]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blobdiff - drivers/net/e1000/e1000_main.c
e1000: omit stats for broken counter in 82543
[mirror_ubuntu-artful-kernel.git] / drivers / net / e1000 / e1000_main.c
index 3f6a752700a101a93ee18fef71518492a1e233ed..0816de267610a58fa9250bb595ddc6865b8ea7de 100644 (file)
@@ -1,25 +1,24 @@
 /*******************************************************************************
 
-  
-  Copyright(c) 1999 - 2006 Intel Corporation. All rights reserved.
-  
-  This program is free software; you can redistribute it and/or modify it 
-  under the terms of the GNU General Public License as published by the Free 
-  Software Foundation; either version 2 of the License, or (at your option) 
-  any later version.
-  
-  This program is distributed in the hope that it will be useful, but WITHOUT 
-  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 
-  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for 
+  Intel PRO/1000 Linux driver
+  Copyright(c) 1999 - 2006 Intel Corporation.
+
+  This program is free software; you can redistribute it and/or modify it
+  under the terms and conditions of the GNU General Public License,
+  version 2, as published by the Free Software Foundation.
+
+  This program is distributed in the hope it will be useful, but WITHOUT
+  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
   more details.
-  
+
   You should have received a copy of the GNU General Public License along with
-  this program; if not, write to the Free Software Foundation, Inc., 59 
-  Temple Place - Suite 330, Boston, MA  02111-1307, USA.
-  
-  The full GNU General Public License is included in this distribution in the
-  file called LICENSE.
-  
+  this program; if not, write to the Free Software Foundation, Inc.,
+  51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+
+  The full GNU General Public License is included in this distribution in
+  the file called "COPYING".
+
   Contact Information:
   Linux NICS <linux.nics@intel.com>
   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
@@ -28,6 +27,7 @@
 *******************************************************************************/
 
 #include "e1000.h"
+#include <net/ip6_checksum.h>
 
 char e1000_driver_name[] = "e1000";
 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
@@ -36,7 +36,7 @@ static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
 #else
 #define DRIVERNAPI "-NAPI"
 #endif
-#define DRV_VERSION "7.2.7-k2"DRIVERNAPI
+#define DRV_VERSION "7.3.15-k2"DRIVERNAPI
 char e1000_driver_version[] = DRV_VERSION;
 static char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
 
@@ -104,22 +104,33 @@ static struct pci_device_id e1000_pci_tbl[] = {
        INTEL_E1000_ETHERNET_DEVICE(0x10B9),
        INTEL_E1000_ETHERNET_DEVICE(0x10BA),
        INTEL_E1000_ETHERNET_DEVICE(0x10BB),
+       INTEL_E1000_ETHERNET_DEVICE(0x10BC),
+       INTEL_E1000_ETHERNET_DEVICE(0x10C4),
+       INTEL_E1000_ETHERNET_DEVICE(0x10C5),
        /* required last entry */
        {0,}
 };
 
 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
 
+int e1000_up(struct e1000_adapter *adapter);
+void e1000_down(struct e1000_adapter *adapter);
+void e1000_reinit_locked(struct e1000_adapter *adapter);
+void e1000_reset(struct e1000_adapter *adapter);
+int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
+int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
+int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
+void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
+void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
-                                    struct e1000_tx_ring *txdr);
+                             struct e1000_tx_ring *txdr);
 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
-                                    struct e1000_rx_ring *rxdr);
+                             struct e1000_rx_ring *rxdr);
 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
-                                    struct e1000_tx_ring *tx_ring);
+                             struct e1000_tx_ring *tx_ring);
 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
-                                    struct e1000_rx_ring *rx_ring);
-
-/* Local Function Prototypes */
+                             struct e1000_rx_ring *rx_ring);
+void e1000_update_stats(struct e1000_adapter *adapter);
 
 static int e1000_init_module(void);
 static void e1000_exit_module(void);
@@ -146,7 +157,10 @@ static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
 static int e1000_set_mac(struct net_device *netdev, void *p);
-static irqreturn_t e1000_intr(int irq, void *data, struct pt_regs *regs);
+static irqreturn_t e1000_intr(int irq, void *data);
+#ifdef CONFIG_PCI_MSI
+static irqreturn_t e1000_intr_msi(int irq, void *data);
+#endif
 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
                                     struct e1000_tx_ring *tx_ring);
 #ifdef CONFIG_E1000_NAPI
@@ -172,10 +186,11 @@ static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 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);
+void e1000_set_ethtool_ops(struct net_device *netdev);
 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
 static void e1000_tx_timeout(struct net_device *dev);
-static void e1000_reset_task(struct net_device *dev);
+static void e1000_reset_task(struct work_struct *work);
 static void e1000_smartspeed(struct e1000_adapter *adapter);
 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
                                        struct sk_buff *skb);
@@ -196,6 +211,8 @@ static void e1000_shutdown(struct pci_dev *pdev);
 static void e1000_netpoll (struct net_device *netdev);
 #endif
 
+extern void e1000_check_options(struct e1000_adapter *adapter);
+
 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
                      pci_channel_state_t state);
 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
@@ -212,9 +229,9 @@ static struct pci_driver e1000_driver = {
        .id_table = e1000_pci_tbl,
        .probe    = e1000_probe,
        .remove   = __devexit_p(e1000_remove),
+#ifdef CONFIG_PM
        /* Power Managment Hooks */
        .suspend  = e1000_suspend,
-#ifdef CONFIG_PM
        .resume   = e1000_resume,
 #endif
        .shutdown = e1000_shutdown,
@@ -275,7 +292,7 @@ static int e1000_request_irq(struct e1000_adapter *adapter)
 
        flags = IRQF_SHARED;
 #ifdef CONFIG_PCI_MSI
-       if (adapter->hw.mac_type > e1000_82547_rev_2) {
+       if (adapter->hw.mac_type >= e1000_82571) {
                adapter->have_msi = TRUE;
                if ((err = pci_enable_msi(adapter->pdev))) {
                        DPRINTK(PROBE, ERR,
@@ -283,8 +300,14 @@ static int e1000_request_irq(struct e1000_adapter *adapter)
                        adapter->have_msi = FALSE;
                }
        }
-       if (adapter->have_msi)
+       if (adapter->have_msi) {
                flags &= ~IRQF_SHARED;
+               err = request_irq(adapter->pdev->irq, &e1000_intr_msi, flags,
+                                 netdev->name, netdev);
+               if (err)
+                       DPRINTK(PROBE, ERR,
+                              "Unable to allocate interrupt Error: %d\n", err);
+       } else
 #endif
        if ((err = request_irq(adapter->pdev->irq, &e1000_intr, flags,
                               netdev->name, netdev)))
@@ -365,7 +388,7 @@ e1000_update_mng_vlan(struct e1000_adapter *adapter)
  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
  * For ASF and Pass Through versions of f/w this means that the
  * driver is no longer loaded. For AMT version (only with 82573) i
- * of the f/w this means that the netowrk i/f is closed.
+ * of the f/w this means that the network i/f is closed.
  *
  **/
 
