media: add wave511 v4l2 support for starfive jh 7110 platfrom

Signed-off-by: Som Qin <som.qin@starfivetech.com>

10 files changed, 194 insertions, 82 deletions

diff --git a/drivers/media/platform/chips-media/wave5/wave5-helper.c b/drivers/media/platform/chips-media/wave5/wave5-helper.c
index 8433ecab230c..ec710b838dfe 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-helper.c
+++ b/drivers/media/platform/chips-media/wave5/wave5-helper.c
@@ -29,9 +29,6 @@ void wave5_cleanup_instance(struct vpu_instance *inst)
 {
 	int i;
 
-	if (list_is_singular(&inst->list))
-		wave5_vdi_free_sram(inst->dev);
-
 	for (i = 0; i < inst->fbc_buf_count; i++)
 		wave5_vpu_dec_reset_framebuffer(inst, i);
 
diff --git a/drivers/media/platform/chips-media/wave5/wave5-helper.h b/drivers/media/platform/chips-media/wave5/wave5-helper.h
index 6cee1c14d3ce..b17c96670a3d 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-helper.h
+++ b/drivers/media/platform/chips-media/wave5/wave5-helper.h
@@ -11,7 +11,7 @@
 #include "wave5-vpu.h"
 
 #define FMT_TYPES	2
-#define MAX_FMTS	12
+#define MAX_FMTS	6
 
 const char *state_to_str(enum vpu_instance_state state);
 void wave5_cleanup_instance(struct vpu_instance *inst);
diff --git a/drivers/media/platform/chips-media/wave5/wave5-hw.c b/drivers/media/platform/chips-media/wave5/wave5-hw.c
index 3fcb2d92add8..29e15888a21a 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-hw.c
+++ b/drivers/media/platform/chips-media/wave5/wave5-hw.c
@@ -186,12 +186,13 @@ unsigned int wave5_vpu_get_product_id(struct vpu_device *vpu_dev)
 	u32 val = vpu_read_reg(vpu_dev, W5_PRODUCT_NUMBER);
 
 	switch (val) {
+	case WAVE511_CODE:
+		return PRODUCT_ID_511;
 	case WAVE521C_CODE:
 		return PRODUCT_ID_521;
 	case WAVE521_CODE:
 	case WAVE521C_DUAL_CODE:
 	case WAVE521E1_CODE:
-	case WAVE511_CODE:
 	case WAVE517_CODE:
 	case WAVE537_CODE:
 		dev_err(vpu_dev->dev, "Unsupported product id (%x)\n", val);
@@ -607,6 +608,20 @@ static void wave5_get_dec_seq_result(struct vpu_instance *inst, struct dec_initi
 	info->pic_crop_rect.top = (reg_val >> 16) & 0xffff;
 	info->pic_crop_rect.bottom = reg_val & 0xffff;
 
+	info->f_rate_numerator = vpu_read_reg(inst->dev, W5_RET_DEC_FRAME_RATE_NR);
+	info->f_rate_denominator = vpu_read_reg(inst->dev, W5_RET_DEC_FRAME_RATE_DR);
+
+	if (info->f_rate_numerator > 0 && info->f_rate_denominator >0) {
+		if (inst->std == W_HEVC_DEC)
+			info->ns_per_frame = 1000000000 * (u64)info->f_rate_denominator / (u64)info->f_rate_numerator;
+		else if (inst->std == W_AVC_DEC)
+			info->ns_per_frame = 1000000000 * 2 * (u64)info->f_rate_denominator / (u64)info->f_rate_numerator;
+		else
+			info->ns_per_frame = 1000000000 / 30; //30fps
+	} else {
+		info->ns_per_frame = 1000000000 / 30; //30fps
+	}
+
 	reg_val = vpu_read_reg(inst->dev, W5_RET_DEC_COLOR_SAMPLE_INFO);
 	info->luma_bitdepth = reg_val & 0xf;
 	info->chroma_bitdepth = (reg_val >> 4) & 0xf;
diff --git a/drivers/media/platform/chips-media/wave5/wave5-regdefine.h b/drivers/media/platform/chips-media/wave5/wave5-regdefine.h
index a15c6b2c3d8b..28c89b2f1888 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-regdefine.h
+++ b/drivers/media/platform/chips-media/wave5/wave5-regdefine.h
@@ -204,6 +204,8 @@ enum query_opt {
 #define W5_CODE_PARAM                           (W5_REG_BASE + 0x0118)
 #define W5_ADDR_TEMP_BASE                       (W5_REG_BASE + 0x011C)
 #define W5_TEMP_SIZE                            (W5_REG_BASE + 0x0120)
+#define W5_ADDR_SEC_AXI                         (W5_REG_BASE + 0x0124)
+#define W5_SEC_AXI_SIZE                         (W5_REG_BASE + 0x0128)
 #define W5_HW_OPTION                            (W5_REG_BASE + 0x012C)
 #define W5_SEC_AXI_PARAM                        (W5_REG_BASE + 0x0180)
 
