Re: [PATCH v14.1] media: videobuf2: Be more flexible on the number of queue stored buffers

[Benjamin Gaignard <benjamin.gaignard@collabora.com>]

Le 08/11/2023 à 11:24, Tomasz Figa a écrit :
> On Mon, Nov 06, 2023 at 03:39:40PM +0100, Benjamin Gaignard wrote:
>> Add 'max_num_buffers' field in vb2_queue struct to let drivers decide
>> how many buffers could be stored in a queue.
>> This require 'bufs' array to be allocated at queue init time and freed
>> when releasing the queue.
>> By default VB2_MAX_FRAME remains the limit.
>>
>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
>> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
>> ---
>> version 14.1:
>> - Do not change the number of freed buffers in vb2_core_queue_release().
>>
>>   .../media/common/videobuf2/videobuf2-core.c   | 39 +++++++++++++++----
>>   .../media/common/videobuf2/videobuf2-v4l2.c   |  6 +--
>>   include/media/videobuf2-core.h                | 10 ++++-
>>   3 files changed, 43 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/media/common/videobuf2/videobuf2-core.c b/drivers/media/common/videobuf2/videobuf2-core.c
>> index c5c5ae4d213d..5711c6a130fd 100644
>> --- a/drivers/media/common/videobuf2/videobuf2-core.c
>> +++ b/drivers/media/common/videobuf2/videobuf2-core.c
>> @@ -416,7 +416,7 @@ static void init_buffer_cache_hints(struct vb2_queue *q, struct vb2_buffer *vb)
>>    */
>>   static void vb2_queue_add_buffer(struct vb2_queue *q, struct vb2_buffer *vb, unsigned int index)
>>   {
>> -	WARN_ON(index >= VB2_MAX_FRAME || q->bufs[index]);
>> +	WARN_ON(index >= q->max_num_buffers || q->bufs[index]);
>>   
>>   	q->bufs[index] = vb;
>>   	vb->index = index;
>> @@ -449,9 +449,9 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum vb2_memory memory,
>>   	struct vb2_buffer *vb;
>>   	int ret;
>>   
>> -	/* Ensure that q->num_buffers+num_buffers is below VB2_MAX_FRAME */
>> +	/* Ensure that the number of already queue + num_buffers is below q->max_num_buffers */
> Perhaps "the number of buffers already in the queue"?

I will do that in the next version.

>
>>   	num_buffers = min_t(unsigned int, num_buffers,
>> -			    VB2_MAX_FRAME - q_num_buffers);
>> +			    q->max_num_buffers - q_num_buffers);
>>   
>>   	for (buffer = 0; buffer < num_buffers; ++buffer) {
>>   		/* Allocate vb2 buffer structures */
>> @@ -813,7 +813,7 @@ int vb2_core_reqbufs(struct vb2_queue *q, enum vb2_memory memory,
>>   	unsigned plane_sizes[VB2_MAX_PLANES] = { };
>>   	bool non_coherent_mem = flags & V4L2_MEMORY_FLAG_NON_COHERENT;
>>   	unsigned int i;
>> -	int ret;
>> +	int ret = 0;
>>   
>>   	if (q->streaming) {
>>   		dprintk(q, 1, "streaming active\n");
>> @@ -857,17 +857,22 @@ int vb2_core_reqbufs(struct vb2_queue *q, enum vb2_memory memory,
>>   	/*
>>   	 * Make sure the requested values and current defaults are sane.
>>   	 */
>> -	WARN_ON(q->min_buffers_needed > VB2_MAX_FRAME);
> Do we really want to remove this warning completely?

Yes because VB2_MAX_FRAME is no more relevant.

>
>>   	num_buffers = max_t(unsigned int, *count, q->min_buffers_needed);
>> -	num_buffers = min_t(unsigned int, num_buffers, VB2_MAX_FRAME);
>> +	num_buffers = min_t(unsigned int, num_buffers, q->max_num_buffers);
>>   	memset(q->alloc_devs, 0, sizeof(q->alloc_devs));
>>   	/*
>>   	 * Set this now to ensure that drivers see the correct q->memory value
>>   	 * in the queue_setup op.
>>   	 */
>>   	mutex_lock(&q->mmap_lock);
>> +	if (!q->bufs)
>> +		q->bufs = kcalloc(q->max_num_buffers, sizeof(*q->bufs), GFP_KERNEL);
> Shouldn't this happen in core code rather than the v4l2-specific ioctl
> helper? Since we just allocate the maximum possible size, then maybe
> vb2_core_queue_init()?

Hans had already suggest that in a previous version but it appear that
vb2_core_queue_init() and vb2_core_queue_release() aren't balanced so
we got cases where queue aren't initialized before reqbufs or create_bufs
that why I had to put this allocation here.

>
>> +	if (!q->bufs)
>> +		ret = -ENOMEM;
>>   	q->memory = memory;
>>   	mutex_unlock(&q->mmap_lock);
>> +	if (ret)
>> +		return ret;
>>   	set_queue_coherency(q, non_coherent_mem);
>>   
>>   	/*
>> @@ -976,7 +981,7 @@ int vb2_core_create_bufs(struct vb2_queue *q, enum vb2_memory memory,
>>   	bool no_previous_buffers = !q_num_bufs;
>>   	int ret = 0;
>>   
>> -	if (q_num_bufs == VB2_MAX_FRAME) {
>> +	if (q->num_buffers == q->max_num_buffers) {
>>   		dprintk(q, 1, "maximum number of buffers already allocated\n");
>>   		return -ENOBUFS;
>>   	}
>> @@ -993,7 +998,13 @@ int vb2_core_create_bufs(struct vb2_queue *q, enum vb2_memory memory,
>>   		 */
>>   		mutex_lock(&q->mmap_lock);
>>   		q->memory = memory;
>> +		if (!q->bufs)
>> +			q->bufs = kcalloc(q->max_num_buffers, sizeof(*q->bufs), GFP_KERNEL);
> Ditto.
>
>> +		if (!q->bufs)
>> +			ret = -ENOMEM;
>>   		mutex_unlock(&q->mmap_lock);
>> +		if (ret)
>> +			return ret;
>>   		q->waiting_for_buffers = !q->is_output;
>>   		set_queue_coherency(q, non_coherent_mem);
>>   	} else {
>> @@ -1005,7 +1016,7 @@ int vb2_core_create_bufs(struct vb2_queue *q, enum vb2_memory memory,
>>   			return -EINVAL;
>>   	}
>>   
>> -	num_buffers = min(*count, VB2_MAX_FRAME - q_num_bufs);
>> +	num_buffers = min(*count, q->max_num_buffers - q_num_bufs);
>>   
>>   	if (requested_planes && requested_sizes) {
>>   		num_planes = requested_planes;
>> @@ -2465,6 +2476,12 @@ int vb2_core_queue_init(struct vb2_queue *q)
>>   	/*
>>   	 * Sanity check
>>   	 */
>> +	if (!q->max_num_buffers)
>> +		q->max_num_buffers = VB2_MAX_FRAME;
> Can we add a comment here to explain that this is for backwards
> compatibility with drivers which don't support more buffers?
>
> Actually, we should probably document in kerneldoc for vb2_queue that 0 is
> an allowed and special value.

I will do that.

>
>> +
>> +	/* The maximum is limited by offset cookie encoding pattern */
>> +	q->max_num_buffers = min_t(unsigned int, q->max_num_buffers, MAX_BUFFER_INDEX);
>> +
>>   	if (WARN_ON(!q)			  ||
>>   	    WARN_ON(!q->ops)		  ||
>>   	    WARN_ON(!q->mem_ops)	  ||
>> @@ -2474,6 +2491,10 @@ int vb2_core_queue_init(struct vb2_queue *q)
>>   	    WARN_ON(!q->ops->buf_queue))
>>   		return -EINVAL;
>>   
>> +	if (WARN_ON(q->max_num_buffers > MAX_BUFFER_INDEX) ||
> Hmm, how is this possible?

MAX_BUFFER_INDEX depends on PAGE_SHIFT and, on some architectures,
it can goes up to 15. In this MAX_BUFFER_INDEX is only equal to 512,
that why this check in needed.

>
>> +	    WARN_ON(q->min_buffers_needed > q->max_num_buffers))
>> +		return -EINVAL;
> I have a loose recollection that it's allowed for a driver to change this
> value depending on the configuration. You may want to double check if any
> driver doesn't do so already if we want to disallow that. (and also
> document that it's not allowed)

I don't think any driver change is value given the configuration but Hans wants
to clarify the usage of this field on another series.

