1 files changed, 577 insertions, 0 deletions

diff --git a/drivers/media/platform/coda/coda-jpeg.c b/drivers/media/platform/coda/coda-jpeg.c
index 92234fd1f4fd..00d19859db50 100644
--- a/drivers/media/platform/coda/coda-jpeg.c
+++ b/drivers/media/platform/coda/coda-jpeg.c
@@ -15,6 +15,7 @@
 
 #include <media/v4l2-common.h>
 #include <media/v4l2-fh.h>
+#include <media/v4l2-jpeg.h>
 #include <media/v4l2-mem2mem.h>
 #include <media/videobuf2-core.h>
 #include <media/videobuf2-dma-contig.h>
@@ -23,10 +24,12 @@
 #include "trace.h"
 
 #define SOI_MARKER	0xffd8
+#define APP9_MARKER	0xffe9
 #define DRI_MARKER	0xffdd
 #define DQT_MARKER	0xffdb
 #define DHT_MARKER	0xffc4
 #define SOF_MARKER	0xffc0
+#define SOS_MARKER	0xffda
 #define EOI_MARKER	0xffd9
 
 enum {
@@ -37,6 +40,18 @@ enum {
 	CODA9_JPEG_FORMAT_400,
 };
 
+struct coda_huff_tab {
+	u8 luma_dc[16 + 12];
+	u8 chroma_dc[16 + 12];
+	u8 luma_ac[16 + 162];
+	u8 chroma_ac[16 + 162];
+
+	/* DC Luma, DC Chroma, AC Luma, AC Chroma */
+	s16	min[4 * 16];
+	s16	max[4 * 16];
+	s8	ptr[4 * 16];
+};
+
 #define CODA9_JPEG_ENC_HUFF_DATA_SIZE	(256 + 256 + 16 + 16)
 
 /*
@@ -247,6 +262,291 @@ bool coda_jpeg_check_buffer(struct coda_ctx *ctx, struct vb2_buffer *vb)
 	return false;
 }
 
+static int coda9_jpeg_gen_dec_huff_tab(struct coda_ctx *ctx, int tab_num);
+
+int coda_jpeg_decode_header(struct coda_ctx *ctx, struct vb2_buffer *vb)
+{
+	struct coda_dev *dev = ctx->dev;
+	u8 *buf = vb2_plane_vaddr(vb, 0);
+	size_t len = vb2_get_plane_payload(vb, 0);
+	struct v4l2_jpeg_scan_header scan_header;
+	struct v4l2_jpeg_reference quantization_tables[4] = { };
+	struct v4l2_jpeg_reference huffman_tables[4] = { };
+	struct v4l2_jpeg_header header = {
+		.scan = &scan_header,
+		.quantization_tables = quantization_tables,
+		.huffman_tables = huffman_tables,
+	};
+	struct coda_q_data *q_data_src;
+	struct coda_huff_tab *huff_tab;
+	int i, j, ret;
+
+	ret = v4l2_jpeg_parse_header(buf, len, &header);
+	if (ret < 0) {
+		v4l2_err(&dev->v4l2_dev, "failed to parse header\n");
+		return ret;
+	}
+
+	ctx->params.jpeg_restart_interval = header.restart_interval;
+
+	/* check frame header */
+	if (header.frame.height > ctx->codec->max_h ||
+	    header.frame.width > ctx->codec->max_w) {
+		v4l2_err(&dev->v4l2_dev, "invalid dimensions: %dx%d\n",
+			 header.frame.width, header.frame.height);
+		return -EINVAL;
+	}
+
+	q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
+	if (header.frame.height != q_data_src->height ||
+	    header.frame.width != q_data_src->width) {
+		v4l2_err(&dev->v4l2_dev,
+			 "dimensions don't match format: %dx%d\n",
+			 header.frame.width, header.frame.height);
+		return -EINVAL;
+	}
+
+	if (header.frame.num_components != 3) {
+		v4l2_err(&dev->v4l2_dev,
