Merge branch 'add-flow-director-for-txgbe'

Jiawen Wu says:

====================
add flow director for txgbe

Add flow director support for Wangxun 10Gb NICs.

v2 -> v3: https://lore.kernel.org/all/20240605020852.24144-1-jiawenwu@trustnetic.com/
- Wrap the code at 80 chars where possible. (Jakub Kicinski)
- Add function description address on kernel-doc. (Jakub Kicinski)
- Correct return code. (Simon Horman)
- Remove redundant size check. (Hariprasad Kelam)

v1 -> v2: https://lore.kernel.org/all/20240529093821.27108-1-jiawenwu@trustnetic.com/
- Fix build warnings reported by kernel test robot.
====================

Link: https://lore.kernel.org/r/20240618101609.3580-1-jiawenwu@trustnetic.comSigned-off-by: Paolo Abeni <pabeni@redhat.com>

drivers/net/ethernet/wangxun/libwx/wx_ethtool.c

@@ -43,6 +43,11 @@ static const struct wx_stats wx_gstrings_stats[] = {
 	WX_STAT("alloc_rx_buff_failed", alloc_rx_buff_failed),
 };
 
+static const struct wx_stats wx_gstrings_fdir_stats[] = {
+	WX_STAT("fdir_match", stats.fdirmatch),
+	WX_STAT("fdir_miss", stats.fdirmiss),
+};
+
 /* drivers allocates num_tx_queues and num_rx_queues symmetrically so
  * we set the num_rx_queues to evaluate to num_tx_queues. This is
  * used because we do not have a good way to get the max number of
@@ -55,13 +60,17 @@ static const struct wx_stats wx_gstrings_stats[] = {
 		(WX_NUM_TX_QUEUES + WX_NUM_RX_QUEUES) * \
 		(sizeof(struct wx_queue_stats) / sizeof(u64)))
 #define WX_GLOBAL_STATS_LEN  ARRAY_SIZE(wx_gstrings_stats)
+#define WX_FDIR_STATS_LEN  ARRAY_SIZE(wx_gstrings_fdir_stats)
 #define WX_STATS_LEN (WX_GLOBAL_STATS_LEN + WX_QUEUE_STATS_LEN)
 
 int wx_get_sset_count(struct net_device *netdev, int sset)
 {
+	struct wx *wx = netdev_priv(netdev);
+
 	switch (sset) {
 	case ETH_SS_STATS:
-		return WX_STATS_LEN;
+		return (wx->mac.type == wx_mac_sp) ?
+			WX_STATS_LEN + WX_FDIR_STATS_LEN : WX_STATS_LEN;
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -70,6 +79,7 @@ EXPORT_SYMBOL(wx_get_sset_count);
 
 void wx_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
 {
+	struct wx *wx = netdev_priv(netdev);
 	u8 *p = data;
 	int i;
 
@@ -77,6 +87,10 @@ void wx_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
 	case ETH_SS_STATS:
 		for (i = 0; i < WX_GLOBAL_STATS_LEN; i++)
 			ethtool_puts(&p, wx_gstrings_stats[i].stat_string);
+		if (wx->mac.type == wx_mac_sp) {
+			for (i = 0; i < WX_FDIR_STATS_LEN; i++)
+				ethtool_puts(&p, wx_gstrings_fdir_stats[i].stat_string);
+		}
 		for (i = 0; i < netdev->num_tx_queues; i++) {
 			ethtool_sprintf(&p, "tx_queue_%u_packets", i);
 			ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
@@ -96,7 +110,7 @@ void wx_get_ethtool_stats(struct net_device *netdev,
 	struct wx *wx = netdev_priv(netdev);
 	struct wx_ring *ring;
 	unsigned int start;
-	int i, j;
+	int i, j, k;
 	char *p;
 
 	wx_update_stats(wx);
@@ -107,6 +121,13 @@ void wx_get_ethtool_stats(struct net_device *netdev,
 			   sizeof(u64)) ? *(u64 *)p : *(u32 *)p;
 	}
 
+	if (wx->mac.type == wx_mac_sp) {
+		for (k = 0; k < WX_FDIR_STATS_LEN; k++) {
+			p = (char *)wx + wx_gstrings_fdir_stats[k].stat_offset;
+			data[i++] = *(u64 *)p;
+		}
+	}
+
 	for (j = 0; j < netdev->num_tx_queues; j++) {
 		ring = wx->tx_ring[j];
 		if (!ring) {
@@ -172,17 +193,21 @@ EXPORT_SYMBOL(wx_get_pause_stats);
 
 void wx_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *info)
 {
+	unsigned int stats_len = WX_STATS_LEN;
 	struct wx *wx = netdev_priv(netdev);
 
+	if (wx->mac.type == wx_mac_sp)
+		stats_len += WX_FDIR_STATS_LEN;
+
 	strscpy(info->driver, wx->driver_name, sizeof(info->driver));
 	strscpy(info->fw_version, wx->eeprom_id, sizeof(info->fw_version));
 	strscpy(info->bus_info, pci_name(wx->pdev), sizeof(info->bus_info));
 	if (wx->num_tx_queues <= WX_NUM_TX_QUEUES) {
-		info->n_stats = WX_STATS_LEN -
+		info->n_stats = stats_len -
 				   (WX_NUM_TX_QUEUES - wx->num_tx_queues) *
 				   (sizeof(struct wx_queue_stats) / sizeof(u64)) * 2;
 	} else {
-		info->n_stats = WX_STATS_LEN;
+		info->n_stats = stats_len;
 	}
 }
 EXPORT_SYMBOL(wx_get_drvinfo);
@@ -383,6 +408,9 @@ void wx_get_channels(struct net_device *dev,
 
 	/* record RSS queues */
 	ch->combined_count = wx->ring_feature[RING_F_RSS].indices;
+
+	if (test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags))
+		ch->combined_count = wx->ring_feature[RING_F_FDIR].indices;
 }
 EXPORT_SYMBOL(wx_get_channels);
 
@@ -400,6 +428,9 @@ int wx_set_channels(struct net_device *dev,
 	if (count > wx_max_channels(wx))
 		return -EINVAL;
 
+	if (test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags))
+		wx->ring_feature[RING_F_FDIR].limit = count;
+
 	wx->ring_feature[RING_F_RSS].limit = count;
 
 	return 0;

drivers/net/ethernet/wangxun/libwx/wx_hw.c

@@ -1147,8 +1147,15 @@ static void wx_enable_rx(struct wx *wx)
 static void wx_set_rxpba(struct wx *wx)
 {
 	u32 rxpktsize, txpktsize, txpbthresh;
+	u32 pbsize = wx->mac.rx_pb_size;
 
-	rxpktsize = wx->mac.rx_pb_size << WX_RDB_PB_SZ_SHIFT;
+	if (test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags)) {
+		if (test_bit(WX_FLAG_FDIR_HASH, wx->flags) ||
+		    test_bit(WX_FLAG_FDIR_PERFECT, wx->flags))
+			pbsize -= 64; /* Default 64KB */
+	}
+
+	rxpktsize = pbsize << WX_RDB_PB_SZ_SHIFT;
 	wr32(wx, WX_RDB_PB_SZ(0), rxpktsize);
 
 	/* Only support an equally distributed Tx packet buffer strategy. */
@@ -1261,7 +1268,7 @@ static void wx_configure_port(struct wx *wx)
  *  Stops the receive data path and waits for the HW to internally empty
  *  the Rx security block
  **/
-static int wx_disable_sec_rx_path(struct wx *wx)
+int wx_disable_sec_rx_path(struct wx *wx)
 {
 	u32 secrx;
 
@@ -1271,6 +1278,7 @@ static int wx_disable_sec_rx_path(struct wx *wx)
 	return read_poll_timeout(rd32, secrx, secrx & WX_RSC_ST_RSEC_RDY,
 				 1000, 40000, false, wx, WX_RSC_ST);
 }
+EXPORT_SYMBOL(wx_disable_sec_rx_path);
 
 /**
  *  wx_enable_sec_rx_path - Enables the receive data path
@@ -1278,11 +1286,12 @@ static int wx_disable_sec_rx_path(struct wx *wx)
  *
  *  Enables the receive data path.
  **/
-static void wx_enable_sec_rx_path(struct wx *wx)
+void wx_enable_sec_rx_path(struct wx *wx)
 {
 	wr32m(wx, WX_RSC_CTL, WX_RSC_CTL_RX_DIS, 0);
 	WX_WRITE_FLUSH(wx);
 }
+EXPORT_SYMBOL(wx_enable_sec_rx_path);
 
 static void wx_vlan_strip_control(struct wx *wx, bool enable)
 {
@@ -1499,6 +1508,13 @@ static void wx_configure_tx_ring(struct wx *wx,
 		txdctl |= ring->count / 128 << WX_PX_TR_CFG_TR_SIZE_SHIFT;
 	txdctl |= 0x20 << WX_PX_TR_CFG_WTHRESH_SHIFT;
 
+	ring->atr_count = 0;
+	if (test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags) &&
+	    test_bit(WX_FLAG_FDIR_HASH, wx->flags))
+		ring->atr_sample_rate = wx->atr_sample_rate;
+	else
+		ring->atr_sample_rate = 0;
+
 	/* reinitialize tx_buffer_info */
 	memset(ring->tx_buffer_info, 0,
 	       sizeof(struct wx_tx_buffer) * ring->count);
@@ -1732,7 +1748,9 @@ void wx_configure(struct wx *wx)
 
 	wx_set_rx_mode(wx->netdev);
 	wx_restore_vlan(wx);
-	wx_enable_sec_rx_path(wx);
+
+	if (test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags))
+		wx->configure_fdir(wx);
 
 	wx_configure_tx(wx);
 	wx_configure_rx(wx);
@@ -1959,6 +1977,7 @@ int wx_sw_init(struct wx *wx)
 	}
 
 	bitmap_zero(wx->state, WX_STATE_NBITS);
+	bitmap_zero(wx->flags, WX_PF_FLAGS_NBITS);
 
 	return 0;
 }
@@ -2333,6 +2352,11 @@ void wx_update_stats(struct wx *wx)
 	hwstats->b2ogprc += rd32(wx, WX_RDM_BMC2OS_CNT);
 	hwstats->rdmdrop += rd32(wx, WX_RDM_DRP_PKT);
 
