Re: [RFC PATCH v6 6/9] media: tegra: Add Tegra210 Video input driver

[Sowjanya Komatineni <skomatineni@nvidia.com>]

On 4/6/20 4:48 PM, Sowjanya Komatineni wrote:
>
> On 4/6/20 4:18 PM, Dmitry Osipenko wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> 07.04.2020 01:07, Sowjanya Komatineni пишет:
>>> On 4/6/20 3:00 PM, Sowjanya Komatineni wrote:
>>>> On 4/6/20 2:39 PM, Sowjanya Komatineni wrote:
>>>>> On 4/6/20 2:15 PM, Sowjanya Komatineni wrote:
>>>>>> On 4/6/20 2:11 PM, Dmitry Osipenko wrote:
>>>>>>> External email: Use caution opening links or attachments
>>>>>>>
>>>>>>>
>>>>>>> 07.04.2020 00:02, Sowjanya Komatineni пишет:
>>>>>>>>>>>>> Am I understanding correctly that this thread will take 100%
>>>>>>>>>>>>> CPU,
>>>>>>>>>>>>> spinning here, if more than 2 frame-captures queued?
>>>>>>>>>>>> on more than 2 frames captures, it breaks thread and on next
>>>>>>>>>>>> wakeup it
>>>>>>>>>>>> continues
>>>>>>>>>>> The wait_event() won't wait if condition is true.
>>>>>>>>>> condition is checked when waitqueue is woken up
>>>>>>>>> https://elixir.bootlin.com/linux/v5.6.2/source/include/linux/wait.h#L462 
>>>>>>>>>
>>>>>>>>>
>>>>>>>> process is put to sleep until the condition evaluates to true or
>>>>>>>> signal
>>>>>>>> is received.
>>>>>>>>
>>>>>>>> condition is checked each time the waitqueue head is woken up.
>>>>>>> This is a wrong assumption in accordance to the code.
>>>> process is in sleep until the condition is evaluated and when
>>>> condition is true wakeup still happens only when wake_up on waitqueue
>>>> is called
>>>>
>>>> This is the reason for using this to prevent blocking while waiting
>>>> for the buffers.
>>> w.r.t capture list update, wakeup happens when wake_up on waitqueue is
>>> called.
>>>
>>> wakeup also happens on kthread stop signal event.
>>>
>>>>
>>>>>> when every buffer is available as long as we are in streaming, we
>>>>>> should process it.
>>>>>>
>>>>>> So if wake up happens when list has buffer, it will be processed but
>>>>>> at a time we limit processing 2 simultaneous buffer capture starts
>>>>>> only.
>>>>>>
>>>>> Fixing typo.
>>>>>
>>>>> I meant when ever buffer is available as long as we are in streaming,
>>>>> we should process it.
>>>>>
>>>>> So capture thread processes as long as buffers are available from
>>>>> user space limiting 2 simultaneous trigger of captures and thread
>>>>> will be in sleep when capture buffers list is empty or no stop thread
>>>>> is signaled.
>> IIUC, the waiting won't happen if more than 2 captures are queued and
>> thread will be spinning until captures are processed.
>>
>> I think you need a semaphore with resource count = 2.
> we hold on to issuing capture if more than 2 buffers are queued and it 
> continues only after fifo has min 1 slot empty
caps_inflight gets updated based on requested frame and finished frames 
and capture will happen only for 2 frames at a time but not more