media: coda: jpeg: add CODA960 JPEG decoder support

This patch adds JPEG decoding support for CODA960, handling the JPEG
hardware directly. A separate JPEG decoder video device is created due
to the separate hardware unit and different supported pixel formats.
While the hardware can not change subsampling on the fly, it can decode
4:2:2 subsampled JPEG images into YUV422P.

[m.felsch@pengutronix.de: fix qsequence counting by explicitly
 checking for the !use_bit case]
[hverkuil-cisco@xs4all.nl: fix 'warning: missing braces around initializer']
[mchehab+huawei@kernel.org: fix 'BRACES: Unbalanced braces around else statement']
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>

drivers/media/platform/Kconfig

@@ -187,6 +187,7 @@ config VIDEO_CODA
 	select SRAM
 	select VIDEOBUF2_DMA_CONTIG
 	select VIDEOBUF2_VMALLOC
+	select V4L2_JPEG_HELPER
 	select V4L2_MEM2MEM_DEV
 	select GENERIC_ALLOCATOR
 	help

drivers/media/platform/coda/coda-common.c

@@ -159,6 +159,7 @@ static const struct coda_codec coda9_codecs[] = {
 	CODA_CODEC(CODA9_MODE_DECODE_H264, V4L2_PIX_FMT_H264,   V4L2_PIX_FMT_YUV420, 1920, 1088),
 	CODA_CODEC(CODA9_MODE_DECODE_MP2,  V4L2_PIX_FMT_MPEG2,  V4L2_PIX_FMT_YUV420, 1920, 1088),
 	CODA_CODEC(CODA9_MODE_DECODE_MP4,  V4L2_PIX_FMT_MPEG4,  V4L2_PIX_FMT_YUV420, 1920, 1088),
+	CODA_CODEC(CODA9_MODE_DECODE_MJPG, V4L2_PIX_FMT_JPEG,	V4L2_PIX_FMT_YUV420, 8192, 8192),
 };
 
 struct coda_video_device {
@@ -252,6 +253,22 @@ static const struct coda_video_device coda9_jpeg_encoder = {
 	},
 };
 
+static const struct coda_video_device coda9_jpeg_decoder = {
+	.name = "coda-jpeg-decoder",
+	.type = CODA_INST_DECODER,
+	.ops = &coda9_jpeg_decode_ops,
+	.direct = true,
+	.src_formats = {
+		V4L2_PIX_FMT_JPEG,
+	},
+	.dst_formats = {
+		V4L2_PIX_FMT_NV12,
+		V4L2_PIX_FMT_YUV420,
+		V4L2_PIX_FMT_YVU420,
+		V4L2_PIX_FMT_YUV422P,
+	},
+};
+
 static const struct coda_video_device *codadx6_video_devices[] = {
 	&coda_bit_encoder,
 };
@@ -270,6 +287,7 @@ static const struct coda_video_device *coda7_video_devices[] = {
 
 static const struct coda_video_device *coda9_video_devices[] = {
 	&coda9_jpeg_encoder,
+	&coda9_jpeg_decoder,
 	&coda_bit_encoder,
 	&coda_bit_decoder,
 };
@@ -411,6 +429,12 @@ static int coda_querycap(struct file *file, void *priv,
 	return 0;
 }
 
+static const u32 coda_formats_420[CODA_MAX_FORMATS] = {
+		V4L2_PIX_FMT_NV12,
+		V4L2_PIX_FMT_YUV420,
+		V4L2_PIX_FMT_YVU420,
+};
+
 static int coda_enum_fmt(struct file *file, void *priv,
 			 struct v4l2_fmtdesc *f)
 {
@@ -421,10 +445,33 @@ static int coda_enum_fmt(struct file *file, void *priv,
 
 	if (f->type == V4L2_BUF_TYPE_VIDEO_OUTPUT)
 		formats = cvd->src_formats;
-	else if (f->type == V4L2_BUF_TYPE_VIDEO_CAPTURE)
+	else if (f->type == V4L2_BUF_TYPE_VIDEO_CAPTURE) {
+		struct coda_q_data *q_data_src;
+		struct vb2_queue *src_vq;
+
 		formats = cvd->dst_formats;
-	else
+
+		/*
+		 * If the source format is already fixed, only allow the same
+		 * chroma subsampling.
+		 */
+		q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
+		src_vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx,
+					 V4L2_BUF_TYPE_VIDEO_OUTPUT);
+		if (q_data_src->fourcc == V4L2_PIX_FMT_JPEG &&
+		    vb2_is_streaming(src_vq)) {
+			if (ctx->params.jpeg_chroma_subsampling ==
+			    V4L2_JPEG_CHROMA_SUBSAMPLING_420) {
+				formats = coda_formats_420;
+			} else if (ctx->params.jpeg_chroma_subsampling ==
+				   V4L2_JPEG_CHROMA_SUBSAMPLING_422) {
+				f->pixelformat = V4L2_PIX_FMT_YUV422P;
+				return f->index ? -EINVAL : 0;
+			}
+		}
+	} else {
 		return -EINVAL;
+	}
 
 	if (f->index >= CODA_MAX_FORMATS || formats[f->index] == 0)
 		return -EINVAL;
@@ -614,12 +661,23 @@ static int coda_try_fmt_vid_cap(struct file *file, void *priv,
 
 	/*
 	 * If the source format is already fixed, only allow the same output
-	 * resolution
+	 * resolution. When decoding JPEG images, we also have to make sure to
+	 * use the same chroma subsampling.
 	 */
 	src_vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
 	if (vb2_is_streaming(src_vq)) {
 		f->fmt.pix.width = q_data_src->width;
 		f->fmt.pix.height = q_data_src->height;
+
+		if (q_data_src->fourcc == V4L2_PIX_FMT_JPEG) {
+			if (ctx->params.jpeg_chroma_subsampling ==
+			    V4L2_JPEG_CHROMA_SUBSAMPLING_420 &&
+			    f->fmt.pix.pixelformat == V4L2_PIX_FMT_YUV422P)
+				f->fmt.pix.pixelformat = V4L2_PIX_FMT_NV12;
+			else if (ctx->params.jpeg_chroma_subsampling ==
+				 V4L2_JPEG_CHROMA_SUBSAMPLING_422)
+				f->fmt.pix.pixelformat = V4L2_PIX_FMT_YUV422P;
+		}
 	}
 
 	f->fmt.pix.colorspace = ctx->colorspace;
@@ -637,12 +695,18 @@ static int coda_try_fmt_vid_cap(struct file *file, void *priv,
 	if (ret < 0)
 		return ret;
 
-	/* The decoders always write complete macroblocks */
+	/* The decoders always write complete macroblocks or MCUs */
 	if (ctx->inst_type == CODA_INST_DECODER) {
-		f->fmt.pix.height = round_up(f->fmt.pix.height, 16);
 		f->fmt.pix.bytesperline = round_up(f->fmt.pix.width, 16);
-		f->fmt.pix.sizeimage = f->fmt.pix.bytesperline *
-				       f->fmt.pix.height * 3 / 2;
+		f->fmt.pix.height = round_up(f->fmt.pix.height, 16);
+		if (codec->src_fourcc == V4L2_PIX_FMT_JPEG &&
+		    f->fmt.pix.pixelformat == V4L2_PIX_FMT_YUV422P) {
+			f->fmt.pix.sizeimage = f->fmt.pix.bytesperline *
+					       f->fmt.pix.height * 2;
+		} else {
+			f->fmt.pix.sizeimage = f->fmt.pix.bytesperline *
+					       f->fmt.pix.height * 3 / 2;
+		}
 
 		ret = coda_try_fmt_vdoa(ctx, f, &use_vdoa);
 		if (ret < 0)
@@ -747,6 +811,7 @@ static int coda_s_fmt(struct coda_ctx *ctx, struct v4l2_format *f,
 		/* else fall through */
 	case V4L2_PIX_FMT_YUV420:
 	case V4L2_PIX_FMT_YVU420:
+	case V4L2_PIX_FMT_YUV422P:
 		ctx->tiled_map_type = GDI_LINEAR_FRAME_MAP;
 		break;
 	default:
@@ -1894,6 +1959,42 @@ static int coda_start_streaming(struct vb2_queue *q, unsigned int count)
 			}
 		}
 