+			 "unsupported number of components: %d\n",
+			 header.frame.num_components);
+		return -EINVAL;
+	}
+
+	/* install quantization tables */
+	if (quantization_tables[3].start) {
+		v4l2_err(&dev->v4l2_dev,
+			 "only 3 quantization tables supported\n");
+		return -EINVAL;
+	}
+	for (i = 0; i < 3; i++) {
+		if (!quantization_tables[i].start)
+			continue;
+		if (quantization_tables[i].length != 64) {
+			v4l2_err(&dev->v4l2_dev,
+				 "only 8-bit quantization tables supported\n");
+			continue;
+		}
+		if (!ctx->params.jpeg_qmat_tab[i])
+			ctx->params.jpeg_qmat_tab[i] = kmalloc(64, GFP_KERNEL);
+		memcpy(ctx->params.jpeg_qmat_tab[i],
+		       quantization_tables[i].start, 64);
+	}
+
+	/* install Huffman tables */
+	for (i = 0; i < 4; i++) {
+		if (!huffman_tables[i].start) {
+			v4l2_err(&dev->v4l2_dev, "missing Huffman table\n");
+			return -EINVAL;
+		}
+		/* AC tables should be between 17 -> 178, DC between 17 -> 28 */
+		if (huffman_tables[i].length < 17 ||
+		    huffman_tables[i].length > 178 ||
+		    ((i & 2) == 0 && huffman_tables[i].length > 28)) {
+			v4l2_err(&dev->v4l2_dev,
+				 "invalid Huffman table %d length: %zu\n",
+				 i, huffman_tables[i].length);
+			return -EINVAL;
+		}
+	}
+	huff_tab = ctx->params.jpeg_huff_tab;
+	if (!huff_tab) {
+		huff_tab = kzalloc(sizeof(struct coda_huff_tab), GFP_KERNEL);
+		if (!huff_tab)
+			return -ENOMEM;
+		ctx->params.jpeg_huff_tab = huff_tab;
+	}
+
+	memset(huff_tab, 0, sizeof(*huff_tab));
+	memcpy(huff_tab->luma_dc, huffman_tables[0].start, huffman_tables[0].length);
+	memcpy(huff_tab->chroma_dc, huffman_tables[1].start, huffman_tables[1].length);
+	memcpy(huff_tab->luma_ac, huffman_tables[2].start, huffman_tables[2].length);
+	memcpy(huff_tab->chroma_ac, huffman_tables[3].start, huffman_tables[3].length);
+
+	/* check scan header */
+	for (i = 0; i < scan_header.num_components; i++) {
+		struct v4l2_jpeg_scan_component_spec *scan_component;
+
+		scan_component = &scan_header.component[i];
+		for (j = 0; j < header.frame.num_components; j++) {
+			if (header.frame.component[j].component_identifier ==
+			    scan_component->component_selector)
+				break;
+		}
+		if (j == header.frame.num_components)
+			continue;
+
+		ctx->params.jpeg_huff_dc_index[j] =
+			scan_component->dc_entropy_coding_table_selector;
+		ctx->params.jpeg_huff_ac_index[j] =
+			scan_component->ac_entropy_coding_table_selector;
+	}
+
+	/* Generate Huffman table information */
+	for (i = 0; i < 4; i++)
+		coda9_jpeg_gen_dec_huff_tab(ctx, i);
+
+	/* start of entropy coded segment */
+	ctx->jpeg_ecs_offset = header.ecs_offset;
+
+	switch (header.frame.subsampling) {
+	case V4L2_JPEG_CHROMA_SUBSAMPLING_420:
+	case V4L2_JPEG_CHROMA_SUBSAMPLING_422:
+		ctx->params.jpeg_chroma_subsampling = header.frame.subsampling;
