]> err.no Git - linux-2.6/commitdiff
[PATCH] ibmveth lockless TX
authorSantiago Leon <santil@us.ibm.com>
Wed, 26 Oct 2005 16:47:16 +0000 (10:47 -0600)
committerJeff Garzik <jgarzik@pobox.com>
Fri, 28 Oct 2005 20:07:30 +0000 (16:07 -0400)
This patch adds the lockless TX feature to the ibmveth driver.  The
hypervisor has its own locking so the only change that is necessary is
to protect the statistics counters.

Signed-off-by: Santiago Leon <santil@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
drivers/net/ibmveth.c
drivers/net/ibmveth.h

index aea1598b2253194b941bd1b96c521bd7698c0145..987bcba01889babb7e0a74e664381b16793988cb 100644 (file)
@@ -621,12 +621,18 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
        unsigned long lpar_rc;
        int nfrags = 0, curfrag;
        unsigned long correlator;
+       unsigned long flags;
        unsigned int retry_count;
+       unsigned int tx_dropped = 0;
+       unsigned int tx_bytes = 0;
+       unsigned int tx_packets = 0;
+       unsigned int tx_send_failed = 0;
+       unsigned int tx_map_failed = 0;
+
 
        if ((skb_shinfo(skb)->nr_frags + 1) > IbmVethMaxSendFrags) {
-               adapter->stats.tx_dropped++;
-               dev_kfree_skb(skb);
-               return 0;
+               tx_dropped++;
+               goto out;
        }
 
        memset(&desc, 0, sizeof(desc));
@@ -645,10 +651,9 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 
        if(dma_mapping_error(desc[0].fields.address)) {
                ibmveth_error_printk("tx: unable to map initial fragment\n");
-               adapter->tx_map_failed++;
-               adapter->stats.tx_dropped++;
-               dev_kfree_skb(skb);
-               return 0;
+               tx_map_failed++;
+               tx_dropped++;
+               goto out;
        }
 
        curfrag = nfrags;
@@ -665,8 +670,8 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 
                if(dma_mapping_error(desc[curfrag+1].fields.address)) {
                        ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
-                       adapter->tx_map_failed++;
-                       adapter->stats.tx_dropped++;
+                       tx_map_failed++;
+                       tx_dropped++;
                        /* Free all the mappings we just created */
                        while(curfrag < nfrags) {
                                dma_unmap_single(&adapter->vdev->dev,
@@ -675,8 +680,7 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
                                                 DMA_TO_DEVICE);
                                curfrag++;
                        }
-                       dev_kfree_skb(skb);
-                       return 0;
+                       goto out;
                }
        }
 
@@ -701,11 +705,11 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
                        ibmveth_error_printk("tx: desc[%i] valid=%d, len=%d, address=0x%d\n", i,
                                             desc[i].fields.valid, desc[i].fields.length, desc[i].fields.address);
                }
-               adapter->tx_send_failed++;
-               adapter->stats.tx_dropped++;
+               tx_send_failed++;
+               tx_dropped++;
        } else {
-               adapter->stats.tx_packets++;
-               adapter->stats.tx_bytes += skb->len;
+               tx_packets++;
+               tx_bytes += skb->len;
                netdev->trans_start = jiffies;
        }
 
@@ -715,6 +719,14 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
                                desc[nfrags].fields.length, DMA_TO_DEVICE);
        } while(--nfrags >= 0);
 
+out:   spin_lock_irqsave(&adapter->stats_lock, flags);
+       adapter->stats.tx_dropped += tx_dropped;
+       adapter->stats.tx_bytes += tx_bytes;
+       adapter->stats.tx_packets += tx_packets;
+       adapter->tx_send_failed += tx_send_failed;
+       adapter->tx_map_failed += tx_map_failed;
+       spin_unlock_irqrestore(&adapter->stats_lock, flags);
+
        dev_kfree_skb(skb);
        return 0;
 }
@@ -980,6 +992,8 @@ static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_
        netdev->ethtool_ops           = &netdev_ethtool_ops;
        netdev->change_mtu         = ibmveth_change_mtu;
        SET_NETDEV_DEV(netdev, &dev->dev);
+       netdev->features |= NETIF_F_LLTX; 
+       spin_lock_init(&adapter->stats_lock);
 
        memcpy(&netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
 
index a3ea02937d63a09b046c531e91a4899476f9e017..46919a814fcac44fce8fb60b0fec6530d31e4c82 100644 (file)
@@ -131,6 +131,7 @@ struct ibmveth_adapter {
     u64 tx_linearize_failed;
     u64 tx_map_failed;
     u64 tx_send_failed;
+    spinlock_t stats_lock;
 };
 
 struct ibmveth_buf_desc_fields {