@@ -406,7 +429,7 @@ e1000_release_hw_control(struct e1000_adapter *adapter)
  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
  * For ASF and Pass Through versions of f/w this means that
  * the driver is loaded. For AMT version (only with 82573)
- * of the f/w this means that the netowrk i/f is open.
+ * of the f/w this means that the network i/f is open.
  *
  **/
 
@@ -416,6 +439,7 @@ e1000_get_hw_control(struct e1000_adapter *adapter)
        uint32_t ctrl_ext;
        uint32_t swsm;
        uint32_t extcnf;
+
        /* Let firmware know the driver has taken over */
        switch (adapter->hw.mac_type) {
        case e1000_82571:
@@ -466,13 +490,14 @@ e1000_up(struct e1000_adapter *adapter)
 
        adapter->tx_queue_len = netdev->tx_queue_len;
 
-       mod_timer(&adapter->watchdog_timer, jiffies);
-
 #ifdef CONFIG_E1000_NAPI
        netif_poll_enable(netdev);
 #endif
        e1000_irq_enable(adapter);
 
+       clear_bit(__E1000_DOWN, &adapter->flags);
+
+       mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
        return 0;
 }
 
@@ -502,25 +527,48 @@ void e1000_power_up_phy(struct e1000_adapter *adapter)
 
 static void e1000_power_down_phy(struct e1000_adapter *adapter)
 {
-       boolean_t mng_mode_enabled = (adapter->hw.mac_type >= e1000_82571) &&
-                                     e1000_check_mng_mode(&adapter->hw);
-       /* Power down the PHY so no link is implied when interface is down
-        * The PHY cannot be powered down if any of the following is TRUE
+       /* Power down the PHY so no link is implied when interface is down *
+        * The PHY cannot be powered down if any of the following is TRUE *
         * (a) WoL is enabled
         * (b) AMT is active
         * (c) SoL/IDER session is active */
        if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
-           adapter->hw.mac_type != e1000_ich8lan &&
-           adapter->hw.media_type == e1000_media_type_copper &&
-           !(E1000_READ_REG(&adapter->hw, MANC) & E1000_MANC_SMBUS_EN) &&
-           !mng_mode_enabled &&
-           !e1000_check_phy_reset_block(&adapter->hw)) {
+          adapter->hw.media_type == e1000_media_type_copper) {
                uint16_t mii_reg = 0;
+
+               switch (adapter->hw.mac_type) {
+               case e1000_82540:
+               case e1000_82545:
+               case e1000_82545_rev_3:
+               case e1000_82546:
+               case e1000_82546_rev_3:
+               case e1000_82541:
+               case e1000_82541_rev_2:
+               case e1000_82547:
+               case e1000_82547_rev_2:
+                       if (E1000_READ_REG(&adapter->hw, MANC) &
+                           E1000_MANC_SMBUS_EN)
+                               goto out;
+                       break;
+               case e1000_82571:
+               case e1000_82572:
+               case e1000_82573:
+               case e1000_80003es2lan:
+               case e1000_ich8lan:
+                       if (e1000_check_mng_mode(&adapter->hw) ||
+                           e1000_check_phy_reset_block(&adapter->hw))
+                               goto out;
+                       break;
+               default:
+                       goto out;
+               }
                e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
                mii_reg |= MII_CR_POWER_DOWN;
                e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
                mdelay(1);
        }
+out:
+       return;
 }
 
 void
@@ -528,6 +576,10 @@ e1000_down(struct e1000_adapter *adapter)
 {
        struct net_device *netdev = adapter->netdev;
 
+       /* signal that we're down so the interrupt handler does not
+        * reschedule our watchdog timer */
+       set_bit(__E1000_DOWN, &adapter->flags);
+
        e1000_irq_disable(adapter);
 
        del_timer_sync(&adapter->tx_fifo_stall_timer);
@@ -629,6 +681,7 @@ e1000_reset(struct e1000_adapter *adapter)
        e1000_reset_hw(&adapter->hw);
        if (adapter->hw.mac_type >= e1000_82544)
                E1000_WRITE_REG(&adapter->hw, WUC, 0);
+
        if (e1000_init_hw(&adapter->hw))
                DPRINTK(PROBE, ERR, "Hardware Error\n");
        e1000_update_mng_vlan(adapter);
@@ -652,9 +705,10 @@ e1000_reset(struct e1000_adapter *adapter)
                                    phy_data);
        }
 
-       if (adapter->hw.mac_type < e1000_ich8lan)
-       /* FIXME: this code is duplicate and wrong for PCI Express */
-       if (adapter->en_mng_pt) {
+       if ((adapter->en_mng_pt) &&
+           (adapter->hw.mac_type >= e1000_82540) &&
+           (adapter->hw.mac_type < e1000_82571) &&
+           (adapter->hw.media_type == e1000_media_type_copper)) {
                manc = E1000_READ_REG(&adapter->hw, MANC);
                manc |= (E1000_MANC_ARP_EN | E1000_MANC_EN_MNG2HOST);
                E1000_WRITE_REG(&adapter->hw, MANC, manc);
@@ -760,7 +814,7 @@ e1000_probe(struct pci_dev *pdev,
 #ifdef CONFIG_NET_POLL_CONTROLLER
        netdev->poll_controller = e1000_netpoll;
 #endif
-       strcpy(netdev->name, pci_name(pdev));
+       strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 
        netdev->mem_start = mmio_start;
        netdev->mem_end = mmio_start + mmio_len;
@@ -803,9 +857,9 @@ e1000_probe(struct pci_dev *pdev,
           (adapter->hw.mac_type != e1000_82547))
                netdev->features |= NETIF_F_TSO;
 
-#ifdef NETIF_F_TSO_IPV6
+#ifdef NETIF_F_TSO6
        if (adapter->hw.mac_type > e1000_82547_rev_2)
-               netdev->features |= NETIF_F_TSO_IPV6;
+               netdev->features |= NETIF_F_TSO6;
 #endif
 #endif
        if (pci_using_dac)
@@ -860,13 +914,7 @@ e1000_probe(struct pci_dev *pdev,
        adapter->phy_info_timer.function = &e1000_update_phy_info;
        adapter->phy_info_timer.data = (unsigned long) adapter;
 
