Re: [RFCv4 13/21] media: videobuf2-v4l2: support for requests

[Hans Verkuil]

On 02/23/2018 08:33 AM, Tomasz Figa wrote:
> On Fri, Feb 23, 2018 at 4:21 PM, Hans Verkuil <hverkuil@xxxxxxxxx> wrote:
>> On 02/23/2018 07:34 AM, Tomasz Figa wrote:
>>> On Wed, Feb 21, 2018 at 1:18 AM, Hans Verkuil <hverkuil@xxxxxxxxx> wrote:
>>>> On 02/20/2018 05:44 AM, Alexandre Courbot wrote:
>>>>> Add a new vb2_qbuf_request() (a request-aware version of vb2_qbuf())
>>>>> that request-aware drivers can call to queue a buffer into a request
>>>>> instead of directly into the vb2 queue if relevent.
>>>>>
>>>>> This function expects that drivers invoking it are using instances of
>>>>> v4l2_request_entity and v4l2_request_entity_data to describe their
>>>>> entity and entity data respectively.
>>>>>
>>>>> Also add the vb2_request_submit() helper function which drivers can
>>>>> invoke in order to queue all the buffers previously queued into a
>>>>> request into the regular vb2 queue.
>>>>>
>>>>> Signed-off-by: Alexandre Courbot <acourbot@xxxxxxxxxxxx>
>>>>> ---
>>>>>  .../media/common/videobuf2/videobuf2-v4l2.c   | 129 +++++++++++++++++-
>>>>>  include/media/videobuf2-v4l2.h                |  59 ++++++++
>>>>>  2 files changed, 187 insertions(+), 1 deletion(-)
>>>>>
>>>>
>>>> <snip>
>>>>
>>>>> @@ -776,10 +899,14 @@ EXPORT_SYMBOL_GPL(vb2_ioctl_querybuf);
>>>>>  int vb2_ioctl_qbuf(struct file *file, void *priv, struct v4l2_buffer *p)
>>>>>  {
>>>>>       struct video_device *vdev = video_devdata(file);
>>>>> +     struct v4l2_fh *fh = NULL;
>>>>> +
>>>>> +     if (test_bit(V4L2_FL_USES_V4L2_FH, &vdev->flags))
>>>>> +             fh = file->private_data;
>>>>
>>>> No need for this. All drivers using vb2 will also use v4l2_fh.
>>>>
>>>>>
>>>>>       if (vb2_queue_is_busy(vdev, file))
>>>>>               return -EBUSY;
>>>>> -     return vb2_qbuf(vdev->queue, p);
>>>>> +     return vb2_qbuf_request(vdev->queue, p, fh ? fh->entity : NULL);
>>>>>  }
>>>>>  EXPORT_SYMBOL_GPL(vb2_ioctl_qbuf);
>>>>>
>>>>> diff --git a/include/media/videobuf2-v4l2.h b/include/media/videobuf2-v4l2.h
>>>>> index 3d5e2d739f05..d4dfa266a0da 100644
>>>>> --- a/include/media/videobuf2-v4l2.h
>>>>> +++ b/include/media/videobuf2-v4l2.h
>>>>> @@ -23,6 +23,12 @@
>>>>>  #error VB2_MAX_PLANES != VIDEO_MAX_PLANES
>>>>>  #endif
>>>>>
>>>>> +struct media_entity;
>>>>> +struct v4l2_fh;
>>>>> +struct media_request;
>>>>> +struct media_request_entity;
>>>>> +struct v4l2_request_entity_data;
>>>>> +
>>>>>  /**
>>>>>   * struct vb2_v4l2_buffer - video buffer information for v4l2.
>>>>>   *
>>>>> @@ -116,6 +122,59 @@ int vb2_prepare_buf(struct vb2_queue *q, struct v4l2_buffer *b);
>>>>>   */
>>>>>  int vb2_qbuf(struct vb2_queue *q, struct v4l2_buffer *b);
>>>>>
>>>>> +#if IS_ENABLED(CONFIG_MEDIA_REQUEST_API)
>>>>> +
>>>>> +/**
>>>>> + * vb2_qbuf_request() - Queue a buffer, with request support
>>>>> + * @q:               pointer to &struct vb2_queue with videobuf2 queue.
>>>>> + * @b:               buffer structure passed from userspace to
>>>>> + *           &v4l2_ioctl_ops->vidioc_qbuf handler in driver
>>>>> + * @entity:  request entity to queue for if requests are used.
>>>>> + *
>>>>> + * Should be called from &v4l2_ioctl_ops->vidioc_qbuf handler of a driver.
>>>>> + *
>>>>> + * If requests are not in use, calling this is equivalent to calling vb2_qbuf().
>>>>> + *
>>>>> + * If the request_fd member of b is set, then the buffer represented by b is
>>>>> + * queued in the request instead of the vb2 queue. The buffer will be passed
>>>>> + * to the vb2 queue when the request is submitted.
>>>>
>>>> I would definitely also prepare the buffer at this time. That way you'll see any
>>>> errors relating to the prepare early on.
>>>
>>> Would the prepare operation be completely independent of other state?
>>> I can see a case when how the buffer is to be prepared may depend on
>>> values of some controls. If so, it would be only possible at request
>>> submission time. Alternatively, the application would have to be
>>> mandated to include any controls that may affect buffer preparing in
>>> the request and before the QBUF is called.
>>
>> The buffer is just memory. Controls play no role here. So the prepare
>> operation is indeed independent of other state. Anything else would make
>> this horrible complicated. And besides, with buffers allocated by other
>> subsystems (dmabuf) how could controls from our subsystems ever affect
>> those? The videobuf(2) frameworks have always just operated on memory
>> buffers without any knowledge of what it contains.
> 
> What you said applies to the videobuf(2) frameworks, but driver
> callback is explicitly defined as having access to the buffer
> contents:
> 
>  * @buf_prepare: called every time the buffer is queued from userspace
>  * and from the VIDIOC_PREPARE_BUF() ioctl; drivers may
>  * perform any initialization required before each
>  * hardware operation in this callback; drivers can
>  * access/modify the buffer here as it is still synced for
>  * the CPU; drivers that support VIDIOC_CREATE_BUFS() must
>  * also validate the buffer size; if an error is returned,
>  * the buffer will not be queued in driver; optional.

Ah, good point.

So the prepare step should be taken when the request is submitted.

You definitely want to know at submit time if all the buffers in the
request can be prepared so you can return an error.

Userspace can also prepare the buffer with VIDIOC_BUF_PREPARE before
queuing it into the request if it wants to.

Regards,

	Hans