diff --git a/drivers/media/platform/chips-media/wave5/wave5-vdi.c b/drivers/media/platform/chips-media/wave5/wave5-vdi.c
index 3809f70bc0b4..beae8c747b3d 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vdi.c
+++ b/drivers/media/platform/chips-media/wave5/wave5-vdi.c
@@ -11,6 +11,17 @@
 #include "wave5-regdefine.h"
 #include <linux/delay.h>
 
+extern void sifive_l2_flush64_range(unsigned long start, unsigned long len);
+extern void sifive_ccache_flush_entire(void);
+
+void wave5_flush_l2_cache(unsigned long start, unsigned long len)
+{
+	if (len >= 0x80000)
+		sifive_ccache_flush_entire();
+	else
+		sifive_l2_flush64_range(start, len);
+}
+
 static int wave5_vdi_allocate_common_memory(struct device *dev)
 {
 	struct vpu_device *vpu_dev = dev_get_drvdata(dev);
@@ -89,6 +100,7 @@ int wave5_vdi_clear_memory(struct vpu_device *vpu_dev, struct vpu_buf *vb)
 	}
 
 	memset(vb->vaddr, 0, vb->size);
+	wave5_flush_l2_cache(vb->daddr, vb->size);
 	return vb->size;
 }
 
@@ -106,6 +118,7 @@ int wave5_vdi_write_memory(struct vpu_device *vpu_dev, struct vpu_buf *vb, size_
 	}
 
 	memcpy(vb->vaddr + offset, data, len);
+	wave5_flush_l2_cache(vb->daddr + offset, len);
 
 	return len;
 }
@@ -126,6 +139,8 @@ int wave5_vdi_allocate_dma_memory(struct vpu_device *vpu_dev, struct vpu_buf *vb
 	vb->vaddr = vaddr;
 	vb->daddr = daddr;
 
+	wave5_flush_l2_cache(daddr, vb->size);
+
 	return 0;
 }
 
diff --git a/drivers/media/platform/chips-media/wave5/wave5-vdi.h b/drivers/media/platform/chips-media/wave5/wave5-vdi.h
index 3984ef3f1f96..3dbc3376c716 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vdi.h
+++ b/drivers/media/platform/chips-media/wave5/wave5-vdi.h
@@ -31,5 +31,6 @@ struct vpu_buf {
 
 int wave5_vdi_init(struct device *dev);
 int wave5_vdi_release(struct device *dev);	//this function may be called only at system off.
+void wave5_flush_l2_cache(unsigned long start, unsigned long len);
 