-       INIT_WORK(&adapter->reset_task,
-               (void (*)(void *))e1000_reset_task, netdev);
-
-       /* we're going to reset, so assume we have no link for now */
-
-       netif_carrier_off(netdev);
-       netif_stop_queue(netdev);
+       INIT_WORK(&adapter->reset_task, e1000_reset_task);
 
        e1000_check_options(adapter);
 
@@ -925,6 +973,7 @@ e1000_probe(struct pci_dev *pdev,
                break;
        case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
        case E1000_DEV_ID_82571EB_QUAD_COPPER:
+       case E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE:
                /* if quad port adapter, disable WoL on all but port A */
                if (global_quad_port_a != 0)
                        adapter->eeprom_wol = 0;
@@ -974,6 +1023,10 @@ e1000_probe(struct pci_dev *pdev,
        if ((err = register_netdev(netdev)))
                goto err_register;
 
+       /* tell the stack to leave us alone until e1000_open() is called */
+       netif_carrier_off(netdev);
+       netif_stop_queue(netdev);
+
        DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
 
        cards_found++;
@@ -1033,8 +1086,8 @@ e1000_remove(struct pci_dev *pdev)
        flush_scheduled_work();
 
        if (adapter->hw.mac_type >= e1000_82540 &&
-          adapter->hw.mac_type != e1000_ich8lan &&
-          adapter->hw.media_type == e1000_media_type_copper) {
+           adapter->hw.mac_type < e1000_82571 &&
+           adapter->hw.media_type == e1000_media_type_copper) {
                manc = E1000_READ_REG(&adapter->hw, MANC);
                if (manc & E1000_MANC_SMBUS_EN) {
                        manc |= E1000_MANC_ARP_EN;
@@ -1161,6 +1214,8 @@ e1000_sw_init(struct e1000_adapter *adapter)
        atomic_set(&adapter->irq_sem, 1);
        spin_lock_init(&adapter->stats_lock);
 
+       set_bit(__E1000_DOWN, &adapter->flags);
+
        return 0;
 }
 
@@ -1226,16 +1281,14 @@ e1000_open(struct net_device *netdev)
        int err;
 
        /* disallow open during test */
-       if (test_bit(__E1000_DRIVER_TESTING, &adapter->flags))
+       if (test_bit(__E1000_TESTING, &adapter->flags))
                return -EBUSY;
 
        /* allocate transmit descriptors */
-
        if ((err = e1000_setup_all_tx_resources(adapter)))
                goto err_setup_tx;
 
        /* allocate receive descriptors */
-
        if ((err = e1000_setup_all_rx_resources(adapter)))
                goto err_setup_rx;
 
@@ -1299,8 +1352,12 @@ e1000_close(struct net_device *netdev)
        e1000_free_all_tx_resources(adapter);
        e1000_free_all_rx_resources(adapter);
 
+       /* kill manageability vlan ID if supported, but not if a vlan with
+        * the same ID is registered on the host OS (let 8021q kill it) */
        if ((adapter->hw.mng_cookie.status &
-                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
+                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
+            !(adapter->vlgrp &&
+                         adapter->vlgrp->vlan_devices[adapter->mng_vlan_id])) {
                e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
        }
 
@@ -1510,27 +1567,16 @@ e1000_configure_tx(struct e1000_adapter *adapter)
        /* Program the Transmit Control Register */
 
        tctl = E1000_READ_REG(hw, TCTL);
-
        tctl &= ~E1000_TCTL_CT;
        tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
                (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
 
-#ifdef DISABLE_MULR
-       /* disable Multiple Reads for debugging */
-       tctl &= ~E1000_TCTL_MULR;
-#endif
-
        if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
                tarc = E1000_READ_REG(hw, TARC0);
-               tarc |= ((1 << 25) | (1 << 21));
+               /* set the speed mode bit, we'll clear it if we're not at
+                * gigabit link later */
+               tarc |= (1 << 21);
                E1000_WRITE_REG(hw, TARC0, tarc);
-               tarc = E1000_READ_REG(hw, TARC1);
-               tarc |= (1 << 25);
-               if (tctl & E1000_TCTL_MULR)
-                       tarc &= ~(1 << 28);
-               else
-                       tarc |= (1 << 28);
-               E1000_WRITE_REG(hw, TARC1, tarc);
        } else if (hw->mac_type == e1000_80003es2lan) {
                tarc = E1000_READ_REG(hw, TARC0);
                tarc |= 1;
@@ -1543,8 +1589,11 @@ e1000_configure_tx(struct e1000_adapter *adapter)
        e1000_config_collision_dist(hw);
 
        /* Setup Transmit Descriptor Settings for eop descriptor */
-       adapter->txd_cmd = E1000_TXD_CMD_IDE | E1000_TXD_CMD_EOP |
-               E1000_TXD_CMD_IFCS;
+       adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
+
+       /* only set IDE if we are delaying interrupts using the timers */
+       if (adapter->tx_int_delay)
+               adapter->txd_cmd |= E1000_TXD_CMD_IDE;
 
        if (hw->mac_type < e1000_82543)
                adapter->txd_cmd |= E1000_TXD_CMD_RPS;
@@ -1768,9 +1817,11 @@ e1000_setup_rctl(struct e1000_adapter *adapter)
         * followed by the page buffers.  Therefore, skb->data is
         * sized to hold the largest protocol header.
         */
+       /* allocations using alloc_page take too long for regular MTU
+        * so only enable packet split for jumbo frames */
        pages = PAGE_USE_COUNT(adapter->netdev->mtu);
-       if ((adapter->hw.mac_type > e1000_82547_rev_2) && (pages <= 3) &&
-           PAGE_SIZE <= 16384)
+       if ((adapter->hw.mac_type >= e1000_82571) && (pages <= 3) &&
+           PAGE_SIZE <= 16384 && (rctl & E1000_RCTL_LPE))
                adapter->rx_ps_pages = pages;
        else
                adapter->rx_ps_pages = 0;
@@ -1779,8 +1830,11 @@ e1000_setup_rctl(struct e1000_adapter *adapter)
                /* Configure extra packet-split registers */
                rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
                rfctl |= E1000_RFCTL_EXTEN;
-               /* disable IPv6 packet split support */
-               rfctl |= E1000_RFCTL_IPV6_DIS;
+               /* disable packet split support for IPv6 extension headers,
+                * because some malformed IPv6 headers can hang the RX */
+               rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
+                         E1000_RFCTL_NEW_IPV6_EXT_DIS);
+
                E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
 
                rctl |= E1000_RCTL_DTYP_PS;
@@ -1843,7 +1897,7 @@ e1000_configure_rx(struct e1000_adapter *adapter)
 
        if (hw->mac_type >= e1000_82540) {
                E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
-               if (adapter->itr > 1)
+               if (adapter->itr_setting != 0)
                        E1000_WRITE_REG(hw, ITR,
                                1000000000 / (adapter->itr * 256));
        }
@@ -1853,11 +1907,11 @@ e1000_configure_rx(struct e1000_adapter *adapter)
                /* Reset delay timers after every interrupt */
                ctrl_ext |= E1000_CTRL_EXT_INT_TIMER_CLR;
 #ifdef CONFIG_E1000_NAPI
-               /* Auto-Mask interrupts upon ICR read. */
+               /* Auto-Mask interrupts upon ICR access */
                ctrl_ext |= E1000_CTRL_EXT_IAME;
+               E1000_WRITE_REG(hw, IAM, 0xffffffff);
 #endif
                E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
-               E1000_WRITE_REG(hw, IAM, ~0);
                E1000_WRITE_FLUSH(hw);
        }
 
@@ -1896,6 +1950,12 @@ e1000_configure_rx(struct e1000_adapter *adapter)
                E1000_WRITE_REG(hw, RXCSUM, rxcsum);
        }
 
+       /* enable early receives on 82573, only takes effect if using > 2048
+        * byte total frame size.  for example only for jumbo frames */
+#define E1000_ERT_2048 0x100
+       if (hw->mac_type == e1000_82573)
+               E1000_WRITE_REG(hw, ERT, E1000_ERT_2048);
+
        /* Enable Receives */
        E1000_WRITE_REG(hw, RCTL, rctl);
 }
@@ -1949,10 +2009,13 @@ e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
                                buffer_info->dma,
                                buffer_info->length,
                                PCI_DMA_TODEVICE);
+               buffer_info->dma = 0;
        }
-       if (buffer_info->skb)
+       if (buffer_info->skb) {
                dev_kfree_skb_any(buffer_info->skb);
-       memset(buffer_info, 0, sizeof(struct e1000_buffer));
+               buffer_info->skb = NULL;
+       }
+       /* buffer_info must be completely set up in the transmit path */
 }
 
 /**
@@ -2376,6 +2439,7 @@ e1000_watchdog(unsigned long data)
                DPRINTK(LINK, INFO,
                        "Gigabit has been disabled, downgrading speed\n");
        }
+
        if (adapter->hw.mac_type == e1000_82573) {
                e1000_enable_tx_pkt_filtering(&adapter->hw);
                if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
@@ -2420,13 +2484,12 @@ e1000_watchdog(unsigned long data)
                        if ((adapter->hw.mac_type == e1000_82571 ||
                             adapter->hw.mac_type == e1000_82572) &&
                            txb2b == 0) {
-#define SPEED_MODE_BIT (1 << 21)
                                uint32_t tarc0;
                                tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
-                               tarc0 &= ~SPEED_MODE_BIT;
+                               tarc0 &= ~(1 << 21);
                                E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
                        }
-                               
+
 #ifdef NETIF_F_TSO
                        /* disable TSO for pcie and 10/100 speeds, to avoid
                         * some hardware issues */
@@ -2438,9 +2501,15 @@ e1000_watchdog(unsigned long data)
                                        DPRINTK(PROBE,INFO,
                                        "10/100 speed: disabling TSO\n");
                                        netdev->features &= ~NETIF_F_TSO;
+#ifdef NETIF_F_TSO6
+                                       netdev->features &= ~NETIF_F_TSO6;
+#endif
                                        break;
                                case SPEED_1000:
                                        netdev->features |= NETIF_F_TSO;
+#ifdef NETIF_F_TSO6
+                                       netdev->features |= NETIF_F_TSO6;
+#endif
                                        break;
                                default:
                                        /* oops */
@@ -2507,19 +2576,6 @@ e1000_watchdog(unsigned long data)
                }
        }
 
-       /* Dynamic mode for Interrupt Throttle Rate (ITR) */
-       if (adapter->hw.mac_type >= e1000_82540 && adapter->itr == 1) {
-               /* Symmetric Tx/Rx gets a reduced ITR=2000; Total
-                * asymmetrical Tx or Rx gets ITR=8000; everyone
-                * else is between 2000-8000. */
-               uint32_t goc = (adapter->gotcl + adapter->gorcl) / 10000;
-               uint32_t dif = (adapter->gotcl > adapter->gorcl ?
-                       adapter->gotcl - adapter->gorcl :
-                       adapter->gorcl - adapter->gotcl) / 10000;
-               uint32_t itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
-               E1000_WRITE_REG(&adapter->hw, ITR, 1000000000 / (itr * 256));
-       }
-
        /* Cause software interrupt to ensure rx ring is cleaned */
        E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
 
@@ -2535,6 +2591,143 @@ e1000_watchdog(unsigned long data)
        mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
 }
 
