1 files changed, 187 insertions, 203 deletions

diff --git a/drivers/media/platform/chips-media/wave5/wave5-vpu-dec.c b/drivers/media/platform/chips-media/wave5/wave5-vpu-dec.c
index ef227af72348..fd71f0c43ac3 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vpu-dec.c
+++ b/drivers/media/platform/chips-media/wave5/wave5-vpu-dec.c
@@ -5,116 +5,98 @@
  * Copyright (C) 2021-2023 CHIPS&MEDIA INC
  */
 
+#include <linux/pm_runtime.h>
 #include "wave5-helper.h"
 
 #define VPU_DEC_DEV_NAME "C&M Wave5 VPU decoder"
 #define VPU_DEC_DRV_NAME "wave5-dec"
 
-#define DEFAULT_SRC_SIZE(width, height) ({			\
-	(width) * (height) / 8 * 3;					\
-})
+static const struct v4l2_frmsize_stepwise dec_hevc_frmsize = {
+	.min_width = W5_MIN_DEC_PIC_8_WIDTH,
+	.max_width = W5_MAX_DEC_PIC_WIDTH,
+	.step_width = W5_DEC_CODEC_STEP_WIDTH,
+	.min_height = W5_MIN_DEC_PIC_8_HEIGHT,
+	.max_height = W5_MAX_DEC_PIC_HEIGHT,
+	.step_height = W5_DEC_CODEC_STEP_HEIGHT,
+};
+
+static const struct v4l2_frmsize_stepwise dec_h264_frmsize = {
+	.min_width = W5_MIN_DEC_PIC_32_WIDTH,
+	.max_width = W5_MAX_DEC_PIC_WIDTH,
+	.step_width = W5_DEC_CODEC_STEP_WIDTH,
+	.min_height = W5_MIN_DEC_PIC_32_HEIGHT,
+	.max_height = W5_MAX_DEC_PIC_HEIGHT,
+	.step_height = W5_DEC_CODEC_STEP_HEIGHT,
+};
+
+static const struct v4l2_frmsize_stepwise dec_raw_frmsize = {
+	.min_width = W5_MIN_DEC_PIC_8_WIDTH,
+	.max_width = W5_MAX_DEC_PIC_WIDTH,
+	.step_width = W5_DEC_RAW_STEP_WIDTH,
+	.min_height = W5_MIN_DEC_PIC_8_HEIGHT,
+	.max_height = W5_MAX_DEC_PIC_HEIGHT,
+	.step_height = W5_DEC_RAW_STEP_HEIGHT,
+};
 
 static const struct vpu_format dec_fmt_list[FMT_TYPES][MAX_FMTS] = {
 	[VPU_FMT_TYPE_CODEC] = {
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_HEVC,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_hevc_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_H264,
-			.max_width = 8192,
-			.min_width = 32,
-			.max_height = 4320,
-			.min_height = 32,
+			.v4l2_frmsize = &dec_h264_frmsize,
 		},
 	},
 	[VPU_FMT_TYPE_RAW] = {
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV420,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV12,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV21,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV422P,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV16,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV61,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV420M,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV12M,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV21M,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV422M,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV16M,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_NV61M,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
+			.v4l2_frmsize = &dec_raw_frmsize,
 		},
 	}
 };
@@ -233,74 +215,6 @@ static void wave5_handle_src_buffer(struct vpu_instance *inst, dma_addr_t rd_ptr
 	inst->remaining_consumed_bytes = consumed_bytes;
 }
 
