Merge branch 'fec-XDP_TX'

Wei Fang says:

====================
net: fec: add XDP_TX feature support

This patch set is to support the XDP_TX feature of FEC driver, the first
patch is add initial XDP_TX support, and the second patch improves the
performance of XDP_TX by not using xdp_convert_buff_to_frame(). Please
refer to the commit message of each patch for more details.
====================

Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 132 insertions, 61 deletions

diff --git a/drivers/net/ethernet/freescale/fec.h b/drivers/net/ethernet/freescale/fec.h
index 5ca9906d7c6a3c..a8fbcada6b01ff 100644
--- a/drivers/net/ethernet/freescale/fec.h
+++ b/drivers/net/ethernet/freescale/fec.h
@@ -548,13 +548,11 @@ enum {
 enum fec_txbuf_type {
 	FEC_TXBUF_T_SKB,
 	FEC_TXBUF_T_XDP_NDO,
+	FEC_TXBUF_T_XDP_TX,
 };
 
 struct fec_tx_buffer {
-	union {
-		struct sk_buff *skb;
-		struct xdp_frame *xdp;
-	};
+	void *buf_p;
 	enum fec_txbuf_type type;
 };
 
diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
index 3bd0bf03aedbf0..f77105f017c1e3 100644
--- a/drivers/net/ethernet/freescale/fec_main.c
+++ b/drivers/net/ethernet/freescale/fec_main.c
@@ -69,6 +69,7 @@
 #include <soc/imx/cpuidle.h>
 #include <linux/filter.h>
 #include <linux/bpf.h>
+#include <linux/bpf_trace.h>
 
 #include <asm/cacheflush.h>
 
@@ -76,6 +77,9 @@
 
 static void set_multicast_list(struct net_device *ndev);
 static void fec_enet_itr_coal_set(struct net_device *ndev);
+static int fec_enet_xdp_tx_xmit(struct fec_enet_private *fep,
+				int cpu, struct xdp_buff *xdp,
+				u32 dma_sync_len);
 
 #define DRIVER_NAME	"fec"
 
@@ -396,7 +400,7 @@ static void fec_dump(struct net_device *ndev)
 			fec16_to_cpu(bdp->cbd_sc),
 			fec32_to_cpu(bdp->cbd_bufaddr),
 			fec16_to_cpu(bdp->cbd_datlen),
-			txq->tx_buf[index].skb);
+			txq->tx_buf[index].buf_p);
 		bdp = fec_enet_get_nextdesc(bdp, &txq->bd);
 		index++;
 	} while (bdp != txq->bd.base);
@@ -653,7 +657,7 @@ static int fec_enet_txq_submit_skb(struct fec_enet_priv_tx_q *txq,
 
 	index = fec_enet_get_bd_index(last_bdp, &txq->bd);
 	/* Save skb pointer */
-	txq->tx_buf[index].skb = skb;
+	txq->tx_buf[index].buf_p = skb;
 
 	/* Make sure the updates to rest of the descriptor are performed before
 	 * transferring ownership.
@@ -859,7 +863,7 @@ static int fec_enet_txq_submit_tso(struct fec_enet_priv_tx_q *txq,
 	}
 
 	/* Save skb pointer */
-	txq->tx_buf[index].skb = skb;
+	txq->tx_buf[index].buf_p = skb;
 
 	skb_tx_timestamp(skb);
 	txq->bd.cur = bdp;
@@ -956,26 +960,27 @@ static void fec_enet_bd_init(struct net_device *dev)
 							 fec32_to_cpu(bdp->cbd_bufaddr),
 							 fec16_to_cpu(bdp->cbd_datlen),
 							 DMA_TO_DEVICE);
-				if (txq->tx_buf[i].skb) {
-					dev_kfree_skb_any(txq->tx_buf[i].skb);
-					txq->tx_buf[i].skb = NULL;
-				}
-			} else {
+				if (txq->tx_buf[i].buf_p)
+					dev_kfree_skb_any(txq->tx_buf[i].buf_p);
+			} else if (txq->tx_buf[i].type == FEC_TXBUF_T_XDP_NDO) {
 				if (bdp->cbd_bufaddr)
 					dma_unmap_single(&fep->pdev->dev,
 							 fec32_to_cpu(bdp->cbd_bufaddr),
 							 fec16_to_cpu(bdp->cbd_datlen),
 							 DMA_TO_DEVICE);
 
-				if (txq->tx_buf[i].xdp) {
-					xdp_return_frame(txq->tx_buf[i].xdp);
-					txq->tx_buf[i].xdp = NULL;
-				}
+				if (txq->tx_buf[i].buf_p)
+					xdp_return_frame(txq->tx_buf[i].buf_p);
+			} else {
+				struct page *page = txq->tx_buf[i].buf_p;
 
-				/* restore default tx buffer type: FEC_TXBUF_T_SKB */
-				txq->tx_buf[i].type = FEC_TXBUF_T_SKB;
+				if (page)
+					page_pool_put_page(page->pp, page, 0, false);
 			}
 
+			txq->tx_buf[i].buf_p = NULL;
+			/* restore default tx buffer type: FEC_TXBUF_T_SKB */
+			txq->tx_buf[i].type = FEC_TXBUF_T_SKB;
 			bdp->cbd_bufaddr = cpu_to_fec32(0);
 			bdp = fec_enet_get_nextdesc(bdp, &txq->bd);
 		}
@@ -1382,6 +1387,8 @@ fec_enet_tx_queue(struct net_device *ndev, u16 queue_id, int budget)
 	struct netdev_queue *nq;
 	int	index = 0;
 	int	entries_free;
+	struct page *page;
+	int frame_len;
 
 	fep = netdev_priv(ndev);
 
@@ -1403,8 +1410,7 @@ fec_enet_tx_queue(struct net_device *ndev, u16 queue_id, int budget)
 		index = fec_enet_get_bd_index(bdp, &txq->bd);
 
 		if (txq->tx_buf[index].type == FEC_TXBUF_T_SKB) {
-			skb = txq->tx_buf[index].skb;
-			txq->tx_buf[index].skb = NULL;
+			skb = txq->tx_buf[index].buf_p;
 			if (bdp->cbd_bufaddr &&
 			    !IS_TSO_HEADER(txq, fec32_to_cpu(bdp->cbd_bufaddr)))
 				dma_unmap_single(&fep->pdev->dev,
@@ -1423,17 +1429,24 @@ fec_enet_tx_queue(struct net_device *ndev, u16 queue_id, int budget)
 			if (unlikely(!budget))
 				break;
 
-			xdpf = txq->tx_buf[index].xdp;
-			if (bdp->cbd_bufaddr)
-				dma_unmap_single(&fep->pdev->dev,
-						 fec32_to_cpu(bdp->cbd_bufaddr),
-						 fec16_to_cpu(bdp->cbd_datlen),
-						 DMA_TO_DEVICE);
+			if (txq->tx_buf[index].type == FEC_TXBUF_T_XDP_NDO) {
+				xdpf = txq->tx_buf[index].buf_p;
+				if (bdp->cbd_bufaddr)
+					dma_unmap_single(&fep->pdev->dev,
+							 fec32_to_cpu(bdp->cbd_bufaddr),
+							 fec16_to_cpu(bdp->cbd_datlen),
+							 DMA_TO_DEVICE);
+			} else {
+				page = txq->tx_buf[index].buf_p;
+			}
+
 			bdp->cbd_bufaddr = cpu_to_fec32(0);
-			if (!xdpf) {
+			if (unlikely(!txq->tx_buf[index].buf_p)) {
 				txq->tx_buf[index].type = FEC_TXBUF_T_SKB;
 				goto tx_buf_done;
 			}
+
+			frame_len = fec16_to_cpu(bdp->cbd_datlen);
 		}
 
 		/* Check for errors. */
@@ -1457,7 +1470,7 @@ fec_enet_tx_queue(struct net_device *ndev, u16 queue_id, int budget)
 			if (txq->tx_buf[index].type == FEC_TXBUF_T_SKB)
 				ndev->stats.tx_bytes += skb->len;
 			else
-				ndev->stats.tx_bytes += xdpf->len;
+				ndev->stats.tx_bytes += frame_len;
 		}
 
 		/* Deferred means some collisions occurred during transmit,
@@ -1482,14 +1495,17 @@ fec_enet_tx_queue(struct net_device *ndev, u16 queue_id, int budget)
 
 			/* Free the sk buffer associated with this last transmit */
 			dev_kfree_skb_any(skb);
-		} else {
-			xdp_return_frame(xdpf);
-
-			txq->tx_buf[index].xdp = NULL;
-			/* restore default tx buffer type: FEC_TXBUF_T_SKB */
-			txq->tx_buf[index].type = FEC_TXBUF_T_SKB;
+		} else if (txq->tx_buf[index].type == FEC_TXBUF_T_XDP_NDO) {
+			xdp_return_frame_rx_napi(xdpf);
+		} else { /* recycle pages of XDP_TX frames */
+			/* The dma_sync_size = 0 as XDP_TX has already synced DMA for_device */
+			page_pool_put_page(page->pp, page, 0, true);
 		}
 
+		txq->tx_buf[index].buf_p = NULL;
+		/* restore default tx buffer type: FEC_TXBUF_T_SKB */
+		txq->tx_buf[index].type = FEC_TXBUF_T_SKB;
+
 tx_buf_done:
 		/* Make sure the update to bdp and tx_buf are performed
 		 * before dirty_tx
@@ -1542,7 +1558,7 @@ static void fec_enet_update_cbd(struct fec_enet_priv_rx_q *rxq,
 
 static u32
 fec_enet_run_xdp(struct fec_enet_private *fep, struct bpf_prog *prog,
-		 struct xdp_buff *xdp, struct fec_enet_priv_rx_q *rxq, int index)
+		 struct xdp_buff *xdp, struct fec_enet_priv_rx_q *rxq, int cpu)
 {
 	unsigned int sync, len = xdp->data_end - xdp->data;
 	u32 ret = FEC_ENET_XDP_PASS;
@@ -1552,8 +1568,10 @@ fec_enet_run_xdp(struct fec_enet_private *fep, struct bpf_prog *prog,
 
 	act = bpf_prog_run_xdp(prog, xdp);
 
-	/* Due xdp_adjust_tail: DMA sync for_device cover max len CPU touch */
-	sync = xdp->data_end - xdp->data_hard_start - FEC_ENET_XDP_HEADROOM;
+	/* Due xdp_adjust_tail and xdp_adjust_head: DMA sync for_device cover
+	 * max len CPU touch
+	 */
+	sync = xdp->data_end - xdp->data;
 	sync = max(sync, len);
 
 	switch (act) {
@@ -1574,11 +1592,19 @@ fec_enet_run_xdp(struct fec_enet_private *fep, struct bpf_prog *prog,
 		}
 		break;
 
-	default:
-		bpf_warn_invalid_xdp_action(fep->netdev, prog, act);
-		fallthrough;
-
 	case XDP_TX:
+		err = fec_enet_xdp_tx_xmit(fep, cpu, xdp, sync);
+		if (unlikely(err)) {
+			ret = FEC_ENET_XDP_CONSUMED;
+			page = virt_to_head_page(xdp->data);
+			page_pool_put_page(rxq->page_pool, page, sync, true);
+			trace_xdp_exception(fep->netdev, prog, act);
+		} else {
+			ret = FEC_ENET_XDP_TX;
+		}
+		break;
+
+	default:
 		bpf_warn_invalid_xdp_action(fep->netdev, prog, act);
 		fallthrough;
 
@@ -1620,6 +1646,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
 	struct bpf_prog *xdp_prog = READ_ONCE(fep->xdp_prog);
 	u32 ret, xdp_result = FEC_ENET_XDP_PASS;
 	u32 data_start = FEC_ENET_XDP_HEADROOM;
+	int cpu = smp_processor_id();
 	struct xdp_buff xdp;
 	struct page *page;
 	u32 sub_len = 4;
@@ -1698,7 +1725,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
 			/* subtract 16bit shift and FCS */
 			xdp_prepare_buff(&xdp, page_address(page),
 					 data_start, pkt_len - sub_len, false);
-			ret = fec_enet_run_xdp(fep, xdp_prog, &xdp, rxq, index);
+			ret = fec_enet_run_xdp(fep, xdp_prog, &xdp, rxq, cpu);
 			xdp_result |= ret;
 			if (ret != FEC_ENET_XDP_PASS)
 				goto rx_processing_done;
@@ -3208,7 +3235,6 @@ static void fec_enet_free_buffers(struct net_device *ndev)
 {
 	struct fec_enet_private *fep = netdev_priv(ndev);
 	unsigned int i;
-	struct sk_buff *skb;
 	struct fec_enet_priv_tx_q *txq;
 	struct fec_enet_priv_rx_q *rxq;
 	unsigned int q;
@@ -3233,18 +3259,23 @@ static void fec_enet_free_buffers(struct net_device *ndev)
 			kfree(txq->tx_bounce[i]);
 			txq->tx_bounce[i] = NULL;
 
+			if (!txq->tx_buf[i].buf_p) {
+				txq->tx_buf[i].type = FEC_TXBUF_T_SKB;
+				continue;
+			}
+
 			if (txq->tx_buf[i].type == FEC_TXBUF_T_SKB) {
-				skb = txq->tx_buf[i].skb;
-				txq->tx_buf[i].skb = NULL;
-				dev_kfree_skb(skb);
+				dev_kfree_skb(txq->tx_buf[i].buf_p);
+			} else if (txq->tx_buf[i].type == FEC_TXBUF_T_XDP_NDO) {
+				xdp_return_frame(txq->tx_buf[i].buf_p);
 			} else {
-				if (txq->tx_buf[i].xdp) {
-					xdp_return_frame(txq->tx_buf[i].xdp);
-					txq->tx_buf[i].xdp = NULL;
-				}
+				struct page *page = txq->tx_buf[i].buf_p;
 
-				txq->tx_buf[i].type = FEC_TXBUF_T_SKB;
+				page_pool_put_page(page->pp, page, 0, false);
 			}
+
+			txq->tx_buf[i].buf_p = NULL;
+			txq->tx_buf[i].type = FEC_TXBUF_T_SKB;
 		}
 	}
 }
@@ -3767,12 +3798,14 @@ fec_enet_xdp_get_tx_queue(struct fec_enet_private *fep, int index)
 
 static int fec_enet_txq_xmit_frame(struct fec_enet_private *fep,
 				   struct fec_enet_priv_tx_q *txq,
-				   struct xdp_frame *frame)
+				   void *frame, u32 dma_sync_len,
+				   bool ndo_xmit)
 {
 	unsigned int index, status, estatus;
 	struct bufdesc *bdp;
 	dma_addr_t dma_addr;
 	int entries_free;
+	u16 frame_len;
 
 	entries_free = fec_enet_get_free_txdesc_num(txq);
 	if (entries_free < MAX_SKB_FRAGS + 1) {
@@ -3787,17 +3820,37 @@ static int fec_enet_txq_xmit_frame(struct fec_enet_private *fep,
 
 	index = fec_enet_get_bd_index(bdp, &txq->bd);
 
-	dma_addr = dma_map_single(&fep->pdev->dev, frame->data,
-				  frame->len, DMA_TO_DEVICE);
-	if (dma_mapping_error(&fep->pdev->dev, dma_addr))
-		return -ENOMEM;
+	if (ndo_xmit) {
+		struct xdp_frame *xdpf = frame;
+
+		dma_addr = dma_map_single(&fep->pdev->dev, xdpf->data,
+					  xdpf->len, DMA_TO_DEVICE);
+		if (dma_mapping_error(&fep->pdev->dev, dma_addr))
+			return -ENOMEM;
+
+		frame_len = xdpf->len;
+		txq->tx_buf[index].buf_p = xdpf;
+		txq->tx_buf[index].type = FEC_TXBUF_T_XDP_NDO;
+	} else {
+		struct xdp_buff *xdpb = frame;
+		struct page *page;
+
+		page = virt_to_page(xdpb->data);
+		dma_addr = page_pool_get_dma_addr(page) +
+			   (xdpb->data - xdpb->data_hard_start);
+		dma_sync_single_for_device(&fep->pdev->dev, dma_addr,
+					   dma_sync_len, DMA_BIDIRECTIONAL);
+		frame_len = xdpb->data_end - xdpb->data;
+		txq->tx_buf[index].buf_p = page;
+		txq->tx_buf[index].type = FEC_TXBUF_T_XDP_TX;
+	}
 
 	status |= (BD_ENET_TX_INTR | BD_ENET_TX_LAST);
 	if (fep->bufdesc_ex)
 		estatus = BD_ENET_TX_INT;
 
 	bdp->cbd_bufaddr = cpu_to_fec32(dma_addr);
-	bdp->cbd_datlen = cpu_to_fec16(frame->len);
+	bdp->cbd_datlen = cpu_to_fec16(frame_len);
 
 	if (fep->bufdesc_ex) {
 		struct bufdesc_ex *ebdp = (struct bufdesc_ex *)bdp;
@@ -3809,9 +3862,6 @@ static int fec_enet_txq_xmit_frame(struct fec_enet_private *fep,
 		ebdp->cbd_esc = cpu_to_fec32(estatus);
 	}
 
-	txq->tx_buf[index].type = FEC_TXBUF_T_XDP_NDO;
-	txq->tx_buf[index].xdp = frame;
-
 	/* Make sure the updates to rest of the descriptor are performed before
 	 * transferring ownership.
 	 */
@@ -3837,6 +3887,29 @@ static int fec_enet_txq_xmit_frame(struct fec_enet_private *fep,
 	return 0;
 }
 
+static int fec_enet_xdp_tx_xmit(struct fec_enet_private *fep,
+				int cpu, struct xdp_buff *xdp,
+				u32 dma_sync_len)
+{
+	struct fec_enet_priv_tx_q *txq;
+	struct netdev_queue *nq;
+	int queue, ret;
+
+	queue = fec_enet_xdp_get_tx_queue(fep, cpu);
+	txq = fep->tx_queue[queue];
+	nq = netdev_get_tx_queue(fep->netdev, queue);
+
+	__netif_tx_lock(nq, cpu);
+
+	/* Avoid tx timeout as XDP shares the queue with kernel stack */
+	txq_trans_cond_update(nq);
+	ret = fec_enet_txq_xmit_frame(fep, txq, xdp, dma_sync_len, false);
+
+	__netif_tx_unlock(nq);
+
+	return ret;
+}
+
 static int fec_enet_xdp_xmit(struct net_device *dev,
 			     int num_frames,
 			     struct xdp_frame **frames,
@@ -3859,7 +3932,7 @@ static int fec_enet_xdp_xmit(struct net_device *dev,
 	/* Avoid tx timeout as XDP shares the queue with kernel stack */
 	txq_trans_cond_update(nq);
 	for (i = 0; i < num_frames; i++) {
-		if (fec_enet_txq_xmit_frame(fep, txq, frames[i]) < 0)
+		if (fec_enet_txq_xmit_frame(fep, txq, frames[i], 0, true) < 0)
 			break;
 		sent_frames++;
 	}