nfp: add support for NFDK data path

Add new data path.  The TX is completely different, each packet
has multiple descriptor entries (between 2 and 32).  TX ring is
divided into blocks 32 descriptor, and descritors of one packet
can't cross block bounds. The RX side is the same for now.

ABI version 5 or later is required.  There is no support for
VLAN insertion on TX. XDP_TX action and AF_XDP zero-copy is not
implemented in NFDK path.

Changes to Jakub's work:
* Move statistics of hw_csum_tx after jumbo packet's segmentation.
* Set L3_CSUM flag to enable recaculating of L3 header checksum
in ipv4 case.
* Mark the case of TSO a packet with metadata prepended as
unsupported.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Xingfeng Hu <xingfeng.hu@corigine.com>
Signed-off-by: Yinjun Zhang <yinjun.zhang@corigine.com>
Signed-off-by: Dianchao Wang <dianchao.wang@corigine.com>
Signed-off-by: Fei Qin <fei.qin@corigine.com>
Signed-off-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/netronome/nfp/Makefile

@@ -23,6 +23,8 @@ nfp-objs := \
 	    nfd3/dp.o \
 	    nfd3/rings.o \
 	    nfd3/xsk.o \
+	    nfdk/dp.o \
+	    nfdk/rings.o \
 	    nfp_app.o \
 	    nfp_app_nic.o \
 	    nfp_devlink.o \

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

+/* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */
+/* Copyright (C) 2019 Netronome Systems, Inc. */
+
+#ifndef _NFP_DP_NFDK_H_
+#define _NFP_DP_NFDK_H_
+
+#include <linux/bitops.h>
+#include <linux/types.h>
+
+#define NFDK_TX_DESC_PER_SIMPLE_PKT	2
+
+#define NFDK_TX_MAX_DATA_PER_HEAD	SZ_4K
+#define NFDK_TX_MAX_DATA_PER_DESC	SZ_16K
+#define NFDK_TX_DESC_BLOCK_SZ		256
+#define NFDK_TX_DESC_BLOCK_CNT		(NFDK_TX_DESC_BLOCK_SZ /	\
+					 sizeof(struct nfp_nfdk_tx_desc))
+#define NFDK_TX_DESC_STOP_CNT		(NFDK_TX_DESC_BLOCK_CNT *	\
+					 NFDK_TX_DESC_PER_SIMPLE_PKT)
+#define NFDK_TX_MAX_DATA_PER_BLOCK	SZ_64K
+#define NFDK_TX_DESC_GATHER_MAX		17
+
+/* TX descriptor format */
+
+#define NFDK_DESC_TX_MSS_MASK		GENMASK(13, 0)
+
+#define NFDK_DESC_TX_CHAIN_META		BIT(3)
+#define NFDK_DESC_TX_ENCAP		BIT(2)
+#define NFDK_DESC_TX_L4_CSUM		BIT(1)
+#define NFDK_DESC_TX_L3_CSUM		BIT(0)
+
+#define NFDK_DESC_TX_DMA_LEN_HEAD	GENMASK(11, 0)
+#define NFDK_DESC_TX_TYPE_HEAD		GENMASK(15, 12)
+#define NFDK_DESC_TX_DMA_LEN		GENMASK(13, 0)
+#define NFDK_DESC_TX_TYPE_NOP		0
+#define NFDK_DESC_TX_TYPE_GATHER	1
+#define NFDK_DESC_TX_TYPE_TSO		2
+#define NFDK_DESC_TX_TYPE_SIMPLE	8
+#define NFDK_DESC_TX_EOP		BIT(14)
+
+#define NFDK_META_LEN			GENMASK(7, 0)
+#define NFDK_META_FIELDS		GENMASK(31, 8)
+
+#define D_BLOCK_CPL(idx)		(NFDK_TX_DESC_BLOCK_CNT -	\
+					 (idx) % NFDK_TX_DESC_BLOCK_CNT)
+
+struct nfp_nfdk_tx_desc {
+	union {
+		struct {
+			u8 dma_addr_hi;  /* High bits of host buf address */
+			u8 padding;  /* Must be zero */
+			__le16 dma_len_type; /* Length to DMA for this desc */
+			__le32 dma_addr_lo;  /* Low 32bit of host buf addr */
+		};
+
+		struct {
+			__le16 mss;	/* MSS to be used for LSO */
+			u8 lso_hdrlen;  /* LSO, TCP payload offset */
+			u8 lso_totsegs; /* LSO, total segments */
+			u8 l3_offset;   /* L3 header offset */
+			u8 l4_offset;   /* L4 header offset */
+			__le16 lso_meta_res; /* Rsvd bits in TSO metadata */
+		};
+
+		struct {
+			u8 flags;	/* TX Flags, see @NFDK_DESC_TX_* */
+			u8 reserved[7];	/* meta byte placeholder */
+		};
+
+		__le32 vals[2];
+		__le64 raw;
+	};
+};
+
+struct nfp_nfdk_tx_buf {
+	union {
+		/* First slot */
+		union {
+			struct sk_buff *skb;
+			void *frag;
+		};
+
+		/* 1 + nr_frags next slots */
+		dma_addr_t dma_addr;
+
+		/* TSO (optional) */
+		struct {
+			u32 pkt_cnt;
+			u32 real_len;
+		};
+
+		u64 raw;
+	};
+};
+
+static inline int nfp_nfdk_headlen_to_segs(unsigned int headlen)
+{
+	/* First descriptor fits less data, so adjust for that */
+	return DIV_ROUND_UP(headlen +
+			    NFDK_TX_MAX_DATA_PER_DESC -
+			    NFDK_TX_MAX_DATA_PER_HEAD,
+			    NFDK_TX_MAX_DATA_PER_DESC);
+}
+
+int nfp_nfdk_poll(struct napi_struct *napi, int budget);
+netdev_tx_t nfp_nfdk_tx(struct sk_buff *skb, struct net_device *netdev);
+bool
+nfp_nfdk_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
+		     struct sk_buff *skb, bool old);
+void nfp_nfdk_ctrl_poll(struct tasklet_struct *t);
+void nfp_nfdk_rx_ring_fill_freelist(struct nfp_net_dp *dp,
+				    struct nfp_net_rx_ring *rx_ring);
+#endif

