e1000: fix flow control thresholds
[safe/jmp/linux-2.6] / drivers / net / e1000 / e1000_main.c
index 2ab44db..d7df00c 100644 (file)
@@ -31,7 +31,7 @@
 
 char e1000_driver_name[] = "e1000";
 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
-#define DRV_VERSION "7.3.20-k3-NAPI"
+#define DRV_VERSION "7.3.21-k3-NAPI"
 const char e1000_driver_version[] = DRV_VERSION;
 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
 
@@ -137,9 +137,15 @@ static int e1000_clean(struct napi_struct *napi, int budget);
 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
                               struct e1000_rx_ring *rx_ring,
                               int *work_done, int work_to_do);
+static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
+                                    struct e1000_rx_ring *rx_ring,
+                                    int *work_done, int work_to_do);
 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
-                                   struct e1000_rx_ring *rx_ring,
+                                  struct e1000_rx_ring *rx_ring,
                                   int cleaned_count);
+static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
+                                        struct e1000_rx_ring *rx_ring,
+                                        int cleaned_count);
 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
                           int cmd);
@@ -156,8 +162,8 @@ static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
 static void e1000_restore_vlan(struct e1000_adapter *adapter);
 
-static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
 #ifdef CONFIG_PM
+static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
 static int e1000_resume(struct pci_dev *pdev);
 #endif
 static void e1000_shutdown(struct pci_dev *pdev);
@@ -498,6 +504,8 @@ int e1000_up(struct e1000_adapter *adapter)
 
        e1000_irq_enable(adapter);
 
+       netif_wake_queue(adapter->netdev);
+
        /* fire a link change interrupt to start the watchdog */
        ew32(ICS, E1000_ICS_LSC);
        return 0;
@@ -577,12 +585,30 @@ out:
 
 void e1000_down(struct e1000_adapter *adapter)
 {
+       struct e1000_hw *hw = &adapter->hw;
        struct net_device *netdev = adapter->netdev;
+       u32 rctl, tctl;
 
        /* signal that we're down so the interrupt handler does not
         * reschedule our watchdog timer */
        set_bit(__E1000_DOWN, &adapter->flags);
 
+       /* disable receives in the hardware */
+       rctl = er32(RCTL);
+       ew32(RCTL, rctl & ~E1000_RCTL_EN);
+       /* flush and sleep below */
+
+       /* can be netif_tx_disable when NETIF_F_LLTX is removed */
+       netif_stop_queue(netdev);
+
+       /* disable transmits in the hardware */
+       tctl = er32(TCTL);
+       tctl &= ~E1000_TCTL_EN;
+       ew32(TCTL, tctl);
+       /* flush both disables and wait for them to finish */
+       E1000_WRITE_FLUSH();
+       msleep(10);
+
        napi_disable(&adapter->napi);
 
        e1000_irq_disable(adapter);
@@ -595,7 +621,6 @@ void e1000_down(struct e1000_adapter *adapter)
        adapter->link_speed = 0;
        adapter->link_duplex = 0;
        netif_carrier_off(netdev);
-       netif_stop_queue(netdev);
 
        e1000_reset(adapter);
        e1000_clean_all_tx_rings(adapter);
@@ -616,8 +641,8 @@ void e1000_reset(struct e1000_adapter *adapter)
 {
        struct e1000_hw *hw = &adapter->hw;
        u32 pba = 0, tx_space, min_tx_space, min_rx_space;
-       u16 fc_high_water_mark = E1000_FC_HIGH_DIFF;
        bool legacy_pba_adjust = false;
+       u16 hwm;
 
        /* Repartition Pba for greater than 9k mtu
         * To take effect CTRL.RST is required.
@@ -661,7 +686,7 @@ void e1000_reset(struct e1000_adapter *adapter)
        }
 
        if (legacy_pba_adjust) {
-               if (adapter->netdev->mtu > E1000_RXBUFFER_8192)
+               if (hw->max_frame_size > E1000_RXBUFFER_8192)
                        pba -= 8; /* allocate more FIFO for Tx */
 
                if (hw->mac_type == e1000_82547) {
@@ -671,14 +696,14 @@ void e1000_reset(struct e1000_adapter *adapter)
                                (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
                        atomic_set(&adapter->tx_fifo_stall, 0);
                }
-       } else if (hw->max_frame_size > MAXIMUM_ETHERNET_FRAME_SIZE) {
+       } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
                /* adjust PBA for jumbo frames */
                ew32(PBA, pba);
 
                /* To maintain wire speed transmits, the Tx FIFO should be
-                * large enough to accomodate two full transmit packets,
+                * large enough to accommodate two full transmit packets,
                 * rounded up to the next 1KB and expressed in KB.  Likewise,
-                * the Rx FIFO should be large enough to accomodate at least
+                * the Rx FIFO should be large enough to accommodate at least
                 * one full receive packet and is similarly rounded up and
                 * expressed in KB. */
                pba = er32(PBA);
@@ -686,13 +711,17 @@ void e1000_reset(struct e1000_adapter *adapter)
                tx_space = pba >> 16;
                /* lower 16 bits has Rx packet buffer allocation size in KB */
                pba &= 0xffff;
-               /* don't include ethernet FCS because hardware appends/strips */
-               min_rx_space = adapter->netdev->mtu + ENET_HEADER_SIZE +
-                              VLAN_TAG_SIZE;
-               min_tx_space = min_rx_space;
-               min_tx_space *= 2;
+               /*
+                * the tx fifo also stores 16 bytes of information about the tx
+                * but don't include ethernet FCS because hardware appends it
+                */
+               min_tx_space = (hw->max_frame_size +
+                               sizeof(struct e1000_tx_desc) -
+                               ETH_FCS_LEN) * 2;
                min_tx_space = ALIGN(min_tx_space, 1024);
                min_tx_space >>= 10;
+               /* software strips receive CRC, so leave room for it */
+               min_rx_space = hw->max_frame_size;
                min_rx_space = ALIGN(min_rx_space, 1024);
                min_rx_space >>= 10;
 
@@ -729,23 +758,22 @@ void e1000_reset(struct e1000_adapter *adapter)
 
        ew32(PBA, pba);
 
-       /* flow control settings */
-       /* Set the FC high water mark to 90% of the FIFO size.
-        * Required to clear last 3 LSB */
-       fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
-       /* We can't use 90% on small FIFOs because the remainder
-        * would be less than 1 full frame.  In this case, we size
-        * it to allow at least a full frame above the high water
-        *  mark. */
-       if (pba < E1000_PBA_16K)
-               fc_high_water_mark = (pba * 1024) - 1600;
-
-       hw->fc_high_water = fc_high_water_mark;
-       hw->fc_low_water = fc_high_water_mark - 8;
-       if (hw->mac_type == e1000_80003es2lan)
-               hw->fc_pause_time = 0xFFFF;
-       else
-               hw->fc_pause_time = E1000_FC_PAUSE_TIME;
+       /*
+        * flow control settings:
+        * The high water mark must be low enough to fit one full frame
+        * (or the size used for early receive) above it in the Rx FIFO.
+        * Set it to the lower of:
+        * - 90% of the Rx FIFO size, and
+        * - the full Rx FIFO size minus the early receive size (for parts
+        *   with ERT support assuming ERT set to E1000_ERT_2048), or
+        * - the full Rx FIFO size minus one full frame
+        */
+       hwm = min(((pba << 10) * 9 / 10),
+                 ((pba << 10) - hw->max_frame_size));
+
+       hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
+       hw->fc_low_water = hw->fc_high_water - 8;
+       hw->fc_pause_time = E1000_FC_PAUSE_TIME;
        hw->fc_send_xon = 1;
        hw->fc = hw->original_fc;
 
