libcamera: pipeline: rkisp1: Queue parameters even if they are not ready

If the IPA has not filled in the parameters buffer still queue it to
hardware. Not queuing the buffer results in the pipeline and hardware
going out of sync.

This is not a permanent fix of the problem and a todo is added to fix it
properly. This change does not make the situation worse as the state of
the pipeline is just as unknown as if no param buffer is queued as if one
with old content in it.

Signed-off-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>

1 files changed, 14 insertions, 4 deletions

diff --git a/src/libcamera/pipeline/rkisp1/rkisp1.cpp b/src/libcamera/pipeline/rkisp1/rkisp1.cpp
index 2f909cef..3d37677f 100644
--- a/src/libcamera/pipeline/rkisp1/rkisp1.cpp
+++ b/src/libcamera/pipeline/rkisp1/rkisp1.cpp
@@ -351,13 +351,23 @@ protected:
 		if (!info)
 			LOG(RkISP1, Fatal) << "Frame not known";
 
-		if (info->paramFilled)
-			pipe_->param_->queueBuffer(info->paramBuffer);
-		else
+		/*
+		 * \todo: If parameters are not filled a better method to handle
+		 * the situation than queuing a buffer with unknown content
+		 * should be used.
+		 *
+		 * It seems excessive to keep an internal zeroed scratch
+		 * parameters buffer around as this should not happen unless the
+		 * devices is under too much load. Perhaps failing the request
+		 * and returning it to the application with an error code is
+		 * better than queue it to hardware?
+		 */
+		if (!info->paramFilled)
 			LOG(RkISP1, Error)
 				<< "Parameters not ready on time for frame "
-				<< frame() << ", ignore parameters.";
+				<< frame();
 
+		pipe_->param_->queueBuffer(info->paramBuffer);
 		pipe_->stat_->queueBuffer(info->statBuffer);
 		pipe_->video_->queueBuffer(info->videoBuffer);
 	}