 #endif //#ifndef _VDI_H_
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..7a8d0ea9193c 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vpu-dec.c
+++ b/drivers/media/platform/chips-media/wave5/wave5-vpu-dec.c
@@ -33,48 +33,6 @@ static const struct vpu_format dec_fmt_list[FMT_TYPES][MAX_FMTS] = {
 	},
 	[VPU_FMT_TYPE_RAW] = {
 		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV420,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
-		},
-		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_NV12,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
-		},
-		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_NV21,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
-		},
-		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV422P,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
-		},
-		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_NV16,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
-		},
-		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_NV61,
-			.max_width = 8192,
-			.min_width = 8,
-			.max_height = 4320,
-			.min_height = 8,
-		},
-		{
 			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV420M,
 			.max_width = 8192,
 			.min_width = 8,
@@ -96,21 +54,21 @@ static const struct vpu_format dec_fmt_list[FMT_TYPES][MAX_FMTS] = {
 			.min_height = 8,
 		},
 		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV422M,
+			.v4l2_pix_fmt = V4L2_PIX_FMT_YUV420,
 			.max_width = 8192,
 			.min_width = 8,
 			.max_height = 4320,
 			.min_height = 8,
 		},
 		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_NV16M,
+			.v4l2_pix_fmt = V4L2_PIX_FMT_NV12,
 			.max_width = 8192,
 			.min_width = 8,
 			.max_height = 4320,
 			.min_height = 8,
 		},
 		{
-			.v4l2_pix_fmt = V4L2_PIX_FMT_NV61M,
+			.v4l2_pix_fmt = V4L2_PIX_FMT_NV21,
 			.max_width = 8192,
 			.min_width = 8,
 			.max_height = 4320,
@@ -187,6 +145,8 @@ static void wave5_handle_src_buffer(struct vpu_instance *inst, dma_addr_t rd_ptr
 	struct v4l2_m2m_ctx *m2m_ctx = inst->v4l2_fh.m2m_ctx;
 	struct v4l2_m2m_buffer *buf, *n;
 	size_t consumed_bytes = 0;
+	int i, j, ret;
+	u64 flag;
 
 	if (rd_ptr >= inst->last_rd_ptr) {
 		consumed_bytes = rd_ptr - inst->last_rd_ptr;
@@ -213,7 +173,33 @@ static void wave5_handle_src_buffer(struct vpu_instance *inst, dma_addr_t rd_ptr
 		dev_dbg(inst->dev->dev, "%s: removing src buffer %i",
 			__func__, src_buf->vb2_buf.index);
 		src_buf = v4l2_m2m_src_buf_remove(m2m_ctx);
-		inst->timestamp = src_buf->vb2_buf.timestamp;
+
+		if (!inst->monotonic_timestamp && !src_buf->vb2_buf.timestamp) {
+			inst->timestamp_zero_cnt++;
+			if (inst->timestamp_zero_cnt > 1) {
+				inst->monotonic_timestamp = TRUE;
+			}
+		}
+
+		if(!inst->monotonic_timestamp) {
+			ret = mutex_lock_interruptible(&inst->time_stamp.lock);
+			if (ret) {
+				dev_err(inst->dev->dev, "%s: lock err\n", __func__);
+				return;
+			}
+			inst->time_stamp.buf[inst->time_stamp.cnt] = src_buf->vb2_buf.timestamp;
+			inst->time_stamp.cnt++;
+
+			for (i = 1; i < inst->time_stamp.cnt; i++) {
+				flag = inst->time_stamp.buf[i];
+				for (j = i - 1; j >= 0 && inst->time_stamp.buf[j] < flag ; j--) {
+					inst->time_stamp.buf[j + 1] = inst->time_stamp.buf[j];
+				}
+				inst->time_stamp.buf[j + 1] = flag;
+			}
+			mutex_unlock(&inst->time_stamp.lock);
+		}
+
 		v4l2_m2m_buf_done(src_buf, VB2_BUF_STATE_DONE);
 		consumed_bytes -= src_size;
 
@@ -295,8 +281,7 @@ static void wave5_update_pix_fmt(struct v4l2_pix_format_mplane *pix_mp, unsigned
 		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);
+		pix_mp->plane_fmt[0].sizeimage = width * height;
 		break;
 	}
 }
@@ -395,8 +380,8 @@ 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->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);
 	}
@@ -415,6 +400,7 @@ static void wave5_vpu_dec_finish_decode(struct vpu_instance *inst)
 	struct vb2_v4l2_buffer *disp_buf = NULL;
 	struct vb2_queue *dst_vq = v4l2_m2m_get_dst_vq(m2m_ctx);
 	struct queue_status_info q_status;
+	u32 stride, height;
 
 	dev_dbg(inst->dev->dev, "%s: Fetch output info from firmware.", __func__);
 