+		break;
+	default:
+		v4l2_err(&dev->v4l2_dev, "chroma subsampling not supported: %d",
+			 header.frame.subsampling);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static inline void coda9_jpeg_write_huff_values(struct coda_dev *dev, u8 *bits,
+						int num_values)
+{
+	s8 *values = (s8 *)(bits + 16);
+	int huff_length, i;
+
+	for (huff_length = 0, i = 0; i < 16; i++)
+		huff_length += bits[i];
+	for (i = huff_length; i < num_values; i++)
+		values[i] = -1;
+	for (i = 0; i < num_values; i++)
+		coda_write(dev, (s32)values[i], CODA9_REG_JPEG_HUFF_DATA);
+}
+
+static int coda9_jpeg_dec_huff_setup(struct coda_ctx *ctx)
+{
+	struct coda_huff_tab *huff_tab = ctx->params.jpeg_huff_tab;
+	struct coda_dev *dev = ctx->dev;
+	s16 *huff_min = huff_tab->min;
+	s16 *huff_max = huff_tab->max;
+	s8 *huff_ptr = huff_tab->ptr;
+	int i;
+
+	/* MIN Tables */
+	coda_write(dev, 0x003, CODA9_REG_JPEG_HUFF_CTRL);
+	coda_write(dev, 0x000, CODA9_REG_JPEG_HUFF_ADDR);
+	for (i = 0; i < 4 * 16; i++)
+		coda_write(dev, (s32)huff_min[i], CODA9_REG_JPEG_HUFF_DATA);
+
+	/* MAX Tables */
+	coda_write(dev, 0x403, CODA9_REG_JPEG_HUFF_CTRL);
+	coda_write(dev, 0x440, CODA9_REG_JPEG_HUFF_ADDR);
+	for (i = 0; i < 4 * 16; i++)
+		coda_write(dev, (s32)huff_max[i], CODA9_REG_JPEG_HUFF_DATA);
+
+	/* PTR Tables */
+	coda_write(dev, 0x803, CODA9_REG_JPEG_HUFF_CTRL);
+	coda_write(dev, 0x880, CODA9_REG_JPEG_HUFF_ADDR);
+	for (i = 0; i < 4 * 16; i++)
+		coda_write(dev, (s32)huff_ptr[i], CODA9_REG_JPEG_HUFF_DATA);
+
+	/* VAL Tables: DC Luma, DC Chroma, AC Luma, AC Chroma */
+	coda_write(dev, 0xc03, CODA9_REG_JPEG_HUFF_CTRL);
+	coda9_jpeg_write_huff_values(dev, huff_tab->luma_dc, 12);
+	coda9_jpeg_write_huff_values(dev, huff_tab->chroma_dc, 12);
+	coda9_jpeg_write_huff_values(dev, huff_tab->luma_ac, 162);
+	coda9_jpeg_write_huff_values(dev, huff_tab->chroma_ac, 162);
+	coda_write(dev, 0x000, CODA9_REG_JPEG_HUFF_CTRL);
+	return 0;
+}
+
+static inline void coda9_jpeg_write_qmat_tab(struct coda_dev *dev,
+					     u8 *qmat, int index)
+{
+	int i;
+
+	coda_write(dev, index | 0x3, CODA9_REG_JPEG_QMAT_CTRL);
+	for (i = 0; i < 64; i++)
+		coda_write(dev, qmat[i], CODA9_REG_JPEG_QMAT_DATA);
+	coda_write(dev, 0, CODA9_REG_JPEG_QMAT_CTRL);
+}
+
+static void coda9_jpeg_qmat_setup(struct coda_ctx *ctx)
+{
+	struct coda_dev *dev = ctx->dev;
+	int *qmat_index = ctx->params.jpeg_qmat_index;
+	u8 **qmat_tab = ctx->params.jpeg_qmat_tab;
+
+	coda9_jpeg_write_qmat_tab(dev, qmat_tab[qmat_index[0]], 0x00);
+	coda9_jpeg_write_qmat_tab(dev, qmat_tab[qmat_index[1]], 0x40);
+	coda9_jpeg_write_qmat_tab(dev, qmat_tab[qmat_index[2]], 0x80);
+}
+
+static void coda9_jpeg_dec_bbc_gbu_setup(struct coda_ctx *ctx,
+					 struct vb2_buffer *buf, u32 ecs_offset)
