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.com
Signed-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;
 	}
@@ -70,13 +79,18 @@ 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;
 
 	switch (stringset) {
 	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;
 
@@ -1271,18 +1278,20 @@ 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
  *  @wx: pointer to private structure
  *
  *  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,