@@ -888,6 +916,26 @@ static int e1000_is_need_ioport(struct pci_dev *pdev)
        }
 }
 
+static const struct net_device_ops e1000_netdev_ops = {
+       .ndo_open               = e1000_open,
+       .ndo_stop               = e1000_close,
+       .ndo_start_xmit         = e1000_xmit_frame,
+       .ndo_get_stats          = e1000_get_stats,
+       .ndo_set_rx_mode        = e1000_set_rx_mode,
+       .ndo_set_mac_address    = e1000_set_mac,
+       .ndo_tx_timeout         = e1000_tx_timeout,
+       .ndo_change_mtu         = e1000_change_mtu,
+       .ndo_do_ioctl           = e1000_ioctl,
+       .ndo_validate_addr      = eth_validate_addr,
+
+       .ndo_vlan_rx_register   = e1000_vlan_rx_register,
+       .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
+       .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = e1000_netpoll,
+#endif
+};
+
 /**
  * e1000_probe - Device Initialization Routine
  * @pdev: PCI device information struct
@@ -912,7 +960,6 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
        u16 eeprom_data = 0;
        u16 eeprom_apme_mask = E1000_EEPROM_APME;
        int bars, need_ioport;
-       DECLARE_MAC_BUF(mac);
 
        /* do not allocate ioport bars when not needed */
        need_ioport = e1000_is_need_ioport(pdev);
@@ -921,18 +968,18 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
                err = pci_enable_device(pdev);
        } else {
                bars = pci_select_bars(pdev, IORESOURCE_MEM);
-               err = pci_enable_device(pdev);
+               err = pci_enable_device_mem(pdev);
        }
        if (err)
                return err;
 
-       if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
-           !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
+       if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
+           !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
                pci_using_dac = 1;
        } else {
-               err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+               err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
                if (err) {
-                       err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
+                       err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
                        if (err) {
                                E1000_ERR("No usable DMA configuration, "
                                          "aborting\n");
@@ -967,8 +1014,7 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
        hw->back = adapter;
 
        err = -EIO;
-       hw->hw_addr = ioremap(pci_resource_start(pdev, BAR_0),
-                             pci_resource_len(pdev, BAR_0));
+       hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
        if (!hw->hw_addr)
                goto err_ioremap;
 
@@ -983,24 +1029,11 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
                }
        }
 
-       netdev->open = &e1000_open;
-       netdev->stop = &e1000_close;
-       netdev->hard_start_xmit = &e1000_xmit_frame;
-       netdev->get_stats = &e1000_get_stats;
-       netdev->set_rx_mode = &e1000_set_rx_mode;
-       netdev->set_mac_address = &e1000_set_mac;
-       netdev->change_mtu = &e1000_change_mtu;
-       netdev->do_ioctl = &e1000_ioctl;
+       netdev->netdev_ops = &e1000_netdev_ops;
        e1000_set_ethtool_ops(netdev);
-       netdev->tx_timeout = &e1000_tx_timeout;
        netdev->watchdog_timeo = 5 * HZ;
        netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
-       netdev->vlan_rx_register = e1000_vlan_rx_register;
-       netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
-       netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
-#ifdef CONFIG_NET_POLL_CONTROLLER
-       netdev->poll_controller = e1000_netpoll;
-#endif
+
        strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 
        adapter->bd_number = cards_found;
@@ -1016,9 +1049,7 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
         * because it depends on mac_type */
        if ((hw->mac_type == e1000_ich8lan) &&
           (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
-               hw->flash_address =
-                       ioremap(pci_resource_start(pdev, 1),
-                               pci_resource_len(pdev, 1));
+               hw->flash_address = pci_ioremap_bar(pdev, 1);
                if (!hw->flash_address)
                        goto err_flashmap;
        }
@@ -1045,7 +1076,10 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
        if (pci_using_dac)
                netdev->features |= NETIF_F_HIGHDMA;
 
-       netdev->features |= NETIF_F_LLTX;
+       netdev->vlan_features |= NETIF_F_TSO;
+       netdev->vlan_features |= NETIF_F_TSO6;
+       netdev->vlan_features |= NETIF_F_HW_CSUM;
+       netdev->vlan_features |= NETIF_F_SG;
 
        adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
 
@@ -1174,6 +1208,7 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
 
        /* initialize the wol settings based on the eeprom settings */
        adapter->wol = adapter->eeprom_wol;
+       device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
 
        /* print bus type/speed/width info */
        DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
@@ -1189,7 +1224,7 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
                 (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
                 "32-bit"));
 
-       printk("%s\n", print_mac(mac, netdev->dev_addr));
+       printk("%pM\n", netdev->dev_addr);
 
        if (hw->bus_type == e1000_bus_type_pci_express) {
                DPRINTK(PROBE, WARNING, "This device (id %04x:%04x) will no "
@@ -1210,15 +1245,14 @@ static int __devinit e1000_probe(struct pci_dev *pdev,
            !e1000_check_mng_mode(hw))
                e1000_get_hw_control(adapter);
 
-       /* tell the stack to leave us alone until e1000_open() is called */
-       netif_carrier_off(netdev);
-       netif_stop_queue(netdev);
-
        strcpy(netdev->name, "eth%d");
        err = register_netdev(netdev);
        if (err)
                goto err_register;
 
+       /* carrier off reporting is important to ethtool even BEFORE open */
+       netif_carrier_off(netdev);
+
        DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
 
        cards_found++;
@@ -1233,12 +1267,8 @@ err_eeprom:
        if (hw->flash_address)
                iounmap(hw->flash_address);
 err_flashmap:
-       for (i = 0; i < adapter->num_rx_queues; i++)
-               dev_put(&adapter->polling_netdev[i]);
-
        kfree(adapter->tx_ring);
        kfree(adapter->rx_ring);
-       kfree(adapter->polling_netdev);
 err_sw_init:
        iounmap(hw->hw_addr);
 err_ioremap:
@@ -1266,7 +1296,6 @@ static void __devexit e1000_remove(struct pci_dev *pdev)
        struct net_device *netdev = pci_get_drvdata(pdev);
        struct e1000_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
-       int i;
 
        cancel_work_sync(&adapter->reset_task);
 
@@ -1276,9 +1305,6 @@ static void __devexit e1000_remove(struct pci_dev *pdev)
         * would have already happened in close and is redundant. */
        e1000_release_hw_control(adapter);
 
-       for (i = 0; i < adapter->num_rx_queues; i++)
-               dev_put(&adapter->polling_netdev[i]);
-
        unregister_netdev(netdev);
 
        if (!e1000_check_phy_reset_block(hw))
@@ -1286,7 +1312,6 @@ static void __devexit e1000_remove(struct pci_dev *pdev)
 
        kfree(adapter->tx_ring);
        kfree(adapter->rx_ring);
-       kfree(adapter->polling_netdev);
 
        iounmap(hw->hw_addr);
        if (hw->flash_address)
@@ -1312,7 +1337,6 @@ static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
        struct e1000_hw *hw = &adapter->hw;
        struct net_device *netdev = adapter->netdev;
        struct pci_dev *pdev = adapter->pdev;
-       int i;
 