drivers/net/ethernet/netronome/nfp/nfdk/rings.c

+// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+/* Copyright (C) 2019 Netronome Systems, Inc. */
+
+#include <linux/seq_file.h>
+
+#include "../nfp_net.h"
+#include "../nfp_net_dp.h"
+#include "nfdk.h"
+
+static void
+nfp_nfdk_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
+{
+	struct device *dev = dp->dev;
+	struct netdev_queue *nd_q;
+
+	while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
+		const skb_frag_t *frag, *fend;
+		unsigned int size, n_descs = 1;
+		struct nfp_nfdk_tx_buf *txbuf;
+		int nr_frags, rd_idx;
+		struct sk_buff *skb;
+
+		rd_idx = D_IDX(tx_ring, tx_ring->rd_p);
+		txbuf = &tx_ring->ktxbufs[rd_idx];
+
+		skb = txbuf->skb;
+		if (!skb) {
+			n_descs = D_BLOCK_CPL(tx_ring->rd_p);
+			goto next;
+		}
+
+		nr_frags = skb_shinfo(skb)->nr_frags;
+		txbuf++;
+
+		/* Unmap head */
+		size = skb_headlen(skb);
+		dma_unmap_single(dev, txbuf->dma_addr, size, DMA_TO_DEVICE);
+		n_descs += nfp_nfdk_headlen_to_segs(size);
+		txbuf++;
+
+		frag = skb_shinfo(skb)->frags;
+		fend = frag + nr_frags;
+		for (; frag < fend; frag++) {
+			size = skb_frag_size(frag);
+			dma_unmap_page(dev, txbuf->dma_addr,
+				       skb_frag_size(frag), DMA_TO_DEVICE);
+			n_descs += DIV_ROUND_UP(size,
+						NFDK_TX_MAX_DATA_PER_DESC);
+			txbuf++;
+		}
+
+		if (skb_is_gso(skb))
+			n_descs++;
+
+		dev_kfree_skb_any(skb);
+next:
+		tx_ring->rd_p += n_descs;
+	}
+
+	memset(tx_ring->txds, 0, tx_ring->size);
+	tx_ring->data_pending = 0;
+	tx_ring->wr_p = 0;
+	tx_ring->rd_p = 0;
+	tx_ring->qcp_rd_p = 0;
+	tx_ring->wr_ptr_add = 0;
+
+	if (tx_ring->is_xdp || !dp->netdev)
+		return;
+
+	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
+	netdev_tx_reset_queue(nd_q);
+}
+
+static void nfp_nfdk_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
+{
+	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
+	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
+
+	kvfree(tx_ring->ktxbufs);
+
+	if (tx_ring->ktxds)
+		dma_free_coherent(dp->dev, tx_ring->size,
+				  tx_ring->ktxds, tx_ring->dma);
+
+	tx_ring->cnt = 0;
+	tx_ring->txbufs = NULL;
+	tx_ring->txds = NULL;
+	tx_ring->dma = 0;
+	tx_ring->size = 0;
+}
+
+static int
+nfp_nfdk_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
+{
+	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
+
+	tx_ring->cnt = dp->txd_cnt * NFDK_TX_DESC_PER_SIMPLE_PKT;
+	tx_ring->size = array_size(tx_ring->cnt, sizeof(*tx_ring->ktxds));
+	tx_ring->ktxds = dma_alloc_coherent(dp->dev, tx_ring->size,
+					    &tx_ring->dma,
+					    GFP_KERNEL | __GFP_NOWARN);
+	if (!tx_ring->ktxds) {
+		netdev_warn(dp->netdev, "failed to allocate TX descriptor ring memory, requested descriptor count: %d, consider lowering descriptor count\n",
+			    tx_ring->cnt);
+		goto err_alloc;
+	}
+
+	tx_ring->ktxbufs = kvcalloc(tx_ring->cnt, sizeof(*tx_ring->ktxbufs),
+				    GFP_KERNEL);
+	if (!tx_ring->ktxbufs)
+		goto err_alloc;
+
+	if (!tx_ring->is_xdp && dp->netdev)
+		netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
+				    tx_ring->idx);
+
+	return 0;
+
+err_alloc:
+	nfp_nfdk_tx_ring_free(tx_ring);
+	return -ENOMEM;
+}
+
+static void
+nfp_nfdk_tx_ring_bufs_free(struct nfp_net_dp *dp,
+			   struct nfp_net_tx_ring *tx_ring)
+{
+}
+
+static int
+nfp_nfdk_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
+			    struct nfp_net_tx_ring *tx_ring)
+{
+	return 0;
+}
+
+static void
+nfp_nfdk_print_tx_descs(struct seq_file *file,
+			struct nfp_net_r_vector *r_vec,
+			struct nfp_net_tx_ring *tx_ring,
+			u32 d_rd_p, u32 d_wr_p)
+{
+	struct nfp_nfdk_tx_desc *txd;
+	u32 txd_cnt = tx_ring->cnt;
+	int i;
+
+	for (i = 0; i < txd_cnt; i++) {
+		txd = &tx_ring->ktxds[i];
+
+		seq_printf(file, "%04d: 0x%08x 0x%08x 0x%016llx", i,
+			   txd->vals[0], txd->vals[1], tx_ring->ktxbufs[i].raw);
+
+		if (i == tx_ring->rd_p % txd_cnt)
+			seq_puts(file, " H_RD");
+		if (i == tx_ring->wr_p % txd_cnt)
+			seq_puts(file, " H_WR");
+		if (i == d_rd_p % txd_cnt)
+			seq_puts(file, " D_RD");
+		if (i == d_wr_p % txd_cnt)
+			seq_puts(file, " D_WR");
+
+		seq_putc(file, '\n');
+	}
+}
+
+#define NFP_NFDK_CFG_CTRL_SUPPORTED					\
+	(NFP_NET_CFG_CTRL_ENABLE | NFP_NET_CFG_CTRL_PROMISC |		\
+	 NFP_NET_CFG_CTRL_L2BC | NFP_NET_CFG_CTRL_L2MC |		\
+	 NFP_NET_CFG_CTRL_RXCSUM | NFP_NET_CFG_CTRL_TXCSUM |		\
+	 NFP_NET_CFG_CTRL_RXVLAN |					\
+	 NFP_NET_CFG_CTRL_GATHER | NFP_NET_CFG_CTRL_LSO |		\
+	 NFP_NET_CFG_CTRL_CTAG_FILTER | NFP_NET_CFG_CTRL_CMSG_DATA |	\
+	 NFP_NET_CFG_CTRL_RINGCFG | NFP_NET_CFG_CTRL_IRQMOD |		\
+	 NFP_NET_CFG_CTRL_TXRWB |					\
+	 NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE |		\
+	 NFP_NET_CFG_CTRL_BPF | NFP_NET_CFG_CTRL_LSO2 |			\
+	 NFP_NET_CFG_CTRL_RSS2 | NFP_NET_CFG_CTRL_CSUM_COMPLETE |	\
+	 NFP_NET_CFG_CTRL_LIVE_ADDR)
+
+const struct nfp_dp_ops nfp_nfdk_ops = {
+	.version		= NFP_NFD_VER_NFDK,
+	.tx_min_desc_per_pkt	= NFDK_TX_DESC_PER_SIMPLE_PKT,
+	.cap_mask		= NFP_NFDK_CFG_CTRL_SUPPORTED,
+	.poll			= nfp_nfdk_poll,
+	.ctrl_poll		= nfp_nfdk_ctrl_poll,
+	.xmit			= nfp_nfdk_tx,
+	.ctrl_tx_one		= nfp_nfdk_ctrl_tx_one,
+	.rx_ring_fill_freelist	= nfp_nfdk_rx_ring_fill_freelist,
+	.tx_ring_alloc		= nfp_nfdk_tx_ring_alloc,
+	.tx_ring_reset		= nfp_nfdk_tx_ring_reset,
+	.tx_ring_free		= nfp_nfdk_tx_ring_free,
+	.tx_ring_bufs_alloc	= nfp_nfdk_tx_ring_bufs_alloc,
+	.tx_ring_bufs_free	= nfp_nfdk_tx_ring_bufs_free,
+	.print_tx_descs		= nfp_nfdk_print_tx_descs
+};