@@ -446,7 +432,6 @@ static void wave5_vpu_dec_finish_decode(struct vpu_instance *inst)
 						       dec_info.index_frame_decoded);
 		if (vb) {
 			dec_buf = to_vb2_v4l2_buffer(vb);
-			dec_buf->vb2_buf.timestamp = inst->timestamp;
 		} else {
 			dev_warn(inst->dev->dev, "%s: invalid decoded frame index %i",
 				 __func__, dec_info.index_frame_decoded);
@@ -463,27 +448,47 @@ static void wave5_vpu_dec_finish_decode(struct vpu_instance *inst)
 	/* If there is anything to display, do that now */
 	if (disp_buf) {
 		struct vpu_dst_buffer *dst_vpu_buf = wave5_to_vpu_dst_buf(disp_buf);
+		stride = dec_info.disp_frame.stride;
+		height = dec_info.disp_pic_height -
+				dec_info.rc_display.bottom;
+		dev_dbg(inst->dev->dev, "%s %d disp_pic_height %u rc_display.bottom %u\n",
+				__func__, __LINE__, dec_info.disp_pic_height, dec_info.rc_display.bottom);
+		dev_dbg(inst->dev->dev, "%s %d stride %u height %u num %d\n", __func__, __LINE__, stride, height,inst->dst_fmt.num_planes);
 
 		if (inst->dst_fmt.num_planes == 1) {
 			vb2_set_plane_payload(&disp_buf->vb2_buf, 0,
-					      inst->dst_fmt.plane_fmt[0].sizeimage);
+					      (stride * height * 3 / 2));
 		} else if (inst->dst_fmt.num_planes == 2) {
 			vb2_set_plane_payload(&disp_buf->vb2_buf, 0,
-					      inst->dst_fmt.plane_fmt[0].sizeimage);
+					      (stride * height));
 			vb2_set_plane_payload(&disp_buf->vb2_buf, 1,
-					      inst->dst_fmt.plane_fmt[1].sizeimage);
+					      ((stride / 2) * height));
 		} else if (inst->dst_fmt.num_planes == 3) {
 			vb2_set_plane_payload(&disp_buf->vb2_buf, 0,
-					      inst->dst_fmt.plane_fmt[0].sizeimage);
+					      (stride * height));
 			vb2_set_plane_payload(&disp_buf->vb2_buf, 1,
-					      inst->dst_fmt.plane_fmt[1].sizeimage);
+					      ((stride / 2) * (height / 2)));
 			vb2_set_plane_payload(&disp_buf->vb2_buf, 2,
-					      inst->dst_fmt.plane_fmt[2].sizeimage);
+					      ((stride / 2) * (height / 2)));
 		}
 
 		/* TODO implement interlace support */
 		disp_buf->field = V4L2_FIELD_NONE;
 		dst_vpu_buf->display = true;
+
+		if (!inst->monotonic_timestamp) {
+			ret = mutex_lock_interruptible(&inst->time_stamp.lock);
+			if (ret) {
+				dev_err(inst->dev->dev, "%s: lock err\n", __func__);
+				return;
+			}
+			disp_buf->vb2_buf.timestamp = inst->time_stamp.buf[inst->time_stamp.cnt - 1];
+			inst->time_stamp.cnt--;
+			mutex_unlock(&inst->time_stamp.lock);
+		} else {
+			disp_buf->vb2_buf.timestamp = inst->timestamp_cnt++ * inst->codec_info->dec_info.initial_info.ns_per_frame;
+		}
+
 		v4l2_m2m_buf_done(disp_buf, VB2_BUF_STATE_DONE);
 
 		dev_dbg(inst->dev->dev, "%s: frame_cycle %8u (payload %lu)\n",
@@ -573,6 +578,7 @@ 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;
+	struct dec_initial_info *initial_info;
 	const struct vpu_format *vpu_fmt;
 	int width, height;
 
@@ -597,8 +603,10 @@ static int wave5_vpu_dec_try_fmt_cap(struct file *file, void *fh, struct v4l2_fo
 	}
 
 	if (p_dec_info->initial_info_obtained) {
+		initial_info = &inst->codec_info->dec_info.initial_info;
 		width = inst->dst_fmt.width;
-		height = inst->dst_fmt.height;
+		height = initial_info->pic_height -
+			initial_info->pic_crop_rect.top - initial_info->pic_crop_rect.bottom;
 	}
 
 	wave5_update_pix_fmt(&f->fmt.pix_mp, width, height);
@@ -975,6 +983,7 @@ static const struct v4l2_ioctl_ops wave5_vpu_dec_ioctl_ops = {
 	.vidioc_s_selection = wave5_vpu_dec_s_selection,
 
 	.vidioc_reqbufs = v4l2_m2m_ioctl_reqbufs,
+	.vidioc_create_bufs = v4l2_m2m_ioctl_create_bufs,
 	/*
 	 * Firmware does not support CREATE_BUFS for CAPTURE queue. Since
 	 * there is no immediate use-case for supporting CREATE_BUFS on
@@ -1063,7 +1072,13 @@ 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;
+		if (inst->codec_info->dec_info.initial_info.luma_bitdepth > 8 ||
+			inst->codec_info->dec_info.initial_info.chroma_bitdepth > 8) {
+			fb_stride = ALIGN(ALIGN(inst->dst_fmt.width, 16) * 5, 32) / 4;
+			fb_stride = ALIGN(fb_stride, 32);
+		} else {
+			fb_stride = inst->dst_fmt.width;
+		}
 		fb_height = ALIGN(inst->dst_fmt.height, 32);
 		luma_size = fb_stride * fb_height;
 
@@ -1144,6 +1159,9 @@ static int wave5_prepare_fb(struct vpu_instance *inst)
 		frame->stride = fb_stride;
 		frame->map_type = LINEAR_FRAME_MAP;
 		frame->update_fb_info = true;
+
+		if (frame->size < inst->dev->l2_cache_size)
+			wave5_flush_l2_cache(frame->buf_y, frame->size);
 	}
 
 	ret = wave5_vpu_dec_register_frame_buffer_ex(inst, non_linear_num, linear_num,
@@ -1232,6 +1250,9 @@ static int fill_ringbuffer(struct vpu_instance *inst)
 		dma_addr_t wr_ptr = 0;
 		size_t remain_size = 0;
 
+		if (src_size == vb2_plane_size(&vbuf->vb2_buf, 0))
+			src_size = 0;
+
 		if (vpu_buf->consumed) {
 			dev_dbg(inst->dev->dev, "already copied src buf (%u) to the ring buffer\n",
 				vbuf->vb2_buf.index);
@@ -1307,6 +1328,7 @@ static void wave5_vpu_dec_buf_queue_dst(struct vb2_buffer *vb)
 	struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
 	struct vpu_instance *inst = vb2_get_drv_priv(vb->vb2_queue);
 	struct v4l2_m2m_ctx *m2m_ctx = inst->v4l2_fh.m2m_ctx;
+	struct frame_buffer *frame_buf;
 
 	vbuf->sequence = inst->queued_dst_buf_num++;
 
@@ -1314,6 +1336,10 @@ static void wave5_vpu_dec_buf_queue_dst(struct vb2_buffer *vb)
 		struct vpu_dst_buffer *vpu_buf = wave5_to_vpu_dst_buf(vbuf);
 		int ret;
 
+		frame_buf = &inst->frame_buf[vb->index + inst->fbc_buf_count];
+		if (frame_buf->size < inst->dev->l2_cache_size)
+			wave5_flush_l2_cache(frame_buf->buf_y, frame_buf->size);
+
 		/*
 		 * The buffer is already registered just clear the display flag
 		 * to let the firmware know it can be used.
@@ -1417,7 +1443,7 @@ static int wave5_vpu_dec_start_streaming(struct vb2_queue *q, unsigned int count
 			goto return_buffers;
 
 		if (inst->state == VPU_INST_STATE_INIT_SEQ) {
-			if (initial_info->luma_bitdepth != 8) {
+			if (initial_info->luma_bitdepth != 8 && initial_info->luma_bitdepth != 10) {
 				dev_info(inst->dev->dev, "%s: no support for %d bit depth",
 					 __func__, initial_info->luma_bitdepth);
 				ret = -EINVAL;
@@ -1584,21 +1610,25 @@ static int initialize_sequence(struct vpu_instance *inst)
 	struct dec_initial_info initial_info;
 	int ret = 0;
 
+	inst->time_stamp.cnt = 0;
+	mutex_init(&inst->time_stamp.lock);
+	memset(&inst->time_stamp.buf, 0, sizeof(MAX_TIMESTAMP_CIR_BUF));
+
 	memset(&initial_info, 0, sizeof(struct dec_initial_info));
 
 	ret = wave5_vpu_dec_issue_seq_init(inst);
 	if (ret) {
-		dev_dbg(inst->dev->dev, "%s: wave5_vpu_dec_issue_seq_init, fail: %d\n",
+		dev_err(inst->dev->dev, "%s: wave5_vpu_dec_issue_seq_init, fail: %d\n",
 			__func__, ret);
 		return ret;
 	}
 
 	if (wave5_vpu_wait_interrupt(inst, VPU_DEC_TIMEOUT) < 0)
-		dev_dbg(inst->dev->dev, "%s: failed to call vpu_wait_interrupt()\n", __func__);
+		dev_err(inst->dev->dev, "%s: failed to call vpu_wait_interrupt()\n", __func__);
 
 	ret = wave5_vpu_dec_complete_seq_init(inst, &initial_info);
 	if (ret) {
-		dev_dbg(inst->dev->dev, "%s: vpu_dec_complete_seq_init, fail: %d, reason: %u\n",
+		dev_err(inst->dev->dev, "%s: vpu_dec_complete_seq_init, fail: %d, reason: %u\n",
 			__func__, ret, initial_info.seq_init_err_reason);
 		wave5_handle_src_buffer(inst, initial_info.rd_ptr);
 		return ret;
@@ -1725,6 +1755,8 @@ static void wave5_vpu_dec_job_abort(void *priv)
 	if (ret)
 		dev_warn(inst->dev->dev,
 			 "Setting EOS for the bitstream, fail: %d\n", ret);
+
+	v4l2_m2m_job_finish(inst->v4l2_m2m_dev, inst->v4l2_fh.m2m_ctx);
 }
 
 static int wave5_vpu_dec_job_ready(void *priv)
@@ -1865,8 +1897,6 @@ static int wave5_vpu_open_dec(struct file *filp)
 		goto cleanup_inst;
 	}
 
-	wave5_vdi_allocate_sram(inst->dev);
-
 	return 0;
 
 cleanup_inst:
diff --git a/drivers/media/platform/chips-media/wave5/wave5-vpu.c b/drivers/media/platform/chips-media/wave5/wave5-vpu.c
index bfe4caa79cc9..3a20bc1c0a8c 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vpu.c
+++ b/drivers/media/platform/chips-media/wave5/wave5-vpu.c
@@ -8,6 +8,7 @@
 #include <linux/module.h>
 #include <linux/platform_device.h>
 #include <linux/clk.h>
+#include <linux/reset.h>
 #include <linux/firmware.h>
 #include <linux/interrupt.h>
 #include "wave5-vpu.h"
@@ -117,11 +118,17 @@ static int wave5_vpu_load_firmware(struct device *dev, const char *fw_name,
 	return 0;
 }
 
