qlcnic: Code optimization patch

Optimized code resulted in achieving lower CPU utilization on transmit path
and higher throughput for small packet sizes (64 bytes).
Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/qlcnic/qlcnic.h

@@ -434,50 +434,49 @@ struct qlcnic_adapter_stats {
  * be one Rcv Descriptor for normal packets, one for jumbo and may be others.
  */
 struct qlcnic_host_rds_ring {
-	u32 producer;
+	void __iomem *crb_rcv_producer;
+	struct rcv_desc *desc_head;
+	struct qlcnic_rx_buffer *rx_buf_arr;
 	u32 num_desc;
+	u32 producer;
 	u32 dma_size;
 	u32 skb_size;
 	u32 flags;
-	void __iomem *crb_rcv_producer;
-	struct rcv_desc *desc_head;
-	struct qlcnic_rx_buffer *rx_buf_arr;
 	struct list_head free_list;
 	spinlock_t lock;
 	dma_addr_t phys_addr;
-};
+} ____cacheline_internodealigned_in_smp;
 
 struct qlcnic_host_sds_ring {
 	u32 consumer;
 	u32 num_desc;
 	void __iomem *crb_sts_consumer;
-	void __iomem *crb_intr_mask;
 
 	struct status_desc *desc_head;
 	struct qlcnic_adapter *adapter;
 	struct napi_struct napi;
 	struct list_head free_list[NUM_RCV_DESC_RINGS];
 
+	void __iomem *crb_intr_mask;
 	int irq;
 
 	dma_addr_t phys_addr;
 	char name[IFNAMSIZ+4];
-};
+} ____cacheline_internodealigned_in_smp;
 
 struct qlcnic_host_tx_ring {
 	u32 producer;
-	__le32 *hw_consumer;
 	u32 sw_consumer;
-	void __iomem *crb_cmd_producer;
 	u32 num_desc;
-
-	struct netdev_queue *txq;
-
-	struct qlcnic_cmd_buffer *cmd_buf_arr;
+	void __iomem *crb_cmd_producer;
 	struct cmd_desc_type0 *desc_head;
+	struct qlcnic_cmd_buffer *cmd_buf_arr;
+	__le32 *hw_consumer;
+
 	dma_addr_t phys_addr;
 	dma_addr_t hw_cons_phys_addr;
-};
+	struct netdev_queue *txq;
+} ____cacheline_internodealigned_in_smp;
 
 /*
  * Receive context. There is one such structure per instance of the
@@ -1328,8 +1327,7 @@ static const struct qlcnic_brdinfo qlcnic_boards[] = {
 
 static inline u32 qlcnic_tx_avail(struct qlcnic_host_tx_ring *tx_ring)
 {
-	smp_mb();
-	if (tx_ring->producer < tx_ring->sw_consumer)
+	if (likely(tx_ring->producer < tx_ring->sw_consumer))
 		return tx_ring->sw_consumer - tx_ring->producer;
 	else
 		return tx_ring->sw_consumer + tx_ring->num_desc -

drivers/net/qlcnic/qlcnic_main.c

@@ -1861,6 +1861,7 @@ static void qlcnic_change_filter(struct qlcnic_adapter *adapter,
 	vlan_req->vlan_id = vlan_id;
 
 	tx_ring->producer = get_next_index(producer, tx_ring->num_desc);
+	smp_mb();
 }
 
 #define QLCNIC_MAC_HASH(MAC)\
@@ -1921,93 +1922,80 @@ qlcnic_send_filter(struct qlcnic_adapter *adapter,
 	spin_unlock(&adapter->mac_learn_lock);
 }
 
