ixgbe: Only DMA sync frame length
authorAlexander Duyck <alexander.h.duyck@intel.com>
Tue, 17 Jan 2017 16:35:44 +0000 (08:35 -0800)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Thu, 16 Feb 2017 12:02:44 +0000 (04:02 -0800)
On some platforms, syncing a buffer for DMA is expensive. Rather than
sync the whole 2K receive buffer, only synchronise the length of the
frame, which will typically be the MTU, or a much smaller TCP ACK.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

index a19dda5711ae6243a4b6d0aaca76b776d62bb820..dde2c852e01d2fc7fc1e6012a356fdf34e38614c 100644 (file)
@@ -1841,7 +1841,7 @@ static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
                dma_sync_single_range_for_cpu(rx_ring->dev,
                                              IXGBE_CB(skb)->dma,
                                              frag->page_offset,
-                                             ixgbe_rx_bufsz(rx_ring),
+                                             skb_frag_size(frag),
                                              DMA_FROM_DEVICE);
        }
        IXGBE_CB(skb)->dma = 0;
@@ -1983,12 +1983,11 @@ static bool ixgbe_can_reuse_rx_page(struct ixgbe_ring *rx_ring,
  **/
 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
                              struct ixgbe_rx_buffer *rx_buffer,
-                             union ixgbe_adv_rx_desc *rx_desc,
+                             unsigned int size,
                              struct sk_buff *skb)
 {
        struct page *page = rx_buffer->page;
        unsigned char *va = page_address(page) + rx_buffer->page_offset;
-       unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 #if (PAGE_SIZE < 8192)
        unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
 #else
@@ -2020,6 +2019,7 @@ add_tail_frag:
 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
                                             union ixgbe_adv_rx_desc *rx_desc)
 {
+       unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
        struct ixgbe_rx_buffer *rx_buffer;
        struct sk_buff *skb;
        struct page *page;
@@ -2074,14 +2074,14 @@ dma_sync:
                dma_sync_single_range_for_cpu(rx_ring->dev,
                                              rx_buffer->dma,
                                              rx_buffer->page_offset,
-                                             ixgbe_rx_bufsz(rx_ring),
+                                             size,
                                              DMA_FROM_DEVICE);
 
                rx_buffer->skb = NULL;
        }
 
        /* pull page into skb */
-       if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
+       if (ixgbe_add_rx_frag(rx_ring, rx_buffer, size, skb)) {
                /* hand second half of page back to the ring */
                ixgbe_reuse_rx_page(rx_ring, rx_buffer);
        } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {