Message-ID: <1551462396-5115-3-git-send-email-ioana.ciornei@nxp.com>
Date:   Fri, 1 Mar 2019 17:47:24 +0000
From:   Ioana Ciornei <ioana.ciornei@....com>
To:     "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "davem@...emloft.net" <davem@...emloft.net>
CC:     Ioana Ciocoi Radulescu <ruxandra.radulescu@....com>,
        "brouer@...hat.com" <brouer@...hat.com>,
        "ilias.apalodimas@...aro.org" <ilias.apalodimas@...aro.org>,
        "toke@...hat.com" <toke@...hat.com>,
        Ioana Ciornei <ioana.ciornei@....com>
Subject: [PATCH v2 2/2] dpaa2-eth: add XDP_REDIRECT support

From: Ioana Radulescu <ruxandra.radulescu@....com>

Implement support for the XDP_REDIRECT action.

The redirected frame is transmitted and confirmed on the regular Tx/Tx
conf queues. Frame is marked with the "XDP" type in the software
annotation, since it requires special treatment.

We don't have good hardware support for TX batching, so the
XDP_XMIT_FLUSH flag doesn't make a difference for now; ndo_xdp_xmit
performs the actual Tx operation on the spot.

Signed-off-by: Ioana Ciornei <ioana.ciornei@....com>
Signed-off-by: Ioana Radulescu <ruxandra.radulescu@....com>
---
Changes in v2:
  - add missing xdp_do_flush_map() call

 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c   | 175 +++++++++++++++++++--
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h   |  14 ++
 .../net/ethernet/freescale/dpaa2/dpaa2-ethtool.c   |   1 +
 3 files changed, 176 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
index 3acfd8c..81085e7 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
@@ -296,6 +296,7 @@ static u32 run_xdp(struct dpaa2_eth_priv *priv,
 	xdp.data_end = xdp.data + dpaa2_fd_get_len(fd);
 	xdp.data_hard_start = xdp.data - XDP_PACKET_HEADROOM;
 	xdp_set_data_meta_invalid(&xdp);
+	xdp.rxq = &ch->xdp_rxq;
 
 	xdp_act = bpf_prog_run_xdp(xdp_prog, &xdp);
 
@@ -328,8 +329,20 @@ static u32 run_xdp(struct dpaa2_eth_priv *priv,
 		xdp_release_buf(priv, ch, addr);
 		ch->stats.xdp_drop++;
 		break;
+	case XDP_REDIRECT:
+		dma_unmap_page(priv->net_dev->dev.parent, addr,
+			       DPAA2_ETH_RX_BUF_SIZE, DMA_BIDIRECTIONAL);
+		ch->buf_count--;
+		xdp.data_hard_start = vaddr;
+		err = xdp_do_redirect(priv->net_dev, &xdp, xdp_prog);
+		if (unlikely(err))
+			ch->stats.xdp_drop++;
+		else
+			ch->stats.xdp_redirect++;
+		break;
 	}
 
+	ch->xdp.res |= xdp_act;
 out:
 	rcu_read_unlock();
 	return xdp_act;
@@ -657,27 +670,35 @@ static int build_single_fd(struct dpaa2_eth_priv *priv,
  * dpaa2_eth_tx().
  */
 static void free_tx_fd(const struct dpaa2_eth_priv *priv,
+		       struct dpaa2_eth_fq *fq,
 		       const struct dpaa2_fd *fd, bool in_napi)
 {
 	struct device *dev = priv->net_dev->dev.parent;
 	dma_addr_t fd_addr;
-	struct sk_buff *skb;
+	struct sk_buff *skb = NULL;
 	unsigned char *buffer_start;
 	struct dpaa2_eth_swa *swa;
 	u8 fd_format = dpaa2_fd_get_format(fd);
+	u32 fd_len = dpaa2_fd_get_len(fd);
 
 	fd_addr = dpaa2_fd_get_addr(fd);
 	buffer_start = dpaa2_iova_to_virt(priv->iommu_domain, fd_addr);
 	swa = (struct dpaa2_eth_swa *)buffer_start;
 
 	if (fd_format == dpaa2_fd_single) {
-		skb = swa->single.skb;
-		/* Accessing the skb buffer is safe before dma unmap, because
-		 * we didn't map the actual skb shell.
-		 */
-		dma_unmap_single(dev, fd_addr,
-				 skb_tail_pointer(skb) - buffer_start,
-				 DMA_BIDIRECTIONAL);
+		if (swa->type == DPAA2_ETH_SWA_SINGLE) {
+			skb = swa->single.skb;
+			/* Accessing the skb buffer is safe before dma unmap,
+			 * because we didn't map the actual skb shell.
+			 */
+			dma_unmap_single(dev, fd_addr,
+					 skb_tail_pointer(skb) - buffer_start,
+					 DMA_BIDIRECTIONAL);
+		} else {
+			WARN_ONCE(swa->type != DPAA2_ETH_SWA_XDP, "Wrong SWA type");
+			dma_unmap_single(dev, fd_addr, swa->xdp.dma_size,
+					 DMA_BIDIRECTIONAL);
+		}
 	} else if (fd_format == dpaa2_fd_sg) {
 		skb = swa->sg.skb;
 