+static const struct of_device_id sifive_l2_ids[] = {
+	{ .compatible = "sifive,fu740-c000-ccache" },
+	{ /* end of table */ },
+};
+
 static int wave5_vpu_probe(struct platform_device *pdev)
 {
 	int ret;
 	struct vpu_device *dev;
 	const struct wave5_match_data *match_data;
+	struct device_node *np;
 	u32 fw_revision;
 
 	match_data = device_get_match_data(&pdev->dev);
@@ -160,28 +167,32 @@ static int wave5_vpu_probe(struct platform_device *pdev)
 	}
 	dev->num_clks = ret;
 
+	dev->resets = devm_reset_control_array_get_exclusive(&pdev->dev);
+	if (IS_ERR(dev->resets)) {
+		dev_err(&pdev->dev, "faied to get vpu reset controls\n");
+		return -ENODEV;
+	}
+
+	dev->sram_buf.daddr = VDI_SRAM_BASE_ADDR;
+	dev->sram_buf.size = VDI_WAVE511_SRAM_SIZE;
+
 	ret = clk_bulk_prepare_enable(dev->num_clks, dev->clks);
 	if (ret) {
 		dev_err(&pdev->dev, "Enabling clocks, fail: %d\n", ret);
 		return ret;
 	}
 
-	ret = of_property_read_u32(pdev->dev.of_node, "sram-size",
-				   &dev->sram_size);
+	ret = reset_control_deassert(dev->resets);
 	if (ret) {
-		dev_warn(&pdev->dev, "sram-size not found\n");
-		dev->sram_size = 0;
+		dev_err(&pdev->dev, "Reset deassert, fail: %d\n", ret);
+		goto  err_clk_dis;
 	}
 
-	dev->sram_pool = of_gen_pool_get(pdev->dev.of_node, "sram", 0);
-	if (!dev->sram_pool)
-		dev_warn(&pdev->dev, "sram node not found\n");
-
 	dev->product_code = wave5_vdi_read_register(dev, VPU_PRODUCT_CODE_REGISTER);
 	ret = wave5_vdi_init(&pdev->dev);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "wave5_vdi_init, fail: %d\n", ret);
-		goto err_clk_dis;
+		goto err_rst_dis;
 	}
 	dev->product = wave5_vpu_get_product_id(dev);
 
@@ -227,11 +238,21 @@ static int wave5_vpu_probe(struct platform_device *pdev)
 		goto err_enc_unreg;
 	}
 
-	dev_info(&pdev->dev, "Added wave5 driver with caps: %s %s\n",
+	np = of_find_matching_node(NULL, sifive_l2_ids);
+	if (!np) {
+		dev_err(&pdev->dev, "find cache node, fail\n");
+		goto err_enc_unreg;
+	}
+
+	ret = of_property_read_u32(np, "cache-size", &dev->l2_cache_size);
+	if (ret)
+		dev->l2_cache_size = 0x200000;
+
+	dev_dbg(&pdev->dev, "Added wave5 driver with caps: %s %s\n",
 		 (match_data->flags & WAVE5_IS_ENC) ? "'ENCODE'" : "",
 		 (match_data->flags & WAVE5_IS_DEC) ? "'DECODE'" : "");
-	dev_info(&pdev->dev, "Product Code:      0x%x\n", dev->product_code);
-	dev_info(&pdev->dev, "Firmware Revision: %u\n", fw_revision);
+	dev_dbg(&pdev->dev, "Product Code:      0x%x\n", dev->product_code);
+	dev_dbg(&pdev->dev, "Firmware Revision: %u\n", fw_revision);
 	return 0;
 
 err_enc_unreg:
@@ -244,6 +265,8 @@ err_v4l2_unregister:
 	v4l2_device_unregister(&dev->v4l2_dev);
 err_vdi_release:
 	wave5_vdi_release(&pdev->dev);
+err_rst_dis:
+	reset_control_assert(dev->resets);
 err_clk_dis:
 	clk_bulk_disable_unprepare(dev->num_clks, dev->clks);
 
@@ -271,8 +294,14 @@ static const struct wave5_match_data ti_wave521c_data = {
 	.fw_name = "cnm/wave521c_k3_codec_fw.bin",
 };
 
