Some cavlc bistream will decode fail when the frame size is small than
20 bytes. Need to add pending data at the end of the bitstream.

For the minimum size of mapped memory is 256 bytes(16x16), adding four bytes data
won't lead to access unknown virtual memory.

Fixes: 59fba9eed5a7 ("media: mediatek: vcodec: support stateless H.264 decoding for mt8192")
Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
---
compared with v1:
- add detail comments for function: vdec_h264_insert_startcode.
- re-write commit message.
---
 .../vcodec/vdec/vdec_h264_req_multi_if.c      | 32 +++++++++++++++++--
 1 file changed, 29 insertions(+), 3 deletions(-)

diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
index 4cc92700692b..18e048755d11 100644
--- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
+++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
@@ -539,6 +539,29 @@ static int vdec_h264_slice_core_decode(struct vdec_lat_buf *lat_buf)
 	return 0;
 }
 
+static void vdec_h264_insert_startcode(struct mtk_vcodec_dev *vcodec_dev, unsigned char *buf,
+				       size_t *bs_size, struct mtk_h264_pps_param *pps)
+{
+	struct device *dev = &vcodec_dev->plat_dev->dev;
+
+	/* Need to add pending data at the end of bitstream when bs_sz is small than
+	 * 20 bytes for cavlc bitstream, or lat will decode fail. This pending data is
+	 * useful for mt8192 and mt8195 platform.
+	 *
+	 * cavlc bitstream when entropy_coding_mode_flag is false.
+	 */
+	if (pps->entropy_coding_mode_flag || *bs_size > 20 ||
+	    !(of_device_is_compatible(dev->of_node, "mediatek,mt8192-vcodec-dec") ||
+	    of_device_is_compatible(dev->of_node, "mediatek,mt8195-vcodec-dec")))
+		return;
+
+	buf[*bs_size] = 0;
+	buf[*bs_size + 1] = 0;
+	buf[*bs_size + 2] = 1;
+	buf[*bs_size + 3] = 0xff;
+	(*bs_size) += 4;
+}
+
 static int vdec_h264_slice_lat_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
 				      struct vdec_fb *fb, bool *res_chg)
 {
@@ -582,9 +605,6 @@ static int vdec_h264_slice_lat_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
 	}
 
 	inst->vsi->dec.nal_info = buf[nal_start_idx];
-	inst->vsi->dec.bs_buf_addr = (u64)bs->dma_addr;
-	inst->vsi->dec.bs_buf_size = bs->size;
-
 	lat_buf->src_buf_req = src_buf_info->m2m_buf.vb.vb2_buf.req_obj.req;
 	v4l2_m2m_buf_copy_metadata(&src_buf_info->m2m_buf.vb, &lat_buf->ts_info, true);
 
@@ -592,6 +612,12 @@ static int vdec_h264_slice_lat_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
 	if (err)
 		goto err_free_fb_out;
 
+	vdec_h264_insert_startcode(inst->ctx->dev, buf, &bs->size,
+				   &share_info->h264_slice_params.pps);
+
+	inst->vsi->dec.bs_buf_addr = (uint64_t)bs->dma_addr;
+	inst->vsi->dec.bs_buf_size = bs->size;
+
 	*res_chg = inst->resolution_changed;
 	if (inst->resolution_changed) {
 		mtk_vcodec_debug(inst, "- resolution changed -");
-- 
2.25.1