>
>> +
>>   	if (WARN_ON(q->requires_requests && !q->supports_requests))
>>   		return -EINVAL;
>>   
>> @@ -2520,6 +2541,8 @@ void vb2_core_queue_release(struct vb2_queue *q)
>>   	__vb2_queue_cancel(q);
>>   	mutex_lock(&q->mmap_lock);
>>   	__vb2_queue_free(q, vb2_get_num_buffers(q));
>> +	kfree(q->bufs);
>> +	q->bufs = NULL;
>>   	q->num_buffers = 0;
>>   	mutex_unlock(&q->mmap_lock);
>>   }
>> diff --git a/drivers/media/common/videobuf2/videobuf2-v4l2.c b/drivers/media/common/videobuf2/videobuf2-v4l2.c
>> index 7d798fb15c0b..f3cf4b235c1f 100644
>> --- a/drivers/media/common/videobuf2/videobuf2-v4l2.c
>> +++ b/drivers/media/common/videobuf2/videobuf2-v4l2.c
>> @@ -627,7 +627,7 @@ struct vb2_buffer *vb2_find_buffer(struct vb2_queue *q, u64 timestamp)
>>   	 * This loop doesn't scale if there is a really large number of buffers.
>>   	 * Maybe something more efficient will be needed in this case.
>>   	 */
>> -	for (i = 0; i < vb2_get_num_buffers(q); i++) {
>> +	for (i = 0; i < q->max_num_buffers; i++) {
>>   		vb2 = vb2_get_buffer(q, i);
>>   
>>   		if (!vb2)
>> @@ -1142,7 +1142,7 @@ int _vb2_fop_release(struct file *file, struct mutex *lock)
>>   
>>   	if (lock)
>>   		mutex_lock(lock);
>> -	if (file->private_data == vdev->queue->owner) {
>> +	if (!vdev->queue->owner || file->private_data == vdev->queue->owner) {
>>   		vb2_queue_release(vdev->queue);
>>   		vdev->queue->owner = NULL;
>>   	}
>> @@ -1270,7 +1270,7 @@ void vb2_video_unregister_device(struct video_device *vdev)
>>   	 */
>>   	get_device(&vdev->dev);
>>   	video_unregister_device(vdev);
>> -	if (vdev->queue && vdev->queue->owner) {
>> +	if (vdev->queue) {
>>   		struct mutex *lock = vdev->queue->lock ?
>>   			vdev->queue->lock : vdev->lock;
>>   
>> diff --git a/include/media/videobuf2-core.h b/include/media/videobuf2-core.h
>> index 8f9d9e4af5b1..e77a397195f2 100644
>> --- a/include/media/videobuf2-core.h
>> +++ b/include/media/videobuf2-core.h
>> @@ -558,6 +558,7 @@ struct vb2_buf_ops {
>>    * @dma_dir:	DMA mapping direction.
>>    * @bufs:	videobuf2 buffer structures
>>    * @num_buffers: number of allocated/used buffers
>> + * @max_num_buffers: upper limit of number of allocated/used buffers
>>    * @queued_list: list of buffers currently queued from userspace
>>    * @queued_count: number of buffers queued and ready for streaming.
>>    * @owned_by_drv_count: number of buffers owned by the driver
>> @@ -619,8 +620,9 @@ struct vb2_queue {
>>   	struct mutex			mmap_lock;
>>   	unsigned int			memory;
>>   	enum dma_data_direction		dma_dir;
>> -	struct vb2_buffer		*bufs[VB2_MAX_FRAME];
>> +	struct vb2_buffer		**bufs;
>>   	unsigned int			num_buffers;
>> +	unsigned int			max_num_buffers;
>>   
>>   	struct list_head		queued_list;
>>   	unsigned int			queued_count;
>> @@ -1248,6 +1250,12 @@ static inline void vb2_clear_last_buffer_dequeued(struct vb2_queue *q)
>>   static inline struct vb2_buffer *vb2_get_buffer(struct vb2_queue *q,
>>   						unsigned int index)
>>   {
>> +	if (!q->bufs)
>> +		return NULL;
>> +
>> +	if (index >= q->max_num_buffers)
> Wouldn't this be already prevented by the condition below?

yes but the series will remove q->num_buffers after this patch
so for me it make sense to introduce this check now.

Regards,
Benjamin

>
>> +		return NULL;
>> +
>>   	if (index < q->num_buffers)
>>   		return q->bufs[index];
>>   	return NULL;
>> -- 
>> 2.39.2
>>
> Best regards,
> Tomasz