        /* PCI config space info */
 
@@ -1369,13 +1393,6 @@ static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
                return -ENOMEM;
        }
 
-       for (i = 0; i < adapter->num_rx_queues; i++) {
-               adapter->polling_netdev[i].priv = adapter;
-               dev_hold(&adapter->polling_netdev[i]);
-               set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
-       }
-       spin_lock_init(&adapter->tx_queue_lock);
-
        /* Explicitly disable IRQ since the NIC can be in any state. */
        e1000_irq_disable(adapter);
 
@@ -1391,8 +1408,7 @@ static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
  * @adapter: board private structure to initialize
  *
  * We allocate one ring per queue at run-time since we don't know the
- * number of queues at compile-time.  The polling_netdev array is
- * intended for Multiqueue, but should work fine with a single queue.
+ * number of queues at compile-time.
  **/
 
 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
@@ -1409,15 +1425,6 @@ static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
                return -ENOMEM;
        }
 
-       adapter->polling_netdev = kcalloc(adapter->num_rx_queues,
-                                         sizeof(struct net_device),
-                                         GFP_KERNEL);
-       if (!adapter->polling_netdev) {
-               kfree(adapter->tx_ring);
-               kfree(adapter->rx_ring);
-               return -ENOMEM;
-       }
-
        return E1000_SUCCESS;
 }
 
@@ -1444,6 +1451,8 @@ static int e1000_open(struct net_device *netdev)
        if (test_bit(__E1000_TESTING, &adapter->flags))
                return -EBUSY;
 
+       netif_carrier_off(netdev);
+
        /* allocate transmit descriptors */
        err = e1000_setup_all_tx_resources(adapter);
        if (err)
@@ -1640,7 +1649,6 @@ setup_tx_desc_die:
 
        txdr->next_to_use = 0;
        txdr->next_to_clean = 0;
-       spin_lock_init(&txdr->tx_lock);
 
        return 0;
 }
@@ -1863,6 +1871,7 @@ setup_rx_desc_die:
 
        rxdr->next_to_clean = 0;
        rxdr->next_to_use = 0;
+       rxdr->rx_skb_top = NULL;
 
        return 0;
 }
@@ -1898,8 +1907,6 @@ int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
  * e1000_setup_rctl - configure the receive control registers
  * @adapter: Board private structure
  **/
-#define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
-                       (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
 static void e1000_setup_rctl(struct e1000_adapter *adapter)
 {
        struct e1000_hw *hw = &adapter->hw;
@@ -1971,10 +1978,17 @@ static void e1000_configure_rx(struct e1000_adapter *adapter)
        struct e1000_hw *hw = &adapter->hw;
        u32 rdlen, rctl, rxcsum, ctrl_ext;
 
-       rdlen = adapter->rx_ring[0].count *
-               sizeof(struct e1000_rx_desc);
-       adapter->clean_rx = e1000_clean_rx_irq;
-       adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
+       if (adapter->netdev->mtu > ETH_DATA_LEN) {
+               rdlen = adapter->rx_ring[0].count *
+                       sizeof(struct e1000_rx_desc);
+               adapter->clean_rx = e1000_clean_jumbo_rx_irq;
+               adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
+       } else {
+               rdlen = adapter->rx_ring[0].count *
+                       sizeof(struct e1000_rx_desc);
+               adapter->clean_rx = e1000_clean_rx_irq;
+               adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
+       }
 
        /* disable receives while setting up the descriptors */
        rctl = er32(RCTL);
@@ -2072,17 +2086,14 @@ void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
                                             struct e1000_buffer *buffer_info)
 {
-       if (buffer_info->dma) {
-               pci_unmap_page(adapter->pdev,
-                               buffer_info->dma,
-                               buffer_info->length,
-                               PCI_DMA_TODEVICE);
-               buffer_info->dma = 0;
-       }
+       buffer_info->dma = 0;
        if (buffer_info->skb) {
+               skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
+                             DMA_TO_DEVICE);
                dev_kfree_skb_any(buffer_info->skb);
                buffer_info->skb = NULL;
        }
+       buffer_info->time_stamp = 0;
        /* buffer_info must be completely set up in the transmit path */
 }
 
@@ -2191,22 +2202,39 @@ static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
        /* Free all the Rx ring sk_buffs */
        for (i = 0; i < rx_ring->count; i++) {
                buffer_info = &rx_ring->buffer_info[i];
-               if (buffer_info->skb) {
-                       pci_unmap_single(pdev,
-                                        buffer_info->dma,
-                                        buffer_info->length,
-                                        PCI_DMA_FROMDEVICE);
+               if (buffer_info->dma &&
+                   adapter->clean_rx == e1000_clean_rx_irq) {
+                       pci_unmap_single(pdev, buffer_info->dma,
+                                        buffer_info->length,
+                                        PCI_DMA_FROMDEVICE);
+               } else if (buffer_info->dma &&
+                          adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
+                       pci_unmap_page(pdev, buffer_info->dma,
+                                      buffer_info->length,
+                                      PCI_DMA_FROMDEVICE);
+               }
 
+               buffer_info->dma = 0;
+               if (buffer_info->page) {
+                       put_page(buffer_info->page);
+                       buffer_info->page = NULL;
+               }
+               if (buffer_info->skb) {
                        dev_kfree_skb(buffer_info->skb);
                        buffer_info->skb = NULL;
                }
        }
 
+       /* there also may be some cached data from a chained receive */
+       if (rx_ring->rx_skb_top) {
+               dev_kfree_skb(rx_ring->rx_skb_top);
+               rx_ring->rx_skb_top = NULL;
+       }
+
        size = sizeof(struct e1000_buffer) * rx_ring->count;
        memset(rx_ring->buffer_info, 0, size);
 
        /* Zero out the descriptor ring */
-
        memset(rx_ring->desc, 0, rx_ring->size);
 
        rx_ring->next_to_clean = 0;
@@ -2336,7 +2364,8 @@ static void e1000_set_rx_mode(struct net_device *netdev)
 {
        struct e1000_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
-       struct dev_addr_list *uc_ptr;
+       struct netdev_hw_addr *ha;
+       bool use_uc = false;
        struct dev_addr_list *mc_ptr;
        u32 rctl;
        u32 hash_value;
@@ -2344,6 +2373,12 @@ static void e1000_set_rx_mode(struct net_device *netdev)
        int mta_reg_count = (hw->mac_type == e1000_ich8lan) ?
                                E1000_NUM_MTA_REGISTERS_ICH8LAN :
                                E1000_NUM_MTA_REGISTERS;
+       u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
+
+       if (!mcarray) {
+               DPRINTK(PROBE, ERR, "memory allocation failed\n");
+               return;
+       }
 
        if (hw->mac_type == e1000_ich8lan)
                rar_entries = E1000_RAR_ENTRIES_ICH8LAN;
@@ -2369,12 +2404,11 @@ static void e1000_set_rx_mode(struct net_device *netdev)
                        rctl |= E1000_RCTL_VFE;
        }
 