+static const struct wave5_match_data sfdec_match_data = {
+	.flags = WAVE5_IS_DEC,
+	.fw_name = "wave511_dec_fw.bin",
+};
+
 static const struct of_device_id wave5_dt_ids[] = {
 	{ .compatible = "ti,k3-j721s2-wave521c", .data = &ti_wave521c_data },
+	{ .compatible = "starfive,vdec", .data = &sfdec_match_data },
 	{ /* sentinel */ }
 };
 MODULE_DEVICE_TABLE(of, wave5_dt_ids);
diff --git a/drivers/media/platform/chips-media/wave5/wave5-vpu.h b/drivers/media/platform/chips-media/wave5/wave5-vpu.h
index 32b7fd3730b5..fe1de3e8a22c 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vpu.h
+++ b/drivers/media/platform/chips-media/wave5/wave5-vpu.h
@@ -20,6 +20,9 @@
 #define VPU_BUF_SYNC_TO_DEVICE 0
 #define VPU_BUF_SYNC_FROM_DEVICE 1
 
+#define VDI_SRAM_BASE_ADDR          0x00000000
+#define VDI_WAVE511_SRAM_SIZE       0x2D000
+
 struct vpu_src_buffer {
 	struct v4l2_m2m_buffer	v4l2_m2m_buf;
 	struct list_head	list;
diff --git a/drivers/media/platform/chips-media/wave5/wave5-vpuapi.h b/drivers/media/platform/chips-media/wave5/wave5-vpuapi.h
index 352f6e904e50..3c70d4572455 100644
--- a/drivers/media/platform/chips-media/wave5/wave5-vpuapi.h
+++ b/drivers/media/platform/chips-media/wave5/wave5-vpuapi.h
@@ -44,6 +44,8 @@ enum vpu_instance_state {
 
 #define MAX_REG_FRAME (WAVE5_MAX_FBS * 2)
 
+#define MAX_TIMESTAMP_CIR_BUF 30
+
 #define WAVE5_DEC_HEVC_BUF_SIZE(_w, _h) (DIV_ROUND_UP(_w, 64) * DIV_ROUND_UP(_h, 64) * 256 + 64)
 #define WAVE5_DEC_AVC_BUF_SIZE(_w, _h) ((((ALIGN(_w, 256) / 16) * (ALIGN(_h, 16) / 16)) + 16) * 80)
 
@@ -366,6 +368,9 @@ struct dec_open_param {
 struct dec_initial_info {
 	u32 pic_width;
 	u32 pic_height;
+	s32 f_rate_numerator; /* the numerator part of frame rate fraction */
+	s32 f_rate_denominator; /* the denominator part of frame rate fraction */
+	u64 ns_per_frame;
 	struct vpu_rect pic_crop_rect;
 	u32 min_frame_buffer_count; /* between 1 to 16 */
 
@@ -754,8 +759,10 @@ struct vpu_device {
 	struct vpu_buf sram_buf;
 	void __iomem *vdb_register;
 	u32 product_code;
+	u32 l2_cache_size;
 	struct ida inst_ida;
 	struct clk_bulk_data *clks;
+	struct reset_control *resets;
 	int num_clks;
 };
 
@@ -765,6 +772,15 @@ struct vpu_instance_ops {
 	void (*finish_process)(struct vpu_instance *inst);
 };
 
+/* for support GStreamer ver 1.20 over
+ * too old frame, eos sent too early
+ */
+struct timestamp_circ_buf {
+	u64 buf[MAX_TIMESTAMP_CIR_BUF];
+	struct mutex lock;
+	int cnt;
+};
+
 struct vpu_instance {
 	struct list_head list;
 	struct v4l2_fh v4l2_fh;
@@ -800,6 +816,10 @@ struct vpu_instance {
 	struct list_head avail_dst_bufs;
 	struct v4l2_rect conf_win;
 	u64 timestamp;
+	struct timestamp_circ_buf time_stamp;
+	u64 timestamp_cnt;
+	u32 timestamp_zero_cnt;
+	bool monotonic_timestamp;
 	enum frame_buffer_format output_format;
 	bool cbcr_interleave;
 	bool nv21;