[PATCH] e1000: Added RX buffer enhancements
authorJesse Brandeburg <jesse.brandeburg@intel.com>
Wed, 18 Jan 2006 21:01:32 +0000 (13:01 -0800)
committerJeff Garzik <jgarzik@pobox.com>
Wed, 18 Jan 2006 21:17:57 +0000 (16:17 -0500)
Align the prefetches to a dword to help speed them up.
Recycle skb's and early replenish.
Force memory writes to complete before fetching more descriptors.

Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: John Ronciak <john.ronciak@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
drivers/net/e1000/e1000_main.c

index 73b2a7be212641c799cb187d3768656e3bb0670f..53a9cca06905f6e036c6ce3dc087551ed001c880 100644 (file)
@@ -1653,23 +1653,8 @@ e1000_setup_rctl(struct e1000_adapter *adapter)
                rctl |= adapter->rx_buffer_len << 0x11;
        } else {
                rctl &= ~E1000_RCTL_SZ_4096;
-               rctl |= E1000_RCTL_BSEX; 
-               switch (adapter->rx_buffer_len) {
-               case E1000_RXBUFFER_2048:
-               default:
-                       rctl |= E1000_RCTL_SZ_2048;
-                       rctl &= ~E1000_RCTL_BSEX;
-                       break;
-               case E1000_RXBUFFER_4096:
-                       rctl |= E1000_RCTL_SZ_4096;
-                       break;
-               case E1000_RXBUFFER_8192:
-                       rctl |= E1000_RCTL_SZ_8192;
-                       break;
-               case E1000_RXBUFFER_16384:
-                       rctl |= E1000_RCTL_SZ_16384;
-                       break;
-               }
+               rctl &= ~E1000_RCTL_BSEX;
+               rctl |= E1000_RCTL_SZ_2048;
        }
 
 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
@@ -3571,7 +3556,6 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
        struct pci_dev *pdev = adapter->pdev;
        struct e1000_rx_desc *rx_desc;
        struct e1000_buffer *buffer_info;
-       struct sk_buff *skb;
        unsigned long flags;
        uint32_t length;
        uint8_t last_byte;
@@ -3581,9 +3565,10 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
 
        i = rx_ring->next_to_clean;
        rx_desc = E1000_RX_DESC(*rx_ring, i);
+       buffer_info = &rx_ring->buffer_info[i];
 
-       while(rx_desc->status & E1000_RXD_STAT_DD) {
-               buffer_info = &rx_ring->buffer_info[i];
+       while (rx_desc->status & E1000_RXD_STAT_DD) {
+               struct sk_buff *skb;
                u8 status;
 #ifdef CONFIG_E1000_NAPI
                if(*work_done >= work_to_do)
@@ -3591,6 +3576,7 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                (*work_done)++;
 #endif
                status = rx_desc->status;
+               skb = buffer_info->skb;
                cleaned = TRUE;
                cleaned_count++;
                pci_unmap_single(pdev,
@@ -3598,20 +3584,50 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                                 buffer_info->length,
                                 PCI_DMA_FROMDEVICE);
 
-               skb = buffer_info->skb;
                length = le16_to_cpu(rx_desc->length);
 
-               if(unlikely(!(rx_desc->status & E1000_RXD_STAT_EOP))) {
-                       /* All receives must fit into a single buffer */
-                       E1000_DBG("%s: Receive packet consumed multiple"
-                                 " buffers\n", netdev->name);
-                       dev_kfree_skb_irq(skb);
+               skb_put(skb, length);
+
+               if (!(status & E1000_RXD_STAT_EOP)) {
+                       if (!rx_ring->rx_skb_top) {
+                               rx_ring->rx_skb_top = skb;
+                               rx_ring->rx_skb_top->len = length;
+                               rx_ring->rx_skb_prev = skb;
+                       } else {
+                               if (skb_shinfo(rx_ring->rx_skb_top)->frag_list) {
+                                       rx_ring->rx_skb_prev->next = skb;
+                                       skb->prev = rx_ring->rx_skb_prev;
+                               } else {
+                                       skb_shinfo(rx_ring->rx_skb_top)->frag_list = skb;
+                               }
+                               rx_ring->rx_skb_prev = skb;
+                               rx_ring->rx_skb_top->data_len += length;
+                       }
                        goto next_desc;
+               } else {
+                       if (rx_ring->rx_skb_top) {
+                               if (skb_shinfo(rx_ring->rx_skb_top)
+                                                       ->frag_list) {
+                                       rx_ring->rx_skb_prev->next = skb;
+                                       skb->prev = rx_ring->rx_skb_prev;
+                               } else
+                                       skb_shinfo(rx_ring->rx_skb_top)
+                                                       ->frag_list = skb;
+
+                               rx_ring->rx_skb_top->data_len += length;
+                               rx_ring->rx_skb_top->len +=
+                                       rx_ring->rx_skb_top->data_len;
+
+                               skb = rx_ring->rx_skb_top;
+                               multi_descriptor = TRUE;
+                               rx_ring->rx_skb_top = NULL;
+                               rx_ring->rx_skb_prev = NULL;
+                       }
                }
 