+enum latency_range {
+       lowest_latency = 0,
+       low_latency = 1,
+       bulk_latency = 2,
+       latency_invalid = 255
+};
+
+/**
+ * e1000_update_itr - update the dynamic ITR value based on statistics
+ *      Stores a new ITR value based on packets and byte
+ *      counts during the last interrupt.  The advantage of per interrupt
+ *      computation is faster updates and more accurate ITR for the current
+ *      traffic pattern.  Constants in this function were computed
+ *      based on theoretical maximum wire speed and thresholds were set based
+ *      on testing data as well as attempting to minimize response time
+ *      while increasing bulk throughput.
+ *      this functionality is controlled by the InterruptThrottleRate module
+ *      parameter (see e1000_param.c)
+ * @adapter: pointer to adapter
+ * @itr_setting: current adapter->itr
+ * @packets: the number of packets during this measurement interval
+ * @bytes: the number of bytes during this measurement interval
+ **/
+static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
+                                   uint16_t itr_setting,
+                                   int packets,
+                                   int bytes)
+{
+       unsigned int retval = itr_setting;
+       struct e1000_hw *hw = &adapter->hw;
+
+       if (unlikely(hw->mac_type < e1000_82540))
+               goto update_itr_done;
+
+       if (packets == 0)
+               goto update_itr_done;
+
+       switch (itr_setting) {
+       case lowest_latency:
+               /* jumbo frames get bulk treatment*/
+               if (bytes/packets > 8000)
+                       retval = bulk_latency;
+               else if ((packets < 5) && (bytes > 512))
+                       retval = low_latency;
+               break;
+       case low_latency:  /* 50 usec aka 20000 ints/s */
+               if (bytes > 10000) {
+                       /* jumbo frames need bulk latency setting */
+                       if (bytes/packets > 8000)
+                               retval = bulk_latency;
+                       else if ((packets < 10) || ((bytes/packets) > 1200))
+                               retval = bulk_latency;
+                       else if ((packets > 35))
+                               retval = lowest_latency;
+               } else if (bytes/packets > 2000)
+                       retval = bulk_latency;
+               else if (packets <= 2 && bytes < 512)
+                       retval = lowest_latency;
+               break;
+       case bulk_latency: /* 250 usec aka 4000 ints/s */
+               if (bytes > 25000) {
+                       if (packets > 35)
+                               retval = low_latency;
+               } else if (bytes < 6000) {
+                       retval = low_latency;
+               }
+               break;
+       }
+
+update_itr_done:
+       return retval;
+}
+
+static void e1000_set_itr(struct e1000_adapter *adapter)
+{
+       struct e1000_hw *hw = &adapter->hw;
+       uint16_t current_itr;
+       uint32_t new_itr = adapter->itr;
+
+       if (unlikely(hw->mac_type < e1000_82540))
+               return;
+
+       /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
+       if (unlikely(adapter->link_speed != SPEED_1000)) {
+               current_itr = 0;
+               new_itr = 4000;
+               goto set_itr_now;
+       }
+
+       adapter->tx_itr = e1000_update_itr(adapter,
+                                   adapter->tx_itr,
+                                   adapter->total_tx_packets,
+                                   adapter->total_tx_bytes);
+       /* conservative mode (itr 3) eliminates the lowest_latency setting */
+       if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
+               adapter->tx_itr = low_latency;
+
+       adapter->rx_itr = e1000_update_itr(adapter,
+                                   adapter->rx_itr,
+                                   adapter->total_rx_packets,
+                                   adapter->total_rx_bytes);
+       /* conservative mode (itr 3) eliminates the lowest_latency setting */
+       if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
+               adapter->rx_itr = low_latency;
+
+       current_itr = max(adapter->rx_itr, adapter->tx_itr);
+
+       switch (current_itr) {
+       /* counts and packets in update_itr are dependent on these numbers */
+       case lowest_latency:
+               new_itr = 70000;
+               break;
+       case low_latency:
+               new_itr = 20000; /* aka hwitr = ~200 */
+               break;
+       case bulk_latency:
+               new_itr = 4000;
+               break;
+       default:
+               break;
+       }
+
+set_itr_now:
+       if (new_itr != adapter->itr) {
+               /* this attempts to bias the interrupt rate towards Bulk
+                * by adding intermediate steps when interrupt rate is
+                * increasing */
+               new_itr = new_itr > adapter->itr ?
+                            min(adapter->itr + (new_itr >> 2), new_itr) :
+                            new_itr;
+               adapter->itr = new_itr;
+               E1000_WRITE_REG(hw, ITR, 1000000000 / (new_itr * 256));
+       }
+
+       return;
+}
+
 #define E1000_TX_FLAGS_CSUM            0x00000001
 #define E1000_TX_FLAGS_VLAN            0x00000002
 #define E1000_TX_FLAGS_TSO             0x00000004
@@ -2575,7 +2768,7 @@ e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
                                                   0);
                        cmd_length = E1000_TXD_CMD_IP;
                        ipcse = skb->h.raw - skb->data - 1;
-#ifdef NETIF_F_TSO_IPV6
+#ifdef NETIF_F_TSO6
                } else if (skb->protocol == htons(ETH_P_IPV6)) {
                        skb->nh.ipv6h->payload_len = 0;
                        skb->h.th->check =
@@ -2611,6 +2804,7 @@ e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
                context_desc->cmd_and_length = cpu_to_le32(cmd_length);
 
                buffer_info->time_stamp = jiffies;
+               buffer_info->next_to_watch = i;
 
                if (++i == tx_ring->count) i = 0;
                tx_ring->next_to_use = i;
@@ -2639,12 +2833,13 @@ e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
                context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
 
                context_desc->upper_setup.tcp_fields.tucss = css;
-               context_desc->upper_setup.tcp_fields.tucso = css + skb->csum;
+               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);
 
                buffer_info->time_stamp = jiffies;
+               buffer_info->next_to_watch = i;
 
                if (unlikely(++i == tx_ring->count)) i = 0;
                tx_ring->next_to_use = i;
@@ -2713,6 +2908,7 @@ e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
                                size,
                                PCI_DMA_TODEVICE);
                buffer_info->time_stamp = jiffies;
+               buffer_info->next_to_watch = i;
 
                len -= size;
                offset += size;
@@ -2752,6 +2948,7 @@ e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
                                        size,
                                        PCI_DMA_TODEVICE);
                        buffer_info->time_stamp = jiffies;
+                       buffer_info->next_to_watch = i;
 
                        len -= size;
                        offset += size;
@@ -2817,6 +3014,9 @@ e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
 
        tx_ring->next_to_use = i;
        writel(i, adapter->hw.hw_addr + tx_ring->tdt);
+       /* we need this if more than one processor can write to our tail
+        * at a time, it syncronizes IO on IA64/Altix systems */
+       mmiowb();
 }
 
 /**
@@ -2892,6 +3092,36 @@ e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
        return 0;
 }
 
+static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
+{
+       struct e1000_adapter *adapter = netdev_priv(netdev);
+       struct e1000_tx_ring *tx_ring = adapter->tx_ring;
+
+       netif_stop_queue(netdev);
+       /* Herbert's original patch had:
+        *  smp_mb__after_netif_stop_queue();
+        * but since that doesn't exist yet, just open code it. */
+       smp_mb();
+
+       /* We need to check again in a case another CPU has just
+        * made room available. */
+       if (likely(E1000_DESC_UNUSED(tx_ring) < size))
+               return -EBUSY;
+
+       /* A reprieve! */
+       netif_start_queue(netdev);
+       ++adapter->restart_queue;
+       return 0;
+}
+
+static int e1000_maybe_stop_tx(struct net_device *netdev,
+                               struct e1000_tx_ring *tx_ring, int size)
+{
+       if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
+               return 0;
+       return __e1000_maybe_stop_tx(netdev, size);
+}
+
 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
 static int
 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
@@ -2910,6 +3140,10 @@ e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        unsigned int f;
        len -= skb->data_len;
 
