igc: Add support for TSO

TCP segmentation offload allows a device to segment a single frame
into multiple frames with a data payload size specified in socket buffer.
As a result we can now send data approximately up to seven percents fast
than was previously possible on my system.
Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

drivers/net/ethernet/intel/igc/igc_defines.h

@@ -282,6 +282,10 @@
 #define IGC_TXD_STAT_TC		0x00000004 /* Tx Underrun */
 #define IGC_TXD_EXTCMD_TSTAMP	0x00000010 /* IEEE1588 Timestamp packet */
 
+/* IPSec Encrypt Enable */
+#define IGC_ADVTXD_L4LEN_SHIFT	8  /* Adv ctxt L4LEN shift */
+#define IGC_ADVTXD_MSS_SHIFT	16 /* Adv ctxt MSS shift */
+
 /* Transmit Control */
 #define IGC_TCTL_EN		0x00000002 /* enable Tx */
 #define IGC_TCTL_PSP		0x00000008 /* pad short packets */

drivers/net/ethernet/intel/igc/igc_main.c

@@ -1000,6 +1000,10 @@ static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
 		       IGC_ADVTXD_DCMD_DEXT |
 		       IGC_ADVTXD_DCMD_IFCS;
 
+	/* set segmentation bits for TSO */
+	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSO,
+				 (IGC_ADVTXD_DCMD_TSE));
+
 	/* set timestamp bit if present */
 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP,
 				 (IGC_ADVTXD_MAC_TSTAMP));
@@ -1171,6 +1175,100 @@ static int igc_tx_map(struct igc_ring *tx_ring,
 	return -1;
 }
 
+static int igc_tso(struct igc_ring *tx_ring,
+		   struct igc_tx_buffer *first,
+		   u8 *hdr_len)
+{
+	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
+	struct sk_buff *skb = first->skb;
+	union {
+		struct iphdr *v4;
+		struct ipv6hdr *v6;
+		unsigned char *hdr;
+	} ip;
+	union {
+		struct tcphdr *tcp;
+		struct udphdr *udp;
+		unsigned char *hdr;
+	} l4;
+	u32 paylen, l4_offset;
+	int err;
+
+	if (skb->ip_summed != CHECKSUM_PARTIAL)
+		return 0;
+
+	if (!skb_is_gso(skb))
+		return 0;
+
+	err = skb_cow_head(skb, 0);
+	if (err < 0)
+		return err;
+
+	ip.hdr = skb_network_header(skb);
+	l4.hdr = skb_checksum_start(skb);
+
+	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
+	type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
+
+	/* initialize outer IP header fields */
+	if (ip.v4->version == 4) {
+		unsigned char *csum_start = skb_checksum_start(skb);
+		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
+
+		/* IP header will have to cancel out any data that
+		 * is not a part of the outer IP header
+		 */
+		ip.v4->check = csum_fold(csum_partial(trans_start,
+						      csum_start - trans_start,
+						      0));
+		type_tucmd |= IGC_ADVTXD_TUCMD_IPV4;
+
+		ip.v4->tot_len = 0;
+		first->tx_flags |= IGC_TX_FLAGS_TSO |
+				   IGC_TX_FLAGS_CSUM |
+				   IGC_TX_FLAGS_IPV4;
+	} else {
+		ip.v6->payload_len = 0;
+		first->tx_flags |= IGC_TX_FLAGS_TSO |
+				   IGC_TX_FLAGS_CSUM;
+	}
+
+	/* determine offset of inner transport header */
+	l4_offset = l4.hdr - skb->data;
+
+	/* remove payload length from inner checksum */
+	paylen = skb->len - l4_offset;
+	if (type_tucmd & IGC_ADVTXD_TUCMD_L4T_TCP) {
+		/* compute length of segmentation header */
+		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
+		csum_replace_by_diff(&l4.tcp->check,
+				     (__force __wsum)htonl(paylen));
+	} else {
+		/* compute length of segmentation header */
+		*hdr_len = sizeof(*l4.udp) + l4_offset;
+		csum_replace_by_diff(&l4.udp->check,
+				     (__force __wsum)htonl(paylen));
+	}
+
+	/* update gso size and bytecount with header size */
+	first->gso_segs = skb_shinfo(skb)->gso_segs;
+	first->bytecount += (first->gso_segs - 1) * *hdr_len;
+
+	/* MSS L4LEN IDX */
+	mss_l4len_idx = (*hdr_len - l4_offset) << IGC_ADVTXD_L4LEN_SHIFT;
+	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IGC_ADVTXD_MSS_SHIFT;
+
+	/* VLAN MACLEN IPLEN */
+	vlan_macip_lens = l4.hdr - ip.hdr;
+	vlan_macip_lens |= (ip.hdr - skb->data) << IGC_ADVTXD_MACLEN_SHIFT;
+	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
+
+	igc_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
+			type_tucmd, mss_l4len_idx);
+
+	return 1;
+}
+
 static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
 				       struct igc_ring *tx_ring)
 {
@@ -1180,6 +1278,7 @@ static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
 	u32 tx_flags = 0;
 	unsigned short f;
 	u8 hdr_len = 0;
+	int tso = 0;
 
 	/* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD,
 	 *	+ 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD,
@@ -1226,10 +1325,20 @@ static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
 	first->tx_flags = tx_flags;
 	first->protocol = protocol;
 
-	igc_tx_csum(tx_ring, first);
+	tso = igc_tso(tx_ring, first, &hdr_len);
+	if (tso < 0)
+		goto out_drop;
+	else if (!tso)
+		igc_tx_csum(tx_ring, first);
 
 	igc_tx_map(tx_ring, first, hdr_len);
 
+	return NETDEV_TX_OK;
+
+out_drop:
+	dev_kfree_skb_any(first->skb);
+	first->skb = NULL;
+
 	return NETDEV_TX_OK;
 }
 
@@ -4589,6 +4698,8 @@ static int igc_probe(struct pci_dev *pdev,
 
 	/* Add supported features to the features list*/
 	netdev->features |= NETIF_F_SG;
+	netdev->features |= NETIF_F_TSO;
+	netdev->features |= NETIF_F_TSO6;
 	netdev->features |= NETIF_F_RXCSUM;
 	netdev->features |= NETIF_F_HW_CSUM;
 	netdev->features |= NETIF_F_SCTP_CRC;