                if(unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
                        last_byte = *(skb->data + length - 1);
-                       if(TBI_ACCEPT(&adapter->hw, rx_desc->status,
+                       if (TBI_ACCEPT(&adapter->hw, status,
                                      rx_desc->errors, length, last_byte)) {
                                spin_lock_irqsave(&adapter->stats_lock, flags);
                                e1000_tbi_adjust_stats(&adapter->hw,
@@ -3668,7 +3684,7 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                }
 #else /* CONFIG_E1000_NAPI */
                if(unlikely(adapter->vlgrp &&
-                           (rx_desc->status & E1000_RXD_STAT_VP))) {
+                           (status & E1000_RXD_STAT_VP))) {
                        vlan_hwaccel_rx(skb, adapter->vlgrp,
                                        le16_to_cpu(rx_desc->special) &
                                        E1000_RXD_SPC_VLAN_MASK);
@@ -3795,12 +3811,8 @@ e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
                skb->protocol = eth_type_trans(skb, netdev);
 
                if(likely(rx_desc->wb.upper.header_status &
-                         E1000_RXDPS_HDRSTAT_HDRSP)) {
+                         E1000_RXDPS_HDRSTAT_HDRSP))
                        adapter->rx_hdr_split++;
-#ifdef HAVE_RX_ZERO_COPY
-                       skb_shinfo(skb)->zero_copy = TRUE;
-#endif
-               }
 #ifdef CONFIG_E1000_NAPI
                if(unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
                        vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
@@ -3940,20 +3952,22 @@ 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)) == i)) {
-                       /* 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->rdt);
-               }
-
                if(unlikely(++i == rx_ring->count)) i = 0;
                buffer_info = &rx_ring->buffer_info[i];
        }
 
-       rx_ring->next_to_use = i;
+       if (likely(rx_ring->next_to_use != i)) {
+               rx_ring->next_to_use = i;
+               if (unlikely(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->rdt);
+       }
 }
 
 /**
@@ -3988,8 +4002,10 @@ e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
                                if (likely(!ps_page->ps_page[j])) {
                                        ps_page->ps_page[j] =
                                                alloc_page(GFP_ATOMIC);
-                                       if (unlikely(!ps_page->ps_page[j]))
+                                       if (unlikely(!ps_page->ps_page[j])) {
+                                               adapter->alloc_rx_buff_failed++;
                                                goto no_buffers;
+                                       }
                                        ps_page_dma->ps_page_dma[j] =
                                                pci_map_page(pdev,
                                                            ps_page->ps_page[j],
@@ -4008,8 +4024,10 @@ e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 
                skb = dev_alloc_skb(adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 
-               if(unlikely(!skb))
+               if (unlikely(!skb)) {
+                       adapter->alloc_rx_buff_failed++;
                        break;
+               }
 
                /* Make buffer alignment 2 beyond a 16 byte boundary
                 * this will result in a 16 byte aligned IP header after
@@ -4027,19 +4045,6 @@ 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)) == i)) {
-                       /* 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->rdt);
-               }
-
                if(unlikely(++i == rx_ring->count)) i = 0;
                buffer_info = &rx_ring->buffer_info[i];
                ps_page = &rx_ring->ps_page[i];
@@ -4047,7 +4052,21 @@ e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
        }
 
 no_buffers:
-       rx_ring->next_to_use = i;
+       if (likely(rx_ring->next_to_use != i)) {
+               rx_ring->next_to_use = i;
+               if (unlikely(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();
+               /* 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->rdt);
+       }
 }
 
 /**