ice: Add support for XDP

Add support for XDP. Implement ndo_bpf and ndo_xdp_xmit.  Upon load of
an XDP program, allocate additional Tx rings for dedicated XDP use.
The following actions are supported: XDP_TX, XDP_DROP, XDP_REDIRECT,
XDP_PASS, and XDP_ABORTED.
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

drivers/net/ethernet/intel/ice/ice.h

@@ -29,8 +29,10 @@
 #include <linux/ip.h>
 #include <linux/sctp.h>
 #include <linux/ipv6.h>
+#include <linux/pkt_sched.h>
 #include <linux/if_bridge.h>
 #include <linux/ctype.h>
+#include <linux/bpf.h>
 #include <linux/avf/virtchnl.h>
 #include <net/ipv6.h>
 #include "ice_devids.h"
@@ -78,8 +80,7 @@ extern const char ice_drv_ver[];
 
 #define ICE_DFLT_NETIF_M (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
 
-#define ICE_MAX_MTU	(ICE_AQ_SET_MAC_FRAME_SIZE_MAX - \
-			(ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2)))
+#define ICE_MAX_MTU	(ICE_AQ_SET_MAC_FRAME_SIZE_MAX - ICE_ETH_PKT_HDR_PAD)
 
 #define ICE_UP_TABLE_TRANSLATE(val, i) \
 		(((val) << ICE_AQ_VSI_UP_TABLE_UP##i##_S) & \
@@ -282,6 +283,10 @@ struct ice_vsi {
 	u16 num_rx_desc;
 	u16 num_tx_desc;
 	struct ice_tc_cfg tc_cfg;
+	struct bpf_prog *xdp_prog;
+	struct ice_ring **xdp_rings;	 /* XDP ring array */
+	u16 num_xdp_txq;		 /* Used XDP queues */
+	u8 xdp_mapping_mode;		 /* ICE_MAP_MODE_[CONTIG|SCATTER] */
 } ____cacheline_internodealigned_in_smp;
 
 /* struct that defines an interrupt vector */
@@ -425,6 +430,16 @@ static inline struct ice_pf *ice_netdev_to_pf(struct net_device *netdev)
 	return np->vsi->back;
 }
 
+static inline bool ice_is_xdp_ena_vsi(struct ice_vsi *vsi)
+{
+	return !!vsi->xdp_prog;
+}
+
+static inline void ice_set_ring_xdp(struct ice_ring *ring)
+{
+	ring->flags |= ICE_TX_FLAGS_RING_XDP;
+}
+
 /**
  * ice_get_main_vsi - Get the PF VSI
  * @pf: PF instance
@@ -451,6 +466,11 @@ int ice_up(struct ice_vsi *vsi);
 int ice_down(struct ice_vsi *vsi);
 int ice_vsi_cfg(struct ice_vsi *vsi);
 struct ice_vsi *ice_lb_vsi_setup(struct ice_pf *pf, struct ice_port_info *pi);
+int ice_prepare_xdp_rings(struct ice_vsi *vsi, struct bpf_prog *prog);
+int ice_destroy_xdp_rings(struct ice_vsi *vsi);
+int
+ice_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
+	     u32 flags);
 int ice_set_rss(struct ice_vsi *vsi, u8 *seed, u8 *lut, u16 lut_size);
 int ice_get_rss(struct ice_vsi *vsi, u8 *seed, u8 *lut, u16 lut_size);
 void ice_fill_rss_lut(u8 *lut, u16 rss_table_size, u16 rss_size);

drivers/net/ethernet/intel/ice/ice_base.c

@@ -198,6 +198,9 @@ static void ice_cfg_itr_gran(struct ice_hw *hw)
  */
 static u16 ice_calc_q_handle(struct ice_vsi *vsi, struct ice_ring *ring, u8 tc)
 {
+	WARN_ONCE(ice_ring_is_xdp(ring) && tc,
+		  "XDP ring can't belong to TC other than 0");
+
 	/* Idea here for calculation is that we subtract the number of queue
 	 * count from TC that ring belongs to from it's absolute queue index
 	 * and as a result we get the queue's index within TC.
@@ -287,6 +290,22 @@ int ice_setup_rx_ctx(struct ice_ring *ring)
 	/* clear the context structure first */
 	memset(&rlan_ctx, 0, sizeof(rlan_ctx));
 