-static void
-qlcnic_tso_check(struct net_device *netdev,
-		struct qlcnic_host_tx_ring *tx_ring,
+static int
+qlcnic_tx_pkt(struct qlcnic_adapter *adapter,
 		struct cmd_desc_type0 *first_desc,
 		struct sk_buff *skb)
 {
-	u8 opcode = TX_ETHER_PKT;
-	__be16 protocol = skb->protocol;
-	u16 flags = 0;
-	int copied, offset, copy_len, hdr_len = 0, tso = 0;
+	u8 opcode = 0, hdr_len = 0;
+	u16 flags = 0, vlan_tci = 0;
+	int copied, offset, copy_len;
 	struct cmd_desc_type0 *hwdesc;
 	struct vlan_ethhdr *vh;
-	struct qlcnic_adapter *adapter = netdev_priv(netdev);
+	struct qlcnic_host_tx_ring *tx_ring = adapter->tx_ring;
+	u16 protocol = ntohs(skb->protocol);
 	u32 producer = tx_ring->producer;
-	__le16 vlan_oob = first_desc->flags_opcode &
-				cpu_to_le16(FLAGS_VLAN_OOB);
+
+	if (protocol == ETH_P_8021Q) {
+		vh = (struct vlan_ethhdr *)skb->data;
+		flags = FLAGS_VLAN_TAGGED;
+		vlan_tci = vh->h_vlan_TCI;
+	} else if (vlan_tx_tag_present(skb)) {
+		flags = FLAGS_VLAN_OOB;
+		vlan_tci = vlan_tx_tag_get(skb);
+	}
+	if (unlikely(adapter->pvid)) {
+		if (vlan_tci && !(adapter->flags & QLCNIC_TAGGING_ENABLED))
+			return -EIO;
+		if (vlan_tci && (adapter->flags & QLCNIC_TAGGING_ENABLED))
+			goto set_flags;
+
+		flags = FLAGS_VLAN_OOB;
+		vlan_tci = adapter->pvid;
+	}
+set_flags:
+	qlcnic_set_tx_vlan_tci(first_desc, vlan_tci);
+	qlcnic_set_tx_flags_opcode(first_desc, flags, opcode);
 
 	if (*(skb->data) & BIT_0) {
 		flags |= BIT_0;
 		memcpy(&first_desc->eth_addr, skb->data, ETH_ALEN);
 	}
-
-	if ((netdev->features & (NETIF_F_TSO | NETIF_F_TSO6)) &&
+	opcode = TX_ETHER_PKT;
+	if ((adapter->netdev->features & (NETIF_F_TSO | NETIF_F_TSO6)) &&
 			skb_shinfo(skb)->gso_size > 0) {
 
 		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 
 		first_desc->mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
 		first_desc->total_hdr_length = hdr_len;
-		if (vlan_oob) {
-			first_desc->total_hdr_length += VLAN_HLEN;
-			first_desc->tcp_hdr_offset = VLAN_HLEN;
-			first_desc->ip_hdr_offset = VLAN_HLEN;
-			/* Only in case of TSO on vlan device */
-			flags |= FLAGS_VLAN_TAGGED;
-		}
 
-		opcode = (protocol == cpu_to_be16(ETH_P_IPV6)) ?
-				TX_TCP_LSO6 : TX_TCP_LSO;
-		tso = 1;
-
-	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
-		u8 l4proto;
-
-		if (protocol == cpu_to_be16(ETH_P_IP)) {
-			l4proto = ip_hdr(skb)->protocol;
-
-			if (l4proto == IPPROTO_TCP)
-				opcode = TX_TCP_PKT;
-			else if (l4proto == IPPROTO_UDP)
-				opcode = TX_UDP_PKT;
-		} else if (protocol == cpu_to_be16(ETH_P_IPV6)) {
-			l4proto = ipv6_hdr(skb)->nexthdr;
-
-			if (l4proto == IPPROTO_TCP)
-				opcode = TX_TCPV6_PKT;
-			else if (l4proto == IPPROTO_UDP)
-				opcode = TX_UDPV6_PKT;
-		}
-	}
-
-	first_desc->tcp_hdr_offset += skb_transport_offset(skb);
-	first_desc->ip_hdr_offset += skb_network_offset(skb);
-	qlcnic_set_tx_flags_opcode(first_desc, flags, opcode);
-
-	if (!tso)
-		return;
+		opcode = (protocol == ETH_P_IPV6) ? TX_TCP_LSO6 : TX_TCP_LSO;
 
