Re: [PATCH v2 2/2] drm/i915: Merge of visible and !visible paths for primary planes

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Wed, Sep 10, 2014 at 09:25:29PM +0300, Ville Syrjälä wrote:
> On Wed, Sep 10, 2014 at 12:04:18PM -0300, Gustavo Padovan wrote:
> > From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> > 
> > Fold intel_pipe_set_base() in the update primary plane path merging
> > pieces of code that are common to both paths.
> > 
> > Basically the the pin/unpin procedures are the same for both paths
> > and some checks can also be shared (some of the were moved to the
> > check() stage)
> > 
> > v2: take Ville's comments:
> > 	- remove unecessary plane check
> > 	- move mutex lock to inside the conditional
> > 	- make the pin fail message a debug one
> > 	- add a fixme for the fastboot hack
> > 	- call intel_frontbuffer_flip() after FBC update
> > 
> > Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
> > ---
> >  drivers/gpu/drm/i915/intel_display.c | 93 ++++++++++++++++++++++++------------
> >  1 file changed, 62 insertions(+), 31 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> > index b78f00a..1a001bf 100644
> > --- a/drivers/gpu/drm/i915/intel_display.c
> > +++ b/drivers/gpu/drm/i915/intel_display.c
> > @@ -11824,12 +11824,23 @@ intel_check_primary_plane(struct drm_plane *plane,
> >  	struct drm_rect *dest = &state->dst;
> >  	struct drm_rect *src = &state->src;
> >  	const struct drm_rect *clip = &state->clip;
> > +	int ret;
> >  
> > -	return drm_plane_helper_check_update(plane, crtc, fb,
> > +	ret = drm_plane_helper_check_update(plane, crtc, fb,
> >  					    src, dest, clip,
> >  					    DRM_PLANE_HELPER_NO_SCALING,
> >  					    DRM_PLANE_HELPER_NO_SCALING,
> >  					    false, true, &state->visible);
> > +	if (ret)
> > +		return ret;
> > +
> > +	/* no fb bound */
> > +	if (state->visible && !fb) {
> > +		DRM_ERROR("No FB bound\n");
> > +		return -EINVAL;
> > +	}
> > +
> > +	return 0;
> >  }
> >  
> >  static int
> > @@ -11841,14 +11852,34 @@ intel_commit_primary_plane(struct drm_plane *plane,
> >  	struct drm_device *dev = crtc->dev;
> >  	struct drm_i915_private *dev_priv = dev->dev_private;
> >  	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
> > +	enum pipe pipe = intel_crtc->pipe;
> > +	struct drm_framebuffer *old_fb = plane->fb;
> >  	struct drm_i915_gem_object *obj = intel_fb_obj(fb);
> >  	struct drm_i915_gem_object *old_obj = intel_fb_obj(plane->fb);
> >  	struct intel_plane *intel_plane = to_intel_plane(plane);
> >  	struct drm_rect *src = &state->src;
> > -	int ret;
> > +	int ret = 0;
> >  
> >  	intel_crtc_wait_for_pending_flips(crtc);
> >  
> > +	if (intel_crtc_has_pending_flip(crtc)) {
> > +		DRM_ERROR("pipe is still busy with an old pageflip\n");
> > +		return -EBUSY;
> > +	}
> > +
> > +	if (plane->fb != fb) {
> > +		mutex_lock(&dev->struct_mutex);
> > +		ret = intel_pin_and_fence_fb_obj(dev, obj, NULL);
> > +		if (ret == 0)
> > +			i915_gem_track_fb(old_obj, obj,
> > +					  INTEL_FRONTBUFFER_PRIMARY(pipe));
> > +		mutex_unlock(&dev->struct_mutex);
> > +	}
> > +	if (ret != 0) {
> > +		DRM_DEBUG_KMS("pin & fence failed\n");
> > +		return ret;
> > +	}
> 
> I'd move this error handling also inside the 'plane->fb != fb' block.
> That avoids the ret=0 initialization, and more importantly it will
> match what you did in the sprite code. Thanks to your efforts I'm 
> already dreaming about sharing buffer pinning codes across
> different plane types...
> 
> > +
> >  	/*
> >  	 * If clipping results in a non-visible primary plane, we'll disable
> >  	 * the primary plane.  Note that this is a bit different than what
> > @@ -11856,33 +11887,9 @@ intel_commit_primary_plane(struct drm_plane *plane,
> >  	 * because plane->fb still gets set and pinned.
> >  	 */
> >  	if (!state->visible) {
> > -		mutex_lock(&dev->struct_mutex);
> > -
> > -		/*
> > -		 * Try to pin the new fb first so that we can bail out if we
> > -		 * fail.
> > -		 */
> > -		if (plane->fb != fb) {
> > -			ret = intel_pin_and_fence_fb_obj(dev, obj, NULL);
> > -			if (ret) {
> > -				mutex_unlock(&dev->struct_mutex);
> > -				return ret;
> > -			}
> > -		}
> > -
> > -		i915_gem_track_fb(old_obj, obj,
> > -				  INTEL_FRONTBUFFER_PRIMARY(intel_crtc->pipe));
> > -
> >  		if (intel_crtc->primary_enabled)
> >  			intel_disable_primary_hw_plane(plane, crtc);
> >  
> > -
> > -		if (plane->fb != fb)
> > -			if (plane->fb)
> > -				intel_unpin_fb_obj(old_obj);
> > -
> > -		mutex_unlock(&dev->struct_mutex);
> > -
> >  	} else {
> >  		if (intel_crtc && intel_crtc->active &&
> >  		    intel_crtc->primary_enabled) {
> > @@ -11902,12 +11909,36 @@ intel_commit_primary_plane(struct drm_plane *plane,
> >  				intel_disable_fbc(dev);
> >  			}
> >  		}
> > -		ret = intel_pipe_set_base(crtc, src->x1, src->y1, fb);
> > -		if (ret)
> > -			return ret;
> >  
> > -		if (!intel_crtc->primary_enabled)
> > -			intel_enable_primary_hw_plane(plane, crtc);
> > +		/* FIXME: kill this fastboot hack */
> > +		intel_update_pipe_size(intel_crtc);
> > +
> > +		intel_crtc->primary_enabled = true;
> > +
> > +		dev_priv->display.update_primary_plane(crtc, fb, src->x1, src->y1);
> > +
> > +		if (intel_crtc->active)
> > +			intel_frontbuffer_flip(dev,
> > +					       INTEL_FRONTBUFFER_PRIMARY(pipe));
> > +
> > +		mutex_lock(&dev->struct_mutex);
> > +		intel_update_fbc(dev);
> > +		mutex_unlock(&dev->struct_mutex);
> 
> This is still done too early. We need primary->fb to be set before since
> intel_update_fbc() will consult it to determine if it can now enable fbc.
> 
> Hmm. Actaully I think we may want to reorganize this code a bit more so
> that it more closely resembles the sprite code. So I'm thinking
> something like this:
> 
> crtc->primary->fb = fb;
> crtc->x = ...;
> crtc->y = ...;
> 
> intel_plane->crtc_x = ...;
> intel_plane->crtc_y = ...;
> ...
> 
> if (crtc->active) {
> 	if (fbc.plane == plane && ...) {
> 		disable_fbc();
> 
> 	if (visible)
> 		update_pipe_size();
> 
> 	primary_enabled = visible;
> 	.update_primary_plane()