+	ring->rx_buf_len = vsi->rx_buf_len;
+
+	if (ring->vsi->type == ICE_VSI_PF) {
+		if (!xdp_rxq_info_is_reg(&ring->xdp_rxq))
+			xdp_rxq_info_reg(&ring->xdp_rxq, ring->netdev,
+					 ring->q_index);
+
+		err = xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
+						 MEM_TYPE_PAGE_SHARED, NULL);
+		if (err)
+			return err;
+	}
+	/* Receive Queue Base Address.
+	 * Indicates the starting address of the descriptor queue defined in
+	 * 128 Byte units.
+	 */
 	rlan_ctx.base = ring->dma >> 7;
 
 	rlan_ctx.qlen = ring->count;
@@ -294,7 +313,7 @@ int ice_setup_rx_ctx(struct ice_ring *ring)
 	/* Receive Packet Data Buffer Size.
 	 * The Packet Data Buffer Size is defined in 128 byte units.
 	 */
-	rlan_ctx.dbuf = vsi->rx_buf_len >> ICE_RLAN_CTX_DBUF_S;
+	rlan_ctx.dbuf = ring->rx_buf_len >> ICE_RLAN_CTX_DBUF_S;
 
 	/* use 32 byte descriptors */
 	rlan_ctx.dsize = 1;
@@ -657,6 +676,13 @@ ice_cfg_txq_interrupt(struct ice_vsi *vsi, u16 txq, u16 msix_idx, u16 itr_idx)
 	      ((msix_idx << QINT_TQCTL_MSIX_INDX_S) & QINT_TQCTL_MSIX_INDX_M);
 
 	wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), val);
+	if (ice_is_xdp_ena_vsi(vsi)) {
+		u32 xdp_txq = txq + vsi->num_xdp_txq;
+
+		wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]),
+		     val);
+	}
+	ice_flush(hw);
 }
 
 /**

drivers/net/ethernet/intel/ice/ice_ethtool.c

@@ -2577,6 +2577,7 @@ ice_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
 {
 	struct ice_ring *tx_rings = NULL, *rx_rings = NULL;
 	struct ice_netdev_priv *np = netdev_priv(netdev);
+	struct ice_ring *xdp_rings = NULL;
 	struct ice_vsi *vsi = np->vsi;
 	struct ice_pf *pf = vsi->back;
 	int i, timeout = 50, err = 0;
@@ -2624,6 +2625,11 @@ ice_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
 			vsi->tx_rings[i]->count = new_tx_cnt;
 		for (i = 0; i < vsi->alloc_rxq; i++)
 			vsi->rx_rings[i]->count = new_rx_cnt;
+		if (ice_is_xdp_ena_vsi(vsi))
+			for (i = 0; i < vsi->num_xdp_txq; i++)
+				vsi->xdp_rings[i]->count = new_tx_cnt;
+		vsi->num_tx_desc = new_tx_cnt;
+		vsi->num_rx_desc = new_rx_cnt;
 		netdev_dbg(netdev, "Link is down, descriptor count change happens when link is brought up\n");
 		goto done;
 	}
@@ -2650,15 +2656,43 @@ ice_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
 		tx_rings[i].tx_buf = NULL;
 		err = ice_setup_tx_ring(&tx_rings[i]);
 		if (err) {
-			while (i) {
-				i--;
+			while (i--)
 				ice_clean_tx_ring(&tx_rings[i]);
-			}
 			devm_kfree(&pf->pdev->dev, tx_rings);
 			goto done;
 		}
 	}
 
