diff mbox series

[1/4] media: staging: rkisp1: cap: don't set next buffer from rkisp1_vb2_buf_queue

Message ID 20200714123832.28011-2-dafna.hirschfeld@collabora.com (mailing list archive)
State New, archived
Headers show
Series media: staging: rkisp1: fix possible race conditions in capture | expand

Commit Message

Dafna Hirschfeld July 14, 2020, 12:38 p.m. UTC
The function 'rkisp1_vb2_buf_queue' sets the next buffer directly
in case the capture is already streaming but no frame yet arrived
from the sensor. This is an optimization that tries to avoid
dropping a frame.
The call atomic_read(&cap->rkisp1->isp.frame_sequence) is used
to check if a frame arrived. Reading the 'frame_sequence' should
be avoided outside irq handlers to avoid race conditions.

This patch removes this optimization. Dropping of the first
frames can be avoided if userspace queues the buffers before
start streaming. If userspace starts queueing buffers
only after calling 'streamon' he risks frame drops anyway.

Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@collabora.com>
---
 drivers/staging/media/rkisp1/rkisp1-capture.c | 13 +------------
 1 file changed, 1 insertion(+), 12 deletions(-)

Comments

Helen Mae Koike Fornazier July 14, 2020, 12:48 p.m. UTC | #1
Hi Dafna,

On 7/14/20 9:38 AM, Dafna Hirschfeld wrote:
> The function 'rkisp1_vb2_buf_queue' sets the next buffer directly
> in case the capture is already streaming but no frame yet arrived
> from the sensor. This is an optimization that tries to avoid
> dropping a frame.
> The call atomic_read(&cap->rkisp1->isp.frame_sequence) is used
> to check if a frame arrived. Reading the 'frame_sequence' should
> be avoided outside irq handlers to avoid race conditions.
> 
> This patch removes this optimization. Dropping of the first
> frames can be avoided if userspace queues the buffers before
> start streaming. If userspace starts queueing buffers
> only after calling 'streamon' he risks frame drops anyway.
> 
> Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@collabora.com>
> ---
>  drivers/staging/media/rkisp1/rkisp1-capture.c | 13 +------------
>  1 file changed, 1 insertion(+), 12 deletions(-)
> 
> diff --git a/drivers/staging/media/rkisp1/rkisp1-capture.c b/drivers/staging/media/rkisp1/rkisp1-capture.c
> index 793ec884c894..572b0949c81f 100644
> --- a/drivers/staging/media/rkisp1/rkisp1-capture.c
> +++ b/drivers/staging/media/rkisp1/rkisp1-capture.c
> @@ -743,18 +743,7 @@ static void rkisp1_vb2_buf_queue(struct vb2_buffer *vb)
>  		     ispbuf->buff_addr[RKISP1_PLANE_CB]);
>  
>  	spin_lock_irqsave(&cap->buf.lock, flags);
> -
> -	/*
> -	 * If there's no next buffer assigned, queue this buffer directly
> -	 * as the next buffer, and update the memory interface.
> -	 */
> -	if (cap->is_streaming && !cap->buf.next &&
> -	    atomic_read(&cap->rkisp1->isp.frame_sequence) == -1) {
> -		cap->buf.next = ispbuf;
> -		rkisp1_set_next_buf(cap);

Doesn't this mean we'll always drop the first frame? Since the first user buffer will only be configured to
the hardware after receiving the first frame? So the first frame will always go to the dummy buffer, no?

Thanks
Helen

> -	} else {
> -		list_add_tail(&ispbuf->queue, &cap->buf.queue);
> -	}
> +	list_add_tail(&ispbuf->queue, &cap->buf.queue);
>  	spin_unlock_irqrestore(&cap->buf.lock, flags);
>  }
>  
>
Helen Mae Koike Fornazier July 14, 2020, 3:11 p.m. UTC | #2
On 7/14/20 9:48 AM, Helen Koike wrote:
> Hi Dafna,
> 
> On 7/14/20 9:38 AM, Dafna Hirschfeld wrote:
>> The function 'rkisp1_vb2_buf_queue' sets the next buffer directly
>> in case the capture is already streaming but no frame yet arrived
>> from the sensor. This is an optimization that tries to avoid
>> dropping a frame.
>> The call atomic_read(&cap->rkisp1->isp.frame_sequence) is used
>> to check if a frame arrived. Reading the 'frame_sequence' should
>> be avoided outside irq handlers to avoid race conditions.
>>
>> This patch removes this optimization. Dropping of the first
>> frames can be avoided if userspace queues the buffers before
>> start streaming. If userspace starts queueing buffers
>> only after calling 'streamon' he risks frame drops anyway.
>>
>> Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@collabora.com>
>> ---
>>  drivers/staging/media/rkisp1/rkisp1-capture.c | 13 +------------
>>  1 file changed, 1 insertion(+), 12 deletions(-)
>>
>> diff --git a/drivers/staging/media/rkisp1/rkisp1-capture.c b/drivers/staging/media/rkisp1/rkisp1-capture.c
>> index 793ec884c894..572b0949c81f 100644
>> --- a/drivers/staging/media/rkisp1/rkisp1-capture.c
>> +++ b/drivers/staging/media/rkisp1/rkisp1-capture.c
>> @@ -743,18 +743,7 @@ static void rkisp1_vb2_buf_queue(struct vb2_buffer *vb)
>>  		     ispbuf->buff_addr[RKISP1_PLANE_CB]);
>>  
>>  	spin_lock_irqsave(&cap->buf.lock, flags);
>> -
>> -	/*
>> -	 * If there's no next buffer assigned, queue this buffer directly
>> -	 * as the next buffer, and update the memory interface.
>> -	 */
>> -	if (cap->is_streaming && !cap->buf.next &&
>> -	    atomic_read(&cap->rkisp1->isp.frame_sequence) == -1) {
>> -		cap->buf.next = ispbuf;
>> -		rkisp1_set_next_buf(cap);
> 
> Doesn't this mean we'll always drop the first frame? Since the first user buffer will only be configured to
> the hardware after receiving the first frame? So the first frame will always go to the dummy buffer, no?

