Re: [PATCH v10 0/4] Media Device Allocator API

[Hans Verkuil]

On 2/1/19 1:46 AM, shuah wrote:
> Hi Hans,
> 
> On 1/30/19 12:42 AM, Hans Verkuil wrote:
>> On 1/30/19 2:50 AM, shuah wrote:
>>> On 1/29/19 2:43 AM, Hans Verkuil wrote:
>>>> On 1/29/19 12:48 AM, shuah wrote:
>>>>> Hi Hans,
>>>>>
>>>>> On 1/28/19 5:03 AM, Hans Verkuil wrote:
>>>>>> Hi Shuah,
>>>>>>
>>>>>> On 1/24/19 9:32 PM, Shuah Khan wrote:
>>>>>>> Media Device Allocator API to allows multiple drivers share a media device.
>>>>>>> This API solves a very common use-case for media devices where one physical
>>>>>>> device (an USB stick) provides both audio and video. When such media device
>>>>>>> exposes a standard USB Audio class, a proprietary Video class, two or more
>>>>>>> independent drivers will share a single physical USB bridge. In such cases,
>>>>>>> it is necessary to coordinate access to the shared resource.
>>>>>>>
>>>>>>> Using this API, drivers can allocate a media device with the shared struct
>>>>>>> device as the key. Once the media device is allocated by a driver, other
>>>>>>> drivers can get a reference to it. The media device is released when all
>>>>>>> the references are released.
>>>>>>>
>>>>>>> - This patch series is tested on 5.0-rc3 and addresses comments on
>>>>>>> ÂÂÂÂ v9 series from Hans Verkuil.
>>>>>>> - v9 was tested on 4.20-rc6.
>>>>>>> - Tested sharing resources with kaffeine, vlc, xawtv, tvtime, and
>>>>>>> ÂÂÂÂ arecord. When analog is streaming, digital and audio user-space
>>>>>>> ÂÂÂÂ applications detect that the tuner is busy and exit. When digital
>>>>>>> ÂÂÂÂ is streaming, analog and audio applications detect that the tuner is
>>>>>>> ÂÂÂÂ busy and exit. When arecord is owns the tuner, digital and analog
>>>>>>> ÂÂÂÂ detect that the tuner is busy and exit.
>>>>>>
>>>>>> I've been doing some testing with my au0828, and I am confused about one
>>>>>> thing, probably because it has been too long ago since I last looked into
>>>>>> this in detail:
>>>>>>
>>>>>
>>>>> Great.
>>>>>
>>>>>> Why can't I change the tuner frequency if arecord (and only arecord) is
>>>>>> streaming audio? If arecord is streaming, then it is recording the audio
>>>>>> from the analog TV tuner, right? So changing the analog TV frequency
>>>>>> should be fine.
>>>>>>
>>>>>
>>>>> Changing analog TV frequency would be s_frequency. The way it works is
>>>>> any s_* calls would require holding the pipeline. In Analog TV case, it
>>>>> would mean holding both audio and video pipelines for any changes
>>>>> including TV.
>>>>>
>>>>> As I recall, we discussed this design and the decision was to make all
>>>>> s_* calls interfaces to hold the tuner. A special exception is g_tuner
>>>>> in case of au0828. au0828 initializes the tuner from s_* interfaces and
>>>>> its g_tuner interfaces. Allowing s_frequency to proceed will disrupt the
>>>>> arecord audio stream.
>>>>>
>>>>> Query (q_*) works just fine without holding the pipeline. I limited the
>>>>> analog holds to just the ones that are required. The current set is
>>>>> required to avoid audio stream disruptions.
>>>>
>>>> So I am not sure about that ('avoid audio stream disruptions'): if I
>>>> stream video AND use arecord, then I can just set the frequency while
>>>> streaming. Doesn't that interrupt audio as well? And are you sure changing
>>>> the tuner frequency actually disrupts audio? And if audio is disrupted,
>>>> are we talking about a glitch or is audio permanently disrupted?
>>>
>>> I think it is a glitch. I will run some tests and let you know.
>>>>
>>>> That's basically the inconsistent behavior I noticed: just running arecord
>>>> will prevent me from changing the frequency, but if I run arecord and stream
>>>> video, then it is suddenly OK to change the frequency.
>>>
>>> How are you changing frequency? I want to duplicate what you are doing.
>>
>> v4l2-ctl -f <freq>
> 
> I am not seeing the inconsistent behavior. Here are my results.
> 
> 1. Started acecord and while it is running:
> 
> arecord -M -D plughw:2,0 -c2Â -f S16_LE -t wav foo.wav
> Recording WAVE 'foo.wav' : Signed 16 bit Little Endian, Rate 8000 Hz, Stereo
> 
> 2. Ran v4l2-ctl -f as follows:
> 
> v4l2-ctl -f 700
> VIDIOC_G_TUNER: failed: Device or resource busy
> VIDIOC_S_FREQUENCY: failed: Device or resource busy
> 
> Based on the current implementation, it failed with resource
> busy as expected.
> 
> 3. Started v4l2-ctl as follows:
> 
> Âv4l2-ctl --stream-mmap --stream-count=100 -d /dev/video0
> VIDIOC_STREAMON: failed: Device or resource busy

