.com> writes:
> 
> > We are about to use vc4_plane_mode_set() in the async check path, and
> > async updates require that LBM size stay the same since they reuse the
> > LBM from the previous state. So we definitely don't want to allocate a
> > new LBM region that we know for sure will be free right away.
> >
> > Move the LBM allocation out of vc4_plane_mode_set() and call the new
> > function (vc4_plane_update_lbm()) from vc4_plane_atomic_check().
> >
> > Signed-off-by: Boris Brezillon <boris.brezillon at bootlin.com>
> > ---
> >  drivers/gpu/drm/vc4/vc4_drv.h   |  1 +
> >  drivers/gpu/drm/vc4/vc4_plane.c | 82 ++++++++++++++++++++++-----------
> >  2 files changed, 55 insertions(+), 28 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h
> > index bd6ef1f31822..9ed05fb61eb6 100644
> > --- a/drivers/gpu/drm/vc4/vc4_drv.h
> > +++ b/drivers/gpu/drm/vc4/vc4_drv.h
> > @@ -338,6 +338,7 @@ struct vc4_plane_state {
> >  	u32 pos0_offset;
> >  	u32 pos2_offset;
> >  	u32 ptr0_offset;
> > +	u32 lbm_offset;
> >  
> >  	/* Offset where the plane's dlist was last stored in the
> >  	 * hardware at vc4_crtc_atomic_flush() time.
> > diff --git a/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c
> > index f6e3e8d33115..392a51f2bf8f 100644
> > --- a/drivers/gpu/drm/vc4/vc4_plane.c
> > +++ b/drivers/gpu/drm/vc4/vc4_plane.c
> > @@ -452,6 +452,43 @@ static void vc4_write_scaling_parameters(struct drm_plane_state *state,
> >  	}
> >  }
> >  
> > +static int vc4_plane_update_lbm(struct drm_plane_state *state)
> > +{
> > +	struct vc4_dev *vc4 = to_vc4_dev(state->plane->dev);
> > +	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
> > +	unsigned long irqflags;
> > +	u32 lbm_size;
> > +
> > +	lbm_size = vc4_lbm_size(state);
> > +	if (!lbm_size)
> > +		return 0;
> > +
> > +	if (WARN_ON(!vc4_state->lbm_offset))
> > +		return -EINVAL;
> > +
> > +	/* Allocate the LBM memory that the HVS will use for temporary
> > +	 * storage due to our scaling/format conversion.
> > +	 */
> > +	if (!vc4_state->lbm.allocated) {
> > +		int ret;
> > +
> > +		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
> > +		ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
> > +						 &vc4_state->lbm,
> > +						 lbm_size, 32, 0, 0);
> > +		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
> > +
> > +		if (ret)
> > +			return ret;
> > +	} else {
> > +		WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
> > +	}
> > +
> > +	vc4_state->dlist[vc4_state->lbm_offset] = vc4_state->lbm.start;
> > +
> > +	return 0;
> > +}
> > +
> >  /* Writes out a full display list for an active plane to the plane's
> >   * private dlist state.
> >   */
> > @@ -469,31 +506,12 @@ static int vc4_plane_mode_set(struct drm_plane *plane,
> >  	bool mix_plane_alpha;
> >  	bool covers_screen;
> >  	u32 scl0, scl1, pitch0;
> > -	u32 lbm_size, tiling;
> > -	unsigned long irqflags;
> > +	u32 tiling;
> >  	u32 hvs_format = format->hvs;
> >  	int ret, i;
> >  
> > -	ret = vc4_plane_setup_clipping_and_scaling(state);
> > -	if (ret)
> > -		return ret;
> > -
> > -	/* Allocate the LBM memory that the HVS will use for temporary
> > -	 * storage due to our scaling/format conversion.
> > -	 */
> > -	lbm_size = vc4_lbm_size(state);
> > -	if (lbm_size) {
> > -		if (!vc4_state->lbm.allocated) {
> > -			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
> > -			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
> > -							 &vc4_state->lbm,
> > -							 lbm_size, 32, 0, 0);
> > -			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
> > -		} else {
> > -			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
> > -		}
> > -	}
> >  
> > +	ret = vc4_plane_setup_clipping_and_scaling(state);
> >  	if (ret)
> >  		return ret;
> >  
> > @@ -717,15 +735,18 @@ static int vc4_plane_mode_set(struct drm_plane *plane,
> >  		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
> >  	}
> >  
> > +	vc4_state->lbm_offset = 0;
> > +
> >  	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
> >  	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
> >  	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
> >  	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
> > -		/* LBM Base Address. */
> > +		/* Reserve a slot for the LBM Base Address. The real value will
> > +		 * be set when calling vc4_plane_update_lbm().
> > +		 */
> >  		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
> > -		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
> > -			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
> > -		}
> > +		    vc4_state->y_scaling[1] != VC4_SCALING_NONE)
> > +			vc4_state->lbm_offset = vc4_state->dlist_count++;
> >  
> >  		if (num_planes > 1) {
> >  			/* Emit Cb/Cr as channel 0 and Y as channel
> > @@ -785,13 +806,18 @@ static int vc4_plane_atomic_check(struct drm_plane *plane,
> >  				  struct drm_plane_state *state)
> >  {
> >  	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
> > +	int ret;
> >  
> >  	vc4_state->dlist_count = 0;
> >  
> > -	if (plane_enabled(state))
> > -		return vc4_plane_mode_set(plane, state);
> > -	else
> > +	if (!plane_enabled(state))
> >  		return 0;
> > +
> > +	ret = vc4_plane_mode_set(plane, state);
> > +	if (ret)
> > +		return ret;
> > +
> > +	return vc4_plane_update_lbm(state);
> >  }  
> 
> Couldn't you call update_lbm() (maybe called allocate_lbm()) first, and
> then leave the dlist setup as is without needing lbm_offset?

I'll anyway need lbm_offset for the dlist check I do in async_check().