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

@@ -108,6 +108,9 @@ struct xsk_buff_pool;
 struct nfp_nfd3_tx_desc;
 struct nfp_nfd3_tx_buf;
 
+struct nfp_nfdk_tx_desc;
+struct nfp_nfdk_tx_buf;
+
 /* Convenience macro for wrapping descriptor index on ring size */
 #define D_IDX(ring, idx)	((idx) & ((ring)->cnt - 1))
 
@@ -125,6 +128,7 @@ struct nfp_nfd3_tx_buf;
  * struct nfp_net_tx_ring - TX ring structure
  * @r_vec:      Back pointer to ring vector structure
  * @idx:        Ring index from Linux's perspective
+ * @data_pending: number of bytes added to current block (NFDK only)
  * @qcp_q:      Pointer to base of the QCP TX queue
  * @txrwb:	TX pointer write back area
  * @cnt:        Size of the queue in number of descriptors
@@ -133,8 +137,10 @@ struct nfp_nfd3_tx_buf;
  * @qcp_rd_p:   Local copy of QCP TX queue read pointer
  * @wr_ptr_add:	Accumulated number of buffers to add to QCP write pointer
  *		(used for .xmit_more delayed kick)
- * @txbufs:     Array of transmitted TX buffers, to free on transmit
- * @txds:       Virtual address of TX ring in host memory
+ * @txbufs:	Array of transmitted TX buffers, to free on transmit (NFD3)
+ * @ktxbufs:	Array of transmitted TX buffers, to free on transmit (NFDK)
+ * @txds:	Virtual address of TX ring in host memory (NFD3)
+ * @ktxds:	Virtual address of TX ring in host memory (NFDK)
  *
  * @qcidx:      Queue Controller Peripheral (QCP) queue index for the TX queue
  * @dma:        DMA address of the TX ring
@@ -144,7 +150,8 @@ struct nfp_nfd3_tx_buf;
 struct nfp_net_tx_ring {
 	struct nfp_net_r_vector *r_vec;
 
-	u32 idx;
+	u16 idx;
+	u16 data_pending;
 	u8 __iomem *qcp_q;
 	u64 *txrwb;
 
@@ -155,8 +162,14 @@ struct nfp_net_tx_ring {
 
 	u32 wr_ptr_add;
 
-	struct nfp_nfd3_tx_buf *txbufs;
-	struct nfp_nfd3_tx_desc *txds;
+	union {
+		struct nfp_nfd3_tx_buf *txbufs;
+		struct nfp_nfdk_tx_buf *ktxbufs;
+	};
+	union {
+		struct nfp_nfd3_tx_desc *txds;
+		struct nfp_nfdk_tx_desc *ktxds;
+	};
 
 	/* Cold data follows */
 	int qcidx;
@@ -860,10 +873,12 @@ static inline void nn_ctrl_bar_unlock(struct nfp_net *nn)
 extern const char nfp_driver_version[];
 
 extern const struct net_device_ops nfp_nfd3_netdev_ops;
+extern const struct net_device_ops nfp_nfdk_netdev_ops;
 
 static inline bool nfp_netdev_is_nfp_net(struct net_device *netdev)
 {
-	return netdev->netdev_ops == &nfp_nfd3_netdev_ops;
+	return netdev->netdev_ops == &nfp_nfd3_netdev_ops ||
+	       netdev->netdev_ops == &nfp_nfdk_netdev_ops;
 }
 
 static inline int nfp_net_coalesce_para_check(u32 usecs, u32 pkts)

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

@@ -1920,6 +1920,33 @@ const struct net_device_ops nfp_nfd3_netdev_ops = {
 	.ndo_get_devlink_port	= nfp_devlink_get_devlink_port,
 };
 
+const struct net_device_ops nfp_nfdk_netdev_ops = {
+	.ndo_init		= nfp_app_ndo_init,
+	.ndo_uninit		= nfp_app_ndo_uninit,
+	.ndo_open		= nfp_net_netdev_open,
+	.ndo_stop		= nfp_net_netdev_close,
+	.ndo_start_xmit		= nfp_net_tx,
+	.ndo_get_stats64	= nfp_net_stat64,
+	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
+	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
+	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
+	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
+	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
+	.ndo_set_vf_trust	= nfp_app_set_vf_trust,
+	.ndo_get_vf_config	= nfp_app_get_vf_config,
+	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
+	.ndo_setup_tc		= nfp_port_setup_tc,
+	.ndo_tx_timeout		= nfp_net_tx_timeout,
+	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
+	.ndo_change_mtu		= nfp_net_change_mtu,
+	.ndo_set_mac_address	= nfp_net_set_mac_address,
+	.ndo_set_features	= nfp_net_set_features,
+	.ndo_features_check	= nfp_net_features_check,
+	.ndo_get_phys_port_name	= nfp_net_get_phys_port_name,
+	.ndo_bpf		= nfp_net_xdp,
+	.ndo_get_devlink_port	= nfp_devlink_get_devlink_port,
+};
+
 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table)
 {
 	struct nfp_net *nn = netdev_priv(netdev);
@@ -2042,6 +2069,16 @@ nfp_net_alloc(struct pci_dev *pdev, const struct nfp_dev_info *dev_info,
 	case NFP_NET_CFG_VERSION_DP_NFD3:
 		nn->dp.ops = &nfp_nfd3_ops;
 		break;
+	case NFP_NET_CFG_VERSION_DP_NFDK:
+		if (nn->fw_ver.major < 5) {
+			dev_err(&pdev->dev,
+				"NFDK must use ABI 5 or newer, found: %d\n",
+				nn->fw_ver.major);
+			err = -EINVAL;
+			goto err_free_nn;
+		}
+		nn->dp.ops = &nfp_nfdk_ops;
+		break;
 	default:
 		err = -EINVAL;
 		goto err_free_nn;
@@ -2268,6 +2305,9 @@ static void nfp_net_netdev_init(struct nfp_net *nn)
 	case NFP_NFD_VER_NFD3:
 		netdev->netdev_ops = &nfp_nfd3_netdev_ops;
 		break;
+	case NFP_NFD_VER_NFDK:
+		netdev->netdev_ops = &nfp_nfdk_netdev_ops;
+		break;
 	}
 
 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);

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

@@ -151,6 +151,7 @@
 #define NFP_NET_CFG_VERSION		0x0030
 #define   NFP_NET_CFG_VERSION_RESERVED_MASK	(0xfe << 24)
 #define   NFP_NET_CFG_VERSION_DP_NFD3		0
+#define   NFP_NET_CFG_VERSION_DP_NFDK		1
 #define   NFP_NET_CFG_VERSION_DP_MASK		1
 #define   NFP_NET_CFG_VERSION_CLASS_MASK  (0xff << 16)
 #define   NFP_NET_CFG_VERSION_CLASS(x)	  (((x) & 0xff) << 16)

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

@@ -109,6 +109,7 @@ void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring);
 
 enum nfp_nfd_version {
 	NFP_NFD_VER_NFD3,
+	NFP_NFD_VER_NFDK,
 };
 
 /**
@@ -207,6 +208,7 @@ nfp_net_debugfs_print_tx_descs(struct seq_file *file, struct nfp_net_dp *dp,
 }
 
 extern const struct nfp_dp_ops nfp_nfd3_ops;
+extern const struct nfp_dp_ops nfp_nfdk_ops;
 
 netdev_tx_t nfp_net_tx(struct sk_buff *skb, struct net_device *netdev);
 

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

@@ -112,6 +112,10 @@ int nfp_net_xsk_setup_pool(struct net_device *netdev,
 	struct nfp_net_dp *dp;
 	int err;
 
+	/* NFDK doesn't implement xsk yet. */
+	if (nn->dp.ops->version == NFP_NFD_VER_NFDK)
+		return -EOPNOTSUPP;
+
 	/* Reject on old FWs so we can drop some checks on datapath. */
 	if (nn->dp.rx_offset != NFP_NET_CFG_RX_OFFSET_DYNAMIC)
 		return -EOPNOTSUPP;