Il 18/10/22 13:41, Yunfei Dong ha scritto:
> Some cavlc bistream will decode fail when the frame size is small than

s/small/smaller/g

> 20 bytes. Need to add pending data at the end of the bitstream.
> 
> For the minimum size of mapped memory is 256 bytes(16x16), adding four bytes data
> won't lead to access unknown virtual memory.
> 
> Fixes: 59fba9eed5a7 ("media: mediatek: vcodec: support stateless H.264 decoding for mt8192")
> Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
> ---
> compared with v1:
> - add detail comments for function: vdec_h264_insert_startcode.
> - re-write commit message.
> ---
>   .../vcodec/vdec/vdec_h264_req_multi_if.c      | 32 +++++++++++++++++--
>   1 file changed, 29 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
> index 4cc92700692b..18e048755d11 100644
> --- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
> +++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
> @@ -539,6 +539,29 @@ static int vdec_h264_slice_core_decode(struct vdec_lat_buf *lat_buf)
>   	return 0;
>   }
>   
> +static void vdec_h264_insert_startcode(struct mtk_vcodec_dev *vcodec_dev, unsigned char *buf,
> +				       size_t *bs_size, struct mtk_h264_pps_param *pps)
> +{
> +	struct device *dev = &vcodec_dev->plat_dev->dev;
> +
> +	/* Need to add pending data at the end of bitstream when bs_sz is small than
> +	 * 20 bytes for cavlc bitstream, or lat will decode fail. This pending data is
> +	 * useful for mt8192 and mt8195 platform.

What is the reason why other SoCs don't need this?

> +	 *
> +	 * cavlc bitstream when entropy_coding_mode_flag is false.
> +	 */
> +	if (pps->entropy_coding_mode_flag || *bs_size > 20 ||
> +	    !(of_device_is_compatible(dev->of_node, "mediatek,mt8192-vcodec-dec") ||
> +	    of_device_is_compatible(dev->of_node, "mediatek,mt8195-vcodec-dec")))

I'm not comfortable seeing of_device_is_compatible... this list will grow whenever
a new SoC needing this appears.
Please think about a good name for a flag/quirk, or a bool, in the platform data
for these two SoCs and use it.

Regards,
Angelo

Hi AngeloGioacchino,

Thanks for your suggestion.

Fixed in patch v3, could you please help to review it again.
For this flag only be used in H264, add the flag in H264 instance.

Best Regards,
Yunfei Dong 

On Mon, 2022-11-14 at 12:08 +0100, AngeloGioacchino Del Regno wrote:
> Il 18/10/22 13:41, Yunfei Dong ha scritto:
> > Some cavlc bistream will decode fail when the frame size is small
> > than
> 
> s/small/smaller/g
> 
> > 20 bytes. Need to add pending data at the end of the bitstream.
> > 
> > For the minimum size of mapped memory is 256 bytes(16x16), adding
> > four bytes data
> > won't lead to access unknown virtual memory.
> > 
> > Fixes: 59fba9eed5a7 ("media: mediatek: vcodec: support stateless
> > H.264 decoding for mt8192")
> > Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
> > ---
> > compared with v1:
> > - add detail comments for function: vdec_h264_insert_startcode.
> > - re-write commit message.
> > ---
> >   .../vcodec/vdec/vdec_h264_req_multi_if.c      | 32
> > +++++++++++++++++--
> >   1 file changed, 29 insertions(+), 3 deletions(-)
> > 
> > diff --git
> > a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > index 4cc92700692b..18e048755d11 100644
> > ---
> > a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > +++
> > b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > @@ -539,6 +539,29 @@ static int vdec_h264_slice_core_decode(struct
> > vdec_lat_buf *lat_buf)
> >   	return 0;
> >   }
> >   
> > +static void vdec_h264_insert_startcode(struct mtk_vcodec_dev
> > *vcodec_dev, unsigned char *buf,
> > +				       size_t *bs_size, struct
> > mtk_h264_pps_param *pps)
> > +{
> > +	struct device *dev = &vcodec_dev->plat_dev->dev;
> > +
> > +	/* Need to add pending data at the end of bitstream when bs_sz
> > is small than
> > +	 * 20 bytes for cavlc bitstream, or lat will decode fail. This
> > pending data is
> > +	 * useful for mt8192 and mt8195 platform.
> 
> What is the reason why other SoCs don't need this?
> 
> > +	 *
> > +	 * cavlc bitstream when entropy_coding_mode_flag is false.
> > +	 */
> > +	if (pps->entropy_coding_mode_flag || *bs_size > 20 ||
> > +	    !(of_device_is_compatible(dev->of_node, "mediatek,mt8192-
> > vcodec-dec") ||
> > +	    of_device_is_compatible(dev->of_node, "mediatek,mt8195-
> > vcodec-dec")))
> 
> I'm not comfortable seeing of_device_is_compatible... this list will
> grow whenever
> a new SoC needing this appears.
> Please think about a good name for a flag/quirk, or a bool, in the
> platform data
> for these two SoCs and use it.
> 
> Regards,
> Angelo
> 
> 