+	if (wx->mac.type == wx_mac_sp) {
+		hwstats->fdirmatch += rd32(wx, WX_RDB_FDIR_MATCH);
+		hwstats->fdirmiss += rd32(wx, WX_RDB_FDIR_MISS);
+	}
+
 	for (i = 0; i < wx->mac.max_rx_queues; i++)
 		hwstats->qmprc += rd32(wx, WX_PX_MPRC(i));
 }

drivers/net/ethernet/wangxun/libwx/wx_hw.h

@@ -28,6 +28,8 @@ void wx_mac_set_default_filter(struct wx *wx, u8 *addr);
 void wx_flush_sw_mac_table(struct wx *wx);
 int wx_set_mac(struct net_device *netdev, void *p);
 void wx_disable_rx(struct wx *wx);
+int wx_disable_sec_rx_path(struct wx *wx);
+void wx_enable_sec_rx_path(struct wx *wx);
 void wx_set_rx_mode(struct net_device *netdev);
 int wx_change_mtu(struct net_device *netdev, int new_mtu);
 void wx_disable_rx_queue(struct wx *wx, struct wx_ring *ring);

drivers/net/ethernet/wangxun/libwx/wx_lib.c

@@ -148,10 +148,11 @@ static struct wx_dec_ptype wx_ptype_lookup[256] = {
 	[0xFD] = WX_PTT(IP, IPV6, IGMV, IPV6, SCTP, PAY4),
 };
 
-static struct wx_dec_ptype wx_decode_ptype(const u8 ptype)
+struct wx_dec_ptype wx_decode_ptype(const u8 ptype)
 {
 	return wx_ptype_lookup[ptype];
 }
+EXPORT_SYMBOL(wx_decode_ptype);
 
 /* wx_test_staterr - tests bits in Rx descriptor status and error fields */
 static __le32 wx_test_staterr(union wx_rx_desc *rx_desc,
@@ -1453,6 +1454,7 @@ static void wx_tx_csum(struct wx_ring *tx_ring, struct wx_tx_buffer *first,
 static netdev_tx_t wx_xmit_frame_ring(struct sk_buff *skb,
 				      struct wx_ring *tx_ring)
 {
+	struct wx *wx = netdev_priv(tx_ring->netdev);
 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
 	struct wx_tx_buffer *first;
 	u8 hdr_len = 0, ptype;
@@ -1498,6 +1500,10 @@ static netdev_tx_t wx_xmit_frame_ring(struct sk_buff *skb,
 		goto out_drop;
 	else if (!tso)
 		wx_tx_csum(tx_ring, first, ptype);
+
+	if (test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags) && tx_ring->atr_sample_rate)
+		wx->atr(tx_ring, first, ptype);
+
 	wx_tx_map(tx_ring, first, hdr_len);
 
 	return NETDEV_TX_OK;
@@ -1574,8 +1580,27 @@ static void wx_set_rss_queues(struct wx *wx)
 	f = &wx->ring_feature[RING_F_RSS];
 	f->indices = f->limit;
 
-	wx->num_rx_queues = f->limit;
-	wx->num_tx_queues = f->limit;
+	if (!(test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags)))
+		goto out;
+
+	clear_bit(WX_FLAG_FDIR_HASH, wx->flags);
+
+	/* Use Flow Director in addition to RSS to ensure the best
+	 * distribution of flows across cores, even when an FDIR flow
+	 * isn't matched.
+	 */
+	if (f->indices > 1) {
+		f = &wx->ring_feature[RING_F_FDIR];
+
+		f->indices = f->limit;
+
+		if (!(test_bit(WX_FLAG_FDIR_PERFECT, wx->flags)))
+			set_bit(WX_FLAG_FDIR_HASH, wx->flags);
+	}
+
+out:
+	wx->num_rx_queues = f->indices;
+	wx->num_tx_queues = f->indices;
 }
 
 static void wx_set_num_queues(struct wx *wx)
@@ -2680,6 +2705,7 @@ int wx_set_features(struct net_device *netdev, netdev_features_t features)
 {
 	netdev_features_t changed = netdev->features ^ features;
 	struct wx *wx = netdev_priv(netdev);
+	bool need_reset = false;
 
 	if (features & NETIF_F_RXHASH) {
 		wr32m(wx, WX_RDB_RA_CTL, WX_RDB_RA_CTL_RSS_EN,
@@ -2697,6 +2723,36 @@ int wx_set_features(struct net_device *netdev, netdev_features_t features)
 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER))
 		wx_set_rx_mode(netdev);
 
+	if (!(test_bit(WX_FLAG_FDIR_CAPABLE, wx->flags)))
+		return 0;
+
+	/* Check if Flow Director n-tuple support was enabled or disabled.  If
+	 * the state changed, we need to reset.
+	 */
+	switch (features & NETIF_F_NTUPLE) {
+	case NETIF_F_NTUPLE:
+		/* turn off ATR, enable perfect filters and reset */
+		if (!(test_and_set_bit(WX_FLAG_FDIR_PERFECT, wx->flags)))
+			need_reset = true;
+
+		clear_bit(WX_FLAG_FDIR_HASH, wx->flags);
+		break;
+	default:
+		/* turn off perfect filters, enable ATR and reset */
+		if (test_and_clear_bit(WX_FLAG_FDIR_PERFECT, wx->flags))
+			need_reset = true;
+
+		/* We cannot enable ATR if RSS is disabled */
+		if (wx->ring_feature[RING_F_RSS].limit <= 1)
+			break;
+
+		set_bit(WX_FLAG_FDIR_HASH, wx->flags);
+		break;
+	}
+
+	if (need_reset)
+		wx->do_reset(netdev);
+
 	return 0;
 }
 EXPORT_SYMBOL(wx_set_features);

drivers/net/ethernet/wangxun/libwx/wx_lib.h

@@ -7,6 +7,7 @@
 #ifndef _WX_LIB_H_
 #define _WX_LIB_H_
 
+struct wx_dec_ptype wx_decode_ptype(const u8 ptype);
 void wx_alloc_rx_buffers(struct wx_ring *rx_ring, u16 cleaned_count);
 u16 wx_desc_unused(struct wx_ring *ring);
 netdev_tx_t wx_xmit_frame(struct sk_buff *skb,

drivers/net/ethernet/wangxun/libwx/wx_type.h

@@ -157,6 +157,8 @@
 #define WX_RDB_RA_CTL_RSS_IPV6_TCP   BIT(21)
 #define WX_RDB_RA_CTL_RSS_IPV4_UDP   BIT(22)
 #define WX_RDB_RA_CTL_RSS_IPV6_UDP   BIT(23)
+#define WX_RDB_FDIR_MATCH            0x19558
+#define WX_RDB_FDIR_MISS             0x1955C
 
 /******************************* PSR Registers *******************************/
 /* psr control */
@@ -503,6 +505,34 @@ enum WX_MSCA_CMD_value {
 #define WX_PTYPE_TYP_TCP             0x04
 #define WX_PTYPE_TYP_SCTP            0x05
 
+/* Packet type non-ip values */
+enum wx_l2_ptypes {
+	WX_PTYPE_L2_ABORTED = (WX_PTYPE_PKT_MAC),
+	WX_PTYPE_L2_MAC = (WX_PTYPE_PKT_MAC | WX_PTYPE_TYP_MAC),
+
+	WX_PTYPE_L2_IPV4_FRAG = (WX_PTYPE_PKT_IP | WX_PTYPE_TYP_IPFRAG),
+	WX_PTYPE_L2_IPV4 = (WX_PTYPE_PKT_IP | WX_PTYPE_TYP_IP),
+	WX_PTYPE_L2_IPV4_UDP = (WX_PTYPE_PKT_IP | WX_PTYPE_TYP_UDP),
+	WX_PTYPE_L2_IPV4_TCP = (WX_PTYPE_PKT_IP | WX_PTYPE_TYP_TCP),
+	WX_PTYPE_L2_IPV4_SCTP = (WX_PTYPE_PKT_IP | WX_PTYPE_TYP_SCTP),
+	WX_PTYPE_L2_IPV6_FRAG = (WX_PTYPE_PKT_IP | WX_PTYPE_PKT_IPV6 |
+				 WX_PTYPE_TYP_IPFRAG),
+	WX_PTYPE_L2_IPV6 = (WX_PTYPE_PKT_IP | WX_PTYPE_PKT_IPV6 |
+			    WX_PTYPE_TYP_IP),
+	WX_PTYPE_L2_IPV6_UDP = (WX_PTYPE_PKT_IP | WX_PTYPE_PKT_IPV6 |
+				WX_PTYPE_TYP_UDP),
+	WX_PTYPE_L2_IPV6_TCP = (WX_PTYPE_PKT_IP | WX_PTYPE_PKT_IPV6 |
+				WX_PTYPE_TYP_TCP),
+	WX_PTYPE_L2_IPV6_SCTP = (WX_PTYPE_PKT_IP | WX_PTYPE_PKT_IPV6 |
+				 WX_PTYPE_TYP_SCTP),
+
+	WX_PTYPE_L2_TUN4_MAC = (WX_PTYPE_TUN_IPV4 | WX_PTYPE_PKT_IGM),
+	WX_PTYPE_L2_TUN6_MAC = (WX_PTYPE_TUN_IPV6 | WX_PTYPE_PKT_IGM),
+};
+
+#define WX_PTYPE_PKT(_pt)            ((_pt) & 0x30)
+#define WX_PTYPE_TYPL4(_pt)          ((_pt) & 0x07)
+
 #define WX_RXD_PKTTYPE(_rxd) \
 	((le32_to_cpu((_rxd)->wb.lower.lo_dword.data) >> 9) & 0xFF)
 #define WX_RXD_IPV6EX(_rxd) \
@@ -552,6 +582,9 @@ enum wx_tx_flags {
 	WX_TX_FLAGS_OUTER_IPV4	= 0x100,
 	WX_TX_FLAGS_LINKSEC	= 0x200,
 	WX_TX_FLAGS_IPSEC	= 0x400,
+
+	/* software defined flags */
+	WX_TX_FLAGS_SW_VLAN	= 0x40,
 };
 
 /* VLAN info */
@@ -900,7 +933,13 @@ struct wx_ring {
 					 */
 	u16 next_to_use;
 	u16 next_to_clean;
-	u16 next_to_alloc;
+	union {
+		u16 next_to_alloc;
+		struct {
+			u8 atr_sample_rate;
+			u8 atr_count;
+		};
+	};
 
 	struct wx_queue_stats stats;
 	struct u64_stats_sync syncp;
@@ -939,6 +978,7 @@ struct wx_ring_feature {
 enum wx_ring_f_enum {
 	RING_F_NONE = 0,
 	RING_F_RSS,
+	RING_F_FDIR,
 	RING_F_ARRAY_SIZE  /* must be last in enum set */
 };
 
