ixgbe: cleanup flow director hash computation to improve performance

This change cleans up the layout of the flow director data, and the
algorithm used to calculate the hash resulting in a 35x / 3500% performance
increase versus the old flow director hash computation.  The overall effect
is only a 1% increase in transactions per second though due to the fact
that only 1 packet in 20 are actually hashed upon.

TCP_RR before:
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       60.00    23059.27
16384  87380

TCP_RR after:
Socket Size   Request  Resp.   Elapsed  Trans.
Send   Recv   Size     Size    Time     Rate
bytes  Bytes  bytes    bytes   secs.    per sec

16384  87380  1        1       60.00    23239.98
16384  87380
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ixgbe/ixgbe.h

@@ -526,25 +526,25 @@ extern s32 ixgbe_reinit_fdir_tables_82599(struct ixgbe_hw *hw);
 extern s32 ixgbe_init_fdir_signature_82599(struct ixgbe_hw *hw, u32 pballoc);
 extern s32 ixgbe_init_fdir_perfect_82599(struct ixgbe_hw *hw, u32 pballoc);
 extern s32 ixgbe_fdir_add_signature_filter_82599(struct ixgbe_hw *hw,
-                                                 struct ixgbe_atr_input *input,
+                                                 union ixgbe_atr_input *input,
                                                  u8 queue);
 extern s32 ixgbe_fdir_add_perfect_filter_82599(struct ixgbe_hw *hw,
-                                      struct ixgbe_atr_input *input,
+                                      union ixgbe_atr_input *input,
                                       struct ixgbe_atr_input_masks *input_masks,
                                       u16 soft_id, u8 queue);
-extern s32 ixgbe_atr_set_vlan_id_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_vlan_id_82599(union ixgbe_atr_input *input,
                                        u16 vlan_id);
-extern s32 ixgbe_atr_set_src_ipv4_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_src_ipv4_82599(union ixgbe_atr_input *input,
                                         u32 src_addr);
-extern s32 ixgbe_atr_set_dst_ipv4_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_dst_ipv4_82599(union ixgbe_atr_input *input,
                                         u32 dst_addr);
-extern s32 ixgbe_atr_set_src_port_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_src_port_82599(union ixgbe_atr_input *input,
                                         u16 src_port);
-extern s32 ixgbe_atr_set_dst_port_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_dst_port_82599(union ixgbe_atr_input *input,
                                         u16 dst_port);
-extern s32 ixgbe_atr_set_flex_byte_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_flex_byte_82599(union ixgbe_atr_input *input,
                                          u16 flex_byte);
-extern s32 ixgbe_atr_set_l4type_82599(struct ixgbe_atr_input *input,
+extern s32 ixgbe_atr_set_l4type_82599(union ixgbe_atr_input *input,
                                       u8 l4type);
 extern void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
                                    struct ixgbe_ring *ring);

drivers/net/ixgbe/ixgbe_ethtool.c

@@ -2278,7 +2278,7 @@ static int ixgbe_set_rx_ntuple(struct net_device *dev,
 {
 	struct ixgbe_adapter *adapter = netdev_priv(dev);
 	struct ethtool_rx_ntuple_flow_spec fs = cmd->fs;
-	struct ixgbe_atr_input input_struct;
+	union ixgbe_atr_input input_struct;
 	struct ixgbe_atr_input_masks input_masks;
 	int target_queue;
 
@@ -2293,7 +2293,7 @@ static int ixgbe_set_rx_ntuple(struct net_device *dev,
 	    (fs.action < ETHTOOL_RXNTUPLE_ACTION_DROP))
 		return -EINVAL;
 
-	memset(&input_struct, 0, sizeof(struct ixgbe_atr_input));
+	memset(&input_struct, 0, sizeof(union ixgbe_atr_input));
 	memset(&input_masks, 0, sizeof(struct ixgbe_atr_input_masks));
 
 	input_masks.src_ip_mask = fs.m_u.tcp_ip4_spec.ip4src;

drivers/net/ixgbe/ixgbe_main.c

@@ -6509,21 +6509,20 @@ static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
 		      u8 queue, u32 tx_flags, __be16 protocol)
 {
-	struct ixgbe_atr_input atr_input;
+	union ixgbe_atr_input atr_input;
 	struct iphdr *iph = ip_hdr(skb);
 	struct ethhdr *eth = (struct ethhdr *)skb->data;
 	struct tcphdr *th;
-	u16 vlan_id;
+	__be16 vlan_id;
 
 	/* Right now, we support IPv4 w/ TCP only */
 	if (protocol != htons(ETH_P_IP) ||
 	    iph->protocol != IPPROTO_TCP)
 		return;
 
-	memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
+	memset(&atr_input, 0, sizeof(union ixgbe_atr_input));
 
-	vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
-		   IXGBE_TX_FLAGS_VLAN_SHIFT;
+	vlan_id = htons(tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
 
 	th = tcp_hdr(skb);
 