+	if (!ice_is_xdp_ena_vsi(vsi))
+		goto process_rx;
+
+	/* alloc updated XDP resources */
+	netdev_info(netdev, "Changing XDP descriptor count from %d to %d\n",
+		    vsi->xdp_rings[0]->count, new_tx_cnt);
+
+	xdp_rings = devm_kcalloc(&pf->pdev->dev, vsi->num_xdp_txq,
+				 sizeof(*xdp_rings), GFP_KERNEL);
+	if (!xdp_rings) {
+		err = -ENOMEM;
+		goto free_tx;
+	}
+
+	for (i = 0; i < vsi->num_xdp_txq; i++) {
+		/* clone ring and setup updated count */
+		xdp_rings[i] = *vsi->xdp_rings[i];
+		xdp_rings[i].count = new_tx_cnt;
+		xdp_rings[i].desc = NULL;
+		xdp_rings[i].tx_buf = NULL;
+		err = ice_setup_tx_ring(&xdp_rings[i]);
+		if (err) {
+			while (i--)
+				ice_clean_tx_ring(&xdp_rings[i]);
+			devm_kfree(&pf->pdev->dev, xdp_rings);
+			goto free_tx;
+		}
+		ice_set_ring_xdp(&xdp_rings[i]);
+	}
+
 process_rx:
 	if (new_rx_cnt == vsi->rx_rings[0]->count)
 		goto process_link;
@@ -2737,6 +2771,16 @@ ice_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
 			devm_kfree(&pf->pdev->dev, rx_rings);
 		}
 
+		if (xdp_rings) {
+			for (i = 0; i < vsi->num_xdp_txq; i++) {
+				ice_free_tx_ring(vsi->xdp_rings[i]);
+				*vsi->xdp_rings[i] = xdp_rings[i];
+			}
+			devm_kfree(&pf->pdev->dev, xdp_rings);
+		}
+
+		vsi->num_tx_desc = new_tx_cnt;
+		vsi->num_rx_desc = new_rx_cnt;
 		ice_up(vsi);
 	}
 	goto done;

drivers/net/ethernet/intel/ice/ice_lib.c

@@ -46,7 +46,8 @@ static int ice_vsi_alloc_arrays(struct ice_vsi *vsi)
 	if (!vsi->rx_rings)
 		goto err_rings;
 
-	vsi->txq_map = devm_kcalloc(&pf->pdev->dev, vsi->alloc_txq,
+	/* XDP will have vsi->alloc_txq Tx queues as well, so double the size */
+	vsi->txq_map = devm_kcalloc(&pf->pdev->dev, (2 * vsi->alloc_txq),
 				    sizeof(*vsi->txq_map), GFP_KERNEL);
 
 	if (!vsi->txq_map)
@@ -1183,6 +1184,20 @@ int ice_vsi_kill_vlan(struct ice_vsi *vsi, u16 vid)
 	return err;
 }
 
+/**
+ * ice_vsi_cfg_frame_size - setup max frame size and Rx buffer length
+ * @vsi: VSI
+ */
+void ice_vsi_cfg_frame_size(struct ice_vsi *vsi)
+{
+	if (vsi->netdev && vsi->netdev->mtu > ETH_DATA_LEN)
+		vsi->max_frame = vsi->netdev->mtu + ICE_ETH_PKT_HDR_PAD;
+	else
+		vsi->max_frame = ICE_RXBUF_2048;
+
+	vsi->rx_buf_len = ICE_RXBUF_2048;
+}
+
 /**
  * ice_vsi_cfg_rxqs - Configure the VSI for Rx
  * @vsi: the VSI being configured
@@ -1197,13 +1212,7 @@ int ice_vsi_cfg_rxqs(struct ice_vsi *vsi)
 	if (vsi->type == ICE_VSI_VF)
 		goto setup_rings;
 
-	if (vsi->netdev && vsi->netdev->mtu > ETH_DATA_LEN)
-		vsi->max_frame = vsi->netdev->mtu +
-			ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
-	else
-		vsi->max_frame = ICE_RXBUF_2048;
-
-	vsi->rx_buf_len = ICE_RXBUF_2048;
+	ice_vsi_cfg_frame_size(vsi);
 setup_rings:
 	/* set up individual rings */
 	for (i = 0; i < vsi->num_rxq; i++) {
@@ -1265,6 +1274,18 @@ int ice_vsi_cfg_lan_txqs(struct ice_vsi *vsi)
 	return ice_vsi_cfg_txqs(vsi, vsi->tx_rings);
 }
 