-static void wave5_update_pix_fmt(struct v4l2_pix_format_mplane *pix_mp, unsigned int width,
-				 unsigned int height)
-{
-	switch (pix_mp->pixelformat) {
-	case V4L2_PIX_FMT_YUV420:
-	case V4L2_PIX_FMT_NV12:
-	case V4L2_PIX_FMT_NV21:
-		pix_mp->width = round_up(width, 32);
-		pix_mp->height = round_up(height, 16);
-		pix_mp->plane_fmt[0].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[0].sizeimage = width * height * 3 / 2;
-		break;
-	case V4L2_PIX_FMT_YUV422P:
-	case V4L2_PIX_FMT_NV16:
-	case V4L2_PIX_FMT_NV61:
-		pix_mp->width = round_up(width, 32);
-		pix_mp->height = round_up(height, 16);
-		pix_mp->plane_fmt[0].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[0].sizeimage = width * height * 2;
-		break;
-	case V4L2_PIX_FMT_YUV420M:
-		pix_mp->width = round_up(width, 32);
-		pix_mp->height = round_up(height, 16);
-		pix_mp->plane_fmt[0].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[0].sizeimage = width * height;
-		pix_mp->plane_fmt[1].bytesperline = round_up(width, 32) / 2;
-		pix_mp->plane_fmt[1].sizeimage = width * height / 4;
-		pix_mp->plane_fmt[2].bytesperline = round_up(width, 32) / 2;
-		pix_mp->plane_fmt[2].sizeimage = width * height / 4;
-		break;
-	case V4L2_PIX_FMT_NV12M:
-	case V4L2_PIX_FMT_NV21M:
-		pix_mp->width = round_up(width, 32);
-		pix_mp->height = round_up(height, 16);
-		pix_mp->plane_fmt[0].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[0].sizeimage = width * height;
-		pix_mp->plane_fmt[1].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[1].sizeimage = width * height / 2;
-		break;
-	case V4L2_PIX_FMT_YUV422M:
-		pix_mp->width = round_up(width, 32);
-		pix_mp->height = round_up(height, 16);
-		pix_mp->plane_fmt[0].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[0].sizeimage = width * height;
-		pix_mp->plane_fmt[1].bytesperline = round_up(width, 32) / 2;
-		pix_mp->plane_fmt[1].sizeimage = width * height / 2;
-		pix_mp->plane_fmt[2].bytesperline = round_up(width, 32) / 2;
-		pix_mp->plane_fmt[2].sizeimage = width * height / 2;
-		break;
-	case V4L2_PIX_FMT_NV16M:
-	case V4L2_PIX_FMT_NV61M:
-		pix_mp->width = round_up(width, 32);
-		pix_mp->height = round_up(height, 16);
-		pix_mp->plane_fmt[0].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[0].sizeimage = width * height;
-		pix_mp->plane_fmt[1].bytesperline = round_up(width, 32);
-		pix_mp->plane_fmt[1].sizeimage = width * height;
-		break;
-	default:
-		pix_mp->width = width;
-		pix_mp->height = height;
-		pix_mp->plane_fmt[0].bytesperline = 0;
-		pix_mp->plane_fmt[0].sizeimage = max(DEFAULT_SRC_SIZE(width, height),
-						     pix_mp->plane_fmt[0].sizeimage);
-		break;
-	}
-}
-
 static int start_decode(struct vpu_instance *inst, u32 *fail_res)
 {
 	struct v4l2_m2m_ctx *m2m_ctx = inst->v4l2_fh.m2m_ctx;
@@ -388,6 +302,8 @@ static int handle_dynamic_resolution_change(struct vpu_instance *inst)
 	}
 
 	if (p_dec_info->initial_info_obtained) {
+		const struct vpu_format *vpu_fmt;
+
 		inst->conf_win.left = initial_info->pic_crop_rect.left;
 		inst->conf_win.top = initial_info->pic_crop_rect.top;
 		inst->conf_win.width = initial_info->pic_width -
@@ -395,10 +311,27 @@ static int handle_dynamic_resolution_change(struct vpu_instance *inst)
 		inst->conf_win.height = initial_info->pic_height -
 			initial_info->pic_crop_rect.top - initial_info->pic_crop_rect.bottom;
 