@@ -6531,7 +6530,7 @@ static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
 	ixgbe_atr_set_src_port_82599(&atr_input, th->dest);
 	ixgbe_atr_set_dst_port_82599(&atr_input, th->source);
 	ixgbe_atr_set_flex_byte_82599(&atr_input, eth->h_proto);
-	ixgbe_atr_set_l4type_82599(&atr_input, IXGBE_ATR_L4TYPE_TCP);
+	ixgbe_atr_set_l4type_82599(&atr_input, IXGBE_ATR_FLOW_TYPE_TCPV4);
 	/* src and dst are inverted, think how the receiver sees them */
 	ixgbe_atr_set_src_ipv4_82599(&atr_input, iph->daddr);
 	ixgbe_atr_set_dst_ipv4_82599(&atr_input, iph->saddr);

drivers/net/ixgbe/ixgbe_type.h

@@ -1990,6 +1990,7 @@ enum ixgbe_fdir_pballoc_type {
 #define IXGBE_FDIRCMD_LAST                      0x00000800
 #define IXGBE_FDIRCMD_COLLISION                 0x00001000
 #define IXGBE_FDIRCMD_QUEUE_EN                  0x00008000
+#define IXGBE_FDIRCMD_FLOW_TYPE_SHIFT           5
 #define IXGBE_FDIRCMD_RX_QUEUE_SHIFT            16
 #define IXGBE_FDIRCMD_VT_POOL_SHIFT             24
 #define IXGBE_FDIR_INIT_DONE_POLL               10
@@ -2147,51 +2148,63 @@ typedef u32 ixgbe_physical_layer;
 #define FC_LOW_WATER(MTU)  (2 * (2 * PAUSE_MTU(MTU) + PAUSE_RTT))
 
 /* Software ATR hash keys */
-#define IXGBE_ATR_BUCKET_HASH_KEY    0xE214AD3D
-#define IXGBE_ATR_SIGNATURE_HASH_KEY 0x14364D17
-
-/* Software ATR input stream offsets and masks */
-#define IXGBE_ATR_VLAN_OFFSET       0
-#define IXGBE_ATR_SRC_IPV6_OFFSET   2
-#define IXGBE_ATR_SRC_IPV4_OFFSET  14
-#define IXGBE_ATR_DST_IPV6_OFFSET  18
-#define IXGBE_ATR_DST_IPV4_OFFSET  30
-#define IXGBE_ATR_SRC_PORT_OFFSET  34
-#define IXGBE_ATR_DST_PORT_OFFSET  36
-#define IXGBE_ATR_FLEX_BYTE_OFFSET 38
-#define IXGBE_ATR_VM_POOL_OFFSET   40
-#define IXGBE_ATR_L4TYPE_OFFSET    41
+#define IXGBE_ATR_BUCKET_HASH_KEY    0x3DAD14E2
+#define IXGBE_ATR_SIGNATURE_HASH_KEY 0x174D3614
 
+/* Software ATR input stream values and masks */
+#define IXGBE_ATR_HASH_MASK     0x7fff
 #define IXGBE_ATR_L4TYPE_MASK      0x3
-#define IXGBE_ATR_L4TYPE_IPV6_MASK 0x4
 #define IXGBE_ATR_L4TYPE_UDP       0x1
 #define IXGBE_ATR_L4TYPE_TCP       0x2
 #define IXGBE_ATR_L4TYPE_SCTP      0x3
-#define IXGBE_ATR_HASH_MASK     0x7fff
+#define IXGBE_ATR_L4TYPE_IPV6_MASK 0x4
+enum ixgbe_atr_flow_type {
+	IXGBE_ATR_FLOW_TYPE_IPV4   = 0x0,
+	IXGBE_ATR_FLOW_TYPE_UDPV4  = 0x1,
+	IXGBE_ATR_FLOW_TYPE_TCPV4  = 0x2,
+	IXGBE_ATR_FLOW_TYPE_SCTPV4 = 0x3,
+	IXGBE_ATR_FLOW_TYPE_IPV6   = 0x4,
+	IXGBE_ATR_FLOW_TYPE_UDPV6  = 0x5,
+	IXGBE_ATR_FLOW_TYPE_TCPV6  = 0x6,
+	IXGBE_ATR_FLOW_TYPE_SCTPV6 = 0x7,
+};
 
 /* Flow Director ATR input struct. */
-struct ixgbe_atr_input {
-	/* Byte layout in order, all values with MSB first:
+union ixgbe_atr_input {
+	/*
+	 * Byte layout in order, all values with MSB first:
 	 *
+	 * vm_pool    - 1 byte
+	 * flow_type  - 1 byte
 	 * vlan_id    - 2 bytes
 	 * src_ip     - 16 bytes
 	 * dst_ip     - 16 bytes
 	 * src_port   - 2 bytes
 	 * dst_port   - 2 bytes
 	 * flex_bytes - 2 bytes
-	 * vm_pool    - 1 byte
-	 * l4type     - 1 byte
+	 * rsvd0      - 2 bytes - space reserved must be 0.
 	 */
-	u8 byte_stream[42];
+	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 rsvd0;
+	} formatted;
+	__be32 dword_stream[11];
 };
 
 struct ixgbe_atr_input_masks {
-	u32 src_ip_mask;
-	u32 dst_ip_mask;
-	u16 src_port_mask;
-	u16 dst_port_mask;
-	u16 vlan_id_mask;
-	u16 data_mask;
+	__be32 src_ip_mask;
+	__be32 dst_ip_mask;
+	__be16 src_port_mask;
+	__be16 dst_port_mask;
+	__be16 vlan_id_mask;
+	__be16 data_mask;
 };
 
 enum ixgbe_eeprom_type {