+       /* This goes back to the question of how to logically map a tx queue
+        * to a flow.  Right now, performance is impacted slightly negatively
+        * if using multiple tx queues.  If the stack breaks away from a
+        * single qdisc implementation, we can look at this again. */
        tx_ring = adapter->tx_ring;
 
        if (unlikely(skb->len <= 0)) {
@@ -2917,6 +3151,11 @@ e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
                return NETDEV_TX_OK;
        }
 
+       /* 82571 and newer doesn't need the workaround that limited descriptor
+        * length to 4kB */
+       if (adapter->hw.mac_type >= e1000_82571)
+               max_per_txd = 8192;
+
 #ifdef NETIF_F_TSO
        mss = skb_shinfo(skb)->gso_size;
        /* The controller does a simple calculation to
@@ -2930,9 +3169,9 @@ e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
                max_per_txd = min(mss << 2, max_per_txd);
                max_txd_pwr = fls(max_per_txd) - 1;
 
-       /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
-        * points to just header, pull a few bytes of payload from
-        * frags into skb->data */
+               /* TSO Workaround for 82571/2/3 Controllers -- if skb->data
+               * points to just header, pull a few bytes of payload from
+               * frags into skb->data */
                hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
                if (skb->data_len && (hdr_len == (skb->len - skb->data_len))) {
                        switch (adapter->hw.mac_type) {
@@ -3005,8 +3244,7 @@ e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 
        /* need: count + 2 desc gap to keep tail from touching
         * head, otherwise try next time */
-       if (unlikely(E1000_DESC_UNUSED(tx_ring) < count + 2)) {
-               netif_stop_queue(netdev);
+       if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2))) {
                spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
                return NETDEV_TX_BUSY;
        }
@@ -3014,7 +3252,7 @@ e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        if (unlikely(adapter->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);
+                       mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
                        spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
                        return NETDEV_TX_BUSY;
                }
@@ -3053,8 +3291,7 @@ e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
        netdev->trans_start = jiffies;
 
        /* Make sure there is space in the ring for the next send. */
-       if (unlikely(E1000_DESC_UNUSED(tx_ring) < MAX_SKB_FRAGS + 2))
-               netif_stop_queue(netdev);
+       e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
 
        spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
        return NETDEV_TX_OK;
@@ -3076,9 +3313,10 @@ e1000_tx_timeout(struct net_device *netdev)
 }
 
 static void
-e1000_reset_task(struct net_device *netdev)
+e1000_reset_task(struct work_struct *work)
 {
-       struct e1000_adapter *adapter = netdev_priv(netdev);
+       struct e1000_adapter *adapter =
+               container_of(work, struct e1000_adapter, reset_task);
 
        e1000_reinit_locked(adapter);
 }
@@ -3131,11 +3369,13 @@ e1000_change_mtu(struct net_device *netdev, int new_mtu)
                }
                break;
        case e1000_82573:
-               /* only enable jumbo frames if ASPM is disabled completely
-                * this means both bits must be zero in 0x1A bits 3:2 */
+               /* Jumbo Frames not supported if:
+                * - this is not an 82573L device
+                * - ASPM is enabled in any way (0x1A bits 3:2) */
                e1000_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3, 1,
                                  &eeprom_data);
-               if (eeprom_data & EEPROM_WORD1A_ASPM_MASK) {
+               if ((adapter->hw.device_id != E1000_DEV_ID_82573L) ||
+                   (eeprom_data & EEPROM_WORD1A_ASPM_MASK)) {
                        if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
                                DPRINTK(PROBE, ERR,
                                        "Jumbo Frames not supported.\n");
@@ -3143,6 +3383,8 @@ e1000_change_mtu(struct net_device *netdev, int new_mtu)
                        }
                        break;
                }
+               /* ERT will be enabled later to enable wire speed receives */
+
                /* fall through to get support */
        case e1000_82571:
        case e1000_82572:
@@ -3234,12 +3476,12 @@ e1000_update_stats(struct e1000_adapter *adapter)
        adapter->stats.roc += E1000_READ_REG(hw, ROC);
 
        if (adapter->hw.mac_type != e1000_ich8lan) {
-       adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
-       adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
-       adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
-       adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
-       adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
-       adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
+               adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
+               adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
+               adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
+               adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
+               adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
+               adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
        }
 
        adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
@@ -3270,12 +3512,12 @@ e1000_update_stats(struct e1000_adapter *adapter)
        adapter->stats.tpr += E1000_READ_REG(hw, TPR);
 
        if (adapter->hw.mac_type != e1000_ich8lan) {
-       adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
-       adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
-       adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
-       adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
-       adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
-       adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
+               adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
+               adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
+               adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
+               adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
+               adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
+               adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
        }
 
        adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
@@ -3301,18 +3543,17 @@ e1000_update_stats(struct e1000_adapter *adapter)
                adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
 
                if (adapter->hw.mac_type != e1000_ich8lan) {
-               adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
-               adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
-               adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
-               adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
-               adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
-               adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
-               adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
+                       adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
+                       adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
+                       adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
+                       adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
+                       adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
+                       adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
+                       adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
                }
        }
 
        /* Fill out the OS statistics structure */
-
        adapter->net_stats.rx_packets = adapter->stats.gprc;
        adapter->net_stats.tx_packets = adapter->stats.gptc;
        adapter->net_stats.rx_bytes = adapter->stats.gorcl;
@@ -3328,24 +3569,27 @@ e1000_update_stats(struct e1000_adapter *adapter)
                adapter->stats.crcerrs + adapter->stats.algnerrc +
                adapter->stats.ruc + adapter->stats.roc +
                adapter->stats.cexterr;
-       adapter->net_stats.rx_length_errors = adapter->stats.ruc +
-                                             adapter->stats.roc;
+       adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
+       adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
        adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
        adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
        adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
 
        /* Tx Errors */
-
-       adapter->net_stats.tx_errors = adapter->stats.ecol +
-                                      adapter->stats.latecol;
+       adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
+       adapter->net_stats.tx_errors = adapter->stats.txerrc;
        adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
        adapter->net_stats.tx_window_errors = adapter->stats.latecol;
        adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
+       if (adapter->hw.bad_tx_carr_stats_fd &&
+           adapter->link_duplex == FULL_DUPLEX) {
+               adapter->net_stats.tx_carrier_errors = 0;
+               adapter->stats.tncrs = 0;
+       }
 