Why is this? You have one analog tuner and it delivers independent audio
and video streams. I should be able to start/stop audio and video independently.

And as mentioned above, if I use v4l2-ctl for video streaming, then start arecord,
then that works fine. And I can change the tuner frequency while both are streaming.

But doing this the other way around (first start arecord, then v4l2-ctl) then that
doesn't work.

It makes no sense.

Note that v4l2-ctl does not open audio, it solely deals with video. qv4l2 will
open both audio and video, but for testing audio and video independently you
need to use arecord/v4l2-ctl.

In any case, my understanding of how this should work is that both arecord and
v4l2-ctl should attempt to lock the analog tuner resource, but they can share it.

If DVB is using the tuner, then both arecord and v4l2-ctl should fail with -EBUSY.
Same if either of arecord/v4l2-ctl is running, then DVB should fail with -EBUSY.

BTW, I won't be able to test anything myself until Feb 9th since I'm abroad and
don't have access to my au0828.

Regards,

	Hans

> shuah@deneb:/mnt/data/lkml/v4l-utils/utils/v4l2-ctl$ v4l2-ctl -f 700
> VIDIOC_G_TUNER: failed: Device or resource busy
> VIDIOC_S_FREQUENCY: failed: Device or resource busy
> 
> Based on the current implementation, it failed with resource
> busy as expected.
> 
> 4. After stopping arecord:
> 
> v4l2-ctl --stream-mmap --stream-count=100 -d /dev/video0
> <<<<<<<<<<<<< 11.88 fps
> <<<<<<<<<<<<<<< 13.36 fps
> <<<<<<<<<<<<<<< 14.00 fps
> <<<<<<<<<<<<<<<< 14.35 fps
> <<<<<<<<<<<<<<< 14.57 fps
> <<<<<<<<<<<<<<<< 14.71 fps
> <<<<<<<<<
> 
> 
> Worked as expected.
> 
> 
> 5. After stopping above video streaming:
> 
> arecord -M -D plughw:2,0 -c2Â -f S16_LE -t wav foo.wav
> Recording WAVE 'foo.wav' : Signed 16 bit Little Endian, Rate 8000 Hz, Stereo
> 
> 
> Worked as expected.
> 
>>
>>>
>>>>
>>>> BTW, I think there was also inconsistent behavior in the order of streaming
>>>> audio and video: if I stream video first, then I can stream audio afterwards.
>>>> But if I stream audio first, then (if I remember correctly) I can't start
>>>> video streaming.
>>>>
> 
> Okay this is what I saw:
> 
> While v4l2-ctl --stream-mmap --stream-count=100 -d /dev/video0
> 
> I could start arecord -M -D plughw:2,0 -c2Â -f S16_LE -t wav foo.wav
> 
> I ran strace on v4l2-ctl --stream-mmap --stream-count=100 -d /dev/video0
> and I didn't see AUDIO holds. The only think of here is that
> 
> v4l2-ctl --stream-mmap --stream-count=100 -d /dev/video0
> 
> doesn't open audio device. I didn't see this when tested with other
> video apps, (tvtine, xawtv, vlc). When video is streaming, I see
> arecord failing with device busy.
> 
> As far I can see, exclusion logic is working correctly. s_freq does
> fail now based on the current logic.
> 
> btw I tested all of this on 5.0-rc4
> 
> thanks,
> -- Shuah