On Thu, Jan 02, 2025 at 06:35:00PM +0300, Mikhail Rudenko wrote:
> Currently, the rkisp1 driver always uses coherent DMA allocations for
> video capture buffers. However, on some platforms, using non-coherent
> buffers can improve performance, especially when CPU processing of
> MMAP'ed video buffers is required.
>
> For example, on the Rockchip RK3399 running at maximum CPU frequency,
> the time to memcpy a frame from a 1280x720 XRGB32 MMAP'ed buffer to a
> malloc'ed userspace buffer decreases from 7.7 ms to 1.1 ms when using
> non-coherent DMA allocation. CPU usage also decreases accordingly.
What's the time taken by the cache management operations ?
> This change allows userspace to request the allocation of non-coherent
> buffers. Note that the behavior for existing users will remain unchanged
> unless they explicitly set the V4L2_MEMORY_FLAG_NON_COHERENT flag when
> allocating buffers.
>
> Signed-off-by: Mikhail Rudenko <mike.rudenko@xxxxxxxxx>
> ---
> drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> index 6dcefd144d5abe358323e37ac6133c6134ac636e..c94f7d1d73a92646457a27da20726ec6f92e7717 100644
> --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> @@ -1563,6 +1563,7 @@ static int rkisp1_register_capture(struct rkisp1_capture *cap)
> q->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC;
> q->lock = &node->vlock;
> q->dev = cap->rkisp1->dev;
> + q->allow_cache_hints = 1;
> ret = vb2_queue_init(q);
> if (ret) {
> dev_err(cap->rkisp1->dev,
>
> ---
> base-commit: 40ed9e9b2808beeb835bd0ed971fb364c285d39c
> change-id: 20241231-b4-rkisp-noncoherent-ad6e7c7a68ba
--
Regards,
Laurent Pinchart
