]> bbs.cooldavid.org Git - net-next-2.6.git/commitdiff
ixgbe: move GSO segments and byte count processing into ixgbe_tx_map
authorAlexander Duyck <alexander.h.duyck@intel.com>
Wed, 17 Nov 2010 03:26:47 +0000 (19:26 -0800)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Wed, 17 Nov 2010 03:26:47 +0000 (19:26 -0800)
This change simplifies the work being done by the TX interrupt handler and
pushes it into the tx_map call. This allows for fewer cache misses since
the TX cleanup now accesses almost none of the skb members.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ixgbe/ixgbe.h
drivers/net/ixgbe/ixgbe_main.c

index 4f98486d8c2cdd9f3d4d054767f5032c5c82b8ba..93946b683ad8e22a6b9637ced66a158857e9b5d5 100644 (file)
@@ -128,7 +128,9 @@ struct ixgbe_tx_buffer {
        unsigned long time_stamp;
        u16 length;
        u16 next_to_watch;
-       u16 mapped_as_page;
+       unsigned int bytecount;
+       u16 gso_segs;
+       u8 mapped_as_page;
 };
 
 struct ixgbe_rx_buffer {
index 45d988741fe91cb926ce7c6002dd24998f916c09..480f0b0f038af1c2b35a1fa9cf326402ce3b62ff 100644 (file)
@@ -749,45 +749,23 @@ static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
                bool cleaned = false;
                rmb(); /* read buffer_info after eop_desc */
                for ( ; !cleaned; count++) {
-                       struct sk_buff *skb;
                        tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
                        tx_buffer_info = &tx_ring->tx_buffer_info[i];
+
+                       tx_desc->wb.status = 0;
                        cleaned = (i == eop);
-                       skb = tx_buffer_info->skb;
 
-                       if (cleaned && skb) {
-                               unsigned int segs, bytecount;
-                               unsigned int hlen = skb_headlen(skb);
+                       i++;
+                       if (i == tx_ring->count)
+                               i = 0;
 
-                               /* gso_segs is currently only valid for tcp */
-                               segs = skb_shinfo(skb)->gso_segs ?: 1;
-#ifdef IXGBE_FCOE
-                               /* adjust for FCoE Sequence Offload */
-                               if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
-                                   && skb_is_gso(skb)
-                                   && vlan_get_protocol(skb) ==
-                                   htons(ETH_P_FCOE)) {
-                                       hlen = skb_transport_offset(skb) +
-                                               sizeof(struct fc_frame_header) +
-                                               sizeof(struct fcoe_crc_eof);
-                                       segs = DIV_ROUND_UP(skb->len - hlen,
-                                               skb_shinfo(skb)->gso_size);
-                               }
-#endif /* IXGBE_FCOE */
-                               /* multiply data chunks by size of headers */
-                               bytecount = ((segs - 1) * hlen) + skb->len;
-                               total_packets += segs;
-                               total_bytes += bytecount;
+                       if (cleaned && tx_buffer_info->skb) {
+                               total_bytes += tx_buffer_info->bytecount;
+                               total_packets += tx_buffer_info->gso_segs;
                        }
 
                        ixgbe_unmap_and_free_tx_resource(adapter,
                                                         tx_buffer_info);
-
-                       tx_desc->wb.status = 0;
-
-                       i++;
-                       if (i == tx_ring->count)
-                               i = 0;
                }
 
                eop = tx_ring->tx_buffer_info[i].next_to_watch;
@@ -6015,7 +5993,7 @@ static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
                        struct ixgbe_ring *tx_ring,
                        struct sk_buff *skb, u32 tx_flags,
-                       unsigned int first)
+                       unsigned int first, const u8 hdr_len)
 {
        struct pci_dev *pdev = adapter->pdev;
        struct ixgbe_tx_buffer *tx_buffer_info;
@@ -6024,6 +6002,8 @@ static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
        unsigned int offset = 0, size, count = 0, i;
        unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
        unsigned int f;
+       unsigned int bytecount = skb->len;
+       u16 gso_segs = 1;
 
        i = tx_ring->next_to_use;
 
@@ -6093,6 +6073,19 @@ static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
                        break;
        }
 
+       if (tx_flags & IXGBE_TX_FLAGS_TSO)
+               gso_segs = skb_shinfo(skb)->gso_segs;
+#ifdef IXGBE_FCOE
+       /* adjust for FCoE Sequence Offload */
+       else if (tx_flags & IXGBE_TX_FLAGS_FSO)
+               gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
+                                       skb_shinfo(skb)->gso_size);
+#endif /* IXGBE_FCOE */
+       bytecount += (gso_segs - 1) * hdr_len;
+
+       /* multiply data chunks by size of headers */
+       tx_ring->tx_buffer_info[i].bytecount = bytecount;
+       tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
        tx_ring->tx_buffer_info[i].skb = skb;
        tx_ring->tx_buffer_info[first].next_to_watch = i;
 
@@ -6402,7 +6395,7 @@ netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb, struct net_device *netdev
                        tx_flags |= IXGBE_TX_FLAGS_CSUM;
        }
 
-       count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
+       count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
        if (count) {
                /* add the ATR filter if ATR is on */
                if (tx_ring->atr_sample_rate) {