Re: [PATCH net-next v3 1/4] net: stmmac: Switch to zero-copy in non-XDP RX path

[Furong Xu <0x1207@gmail.com>]

On Sun, 26 Jan 2025 10:41:23 +0200, Ido Schimmel wrote:
 
> SPH is the only scenario in which the driver uses multiple buffers per
> packet?

Yes.

Jumbo mode may use multiple buffers per packet too, but they are
high order pages, just like a single page in a page pool when using
a standard MTU.

> >         pp_params.max_len = dma_conf->dma_buf_sz;  
> 
> Are you sure this is correct? Page pool documentation says that "For
> pages recycled on the XDP xmit and skb paths the page pool will use
> the max_len member of struct page_pool_params to decide how much of
> the page needs to be synced (starting at offset)" [1].

Page pool must sync an area of the buffer because both DMA and CPU may
touch this area, other areas are CPU exclusive, so no sync for them
seems better.

> While "no more than dma_conf->dma_buf_sz bytes will be written into a
> page buffer", for the head buffer they will be written starting at a
> non-zero offset unlike buffers used for the data, no?

Correct, they have different offsets.

The "SPH feature" splits header into buf->page (non-zero offset) and
splits payload into buf->sec_page (zero offset).

For buf->page, pp_params.max_len should be the size of L3/L4 header,
and with a offset of NET_SKB_PAD.

For buf->sec_page, pp_params.max_len should be dma_conf->dma_buf_sz,
and with a offset of 0.

This is always true:
sizeof(L3/L4 header) + NET_SKB_PAD < dma_conf->dma_buf_sz + 0

pp_params.max_len = dma_conf->dma_buf_sz;
make things simpler :)