 		/* For LSO, we need to copy the MAC/IP/TCP headers into
-	 * the descriptor ring
-	 */
+		* the descriptor ring */
 		copied = 0;
 		offset = 2;
 
-	if (vlan_oob) {
+		if (flags & FLAGS_VLAN_OOB) {
+			first_desc->total_hdr_length += VLAN_HLEN;
+			first_desc->tcp_hdr_offset = VLAN_HLEN;
+			first_desc->ip_hdr_offset = VLAN_HLEN;
+			/* Only in case of TSO on vlan device */
+			flags |= FLAGS_VLAN_TAGGED;
+
 			/* Create a TSO vlan header template for firmware */
 
 			hwdesc = &tx_ring->desc_head[producer];
 			tx_ring->cmd_buf_arr[producer].skb = NULL;
 
-		copy_len = min((int)sizeof(struct cmd_desc_type0) - offset,
-				hdr_len + VLAN_HLEN);
+			copy_len = min((int)sizeof(struct cmd_desc_type0) -
+				offset, hdr_len + VLAN_HLEN);
 
-		vh = (struct vlan_ethhdr *)((char *)hwdesc + 2);
+			vh = (struct vlan_ethhdr *)((char *) hwdesc + 2);
 			skb_copy_from_linear_data(skb, vh, 12);
 			vh->h_vlan_proto = htons(ETH_P_8021Q);
-		vh->h_vlan_TCI = (__be16)swab16((u16)first_desc->vlan_TCI);
+			vh->h_vlan_TCI = htons(vlan_tci);
 
 			skb_copy_from_linear_data_offset(skb, 12,
 				(char *)vh + 16, copy_len - 16);
@@ -2020,14 +2008,14 @@ qlcnic_tso_check(struct net_device *netdev,
 
 		while (copied < hdr_len) {
 
-		copy_len = min((int)sizeof(struct cmd_desc_type0) - offset,
-				(hdr_len - copied));
+			copy_len = min((int)sizeof(struct cmd_desc_type0) -
+				offset, (hdr_len - copied));
 
 			hwdesc = &tx_ring->desc_head[producer];
 			tx_ring->cmd_buf_arr[producer].skb = NULL;
 
 			skb_copy_from_linear_data_offset(skb, copied,
-				 (char *)hwdesc + offset, copy_len);
+				 (char *) hwdesc + offset, copy_len);
 
 			copied += copy_len;
 			offset = 0;
@@ -2036,8 +2024,33 @@ qlcnic_tso_check(struct net_device *netdev,
 		}
 
 		tx_ring->producer = producer;
-	barrier();
+		smp_mb();
 		adapter->stats.lso_frames++;
+
+	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
+		u8 l4proto;
+
+		if (protocol == ETH_P_IP) {
+			l4proto = ip_hdr(skb)->protocol;
+
+			if (l4proto == IPPROTO_TCP)
+				opcode = TX_TCP_PKT;
+			else if (l4proto == IPPROTO_UDP)
+				opcode = TX_UDP_PKT;
+		} else if (protocol == ETH_P_IPV6) {
+			l4proto = ipv6_hdr(skb)->nexthdr;
+
+			if (l4proto == IPPROTO_TCP)
+				opcode = TX_TCPV6_PKT;
+			else if (l4proto == IPPROTO_UDP)
+				opcode = TX_UDPV6_PKT;
+		}
+	}
+	first_desc->tcp_hdr_offset += skb_transport_offset(skb);
+	first_desc->ip_hdr_offset += skb_network_offset(skb);
+	qlcnic_set_tx_flags_opcode(first_desc, flags, opcode);
+
+	return 0;
 }
 
 static int
@@ -2088,39 +2101,21 @@ qlcnic_map_tx_skb(struct pci_dev *pdev,
 	return -ENOMEM;
 }
 