+/**
+ * ice_vsi_cfg_xdp_txqs - Configure Tx queues dedicated for XDP in given VSI
+ * @vsi: the VSI being configured
+ *
+ * Return 0 on success and a negative value on error
+ * Configure the Tx queues dedicated for XDP in given VSI for operation.
+ */
+int ice_vsi_cfg_xdp_txqs(struct ice_vsi *vsi)
+{
+	return ice_vsi_cfg_txqs(vsi, vsi->xdp_rings);
+}
+
 /**
  * ice_intrl_usec_to_reg - convert interrupt rate limit to register value
  * @intrl: interrupt rate limit in usecs
@@ -1488,6 +1509,15 @@ ice_vsi_stop_lan_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src,
 	return ice_vsi_stop_tx_rings(vsi, rst_src, rel_vmvf_num, vsi->tx_rings);
 }
 
+/**
+ * ice_vsi_stop_xdp_tx_rings - Disable XDP Tx rings
+ * @vsi: the VSI being configured
+ */
+int ice_vsi_stop_xdp_tx_rings(struct ice_vsi *vsi)
+{
+	return ice_vsi_stop_tx_rings(vsi, ICE_NO_RESET, 0, vsi->xdp_rings);
+}
+
 /**
  * ice_cfg_vlan_pruning - enable or disable VLAN pruning on the VSI
  * @vsi: VSI to enable or disable VLAN pruning on
@@ -1885,6 +1915,11 @@ static void ice_vsi_release_msix(struct ice_vsi *vsi)
 		wr32(hw, GLINT_ITR(ICE_IDX_ITR1, reg_idx), 0);
 		for (q = 0; q < q_vector->num_ring_tx; q++) {
 			wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), 0);
+			if (ice_is_xdp_ena_vsi(vsi)) {
+				u32 xdp_txq = txq + vsi->num_xdp_txq;
+
+				wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]), 0);
+			}
 			txq++;
 		}
 
@@ -2259,6 +2294,11 @@ int ice_vsi_rebuild(struct ice_vsi *vsi)
 		vsi->base_vector = 0;
 	}
 
+	if (ice_is_xdp_ena_vsi(vsi))
+		/* return value check can be skipped here, it always returns
+		 * 0 if reset is in progress
+		 */
+		ice_destroy_xdp_rings(vsi);
 	ice_vsi_put_qs(vsi);
 	ice_vsi_clear_rings(vsi);
 	ice_vsi_free_arrays(vsi);
@@ -2299,6 +2339,12 @@ int ice_vsi_rebuild(struct ice_vsi *vsi)
 			goto err_vectors;
 
 		ice_vsi_map_rings_to_vectors(vsi);
+		if (ice_is_xdp_ena_vsi(vsi)) {
+			vsi->num_xdp_txq = vsi->alloc_txq;
+			ret = ice_prepare_xdp_rings(vsi, vsi->xdp_prog);
+			if (ret)
+				goto err_vectors;
+		}
 		/* Do not exit if configuring RSS had an issue, at least
 		 * receive traffic on first queue. Hence no need to capture
 		 * return value
@@ -2325,9 +2371,13 @@ int ice_vsi_rebuild(struct ice_vsi *vsi)
 	}
 
 	/* configure VSI nodes based on number of queues and TC's */
-	for (i = 0; i < vsi->tc_cfg.numtc; i++)
+	for (i = 0; i < vsi->tc_cfg.numtc; i++) {
 		max_txqs[i] = vsi->alloc_txq;
 
+		if (ice_is_xdp_ena_vsi(vsi))
+			max_txqs[i] += vsi->num_xdp_txq;
+	}
+
 	status = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, vsi->tc_cfg.ena_tc,
 				 max_txqs);
 	if (status) {

drivers/net/ethernet/intel/ice/ice_lib.h

@@ -36,6 +36,10 @@ int
 ice_vsi_stop_lan_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src,
 			  u16 rel_vmvf_num);
 
