On (20/11/25 23:19), Ricardo Ribalda wrote:
[..]
> + if (uvc_urb->pages)
> + dma_sync_sgtable_for_device(stream_to_dmadev(uvc_urb->stream),
> + &uvc_urb->sgt, DMA_FROM_DEVICE);
[..]
> + if (uvc_urb->pages)
> + dma_sync_sgtable_for_cpu(stream_to_dmadev(stream),
> + &uvc_urb->sgt, DMA_FROM_DEVICE);
[..]
> + uvc_urb->pages = dma_alloc_noncontiguous(dma_dev, stream->urb_size,
> + &uvc_urb->dma,
> + gfp_flags | __GFP_NOWARN, 0);
Do we need to pass __GFP_NOWARN? It seems that
dma_alloc_noncontiguous()
__iommu_dma_alloc_noncontiguous()
__iommu_dma_alloc_pages()
does this internally.
> + if (!uvc_urb->pages)
> + return false;
> +
> + uvc_urb->buffer = vmap(uvc_urb->pages,
> + PAGE_ALIGN(stream->urb_size) >> PAGE_SHIFT,
> + VM_DMA_COHERENT, PAGE_KERNEL);
This is not related to Ricardo's patch, just a side note:
I think VM_DMA_COHERENT needs to be renamed. I found it a bit confusing
to see DMA_COHERENT mapping being dma_sync-ed. It turned out that the
flag has different meaning.
-ss