Hi AngeloGioacchino,

Thanks for your detail suggestion.
On Mon, 2022-11-14 at 12:08 +0100, AngeloGioacchino Del Regno wrote:
> Il 18/10/22 13:41, Yunfei Dong ha scritto:
> > Some cavlc bistream will decode fail when the frame size is small
> > than
> 
> s/small/smaller/g

Will fix in next patch.
> 
> > 20 bytes. Need to add pending data at the end of the bitstream.
> > 
> > For the minimum size of mapped memory is 256 bytes(16x16), adding
> > four bytes data
> > won't lead to access unknown virtual memory.
> > 
> > Fixes: 59fba9eed5a7 ("media: mediatek: vcodec: support stateless
> > H.264 decoding for mt8192")
> > Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
> > ---
> > compared with v1:
> > - add detail comments for function: vdec_h264_insert_startcode.
> > - re-write commit message.
> > ---
> >   .../vcodec/vdec/vdec_h264_req_multi_if.c      | 32
> > +++++++++++++++++--
> >   1 file changed, 29 insertions(+), 3 deletions(-)
> > 
> > diff --git
> > a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > index 4cc92700692b..18e048755d11 100644
> > ---
> > a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > +++
> > b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
> > f.c
> > @@ -539,6 +539,29 @@ static int vdec_h264_slice_core_decode(struct
> > vdec_lat_buf *lat_buf)
> >   	return 0;
> >   }
> >   
> > +static void vdec_h264_insert_startcode(struct mtk_vcodec_dev
> > *vcodec_dev, unsigned char *buf,
> > +				       size_t *bs_size, struct
> > mtk_h264_pps_param *pps)
> > +{
> > +	struct device *dev = &vcodec_dev->plat_dev->dev;
> > +
> > +	/* Need to add pending data at the end of bitstream when bs_sz
> > is small than
> > +	 * 20 bytes for cavlc bitstream, or lat will decode fail. This
> > pending data is
> > +	 * useful for mt8192 and mt8195 platform.
> 
> What is the reason why other SoCs don't need this?
> 
For the hardware not add this feature, and will add in the future Soc.
> > +	 *
> > +	 * cavlc bitstream when entropy_coding_mode_flag is false.
> > +	 */
> > +	if (pps->entropy_coding_mode_flag || *bs_size > 20 ||
> > +	    !(of_device_is_compatible(dev->of_node, "mediatek,mt8192-
> > vcodec-dec") ||
> > +	    of_device_is_compatible(dev->of_node, "mediatek,mt8195-
> > vcodec-dec")))
> 
> I'm not comfortable seeing of_device_is_compatible... this list will
> grow whenever
> a new SoC needing this appears.
> Please think about a good name for a flag/quirk, or a bool, in the
> platform data
> for these two SoCs and use it.
> 
For this feature only need to add in these two Socs, and won't grow
anymore. So just want to use compatible to separate, not add one flags.

So you think that using one flag to separate much better?