-       uc_ptr = NULL;
-       if (netdev->uc_count > rar_entries - 1) {
+       if (netdev->uc.count > rar_entries - 1) {
                rctl |= E1000_RCTL_UPE;
        } else if (!(netdev->flags & IFF_PROMISC)) {
                rctl &= ~E1000_RCTL_UPE;
-               uc_ptr = netdev->uc_list;
+               use_uc = true;
        }
 
        ew32(RCTL, rctl);
@@ -2392,13 +2426,20 @@ static void e1000_set_rx_mode(struct net_device *netdev)
         * if there are not 14 addresses, go ahead and clear the filters
         * -- with 82571 controllers only 0-13 entries are filled here
         */
+       i = 1;
+       if (use_uc)
+               list_for_each_entry(ha, &netdev->uc.list, list) {
+                       if (i == rar_entries)
+                               break;
+                       e1000_rar_set(hw, ha->addr, i++);
+               }
+
+       WARN_ON(i == rar_entries);
+
        mc_ptr = netdev->mc_list;
 
-       for (i = 1; i < rar_entries; i++) {
-               if (uc_ptr) {
-                       e1000_rar_set(hw, uc_ptr->da_addr, i);
-                       uc_ptr = uc_ptr->next;
-               } else if (mc_ptr) {
+       for (; i < rar_entries; i++) {
+               if (mc_ptr) {
                        e1000_rar_set(hw, mc_ptr->da_addr, i);
                        mc_ptr = mc_ptr->next;
                } else {
@@ -2408,24 +2449,35 @@ static void e1000_set_rx_mode(struct net_device *netdev)
                        E1000_WRITE_FLUSH();
                }
        }
-       WARN_ON(uc_ptr != NULL);
-
-       /* clear the old settings from the multicast hash table */
-
-       for (i = 0; i < mta_reg_count; i++) {
-               E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
-               E1000_WRITE_FLUSH();
-       }
 
        /* load any remaining addresses into the hash table */
 
        for (; mc_ptr; mc_ptr = mc_ptr->next) {
+               u32 hash_reg, hash_bit, mta;
                hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
-               e1000_mta_set(hw, hash_value);
+               hash_reg = (hash_value >> 5) & 0x7F;
+               hash_bit = hash_value & 0x1F;
+               mta = (1 << hash_bit);
+               mcarray[hash_reg] |= mta;
+       }
+
+       /* write the hash table completely, write from bottom to avoid
+        * both stupid write combining chipsets, and flushing each write */
+       for (i = mta_reg_count - 1; i >= 0 ; i--) {
+               /*
+                * If we are on an 82544 has an errata where writing odd
+                * offsets overwrites the previous even offset, but writing
+                * backwards over the range solves the issue by always
+                * writing the odd offset first
+                */
+               E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
        }
+       E1000_WRITE_FLUSH();
 
        if (hw->mac_type == e1000_82542_rev2_0)
                e1000_leave_82542_rst(adapter);
+
+       kfree(mcarray);
 }
 
 /* Need to wait a few seconds after link up to get diagnostic information from
@@ -2515,10 +2567,11 @@ static void e1000_watchdog(unsigned long data)
                                                   &adapter->link_duplex);
 
                        ctrl = er32(CTRL);
-                       DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s, "
-                               "Flow Control: %s\n",
-                               adapter->link_speed,
-                               adapter->link_duplex == FULL_DUPLEX ?
+                       printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
+                              "Flow Control: %s\n",
+                              netdev->name,
+                              adapter->link_speed,
+                              adapter->link_duplex == FULL_DUPLEX ?
                                "Full Duplex" : "Half Duplex",
                                ((ctrl & E1000_CTRL_TFCE) && (ctrl &
                                E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
@@ -2580,7 +2633,6 @@ static void e1000_watchdog(unsigned long data)
                        ew32(TCTL, tctl);
 
                        netif_carrier_on(netdev);
-                       netif_wake_queue(netdev);
                        mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
                        adapter->smartspeed = 0;
                } else {
@@ -2594,9 +2646,9 @@ static void e1000_watchdog(unsigned long data)
                if (netif_carrier_ok(netdev)) {
                        adapter->link_speed = 0;
                        adapter->link_duplex = 0;
-                       DPRINTK(LINK, INFO, "NIC Link is Down\n");
+                       printk(KERN_INFO "e1000: %s NIC Link is Down\n",
+                              netdev->name);
                        netif_carrier_off(netdev);
-                       netif_stop_queue(netdev);
                        mod_timer(&adapter->phy_info_timer, round_jiffies(jiffies + 2 * HZ));
 
                        /* 80003ES2LAN workaround--
@@ -2634,6 +2686,8 @@ static void e1000_watchdog(unsigned long data)
                         * (Do the reset outside of interrupt context). */
                        adapter->tx_timeout_count++;
                        schedule_work(&adapter->reset_task);
+                       /* return immediately since reset is imminent */
+                       return;
                }
        }
 
@@ -2873,32 +2927,49 @@ static bool e1000_tx_csum(struct e1000_adapter *adapter,
        struct e1000_buffer *buffer_info;
        unsigned int i;
        u8 css;
+       u32 cmd_len = E1000_TXD_CMD_DEXT;
 
-       if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
-               css = skb_transport_offset(skb);
+       if (skb->ip_summed != CHECKSUM_PARTIAL)
+               return false;
 
-               i = tx_ring->next_to_use;
-               buffer_info = &tx_ring->buffer_info[i];
-               context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
+       switch (skb->protocol) {
+       case cpu_to_be16(ETH_P_IP):
+               if (ip_hdr(skb)->protocol == IPPROTO_TCP)
+                       cmd_len |= E1000_TXD_CMD_TCP;
+               break;
+       case cpu_to_be16(ETH_P_IPV6):
+               /* XXX not handling all IPV6 headers */
+               if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
+                       cmd_len |= E1000_TXD_CMD_TCP;
+               break;
+       default:
+               if (unlikely(net_ratelimit()))
+                       DPRINTK(DRV, WARNING,
+                               "checksum_partial proto=%x!\n", skb->protocol);
+               break;
+       }
 
-               context_desc->lower_setup.ip_config = 0;
-               context_desc->upper_setup.tcp_fields.tucss = css;
-               context_desc->upper_setup.tcp_fields.tucso =
-                       css + skb->csum_offset;
-               context_desc->upper_setup.tcp_fields.tucse = 0;
-               context_desc->tcp_seg_setup.data = 0;
-               context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
+       css = skb_transport_offset(skb);
 
-               buffer_info->time_stamp = jiffies;
-               buffer_info->next_to_watch = i;
+       i = tx_ring->next_to_use;
+       buffer_info = &tx_ring->buffer_info[i];
+       context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
 
-               if (unlikely(++i == tx_ring->count)) i = 0;
-               tx_ring->next_to_use = i;
+       context_desc->lower_setup.ip_config = 0;
+       context_desc->upper_setup.tcp_fields.tucss = css;
+       context_desc->upper_setup.tcp_fields.tucso =
+               css + skb->csum_offset;
+       context_desc->upper_setup.tcp_fields.tucse = 0;
+       context_desc->tcp_seg_setup.data = 0;
+       context_desc->cmd_and_length = cpu_to_le32(cmd_len);
 
-               return true;
-       }
+       buffer_info->time_stamp = jiffies;
+       buffer_info->next_to_watch = i;
 
-       return false;
+       if (unlikely(++i == tx_ring->count)) i = 0;
+       tx_ring->next_to_use = i;
+
+       return true;
 }
 
 #define E1000_MAX_TXD_PWR      12
@@ -2912,13 +2983,21 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
 {
        struct e1000_hw *hw = &adapter->hw;
        struct e1000_buffer *buffer_info;
-       unsigned int len = skb->len;
-       unsigned int offset = 0, size, count = 0, i;
+       unsigned int len = skb_headlen(skb);
+       unsigned int offset, size, count = 0, i;
        unsigned int f;
-       len -= skb->data_len;
+       dma_addr_t *map;
 
        i = tx_ring->next_to_use;
 
