Re: [PATCH net-next v2 12/15] idpf: add RX splitq napi poll support

[Jakub Kicinski <kuba@kernel.org>]

On Wed, 14 Jun 2023 10:14:25 -0700 Tony Nguyen wrote:
> +static bool idpf_rx_can_reuse_page(struct idpf_rx_buf *rx_buf)
> +{
> +	unsigned int last_offset = PAGE_SIZE - rx_buf->buf_size;
> +	struct idpf_page_info *pinfo;
> +	unsigned int pagecnt_bias;
> +	struct page *page;
> +
> +	pinfo = &rx_buf->page_info[rx_buf->page_indx];
> +	pagecnt_bias = pinfo->pagecnt_bias;
> +	page = pinfo->page;
> +
> +	if (unlikely(!dev_page_is_reusable(page)))
> +		return false;
> +
> +	if (PAGE_SIZE < 8192) {
> +		/* For 2K buffers, we can reuse the page if we are the
> +		 * owner. For 4K buffers, we can reuse the page if there are
> +		 * no other others.
> +		 */
> +		if (unlikely((page_count(page) - pagecnt_bias) >
> +			     pinfo->reuse_bias))
> +			return false;
> +	} else if (pinfo->page_offset > last_offset) {
> +		return false;
> +	}
> +
> +	/* If we have drained the page fragment pool we need to update
> +	 * the pagecnt_bias and page count so that we fully restock the
> +	 * number of references the driver holds.
> +	 */
> +	if (unlikely(pagecnt_bias == 1)) {
> +		page_ref_add(page, USHRT_MAX - 1);
> +		pinfo->pagecnt_bias = USHRT_MAX;
> +	}
> +
> +	return true;
> +}

If you want to do local recycling you must use the page pool first,
and then share the analysis of how much and why the recycling helps.