+		/*
+		 * Check the first input JPEG buffer to determine chroma
+		 * subsampling.
+		 */
+		if (q_data_src->fourcc == V4L2_PIX_FMT_JPEG) {
+			buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
+			ret = coda_jpeg_decode_header(ctx, &buf->vb2_buf);
+			if (ret < 0) {
+				v4l2_err(v4l2_dev,
+					 "failed to decode JPEG header: %d\n",
+					 ret);
+				goto err;
+			}
+
+			q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
+			q_data_dst->width = round_up(q_data_src->width, 16);
+			q_data_dst->height = round_up(q_data_src->height, 16);
+			q_data_dst->bytesperline = q_data_dst->width;
+			if (ctx->params.jpeg_chroma_subsampling ==
+			    V4L2_JPEG_CHROMA_SUBSAMPLING_420) {
+				q_data_dst->sizeimage =
+						q_data_dst->bytesperline *
+						q_data_dst->height * 3 / 2;
+				if (q_data_dst->fourcc != V4L2_PIX_FMT_YUV420)
+					q_data_dst->fourcc = V4L2_PIX_FMT_NV12;
+			} else {
+				q_data_dst->sizeimage =
+						q_data_dst->bytesperline *
+						q_data_dst->height * 2;
+				q_data_dst->fourcc = V4L2_PIX_FMT_YUV422P;
+			}
+			q_data_dst->rect.left = 0;
+			q_data_dst->rect.top = 0;
+			q_data_dst->rect.width = q_data_src->width;
+			q_data_dst->rect.height = q_data_src->height;
+		}
 		ctx->streamon_out = 1;
 	} else {
 		ctx->streamon_cap = 1;

drivers/media/platform/coda/coda.h

@@ -69,7 +69,7 @@ struct coda_aux_buf {
 
 struct coda_dev {
 	struct v4l2_device	v4l2_dev;
-	struct video_device	vfd[5];
+	struct video_device	vfd[6];
 	struct device		*dev;
 	const struct coda_devtype *devtype;
 	int			firmware;
@@ -123,10 +123,15 @@ struct coda_params {
 	u8			mpeg4_inter_qp;
 	u8			gop_size;
 	int			intra_refresh;
+	enum v4l2_jpeg_chroma_subsampling jpeg_chroma_subsampling;
 	u8			jpeg_quality;
 	u8			jpeg_restart_interval;
 	u8			*jpeg_qmat_tab[3];
+	int			jpeg_qmat_index[3];
+	int			jpeg_huff_dc_index[3];
+	int			jpeg_huff_ac_index[3];
 	u32			*jpeg_huff_data;
+	struct coda_huff_tab	*jpeg_huff_tab;
 	int			codec_mode;
 	int			codec_mode_aux;
 	enum v4l2_mpeg_video_multi_slice_mode slice_mode;
@@ -238,6 +243,7 @@ struct coda_ctx {
 	struct v4l2_fh			fh;
 	int				gopcounter;
 	int				runcounter;
+	int				jpeg_ecs_offset;
 	char				vpu_header[3][64];
 	int				vpu_header_size[3];
 	struct kfifo			bitstream_fifo;
@@ -362,12 +368,14 @@ void coda_update_profile_level_ctrls(struct coda_ctx *ctx, u8 profile_idc,
 				     u8 level_idc);
 
 bool coda_jpeg_check_buffer(struct coda_ctx *ctx, struct vb2_buffer *vb);
+int coda_jpeg_decode_header(struct coda_ctx *ctx, struct vb2_buffer *vb);
 int coda_jpeg_write_tables(struct coda_ctx *ctx);
 void coda_set_jpeg_compression_quality(struct coda_ctx *ctx, int quality);
 
 extern const struct coda_context_ops coda_bit_encode_ops;
 extern const struct coda_context_ops coda_bit_decode_ops;
 extern const struct coda_context_ops coda9_jpeg_encode_ops;
+extern const struct coda_context_ops coda9_jpeg_decode_ops;
 
 irqreturn_t coda_irq_handler(int irq, void *data);
 irqreturn_t coda9_jpeg_irq_handler(int irq, void *data);