+       if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
+               dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
+               return 0;
+       }
+
+       map = skb_shinfo(skb)->dma_maps;
+       offset = 0;
+
        while (len) {
                buffer_info = &tx_ring->buffer_info[i];
                size = min(len, max_per_txd);
@@ -2953,18 +3032,18 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
                        size -= 4;
 
                buffer_info->length = size;
-               buffer_info->dma =
-                       pci_map_single(adapter->pdev,
-                               skb->data + offset,
-                               size,
-                               PCI_DMA_TODEVICE);
+               buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
                buffer_info->time_stamp = jiffies;
                buffer_info->next_to_watch = i;
 
                len -= size;
                offset += size;
                count++;
-               if (unlikely(++i == tx_ring->count)) i = 0;
+               if (len) {
+                       i++;
+                       if (unlikely(i == tx_ring->count))
+                               i = 0;
+               }
        }
 
        for (f = 0; f < nr_frags; f++) {
@@ -2972,9 +3051,13 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
 
                frag = &skb_shinfo(skb)->frags[f];
                len = frag->size;
-               offset = frag->page_offset;
+               offset = 0;
 
                while (len) {
+                       i++;
+                       if (unlikely(i == tx_ring->count))
+                               i = 0;
+
                        buffer_info = &tx_ring->buffer_info[i];
                        size = min(len, max_per_txd);
                        /* Workaround for premature desc write-backs
@@ -2990,23 +3073,16 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
                                size -= 4;
 
                        buffer_info->length = size;
-                       buffer_info->dma =
-                               pci_map_page(adapter->pdev,
-                                       frag->page,
-                                       offset,
-                                       size,
-                                       PCI_DMA_TODEVICE);
+                       buffer_info->dma = map[f] + offset;
                        buffer_info->time_stamp = jiffies;
                        buffer_info->next_to_watch = i;
 
                        len -= size;
                        offset += size;
                        count++;
-                       if (unlikely(++i == tx_ring->count)) i = 0;
                }
        }
 
-       i = (i == 0) ? tx_ring->count - 1 : i - 1;
        tx_ring->buffer_info[i].skb = skb;
        tx_ring->buffer_info[first].next_to_watch = i;
 
@@ -3182,7 +3258,6 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
        unsigned int tx_flags = 0;
        unsigned int len = skb->len - skb->data_len;
-       unsigned long flags;
        unsigned int nr_frags;
        unsigned int mss;
        int count = 0;
@@ -3287,22 +3362,15 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
            (hw->mac_type == e1000_82573))
                e1000_transfer_dhcp_info(adapter, skb);
 
-       if (!spin_trylock_irqsave(&tx_ring->tx_lock, flags))
-               /* Collision - tell upper layer to requeue */
-               return NETDEV_TX_LOCKED;
-
        /* need: count + 2 desc gap to keep tail from touching
         * head, otherwise try next time */
-       if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
-               spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
+       if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
                return NETDEV_TX_BUSY;
-       }
 
        if (unlikely(hw->mac_type == e1000_82547)) {
                if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
                        netif_stop_queue(netdev);
                        mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
-                       spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
                        return NETDEV_TX_BUSY;
                }
        }
@@ -3317,7 +3385,6 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        tso = e1000_tso(adapter, tx_ring, skb);
        if (tso < 0) {
                dev_kfree_skb_any(skb);
-               spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
                return NETDEV_TX_OK;
        }
 
@@ -3333,16 +3400,20 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        if (likely(skb->protocol == htons(ETH_P_IP)))
                tx_flags |= E1000_TX_FLAGS_IPV4;
 
-       e1000_tx_queue(adapter, tx_ring, tx_flags,
-                      e1000_tx_map(adapter, tx_ring, skb, first,
-                                   max_per_txd, nr_frags, mss));
+       count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
+                            nr_frags, mss);
 
-       netdev->trans_start = jiffies;
+       if (count) {
+               e1000_tx_queue(adapter, tx_ring, tx_flags, count);
+               /* Make sure there is space in the ring for the next send. */
+               e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
 
-       /* Make sure there is space in the ring for the next send. */
-       e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
+       } else {
+               dev_kfree_skb_any(skb);
+               tx_ring->buffer_info[first].time_stamp = 0;
+               tx_ring->next_to_use = first;
+       }
 
-       spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
        return NETDEV_TX_OK;
 }
 
@@ -3409,7 +3480,7 @@ static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
        switch (hw->mac_type) {
        case e1000_undefined ... e1000_82542_rev2_1:
        case e1000_ich8lan:
-               if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
+               if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
                        DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
                        return -EINVAL;
                }
@@ -3422,7 +3493,7 @@ static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
                                  &eeprom_data);
                if ((hw->device_id != E1000_DEV_ID_82573L) ||
                    (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
-                       if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
+                       if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
                                DPRINTK(PROBE, ERR,
                                        "Jumbo Frames not supported.\n");
                                return -EINVAL;
@@ -3448,8 +3519,10 @@ static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
 
        /* NOTE: netdev_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 */
+        * larger slab size.
+        * i.e. RXBUFFER_2048 --> size-4096 slab
+        *  however with the new *_jumbo_rx* routines, jumbo receives will use
+        *  fragmented skbs */
 
        if (max_frame <= E1000_RXBUFFER_256)
                adapter->rx_buffer_len = E1000_RXBUFFER_256;
@@ -3459,16 +3532,16 @@ static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
                adapter->rx_buffer_len = E1000_RXBUFFER_1024;
        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)
+       else
+#if (PAGE_SIZE >= E1000_RXBUFFER_16384)
                adapter->rx_buffer_len = E1000_RXBUFFER_16384;
+#elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
+               adapter->rx_buffer_len = PAGE_SIZE;
+#endif
 
        /* adjust allocation if LPE protects us, and we aren't using SBP */
        if (!hw->tbi_compatibility_on &&
-           ((max_frame == MAXIMUM_ETHERNET_FRAME_SIZE) ||
+           ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
             (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
                adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
 
@@ -3684,12 +3757,12 @@ static irqreturn_t e1000_intr_msi(int irq, void *data)
                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
        }
 
-       if (likely(netif_rx_schedule_prep(netdev, &adapter->napi))) {
+       if (likely(napi_schedule_prep(&adapter->napi))) {
                adapter->total_tx_bytes = 0;
                adapter->total_tx_packets = 0;
                adapter->total_rx_bytes = 0;
                adapter->total_rx_packets = 0;
-               __netif_rx_schedule(netdev, &adapter->napi);
+               __napi_schedule(&adapter->napi);
        } else
                e1000_irq_enable(adapter);
 
@@ -3709,7 +3782,7 @@ static irqreturn_t e1000_intr(int irq, void *data)
        struct e1000_hw *hw = &adapter->hw;
        u32 rctl, icr = er32(ICR);
 
-       if (unlikely(!icr))
+       if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
                return IRQ_NONE;  /* Not our interrupt */
 
        /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
@@ -3744,16 +3817,18 @@ static irqreturn_t e1000_intr(int irq, void *data)
                ew32(IMC, ~0);
                E1000_WRITE_FLUSH();
        }
-       if (likely(netif_rx_schedule_prep(netdev, &adapter->napi))) {
+       if (likely(napi_schedule_prep(&adapter->napi))) {
                adapter->total_tx_bytes = 0;
                adapter->total_tx_packets = 0;
                adapter->total_rx_bytes = 0;
                adapter->total_rx_packets = 0;
-               __netif_rx_schedule(netdev, &adapter->napi);
-       } else
+               __napi_schedule(&adapter->napi);
+       } else {
                /* this really should not happen! if it does it is basically a
                 * bug, but not a hard error, so enable ints and continue */
-               e1000_irq_enable(adapter);
+               if (!test_bit(__E1000_DOWN, &adapter->flags))
+                       e1000_irq_enable(adapter);
+       }
 
        return IRQ_HANDLED;
 }
@@ -3768,31 +3843,23 @@ static int e1000_clean(struct napi_struct *napi, int budget)
        struct net_device *poll_dev = adapter->netdev;
        int tx_cleaned = 0, work_done = 0;
 
-       /* Must NOT use netdev_priv macro here. */
-       adapter = poll_dev->priv;
+       adapter = netdev_priv(poll_dev);
 
-       /* e1000_clean is called per-cpu.  This lock protects
-        * tx_ring[0] from being cleaned by multiple cpus
-        * simultaneously.  A failure obtaining the lock means
-        * tx_ring[0] is currently being cleaned anyway. */
-       if (spin_trylock(&adapter->tx_queue_lock)) {
-               tx_cleaned = e1000_clean_tx_irq(adapter,
-                                               &adapter->tx_ring[0]);
-               spin_unlock(&adapter->tx_queue_lock);
-       }
+       tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
 
        adapter->clean_rx(adapter, &adapter->rx_ring[0],
                          &work_done, budget);
 
-       if (tx_cleaned)
+       if (!tx_cleaned)
                work_done = budget;
 
        /* If budget not fully consumed, exit the polling mode */
        if (work_done < budget) {
                if (likely(adapter->itr_setting & 3))
                        e1000_set_itr(adapter);
-               netif_rx_complete(poll_dev, napi);
-               e1000_irq_enable(adapter);
+               napi_complete(napi);
+               if (!test_bit(__E1000_DOWN, &adapter->flags))
+                       e1000_irq_enable(adapter);
        }
 
        return work_done;
@@ -3811,15 +3878,16 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
        struct e1000_buffer *buffer_info;
        unsigned int i, eop;
        unsigned int count = 0;
-       bool cleaned = false;
        unsigned int total_tx_bytes=0, total_tx_packets=0;
 
        i = tx_ring->next_to_clean;
        eop = tx_ring->buffer_info[i].next_to_watch;
        eop_desc = E1000_TX_DESC(*tx_ring, eop);
 
-       while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
-               for (cleaned = false; !cleaned; ) {
+       while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
+              (count < tx_ring->count)) {
+               bool cleaned = false;
+               for ( ; !cleaned; count++) {
                        tx_desc = E1000_TX_DESC(*tx_ring, i);
                        buffer_info = &tx_ring->buffer_info[i];
                        cleaned = (i == eop);
@@ -3842,16 +3910,12 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
 
                eop = tx_ring->buffer_info[i].next_to_watch;
                eop_desc = E1000_TX_DESC(*tx_ring, eop);
-#define E1000_TX_WEIGHT 64
-               /* weight of a sort for tx, to avoid endless transmit cleanup */
-               if (count++ == E1000_TX_WEIGHT)
-                       break;
        }
 
        tx_ring->next_to_clean = i;
 
 #define TX_WAKE_THRESHOLD 32
-       if (unlikely(cleaned && netif_carrier_ok(netdev) &&
+       if (unlikely(count && netif_carrier_ok(netdev) &&
                     E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
                /* Make sure that anybody stopping the queue after this
                 * sees the new next_to_clean.
@@ -3867,8 +3931,8 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
                /* Detect a transmit hang in hardware, this serializes the
                 * check with the clearing of time_stamp and movement of i */
                adapter->detect_tx_hung = false;
-               if (tx_ring->buffer_info[eop].dma &&
-                   time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
+               if (tx_ring->buffer_info[i].time_stamp &&
+                   time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
                               (adapter->tx_timeout_factor * HZ))
                    && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
 
@@ -3890,7 +3954,7 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
                                readl(hw->hw_addr + tx_ring->tdt),
                                tx_ring->next_to_use,
                                tx_ring->next_to_clean,
-                               tx_ring->buffer_info[eop].time_stamp,
+                               tx_ring->buffer_info[i].time_stamp,
                                eop,
                                jiffies,
                                eop_desc->upper.fields.status);
@@ -3901,7 +3965,7 @@ static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
        adapter->total_tx_packets += total_tx_packets;
        adapter->net_stats.tx_bytes += total_tx_bytes;
        adapter->net_stats.tx_packets += total_tx_packets;
-       return cleaned;
+       return (count < tx_ring->count);
 }
 
 /**
@@ -3955,9 +4019,227 @@ static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 }
 