-static int
-qlcnic_check_tx_tagging(struct qlcnic_adapter *adapter,
-			struct sk_buff *skb,
-			struct cmd_desc_type0 *first_desc)
+static void
+qlcnic_unmap_buffers(struct pci_dev *pdev, struct sk_buff *skb,
+			struct qlcnic_cmd_buffer *pbuf)
 {
-	u8 opcode = 0;
-	u16 flags = 0;
-	__be16 protocol = skb->protocol;
-	struct vlan_ethhdr *vh;
+	struct qlcnic_skb_frag *nf = &pbuf->frag_array[0];
+	int nr_frags = skb_shinfo(skb)->nr_frags;
+	int i;
 
-	if (protocol == cpu_to_be16(ETH_P_8021Q)) {
-		vh = (struct vlan_ethhdr *)skb->data;
-		protocol = vh->h_vlan_encapsulated_proto;
-		flags = FLAGS_VLAN_TAGGED;
-		qlcnic_set_tx_vlan_tci(first_desc, ntohs(vh->h_vlan_TCI));
-	} else if (vlan_tx_tag_present(skb)) {
-		flags = FLAGS_VLAN_OOB;
-		qlcnic_set_tx_vlan_tci(first_desc, vlan_tx_tag_get(skb));
+	for (i = 0; i < nr_frags; i++) {
+		nf = &pbuf->frag_array[i+1];
+		pci_unmap_page(pdev, nf->dma, nf->length, PCI_DMA_TODEVICE);
 	}
-	if (unlikely(adapter->pvid)) {
-		if (first_desc->vlan_TCI &&
-				!(adapter->flags & QLCNIC_TAGGING_ENABLED))
-			return -EIO;
-		if (first_desc->vlan_TCI &&
-				(adapter->flags & QLCNIC_TAGGING_ENABLED))
-			goto set_flags;
 
-		flags = FLAGS_VLAN_OOB;
-		qlcnic_set_tx_vlan_tci(first_desc, adapter->pvid);
-	}
-set_flags:
-	qlcnic_set_tx_flags_opcode(first_desc, flags, opcode);
-	return 0;
+	nf = &pbuf->frag_array[0];
+	pci_unmap_single(pdev, nf->dma, skb_headlen(skb), PCI_DMA_TODEVICE);
 }
 
 static inline void
@@ -2144,7 +2139,7 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	int i, k;
 
 	u32 producer;
-	int frag_count, no_of_desc;
+	int frag_count;
 	u32 num_txd = tx_ring->num_desc;
 
 	if (!test_bit(__QLCNIC_DEV_UP, &adapter->state)) {
@@ -2161,12 +2156,8 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 
 	frag_count = skb_shinfo(skb)->nr_frags + 1;
 
-	/* 4 fragments per cmd des */
-	no_of_desc = (frag_count + 3) >> 2;
-
 	if (unlikely(qlcnic_tx_avail(tx_ring) <= TX_STOP_THRESH)) {
 		netif_stop_queue(netdev);
-		smp_mb();
 		if (qlcnic_tx_avail(tx_ring) > TX_STOP_THRESH)
 			netif_start_queue(netdev);
 		else {
@@ -2183,9 +2174,6 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	first_desc = hwdesc = &tx_ring->desc_head[producer];
 	qlcnic_clear_cmddesc((u64 *)hwdesc);
 
-	if (qlcnic_check_tx_tagging(adapter, skb, first_desc))
-		goto drop_packet;
-
 	if (qlcnic_map_tx_skb(pdev, skb, pbuf)) {
 		adapter->stats.tx_dma_map_error++;
 		goto drop_packet;
@@ -2229,8 +2217,10 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	}
 
 	tx_ring->producer = get_next_index(producer, num_txd);
+	smp_mb();
 
-	qlcnic_tso_check(netdev, tx_ring, first_desc, skb);
+	if (unlikely(qlcnic_tx_pkt(adapter, first_desc, skb)))
+		goto unwind_buff;
 
 	if (qlcnic_mac_learn)
 		qlcnic_send_filter(adapter, tx_ring, first_desc, skb);
@@ -2242,6 +2232,8 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 
 	return NETDEV_TX_OK;
 
+unwind_buff:
+	qlcnic_unmap_buffers(pdev, skb, pbuf);
 drop_packet:
 	adapter->stats.txdropped++;
 	dev_kfree_skb_any(skb);