@@ -980,15 +1020,26 @@ struct wx_hw_stats {
 	u64 crcerrs;
 	u64 rlec;
 	u64 qmprc;
+	u64 fdirmatch;
+	u64 fdirmiss;
 };
 
 enum wx_state {
 	WX_STATE_RESETTING,
 	WX_STATE_NBITS,		/* must be last */
 };
+
+enum wx_pf_flags {
+	WX_FLAG_FDIR_CAPABLE,
+	WX_FLAG_FDIR_HASH,
+	WX_FLAG_FDIR_PERFECT,
+	WX_PF_FLAGS_NBITS               /* must be last */
+};
+
 struct wx {
 	unsigned long active_vlans[BITS_TO_LONGS(VLAN_N_VID)];
 	DECLARE_BITMAP(state, WX_STATE_NBITS);
+	DECLARE_BITMAP(flags, WX_PF_FLAGS_NBITS);
 
 	void *priv;
 	u8 __iomem *hw_addr;
@@ -1077,6 +1128,9 @@ struct wx {
 	u64 hw_csum_rx_error;
 	u64 alloc_rx_buff_failed;
 
+	u32 atr_sample_rate;
+	void (*atr)(struct wx_ring *ring, struct wx_tx_buffer *first, u8 ptype);
+	void (*configure_fdir)(struct wx *wx);
 	void (*do_reset)(struct net_device *netdev);
 };
 

drivers/net/ethernet/wangxun/txgbe/Makefile

@@ -10,4 +10,5 @@ txgbe-objs := txgbe_main.o \
               txgbe_hw.o \
               txgbe_phy.o \
               txgbe_irq.o \
+              txgbe_fdir.o \
               txgbe_ethtool.o

drivers/net/ethernet/wangxun/txgbe/txgbe_ethtool.c

@@ -9,6 +9,7 @@
 #include "../libwx/wx_type.h"
 #include "../libwx/wx_lib.h"
 #include "txgbe_type.h"
+#include "txgbe_fdir.h"
 #include "txgbe_ethtool.h"
 
 static int txgbe_set_ringparam(struct net_device *netdev,
@@ -79,6 +80,430 @@ static int txgbe_set_channels(struct net_device *dev,
 	return txgbe_setup_tc(dev, netdev_get_num_tc(dev));
 }
 