 /**
+ * e1000_consume_page - helper function
+ **/
+static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
+                               u16 length)
+{
+       bi->page = NULL;
+       skb->len += length;
+       skb->data_len += length;
+       skb->truesize += length;
+}
+
+/**
+ * e1000_receive_skb - helper function to handle rx indications
+ * @adapter: board private structure
+ * @status: descriptor status field as written by hardware
+ * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
+ * @skb: pointer to sk_buff to be indicated to stack
+ */
+static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
+                             __le16 vlan, struct sk_buff *skb)
+{
+       if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
+               vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
+                                        le16_to_cpu(vlan) &
+                                        E1000_RXD_SPC_VLAN_MASK);
+       } else {
+               netif_receive_skb(skb);
+       }
+}
+
+/**
+ * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
+ * @adapter: board private structure
+ * @rx_ring: ring to clean
+ * @work_done: amount of napi work completed this call
+ * @work_to_do: max amount of work allowed for this call to do
+ *
+ * the return value indicates whether actual cleaning was done, there
+ * is no guarantee that everything was cleaned
+ */
+static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
+                                    struct e1000_rx_ring *rx_ring,
+                                    int *work_done, int work_to_do)
+{
+       struct e1000_hw *hw = &adapter->hw;
+       struct net_device *netdev = adapter->netdev;
+       struct pci_dev *pdev = adapter->pdev;
+       struct e1000_rx_desc *rx_desc, *next_rxd;
+       struct e1000_buffer *buffer_info, *next_buffer;
+       unsigned long irq_flags;
+       u32 length;
+       unsigned int i;
+       int cleaned_count = 0;
+       bool cleaned = false;
+       unsigned int total_rx_bytes=0, total_rx_packets=0;
+
+       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) {
+               struct sk_buff *skb;
+               u8 status;
+
+               if (*work_done >= work_to_do)
+                       break;
+               (*work_done)++;
+
+               status = rx_desc->status;
+               skb = buffer_info->skb;
+               buffer_info->skb = NULL;
+
+               if (++i == rx_ring->count) i = 0;
+               next_rxd = E1000_RX_DESC(*rx_ring, i);
+               prefetch(next_rxd);
+
+               next_buffer = &rx_ring->buffer_info[i];
+
+               cleaned = true;
+               cleaned_count++;
+               pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
+                              PCI_DMA_FROMDEVICE);
+               buffer_info->dma = 0;
+
+               length = le16_to_cpu(rx_desc->length);
+
+               /* errors is only valid for DD + EOP descriptors */
+               if (unlikely((status & E1000_RXD_STAT_EOP) &&
+                   (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
+                       u8 last_byte = *(skb->data + length - 1);
+                       if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
+                                      last_byte)) {
+                               spin_lock_irqsave(&adapter->stats_lock,
+                                                 irq_flags);
+                               e1000_tbi_adjust_stats(hw, &adapter->stats,
+                                                      length, skb->data);
+                               spin_unlock_irqrestore(&adapter->stats_lock,
+                                                      irq_flags);
+                               length--;
+                       } else {
+                               /* recycle both page and skb */
+                               buffer_info->skb = skb;
+                               /* an error means any chain goes out the window
+                                * too */
+                               if (rx_ring->rx_skb_top)
+                                       dev_kfree_skb(rx_ring->rx_skb_top);
+                               rx_ring->rx_skb_top = NULL;
+                               goto next_desc;
+                       }
+               }
+
+#define rxtop rx_ring->rx_skb_top
+               if (!(status & E1000_RXD_STAT_EOP)) {
+                       /* this descriptor is only the beginning (or middle) */
+                       if (!rxtop) {
+                               /* this is the beginning of a chain */
+                               rxtop = skb;
+                               skb_fill_page_desc(rxtop, 0, buffer_info->page,
+                                                  0, length);
+                       } else {
+                               /* this is the middle of a chain */
+                               skb_fill_page_desc(rxtop,
+                                   skb_shinfo(rxtop)->nr_frags,
+                                   buffer_info->page, 0, length);
+                               /* re-use the skb, only consumed the page */
+                               buffer_info->skb = skb;
+                       }
+                       e1000_consume_page(buffer_info, rxtop, length);
+                       goto next_desc;
+               } else {
+                       if (rxtop) {
+                               /* end of the chain */
+                               skb_fill_page_desc(rxtop,
+                                   skb_shinfo(rxtop)->nr_frags,
+                                   buffer_info->page, 0, length);
+                               /* re-use the current skb, we only consumed the
+                                * page */
+                               buffer_info->skb = skb;
+                               skb = rxtop;
+                               rxtop = NULL;
+                               e1000_consume_page(buffer_info, skb, length);
+                       } else {
+                               /* no chain, got EOP, this buf is the packet
+                                * copybreak to save the put_page/alloc_page */
+                               if (length <= copybreak &&
+                                   skb_tailroom(skb) >= length) {
+                                       u8 *vaddr;
+                                       vaddr = kmap_atomic(buffer_info->page,
+                                                           KM_SKB_DATA_SOFTIRQ);
+                                       memcpy(skb_tail_pointer(skb), vaddr, length);
+                                       kunmap_atomic(vaddr,
+                                                     KM_SKB_DATA_SOFTIRQ);
+                                       /* re-use the page, so don't erase
+                                        * buffer_info->page */
+                                       skb_put(skb, length);
+                               } else {
+                                       skb_fill_page_desc(skb, 0,
+                                                          buffer_info->page, 0,
+                                                          length);
+                                       e1000_consume_page(buffer_info, skb,
+                                                          length);
+                               }
+                       }
+               }
+
+               /* Receive Checksum Offload XXX recompute due to CRC strip? */
+               e1000_rx_checksum(adapter,
+                                 (u32)(status) |
+                                 ((u32)(rx_desc->errors) << 24),
+                                 le16_to_cpu(rx_desc->csum), skb);
+
+               pskb_trim(skb, skb->len - 4);
+
+               /* probably a little skewed due to removing CRC */
+               total_rx_bytes += skb->len;
+               total_rx_packets++;
+
+               /* eth type trans needs skb->data to point to something */
+               if (!pskb_may_pull(skb, ETH_HLEN)) {
+                       DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
+                       dev_kfree_skb(skb);
+                       goto next_desc;
+               }
+
+               skb->protocol = eth_type_trans(skb, netdev);
+
+               e1000_receive_skb(adapter, status, rx_desc->special, skb);
+
+next_desc:
+               rx_desc->status = 0;
+
+               /* return some buffers to hardware, one at a time is too slow */
+               if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
+                       adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
+                       cleaned_count = 0;
+               }
+
+               /* use prefetched values */
+               rx_desc = next_rxd;
+               buffer_info = next_buffer;
+       }
+       rx_ring->next_to_clean = i;
+
+       cleaned_count = E1000_DESC_UNUSED(rx_ring);
+       if (cleaned_count)
+               adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
+
+       adapter->total_rx_packets += total_rx_packets;
+       adapter->total_rx_bytes += total_rx_bytes;
+       adapter->net_stats.rx_bytes += total_rx_bytes;
+       adapter->net_stats.rx_packets += total_rx_packets;
+       return cleaned;
+}
+
+/**
  * e1000_clean_rx_irq - Send received data up the network stack; legacy
  * @adapter: board private structure
- **/
+ * @rx_ring: ring to clean
+ * @work_done: amount of napi work completed this call
+ * @work_to_do: max amount of work allowed for this call to do
+ */
 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
                               struct e1000_rx_ring *rx_ring,
                               int *work_done, int work_to_do)
@@ -3969,7 +4251,6 @@ static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
        struct e1000_buffer *buffer_info, *next_buffer;
        unsigned long flags;
        u32 length;
-       u8 last_byte;
        unsigned int i;
        int cleaned_count = 0;
        bool cleaned = false;
@@ -4001,14 +4282,14 @@ static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 
                cleaned = true;
                cleaned_count++;
-               pci_unmap_single(pdev,
-                                buffer_info->dma,
-                                buffer_info->length,
+               pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
                                 PCI_DMA_FROMDEVICE);
+               buffer_info->dma = 0;
 
                length = le16_to_cpu(rx_desc->length);
-
-               if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
+               /* !EOP means multiple descriptors were used to store a single
+                * packet, also make sure the frame isn't just CRC only */
+               if (unlikely(!(status & E1000_RXD_STAT_EOP) || (length <= 4))) {
                        /* All receives must fit into a single buffer */
                        E1000_DBG("%s: Receive packet consumed multiple"
                                  " buffers\n", netdev->name);
@@ -4018,7 +4299,7 @@ static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
                }
 
                if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
-                       last_byte = *(skb->data + length - 1);
+                       u8 last_byte = *(skb->data + length - 1);
                        if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
                                       last_byte)) {
                                spin_lock_irqsave(&adapter->stats_lock, flags);
@@ -4073,15 +4354,7 @@ static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 
                skb->protocol = eth_type_trans(skb, netdev);
 
-               if (unlikely(adapter->vlgrp &&
-                           (status & E1000_RXD_STAT_VP))) {
-                       vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
-                                                le16_to_cpu(rx_desc->special));
-               } else {
-                       netif_receive_skb(skb);
-               }
-
-               netdev->last_rx = jiffies;
+               e1000_receive_skb(adapter, status, rx_desc->special, skb);
 
 next_desc:
                rx_desc->status = 0;
@@ -4110,6 +4383,114 @@ next_desc:
 }
 
 /**
+ * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
+ * @adapter: address of board private structure
+ * @rx_ring: pointer to receive ring structure
+ * @cleaned_count: number of buffers to allocate this pass
+ **/
+
+static void
+e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
+                             struct e1000_rx_ring *rx_ring, int cleaned_count)
+{
+       struct net_device *netdev = adapter->netdev;
+       struct pci_dev *pdev = adapter->pdev;
+       struct e1000_rx_desc *rx_desc;
+       struct e1000_buffer *buffer_info;
+       struct sk_buff *skb;
+       unsigned int i;
+       unsigned int bufsz = 256 -
+                            16 /*for skb_reserve */ -
+                            NET_IP_ALIGN;
+
+       i = rx_ring->next_to_use;
+       buffer_info = &rx_ring->buffer_info[i];
+
+       while (cleaned_count--) {
+               skb = buffer_info->skb;
+               if (skb) {
+                       skb_trim(skb, 0);
+                       goto check_page;
+               }
+
+               skb = netdev_alloc_skb(netdev, bufsz);
+               if (unlikely(!skb)) {
+                       /* Better luck next round */
+                       adapter->alloc_rx_buff_failed++;
+                       break;
+               }
+
+               /* Fix for errata 23, can't cross 64kB boundary */
+               if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
+                       struct sk_buff *oldskb = skb;
+                       DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
+                                            "at %p\n", bufsz, skb->data);
+                       /* Try again, without freeing the previous */
+                       skb = netdev_alloc_skb(netdev, bufsz);
+                       /* Failed allocation, critical failure */
+                       if (!skb) {
+                               dev_kfree_skb(oldskb);
+                               adapter->alloc_rx_buff_failed++;
+                               break;
+                       }
+
+                       if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
+                               /* give up */
+                               dev_kfree_skb(skb);
+                               dev_kfree_skb(oldskb);
+                               break; /* while (cleaned_count--) */
+                       }
+
+                       /* Use new allocation */
+                       dev_kfree_skb(oldskb);
+               }
+               /* Make buffer alignment 2 beyond a 16 byte boundary
+                * this will result in a 16 byte aligned IP header after
+                * the 14 byte MAC header is removed
+                */
+               skb_reserve(skb, NET_IP_ALIGN);
+
+               buffer_info->skb = skb;
+               buffer_info->length = adapter->rx_buffer_len;
+check_page:
+               /* allocate a new page if necessary */
+               if (!buffer_info->page) {
+                       buffer_info->page = alloc_page(GFP_ATOMIC);
+                       if (unlikely(!buffer_info->page)) {
+                               adapter->alloc_rx_buff_failed++;
+                               break;
+                       }
+               }
+
+               if (!buffer_info->dma)
+                       buffer_info->dma = pci_map_page(pdev,
+                                                       buffer_info->page, 0,
+                                                       buffer_info->length,
+                                                       PCI_DMA_FROMDEVICE);
+
+               rx_desc = E1000_RX_DESC(*rx_ring, i);
+               rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
+
+               if (unlikely(++i == rx_ring->count))
+                       i = 0;
+               buffer_info = &rx_ring->buffer_info[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);
+       }
+}
+
+/**
  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
  * @adapter: address of board private structure
  **/
@@ -4154,6 +4535,7 @@ static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
                        /* Failed allocation, critical failure */
                        if (!skb) {
                                dev_kfree_skb(oldskb);
+                               adapter->alloc_rx_buff_failed++;
                                break;
                        }
 
@@ -4161,6 +4543,7 @@ static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
                                /* give up */
                                dev_kfree_skb(skb);
                                dev_kfree_skb(oldskb);
+                               adapter->alloc_rx_buff_failed++;
                                break; /* while !buffer_info->skb */
                        }
 
@@ -4178,9 +4561,14 @@ static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 map_skb:
                buffer_info->dma = pci_map_single(pdev,
                                                  skb->data,
-                                                 adapter->rx_buffer_len,
+                                                 buffer_info->length,
                                                  PCI_DMA_FROMDEVICE);
 
+               /*
+                * XXX if it was allocated cleanly it will never map to a
+                * boundary crossing
+                */
+
                /* Fix for errata 23, can't cross 64kB boundary */
                if (!e1000_check_64k_bound(adapter,
                                        (void *)(unsigned long)buffer_info->dma,
@@ -4195,7 +4583,9 @@ map_skb:
                        pci_unmap_single(pdev, buffer_info->dma,
                                         adapter->rx_buffer_len,
                                         PCI_DMA_FROMDEVICE);
+                       buffer_info->dma = 0;
 
+                       adapter->alloc_rx_buff_failed++;
                        break; /* while !buffer_info->skb */
                }
                rx_desc = E1000_RX_DESC(*rx_ring, i);
@@ -4583,7 +4973,7 @@ int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
        return 0;
 }
 
-static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
+static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
        struct e1000_adapter *adapter = netdev_priv(netdev);
@@ -4646,22 +5036,18 @@ static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
 
                ew32(WUC, E1000_WUC_PME_EN);
                ew32(WUFC, wufc);
-               pci_enable_wake(pdev, PCI_D3hot, 1);
-               pci_enable_wake(pdev, PCI_D3cold, 1);
        } else {
                ew32(WUC, 0);
                ew32(WUFC, 0);
-               pci_enable_wake(pdev, PCI_D3hot, 0);
-               pci_enable_wake(pdev, PCI_D3cold, 0);
        }
 
        e1000_release_manageability(adapter);
 
+       *enable_wake = !!wufc;
+
        /* make sure adapter isn't asleep if manageability is enabled */
-       if (adapter->en_mng_pt) {
-               pci_enable_wake(pdev, PCI_D3hot, 1);
-               pci_enable_wake(pdev, PCI_D3cold, 1);
-       }
+       if (adapter->en_mng_pt)
+               *enable_wake = true;
 
        if (hw->phy_type == e1000_phy_igp_3)
                e1000_phy_powerdown_workaround(hw);
@@ -4675,12 +5061,29 @@ static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
 
        pci_disable_device(pdev);
 
-       pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
        return 0;
 }
 
 #ifdef CONFIG_PM
+static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+       int retval;
+       bool wake;
+
+       retval = __e1000_shutdown(pdev, &wake);
+       if (retval)
+               return retval;
+
+       if (wake) {
+               pci_prepare_to_sleep(pdev);
+       } else {
+               pci_wake_from_d3(pdev, false);
+               pci_set_power_state(pdev, PCI_D3hot);
+       }
+
+       return 0;
+}
+
 static int e1000_resume(struct pci_dev *pdev)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
@@ -4735,7 +5138,14 @@ static int e1000_resume(struct pci_dev *pdev)
 
 static void e1000_shutdown(struct pci_dev *pdev)
 {
-       e1000_suspend(pdev, PMSG_SUSPEND);
+       bool wake;
+
+       __e1000_shutdown(pdev, &wake);
+
+       if (system_state == SYSTEM_POWER_OFF) {
+               pci_wake_from_d3(pdev, wake);
+               pci_set_power_state(pdev, PCI_D3hot);
+       }
 }
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
@@ -4766,10 +5176,13 @@ static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
                                                pci_channel_state_t state)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
-       struct e1000_adapter *adapter = netdev->priv;
+       struct e1000_adapter *adapter = netdev_priv(netdev);
 
        netif_device_detach(netdev);
 
+       if (state == pci_channel_io_perm_failure)
+               return PCI_ERS_RESULT_DISCONNECT;
+
        if (netif_running(netdev))
                e1000_down(adapter);
        pci_disable_device(pdev);
@@ -4788,7 +5201,7 @@ static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
-       struct e1000_adapter *adapter = netdev->priv;
+       struct e1000_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
        int err;
 
@@ -4822,7 +5235,7 @@ static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
 static void e1000_io_resume(struct pci_dev *pdev)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
-       struct e1000_adapter *adapter = netdev->priv;
+       struct e1000_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
 
        e1000_init_manageability(adapter);