Best Regards,
Yunfei Dong

> Regards,
> Angelo
> 
> 

Il 15/11/22 03:00, Yunfei Dong (董云飞) ha scritto:
> Hi AngeloGioacchino,
> 
> Thanks for your detail suggestion.
> On Mon, 2022-11-14 at 12:08 +0100, AngeloGioacchino Del Regno wrote:
>> Il 18/10/22 13:41, Yunfei Dong ha scritto:
>>> Some cavlc bistream will decode fail when the frame size is small
>>> than
>>
>> s/small/smaller/g
> 
> Will fix in next patch.
>>
>>> 20 bytes. Need to add pending data at the end of the bitstream.
>>>
>>> For the minimum size of mapped memory is 256 bytes(16x16), adding
>>> four bytes data
>>> won't lead to access unknown virtual memory.
>>>
>>> Fixes: 59fba9eed5a7 ("media: mediatek: vcodec: support stateless
>>> H.264 decoding for mt8192")
>>> Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
>>> ---
>>> compared with v1:
>>> - add detail comments for function: vdec_h264_insert_startcode.
>>> - re-write commit message.
>>> ---
>>>    .../vcodec/vdec/vdec_h264_req_multi_if.c      | 32
>>> +++++++++++++++++--
>>>    1 file changed, 29 insertions(+), 3 deletions(-)
>>>
>>> diff --git
>>> a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
>>> f.c
>>> b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
>>> f.c
>>> index 4cc92700692b..18e048755d11 100644
>>> ---
>>> a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
>>> f.c
>>> +++
>>> b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_i
>>> f.c
>>> @@ -539,6 +539,29 @@ static int vdec_h264_slice_core_decode(struct
>>> vdec_lat_buf *lat_buf)
>>>    	return 0;
>>>    }
>>>    
>>> +static void vdec_h264_insert_startcode(struct mtk_vcodec_dev
>>> *vcodec_dev, unsigned char *buf,
>>> +				       size_t *bs_size, struct
>>> mtk_h264_pps_param *pps)
>>> +{
>>> +	struct device *dev = &vcodec_dev->plat_dev->dev;
>>> +
>>> +	/* Need to add pending data at the end of bitstream when bs_sz
>>> is small than
>>> +	 * 20 bytes for cavlc bitstream, or lat will decode fail. This
>>> pending data is
>>> +	 * useful for mt8192 and mt8195 platform.
>>
>> What is the reason why other SoCs don't need this?
>>
> For the hardware not add this feature, and will add in the future Soc.
>>> +	 *
>>> +	 * cavlc bitstream when entropy_coding_mode_flag is false.
>>> +	 */
>>> +	if (pps->entropy_coding_mode_flag || *bs_size > 20 ||
>>> +	    !(of_device_is_compatible(dev->of_node, "mediatek,mt8192-
>>> vcodec-dec") ||
>>> +	    of_device_is_compatible(dev->of_node, "mediatek,mt8195-
>>> vcodec-dec")))
>>
>> I'm not comfortable seeing of_device_is_compatible... this list will
>> grow whenever
>> a new SoC needing this appears.
>> Please think about a good name for a flag/quirk, or a bool, in the
>> platform data
>> for these two SoCs and use it.
>>
> For this feature only need to add in these two Socs, and won't grow
> anymore. So just want to use compatible to separate, not add one flags.
> 
> So you think that using one flag to separate much better?
> 

A flag is better: please remember that calls to of_device_is_compatible()
will perform a string comparison which, as you know, as much optimized as
it can be, it's always going to be slower than a simple integer/bool/flag
check.

This means that even for functional (not just cosmetic) reasons we should
not use of_device_is_compatible() here :-)

Cheers,
Angelo

On Tue, Oct 18, 2022 at 7:41 PM Yunfei Dong <yunfei.dong@mediatek.com> wrote:
>
> Some cavlc bistream will decode fail when the frame size is small than
> 20 bytes. Need to add pending data at the end of the bitstream.
>
> For the minimum size of mapped memory is 256 bytes(16x16), adding four bytes data
> won't lead to access unknown virtual memory.
>
> Fixes: 59fba9eed5a7 ("media: mediatek: vcodec: support stateless H.264 decoding for mt8192")
> Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>

Reviewed-by: Chen-Yu Tsai <wenst@chromium.org>