-		wave5_update_pix_fmt(&inst->src_fmt, initial_info->pic_width,
-				     initial_info->pic_height);
-		wave5_update_pix_fmt(&inst->dst_fmt, initial_info->pic_width,
-				     initial_info->pic_height);
+		vpu_fmt = wave5_find_vpu_fmt(inst->src_fmt.pixelformat,
+					     dec_fmt_list[VPU_FMT_TYPE_CODEC]);
+		if (!vpu_fmt)
+			return -EINVAL;
+
+		wave5_update_pix_fmt(&inst->src_fmt,
+				     VPU_FMT_TYPE_CODEC,
+				     initial_info->pic_width,
+				     initial_info->pic_height,
+				     vpu_fmt->v4l2_frmsize);
+
+		vpu_fmt = wave5_find_vpu_fmt(inst->dst_fmt.pixelformat,
+					     dec_fmt_list[VPU_FMT_TYPE_RAW]);
+		if (!vpu_fmt)
+			return -EINVAL;
+
+		wave5_update_pix_fmt(&inst->dst_fmt,
+				     VPU_FMT_TYPE_RAW,
+				     initial_info->pic_width,
+				     initial_info->pic_height,
+				     vpu_fmt->v4l2_frmsize);
 	}
 
 	v4l2_event_queue_fh(fh, &vpu_event_src_ch);
@@ -518,6 +451,8 @@ static void wave5_vpu_dec_finish_decode(struct vpu_instance *inst)
 	if (q_status.report_queue_count == 0 &&
 	    (q_status.instance_queue_count == 0 || dec_info.sequence_changed)) {
 		dev_dbg(inst->dev->dev, "%s: finishing job.\n", __func__);
+		pm_runtime_mark_last_busy(inst->dev->dev);
+		pm_runtime_put_autosuspend(inst->dev->dev);
 		v4l2_m2m_job_finish(inst->v4l2_m2m_dev, m2m_ctx);
 	}
 }
@@ -545,12 +480,12 @@ static int wave5_vpu_dec_enum_framesizes(struct file *f, void *fh, struct v4l2_f
 	}
 
 	fsize->type = V4L2_FRMSIZE_TYPE_CONTINUOUS;
-	fsize->stepwise.min_width = vpu_fmt->min_width;
-	fsize->stepwise.max_width = vpu_fmt->max_width;
-	fsize->stepwise.step_width = 1;
-	fsize->stepwise.min_height = vpu_fmt->min_height;
-	fsize->stepwise.max_height = vpu_fmt->max_height;
-	fsize->stepwise.step_height = 1;
+	fsize->stepwise.min_width = vpu_fmt->v4l2_frmsize->min_width;
+	fsize->stepwise.max_width = vpu_fmt->v4l2_frmsize->max_width;
+	fsize->stepwise.step_width = W5_DEC_CODEC_STEP_WIDTH;
+	fsize->stepwise.min_height = vpu_fmt->v4l2_frmsize->min_height;
+	fsize->stepwise.max_height = vpu_fmt->v4l2_frmsize->max_height;
+	fsize->stepwise.step_height = W5_DEC_CODEC_STEP_HEIGHT;
 
 	return 0;
 }
@@ -573,6 +508,7 @@ static int wave5_vpu_dec_try_fmt_cap(struct file *file, void *fh, struct v4l2_fo
 {
 	struct vpu_instance *inst = wave5_to_vpu_inst(fh);
 	struct dec_info *p_dec_info = &inst->codec_info->dec_info;
+	const struct v4l2_frmsize_stepwise *frmsize;
 	const struct vpu_format *vpu_fmt;
 	int width, height;
 