+int ice_vsi_cfg_xdp_txqs(struct ice_vsi *vsi);
+
+int ice_vsi_stop_xdp_tx_rings(struct ice_vsi *vsi);
+
 int ice_cfg_vlan_pruning(struct ice_vsi *vsi, bool ena, bool vlan_promisc);
 
 void ice_cfg_sw_lldp(struct ice_vsi *vsi, bool tx, bool create);
@@ -79,6 +83,8 @@ void ice_vsi_free_tx_rings(struct ice_vsi *vsi);
 
 int ice_vsi_manage_rss_lut(struct ice_vsi *vsi, bool ena);
 
+void ice_vsi_cfg_frame_size(struct ice_vsi *vsi);
+
 u32 ice_intrl_usec_to_reg(u8 intrl, u8 gran);
 
 char *ice_nvm_version_str(struct ice_hw *hw);

drivers/net/ethernet/intel/ice/ice_txrx.h

@@ -22,6 +22,16 @@
 #define ICE_RX_BUF_WRITE	16	/* Must be power of 2 */
 #define ICE_MAX_TXQ_PER_TXQG	128
 
+static inline __le64
+build_ctob(u64 td_cmd, u64 td_offset, unsigned int size, u64 td_tag)
+{
+	return cpu_to_le64(ICE_TX_DESC_DTYPE_DATA |
+			   (td_cmd    << ICE_TXD_QW1_CMD_S) |
+			   (td_offset << ICE_TXD_QW1_OFFSET_S) |
+			   ((u64)size << ICE_TXD_QW1_TX_BUF_SZ_S) |
+			   (td_tag    << ICE_TXD_QW1_L2TAG1_S));
+}
+
 /* We are assuming that the cache line is always 64 Bytes here for ice.
  * In order to make sure that is a correct assumption there is a check in probe
  * to print a warning if the read from GLPCI_CNF2 tells us that the cache line
@@ -49,12 +59,24 @@
 #define ICE_TX_FLAGS_VLAN_PR_S	29
 #define ICE_TX_FLAGS_VLAN_S	16
 
+#define ICE_XDP_PASS		0
+#define ICE_XDP_CONSUMED	BIT(0)
+#define ICE_XDP_TX		BIT(1)
+#define ICE_XDP_REDIR		BIT(2)
+
 #define ICE_RX_DMA_ATTR \
 	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
 
+#define ICE_ETH_PKT_HDR_PAD	(ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
+
+#define ICE_TXD_LAST_DESC_CMD (ICE_TX_DESC_CMD_EOP | ICE_TX_DESC_CMD_RS)
+
 struct ice_tx_buf {
 	struct ice_tx_desc *next_to_watch;
-	struct sk_buff *skb;
+	union {
+		struct sk_buff *skb;
+		void *raw_buf; /* used for XDP */
+	};
 	unsigned int bytecount;
 	unsigned short gso_segs;
 	u32 tx_flags;
@@ -198,9 +220,14 @@ struct ice_ring {
 	};
 
 	struct rcu_head rcu;		/* to avoid race on free */
+	struct bpf_prog *xdp_prog;
+	/* CL3 - 3rd cacheline starts here */
+	struct xdp_rxq_info xdp_rxq;
 	/* CLX - the below items are only accessed infrequently and should be
 	 * in their own cache line if possible
 	 */
+#define ICE_TX_FLAGS_RING_XDP		BIT(0)
+	u8 flags;
 	dma_addr_t dma;			/* physical address of ring */
 	unsigned int size;		/* length of descriptor ring in bytes */
 	u32 txq_teid;			/* Added Tx queue TEID */
@@ -208,6 +235,11 @@ struct ice_ring {
 	u8 dcb_tc;			/* Traffic class of ring */
 } ____cacheline_internodealigned_in_smp;
 
+static inline bool ice_ring_is_xdp(struct ice_ring *ring)
+{
+	return !!(ring->flags & ICE_TX_FLAGS_RING_XDP);
+}
+
 struct ice_ring_container {
 	/* head of linked-list of rings */
 	struct ice_ring *ring;