+static int txgbe_get_ethtool_fdir_entry(struct txgbe *txgbe,
+					struct ethtool_rxnfc *cmd)
+{
+	struct ethtool_rx_flow_spec *fsp =
+		(struct ethtool_rx_flow_spec *)&cmd->fs;
+	union txgbe_atr_input *mask = &txgbe->fdir_mask;
+	struct txgbe_fdir_filter *rule = NULL;
+	struct hlist_node *node;
+
+	/* report total rule count */
+	cmd->data = (1024 << TXGBE_FDIR_PBALLOC_64K) - 2;
+
+	hlist_for_each_entry_safe(rule, node, &txgbe->fdir_filter_list,
+				  fdir_node) {
+		if (fsp->location <= rule->sw_idx)
+			break;
+	}
+
+	if (!rule || fsp->location != rule->sw_idx)
+		return -EINVAL;
+
+	/* set flow type field */
+	switch (rule->filter.formatted.flow_type) {
+	case TXGBE_ATR_FLOW_TYPE_TCPV4:
+		fsp->flow_type = TCP_V4_FLOW;
+		break;
+	case TXGBE_ATR_FLOW_TYPE_UDPV4:
+		fsp->flow_type = UDP_V4_FLOW;
+		break;
+	case TXGBE_ATR_FLOW_TYPE_SCTPV4:
+		fsp->flow_type = SCTP_V4_FLOW;
+		break;
+	case TXGBE_ATR_FLOW_TYPE_IPV4:
+		fsp->flow_type = IP_USER_FLOW;
+		fsp->h_u.usr_ip4_spec.ip_ver = ETH_RX_NFC_IP4;
+		fsp->h_u.usr_ip4_spec.proto = 0;
+		fsp->m_u.usr_ip4_spec.proto = 0;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	fsp->h_u.tcp_ip4_spec.psrc = rule->filter.formatted.src_port;
+	fsp->m_u.tcp_ip4_spec.psrc = mask->formatted.src_port;
+	fsp->h_u.tcp_ip4_spec.pdst = rule->filter.formatted.dst_port;
+	fsp->m_u.tcp_ip4_spec.pdst = mask->formatted.dst_port;
+	fsp->h_u.tcp_ip4_spec.ip4src = rule->filter.formatted.src_ip[0];
+	fsp->m_u.tcp_ip4_spec.ip4src = mask->formatted.src_ip[0];
+	fsp->h_u.tcp_ip4_spec.ip4dst = rule->filter.formatted.dst_ip[0];
+	fsp->m_u.tcp_ip4_spec.ip4dst = mask->formatted.dst_ip[0];
+	fsp->h_ext.vlan_etype = rule->filter.formatted.flex_bytes;
+	fsp->m_ext.vlan_etype = mask->formatted.flex_bytes;
+	fsp->h_ext.data[1] = htonl(rule->filter.formatted.vm_pool);
+	fsp->m_ext.data[1] = htonl(mask->formatted.vm_pool);
+	fsp->flow_type |= FLOW_EXT;
+
+	/* record action */
+	if (rule->action == TXGBE_RDB_FDIR_DROP_QUEUE)
+		fsp->ring_cookie = RX_CLS_FLOW_DISC;
+	else
+		fsp->ring_cookie = rule->action;
+
+	return 0;
+}
+
+static int txgbe_get_ethtool_fdir_all(struct txgbe *txgbe,
+				      struct ethtool_rxnfc *cmd,
+				      u32 *rule_locs)
+{
+	struct txgbe_fdir_filter *rule;
+	struct hlist_node *node;
+	int cnt = 0;
+
+	/* report total rule count */
+	cmd->data = (1024 << TXGBE_FDIR_PBALLOC_64K) - 2;
+
+	hlist_for_each_entry_safe(rule, node, &txgbe->fdir_filter_list,
+				  fdir_node) {
+		if (cnt == cmd->rule_cnt)
+			return -EMSGSIZE;
+		rule_locs[cnt] = rule->sw_idx;
+		cnt++;
+	}
+
+	cmd->rule_cnt = cnt;
+
+	return 0;
+}
+
+static int txgbe_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *cmd,
+			   u32 *rule_locs)
+{
+	struct wx *wx = netdev_priv(dev);
+	struct txgbe *txgbe = wx->priv;
+	int ret = -EOPNOTSUPP;
+
+	switch (cmd->cmd) {
+	case ETHTOOL_GRXRINGS:
+		cmd->data = wx->num_rx_queues;
+		ret = 0;
+		break;
+	case ETHTOOL_GRXCLSRLCNT:
+		cmd->rule_cnt = txgbe->fdir_filter_count;
+		ret = 0;
+		break;
+	case ETHTOOL_GRXCLSRULE:
+		ret = txgbe_get_ethtool_fdir_entry(txgbe, cmd);
+		break;
+	case ETHTOOL_GRXCLSRLALL:
+		ret = txgbe_get_ethtool_fdir_all(txgbe, cmd, (u32 *)rule_locs);
+		break;
+	default:
+		break;
+	}
+
+	return ret;
+}
+
+static int txgbe_flowspec_to_flow_type(struct ethtool_rx_flow_spec *fsp,
+				       u8 *flow_type)
+{
+	switch (fsp->flow_type & ~FLOW_EXT) {
+	case TCP_V4_FLOW:
+		*flow_type = TXGBE_ATR_FLOW_TYPE_TCPV4;
+		break;
+	case UDP_V4_FLOW:
+		*flow_type = TXGBE_ATR_FLOW_TYPE_UDPV4;
+		break;
+	case SCTP_V4_FLOW:
+		*flow_type = TXGBE_ATR_FLOW_TYPE_SCTPV4;
+		break;
+	case IP_USER_FLOW:
+		switch (fsp->h_u.usr_ip4_spec.proto) {
+		case IPPROTO_TCP:
+			*flow_type = TXGBE_ATR_FLOW_TYPE_TCPV4;
+			break;
+		case IPPROTO_UDP:
+			*flow_type = TXGBE_ATR_FLOW_TYPE_UDPV4;
+			break;
+		case IPPROTO_SCTP:
+			*flow_type = TXGBE_ATR_FLOW_TYPE_SCTPV4;
+			break;
+		case 0:
+			if (!fsp->m_u.usr_ip4_spec.proto) {
+				*flow_type = TXGBE_ATR_FLOW_TYPE_IPV4;
+				break;
+			}
+			fallthrough;
+		default:
+			return -EINVAL;
+		}
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static bool txgbe_match_ethtool_fdir_entry(struct txgbe *txgbe,
+					   struct txgbe_fdir_filter *input)
+{
+	struct txgbe_fdir_filter *rule = NULL;
+	struct hlist_node *node2;
+
+	hlist_for_each_entry_safe(rule, node2, &txgbe->fdir_filter_list,
+				  fdir_node) {
+		if (rule->filter.formatted.bkt_hash ==
+		    input->filter.formatted.bkt_hash &&
+		    rule->action == input->action) {
+			wx_dbg(txgbe->wx, "FDIR entry already exist\n");
+			return true;
+		}
+	}
+	return false;
+}
+
+static int txgbe_update_ethtool_fdir_entry(struct txgbe *txgbe,
+					   struct txgbe_fdir_filter *input,
+					   u16 sw_idx)
+{
+	struct hlist_node *node = NULL, *parent = NULL;
+	struct txgbe_fdir_filter *rule;
+	struct wx *wx = txgbe->wx;
+	bool deleted = false;
+	int err;
+
+	hlist_for_each_entry_safe(rule, node, &txgbe->fdir_filter_list,
+				  fdir_node) {
+		/* hash found, or no matching entry */
+		if (rule->sw_idx >= sw_idx)
+			break;
+		parent = node;
+	}
+
+	/* if there is an old rule occupying our place remove it */
+	if (rule && rule->sw_idx == sw_idx) {
+		/* hardware filters are only configured when interface is up,
+		 * and we should not issue filter commands while the interface
+		 * is down
+		 */
+		if (netif_running(wx->netdev) &&
+		    (!input || rule->filter.formatted.bkt_hash !=
+		     input->filter.formatted.bkt_hash)) {
+			err = txgbe_fdir_erase_perfect_filter(wx,
+							      &rule->filter,
+							      sw_idx);
+			if (err)
+				return -EINVAL;
+		}
+
+		hlist_del(&rule->fdir_node);
+		kfree(rule);
+		txgbe->fdir_filter_count--;
+		deleted = true;
+	}
+
+	/* If we weren't given an input, then this was a request to delete a
+	 * filter. We should return -EINVAL if the filter wasn't found, but
+	 * return 0 if the rule was successfully deleted.
+	 */
+	if (!input)
+		return deleted ? 0 : -EINVAL;
+
+	/* initialize node and set software index */
+	INIT_HLIST_NODE(&input->fdir_node);
+
+	/* add filter to the list */
+	if (parent)
+		hlist_add_behind(&input->fdir_node, parent);
+	else
+		hlist_add_head(&input->fdir_node,
+			       &txgbe->fdir_filter_list);
+
+	/* update counts */
+	txgbe->fdir_filter_count++;
+
+	return 0;
+}
+
+static int txgbe_add_ethtool_fdir_entry(struct txgbe *txgbe,
+					struct ethtool_rxnfc *cmd)
+{
+	struct ethtool_rx_flow_spec *fsp =
+		(struct ethtool_rx_flow_spec *)&cmd->fs;
+	struct txgbe_fdir_filter *input;
+	union txgbe_atr_input mask;
+	struct wx *wx = txgbe->wx;
+	int err = -EINVAL;
+	u16 ptype = 0;
+	u8 queue;
+
+	if (!(test_bit(WX_FLAG_FDIR_PERFECT, wx->flags)))
+		return -EOPNOTSUPP;
+
+	/* ring_cookie is a masked into a set of queues and txgbe pools or
+	 * we use drop index
+	 */
+	if (fsp->ring_cookie == RX_CLS_FLOW_DISC) {
+		queue = TXGBE_RDB_FDIR_DROP_QUEUE;
+	} else {
+		u32 ring = ethtool_get_flow_spec_ring(fsp->ring_cookie);
+
+		if (ring >= wx->num_rx_queues)
+			return -EINVAL;
+
+		/* Map the ring onto the absolute queue index */
+		queue = wx->rx_ring[ring]->reg_idx;
+	}
+
+	/* Don't allow indexes to exist outside of available space */
+	if (fsp->location >= ((1024 << TXGBE_FDIR_PBALLOC_64K) - 2)) {
+		wx_err(wx, "Location out of range\n");
+		return -EINVAL;
+	}
+
+	input = kzalloc(sizeof(*input), GFP_ATOMIC);
+	if (!input)
+		return -ENOMEM;
+
+	memset(&mask, 0, sizeof(union txgbe_atr_input));
+
+	/* set SW index */
+	input->sw_idx = fsp->location;
+
+	/* record flow type */
+	if (txgbe_flowspec_to_flow_type(fsp,
+					&input->filter.formatted.flow_type)) {
+		wx_err(wx, "Unrecognized flow type\n");
+		goto err_out;
+	}
+
+	mask.formatted.flow_type = TXGBE_ATR_L4TYPE_IPV6_MASK |
+				   TXGBE_ATR_L4TYPE_MASK;
+
+	if (input->filter.formatted.flow_type == TXGBE_ATR_FLOW_TYPE_IPV4)
+		mask.formatted.flow_type &= TXGBE_ATR_L4TYPE_IPV6_MASK;
+
+	/* Copy input into formatted structures */
+	input->filter.formatted.src_ip[0] = fsp->h_u.tcp_ip4_spec.ip4src;
+	mask.formatted.src_ip[0] = fsp->m_u.tcp_ip4_spec.ip4src;
+	input->filter.formatted.dst_ip[0] = fsp->h_u.tcp_ip4_spec.ip4dst;
+	mask.formatted.dst_ip[0] = fsp->m_u.tcp_ip4_spec.ip4dst;
+	input->filter.formatted.src_port = fsp->h_u.tcp_ip4_spec.psrc;
+	mask.formatted.src_port = fsp->m_u.tcp_ip4_spec.psrc;
+	input->filter.formatted.dst_port = fsp->h_u.tcp_ip4_spec.pdst;
+	mask.formatted.dst_port = fsp->m_u.tcp_ip4_spec.pdst;
+
+	if (fsp->flow_type & FLOW_EXT) {
+		input->filter.formatted.vm_pool =
+				(unsigned char)ntohl(fsp->h_ext.data[1]);
+		mask.formatted.vm_pool =
+				(unsigned char)ntohl(fsp->m_ext.data[1]);
+		input->filter.formatted.flex_bytes =
+						fsp->h_ext.vlan_etype;
+		mask.formatted.flex_bytes = fsp->m_ext.vlan_etype;
+	}
+
+	switch (input->filter.formatted.flow_type) {
+	case TXGBE_ATR_FLOW_TYPE_TCPV4:
+		ptype = WX_PTYPE_L2_IPV4_TCP;
+		break;
+	case TXGBE_ATR_FLOW_TYPE_UDPV4:
+		ptype = WX_PTYPE_L2_IPV4_UDP;
+		break;
+	case TXGBE_ATR_FLOW_TYPE_SCTPV4:
+		ptype = WX_PTYPE_L2_IPV4_SCTP;
+		break;
+	case TXGBE_ATR_FLOW_TYPE_IPV4:
+		ptype = WX_PTYPE_L2_IPV4;
+		break;
+	default:
+		break;
+	}
+
+	input->filter.formatted.vlan_id = htons(ptype);
+	if (mask.formatted.flow_type & TXGBE_ATR_L4TYPE_MASK)
+		mask.formatted.vlan_id = htons(0xFFFF);
+	else
+		mask.formatted.vlan_id = htons(0xFFF8);
+
+	/* determine if we need to drop or route the packet */
+	if (fsp->ring_cookie == RX_CLS_FLOW_DISC)
+		input->action = TXGBE_RDB_FDIR_DROP_QUEUE;
+	else
+		input->action = fsp->ring_cookie;
+
+	spin_lock(&txgbe->fdir_perfect_lock);
+
+	if (hlist_empty(&txgbe->fdir_filter_list)) {
+		/* save mask and program input mask into HW */
+		memcpy(&txgbe->fdir_mask, &mask, sizeof(mask));
+		err = txgbe_fdir_set_input_mask(wx, &mask);
+		if (err)
+			goto err_unlock;
+	} else if (memcmp(&txgbe->fdir_mask, &mask, sizeof(mask))) {
+		wx_err(wx, "Hardware only supports one mask per port. To change the mask you must first delete all the rules.\n");
+		goto err_unlock;
+	}
+
+	/* apply mask and compute/store hash */
+	txgbe_atr_compute_perfect_hash(&input->filter, &mask);
+
+	/* check if new entry does not exist on filter list */
+	if (txgbe_match_ethtool_fdir_entry(txgbe, input))
+		goto err_unlock;
+
+	/* only program filters to hardware if the net device is running, as
+	 * we store the filters in the Rx buffer which is not allocated when
+	 * the device is down
+	 */
+	if (netif_running(wx->netdev)) {
+		err = txgbe_fdir_write_perfect_filter(wx, &input->filter,
+						      input->sw_idx, queue);
+		if (err)
+			goto err_unlock;
+	}
+
+	txgbe_update_ethtool_fdir_entry(txgbe, input, input->sw_idx);
+
+	spin_unlock(&txgbe->fdir_perfect_lock);
+
+	return 0;
+err_unlock:
+	spin_unlock(&txgbe->fdir_perfect_lock);
+err_out:
+	kfree(input);
+	return err;
+}
+
+static int txgbe_del_ethtool_fdir_entry(struct txgbe *txgbe,
+					struct ethtool_rxnfc *cmd)
+{
+	struct ethtool_rx_flow_spec *fsp =
+		(struct ethtool_rx_flow_spec *)&cmd->fs;
+	int err = 0;
+
+	spin_lock(&txgbe->fdir_perfect_lock);
+	err = txgbe_update_ethtool_fdir_entry(txgbe, NULL, fsp->location);
+	spin_unlock(&txgbe->fdir_perfect_lock);
+
+	return err;
+}
+
+static int txgbe_set_rxnfc(struct net_device *dev, struct ethtool_rxnfc *cmd)
+{
+	struct wx *wx = netdev_priv(dev);
+	struct txgbe *txgbe = wx->priv;
+	int ret = -EOPNOTSUPP;
+
+	switch (cmd->cmd) {
+	case ETHTOOL_SRXCLSRLINS:
+		ret = txgbe_add_ethtool_fdir_entry(txgbe, cmd);
+		break;
+	case ETHTOOL_SRXCLSRLDEL:
+		ret = txgbe_del_ethtool_fdir_entry(txgbe, cmd);
+		break;
+	default:
+		break;
+	}
+
+	return ret;
+}
+
 static const struct ethtool_ops txgbe_ethtool_ops = {
 	.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
 				     ETHTOOL_COALESCE_TX_MAX_FRAMES_IRQ,
@@ -100,6 +525,8 @@ static const struct ethtool_ops txgbe_ethtool_ops = {
 	.set_coalesce		= wx_set_coalesce,
 	.get_channels		= wx_get_channels,
 	.set_channels		= txgbe_set_channels,
+	.get_rxnfc		= txgbe_get_rxnfc,
+	.set_rxnfc		= txgbe_set_rxnfc,
 	.get_msglevel		= wx_get_msglevel,
 	.set_msglevel		= wx_set_msglevel,
 };

drivers/net/ethernet/wangxun/txgbe/txgbe_fdir.c

+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2015 - 2024 Beijing WangXun Technology Co., Ltd. */
+
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/pci.h>
+
+#include "../libwx/wx_type.h"
+#include "../libwx/wx_lib.h"
+#include "../libwx/wx_hw.h"
+#include "txgbe_type.h"
+#include "txgbe_fdir.h"
+
+/* These defines allow us to quickly generate all of the necessary instructions
+ * in the function below by simply calling out TXGBE_COMPUTE_SIG_HASH_ITERATION
+ * for values 0 through 15
+ */
+#define TXGBE_ATR_COMMON_HASH_KEY \
+		(TXGBE_ATR_BUCKET_HASH_KEY & TXGBE_ATR_SIGNATURE_HASH_KEY)
+#define TXGBE_COMPUTE_SIG_HASH_ITERATION(_n) \
+do { \
+	u32 n = (_n); \
+	if (TXGBE_ATR_COMMON_HASH_KEY & (0x01 << n)) \
+		common_hash ^= lo_hash_dword >> n; \
+	else if (TXGBE_ATR_BUCKET_HASH_KEY & (0x01 << n)) \
+		bucket_hash ^= lo_hash_dword >> n; \
+	else if (TXGBE_ATR_SIGNATURE_HASH_KEY & (0x01 << n)) \
+		sig_hash ^= lo_hash_dword << (16 - n); \
+	if (TXGBE_ATR_COMMON_HASH_KEY & (0x01 << (n + 16))) \
+		common_hash ^= hi_hash_dword >> n; \
+	else if (TXGBE_ATR_BUCKET_HASH_KEY & (0x01 << (n + 16))) \
+		bucket_hash ^= hi_hash_dword >> n; \
+	else if (TXGBE_ATR_SIGNATURE_HASH_KEY & (0x01 << (n + 16))) \
+		sig_hash ^= hi_hash_dword << (16 - n); \
+} while (0)
+
+/**
+ *  txgbe_atr_compute_sig_hash - Compute the signature hash
+ *  @input: input bitstream to compute the hash on
+ *  @common: compressed common input dword
+ *  @hash: pointer to the computed hash
+ *
+ *  This function is almost identical to the function above but contains
+ *  several optimizations such as unwinding all of the loops, letting the
+ *  compiler work out all of the conditional ifs since the keys are static
+ *  defines, and computing two keys at once since the hashed dword stream
+ *  will be the same for both keys.
+ **/
+static void txgbe_atr_compute_sig_hash(union txgbe_atr_hash_dword input,
+				       union txgbe_atr_hash_dword common,
+				       u32 *hash)
+{
+	u32 sig_hash = 0, bucket_hash = 0, common_hash = 0;
+	u32 hi_hash_dword, lo_hash_dword, flow_vm_vlan;
+	u32 i;
+
+	/* record the flow_vm_vlan bits as they are a key part to the hash */
+	flow_vm_vlan = ntohl(input.dword);
+
+	/* generate common hash dword */
+	hi_hash_dword = ntohl(common.dword);
+
+	/* low dword is word swapped version of common */
+	lo_hash_dword = (hi_hash_dword >> 16) | (hi_hash_dword << 16);
+
+	/* apply flow ID/VM pool/VLAN ID bits to hash words */
+	hi_hash_dword ^= flow_vm_vlan ^ (flow_vm_vlan >> 16);
+
+	/* Process bits 0 and 16 */
+	TXGBE_COMPUTE_SIG_HASH_ITERATION(0);
+
+	/* apply flow ID/VM pool/VLAN ID bits to lo hash dword, we had to
+	 * delay this because bit 0 of the stream should not be processed
+	 * so we do not add the VLAN until after bit 0 was processed
+	 */
+	lo_hash_dword ^= flow_vm_vlan ^ (flow_vm_vlan << 16);
+
+	/* Process remaining 30 bit of the key */
+	for (i = 1; i <= 15; i++)
+		TXGBE_COMPUTE_SIG_HASH_ITERATION(i);
+
+	/* combine common_hash result with signature and bucket hashes */
+	bucket_hash ^= common_hash;
+	bucket_hash &= TXGBE_ATR_HASH_MASK;
+
+	sig_hash ^= common_hash << 16;
+	sig_hash &= TXGBE_ATR_HASH_MASK << 16;
+
+	/* return completed signature hash */
+	*hash = sig_hash ^ bucket_hash;
+}
+
+#define TXGBE_COMPUTE_BKT_HASH_ITERATION(_n) \
+do { \
+	u32 n = (_n); \
+	if (TXGBE_ATR_BUCKET_HASH_KEY & (0x01 << n)) \
+		bucket_hash ^= lo_hash_dword >> n; \
+	if (TXGBE_ATR_BUCKET_HASH_KEY & (0x01 << (n + 16))) \
+		bucket_hash ^= hi_hash_dword >> n; \
+} while (0)
+
+/**
+ *  txgbe_atr_compute_perfect_hash - Compute the perfect filter hash
+ *  @input: input bitstream to compute the hash on
+ *  @input_mask: mask for the input bitstream
+ *
+ *  This function serves two main purposes.  First it applies the input_mask
+ *  to the atr_input resulting in a cleaned up atr_input data stream.
+ *  Secondly it computes the hash and stores it in the bkt_hash field at
+ *  the end of the input byte stream.  This way it will be available for
+ *  future use without needing to recompute the hash.
+ **/
+void txgbe_atr_compute_perfect_hash(union txgbe_atr_input *input,
+				    union txgbe_atr_input *input_mask)
+{
+	u32 hi_hash_dword, lo_hash_dword, flow_vm_vlan;
+	u32 bucket_hash = 0;
+	__be32 hi_dword = 0;
+	u32 i = 0;
+
+	/* Apply masks to input data */
+	for (i = 0; i < 11; i++)
+		input->dword_stream[i] &= input_mask->dword_stream[i];
+
+	/* record the flow_vm_vlan bits as they are a key part to the hash */
+	flow_vm_vlan = ntohl(input->dword_stream[0]);
+
+	/* generate common hash dword */
+	for (i = 1; i <= 10; i++)
+		hi_dword ^= input->dword_stream[i];
+	hi_hash_dword = ntohl(hi_dword);
+
+	/* low dword is word swapped version of common */
+	lo_hash_dword = (hi_hash_dword >> 16) | (hi_hash_dword << 16);
+
+	/* apply flow ID/VM pool/VLAN ID bits to hash words */
+	hi_hash_dword ^= flow_vm_vlan ^ (flow_vm_vlan >> 16);
+
+	/* Process bits 0 and 16 */
+	TXGBE_COMPUTE_BKT_HASH_ITERATION(0);
+
+	/* apply flow ID/VM pool/VLAN ID bits to lo hash dword, we had to
+	 * delay this because bit 0 of the stream should not be processed
+	 * so we do not add the VLAN until after bit 0 was processed
+	 */
+	lo_hash_dword ^= flow_vm_vlan ^ (flow_vm_vlan << 16);
+
+	/* Process remaining 30 bit of the key */
+	for (i = 1; i <= 15; i++)
+		TXGBE_COMPUTE_BKT_HASH_ITERATION(i);
+
+	/* Limit hash to 13 bits since max bucket count is 8K.
+	 * Store result at the end of the input stream.
+	 */
+	input->formatted.bkt_hash = (__force __be16)(bucket_hash & 0x1FFF);
+}
+
+static int txgbe_fdir_check_cmd_complete(struct wx *wx)
+{
+	u32 val;
+
+	return read_poll_timeout_atomic(rd32, val,
+					!(val & TXGBE_RDB_FDIR_CMD_CMD_MASK),
+					10, 100, false,
+					wx, TXGBE_RDB_FDIR_CMD);
+}
+
+/**
+ *  txgbe_fdir_add_signature_filter - Adds a signature hash filter
+ *  @wx: pointer to hardware structure
+ *  @input: unique input dword
+ *  @common: compressed common input dword
+ *  @queue: queue index to direct traffic to
+ *
+ *  @return: 0 on success and negative on failure
+ **/
+static int txgbe_fdir_add_signature_filter(struct wx *wx,
+					   union txgbe_atr_hash_dword input,
+					   union txgbe_atr_hash_dword common,
+					   u8 queue)
+{
+	u32 fdirhashcmd, fdircmd;
+	u8 flow_type;
+	int err;
+
+	/* Get the flow_type in order to program FDIRCMD properly
+	 * lowest 2 bits are FDIRCMD.L4TYPE, third lowest bit is FDIRCMD.IPV6
+	 * fifth is FDIRCMD.TUNNEL_FILTER
+	 */
+	flow_type = input.formatted.flow_type;
+	switch (flow_type) {
+	case TXGBE_ATR_FLOW_TYPE_TCPV4:
+	case TXGBE_ATR_FLOW_TYPE_UDPV4:
+	case TXGBE_ATR_FLOW_TYPE_SCTPV4:
+	case TXGBE_ATR_FLOW_TYPE_TCPV6:
+	case TXGBE_ATR_FLOW_TYPE_UDPV6:
+	case TXGBE_ATR_FLOW_TYPE_SCTPV6:
+		break;
+	default:
+		wx_err(wx, "Error on flow type input\n");
+		return -EINVAL;
+	}
+
+	/* configure FDIRCMD register */
+	fdircmd = TXGBE_RDB_FDIR_CMD_CMD_ADD_FLOW |
+		  TXGBE_RDB_FDIR_CMD_FILTER_UPDATE |
+		  TXGBE_RDB_FDIR_CMD_LAST | TXGBE_RDB_FDIR_CMD_QUEUE_EN;
+	fdircmd |= TXGBE_RDB_FDIR_CMD_FLOW_TYPE(flow_type);
+	fdircmd |= TXGBE_RDB_FDIR_CMD_RX_QUEUE(queue);
+
+	txgbe_atr_compute_sig_hash(input, common, &fdirhashcmd);
+	fdirhashcmd |= TXGBE_RDB_FDIR_HASH_BUCKET_VALID;
+	wr32(wx, TXGBE_RDB_FDIR_HASH, fdirhashcmd);
+	wr32(wx, TXGBE_RDB_FDIR_CMD, fdircmd);
+
+	wx_dbg(wx, "Tx Queue=%x hash=%x\n", queue, (u32)fdirhashcmd);
+
+	err = txgbe_fdir_check_cmd_complete(wx);
+	if (err)
+		wx_err(wx, "Flow Director command did not complete!\n");
+
+	return err;
+}
+
+void txgbe_atr(struct wx_ring *ring, struct wx_tx_buffer *first, u8 ptype)
+{
+	union txgbe_atr_hash_dword common = { .dword = 0 };
+	union txgbe_atr_hash_dword input = { .dword = 0 };
+	struct wx_q_vector *q_vector = ring->q_vector;
+	struct wx_dec_ptype dptype;
+	union network_header {
+		struct ipv6hdr *ipv6;
+		struct iphdr *ipv4;
+		void *raw;
+	} hdr;
+	struct tcphdr *th;
+
+	/* if ring doesn't have a interrupt vector, cannot perform ATR */
+	if (!q_vector)
+		return;
+
+	ring->atr_count++;
+	dptype = wx_decode_ptype(ptype);
+	if (dptype.etype) {
+		if (WX_PTYPE_TYPL4(ptype) != WX_PTYPE_TYP_TCP)
+			return;
+		hdr.raw = (void *)skb_inner_network_header(first->skb);
+		th = inner_tcp_hdr(first->skb);
+	} else {
+		if (WX_PTYPE_PKT(ptype) != WX_PTYPE_PKT_IP ||
+		    WX_PTYPE_TYPL4(ptype) != WX_PTYPE_TYP_TCP)
+			return;
+		hdr.raw = (void *)skb_network_header(first->skb);
+		th = tcp_hdr(first->skb);
+	}
+
+	/* skip this packet since it is invalid or the socket is closing */
+	if (!th || th->fin)
+		return;
+
+	/* sample on all syn packets or once every atr sample count */
+	if (!th->syn && ring->atr_count < ring->atr_sample_rate)
+		return;
+
+	/* reset sample count */
+	ring->atr_count = 0;
+
+	/* src and dst are inverted, think how the receiver sees them
+	 *
+	 * The input is broken into two sections, a non-compressed section
+	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
+	 * is XORed together and stored in the compressed dword.
+	 */
+	input.formatted.vlan_id = htons((u16)ptype);
+
+	/* since src port and flex bytes occupy the same word XOR them together
+	 * and write the value to source port portion of compressed dword
+	 */
+	if (first->tx_flags & WX_TX_FLAGS_SW_VLAN)
+		common.port.src ^= th->dest ^ first->skb->protocol;
+	else if (first->tx_flags & WX_TX_FLAGS_HW_VLAN)
+		common.port.src ^= th->dest ^ first->skb->vlan_proto;
+	else
+		common.port.src ^= th->dest ^ first->protocol;
+	common.port.dst ^= th->source;
+
+	if (WX_PTYPE_PKT_IPV6 & WX_PTYPE_PKT(ptype)) {
+		input.formatted.flow_type = TXGBE_ATR_FLOW_TYPE_TCPV6;
+		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
+					 hdr.ipv6->saddr.s6_addr32[1] ^
+					 hdr.ipv6->saddr.s6_addr32[2] ^
+					 hdr.ipv6->saddr.s6_addr32[3] ^
+					 hdr.ipv6->daddr.s6_addr32[0] ^
+					 hdr.ipv6->daddr.s6_addr32[1] ^
+					 hdr.ipv6->daddr.s6_addr32[2] ^
+					 hdr.ipv6->daddr.s6_addr32[3];
+	} else {
+		input.formatted.flow_type = TXGBE_ATR_FLOW_TYPE_TCPV4;
+		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
+	}
+
+	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
+	txgbe_fdir_add_signature_filter(q_vector->wx, input, common,