@@ -586,14 +522,12 @@ static int wave5_vpu_dec_try_fmt_cap(struct file *file, void *fh, struct v4l2_fo
 		width = inst->dst_fmt.width;
 		height = inst->dst_fmt.height;
 		f->fmt.pix_mp.pixelformat = inst->dst_fmt.pixelformat;
-		f->fmt.pix_mp.num_planes = inst->dst_fmt.num_planes;
+		frmsize = &dec_raw_frmsize;
 	} else {
-		const struct v4l2_format_info *info = v4l2_format_info(vpu_fmt->v4l2_pix_fmt);
-
-		width = clamp(f->fmt.pix_mp.width, vpu_fmt->min_width, vpu_fmt->max_width);
-		height = clamp(f->fmt.pix_mp.height, vpu_fmt->min_height, vpu_fmt->max_height);
+		width = f->fmt.pix_mp.width;
+		height = f->fmt.pix_mp.height;
 		f->fmt.pix_mp.pixelformat = vpu_fmt->v4l2_pix_fmt;
-		f->fmt.pix_mp.num_planes = info->mem_planes;
+		frmsize = vpu_fmt->v4l2_frmsize;
 	}
 
 	if (p_dec_info->initial_info_obtained) {
@@ -601,9 +535,8 @@ static int wave5_vpu_dec_try_fmt_cap(struct file *file, void *fh, struct v4l2_fo
 		height = inst->dst_fmt.height;
 	}
 
-	wave5_update_pix_fmt(&f->fmt.pix_mp, width, height);
-	f->fmt.pix_mp.flags = 0;
-	f->fmt.pix_mp.field = V4L2_FIELD_NONE;
+	wave5_update_pix_fmt(&f->fmt.pix_mp, VPU_FMT_TYPE_RAW,
+			     width, height, frmsize);
 	f->fmt.pix_mp.colorspace = inst->colorspace;
 	f->fmt.pix_mp.ycbcr_enc = inst->ycbcr_enc;
 	f->fmt.pix_mp.quantization = inst->quantization;
@@ -715,7 +648,9 @@ static int wave5_vpu_dec_enum_fmt_out(struct file *file, void *fh, struct v4l2_f
 static int wave5_vpu_dec_try_fmt_out(struct file *file, void *fh, struct v4l2_format *f)
 {
 	struct vpu_instance *inst = wave5_to_vpu_inst(fh);
+	const struct v4l2_frmsize_stepwise *frmsize;
 	const struct vpu_format *vpu_fmt;
+	int width, height;
 
 	dev_dbg(inst->dev->dev,
 		"%s: fourcc: %u width: %u height: %u num_planes: %u colorspace: %u field: %u\n",
@@ -724,20 +659,19 @@ static int wave5_vpu_dec_try_fmt_out(struct file *file, void *fh, struct v4l2_fo
 
 	vpu_fmt = wave5_find_vpu_fmt(f->fmt.pix_mp.pixelformat, dec_fmt_list[VPU_FMT_TYPE_CODEC]);
 	if (!vpu_fmt) {
+		width = inst->src_fmt.width;
+		height = inst->src_fmt.height;
 		f->fmt.pix_mp.pixelformat = inst->src_fmt.pixelformat;
-		f->fmt.pix_mp.num_planes = inst->src_fmt.num_planes;
-		wave5_update_pix_fmt(&f->fmt.pix_mp, inst->src_fmt.width, inst->src_fmt.height);
+		frmsize = &dec_hevc_frmsize;
 	} else {
-		int width = clamp(f->fmt.pix_mp.width, vpu_fmt->min_width, vpu_fmt->max_width);
-		int height = clamp(f->fmt.pix_mp.height, vpu_fmt->min_height, vpu_fmt->max_height);
-
+		width = f->fmt.pix_mp.width;
+		height = f->fmt.pix_mp.height;
 		f->fmt.pix_mp.pixelformat = vpu_fmt->v4l2_pix_fmt;
-		f->fmt.pix_mp.num_planes = 1;
-		wave5_update_pix_fmt(&f->fmt.pix_mp, width, height);
+		frmsize = vpu_fmt->v4l2_frmsize;
 	}
 