+{
+	struct coda_dev *dev = ctx->dev;
+	int page_ptr, word_ptr, bit_ptr;
+	u32 bbc_base_addr, end_addr;
+	int bbc_cur_pos;
+	int ret, val;
+
+	bbc_base_addr = vb2_dma_contig_plane_dma_addr(buf, 0);
+	end_addr = bbc_base_addr + vb2_get_plane_payload(buf, 0);
+
+	page_ptr = ecs_offset / 256;
+	word_ptr = (ecs_offset % 256) / 4;
+	if (page_ptr & 1)
+		word_ptr += 64;
+	bit_ptr = (ecs_offset % 4) * 8;
+	if (word_ptr & 1)
+		bit_ptr += 32;
+	word_ptr &= ~0x1;
+
+	coda_write(dev, end_addr, CODA9_REG_JPEG_BBC_WR_PTR);
+	coda_write(dev, bbc_base_addr, CODA9_REG_JPEG_BBC_BAS_ADDR);
+
+	/* Leave 3 256-byte page margin to avoid a BBC interrupt */
+	coda_write(dev, end_addr + 256 * 3 + 256, CODA9_REG_JPEG_BBC_END_ADDR);
+	val = DIV_ROUND_UP(vb2_plane_size(buf, 0), 256) + 3;
+	coda_write(dev, BIT(31) | val, CODA9_REG_JPEG_BBC_STRM_CTRL);
+
+	bbc_cur_pos = page_ptr;
+	coda_write(dev, bbc_cur_pos, CODA9_REG_JPEG_BBC_CUR_POS);
+	coda_write(dev, bbc_base_addr + (bbc_cur_pos << 8),
+			CODA9_REG_JPEG_BBC_EXT_ADDR);
+	coda_write(dev, (bbc_cur_pos & 1) << 6, CODA9_REG_JPEG_BBC_INT_ADDR);
+	coda_write(dev, 64, CODA9_REG_JPEG_BBC_DATA_CNT);
+	coda_write(dev, 0, CODA9_REG_JPEG_BBC_COMMAND);
+	do {
+		ret = coda_read(dev, CODA9_REG_JPEG_BBC_BUSY);
+	} while (ret == 1);
+
+	bbc_cur_pos++;
+	coda_write(dev, bbc_cur_pos, CODA9_REG_JPEG_BBC_CUR_POS);
+	coda_write(dev, bbc_base_addr + (bbc_cur_pos << 8),
+			CODA9_REG_JPEG_BBC_EXT_ADDR);
+	coda_write(dev, (bbc_cur_pos & 1) << 6, CODA9_REG_JPEG_BBC_INT_ADDR);
+	coda_write(dev, 64, CODA9_REG_JPEG_BBC_DATA_CNT);
+	coda_write(dev, 0, CODA9_REG_JPEG_BBC_COMMAND);
+	do {
+		ret = coda_read(dev, CODA9_REG_JPEG_BBC_BUSY);
+	} while (ret == 1);
+
+	bbc_cur_pos++;
+	coda_write(dev, bbc_cur_pos, CODA9_REG_JPEG_BBC_CUR_POS);
+	coda_write(dev, 1, CODA9_REG_JPEG_BBC_CTRL);
+
+	coda_write(dev, 0, CODA9_REG_JPEG_GBU_TT_CNT);
+	coda_write(dev, word_ptr, CODA9_REG_JPEG_GBU_WD_PTR);
+	coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBSR);
+	coda_write(dev, 127, CODA9_REG_JPEG_GBU_BBER);
+	if (page_ptr & 1) {
+		coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBIR);
+		coda_write(dev, 0, CODA9_REG_JPEG_GBU_BBHR);
+	} else {
+		coda_write(dev, 64, CODA9_REG_JPEG_GBU_BBIR);
+		coda_write(dev, 64, CODA9_REG_JPEG_GBU_BBHR);
+	}
+	coda_write(dev, 4, CODA9_REG_JPEG_GBU_CTRL);
+	coda_write(dev, bit_ptr, CODA9_REG_JPEG_GBU_FF_RPTR);
+	coda_write(dev, 3, CODA9_REG_JPEG_GBU_CTRL);
+}
+
 static const int bus_req_num[] = {
 	[CODA9_JPEG_FORMAT_420] = 2,
 	[CODA9_JPEG_FORMAT_422] = 3,
@@ -345,6 +645,71 @@ out:
 #define DC_TABLE_INDEX1		    2
 #define AC_TABLE_INDEX1		    3
 
+static u8 *coda9_jpeg_get_huff_bits(struct coda_ctx *ctx, int tab_num)
+{
+	struct coda_huff_tab *huff_tab = ctx->params.jpeg_huff_tab;
