On Thu, Sep 12, 2024 at 3:43 PM Xuan Zhuo [off-list ref] wrote:

On Wed, 11 Sep 2024 11:54:25 +0800, Jason Wang [off-list ref] wrote:

quoted

On Tue, Aug 20, 2024 at 3:33 PM Xuan Zhuo [off-list ref] wrote:

quoted

The current configuration sets the virtqueue (vq) to premapped mode,
implying that all buffers submitted to this queue must be mapped ahead
of time. This presents a challenge for the virtnet send queue (sq): the
virtnet driver would be required to keep track of dma information for vq
size * 17, which can be substantial. However, if the premapped mode were
applied on a per-buffer basis, the complexity would be greatly reduced.
With AF_XDP enabled, AF_XDP buffers would become premapped, while kernel
skb buffers could remain unmapped.

Is this only applied to TX or both TX and RX.

For rx, if you mean per-buffer dma buffer, I think it is yes,
rx can reuse this. If you mean should we do premapped for the
normal rx buffers, I think we should, that can reduce the
dma map operations.

quoted

quoted

We can distinguish them by sg_page(sg), When sg_page(sg) is NULL, this
indicates that the driver has performed DMA mapping in advance, allowing
the Virtio core to directly utilize sg_dma_address(sg) without
conducting any internal DMA mapping.

This seems conflict with the code below?

#define sg_is_premapped(sg) (!sg_page(sg))

Sorry, I do not get for you.

The key point is that the sg->page is setted by driver.

Ok, I forget that but let's document this assumption in the changelog.

So I mean if the driver sets sg->page = NULL, then for this sg,
the virtio core can skip dma mapping. If the driver sets
sg->page to the page of the buffer, then the virtio core should
do dma mapping for this sg.

Ok, let's describe this in the changelog.

Thanks