[PATCH] e1000: revert to single descriptor for legacy receive path
authorJeff Kirsher <jeffrey.t.kirsher@intel.com>
Wed, 1 Mar 2006 04:24:07 +0000 (20:24 -0800)
committerroot <root@jk-desktop.jf.intel.com>
Wed, 1 Mar 2006 04:24:07 +0000 (20:24 -0800)
A recent patch attempted to enable more efficient memory usage by using
only 2kB descriptors for jumbo frames.  The method used to implement this
has since been commented upon as "illegal" and in recent kernels even
causes a BUG when receiving ip fragments while using jumbo frames.
This patch simply goes back to the way things were.  We expect some
complaints due to order 3 allocations failing to come back due to this
change.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
drivers/net/e1000/e1000.h
drivers/net/e1000/e1000_main.c

index 27c77306193b2cb14e17589455ec35ca6d2e4576..99baf0e099fcce432bf17ccfb0735df21b7a2847 100644 (file)
@@ -225,9 +225,6 @@ struct e1000_rx_ring {
        struct e1000_ps_page *ps_page;
        struct e1000_ps_page_dma *ps_page_dma;
 
-       struct sk_buff *rx_skb_top;
-       struct sk_buff *rx_skb_prev;
-
        /* cpu for rx queue */
        int cpu;
 
index 31e332935e5a5dd335e14cbf50091f72197103e7..5b7d0f425af21a76966230ebdb97815bee869d21 100644 (file)
@@ -103,7 +103,7 @@ static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
 #else
 #define DRIVERNAPI "-NAPI"
 #endif
-#define DRV_VERSION "6.3.9-k2"DRIVERNAPI
+#define DRV_VERSION "6.3.9-k4"DRIVERNAPI
 char e1000_driver_version[] = DRV_VERSION;
 static char e1000_copyright[] = "Copyright (c) 1999-2005 Intel Corporation.";
 
@@ -1635,8 +1635,6 @@ setup_rx_desc_die:
 
        rxdr->next_to_clean = 0;
        rxdr->next_to_use = 0;
-       rxdr->rx_skb_top = NULL;
-       rxdr->rx_skb_prev = NULL;
 
        return 0;
 }
@@ -1713,8 +1711,23 @@ e1000_setup_rctl(struct e1000_adapter *adapter)
                rctl |= adapter->rx_buffer_len << 0x11;
        } else {
                rctl &= ~E1000_RCTL_SZ_4096;
-               rctl &= ~E1000_RCTL_BSEX;
-               rctl |= E1000_RCTL_SZ_2048;
+               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;
+               }
        }
 
 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
@@ -2107,16 +2120,6 @@ e1000_clean_rx_ring(struct e1000_adapter *adapter,
                }
        }
 
-       /* there also may be some cached data in our adapter */
-       if (rx_ring->rx_skb_top) {
-               dev_kfree_skb(rx_ring->rx_skb_top);
-
-               /* rx_skb_prev will be wiped out by rx_skb_top */
-               rx_ring->rx_skb_top = NULL;
-               rx_ring->rx_skb_prev = NULL;
-       }
-
-
        size = sizeof(struct e1000_buffer) * rx_ring->count;
        memset(rx_ring->buffer_info, 0, size);
        size = sizeof(struct e1000_ps_page) * rx_ring->count;
@@ -3106,24 +3109,27 @@ e1000_change_mtu(struct net_device *netdev, int new_mtu)
                break;
        }
 
-       /* since the driver code now supports splitting a packet across
-        * multiple descriptors, most of the fifo related limitations on
-        * jumbo frame traffic have gone away.
-        * simply use 2k descriptors for everything.
-        *
-        * NOTE: dev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
-        * means we reserve 2 more, this pushes us to allocate from the next
-        * larger slab size
-        * i.e. RXBUFFER_2048 --> size-4096 slab */
-
-       /* recent hardware supports 1KB granularity */
+
        if (adapter->hw.mac_type > e1000_82547_rev_2) {
-               adapter->rx_buffer_len =
-                   ((max_frame < E1000_RXBUFFER_2048) ?
-                       max_frame : E1000_RXBUFFER_2048);
+               adapter->rx_buffer_len = max_frame;
                E1000_ROUNDUP(adapter->rx_buffer_len, 1024);
-       } else
-               adapter->rx_buffer_len = E1000_RXBUFFER_2048;
+       } else {
+               if(unlikely((adapter->hw.mac_type < e1000_82543) &&
+                  (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE))) {
+                       DPRINTK(PROBE, ERR, "Jumbo Frames not supported "
+                                           "on 82542\n");
+                       return -EINVAL;
+               } else {
+                       if(max_frame <= E1000_RXBUFFER_2048)
+                               adapter->rx_buffer_len = E1000_RXBUFFER_2048;
+                       else if(max_frame <= E1000_RXBUFFER_4096)
+                               adapter->rx_buffer_len = E1000_RXBUFFER_4096;
+                       else if(max_frame <= E1000_RXBUFFER_8192)
+                               adapter->rx_buffer_len = E1000_RXBUFFER_8192;
+                       else if(max_frame <= E1000_RXBUFFER_16384)
+                               adapter->rx_buffer_len = E1000_RXBUFFER_16384;
+               }
+       }
 
        netdev->mtu = new_mtu;
 
@@ -3620,7 +3626,7 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
        uint8_t last_byte;
        unsigned int i;
        int cleaned_count = 0;
-       boolean_t cleaned = FALSE, multi_descriptor = FALSE;
+       boolean_t cleaned = FALSE;
 
        i = rx_ring->next_to_clean;
        rx_desc = E1000_RX_DESC(*rx_ring, i);
@@ -3652,43 +3658,12 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
 
                length = le16_to_cpu(rx_desc->length);
 
-               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;
-                       }
+               if (unlikely(!(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);
                        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)) {
@@ -3712,10 +3687,7 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                 * performance for small packets with large amounts
                 * of reassembly being done in the stack */
 #define E1000_CB_LENGTH 256
-               if ((length < E1000_CB_LENGTH) &&
-                  !rx_ring->rx_skb_top &&
-                  /* or maybe (status & E1000_RXD_STAT_EOP) && */
-                  !multi_descriptor) {
+               if (length < E1000_CB_LENGTH) {
                        struct sk_buff *new_skb =
                            dev_alloc_skb(length + NET_IP_ALIGN);
                        if (new_skb) {
@@ -3729,7 +3701,8 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                                skb = new_skb;
                                skb_put(skb, length);
                        }
-               }
+               } else
+                       skb_put(skb, length);
 
                /* end copybreak code */