media: cedrus: h265: Support decoding 10-bit frames

10-bit frames needs extra buffer space when 8-bit capture format is
used. Use previously prepared infrastructure to adjust buffer size.
Signed-off-by: Jernej Skrabec <jernej.skrabec@gmail.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>

drivers/staging/media/sunxi/cedrus/cedrus.c

@@ -45,6 +45,8 @@ static int cedrus_try_ctrl(struct v4l2_ctrl *ctrl)
 	} else if (ctrl->id == V4L2_CID_STATELESS_HEVC_SPS) {
 		const struct v4l2_ctrl_hevc_sps *sps = ctrl->p_new.p_hevc_sps;
 		struct cedrus_ctx *ctx = container_of(ctrl->handler, struct cedrus_ctx, hdl);
+		unsigned int bit_depth;
+		struct vb2_queue *vq;
 
 		if (sps->chroma_format_idc != 1)
 			/* Only 4:2:0 is supported */
@@ -63,6 +65,24 @@ static int cedrus_try_ctrl(struct v4l2_ctrl *ctrl)
 				/* Only 8-bit is supported */
 				return -EINVAL;
 		}
+
+		bit_depth = max(sps->bit_depth_luma_minus8,
+				sps->bit_depth_chroma_minus8) + 8;
+
+		vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx,
+				     V4L2_BUF_TYPE_VIDEO_CAPTURE);
+
+		/*
+		 * Bit depth can't be higher than currently set once
+		 * buffers are allocated.
+		 */
+		if (vb2_is_busy(vq)) {
+			if (ctx->bit_depth < bit_depth)
+				return -EINVAL;
+		} else {
+			ctx->bit_depth = bit_depth;
+			cedrus_reset_cap_format(ctx);
+		}
 	}
 
 	return 0;
@@ -354,6 +374,7 @@ static int cedrus_open(struct file *file)
 	v4l2_fh_init(&ctx->fh, video_devdata(file));
 	file->private_data = &ctx->fh;
 	ctx->dev = dev;
+	ctx->bit_depth = 8;
 
 	ctx->fh.m2m_ctx = v4l2_m2m_ctx_init(dev->m2m_dev, ctx,
 					    &cedrus_queue_init);

drivers/staging/media/sunxi/cedrus/cedrus.h

@@ -119,6 +119,7 @@ struct cedrus_ctx {
 	struct v4l2_pix_format		src_fmt;
 	struct v4l2_pix_format		dst_fmt;
 	struct cedrus_dec_ops		*current_codec;
+	unsigned int			bit_depth;
 
 	struct v4l2_ctrl_handler	hdl;
 	struct v4l2_ctrl		**ctrls;

drivers/staging/media/sunxi/cedrus/cedrus_h265.c

@@ -41,6 +41,19 @@ struct cedrus_h265_sram_pred_weight {
 	__s8	offset;
 } __packed;
 
+static unsigned int cedrus_h265_2bit_size(unsigned int width,
+					  unsigned int height)
+{
+	/*
+	 * Vendor library additionally aligns width and height to 16,
+	 * but all capture formats are already aligned to that anyway,
+	 * so we can skip that here. All formats are also one form of
+	 * YUV 4:2:0 or another, so we can safely assume multiplication
+	 * factor of 1.5.
+	 */
+	return ALIGN(width / 4, 32) * height * 3 / 2;
+}
+
 static enum cedrus_irq_status cedrus_h265_irq_status(struct cedrus_ctx *ctx)
 {
 	struct cedrus_dev *dev = ctx->dev;
@@ -802,6 +815,18 @@ static int cedrus_h265_setup(struct cedrus_ctx *ctx, struct cedrus_run *run)
 						      VE_DEC_H265_SRAM_OFFSET_PRED_WEIGHT_CHROMA_L1);
 	}
 
+	if (ctx->bit_depth > 8) {
+		unsigned int stride = ALIGN(ctx->dst_fmt.width / 4, 32);
+
+		reg = ctx->dst_fmt.sizeimage -
+		      cedrus_h265_2bit_size(ctx->dst_fmt.width,
+					    ctx->dst_fmt.height);
+		cedrus_write(dev, VE_DEC_H265_OFFSET_ADDR_FIRST_OUT, reg);
+
+		reg = VE_DEC_H265_10BIT_CONFIGURE_FIRST_2BIT_STRIDE(stride);
+		cedrus_write(dev, VE_DEC_H265_10BIT_CONFIGURE, reg);
+	}
+
 	/* Enable appropriate interruptions. */
 	cedrus_write(dev, VE_DEC_H265_CTRL, VE_DEC_H265_CTRL_IRQ_MASK);
 
@@ -874,6 +899,15 @@ static void cedrus_h265_trigger(struct cedrus_ctx *ctx)
 	cedrus_write(dev, VE_DEC_H265_TRIGGER, VE_DEC_H265_TRIGGER_DEC_SLICE);
 }
 
+static unsigned int cedrus_h265_extra_cap_size(struct cedrus_ctx *ctx,
+					       struct v4l2_pix_format *pix_fmt)
+{
+	if (ctx->bit_depth > 8)
+		return cedrus_h265_2bit_size(pix_fmt->width, pix_fmt->height);
+
+	return 0;
+}
+
 struct cedrus_dec_ops cedrus_dec_ops_h265 = {
 	.irq_clear	= cedrus_h265_irq_clear,
 	.irq_disable	= cedrus_h265_irq_disable,
@@ -882,4 +916,5 @@ struct cedrus_dec_ops cedrus_dec_ops_h265 = {
 	.start		= cedrus_h265_start,
 	.stop		= cedrus_h265_stop,
 	.trigger	= cedrus_h265_trigger,
+	.extra_cap_size	= cedrus_h265_extra_cap_size,
 };

drivers/staging/media/sunxi/cedrus/cedrus_regs.h

@@ -498,6 +498,22 @@
 
 #define VE_DEC_H265_LOW_ADDR			(VE_ENGINE_DEC_H265 + 0x80)
 
+#define VE_DEC_H265_OFFSET_ADDR_FIRST_OUT	(VE_ENGINE_DEC_H265 + 0x84)
+#define VE_DEC_H265_OFFSET_ADDR_SECOND_OUT	(VE_ENGINE_DEC_H265 + 0x88)
+
+#define VE_DEC_H265_SECOND_OUT_FMT_8BIT_PLUS_2BIT	0
+#define VE_DEC_H265_SECOND_OUT_FMT_P010			1
+#define VE_DEC_H265_SECOND_OUT_FMT_10BIT_4x4_TILED	2
+
+#define VE_DEC_H265_10BIT_CONFIGURE_SECOND_OUT_FMT(v) \
+	SHIFT_AND_MASK_BITS(v, 24, 23)
+#define VE_DEC_H265_10BIT_CONFIGURE_SECOND_2BIT_ENABLE	BIT(22)
+#define VE_DEC_H265_10BIT_CONFIGURE_SECOND_2BIT_STRIDE(v) \
+	SHIFT_AND_MASK_BITS(v, 21, 11)
+#define VE_DEC_H265_10BIT_CONFIGURE_FIRST_2BIT_STRIDE(v) \
+	SHIFT_AND_MASK_BITS(v, 10, 0)
+#define VE_DEC_H265_10BIT_CONFIGURE		(VE_ENGINE_DEC_H265 + 0x8c)
+
 #define VE_DEC_H265_LOW_ADDR_PRIMARY_CHROMA(a) \
 	SHIFT_AND_MASK_BITS(a, 31, 24)
 #define VE_DEC_H265_LOW_ADDR_SECONDARY_CHROMA(a) \