+					ring->queue_index);
+}
+
+int txgbe_fdir_set_input_mask(struct wx *wx, union txgbe_atr_input *input_mask)
+{
+	u32 fdirm = 0, fdirtcpm = 0, flex = 0;
+
+	/* Program the relevant mask registers. If src/dst_port or src/dst_addr
+	 * are zero, then assume a full mask for that field.  Also assume that
+	 * a VLAN of 0 is unspecified, so mask that out as well.  L4type
+	 * cannot be masked out in this implementation.
+	 *
+	 * This also assumes IPv4 only.  IPv6 masking isn't supported at this
+	 * point in time.
+	 */
+
+	/* verify bucket hash is cleared on hash generation */
+	if (input_mask->formatted.bkt_hash)
+		wx_dbg(wx, "bucket hash should always be 0 in mask\n");
+
+	/* Program FDIRM and verify partial masks */
+	switch (input_mask->formatted.vm_pool & 0x7F) {
+	case 0x0:
+		fdirm |= TXGBE_RDB_FDIR_OTHER_MSK_POOL;
+		break;
+	case 0x7F:
+		break;
+	default:
+		wx_err(wx, "Error on vm pool mask\n");
+		return -EINVAL;
+	}
+
+	switch (input_mask->formatted.flow_type & TXGBE_ATR_L4TYPE_MASK) {
+	case 0x0:
+		fdirm |= TXGBE_RDB_FDIR_OTHER_MSK_L4P;
+		if (input_mask->formatted.dst_port ||
+		    input_mask->formatted.src_port) {
+			wx_err(wx, "Error on src/dst port mask\n");
+			return -EINVAL;
+		}
+		break;
+	case TXGBE_ATR_L4TYPE_MASK:
+		break;
+	default:
+		wx_err(wx, "Error on flow type mask\n");
+		return -EINVAL;
+	}
+
+	/* Now mask VM pool and destination IPv6 - bits 5 and 2 */
+	wr32(wx, TXGBE_RDB_FDIR_OTHER_MSK, fdirm);
+
+	flex = rd32(wx, TXGBE_RDB_FDIR_FLEX_CFG(0));
+	flex &= ~TXGBE_RDB_FDIR_FLEX_CFG_FIELD0;
+	flex |= (TXGBE_RDB_FDIR_FLEX_CFG_BASE_MAC |
+		 TXGBE_RDB_FDIR_FLEX_CFG_OFST(0x6));
+
+	switch ((__force u16)input_mask->formatted.flex_bytes & 0xFFFF) {
+	case 0x0000:
+		/* Mask Flex Bytes */
+		flex |= TXGBE_RDB_FDIR_FLEX_CFG_MSK;
+		break;
+	case 0xFFFF:
+		break;
+	default:
+		wx_err(wx, "Error on flexible byte mask\n");
+		return -EINVAL;
+	}
+	wr32(wx, TXGBE_RDB_FDIR_FLEX_CFG(0), flex);
+
+	/* store the TCP/UDP port masks, bit reversed from port layout */
+	fdirtcpm = ntohs(input_mask->formatted.dst_port);
+	fdirtcpm <<= TXGBE_RDB_FDIR_PORT_DESTINATION_SHIFT;
+	fdirtcpm |= ntohs(input_mask->formatted.src_port);
+
+	/* write both the same so that UDP and TCP use the same mask */
+	wr32(wx, TXGBE_RDB_FDIR_TCP_MSK, ~fdirtcpm);
+	wr32(wx, TXGBE_RDB_FDIR_UDP_MSK, ~fdirtcpm);
+	wr32(wx, TXGBE_RDB_FDIR_SCTP_MSK, ~fdirtcpm);
+
+	/* store source and destination IP masks (little-enian) */
+	wr32(wx, TXGBE_RDB_FDIR_SA4_MSK,
+	     ntohl(~input_mask->formatted.src_ip[0]));
+	wr32(wx, TXGBE_RDB_FDIR_DA4_MSK,
+	     ntohl(~input_mask->formatted.dst_ip[0]));
+
+	return 0;
+}
+
+int txgbe_fdir_write_perfect_filter(struct wx *wx,
+				    union txgbe_atr_input *input,
+				    u16 soft_id, u8 queue)
+{
+	u32 fdirport, fdirvlan, fdirhash, fdircmd;
+	int err = 0;
+
+	/* currently IPv6 is not supported, must be programmed with 0 */
+	wr32(wx, TXGBE_RDB_FDIR_IP6(2), ntohl(input->formatted.src_ip[0]));
+	wr32(wx, TXGBE_RDB_FDIR_IP6(1), ntohl(input->formatted.src_ip[1]));
+	wr32(wx, TXGBE_RDB_FDIR_IP6(0), ntohl(input->formatted.src_ip[2]));
+
+	/* record the source address (little-endian) */
+	wr32(wx, TXGBE_RDB_FDIR_SA, ntohl(input->formatted.src_ip[0]));
+
+	/* record the first 32 bits of the destination address
+	 * (little-endian)
+	 */
+	wr32(wx, TXGBE_RDB_FDIR_DA, ntohl(input->formatted.dst_ip[0]));
+
+	/* record source and destination port (little-endian)*/
+	fdirport = ntohs(input->formatted.dst_port);
+	fdirport <<= TXGBE_RDB_FDIR_PORT_DESTINATION_SHIFT;
+	fdirport |= ntohs(input->formatted.src_port);
+	wr32(wx, TXGBE_RDB_FDIR_PORT, fdirport);
+
+	/* record packet type and flex_bytes (little-endian) */
+	fdirvlan = ntohs(input->formatted.flex_bytes);
+	fdirvlan <<= TXGBE_RDB_FDIR_FLEX_FLEX_SHIFT;
+	fdirvlan |= ntohs(input->formatted.vlan_id);
+	wr32(wx, TXGBE_RDB_FDIR_FLEX, fdirvlan);
+
+	/* configure FDIRHASH register */
+	fdirhash = (__force u32)input->formatted.bkt_hash |
+		   TXGBE_RDB_FDIR_HASH_BUCKET_VALID |
+		   TXGBE_RDB_FDIR_HASH_SIG_SW_INDEX(soft_id);
+	wr32(wx, TXGBE_RDB_FDIR_HASH, fdirhash);
+
+	/* flush all previous writes to make certain registers are
+	 * programmed prior to issuing the command
+	 */
+	WX_WRITE_FLUSH(wx);
+
+	/* configure FDIRCMD register */
+	fdircmd = TXGBE_RDB_FDIR_CMD_CMD_ADD_FLOW |
+		  TXGBE_RDB_FDIR_CMD_FILTER_UPDATE |
+		  TXGBE_RDB_FDIR_CMD_LAST | TXGBE_RDB_FDIR_CMD_QUEUE_EN;
+	if (queue == TXGBE_RDB_FDIR_DROP_QUEUE)
+		fdircmd |= TXGBE_RDB_FDIR_CMD_DROP;
+	fdircmd |= TXGBE_RDB_FDIR_CMD_FLOW_TYPE(input->formatted.flow_type);
+	fdircmd |= TXGBE_RDB_FDIR_CMD_RX_QUEUE(queue);
+	fdircmd |= TXGBE_RDB_FDIR_CMD_VT_POOL(input->formatted.vm_pool);
+
+	wr32(wx, TXGBE_RDB_FDIR_CMD, fdircmd);
+	err = txgbe_fdir_check_cmd_complete(wx);
+	if (err)
+		wx_err(wx, "Flow Director command did not complete!\n");
+
+	return err;
+}
+
+int txgbe_fdir_erase_perfect_filter(struct wx *wx,
+				    union txgbe_atr_input *input,
+				    u16 soft_id)
+{
+	u32 fdirhash, fdircmd;
+	int err = 0;
+
+	/* configure FDIRHASH register */
+	fdirhash = (__force u32)input->formatted.bkt_hash;
+	fdirhash |= TXGBE_RDB_FDIR_HASH_SIG_SW_INDEX(soft_id);
+	wr32(wx, TXGBE_RDB_FDIR_HASH, fdirhash);
+
+	/* flush hash to HW */
+	WX_WRITE_FLUSH(wx);
+
+	/* Query if filter is present */
+	wr32(wx, TXGBE_RDB_FDIR_CMD, TXGBE_RDB_FDIR_CMD_CMD_QUERY_REM_FILT);
+
+	err = txgbe_fdir_check_cmd_complete(wx);
+	if (err) {
+		wx_err(wx, "Flow Director command did not complete!\n");
+		return err;
+	}
+
+	fdircmd = rd32(wx, TXGBE_RDB_FDIR_CMD);
+	/* if filter exists in hardware then remove it */
+	if (fdircmd & TXGBE_RDB_FDIR_CMD_FILTER_VALID) {
+		wr32(wx, TXGBE_RDB_FDIR_HASH, fdirhash);
+		WX_WRITE_FLUSH(wx);
+		wr32(wx, TXGBE_RDB_FDIR_CMD,
+		     TXGBE_RDB_FDIR_CMD_CMD_REMOVE_FLOW);
+	}
+
+	return 0;
+}
+
+/**
+ *  txgbe_fdir_enable - Initialize Flow Director control registers
+ *  @wx: pointer to hardware structure
+ *  @fdirctrl: value to write to flow director control register
+ **/
+static void txgbe_fdir_enable(struct wx *wx, u32 fdirctrl)
+{
+	u32 val;
+	int ret;
+
+	/* Prime the keys for hashing */
+	wr32(wx, TXGBE_RDB_FDIR_HKEY, TXGBE_ATR_BUCKET_HASH_KEY);
+	wr32(wx, TXGBE_RDB_FDIR_SKEY, TXGBE_ATR_SIGNATURE_HASH_KEY);
+
+	wr32(wx, TXGBE_RDB_FDIR_CTL, fdirctrl);
+	WX_WRITE_FLUSH(wx);
+	ret = read_poll_timeout(rd32, val, val & TXGBE_RDB_FDIR_CTL_INIT_DONE,
+				1000, 10000, false, wx, TXGBE_RDB_FDIR_CTL);
+
+	if (ret < 0)
+		wx_dbg(wx, "Flow Director poll time exceeded!\n");
+}
+
+/**
+ *  txgbe_init_fdir_signature -Initialize Flow Director sig filters
+ *  @wx: pointer to hardware structure
+ **/
+static void txgbe_init_fdir_signature(struct wx *wx)
+{
+	u32 fdirctrl = TXGBE_FDIR_PBALLOC_64K;
+	u32 flex = 0;
+
+	flex = rd32(wx, TXGBE_RDB_FDIR_FLEX_CFG(0));
+	flex &= ~TXGBE_RDB_FDIR_FLEX_CFG_FIELD0;
+
+	flex |= (TXGBE_RDB_FDIR_FLEX_CFG_BASE_MAC |
+		 TXGBE_RDB_FDIR_FLEX_CFG_OFST(0x6));
+	wr32(wx, TXGBE_RDB_FDIR_FLEX_CFG(0), flex);
+
+	/* Continue setup of fdirctrl register bits:
+	 *  Move the flexible bytes to use the ethertype - shift 6 words
+	 *  Set the maximum length per hash bucket to 0xA filters
+	 *  Send interrupt when 64 filters are left
+	 */
+	fdirctrl |= TXGBE_RDB_FDIR_CTL_HASH_BITS(0xF) |
+		    TXGBE_RDB_FDIR_CTL_MAX_LENGTH(0xA) |
+		    TXGBE_RDB_FDIR_CTL_FULL_THRESH(4);
+
+	/* write hashes and fdirctrl register, poll for completion */
+	txgbe_fdir_enable(wx, fdirctrl);
+}
+
+/**
+ *  txgbe_init_fdir_perfect - Initialize Flow Director perfect filters
+ *  @wx: pointer to hardware structure
+ **/
+static void txgbe_init_fdir_perfect(struct wx *wx)
+{
+	u32 fdirctrl = TXGBE_FDIR_PBALLOC_64K;
+
+	/* Continue setup of fdirctrl register bits:
+	 *  Turn perfect match filtering on
+	 *  Report hash in RSS field of Rx wb descriptor
+	 *  Initialize the drop queue
+	 *  Move the flexible bytes to use the ethertype - shift 6 words
+	 *  Set the maximum length per hash bucket to 0xA filters
+	 *  Send interrupt when 64 (0x4 * 16) filters are left
+	 */
+	fdirctrl |= TXGBE_RDB_FDIR_CTL_PERFECT_MATCH |
+		    TXGBE_RDB_FDIR_CTL_DROP_Q(TXGBE_RDB_FDIR_DROP_QUEUE) |
+		    TXGBE_RDB_FDIR_CTL_HASH_BITS(0xF) |
+		    TXGBE_RDB_FDIR_CTL_MAX_LENGTH(0xA) |
+		    TXGBE_RDB_FDIR_CTL_FULL_THRESH(4);
+
+	/* write hashes and fdirctrl register, poll for completion */
+	txgbe_fdir_enable(wx, fdirctrl);
+}
+
+static void txgbe_fdir_filter_restore(struct wx *wx)
+{
+	struct txgbe_fdir_filter *filter;
+	struct txgbe *txgbe = wx->priv;
+	struct hlist_node *node;
+	u8 queue = 0;
+	int ret = 0;
+
+	spin_lock(&txgbe->fdir_perfect_lock);
+
+	if (!hlist_empty(&txgbe->fdir_filter_list))
+		ret = txgbe_fdir_set_input_mask(wx, &txgbe->fdir_mask);
+
+	if (ret)
+		goto unlock;
+
+	hlist_for_each_entry_safe(filter, node,
+				  &txgbe->fdir_filter_list, fdir_node) {
+		if (filter->action == TXGBE_RDB_FDIR_DROP_QUEUE) {
+			queue = TXGBE_RDB_FDIR_DROP_QUEUE;
+		} else {
+			u32 ring = ethtool_get_flow_spec_ring(filter->action);
+
+			if (ring >= wx->num_rx_queues) {
+				wx_err(wx, "FDIR restore failed, ring:%u\n",
+				       ring);
+				continue;
+			}
+
+			/* Map the ring onto the absolute queue index */
+			queue = wx->rx_ring[ring]->reg_idx;
+		}
+
+		ret = txgbe_fdir_write_perfect_filter(wx,
+						      &filter->filter,
+						      filter->sw_idx,
+						      queue);
+		if (ret)
+			wx_err(wx, "FDIR restore failed, index:%u\n",
+			       filter->sw_idx);
+	}
+
+unlock:
+	spin_unlock(&txgbe->fdir_perfect_lock);
+}
+
+void txgbe_configure_fdir(struct wx *wx)
+{
+	wx_disable_sec_rx_path(wx);
+
+	if (test_bit(WX_FLAG_FDIR_HASH, wx->flags)) {
+		txgbe_init_fdir_signature(wx);
+	} else if (test_bit(WX_FLAG_FDIR_PERFECT, wx->flags)) {
+		txgbe_init_fdir_perfect(wx);
+		txgbe_fdir_filter_restore(wx);
+	}
+
+	wx_enable_sec_rx_path(wx);
+}
+
+void txgbe_fdir_filter_exit(struct wx *wx)
+{
+	struct txgbe_fdir_filter *filter;
+	struct txgbe *txgbe = wx->priv;
+	struct hlist_node *node;
+
+	spin_lock(&txgbe->fdir_perfect_lock);
+
+	hlist_for_each_entry_safe(filter, node,
+				  &txgbe->fdir_filter_list, fdir_node) {
+		hlist_del(&filter->fdir_node);
+		kfree(filter);
+	}
+	txgbe->fdir_filter_count = 0;
+
+	spin_unlock(&txgbe->fdir_perfect_lock);
+}

