e1000e: reduce writes of RX producer ptr
authorTom Herbert <therbert@google.com>
Wed, 5 May 2010 14:02:49 +0000 (14:02 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 6 May 2010 04:11:25 +0000 (21:11 -0700)
Reduce number of writes to RX producer pointer.   When alloc'ing RX
buffers, only write the RX producer pointer once every
E1000_RX_BUFFER_WRITE (16) buffers created.

Signed-off-by: Tom Herbert <therbert@google.com>
Acked-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/e1000e/netdev.c

index 8de64ed8762c99355ad30d81d7eb663e2cd0acdc..b049d1a3c86144cdc01481172b8eb985d675da2e 100644 (file)
@@ -548,26 +548,23 @@ map_skb:
                rx_desc = E1000_RX_DESC(*rx_ring, i);
                rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 
+               if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
+                       /*
+                        * Force memory writes to complete before letting h/w
+                        * know there are new descriptors to fetch.  (Only
+                        * applicable for weak-ordered memory model archs,
+                        * such as IA-64).
+                        */
+                       wmb();
+                       writel(i, adapter->hw.hw_addr + rx_ring->tail);
+               }
                i++;
                if (i == rx_ring->count)
                        i = 0;
                buffer_info = &rx_ring->buffer_info[i];
        }
 
-       if (rx_ring->next_to_use != i) {
-               rx_ring->next_to_use = i;
-               if (i-- == 0)
-                       i = (rx_ring->count - 1);
-
-               /*
-                * Force memory writes to complete before letting h/w
-                * know there are new descriptors to fetch.  (Only
-                * applicable for weak-ordered memory model archs,
-                * such as IA-64).
-                */
-               wmb();
-               writel(i, adapter->hw.hw_addr + rx_ring->tail);
-       }
+       rx_ring->next_to_use = i;
 }
 
 /**
@@ -649,6 +646,17 @@ static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 
                rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 
+               if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
+                       /*
+                        * Force memory writes to complete before letting h/w
+                        * know there are new descriptors to fetch.  (Only
+                        * applicable for weak-ordered memory model archs,
+                        * such as IA-64).
+                        */
+                       wmb();
+                       writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
+               }
+
                i++;
                if (i == rx_ring->count)
                        i = 0;
@@ -656,26 +664,7 @@ static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
        }
 
 no_buffers:
-       if (rx_ring->next_to_use != i) {
-               rx_ring->next_to_use = i;
-
-               if (!(i--))
-                       i = (rx_ring->count - 1);
-
-               /*
-                * Force memory writes to complete before letting h/w
-                * know there are new descriptors to fetch.  (Only
-                * applicable for weak-ordered memory model archs,
-                * such as IA-64).
-                */
-               wmb();
-               /*
-                * Hardware increments by 16 bytes, but packet split
-                * descriptors are 32 bytes...so we increment tail
-                * twice as much.
-                */
-               writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
-       }
+       rx_ring->next_to_use = i;
 }
 
 /**