Re: [RFC][PATCHv2 05/12] videobuf2: handle V4L2_FLAG_MEMORY_NON_CONSISTENT flag

[Hans Verkuil]

On 2/19/20 10:05 AM, Sergey Senozhatsky wrote:
> On (20/02/19 09:48), Hans Verkuil wrote:
> [..]
>>>  int vb2_reqbufs(struct vb2_queue *q, struct v4l2_requestbuffers *req)
>>>  {
>>>  	int ret = vb2_verify_memory_type(q, req->memory, req->type);
>>> +	bool consistent = true;
>>> +
>>> +	if (req->flags & V4L2_FLAG_MEMORY_NON_CONSISTENT)
>>> +		consistent = false;
>>
>> There is no check against allow_cache_hints: if that's 0, then
>> the V4L2_FLAG_MEMORY_NON_CONSISTENT flag should be cleared since it is
>> not supported.
> 
> The check is in set_queue_consistency()

That's the check against the functionality. I'm talking about the API level:
if !q->allow_cache_hints, then clear V4L2_FLAG_MEMORY_NON_CONSISTENT from
req->flags so that, when the ioctl returns to userspace, the application can
tell that that flag was rejected.

Regards,

	Hans

> 
> static void set_queue_consistency(struct vb2_queue *q, bool consistent_mem)
> {
> 	if (!q->allow_cache_hints)
> 		return;
> 
> 	if (consistent_mem)
> 		q->dma_attrs &= ~DMA_ATTR_NON_CONSISTENT;
> 	else
> 		q->dma_attrs |= DMA_ATTR_NON_CONSISTENT;
> }
> 
> I don't explicitly clear DMA_ATTR_NON_CONSISTENT attr for
> !->allow_cache_hints queues just in case if the driver for
> some reason sets that flag. ->allow_cache_hints is, thus,
> only for cases when user-space asks us to set or clear it.
> 
> 	-ss
>