@@ -694,6 +715,16 @@ static void free_tx_fd(const struct dpaa2_eth_priv *priv,
 		return;
 	}
 
+	if (swa->type != DPAA2_ETH_SWA_XDP && in_napi) {
+		fq->dq_frames++;
+		fq->dq_bytes += fd_len;
+	}
+
+	if (swa->type == DPAA2_ETH_SWA_XDP) {
+		xdp_return_frame(swa->xdp.xdpf);
+		return;
+	}
+
 	/* Get the timestamp value */
 	if (priv->tx_tstamp && skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
 		struct skb_shared_hwtstamps shhwtstamps;
@@ -793,7 +824,7 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 	if (unlikely(err < 0)) {
 		percpu_stats->tx_errors++;
 		/* Clean up everything, including freeing the skb */
-		free_tx_fd(priv, &fd, false);
+		free_tx_fd(priv, fq, &fd, false);
 	} else {
 		fd_len = dpaa2_fd_get_len(&fd);
 		percpu_stats->tx_packets++;
@@ -830,12 +861,9 @@ static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
 	percpu_extras->tx_conf_frames++;
 	percpu_extras->tx_conf_bytes += fd_len;
 
-	fq->dq_frames++;
-	fq->dq_bytes += fd_len;
-
 	/* Check frame errors in the FD field */
 	fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
-	free_tx_fd(priv, fd, true);
+	free_tx_fd(priv, fq, fd, true);
 
 	if (likely(!fd_errors))
 		return;
@@ -1083,6 +1111,7 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 	int err;
 
 	ch = container_of(napi, struct dpaa2_eth_channel, napi);
+	ch->xdp.res = 0;
 	priv = ch->priv;
 
 	do {
@@ -1128,7 +1157,7 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 	work_done = max(rx_cleaned, 1);
 
 out:
-	if (txc_fq) {
+	if (txc_fq && txc_fq->dq_frames) {
 		nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
 		netdev_tx_completed_queue(nq, txc_fq->dq_frames,
 					  txc_fq->dq_bytes);
@@ -1136,6 +1165,9 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 		txc_fq->dq_bytes = 0;
 	}
 
+	if (ch->xdp.res & XDP_REDIRECT)
+		xdp_do_flush_map();
+
 	return work_done;
 }
 
@@ -1730,6 +1762,105 @@ static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
 	return 0;
 }
 
+static int dpaa2_eth_xdp_xmit_frame(struct net_device *net_dev,
+				    struct xdp_frame *xdpf)
+{
+	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
+	struct device *dev = net_dev->dev.parent;
+	struct rtnl_link_stats64 *percpu_stats;
+	struct dpaa2_eth_drv_stats *percpu_extras;
+	unsigned int needed_headroom;
+	struct dpaa2_eth_swa *swa;
+	struct dpaa2_eth_fq *fq;
+	struct dpaa2_fd fd;
+	void *buffer_start, *aligned_start;
+	dma_addr_t addr;
+	int err, i;
+
+	/* We require a minimum headroom to be able to transmit the frame.
+	 * Otherwise return an error and let the original net_device handle it
+	 */
+	needed_headroom = dpaa2_eth_needed_headroom(priv, NULL);
+	if (xdpf->headroom < needed_headroom)
+		return -EINVAL;
+
+	percpu_stats = this_cpu_ptr(priv->percpu_stats);
+	percpu_extras = this_cpu_ptr(priv->percpu_extras);
+
+	/* Setup the FD fields */
+	memset(&fd, 0, sizeof(fd));
+
+	/* Align FD address, if possible */
+	buffer_start = xdpf->data - needed_headroom;
+	aligned_start = PTR_ALIGN(buffer_start - DPAA2_ETH_TX_BUF_ALIGN,
+				  DPAA2_ETH_TX_BUF_ALIGN);
+	if (aligned_start >= xdpf->data - xdpf->headroom)
+		buffer_start = aligned_start;
+
+	swa = (struct dpaa2_eth_swa *)buffer_start;
+	/* fill in necessary fields here */
+	swa->type = DPAA2_ETH_SWA_XDP;
+	swa->xdp.dma_size = xdpf->data + xdpf->len - buffer_start;
+	swa->xdp.xdpf = xdpf;
+
+	addr = dma_map_single(dev, buffer_start,
+			      swa->xdp.dma_size,
+			      DMA_BIDIRECTIONAL);
+	if (unlikely(dma_mapping_error(dev, addr))) {
+		percpu_stats->tx_dropped++;
+		return -ENOMEM;
+	}
+
+	dpaa2_fd_set_addr(&fd, addr);
+	dpaa2_fd_set_offset(&fd, xdpf->data - buffer_start);
+	dpaa2_fd_set_len(&fd, xdpf->len);
+	dpaa2_fd_set_format(&fd, dpaa2_fd_single);
+	dpaa2_fd_set_ctrl(&fd, FD_CTRL_PTA);
+
+	fq = &priv->fq[smp_processor_id()];
+	for (i = 0; i < DPAA2_ETH_ENQUEUE_RETRIES; i++) {
+		err = priv->enqueue(priv, fq, &fd, 0);
+		if (err != -EBUSY)
+			break;
+	}
+	percpu_extras->tx_portal_busy += i;
+	if (unlikely(err < 0)) {
+		percpu_stats->tx_errors++;
+		/* let the Rx device handle the cleanup */
+		return err;
+	}
+
+	percpu_stats->tx_packets++;
+	percpu_stats->tx_bytes += dpaa2_fd_get_len(&fd);
+
+	return 0;
+}
+
+static int dpaa2_eth_xdp_xmit(struct net_device *net_dev, int n,
+			      struct xdp_frame **frames, u32 flags)
+{
+	int drops = 0;
+	int i, err;
+
+	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
+		return -EINVAL;
+
+	if (!netif_running(net_dev))
+		return -ENETDOWN;
+
+	for (i = 0; i < n; i++) {
+		struct xdp_frame *xdpf = frames[i];
+
+		err = dpaa2_eth_xdp_xmit_frame(net_dev, xdpf);
+		if (err) {
+			xdp_return_frame_rx_napi(xdpf);
+			drops++;
+		}
+	}
+
+	return n - drops;
+}
+
 static const struct net_device_ops dpaa2_eth_ops = {
 	.ndo_open = dpaa2_eth_open,
 	.ndo_start_xmit = dpaa2_eth_tx,
@@ -1741,6 +1872,7 @@ static int dpaa2_eth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
 	.ndo_do_ioctl = dpaa2_eth_ioctl,
 	.ndo_change_mtu = dpaa2_eth_change_mtu,
 	.ndo_bpf = dpaa2_eth_xdp,
+	.ndo_xdp_xmit = dpaa2_eth_xdp_xmit,
 };
 
