igb: streamline Rx buffer allocation and cleanup
authorAlexander Duyck <alexander.h.duyck@intel.com>
Fri, 26 Aug 2011 07:43:43 +0000 (07:43 +0000)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Tue, 20 Sep 2011 06:57:42 +0000 (23:57 -0700)
This change is meant to streamline the Rx buffer allocation and cleanup.
This is accomplished by reducing the number of writes by only having the Rx
descriptor ring written by software during allocation, and it will only be
read during cleanup.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/igb/igb.h
drivers/net/ethernet/intel/igb/igb_main.c

index 50632b19d2d7a0b30d3d38c38db49f12caeb74d6..b2f2a8ca46e2caafc5fa70083d7aeacb58e23cf6 100644 (file)
@@ -370,7 +370,7 @@ extern void igb_setup_rctl(struct igb_adapter *);
 extern netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *, struct igb_ring *);
 extern void igb_unmap_and_free_tx_resource(struct igb_ring *,
                                           struct igb_buffer *);
-extern void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
+extern void igb_alloc_rx_buffers_adv(struct igb_ring *, u16);
 extern void igb_update_stats(struct igb_adapter *, struct rtnl_link_stats64 *);
 extern bool igb_has_link(struct igb_adapter *adapter);
 extern void igb_set_ethtool_ops(struct net_device *);
index 022c44203501b7bf736c5cd89f9b4e6e3a2f1170..af8c2f783a905470a31fe94fc6aa20dee344ed97 100644 (file)
@@ -3243,16 +3243,15 @@ static void igb_free_all_rx_resources(struct igb_adapter *adapter)
  **/
 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
 {
-       struct igb_buffer *buffer_info;
        unsigned long size;
-       unsigned int i;
+       u16 i;
 
        if (!rx_ring->buffer_info)
                return;
 
        /* Free all the Rx ring sk_buffs */
        for (i = 0; i < rx_ring->count; i++) {
-               buffer_info = &rx_ring->buffer_info[i];
+               struct igb_buffer *buffer_info = &rx_ring->buffer_info[i];
                if (buffer_info->dma) {
                        dma_unmap_single(rx_ring->dev,
                                         buffer_info->dma,
@@ -5764,7 +5763,7 @@ static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
        struct igb_buffer *buffer_info , *next_buffer;
        struct sk_buff *skb;
        bool cleaned = false;
-       int cleaned_count = 0;
+       u16 cleaned_count = igb_desc_unused(rx_ring);
        int current_node = numa_node_id();
        unsigned int total_bytes = 0, total_packets = 0;
        unsigned int i;
@@ -5848,7 +5847,6 @@ static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
                igb_rx_checksum_adv(rx_ring, staterr, skb);
 
                skb->protocol = eth_type_trans(skb, netdev);
-               skb_record_rx_queue(skb, rx_ring->queue_index);
 
                if (staterr & E1000_RXD_STAT_VP) {
                        u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
@@ -5858,8 +5856,6 @@ static bool igb_clean_rx_irq_adv(struct igb_q_vector *q_vector,
                napi_gro_receive(&q_vector->napi, skb);
 
 next_desc:
-               rx_desc->wb.upper.status_error = 0;
-
                /* return some buffers to hardware, one at a time is too slow */
                if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
                        igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
@@ -5873,110 +5869,130 @@ next_desc:
        }
 
        rx_ring->next_to_clean = i;
-       cleaned_count = igb_desc_unused(rx_ring);
-
-       if (cleaned_count)
-               igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
-
-       rx_ring->total_packets += total_packets;
-       rx_ring->total_bytes += total_bytes;
        u64_stats_update_begin(&rx_ring->rx_syncp);
        rx_ring->rx_stats.packets += total_packets;
        rx_ring->rx_stats.bytes += total_bytes;
        u64_stats_update_end(&rx_ring->rx_syncp);
+       rx_ring->total_packets += total_packets;
+       rx_ring->total_bytes += total_bytes;
+
+       if (cleaned_count)
+               igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
+
        return cleaned;
 }
 