+
+	if (!huff_tab)
+		return NULL;
+
+	switch (tab_num) {
+	case DC_TABLE_INDEX0: return huff_tab->luma_dc;
+	case AC_TABLE_INDEX0: return huff_tab->luma_ac;
+	case DC_TABLE_INDEX1: return huff_tab->chroma_dc;
+	case AC_TABLE_INDEX1: return huff_tab->chroma_ac;
+	}
+
+	return NULL;
+}
+
+static int coda9_jpeg_gen_dec_huff_tab(struct coda_ctx *ctx, int tab_num)
+{
+	int ptr_cnt = 0, huff_code = 0, zero_flag = 0, data_flag = 0;
+	u8 *huff_bits;
+	s16 *huff_max;
+	s16 *huff_min;
+	s8 *huff_ptr;
+	int ofs;
+	int i;
+
+	huff_bits = coda9_jpeg_get_huff_bits(ctx, tab_num);
+	if (!huff_bits)
+		return -EINVAL;
+
+	/* DC/AC Luma, DC/AC Chroma -> DC Luma/Chroma, AC Luma/Chroma */
+	ofs = ((tab_num & 1) << 1) | ((tab_num >> 1) & 1);
+	ofs *= 16;
+
+	huff_ptr = ctx->params.jpeg_huff_tab->ptr + ofs;
+	huff_max = ctx->params.jpeg_huff_tab->max + ofs;
+	huff_min = ctx->params.jpeg_huff_tab->min + ofs;
+
+	for (i = 0; i < 16; i++) {
+		if (huff_bits[i]) {
+			huff_ptr[i] = ptr_cnt;
+			ptr_cnt += huff_bits[i];
+			huff_min[i] = huff_code;
+			huff_max[i] = huff_code + (huff_bits[i] - 1);
+			data_flag = 1;
+			zero_flag = 0;
+		} else {
+			huff_ptr[i] = -1;
+			huff_min[i] = -1;
+			huff_max[i] = -1;
+			zero_flag = 1;
+		}
+
+		if (data_flag == 1) {
+			if (zero_flag == 1)
+				huff_code <<= 1;
+			else
+				huff_code = (huff_max[i] + 1) << 1;
+		}
+	}
+
+	return 0;
+}
+
 static int coda9_jpeg_load_huff_tab(struct coda_ctx *ctx)
 {
 	struct {
@@ -880,6 +1245,13 @@ static void coda9_jpeg_finish_encode(struct coda_ctx *ctx)
 	coda_dbg(1, ctx, "job finished: encoded frame (%u)%s\n",
 		 dst_buf->sequence,
 		 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ? " (last)" : "");
+
+	/*
+	 * Reset JPEG processing unit after each encode run to work
+	 * around hangups when switching context between encoder and
+	 * decoder.
+	 */
+	coda_hw_reset(ctx);
 }
 
 static void coda9_jpeg_release(struct coda_ctx *ctx)
@@ -893,6 +1265,7 @@ static void coda9_jpeg_release(struct coda_ctx *ctx)
 	for (i = 0; i < 3; i++)
 		kfree(ctx->params.jpeg_qmat_tab[i]);
 	kfree(ctx->params.jpeg_huff_data);
+	kfree(ctx->params.jpeg_huff_tab);
 }
 
 const struct coda_context_ops coda9_jpeg_encode_ops = {
@@ -903,6 +1276,210 @@ const struct coda_context_ops coda9_jpeg_encode_ops = {
 	.release = coda9_jpeg_release,
 };
 
+/*
+ * Decoder context operations
+ */
+
+static int coda9_jpeg_start_decoding(struct coda_ctx *ctx)
+{
+	ctx->params.jpeg_qmat_index[0] = 0;
+	ctx->params.jpeg_qmat_index[1] = 1;
+	ctx->params.jpeg_qmat_index[2] = 1;
+	ctx->params.jpeg_qmat_tab[0] = luma_q;
+	ctx->params.jpeg_qmat_tab[1] = chroma_q;
+	/* nothing more to do here */
+
+	/* TODO: we could already scan the first header to get the chroma
+	 * format.
