Re: [PATCH v4 2/2] videobuf2-dma-sg: Replace vb2_dma_sg_desc with sg_table

[Marek Szyprowski <m.szyprowski@samsung.com>]

Hello All,

On 2014-01-03 16:51, Ricardo Ribalda Delgado wrote:
> Hello Hans
>
> What if we move the dma_map_sg and dma_unmap_sg to the vb2 interface,
> and there do something like:
>
> n_sg= dma_map_sg()
> if (n_sg=-ENOMEM){
>     split_table() //Breaks down the sg_table into monopages sg
>     n_sg= dma_map_sg()
> }
> if (n_sg=-ENOMEM)
>    return -ENOMEM

dma_map_sg/dma_unmap_sg should be moved to vb2-dma-sg memory allocator. 
The best place for calling them is buf_prepare() and buf_finish() 
callbacks. I think that I've already pointed this some time ago, but 
unfortunately I didn't find enough time to convert existing code.

For solving the problem described by Hans, I think that vb2-dma-sg 
memory allocator should check dma mask of the client device and add 
appropriate GFP_DMA or GFP_DMA32 flags to alloc_pages(). This should fix 
the issues with failed dma_map_sg due to lack of bouncing buffers.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland