nfp: add XDP support in the driver

Add XDP support.  Separate stack's and XDP's TX rings logically.
Add functions for handling XDP_TX and cleanup of XDP's TX rings.
For XDP allocate all RX buffers as separate pages and map them
with DMA_BIDIRECTIONAL.
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

drivers/net/ethernet/netronome/nfp/nfp_net.h

@@ -171,7 +171,10 @@ struct nfp_net_tx_desc {
  *		on the head's buffer). Equal to skb->len for non-TSO packets.
  */
 struct nfp_net_tx_buf {
-	struct sk_buff *skb;
+	union {
+		struct sk_buff *skb;
+		void *frag;
+	};
 	dma_addr_t dma_addr;
 	short int fidx;
 	u16 pkt_cnt;
@@ -341,6 +344,7 @@ struct nfp_net_rx_ring {
  * @napi:           NAPI structure for this ring vec
  * @tx_ring:        Pointer to TX ring
  * @rx_ring:        Pointer to RX ring
+ * @xdp_ring:	    Pointer to an extra TX ring for XDP
  * @irq_idx:        Index into MSI-X table
  * @rx_sync:	    Seqlock for atomic updates of RX stats
  * @rx_pkts:        Number of received packets
@@ -384,6 +388,8 @@ struct nfp_net_r_vector {
 	u64 hw_csum_rx_inner_ok;
 	u64 hw_csum_rx_error;
 
+	struct nfp_net_tx_ring *xdp_ring;
+
 	struct u64_stats_sync tx_sync;
 	u64 tx_pkts;
 	u64 tx_bytes;
@@ -432,6 +438,7 @@ struct nfp_stat_pair {
  * @ctrl:               Local copy of the control register/word.
  * @fl_bufsz:           Currently configured size of the freelist buffers
  * @rx_offset:		Offset in the RX buffers where packet data starts
+ * @xdp_prog:		Installed XDP program
  * @cpp:                Pointer to the CPP handle
  * @nfp_dev_cpp:        Pointer to the NFP Device handle
  * @ctrl_area:          Pointer to the CPP area for the control BAR
@@ -451,6 +458,7 @@ struct nfp_stat_pair {
  * @max_tx_rings:       Maximum number of TX rings supported by the Firmware
  * @max_rx_rings:       Maximum number of RX rings supported by the Firmware
  * @num_tx_rings:       Currently configured number of TX rings
+ * @num_stack_tx_rings:	Number of TX rings used by the stack (not XDP)
  * @num_rx_rings:       Currently configured number of RX rings
  * @txd_cnt:            Size of the TX ring in number of descriptors
  * @rxd_cnt:            Size of the RX ring in number of descriptors
@@ -500,6 +508,8 @@ struct nfp_net {
 
 	u32 rx_offset;
 
+	struct bpf_prog *xdp_prog;
+
 	struct nfp_net_tx_ring *tx_rings;
 	struct nfp_net_rx_ring *rx_rings;
 
@@ -532,6 +542,7 @@ struct nfp_net {
 	unsigned int max_rx_rings;
 
 	unsigned int num_tx_rings;
+	unsigned int num_stack_tx_rings;
 	unsigned int num_rx_rings;
 
 	int stride_tx;
@@ -779,8 +790,8 @@ void nfp_net_coalesce_write_cfg(struct nfp_net *nn);
 int nfp_net_irqs_alloc(struct nfp_net *nn);
 void nfp_net_irqs_disable(struct nfp_net *nn);
 int
-nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_ring_set *rx,
-		      struct nfp_net_ring_set *tx);
+nfp_net_ring_reconfig(struct nfp_net *nn, struct bpf_prog **xdp_prog,
+		      struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx);
 
 #ifdef CONFIG_NFP_NET_DEBUG
 void nfp_net_debugfs_create(void);

drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c

@@ -114,6 +114,16 @@ static const struct file_operations nfp_rx_q_fops = {
 	.llseek = seq_lseek
 };
 
+static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f);
+
+static const struct file_operations nfp_tx_q_fops = {
+	.owner = THIS_MODULE,
+	.open = nfp_net_debugfs_tx_q_open,
+	.release = single_release,
+	.read = seq_read,
+	.llseek = seq_lseek
+};
+
 static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data)
 {
 	struct nfp_net_r_vector *r_vec = file->private;
@@ -126,10 +136,13 @@ static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data)
 
 	rtnl_lock();
 
-	if (!r_vec->nfp_net || !r_vec->tx_ring)
+	if (debugfs_real_fops(file->file) == &nfp_tx_q_fops)
+		tx_ring = r_vec->tx_ring;
+	else
+		tx_ring = r_vec->xdp_ring;
+	if (!r_vec->nfp_net || !tx_ring)
 		goto out;
 	nn = r_vec->nfp_net;
-	tx_ring = r_vec->tx_ring;
 	if (!netif_running(nn->netdev))
 		goto out;
 
@@ -148,9 +161,14 @@ static int nfp_net_debugfs_tx_q_read(struct seq_file *file, void *data)
 			   txd->vals[2], txd->vals[3]);
 
 		skb = READ_ONCE(tx_ring->txbufs[i].skb);
-		if (skb)
-			seq_printf(file, " skb->head=%p skb->data=%p",
-				   skb->head, skb->data);
+		if (skb) {
+			if (tx_ring == r_vec->tx_ring)
+				seq_printf(file, " skb->head=%p skb->data=%p",
+					   skb->head, skb->data);
+			else
+				seq_printf(file, " frag=%p", skb);
+		}
+
 		if (tx_ring->txbufs[i].dma_addr)
 			seq_printf(file, " dma_addr=%pad",
 				   &tx_ring->txbufs[i].dma_addr);
@@ -176,7 +194,7 @@ static int nfp_net_debugfs_tx_q_open(struct inode *inode, struct file *f)
 	return single_open(f, nfp_net_debugfs_tx_q_read, inode->i_private);
 }
 
-static const struct file_operations nfp_tx_q_fops = {
+static const struct file_operations nfp_xdp_q_fops = {
 	.owner = THIS_MODULE,
 	.open = nfp_net_debugfs_tx_q_open,
 	.release = single_release,
@@ -186,7 +204,7 @@ static const struct file_operations nfp_tx_q_fops = {
 
 void nfp_net_debugfs_adapter_add(struct nfp_net *nn)
 {
-	struct dentry *queues, *tx, *rx;
+	struct dentry *queues, *tx, *rx, *xdp;
 	char int_name[16];
 	int i;
 
@@ -204,13 +222,16 @@ void nfp_net_debugfs_adapter_add(struct nfp_net *nn)
 
 	rx = debugfs_create_dir("rx", queues);
 	tx = debugfs_create_dir("tx", queues);
-	if (IS_ERR_OR_NULL(rx) || IS_ERR_OR_NULL(tx))
+	xdp = debugfs_create_dir("xdp", queues);
+	if (IS_ERR_OR_NULL(rx) || IS_ERR_OR_NULL(tx) || IS_ERR_OR_NULL(xdp))
 		return;
 
 	for (i = 0; i < min(nn->max_rx_rings, nn->max_r_vecs); i++) {
 		sprintf(int_name, "%d", i);
 		debugfs_create_file(int_name, S_IRUSR, rx,
 				    &nn->r_vecs[i], &nfp_rx_q_fops);
+		debugfs_create_file(int_name, S_IRUSR, xdp,
+				    &nn->r_vecs[i], &nfp_xdp_q_fops);
 	}
 
 	for (i = 0; i < min(nn->max_tx_rings, nn->max_r_vecs); i++) {

drivers/net/ethernet/netronome/nfp/nfp_net_ethtool.c

@@ -176,7 +176,8 @@ static int nfp_net_set_ring_size(struct nfp_net *nn, u32 rxd_cnt, u32 txd_cnt)
 	if (nn->txd_cnt != txd_cnt)
 		reconfig_tx = &tx;
 
-	return nfp_net_ring_reconfig(nn, reconfig_rx, reconfig_tx);
+	return nfp_net_ring_reconfig(nn, &nn->xdp_prog,
+				     reconfig_rx, reconfig_tx);
 }
 
 static int nfp_net_set_ringparam(struct net_device *netdev,
@@ -639,14 +640,19 @@ static void nfp_net_get_channels(struct net_device *netdev,
 				 struct ethtool_channels *channel)
 {
 	struct nfp_net *nn = netdev_priv(netdev);
+	unsigned int num_tx_rings;
+
+	num_tx_rings = nn->num_tx_rings;
+	if (nn->xdp_prog)
+		num_tx_rings -= nn->num_rx_rings;
 
 	channel->max_rx = min(nn->max_rx_rings, nn->max_r_vecs);
 	channel->max_tx = min(nn->max_tx_rings, nn->max_r_vecs);
 	channel->max_combined = min(channel->max_rx, channel->max_tx);
 	channel->max_other = NFP_NET_NON_Q_VECTORS;
-	channel->combined_count = min(nn->num_rx_rings, nn->num_tx_rings);
+	channel->combined_count = min(nn->num_rx_rings, num_tx_rings);
 	channel->rx_count = nn->num_rx_rings - channel->combined_count;
-	channel->tx_count = nn->num_tx_rings - channel->combined_count;
+	channel->tx_count = num_tx_rings - channel->combined_count;
 	channel->other_count = NFP_NET_NON_Q_VECTORS;
 }
 
@@ -666,10 +672,16 @@ static int nfp_net_set_num_rings(struct nfp_net *nn, unsigned int total_rx,
 
 	if (nn->num_rx_rings != total_rx)
 		reconfig_rx = ℞
-	if (nn->num_tx_rings != total_tx)
+	if (nn->num_stack_tx_rings != total_tx ||
+	    (nn->xdp_prog && reconfig_rx))
 		reconfig_tx = &tx;
 
-	return nfp_net_ring_reconfig(nn, reconfig_rx, reconfig_tx);
+	/* nfp_net_check_config() will catch tx.n_rings > nn->max_tx_rings */
+	if (nn->xdp_prog)
+		tx.n_rings += total_rx;
+
+	return nfp_net_ring_reconfig(nn, &nn->xdp_prog,
+				     reconfig_rx, reconfig_tx);
 }
 
 static int nfp_net_set_channels(struct net_device *netdev,