-	f->fmt.pix_mp.flags = 0;
-	f->fmt.pix_mp.field = V4L2_FIELD_NONE;
+	wave5_update_pix_fmt(&f->fmt.pix_mp, VPU_FMT_TYPE_CODEC,
+			     width, height, frmsize);
 
 	return 0;
 }
@@ -745,6 +679,7 @@ static int wave5_vpu_dec_try_fmt_out(struct file *file, void *fh, struct v4l2_fo
 static int wave5_vpu_dec_s_fmt_out(struct file *file, void *fh, struct v4l2_format *f)
 {
 	struct vpu_instance *inst = wave5_to_vpu_inst(fh);
+	const struct vpu_format *vpu_fmt;
 	int i, ret;
 
 	dev_dbg(inst->dev->dev,
@@ -779,7 +714,13 @@ static int wave5_vpu_dec_s_fmt_out(struct file *file, void *fh, struct v4l2_form
 	inst->quantization = f->fmt.pix_mp.quantization;
 	inst->xfer_func = f->fmt.pix_mp.xfer_func;
 
-	wave5_update_pix_fmt(&inst->dst_fmt, f->fmt.pix_mp.width, f->fmt.pix_mp.height);
+	vpu_fmt = wave5_find_vpu_fmt(inst->dst_fmt.pixelformat, dec_fmt_list[VPU_FMT_TYPE_RAW]);
+	if (!vpu_fmt)
+		return -EINVAL;
+
+	wave5_update_pix_fmt(&inst->dst_fmt, VPU_FMT_TYPE_RAW,
+			     f->fmt.pix_mp.width, f->fmt.pix_mp.height,
+			     vpu_fmt->v4l2_frmsize);
 
 	return 0;
 }
@@ -1002,6 +943,7 @@ static int wave5_vpu_dec_queue_setup(struct vb2_queue *q, unsigned int *num_buff
 	struct vpu_instance *inst = vb2_get_drv_priv(q);
 	struct v4l2_pix_format_mplane inst_format =
 		(q->type == V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE) ? inst->src_fmt : inst->dst_fmt;
+	unsigned int i;
 
 	dev_dbg(inst->dev->dev, "%s: num_buffers: %u | num_planes: %u | type: %u\n", __func__,
 		*num_buffers, *num_planes, q->type);
@@ -1015,31 +957,9 @@ static int wave5_vpu_dec_queue_setup(struct vb2_queue *q, unsigned int *num_buff
 		if (*num_buffers < inst->fbc_buf_count)
 			*num_buffers = inst->fbc_buf_count;
 
-		if (*num_planes == 1) {
-			if (inst->output_format == FORMAT_422)
-				sizes[0] = inst_format.width * inst_format.height * 2;
-			else
-				sizes[0] = inst_format.width * inst_format.height * 3 / 2;
-			dev_dbg(inst->dev->dev, "%s: size[0]: %u\n", __func__, sizes[0]);
-		} else if (*num_planes == 2) {
-			sizes[0] = inst_format.width * inst_format.height;
-			if (inst->output_format == FORMAT_422)
-				sizes[1] = inst_format.width * inst_format.height;
-			else
-				sizes[1] = inst_format.width * inst_format.height / 2;
-			dev_dbg(inst->dev->dev, "%s: size[0]: %u | size[1]: %u\n",
-				__func__, sizes[0], sizes[1]);
-		} else if (*num_planes == 3) {
-			sizes[0] = inst_format.width * inst_format.height;
-			if (inst->output_format == FORMAT_422) {
-				sizes[1] = inst_format.width * inst_format.height / 2;
-				sizes[2] = inst_format.width * inst_format.height / 2;
-			} else {
-				sizes[1] = inst_format.width * inst_format.height / 4;
-				sizes[2] = inst_format.width * inst_format.height / 4;
-			}
-			dev_dbg(inst->dev->dev, "%s: size[0]: %u | size[1]: %u | size[2]: %u\n",
-				__func__, sizes[0], sizes[1], sizes[2]);
+		for (i = 0; i < *num_planes; i++) {
+			sizes[i] = inst_format.plane_fmt[i].sizeimage;
+			dev_dbg(inst->dev->dev, "%s: size[%u]: %u\n", __func__, i, sizes[i]);
 		}
 	}
 
@@ -1055,6 +975,22 @@ static int wave5_prepare_fb(struct vpu_instance *inst)
 	int ret, i;
 	struct v4l2_m2m_buffer *buf, *n;
 	struct v4l2_m2m_ctx *m2m_ctx = inst->v4l2_fh.m2m_ctx;
+	u32 bitdepth = inst->codec_info->dec_info.initial_info.luma_bitdepth;
+
+	switch (bitdepth) {
+	case 8:
+		break;
+	case 10:
+		if (inst->std == W_HEVC_DEC &&
+		    inst->dev->attr.support_hevc10bit_dec)
+			break;
+
+		fallthrough;
+	default:
+		dev_err(inst->dev->dev, "no support for %d bit depth\n", bitdepth);
+
+		return -EINVAL;
+	}
 
 	linear_num = v4l2_m2m_num_dst_bufs_ready(m2m_ctx);
 	non_linear_num = inst->fbc_buf_count;
@@ -1063,7 +999,7 @@ static int wave5_prepare_fb(struct vpu_instance *inst)
 		struct frame_buffer *frame = &inst->frame_buf[i];
 		struct vpu_buf *vframe = &inst->frame_vbuf[i];
 
-		fb_stride = inst->dst_fmt.width;
+		fb_stride = ALIGN(inst->dst_fmt.width * bitdepth / 8, 32);
 		fb_height = ALIGN(inst->dst_fmt.height, 32);
 		luma_size = fb_stride * fb_height;
 
@@ -1382,6 +1318,7 @@ static int wave5_vpu_dec_start_streaming(struct vb2_queue *q, unsigned int count
 	int ret = 0;
 
 	dev_dbg(inst->dev->dev, "%s: type: %u\n", __func__, q->type);
+	pm_runtime_resume_and_get(inst->dev->dev);
 
 	v4l2_m2m_update_start_streaming_state(m2m_ctx, q);
 
@@ -1416,7 +1353,8 @@ static int wave5_vpu_dec_start_streaming(struct vb2_queue *q, unsigned int count
 		if (ret)
 			goto return_buffers;
 
-		if (inst->state == VPU_INST_STATE_INIT_SEQ) {
+		if (inst->state == VPU_INST_STATE_INIT_SEQ &&
+		    inst->dev->product_code == WAVE521C_CODE) {
 			if (initial_info->luma_bitdepth != 8) {
 				dev_info(inst->dev->dev, "%s: no support for %d bit depth",
 					 __func__, initial_info->luma_bitdepth);
@@ -1424,14 +1362,17 @@ static int wave5_vpu_dec_start_streaming(struct vb2_queue *q, unsigned int count
 				goto return_buffers;
 			}
 		}
-	}
 
+	}
+	pm_runtime_mark_last_busy(inst->dev->dev);
+	pm_runtime_put_autosuspend(inst->dev->dev);
 	return ret;
 
 free_bitstream_vbuf:
 	wave5_vdi_free_dma_memory(inst->dev, &inst->bitstream_vbuf);
 return_buffers:
 	wave5_return_bufs(q, VB2_BUF_STATE_QUEUED);
+	pm_runtime_put_autosuspend(inst->dev->dev);
 	return ret;
 }
 
@@ -1442,6 +1383,16 @@ static int streamoff_output(struct vb2_queue *q)
 	struct vb2_v4l2_buffer *buf;
 	int ret;
 	dma_addr_t new_rd_ptr;
+	struct dec_output_info dec_info;
+	unsigned int i;
+
+	for (i = 0; i < v4l2_m2m_num_dst_bufs_ready(m2m_ctx); i++) {
+		ret = wave5_vpu_dec_set_disp_flag(inst, i);
+		if (ret)
+			dev_dbg(inst->dev->dev,
+				"%s: Setting display flag of buf index: %u, fail: %d\n",
+				__func__, i, ret);
+	}
 
 	while ((buf = v4l2_m2m_src_buf_remove(m2m_ctx))) {
 		dev_dbg(inst->dev->dev, "%s: (Multiplanar) buf type %4u | index %4u\n",
@@ -1449,6 +1400,11 @@ static int streamoff_output(struct vb2_queue *q)
 		v4l2_m2m_buf_done(buf, VB2_BUF_STATE_ERROR);
 	}
 
