Re: [PATCH v2] media: mediatek: vcodec: Force capture queue format to MM21

[Nícolas F. R. A. Prado]

On Mon, Feb 27, 2023 at 02:17:08AM +0000, Yunfei Dong (董云飞) wrote:
> Hi Nicolas,
> 
> Thanks for your suggestion.
> On Wed, 2023-02-22 at 16:11 -0500, Nícolas F. R. A. Prado wrote:
> > Hi,
> > 
> > On Tue, Feb 14, 2023 at 02:28:04AM +0000, Yunfei Dong (董云飞) wrote:
> > > Hi Nicolas,
> > > 
> > > Thanks for your suggestion.
> > > On Fri, 2023-02-10 at 10:36 -0500, Nicolas Dufresne wrote:
> > > > Le vendredi 10 février 2023 à 13:55 +0800, Yunfei Dong a écrit :
[..]
> > > > > diff --git
> > > > > a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec.c
> > > > > b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec.c
> > > > > index 641f533c417f..4f5e9c20214f 100644
> > > > > --- a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec.c
> > > > > +++ b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_dec.c
> > > > > @@ -41,7 +41,7 @@ static bool mtk_vdec_get_cap_fmt(struct
> > > > > mtk_vcodec_ctx *ctx, int format_index)
> > > > >  	const struct mtk_video_fmt *fmt;
> > > > >  	struct mtk_q_data *q_data;
> > > > >  	int num_frame_count = 0, i;
> > > > > -	bool ret = true;
> > > > > +	bool ret = false;
> > 
> > This change doesn't do anything, so I'd drop it.
> > 
> This change is useful when capture fourcc is MT21 will return false,
> not support even if scp support mm21 and mt21.

But you have

	default:
		ret = true;
		break;

down below, so you're handling all cases in the switch, and the original value
doesn't matter. It could even be left uninitialized.

> > > > >  
> > > > >  	for (i = 0; i < *dec_pdata->num_formats; i++) {
> > > > >  		if (dec_pdata->vdec_formats[i].type !=
> > > > > MTK_FMT_FRAME)
> > > > > @@ -63,7 +63,7 @@ static bool mtk_vdec_get_cap_fmt(struct
> > > > > mtk_vcodec_ctx *ctx, int format_index)
> > > > >  	case V4L2_PIX_FMT_H264_SLICE:
> > > > >  	case V4L2_PIX_FMT_VP9_FRAME:
> > > > >  		if (fmt->fourcc == V4L2_PIX_FMT_MM21)
> > > > > -			ret = false;
> > > > > +			ret = true;
> > > > 
> > > > This makes the VP8 and the other cases identical, leaving
> > > > anything
> > > > that touches
> > > > MT21 as dead code. I'm not sure, cause I cannot test it, but it
> > > > should in theory
> > > > render MT8192 unusable, unless a new firmware has been submitted
> > > > to
> > > > linux-
> > > > firmware with MM21 support ?
> > > > 
> > > 
> > > If the firmware only support MT21 => won't exist for vp8 need to
> > > use
> > > MM21.
> > 
> > And that's the issue, the scp.img for MT8192 on linux-firmware only
> > supports
> > MT21 [1]. Can you please update it to support both MM21 and MT21?
> > 
> > For MT8195, only MM21 is supported in scp.img [2], but since the
> > hardware
> > supports both MM21 and MT21, the firmware should also support both.
> > So please
> > also update it on linux-firmware.
> > 
> > [1] 
> > https://lore.kernel.org/all/20230112204626.ciaff4amseoidybw@notapiano/
> > [2] 
> > https://lore.kernel.org/all/20230112205825.wb5qcqhh5kwvyi3y@notapiano/
> > 
> > Thanks,
> > Nícolas
> > 
> MT8192 always use MM21 from the beginning, MT21 have not been enabled.

That's not true, and you can verify through the following:

	$ curl https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/plain/mediatek/mt8192/scp.img > /lib/firmware/mediatek/mt8192/scp.img

	$ sha256sum /lib/firmware/mediatek/mt8192/scp.img
	fb9b4727f5e7cd82b2c63a8a2604d02e8c0f390eae2e9fd87aa1ae0797fb01b3  /lib/firmware/mediatek/mt8192/scp.img

	$ v4l2-ctl -d1 --list-formats -D
	Driver Info:
		Driver name      : mtk-vcodec-dec
		Card type        : MT8192 video decoder
		Bus info         : platform:16000000.video-codec
		Driver version   : 6.2.0
		Capabilities     : 0x84204000
			Video Memory-to-Memory Multiplanar
			Streaming
			Extended Pix Format
			Device Capabilities
		Device Caps      : 0x04204000
			Video Memory-to-Memory Multiplanar
			Streaming
			Extended Pix Format
	Media Driver Info:
		Driver name      : mtk-vcodec-dec
		Model            : mtk-vcodec-dec
		Serial           :
		Bus info         : platform:16000000.video-codec
		Media version    : 6.2.0
		Hardware revision: 0x00000000 (0)
		Driver version   : 6.2.0
	Interface Info:
		ID               : 0x0300000c
		Type             : V4L Video
	Entity Info:
		ID               : 0x00000001 (1)
		Name             : mtk-vcodec-dec-source
		Function         : V4L2 I/O
		Pad 0x01000002   : 0: Source
		 Link 0x02000008: to remote pad 0x1000004 of entity 'mtk-vcodec-dec-proc' (Video Decoder): Data, Enabled, Immutable
	ioctl: VIDIOC_ENUM_FMT
		Type: Video Capture Multiplanar

		[0]: 'MT21' (Mediatek Compressed Format, compressed)


So the decoder video node is clearly only exposing MT21 in the capture pad.

Furthermore, with

	$ echo module mtk_vcodec_dec +pmf > /sys/kernel/debug/dynamic_debug/control

and re-running the v4l2-ctl command, you'll see

	[ 3164.998781] mtk_vcodec_dec:fops_vcodec_open: fops_vcodec_open(),212: decoder capability 740

and if you check drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h:

	enum mtk_vdec_format_types {
		MTK_VDEC_FORMAT_MM21 = 0x20,
		MTK_VDEC_FORMAT_MT21C = 0x40,
		MTK_VDEC_FORMAT_H264_SLICE = 0x100,
		MTK_VDEC_FORMAT_VP8_FRAME = 0x200,
		MTK_VDEC_FORMAT_VP9_FRAME = 0x400,
		MTK_VCODEC_INNER_RACING = 0x20000,
	};

meaning the scp.img firmware is exposing the MT21C but not the MM21 format.

So, again, the mt8192 scp.img firmware needs to be fixed.

Thanks,
Nícolas