 static void cdan_cb(struct dpaa2_io_notification_ctx *ctx)
@@ -2353,6 +2485,21 @@ static int setup_rx_flow(struct dpaa2_eth_priv *priv,
 		return err;
 	}
 
+	/* xdp_rxq setup */
+	err = xdp_rxq_info_reg(&fq->channel->xdp_rxq, priv->net_dev,
+			       fq->flowid);
+	if (err) {
+		dev_err(dev, "xdp_rxq_info_reg failed\n");
+		return err;
+	}
+
+	err = xdp_rxq_info_reg_mem_model(&fq->channel->xdp_rxq,
+					 MEM_TYPE_PAGE_ORDER0, NULL);
+	if (err) {
+		dev_err(dev, "xdp_rxq_info_reg_mem_model failed\n");
+		return err;
+	}
+
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
index 423976d..7879622 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
@@ -95,6 +95,7 @@
 enum dpaa2_eth_swa_type {
 	DPAA2_ETH_SWA_SINGLE,
 	DPAA2_ETH_SWA_SG,
+	DPAA2_ETH_SWA_XDP,
 };
 
 /* Must keep this struct smaller than DPAA2_ETH_SWA_SIZE */
@@ -110,6 +111,10 @@ struct dpaa2_eth_swa {
 			int num_sg;
 			int sgt_size;
 		} sg;
+		struct {
+			int dma_size;
+			struct xdp_frame *xdpf;
+		} xdp;
 	};
 };
 
@@ -273,6 +278,7 @@ struct dpaa2_eth_ch_stats {
 	__u64 xdp_drop;
 	__u64 xdp_tx;
 	__u64 xdp_tx_err;
+	__u64 xdp_redirect;
 };
 
 /* Maximum number of queues associated with a DPNI */
@@ -312,6 +318,7 @@ struct dpaa2_eth_ch_xdp {
 	struct bpf_prog *prog;
 	u64 drop_bufs[DPAA2_ETH_BUFS_PER_CMD];
 	int drop_cnt;
+	unsigned int res;
 };
 
 struct dpaa2_eth_channel {
@@ -326,6 +333,7 @@ struct dpaa2_eth_channel {
 	int buf_count;
 	struct dpaa2_eth_ch_stats stats;
 	struct dpaa2_eth_ch_xdp xdp;
+	struct xdp_rxq_info xdp_rxq;
 };
 
 struct dpaa2_eth_dist_fields {
@@ -446,6 +454,12 @@ unsigned int dpaa2_eth_needed_headroom(struct dpaa2_eth_priv *priv,
 {
 	unsigned int headroom = DPAA2_ETH_SWA_SIZE;
 
+	/* If we don't have an skb (e.g. XDP buffer), we only need space for
+	 * the software annotation area
+	 */
+	if (!skb)
+		return headroom;
+
 	/* For non-linear skbs we have no headroom requirement, as we build a
 	 * SG frame with a newly allocated SGT buffer
 	 */
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
index a7389e7..591dfcf 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-ethtool.c
@@ -48,6 +48,7 @@
 	"[drv] xdp drop",
 	"[drv] xdp tx",
 	"[drv] xdp tx errors",
+	"[drv] xdp redirect",
 	/* FQ stats */
 	"[qbman] rx pending frames",
 	"[qbman] rx pending bytes",
-- 
1.9.1

Powered by blists - more mailing lists