+	while (wave5_vpu_dec_get_output_info(inst, &dec_info) == 0) {
+		if (dec_info.index_frame_display >= 0)
+			wave5_vpu_dec_set_disp_flag(inst, dec_info.index_frame_display);
+	}
+
 	ret = wave5_vpu_flush_instance(inst);
 	if (ret)
 		return ret;
@@ -1517,6 +1473,7 @@ static void wave5_vpu_dec_stop_streaming(struct vb2_queue *q)
 	bool check_cmd = TRUE;
 
 	dev_dbg(inst->dev->dev, "%s: type: %u\n", __func__, q->type);
+	pm_runtime_resume_and_get(inst->dev->dev);
 
 	while (check_cmd) {
 		struct queue_status_info q_status;
@@ -1531,7 +1488,7 @@ static void wave5_vpu_dec_stop_streaming(struct vb2_queue *q)
 			break;
 
 		if (wave5_vpu_dec_get_output_info(inst, &dec_output_info))
-			dev_dbg(inst->dev->dev, "Getting decoding results from fw, fail\n");
+			dev_dbg(inst->dev->dev, "there is no output info\n");
 	}
 
 	v4l2_m2m_update_stop_streaming_state(m2m_ctx, q);
@@ -1540,12 +1497,13 @@ static void wave5_vpu_dec_stop_streaming(struct vb2_queue *q)
 		streamoff_output(q);
 	else
 		streamoff_capture(q);
+
+	pm_runtime_mark_last_busy(inst->dev->dev);
+	pm_runtime_put_autosuspend(inst->dev->dev);
 }
 
 static const struct vb2_ops wave5_vpu_dec_vb2_ops = {
 	.queue_setup = wave5_vpu_dec_queue_setup,
-	.wait_prepare = vb2_ops_wait_prepare,
-	.wait_finish = vb2_ops_wait_finish,
 	.buf_queue = wave5_vpu_dec_buf_queue,
 	.start_streaming = wave5_vpu_dec_start_streaming,
 	.stop_streaming = wave5_vpu_dec_stop_streaming,
@@ -1554,20 +1512,15 @@ static const struct vb2_ops wave5_vpu_dec_vb2_ops = {
 static void wave5_set_default_format(struct v4l2_pix_format_mplane *src_fmt,
 				     struct v4l2_pix_format_mplane *dst_fmt)
 {
-	unsigned int dst_pix_fmt = dec_fmt_list[VPU_FMT_TYPE_RAW][0].v4l2_pix_fmt;
-	const struct v4l2_format_info *dst_fmt_info = v4l2_format_info(dst_pix_fmt);
-
 	src_fmt->pixelformat = dec_fmt_list[VPU_FMT_TYPE_CODEC][0].v4l2_pix_fmt;
-	src_fmt->field = V4L2_FIELD_NONE;
-	src_fmt->flags = 0;
-	src_fmt->num_planes = 1;
-	wave5_update_pix_fmt(src_fmt, 720, 480);
-
-	dst_fmt->pixelformat = dst_pix_fmt;
-	dst_fmt->field = V4L2_FIELD_NONE;
-	dst_fmt->flags = 0;
-	dst_fmt->num_planes = dst_fmt_info->mem_planes;
-	wave5_update_pix_fmt(dst_fmt, 736, 480);
+	wave5_update_pix_fmt(src_fmt, VPU_FMT_TYPE_CODEC,
+			     W5_DEF_DEC_PIC_WIDTH, W5_DEF_DEC_PIC_HEIGHT,
+			     &dec_hevc_frmsize);
+
+	dst_fmt->pixelformat = dec_fmt_list[VPU_FMT_TYPE_RAW][0].v4l2_pix_fmt;
+	wave5_update_pix_fmt(dst_fmt, VPU_FMT_TYPE_RAW,
+			     W5_DEF_DEC_PIC_WIDTH, W5_DEF_DEC_PIC_HEIGHT,
+			     &dec_raw_frmsize);
 }
 
