Re: [PATCH v6 2/2] media: platform: Add Aspeed Video Engine driver

From: Hans Verkuil
Date: Mon Dec 03 2018 - 16:09:44 EST


On 12/03/2018 09:37 PM, Eddie James wrote:

<snip>

>>>>> +static int aspeed_video_start(struct aspeed_video *video)
>>>>> +{
>>>>> + int rc;
>>>>> +
>>>>> + aspeed_video_on(video);
>>>>> +
>>>>> + aspeed_video_init_regs(video);
>>>>> +
>>>>> + rc = aspeed_video_get_resolution(video);
>>>>> + if (rc)
>>>>> + return rc;
>>>>> +
>>>>> + /*
>>>>> + * Set the timings here since the device was just opened for the first
>>>>> + * time.
>>>>> + */
>>>>> + video->active_timings = video->detected_timings;
>>>> What happens if no valid signal was detected?
>>>>
>>>> My recommendation is to fallback to some default timings (VGA?) if no valid
>>>> initial timings were found.
>>>>
>>>> The expectation is that applications will always call QUERY_DV_TIMINGS first,
>>>> so it is really not all that important what the initial active_timings are,
>>>> as long as they are valid timings (valid as in: something that the hardware
>>>> can support).
>>> See just above, this call returns with a failure if no signal is
>>> detected, meaning the device cannot be opened. The only valid timings
>>> are the detected timings.
>> That's wrong. You must ALWAYS be able to open the device. If not valid
>> resolution is detected, just fallback to some default.
>
> Why must open always succeed? What use is a video device that cannot
> provide any video?
You always must be able to open the video device so applications can call
QUERYCAP. In fact, any ioctl that returns state information (G_FMT, G_CTRL,
G_INPUT, ENUM_*, etc) can always be called, regardless of whether there is
a video signal or if video streaming is in progress.

With this restriction I cannot even run an application that waits for the
SOURCE_CHANGE event to start streaming, such as 'v4l2-ctl --stream-mmap'
does because the open() will fail immediately.

Sorry, this is really wrong.

Regards,

Hans