+	 */
+
+	return 0;
+}
+
+static int coda9_jpeg_prepare_decode(struct coda_ctx *ctx)
+{
+	struct coda_dev *dev = ctx->dev;
+	int aligned_width, aligned_height;
+	int chroma_format;
+	int ret;
+	u32 val, dst_fourcc;
+	struct coda_q_data *q_data_src, *q_data_dst;
+	struct vb2_v4l2_buffer *src_buf, *dst_buf;
+	int chroma_interleave;
+
+	src_buf = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
+	dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
+	q_data_src = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
+	q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
+	dst_fourcc = q_data_dst->fourcc;
+
+	if (vb2_get_plane_payload(&src_buf->vb2_buf, 0) == 0)
+		vb2_set_plane_payload(&src_buf->vb2_buf, 0,
+				      vb2_plane_size(&src_buf->vb2_buf, 0));
+
+	chroma_format = coda9_jpeg_chroma_format(q_data_dst->fourcc);
+	if (chroma_format < 0) {
+		v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->fh.m2m_ctx);
+		return chroma_format;
+	}
+
+	ret = coda_jpeg_decode_header(ctx, &src_buf->vb2_buf);
+	if (ret < 0) {
+		v4l2_err(&dev->v4l2_dev, "failed to decode JPEG header: %d\n",
+			 ret);
+
+		src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
+		dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
+		v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
+		v4l2_m2m_buf_done(dst_buf, VB2_BUF_STATE_DONE);
+
+		v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->fh.m2m_ctx);
+		return ret;
+	}
+
+	/* Round image dimensions to multiple of MCU size */
+	aligned_width = round_up(q_data_src->width, width_align[chroma_format]);
+	aligned_height = round_up(q_data_src->height, height_align[chroma_format]);
+	if (aligned_width != q_data_dst->bytesperline) {
+		v4l2_err(&dev->v4l2_dev, "stride mismatch: %d != %d\n",
+			 aligned_width, q_data_dst->bytesperline);
+	}
+
+	coda_set_gdi_regs(ctx);
+
+	val = ctx->params.jpeg_huff_ac_index[0] << 12 |
+	      ctx->params.jpeg_huff_ac_index[1] << 11 |
+	      ctx->params.jpeg_huff_ac_index[2] << 10 |
+	      ctx->params.jpeg_huff_dc_index[0] << 9 |
+	      ctx->params.jpeg_huff_dc_index[1] << 8 |
+	      ctx->params.jpeg_huff_dc_index[2] << 7;
+	if (ctx->params.jpeg_huff_tab)
+		val |= CODA9_JPEG_PIC_CTRL_USER_HUFFMAN_EN;
+	coda_write(dev, val, CODA9_REG_JPEG_PIC_CTRL);
+
+	coda_write(dev, aligned_width << 16 | aligned_height,
+			CODA9_REG_JPEG_PIC_SIZE);
+
+	chroma_interleave = (dst_fourcc == V4L2_PIX_FMT_NV12);
+	coda_write(dev, 0, CODA9_REG_JPEG_ROT_INFO);
+	coda_write(dev, bus_req_num[chroma_format], CODA9_REG_JPEG_OP_INFO);
+	coda_write(dev, mcu_info[chroma_format], CODA9_REG_JPEG_MCU_INFO);
+	coda_write(dev, 0, CODA9_REG_JPEG_SCL_INFO);
+	coda_write(dev, chroma_interleave, CODA9_REG_JPEG_DPB_CONFIG);
+	coda_write(dev, ctx->params.jpeg_restart_interval,
+			CODA9_REG_JPEG_RST_INTVAL);
+
+	if (ctx->params.jpeg_huff_tab) {
+		ret = coda9_jpeg_dec_huff_setup(ctx);
+		if (ret < 0) {
+			v4l2_err(&dev->v4l2_dev,
+				 "failed to set up Huffman tables: %d\n", ret);
+			v4l2_m2m_job_finish(ctx->dev->m2m_dev, ctx->fh.m2m_ctx);
+			return ret;
+		}
+	}
+
+	coda9_jpeg_qmat_setup(ctx);
+
+	coda9_jpeg_dec_bbc_gbu_setup(ctx, &src_buf->vb2_buf,
+				     ctx->jpeg_ecs_offset);
+
+	coda_write(dev, 0, CODA9_REG_JPEG_RST_INDEX);
+	coda_write(dev, 0, CODA9_REG_JPEG_RST_COUNT);
+
+	coda_write(dev, 0, CODA9_REG_JPEG_DPCM_DIFF_Y);
+	coda_write(dev, 0, CODA9_REG_JPEG_DPCM_DIFF_CB);
+	coda_write(dev, 0, CODA9_REG_JPEG_DPCM_DIFF_CR);
+
+	coda_write(dev, 0, CODA9_REG_JPEG_ROT_INFO);
+
+	coda_write(dev, 1, CODA9_GDI_CONTROL);
+	do {
+		ret = coda_read(dev, CODA9_GDI_STATUS);
+	} while (!ret);
+
+	val = (chroma_format << 17) | (chroma_interleave << 16) |
+	      q_data_dst->bytesperline;
+	if (ctx->tiled_map_type == GDI_TILED_FRAME_MB_RASTER_MAP)
+		val |= 3 << 20;
+	coda_write(dev, val, CODA9_GDI_INFO_CONTROL);
+
+	coda_write(dev, aligned_width << 16 | aligned_height,
+			CODA9_GDI_INFO_PIC_SIZE);
+
+	coda_write_base(ctx, q_data_dst, dst_buf, CODA9_GDI_INFO_BASE_Y);
+
+	coda_write(dev, 0, CODA9_REG_JPEG_DPB_BASE00);
+	coda_write(dev, 0, CODA9_GDI_CONTROL);
+	coda_write(dev, 1, CODA9_GDI_PIC_INIT_HOST);
+
+	trace_coda_jpeg_run(ctx, src_buf);
+
+	coda_write(dev, 1, CODA9_REG_JPEG_PIC_START);
+
+	return 0;
+}
+
+static void coda9_jpeg_finish_decode(struct coda_ctx *ctx)
+{
+	struct coda_dev *dev = ctx->dev;
+	struct vb2_v4l2_buffer *dst_buf, *src_buf;
+	struct coda_q_data *q_data_dst;
+	u32 err_mb;
+
+	err_mb = coda_read(dev, CODA9_REG_JPEG_PIC_ERRMB);
+	if (err_mb)
+		v4l2_err(&dev->v4l2_dev, "ERRMB: 0x%x\n", err_mb);
+
+	coda_write(dev, 0, CODA9_REG_JPEG_BBC_FLUSH_CMD);
+
+	/*
+	 * Lock to make sure that a decoder stop command running in parallel
+	 * will either already have marked src_buf as last, or it will wake up
+	 * the capture queue after the buffers are returned.
+	 */
+	mutex_lock(&ctx->wakeup_mutex);
+	src_buf = v4l2_m2m_src_buf_remove(ctx->fh.m2m_ctx);
+	dst_buf = v4l2_m2m_dst_buf_remove(ctx->fh.m2m_ctx);
+	dst_buf->sequence = ctx->osequence++;
+
+	trace_coda_jpeg_done(ctx, dst_buf);
+
+	dst_buf->flags &= ~(V4L2_BUF_FLAG_PFRAME | V4L2_BUF_FLAG_LAST);
+	dst_buf->flags |= V4L2_BUF_FLAG_KEYFRAME;
+	dst_buf->flags |= src_buf->flags & V4L2_BUF_FLAG_LAST;
+
+	v4l2_m2m_buf_copy_metadata(src_buf, dst_buf, false);
+
+	q_data_dst = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE);
+	vb2_set_plane_payload(&dst_buf->vb2_buf, 0, q_data_dst->sizeimage);
+
+	v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
+	coda_m2m_buf_done(ctx, dst_buf, err_mb ? VB2_BUF_STATE_ERROR :
+						 VB2_BUF_STATE_DONE);
+
+	mutex_unlock(&ctx->wakeup_mutex);
+
+	coda_dbg(1, ctx, "job finished: decoded frame (%u)%s\n",
+		 dst_buf->sequence,
+		 (dst_buf->flags & V4L2_BUF_FLAG_LAST) ? " (last)" : "");
+
+	/*
+	 * Reset JPEG processing unit after each decode run to work
+	 * around hangups when switching context between encoder and
+	 * decoder.
+	 */
+	coda_hw_reset(ctx);
+}
+
+const struct coda_context_ops coda9_jpeg_decode_ops = {
+	.queue_init = coda_encoder_queue_init, /* non-bitstream operation */
+	.start_streaming = coda9_jpeg_start_decoding,
+	.prepare_run = coda9_jpeg_prepare_decode,
+	.finish_run = coda9_jpeg_finish_decode,
+	.release = coda9_jpeg_release,
+};
+
 irqreturn_t coda9_jpeg_irq_handler(int irq, void *data)
 {
 	struct coda_dev *dev = data;