 static int wave5_vpu_dec_queue_init(void *priv, struct vb2_queue *src_vq, struct vb2_queue *dst_vq)
@@ -1626,7 +1579,7 @@ static void wave5_vpu_dec_device_run(void *priv)
 	int ret = 0;
 
 	dev_dbg(inst->dev->dev, "%s: Fill the ring buffer with new bitstream data", __func__);
-
+	pm_runtime_resume_and_get(inst->dev->dev);
 	ret = fill_ringbuffer(inst);
 	if (ret) {
 		dev_warn(inst->dev->dev, "Filling ring buffer failed\n");
@@ -1709,6 +1662,8 @@ static void wave5_vpu_dec_device_run(void *priv)
 
 finish_job_and_return:
 	dev_dbg(inst->dev->dev, "%s: leave and finish job", __func__);
+	pm_runtime_mark_last_busy(inst->dev->dev);
+	pm_runtime_put_autosuspend(inst->dev->dev);
 	v4l2_m2m_job_finish(inst->v4l2_m2m_dev, m2m_ctx);
 }
 
@@ -1810,7 +1765,6 @@ static int wave5_vpu_open_dec(struct file *filp)
 	v4l2_fh_add(&inst->v4l2_fh);
 
 	INIT_LIST_HEAD(&inst->list);
-	list_add_tail(&inst->list, &dev->instances);
 
 	inst->v4l2_m2m_dev = inst->dev->v4l2_m2m_dec_dev;
 	inst->v4l2_fh.m2m_ctx =
@@ -1865,7 +1819,23 @@ static int wave5_vpu_open_dec(struct file *filp)
 		goto cleanup_inst;
 	}
 
-	wave5_vdi_allocate_sram(inst->dev);
+	/*
+	 * For Wave515 SRAM memory was already allocated
+	 * at wave5_vpu_dec_register_device()
+	 */
+	if (inst->dev->product_code != WAVE515_CODE)
+		wave5_vdi_allocate_sram(inst->dev);
+
+	ret = mutex_lock_interruptible(&dev->dev_lock);
+	if (ret)
+		goto cleanup_inst;
+
+	if (list_empty(&dev->instances))
+		pm_runtime_use_autosuspend(inst->dev->dev);
+
+	list_add_tail(&inst->list, &dev->instances);
+
+	mutex_unlock(&dev->dev_lock);
 
 	return 0;
 
@@ -1893,6 +1863,13 @@ int wave5_vpu_dec_register_device(struct vpu_device *dev)
 	struct video_device *vdev_dec;
 	int ret;
 
+	/*
+	 * Secondary AXI setup for Wave515 is done by INIT_VPU command,
+	 * i.e. wave5_vpu_init(), that's why we allocate SRAM memory early.
+	 */
+	if (dev->product_code == WAVE515_CODE)
+		wave5_vdi_allocate_sram(dev);
+
 	vdev_dec = devm_kzalloc(dev->v4l2_dev.dev, sizeof(*vdev_dec), GFP_KERNEL);
 	if (!vdev_dec)
 		return -ENOMEM;
@@ -1926,6 +1903,13 @@ int wave5_vpu_dec_register_device(struct vpu_device *dev)
 
 void wave5_vpu_dec_unregister_device(struct vpu_device *dev)
 {
+	/*
+	 * Here is a freeing pair for Wave515 SRAM memory allocation
+	 * happened at wave5_vpu_dec_register_device().
+	 */
+	if (dev->product_code == WAVE515_CODE)
+		wave5_vdi_free_sram(dev);
+
 	video_unregister_device(dev->video_dev_dec);
 	if (dev->v4l2_m2m_dec_dev)
 		v4l2_m2m_release(dev->v4l2_m2m_dec_dev);