+static bool igb_alloc_mapped_skb(struct igb_ring *rx_ring,
+                                struct igb_buffer *bi)
+{
+       struct sk_buff *skb = bi->skb;
+       dma_addr_t dma = bi->dma;
+
+       if (dma)
+               return true;
+
+       if (likely(!skb)) {
+               skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
+                                               IGB_RX_HDR_LEN);
+               bi->skb = skb;
+               if (!skb) {
+                       rx_ring->rx_stats.alloc_failed++;
+                       return false;
+               }
+
+               /* initialize skb for ring */
+               skb_record_rx_queue(skb, rx_ring->queue_index);
+       }
+
+       dma = dma_map_single(rx_ring->dev, skb->data,
+                            IGB_RX_HDR_LEN, DMA_FROM_DEVICE);
+
+       if (dma_mapping_error(rx_ring->dev, dma)) {
+               rx_ring->rx_stats.alloc_failed++;
+               return false;
+       }
+
+       bi->dma = dma;
+       return true;
+}
+
+static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
+                                 struct igb_buffer *bi)
+{
+       struct page *page = bi->page;
+       dma_addr_t page_dma = bi->page_dma;
+       unsigned int page_offset = bi->page_offset ^ (PAGE_SIZE / 2);
+
+       if (page_dma)
+               return true;
+
+       if (!page) {
+               page = netdev_alloc_page(rx_ring->netdev);
+               bi->page = page;
+               if (unlikely(!page)) {
+                       rx_ring->rx_stats.alloc_failed++;
+                       return false;
+               }
+       }
+
+       page_dma = dma_map_page(rx_ring->dev, page,
+                               page_offset, PAGE_SIZE / 2,
+                               DMA_FROM_DEVICE);
+
+       if (dma_mapping_error(rx_ring->dev, page_dma)) {
+               rx_ring->rx_stats.alloc_failed++;
+               return false;
+       }
+
+       bi->page_dma = page_dma;
+       bi->page_offset = page_offset;
+       return true;
+}
+
 /**
  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
  * @adapter: address of board private structure
  **/
-void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, int cleaned_count)
+void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring, u16 cleaned_count)
 {
-       struct net_device *netdev = rx_ring->netdev;
        union e1000_adv_rx_desc *rx_desc;
-       struct igb_buffer *buffer_info;
-       struct sk_buff *skb;
-       unsigned int i;
+       struct igb_buffer *bi;
+       u16 i = rx_ring->next_to_use;
 
-       i = rx_ring->next_to_use;
-       buffer_info = &rx_ring->buffer_info[i];
+       rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
+       bi = &rx_ring->buffer_info[i];
+       i -= rx_ring->count;
 
        while (cleaned_count--) {
-               rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
-
-               if (!buffer_info->page_dma) {
-                       if (!buffer_info->page) {
-                               buffer_info->page = netdev_alloc_page(netdev);
-                               if (unlikely(!buffer_info->page)) {
-                                       u64_stats_update_begin(&rx_ring->rx_syncp);
-                                       rx_ring->rx_stats.alloc_failed++;
-                                       u64_stats_update_end(&rx_ring->rx_syncp);
-                                       goto no_buffers;
-                               }
-                               buffer_info->page_offset = 0;
-                       } else {
-                               buffer_info->page_offset ^= PAGE_SIZE / 2;
-                       }
-                       buffer_info->page_dma =
-                               dma_map_page(rx_ring->dev, buffer_info->page,
-                                            buffer_info->page_offset,
-                                            PAGE_SIZE / 2,
-                                            DMA_FROM_DEVICE);
-                       if (dma_mapping_error(rx_ring->dev,
-                                             buffer_info->page_dma)) {
-                               buffer_info->page_dma = 0;
-                               u64_stats_update_begin(&rx_ring->rx_syncp);
-                               rx_ring->rx_stats.alloc_failed++;
-                               u64_stats_update_end(&rx_ring->rx_syncp);
-                               goto no_buffers;
-                       }
-               }
+               if (!igb_alloc_mapped_skb(rx_ring, bi))
+                       break;
 
-               skb = buffer_info->skb;
-               if (!skb) {
-                       skb = netdev_alloc_skb_ip_align(netdev, IGB_RX_HDR_LEN);
-                       if (unlikely(!skb)) {
-                               u64_stats_update_begin(&rx_ring->rx_syncp);
-                               rx_ring->rx_stats.alloc_failed++;
-                               u64_stats_update_end(&rx_ring->rx_syncp);
-                               goto no_buffers;
-                       }
+               /* Refresh the desc even if buffer_addrs didn't change
+                * because each write-back erases this info. */
+               rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
 
-                       buffer_info->skb = skb;
-               }
-               if (!buffer_info->dma) {
-                       buffer_info->dma = dma_map_single(rx_ring->dev,
-                                                         skb->data,
-                                                         IGB_RX_HDR_LEN,
-                                                         DMA_FROM_DEVICE);
-                       if (dma_mapping_error(rx_ring->dev,
-                                             buffer_info->dma)) {
-                               buffer_info->dma = 0;
-                               u64_stats_update_begin(&rx_ring->rx_syncp);
-                               rx_ring->rx_stats.alloc_failed++;
-                               u64_stats_update_end(&rx_ring->rx_syncp);
-                               goto no_buffers;
-                       }
-               }
-               /* Refresh the desc even if buffer_addrs didn't change because
-                * each write-back erases this info. */
-               rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->page_dma);
-               rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
+               if (!igb_alloc_mapped_page(rx_ring, bi))
+                       break;
+
+               rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
 
+               rx_desc++;
+               bi++;
                i++;
-               if (i == rx_ring->count)
-                       i = 0;
-               buffer_info = &rx_ring->buffer_info[i];
+               if (unlikely(!i)) {
+                       rx_desc = E1000_RX_DESC_ADV(*rx_ring, 0);
+                       bi = rx_ring->buffer_info;
+                       i -= rx_ring->count;
+               }
+
+               /* clear the hdr_addr for the next_to_use descriptor */
+               rx_desc->read.hdr_addr = 0;
        }
 
-no_buffers:
+       i += rx_ring->count;
+
        if (rx_ring->next_to_use != i) {
                rx_ring->next_to_use = i;
-               if (i == 0)
-                       i = (rx_ring->count - 1);
-               else
-                       i--;
 
                /* Force memory writes to complete before letting h/w
                 * know there are new descriptors to fetch.  (Only