drivers/net/ethernet/wangxun/txgbe/txgbe_fdir.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (c) 2015 - 2024 Beijing WangXun Technology Co., Ltd. */
+
+#ifndef _TXGBE_FDIR_H_
+#define _TXGBE_FDIR_H_
+
+void txgbe_atr_compute_perfect_hash(union txgbe_atr_input *input,
+				    union txgbe_atr_input *input_mask);
+void txgbe_atr(struct wx_ring *ring, struct wx_tx_buffer *first, u8 ptype);
+int txgbe_fdir_set_input_mask(struct wx *wx, union txgbe_atr_input *input_mask);
+int txgbe_fdir_write_perfect_filter(struct wx *wx,
+				    union txgbe_atr_input *input,
+				    u16 soft_id, u8 queue);
+int txgbe_fdir_erase_perfect_filter(struct wx *wx,
+				    union txgbe_atr_input *input,
+				    u16 soft_id);
+void txgbe_configure_fdir(struct wx *wx);
+void txgbe_fdir_filter_exit(struct wx *wx);
+
+#endif /* _TXGBE_FDIR_H_ */

drivers/net/ethernet/wangxun/txgbe/txgbe_main.c

@@ -18,6 +18,7 @@
 #include "txgbe_hw.h"
 #include "txgbe_phy.h"
 #include "txgbe_irq.h"
+#include "txgbe_fdir.h"
 #include "txgbe_ethtool.h"
 
 char txgbe_driver_name[] = "txgbe";
@@ -257,6 +258,14 @@ static int txgbe_sw_init(struct wx *wx)
 						   num_online_cpus());
 	wx->rss_enabled = true;
 
+	wx->ring_feature[RING_F_FDIR].limit = min_t(int, TXGBE_MAX_FDIR_INDICES,
+						    num_online_cpus());
+	set_bit(WX_FLAG_FDIR_CAPABLE, wx->flags);
+	set_bit(WX_FLAG_FDIR_HASH, wx->flags);
+	wx->atr_sample_rate = TXGBE_DEFAULT_ATR_SAMPLE_RATE;
+	wx->atr = txgbe_atr;
+	wx->configure_fdir = txgbe_configure_fdir;
+
 	/* enable itr by default in dynamic mode */
 	wx->rx_itr_setting = 1;
 	wx->tx_itr_setting = 1;
@@ -274,6 +283,12 @@ static int txgbe_sw_init(struct wx *wx)
 	return 0;
 }
 
+static void txgbe_init_fdir(struct txgbe *txgbe)
+{
+	txgbe->fdir_filter_count = 0;
+	spin_lock_init(&txgbe->fdir_perfect_lock);
+}
+
 /**
  * txgbe_open - Called when a network interface is made active
  * @netdev: network interface device structure
@@ -352,6 +367,7 @@ static int txgbe_close(struct net_device *netdev)
 	txgbe_down(wx);
 	wx_free_irq(wx);
 	wx_free_resources(wx);
+	txgbe_fdir_filter_exit(wx);
 	wx_control_hw(wx, false);
 
 	return 0;
@@ -660,6 +676,8 @@ static int txgbe_probe(struct pci_dev *pdev,
 	txgbe->wx = wx;
 	wx->priv = txgbe;
 
+	txgbe_init_fdir(txgbe);
+
 	err = txgbe_setup_misc_irq(txgbe);
 	if (err)
 		goto err_release_hw;

drivers/net/ethernet/wangxun/txgbe/txgbe_type.h

@@ -89,6 +89,55 @@
 #define TXGBE_XPCS_IDA_ADDR                     0x13000
 #define TXGBE_XPCS_IDA_DATA                     0x13004
 
+/********************************* Flow Director *****************************/
+#define TXGBE_RDB_FDIR_DROP_QUEUE               127
+#define TXGBE_RDB_FDIR_CTL                      0x19500
+#define TXGBE_RDB_FDIR_CTL_INIT_DONE            BIT(3)
+#define TXGBE_RDB_FDIR_CTL_PERFECT_MATCH        BIT(4)
+#define TXGBE_RDB_FDIR_CTL_DROP_Q(v)            FIELD_PREP(GENMASK(14, 8), v)
+#define TXGBE_RDB_FDIR_CTL_HASH_BITS(v)         FIELD_PREP(GENMASK(23, 20), v)
+#define TXGBE_RDB_FDIR_CTL_MAX_LENGTH(v)        FIELD_PREP(GENMASK(27, 24), v)
+#define TXGBE_RDB_FDIR_CTL_FULL_THRESH(v)       FIELD_PREP(GENMASK(31, 28), v)
+#define TXGBE_RDB_FDIR_IP6(_i)                  (0x1950C + ((_i) * 4)) /* 0-2 */
+#define TXGBE_RDB_FDIR_SA                       0x19518
+#define TXGBE_RDB_FDIR_DA                       0x1951C
+#define TXGBE_RDB_FDIR_PORT                     0x19520
+#define TXGBE_RDB_FDIR_PORT_DESTINATION_SHIFT   16
+#define TXGBE_RDB_FDIR_FLEX                     0x19524
+#define TXGBE_RDB_FDIR_FLEX_FLEX_SHIFT          16
+#define TXGBE_RDB_FDIR_HASH                     0x19528
+#define TXGBE_RDB_FDIR_HASH_SIG_SW_INDEX(v)     FIELD_PREP(GENMASK(31, 16), v)
+#define TXGBE_RDB_FDIR_HASH_BUCKET_VALID        BIT(15)
+#define TXGBE_RDB_FDIR_CMD                      0x1952C
+#define TXGBE_RDB_FDIR_CMD_CMD_MASK             GENMASK(1, 0)
+#define TXGBE_RDB_FDIR_CMD_CMD(v)               FIELD_PREP(GENMASK(1, 0), v)
+#define TXGBE_RDB_FDIR_CMD_CMD_ADD_FLOW         TXGBE_RDB_FDIR_CMD_CMD(1)
+#define TXGBE_RDB_FDIR_CMD_CMD_REMOVE_FLOW      TXGBE_RDB_FDIR_CMD_CMD(2)
+#define TXGBE_RDB_FDIR_CMD_CMD_QUERY_REM_FILT   TXGBE_RDB_FDIR_CMD_CMD(3)
+#define TXGBE_RDB_FDIR_CMD_FILTER_VALID         BIT(2)
+#define TXGBE_RDB_FDIR_CMD_FILTER_UPDATE        BIT(3)
+#define TXGBE_RDB_FDIR_CMD_FLOW_TYPE(v)         FIELD_PREP(GENMASK(6, 5), v)
+#define TXGBE_RDB_FDIR_CMD_DROP                 BIT(9)
+#define TXGBE_RDB_FDIR_CMD_LAST                 BIT(11)
+#define TXGBE_RDB_FDIR_CMD_QUEUE_EN             BIT(15)
+#define TXGBE_RDB_FDIR_CMD_RX_QUEUE(v)          FIELD_PREP(GENMASK(22, 16), v)
+#define TXGBE_RDB_FDIR_CMD_VT_POOL(v)           FIELD_PREP(GENMASK(29, 24), v)
+#define TXGBE_RDB_FDIR_DA4_MSK                  0x1953C
+#define TXGBE_RDB_FDIR_SA4_MSK                  0x19540
+#define TXGBE_RDB_FDIR_TCP_MSK                  0x19544
+#define TXGBE_RDB_FDIR_UDP_MSK                  0x19548
+#define TXGBE_RDB_FDIR_SCTP_MSK                 0x19560
+#define TXGBE_RDB_FDIR_HKEY                     0x19568
+#define TXGBE_RDB_FDIR_SKEY                     0x1956C
+#define TXGBE_RDB_FDIR_OTHER_MSK                0x19570
+#define TXGBE_RDB_FDIR_OTHER_MSK_POOL           BIT(2)
+#define TXGBE_RDB_FDIR_OTHER_MSK_L4P            BIT(3)
+#define TXGBE_RDB_FDIR_FLEX_CFG(_i)             (0x19580 + ((_i) * 4))
+#define TXGBE_RDB_FDIR_FLEX_CFG_FIELD0          GENMASK(7, 0)
+#define TXGBE_RDB_FDIR_FLEX_CFG_BASE_MAC        FIELD_PREP(GENMASK(1, 0), 0)
+#define TXGBE_RDB_FDIR_FLEX_CFG_MSK             BIT(2)
+#define TXGBE_RDB_FDIR_FLEX_CFG_OFST(v)         FIELD_PREP(GENMASK(7, 3), v)
+
 /* Checksum and EEPROM pointers */
 #define TXGBE_EEPROM_LAST_WORD                  0x800
 #define TXGBE_EEPROM_CHECKSUM                   0x2F
@@ -112,6 +161,98 @@
 #define TXGBE_SP_RX_PB_SIZE     512
 #define TXGBE_SP_TDB_PB_SZ      (160 * 1024) /* 160KB Packet Buffer */
 
+#define TXGBE_DEFAULT_ATR_SAMPLE_RATE           20
+
+/* Software ATR hash keys */
+#define TXGBE_ATR_BUCKET_HASH_KEY               0x3DAD14E2
+#define TXGBE_ATR_SIGNATURE_HASH_KEY            0x174D3614
+
+/* Software ATR input stream values and masks */
+#define TXGBE_ATR_HASH_MASK                     0x7fff
+#define TXGBE_ATR_L4TYPE_MASK                   0x3
+#define TXGBE_ATR_L4TYPE_UDP                    0x1
+#define TXGBE_ATR_L4TYPE_TCP                    0x2
+#define TXGBE_ATR_L4TYPE_SCTP                   0x3
+#define TXGBE_ATR_L4TYPE_IPV6_MASK              0x4
+#define TXGBE_ATR_L4TYPE_TUNNEL_MASK            0x10
+
+enum txgbe_atr_flow_type {
+	TXGBE_ATR_FLOW_TYPE_IPV4                = 0x0,
+	TXGBE_ATR_FLOW_TYPE_UDPV4               = 0x1,
+	TXGBE_ATR_FLOW_TYPE_TCPV4               = 0x2,
+	TXGBE_ATR_FLOW_TYPE_SCTPV4              = 0x3,
+	TXGBE_ATR_FLOW_TYPE_IPV6                = 0x4,
+	TXGBE_ATR_FLOW_TYPE_UDPV6               = 0x5,
+	TXGBE_ATR_FLOW_TYPE_TCPV6               = 0x6,
+	TXGBE_ATR_FLOW_TYPE_SCTPV6              = 0x7,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_IPV4       = 0x10,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_UDPV4      = 0x11,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_TCPV4      = 0x12,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_SCTPV4     = 0x13,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_IPV6       = 0x14,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_UDPV6      = 0x15,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_TCPV6      = 0x16,
+	TXGBE_ATR_FLOW_TYPE_TUNNELED_SCTPV6     = 0x17,
+};
+
+/* Flow Director ATR input struct. */
+union txgbe_atr_input {
+	/* Byte layout in order, all values with MSB first:
+	 *
+	 * vm_pool    - 1 byte
+	 * flow_type  - 1 byte
+	 * vlan_id    - 2 bytes
+	 * dst_ip     - 16 bytes
+	 * src_ip     - 16 bytes
+	 * src_port   - 2 bytes
+	 * dst_port   - 2 bytes
+	 * flex_bytes - 2 bytes
+	 * bkt_hash   - 2 bytes
+	 */
+	struct {
+		u8 vm_pool;
+		u8 flow_type;
+		__be16 vlan_id;
+		__be32 dst_ip[4];
+		__be32 src_ip[4];
+		__be16 src_port;
+		__be16 dst_port;
+		__be16 flex_bytes;
+		__be16 bkt_hash;
+	} formatted;
+	__be32 dword_stream[11];
+};
+
+/* Flow Director compressed ATR hash input struct */
+union txgbe_atr_hash_dword {
+	struct {
+		u8 vm_pool;
+		u8 flow_type;
+		__be16 vlan_id;
+	} formatted;
+	__be32 ip;
+	struct {
+		__be16 src;
+		__be16 dst;
+	} port;
+	__be16 flex_bytes;
+	__be32 dword;
+};
+
+enum txgbe_fdir_pballoc_type {
+	TXGBE_FDIR_PBALLOC_NONE = 0,
+	TXGBE_FDIR_PBALLOC_64K  = 1,
+	TXGBE_FDIR_PBALLOC_128K = 2,
+	TXGBE_FDIR_PBALLOC_256K = 3,
+};
+
+struct txgbe_fdir_filter {
+	struct hlist_node fdir_node;
+	union txgbe_atr_input filter;
+	u16 sw_idx;
+	u16 action;
+};
+
 /* TX/RX descriptor defines */
 #define TXGBE_DEFAULT_TXD               512
 #define TXGBE_DEFAULT_TX_WORK           256
@@ -196,6 +337,12 @@ struct txgbe {
 	struct gpio_chip *gpio;
 	unsigned int gpio_irq;
 	unsigned int link_irq;
+
+	/* flow director */
+	struct hlist_head fdir_filter_list;
+	union txgbe_atr_input fdir_mask;
+	int fdir_filter_count;
+	spinlock_t fdir_perfect_lock; /* spinlock for FDIR */
 };
 
 #endif /* _TXGBE_TYPE_H_ */