Actually better make that
if (visible) {
	update_pipe_size();
	intel_enable_primary_hw_plane();
} else {
	intel_disable_primary_hw_plane();
}

for now so that we we don't have to duplicate the nasty BDW vblank wait
workaround when enabling the primary plane.

The next step might be to kill off intel_{enable,disable}_primary_hw_plane()
and replace them with intel_primary_plane_setplane() as well, and then we
have to move the vblank hack into intel_primary_plane_setplane() itself.
But let's do baby steps and get this intel_pipe_set_base() killing
finished before we continue too far ahead. Also the sprite plane
covering primary plane case probably needs a bit more thought before
we can do this.

> 
> 	intel_frontbuffer_flip()
> 
> 	update_fbc();
> }
> 
> 
> > +	}
> > +
> > +	crtc->primary->fb = fb;
> > +	crtc->x = src->x1;
> > +	crtc->y = src->y1;
> > +
> > +	if (old_fb) {
> > +		if (intel_crtc->active && old_fb != fb)
> > +			intel_wait_for_vblank(dev, intel_crtc->pipe);
> > +
> > +		if (old_fb != fb) {
> > +			mutex_lock(&dev->struct_mutex);
> > +			intel_unpin_fb_obj(old_obj);
> > +			mutex_unlock(&dev->struct_mutex);
> > +		}
> >  	}
> >  
> >  	intel_plane->crtc_x = state->orig_dst.x1;
> > -- 
> > 1.9.3
> > 
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx@lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
> 
> -- 
> Ville Syrjälä
> Intel OTC
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Ville Syrjälä
Intel OTC