I see this is actually solved in the last patch of this series.

Maybe this can be re-order, so this patch is after 4/4.

With or without this re-ordering:

Acked-by: Helen Koike <helen.koike@collabora.com>

Thanks
Helen

> 
> Thanks
> Helen
> 
>> -	} else {
>> -		list_add_tail(&ispbuf->queue, &cap->buf.queue);
>> -	}
>> +	list_add_tail(&ispbuf->queue, &cap->buf.queue);
>>  	spin_unlock_irqrestore(&cap->buf.lock, flags);
>>  }
>>  
>>
Helen Mae Koike Fornazier July 15, 2020, 10:07 a.m. UTC | #3
On 7/14/20 12:11 PM, Helen Koike wrote:
> 
> 
> On 7/14/20 9:48 AM, Helen Koike wrote:
>> Hi Dafna,
>>
>> On 7/14/20 9:38 AM, Dafna Hirschfeld wrote:
>>> The function 'rkisp1_vb2_buf_queue' sets the next buffer directly
>>> in case the capture is already streaming but no frame yet arrived
>>> from the sensor. This is an optimization that tries to avoid
>>> dropping a frame.
>>> The call atomic_read(&cap->rkisp1->isp.frame_sequence) is used
>>> to check if a frame arrived. Reading the 'frame_sequence' should
>>> be avoided outside irq handlers to avoid race conditions.
>>>
>>> This patch removes this optimization. Dropping of the first
>>> frames can be avoided if userspace queues the buffers before
>>> start streaming. If userspace starts queueing buffers
>>> only after calling 'streamon' he risks frame drops anyway.
>>>
>>> Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@collabora.com>
>>> ---
>>>  drivers/staging/media/rkisp1/rkisp1-capture.c | 13 +------------
>>>  1 file changed, 1 insertion(+), 12 deletions(-)
>>>
>>> diff --git a/drivers/staging/media/rkisp1/rkisp1-capture.c b/drivers/staging/media/rkisp1/rkisp1-capture.c
>>> index 793ec884c894..572b0949c81f 100644
>>> --- a/drivers/staging/media/rkisp1/rkisp1-capture.c
>>> +++ b/drivers/staging/media/rkisp1/rkisp1-capture.c
>>> @@ -743,18 +743,7 @@ static void rkisp1_vb2_buf_queue(struct vb2_buffer *vb)
>>>  		     ispbuf->buff_addr[RKISP1_PLANE_CB]);
>>>  
>>>  	spin_lock_irqsave(&cap->buf.lock, flags);
>>> -
>>> -	/*
>>> -	 * If there's no next buffer assigned, queue this buffer directly
>>> -	 * as the next buffer, and update the memory interface.
>>> -	 */
>>> -	if (cap->is_streaming && !cap->buf.next &&
>>> -	    atomic_read(&cap->rkisp1->isp.frame_sequence) == -1) {
>>> -		cap->buf.next = ispbuf;
>>> -		rkisp1_set_next_buf(cap);
>>
>> Doesn't this mean we'll always drop the first frame? Since the first user buffer will only be configured to
>> the hardware after receiving the first frame? So the first frame will always go to the dummy buffer, no?
> 
> I see this is actually solved in the last patch of this series.
> 
> Maybe this can be re-order, so this patch is after 4/4.
> 
> With or without this re-ordering:

As discussed throught chat, this is not an issue, since we have a minimum amount of buffers that need
to be queued before starting the stream, so we'll never have frame_sequence == -1 (which represents the first frame)
with an empty queue, so no patch re-ordering is needed.

Regards,
Helen

> 
> Acked-by: Helen Koike <helen.koike@collabora.com>
> 
> Thanks
> Helen
> 
>>
>> Thanks
>> Helen
>>
>>> -	} else {
>>> -		list_add_tail(&ispbuf->queue, &cap->buf.queue);
>>> -	}
>>> +	list_add_tail(&ispbuf->queue, &cap->buf.queue);
>>>  	spin_unlock_irqrestore(&cap->buf.lock, flags);
>>>  }
>>>  
>>>
diff mbox series

Patch

diff --git a/drivers/staging/media/rkisp1/rkisp1-capture.c b/drivers/staging/media/rkisp1/rkisp1-capture.c
index 793ec884c894..572b0949c81f 100644
--- a/drivers/staging/media/rkisp1/rkisp1-capture.c
+++ b/drivers/staging/media/rkisp1/rkisp1-capture.c
@@ -743,18 +743,7 @@  static void rkisp1_vb2_buf_queue(struct vb2_buffer *vb)
 		     ispbuf->buff_addr[RKISP1_PLANE_CB]);
 
 	spin_lock_irqsave(&cap->buf.lock, flags);
-
-	/*
-	 * If there's no next buffer assigned, queue this buffer directly
-	 * as the next buffer, and update the memory interface.
-	 */
-	if (cap->is_streaming && !cap->buf.next &&
-	    atomic_read(&cap->rkisp1->isp.frame_sequence) == -1) {
-		cap->buf.next = ispbuf;
-		rkisp1_set_next_buf(cap);
-	} else {
-		list_add_tail(&ispbuf->queue, &cap->buf.queue);
-	}
+	list_add_tail(&ispbuf->queue, &cap->buf.queue);
 	spin_unlock_irqrestore(&cap->buf.lock, flags);
 }