-----Original Message-----
From: Nicolas Dufresne [mailto:nicolas@ndufresne.ca]
Sent: Friday, September 10, 2021 3:41 AM
To: Ming Qian <ming.qian@nxp.com>; mchehab@kernel.org;
shawnguo@kernel.org; robh+dt@kernel.org; s.hauer@pengutronix.de
Cc: hverkuil-cisco@xs4all.nl; kernel@pengutronix.de; festevam@gmail.com;
dl-linux-imx [off-list ref]; Aisheng Dong [off-list ref];
linux-media@vger.kernel.org; linux-kernel@vger.kernel.org;
devicetree@vger.kernel.org; linux-arm-kernel@lists.infradead.org
Subject: Re: [EXT] Re: [PATCH v8 03/15] media:Add v4l2 buf flag codec data.

Caution: EXT Email

Le jeudi 09 septembre 2021 à 02:20 +0000, Ming Qian a écrit :

quoted

quoted

-----Original Message-----
From: Nicolas Dufresne [mailto:nicolas@ndufresne.ca]
Sent: Wednesday, September 8, 2021 9:15 PM
To: Ming Qian <ming.qian@nxp.com>; mchehab@kernel.org;
shawnguo@kernel.org; robh+dt@kernel.org; s.hauer@pengutronix.de
Cc: hverkuil-cisco@xs4all.nl; kernel@pengutronix.de;
festevam@gmail.com; dl-linux-imx [off-list ref]; Aisheng Dong
[off-list ref]; linux-media@vger.kernel.org;
linux-kernel@vger.kernel.org; devicetree@vger.kernel.org;
linux-arm-kernel@lists.infradead.org
Subject: [EXT] Re: [PATCH v8 03/15] media:Add v4l2 buf flag codec data.

Caution: EXT Email

Hi Ming,

thanks for the patch. I'm doing a first pass review of the new APIs
you are introducing.

Le mardi 07 septembre 2021 à 17:49 +0800, Ming Qian a écrit :

quoted

In some decoing scenarios, application may queue a buffer that
only contains codec config data, and the driver needs to know
whether is it a frame or not.
So we add a buf flag to tell this case.

Signed-off-by: Ming Qian <ming.qian@nxp.com>
Signed-off-by: Shijie Qin <redacted>
Signed-off-by: Zhou Peng <eagle.zhou@nxp.com>
---
 Documentation/userspace-api/media/v4l/buffer.rst | 7 +++++++
 include/uapi/linux/videodev2.h                   | 1 +
 2 files changed, 8 insertions(+)

diff --git a/Documentation/userspace-api/media/v4l/buffer.rst

b/Documentation/userspace-api/media/v4l/buffer.rst
index e991ba73d873..11013bcf8a41 100644

--- a/Documentation/userspace-api/media/v4l/buffer.rst
+++ b/Documentation/userspace-api/media/v4l/buffer.rst

@@ -607,6 +607,13 @@ Buffer Flags
      the format. Any subsequent call to the
      :ref:`VIDIOC_DQBUF <VIDIOC_QBUF>` ioctl will not block

anymore,

quoted

quoted

quoted

      but return an ``EPIPE`` error code.
+    * .. _`V4L2-BUF-FLAG-CODECCONFIG`:
+
+      - ``V4L2_BUF_FLAG_CODECCONFIG``
+      - 0x00200000
+      - The buffer only contains codec config data, eg. sps and pps.
+    Applications can set this bit when ``type`` refers to an output
+    stream, this flag is usually used by v4l2 decoder.

Currently, the bottom line is that all decoders needs a frame or
field accompanied with the headers (only Annex B. being defined and
supported for now). Decoders can be more flexible (and some are).
The documentation here is not clear enough, remember that we must not

break compatibility.

quoted

quoted

I think you have to clarify the intention. This flag exist in OMX
and have been source of confusion and inter-operability since the
start.

- If this flag is entirely optional, and is just an optimization,
then adding it this way is fine, but the documentation should be
updated.

- If this flag is required only when the header is split, then this
flag need to be accompanied with a
V4L2_BUF_CAP_SUPPORTS_SPLIT_CODECCONFIG (or similar name, shorter
could be nice). This is so that userspace can detect if that feature
is supported or not.

- If having split codecconfig is strictly needed for your driver,
then this flag is not the proper solution. The only solution I'd see
for that, would be to use something else then V4L2_PIX_FMT_H264 so
that existing userspace are not tricked into trying to use your
driver the wrong way. I think such header could make sense with
H264_NO_SC (though not super clear what this is exactly, it's not
really used), or a clearer new format H264_AVCC/AVCC3

Hi Nicolas,

    This flag is optional, and in fact, our driver doesn't want to
receive a splited header, It's best that every buffer contains a
frame.
    But in some case, the client may enqueue some buffer that only
contains the header data without any frame data.
And our driver need to know this case, for this type of buffer, we
have two cases to handle.
    1. ignore the timestamp.
      2. the amphion decoder needs a ring buffer for decoding, driver
will copy the coded data to the ring buffer, and update the write
pointer, then pass a frame count to firmware, firmware will use the
frame count to determine whether starting decoding a frame or not, if
the frame count is incorrect, it may led to vpu hang. So for this type
of buffer, we won't increase the frame count.

    The flag is required only when the header is split, I agree with
you that it's better to add a capability flag. Actually our driver
doesn't want meet this case, but this situation does exist in some
applications, I add this flag to help handle it.
    Currently we meet this case in android platform.

Thanks, that clarify were this comes from. Perhaps you could drop this from
your initial patchset, and we can handle this separatly ?

OK, I will drop it from this patchset, maybe we can submit a separate patch in the future.

I remain a bit worried about the possible VPU hang, as the door is still wide
open to DoS on this HW from random streams. Have you considered raising
this issue to Amphion ? Perhaps there is a different way you could deal with
partial frames ?

Yes, I agree with you, it's a hardware limitation that the vpu may stall at decoding if there is not enough frame data in the stream buffer.
We have discussed with amphion many times and made a series of workaround in the firmware.
But the root cause is still not fixed.
And now it is almost impossible to improve it as amphion has been acquired, and they're not focused on that anymore

quoted

quoted

quoted

     * .. _`V4L2-BUF-FLAG-REQUEST-FD`:

       - ``V4L2_BUF_FLAG_REQUEST_FD`` diff --git
a/include/uapi/linux/videodev2.h b/include/uapi/linux/videodev2.h
index 167c0e40ec06..5bb0682b4a23
100644

--- a/include/uapi/linux/videodev2.h
+++ b/include/uapi/linux/videodev2.h

@@ -1119,6 +1119,7 @@ static inline __u64 v4l2_timeval_to_ns(const

struct timeval *tv)

quoted

 #define V4L2_BUF_FLAG_TSTAMP_SRC_SOE         0x00010000
 /* mem2mem encoder/decoder */
 #define V4L2_BUF_FLAG_LAST                   0x00100000
+#define V4L2_BUF_FLAG_CODECCONFIG            0x00200000
 /* request_fd is valid */
 #define V4L2_BUF_FLAG_REQUEST_FD             0x00800000

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel