On 12/04/2024 18:38, Julien Panis wrote:
> This patch adds XDP (eXpress Data Path) support to TI AM65 CPSW
> Ethernet driver. The following features are implemented:
> - NETDEV_XDP_ACT_BASIC (XDP_PASS, XDP_TX, XDP_DROP, XDP_ABORTED)
> - NETDEV_XDP_ACT_REDIRECT (XDP_REDIRECT)
> - NETDEV_XDP_ACT_NDO_XMIT (ndo_xdp_xmit callback)
>
> The page pool memory model is used to get better performance.
> Below are benchmark results obtained for the receiver with iperf3 default
> parameters:
> - Without page pool: 495 Mbits/sec
> - With page pool: 605 Mbits/sec (actually 610 Mbits/sec, with a 5 Mbits/sec
> loss due to extra processing in the hot path to handle XDP).
>
> Signed-off-by: Julien Panis <jpanis@xxxxxxxxxxxx>
> ---
> drivers/net/ethernet/ti/am65-cpsw-nuss.c | 659 ++++++++++++++++++++++++++-----
> drivers/net/ethernet/ti/am65-cpsw-nuss.h | 13 +
> 2 files changed, 576 insertions(+), 96 deletions(-)
>
> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> index 9d2f4ac783e4..3c8134006061 100644
> --- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> +++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
> @@ -5,6 +5,7 @@
> *
<snip>
> +
> +static int am65_cpsw_run_xdp(struct am65_cpsw_common *common,
> + struct am65_cpsw_port *port,
> + struct xdp_buff *xdp,
> + int desc_idx, int cpu, int *len)
> +{
> + struct am65_cpsw_rx_chn *rx_chn = &common->rx_chns;
> + struct net_device *ndev = port->ndev;
> + int ret = AM65_CPSW_XDP_CONSUMED;
> + struct am65_cpsw_tx_chn *tx_chn;
> + struct netdev_queue *netif_txq;
> + struct xdp_frame *xdpf;
> + struct bpf_prog *prog;
> + struct page *page;
> + u32 act;
> +
> + prog = READ_ONCE(port->xdp_prog);
> + if (!prog)
> + return AM65_CPSW_XDP_PASS;
> +
> + act = bpf_prog_run_xdp(prog, xdp);
> + /* XDP prog might have changed packet data and boundaries */
> + *len = xdp->data_end - xdp->data;
> +
> + switch (act) {
> + case XDP_PASS:
> + ret = AM65_CPSW_XDP_PASS;
> + goto out;
> + case XDP_TX:
> + tx_chn = &common->tx_chns[cpu % AM65_CPSW_MAX_TX_QUEUES];
> + netif_txq = netdev_get_tx_queue(ndev, tx_chn->id);
> +
> + xdpf = xdp_convert_buff_to_frame(xdp);
> + if (unlikely(!xdpf))
> + break;
> +
> + __netif_tx_lock(netif_txq, cpu);
> + ret = am65_cpsw_xdp_tx_frame(ndev, tx_chn, xdpf,
> + AM65_CPSW_TX_BUF_TYPE_XDP_TX);
> + __netif_tx_unlock(netif_txq);
> + if (ret)
> + break;
> +
> + ndev->stats.rx_bytes += *len;
> + ndev->stats.rx_packets++;
> + ret = AM65_CPSW_XDP_CONSUMED;
> + goto out;
> + case XDP_REDIRECT:
> + if (unlikely(xdp_do_redirect(ndev, xdp, prog)))
> + break;
> +
> + ndev->stats.rx_bytes += *len;
> + ndev->stats.rx_packets++;
> + ret = AM65_CPSW_XDP_REDIRECT;
> + goto out;
> + default:
> + bpf_warn_invalid_xdp_action(ndev, prog, act);
> + fallthrough;
> + case XDP_ABORTED:
> + trace_xdp_exception(ndev, prog, act);
> + fallthrough;
> + case XDP_DROP:
> + ndev->stats.rx_dropped++;
> + }
> +
> + page = virt_to_head_page(xdp->data);
> + am65_cpsw_put_page(rx_chn, page, true, desc_idx);
here you put the page for failures, XDP_ABORTED or XDP_DROP.
> +
> +out:
> + return ret;
> +}
> +
> static void am65_cpsw_nuss_rx_ts(struct sk_buff *skb, u32 *psdata)
> {
> struct skb_shared_hwtstamps *ssh;
> @@ -795,7 +1090,7 @@ static void am65_cpsw_nuss_rx_csum(struct sk_buff *skb, u32 csum_info)
> }
>
> static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_common *common,
> - u32 flow_idx)
> + u32 flow_idx, int cpu)
> {
> struct am65_cpsw_rx_chn *rx_chn = &common->rx_chns;
> u32 buf_dma_len, pkt_len, port_id = 0, csum_info;
> @@ -803,13 +1098,16 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_common *common,
> struct am65_cpsw_ndev_stats *stats;
> struct cppi5_host_desc_t *desc_rx;
> struct device *dev = common->dev;
> - struct sk_buff *skb, *new_skb;
> + struct page *page, *new_page;
> dma_addr_t desc_dma, buf_dma;
> struct am65_cpsw_port *port;
> + int headroom, desc_idx, ret;
> struct net_device *ndev;
> + struct sk_buff *skb;
> + struct xdp_buff xdp;
> + void *page_addr;
> void **swdata;
> u32 *psdata;
> - int ret = 0;
>
> ret = k3_udma_glue_pop_rx_chn(rx_chn->rx_chn, flow_idx, &desc_dma);
> if (ret) {
> @@ -830,7 +1128,8 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_common *common,
> __func__, flow_idx, &desc_dma);
>
> swdata = cppi5_hdesc_get_swdata(desc_rx);
> - skb = *swdata;
> + page_addr = *swdata;
> + page = virt_to_page(page_addr);
> cppi5_hdesc_get_obuf(desc_rx, &buf_dma, &buf_dma_len);
> k3_udma_glue_rx_cppi5_to_dma_addr(rx_chn->rx_chn, &buf_dma);
> pkt_len = cppi5_hdesc_get_pktlen(desc_rx);
> @@ -838,12 +1137,7 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_common *common,
> dev_dbg(dev, "%s rx port_id:%d\n", __func__, port_id);
> port = am65_common_get_port(common, port_id);
> ndev = port->ndev;
> - skb->dev = ndev;
> -
> psdata = cppi5_hdesc_get_psdata(desc_rx);
> - /* add RX timestamp */
> - if (port->rx_ts_enabled)
> - am65_cpsw_nuss_rx_ts(skb, psdata);
> csum_info = psdata[2];
> dev_dbg(dev, "%s rx csum_info:%#x\n", __func__, csum_info);
>
> @@ -851,36 +1145,66 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_common *common,
>
> k3_cppi_desc_pool_free(rx_chn->desc_pool, desc_rx);
>
> - new_skb = netdev_alloc_skb_ip_align(ndev, AM65_CPSW_MAX_PACKET_SIZE);
> - if (new_skb) {
> - ndev_priv = netdev_priv(ndev);
> - am65_cpsw_nuss_set_offload_fwd_mark(skb, ndev_priv->offload_fwd_mark);
> - skb_put(skb, pkt_len);
> - skb->protocol = eth_type_trans(skb, ndev);
> - am65_cpsw_nuss_rx_csum(skb, csum_info);
> - napi_gro_receive(&common->napi_rx, skb);
> -
> - stats = this_cpu_ptr(ndev_priv->stats);
> -
> - u64_stats_update_begin(&stats->syncp);
> - stats->rx_packets++;
> - stats->rx_bytes += pkt_len;
> - u64_stats_update_end(&stats->syncp);
> - kmemleak_not_leak(new_skb);
> - } else {
> - ndev->stats.rx_dropped++;
> - new_skb = skb;
> + desc_idx = am65_cpsw_nuss_desc_idx(rx_chn->desc_pool, desc_rx,
> + rx_chn->dsize_log2);
> +
> + skb = am65_cpsw_build_skb(page_addr, ndev,
> + AM65_CPSW_MAX_PACKET_SIZE);
> + if (unlikely(!skb)) {
> + new_page = page;
> + goto requeue;
> + }
> +
> + if (port->xdp_prog) {
> + xdp_init_buff(&xdp, AM65_CPSW_MAX_PACKET_SIZE, &port->xdp_rxq);
> +
> + xdp_prepare_buff(&xdp, page_addr, skb_headroom(skb),
> + pkt_len, false);
> +
> + ret = am65_cpsw_run_xdp(common, port, &xdp, desc_idx,
> + cpu, &pkt_len);
> + if (ret != AM65_CPSW_XDP_PASS)
For all cases except AM65_CPSW_XDP_PASS, you never requeue the page for RX.
This is why after all pages are exhausted for example with XDP_DROP,
RX will stall forever.
I will send a fixup patch for this.
> + return ret;
> +
> + /* Compute additional headroom to be reserved */
> + headroom = (xdp.data - xdp.data_hard_start) - skb_headroom(skb);
> + skb_reserve(skb, headroom);
> }
>
> + /* Pass skb to netstack if no XDP prog or returned XDP_PASS */
> + if (port->rx_ts_enabled)
> + am65_cpsw_nuss_rx_ts(skb, psdata);
> +
> + ndev_priv = netdev_priv(ndev);
> + am65_cpsw_nuss_set_offload_fwd_mark(skb, ndev_priv->offload_fwd_mark);
> + skb_put(skb, pkt_len);
> + skb_mark_for_recycle(skb);
> + skb->protocol = eth_type_trans(skb, ndev);
> + am65_cpsw_nuss_rx_csum(skb, csum_info);
> + napi_gro_receive(&common->napi_rx, skb);
> +
> + stats = this_cpu_ptr(ndev_priv->stats);
> +
> + u64_stats_update_begin(&stats->syncp);
> + stats->rx_packets++;
> + stats->rx_bytes += pkt_len;
> + u64_stats_update_end(&stats->syncp);
> +
> + new_page = page_pool_dev_alloc_pages(rx_chn->page_pool);
> + if (unlikely(!new_page))
> + return -ENOMEM;
> + rx_chn->pages[desc_idx] = new_page;
> +
> if (netif_dormant(ndev)) {
> - dev_kfree_skb_any(new_skb);
> + am65_cpsw_put_page(rx_chn, new_page, true, desc_idx);
> ndev->stats.rx_dropped++;
> return 0;
> }
>
> - ret = am65_cpsw_nuss_rx_push(common, new_skb);
> +requeue:
> + ret = am65_cpsw_nuss_rx_push(common, new_page);
> if (WARN_ON(ret < 0)) {
> - dev_kfree_skb_any(new_skb);
> + am65_cpsw_put_page(rx_chn, new_page, true, desc_idx);
> ndev->stats.rx_errors++;
> ndev->stats.rx_dropped++;
> }
> @@ -901,6 +1225,8 @@ static int am65_cpsw_nuss_rx_poll(struct napi_struct *napi_rx, int budget)
> {
> struct am65_cpsw_common *common = am65_cpsw_napi_to_common(napi_rx);
> int flow = AM65_CPSW_MAX_RX_FLOWS;
> + int cpu = smp_processor_id();
> + bool xdp_redirect = false;
> int cur_budget, ret;
> int num_rx = 0;
>
<snip>
--
cheers,
-roger