        /* Tx Dropped needs to be maintained elsewhere */
 
        /* Phy Stats */
-
        if (hw->media_type == e1000_media_type_copper) {
                if ((adapter->link_speed == SPEED_1000) &&
                   (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
@@ -3361,16 +3605,104 @@ e1000_update_stats(struct e1000_adapter *adapter)
 
        spin_unlock_irqrestore(&adapter->stats_lock, flags);
 }
+#ifdef CONFIG_PCI_MSI
+
+/**
+ * e1000_intr_msi - Interrupt Handler
+ * @irq: interrupt number
+ * @data: pointer to a network interface device structure
+ **/
+
+static
+irqreturn_t e1000_intr_msi(int irq, void *data)
+{
+       struct net_device *netdev = data;
+       struct e1000_adapter *adapter = netdev_priv(netdev);
+       struct e1000_hw *hw = &adapter->hw;
+#ifndef CONFIG_E1000_NAPI
+       int i;
+#endif
+
+       /* this code avoids the read of ICR but has to get 1000 interrupts
+        * at every link change event before it will notice the change */
+       if (++adapter->detect_link >= 1000) {
+               uint32_t icr = E1000_READ_REG(hw, ICR);
+#ifdef CONFIG_E1000_NAPI
+               /* read ICR disables interrupts using IAM, so keep up with our
+                * enable/disable accounting */
+               atomic_inc(&adapter->irq_sem);
+#endif
+               adapter->detect_link = 0;
+               if ((icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) &&
+                   (icr & E1000_ICR_INT_ASSERTED)) {
+                       hw->get_link_status = 1;
+                       /* 80003ES2LAN workaround--
+                       * For packet buffer work-around on link down event;
+                       * disable receives here in the ISR and
+                       * reset adapter in watchdog
+                       */
+                       if (netif_carrier_ok(netdev) &&
+                           (adapter->hw.mac_type == e1000_80003es2lan)) {
+                               /* disable receives */
+                               uint32_t rctl = E1000_READ_REG(hw, RCTL);
+                               E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
+                       }
+                       /* guard against interrupt when we're going down */
+                       if (!test_bit(__E1000_DOWN, &adapter->flags))
+                               mod_timer(&adapter->watchdog_timer,
+                                         jiffies + 1);
+               }
+       } else {
+               E1000_WRITE_REG(hw, ICR, (0xffffffff & ~(E1000_ICR_RXSEQ |
+                                                        E1000_ICR_LSC)));
+               /* bummer we have to flush here, but things break otherwise as
+                * some event appears to be lost or delayed and throughput
+                * drops.  In almost all tests this flush is un-necessary */
+               E1000_WRITE_FLUSH(hw);
+#ifdef CONFIG_E1000_NAPI
+               /* Interrupt Auto-Mask (IAM)...upon writing ICR, interrupts are
+                * masked.  No need for the IMC write, but it does mean we
+                * should account for it ASAP. */
+               atomic_inc(&adapter->irq_sem);
+#endif
+       }
+
+#ifdef CONFIG_E1000_NAPI
+       if (likely(netif_rx_schedule_prep(netdev))) {
+               adapter->total_tx_bytes = 0;
+               adapter->total_tx_packets = 0;
+               adapter->total_rx_bytes = 0;
+               adapter->total_rx_packets = 0;
+               __netif_rx_schedule(netdev);
+       } else
+               e1000_irq_enable(adapter);
+#else
+       adapter->total_tx_bytes = 0;
+       adapter->total_rx_bytes = 0;
+       adapter->total_tx_packets = 0;
+       adapter->total_rx_packets = 0;
+
+       for (i = 0; i < E1000_MAX_INTR; i++)
+               if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
+                  !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
+                       break;
+
+       if (likely(adapter->itr_setting & 3))
+               e1000_set_itr(adapter);
+#endif
+
+       return IRQ_HANDLED;
+}
+#endif
 
 /**
  * e1000_intr - Interrupt Handler
  * @irq: interrupt number
  * @data: pointer to a network interface device structure
- * @pt_regs: CPU registers structure
  **/
 
 static irqreturn_t
-e1000_intr(int irq, void *data, struct pt_regs *regs)
+e1000_intr(int irq, void *data)
 {
        struct net_device *netdev = data;
        struct e1000_adapter *adapter = netdev_priv(netdev);
@@ -3378,7 +3710,17 @@ e1000_intr(int irq, void *data, struct pt_regs *regs)
        uint32_t rctl, icr = E1000_READ_REG(hw, ICR);
 #ifndef CONFIG_E1000_NAPI
        int i;
-#else
+#endif
+       if (unlikely(!icr))
+               return IRQ_NONE;  /* Not our interrupt */
+
+#ifdef CONFIG_E1000_NAPI
+       /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
+        * not set, then the adapter didn't send an interrupt */
+       if (unlikely(hw->mac_type >= e1000_82571 &&
+                    !(icr & E1000_ICR_INT_ASSERTED)))
+               return IRQ_NONE;
+
        /* Interrupt Auto-Mask...upon reading ICR,
         * interrupts are masked.  No need for the
         * IMC write, but it does mean we should
@@ -3387,14 +3729,6 @@ e1000_intr(int irq, void *data, struct pt_regs *regs)
                atomic_inc(&adapter->irq_sem);
 #endif
 
-       if (unlikely(!icr)) {
-#ifdef CONFIG_E1000_NAPI
-               if (hw->mac_type >= e1000_82571)
-                       e1000_irq_enable(adapter);
-#endif
-               return IRQ_NONE;  /* Not our interrupt */
-       }
-
        if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
                hw->get_link_status = 1;
                /* 80003ES2LAN workaround--
@@ -3408,18 +3742,27 @@ e1000_intr(int irq, void *data, struct pt_regs *regs)
                        rctl = E1000_READ_REG(hw, RCTL);
                        E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
                }
-               mod_timer(&adapter->watchdog_timer, jiffies);
+               /* guard against interrupt when we're going down */
+               if (!test_bit(__E1000_DOWN, &adapter->flags))
+                       mod_timer(&adapter->watchdog_timer, jiffies + 1);
        }
 
 #ifdef CONFIG_E1000_NAPI
        if (unlikely(hw->mac_type < e1000_82571)) {
+               /* disable interrupts, without the synchronize_irq bit */
                atomic_inc(&adapter->irq_sem);
                E1000_WRITE_REG(hw, IMC, ~0);
                E1000_WRITE_FLUSH(hw);
        }
-       if (likely(netif_rx_schedule_prep(netdev)))
+       if (likely(netif_rx_schedule_prep(netdev))) {
+               adapter->total_tx_bytes = 0;
+               adapter->total_tx_packets = 0;
+               adapter->total_rx_bytes = 0;
+               adapter->total_rx_packets = 0;
                __netif_rx_schedule(netdev);
-       else
+       } 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);
 #else
        /* Writing IMC and IMS is needed for 82547.
@@ -3437,16 +3780,23 @@ e1000_intr(int irq, void *data, struct pt_regs *regs)
                E1000_WRITE_REG(hw, IMC, ~0);
        }
 
+       adapter->total_tx_bytes = 0;
+       adapter->total_rx_bytes = 0;
+       adapter->total_tx_packets = 0;
+       adapter->total_rx_packets = 0;
+
        for (i = 0; i < E1000_MAX_INTR; i++)
                if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
                   !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
                        break;
 
+       if (likely(adapter->itr_setting & 3))
+               e1000_set_itr(adapter);
+
        if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
                e1000_irq_enable(adapter);
 
 #endif
-
        return IRQ_HANDLED;
 }
 
@@ -3490,6 +3840,8 @@ e1000_clean(struct net_device *poll_dev, int *budget)
        if ((!tx_cleaned && (work_done == 0)) ||
           !netif_running(poll_dev)) {
 quit_polling:
+               if (likely(adapter->itr_setting & 3))
+                       e1000_set_itr(adapter);
                netif_rx_complete(poll_dev);
                e1000_irq_enable(adapter);
                return 0;
@@ -3516,6 +3868,7 @@ e1000_clean_tx_irq(struct e1000_adapter *adapter,
        unsigned int count = 0;
 #endif
        boolean_t 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;
@@ -3527,13 +3880,19 @@ e1000_clean_tx_irq(struct e1000_adapter *adapter,
                        buffer_info = &tx_ring->buffer_info[i];
                        cleaned = (i == eop);
 
+                       if (cleaned) {
+                               struct sk_buff *skb = buffer_info->skb;
+                               unsigned int segs = skb_shinfo(skb)->gso_segs;
+                               total_tx_packets += segs;
+                               total_tx_packets++;
+                               total_tx_bytes += skb->len;
+                       }
                        e1000_unmap_and_free_tx_resource(adapter, buffer_info);
-                       memset(tx_desc, 0, sizeof(struct e1000_tx_desc));
+                       tx_desc->upper.data = 0;
 
                        if (unlikely(++i == tx_ring->count)) i = 0;
                }
 
-
                eop = tx_ring->buffer_info[i].next_to_watch;
                eop_desc = E1000_TX_DESC(*tx_ring, eop);
 #ifdef CONFIG_E1000_NAPI
@@ -3546,13 +3905,16 @@ e1000_clean_tx_irq(struct e1000_adapter *adapter,
        tx_ring->next_to_clean = i;
 
 #define TX_WAKE_THRESHOLD 32
-       if (unlikely(cleaned && netif_queue_stopped(netdev) &&
-                    netif_carrier_ok(netdev))) {
-               spin_lock(&tx_ring->tx_lock);
-               if (netif_queue_stopped(netdev) &&
-                   (E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))
+       if (unlikely(cleaned && 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.
+                */
+               smp_mb();
+               if (netif_queue_stopped(netdev)) {
                        netif_wake_queue(netdev);
-               spin_unlock(&tx_ring->tx_lock);
+                       ++adapter->restart_queue;
+               }
        }
 
        if (adapter->detect_tx_hung) {
@@ -3590,6 +3952,8 @@ e1000_clean_tx_irq(struct e1000_adapter *adapter,
                        netif_stop_queue(netdev);
                }
        }
+       adapter->total_tx_bytes += total_tx_bytes;
+       adapter->total_tx_packets += total_tx_packets;
        return cleaned;
 }
 
@@ -3669,6 +4033,7 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
        unsigned int i;
        int cleaned_count = 0;
        boolean_t 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);
@@ -3677,6 +4042,7 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
        while (rx_desc->status & E1000_RXD_STAT_DD) {
                struct sk_buff *skb;
                u8 status;
+
 #ifdef CONFIG_E1000_NAPI
                if (*work_done >= work_to_do)
                        break;
@@ -3703,9 +4069,6 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
 
                length = le16_to_cpu(rx_desc->length);
 
-               /* adjust length to remove Ethernet CRC */
-               length -= 4;
-
                if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
                        /* All receives must fit into a single buffer */
                        E1000_DBG("%s: Receive packet consumed multiple"
@@ -3733,6 +4096,14 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                        }
                }
 
+               /* adjust length to remove Ethernet CRC, this must be
+                * done after the TBI_ACCEPT workaround above */
+               length -= 4;
+
+               /* probably a little skewed due to removing CRC */
+               total_rx_bytes += length;
+               total_rx_packets++;
+
                /* code added for copybreak, this should improve
                 * performance for small packets with large amounts
                 * of reassembly being done in the stack */
@@ -3748,12 +4119,11 @@ e1000_clean_rx_irq(struct e1000_adapter *adapter,
                                /* save the skb in buffer_info as good */
                                buffer_info->skb = skb;
                                skb = new_skb;
-                               skb_put(skb, length);
                        }
-               } else
-                       skb_put(skb, length);
-
+                       /* else just continue with the old one */
+               }
                /* end copybreak code */
+               skb_put(skb, length);
 
                /* Receive Checksum Offload */
                e1000_rx_checksum(adapter,
@@ -3802,6 +4172,8 @@ next_desc:
        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;
        return cleaned;
 }
 
@@ -3831,6 +4203,7 @@ e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
        uint32_t length, staterr;
        int cleaned_count = 0;
        boolean_t cleaned = FALSE;
+       unsigned int total_rx_bytes=0, total_rx_packets=0;
 
        i = rx_ring->next_to_clean;
        rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
@@ -3915,7 +4288,7 @@ e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
                        goto copydone;
                } /* if */
                }
-               
+
                for (j = 0; j < adapter->rx_ps_pages; j++) {
                        if (!(length= le16_to_cpu(rx_desc->wb.upper.length[j])))
                                break;
@@ -3935,6 +4308,9 @@ e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
                pskb_trim(skb, skb->len - 4);
 
 copydone:
+               total_rx_bytes += skb->len;
+               total_rx_packets++;
+
                e1000_rx_checksum(adapter, staterr,
                                  le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
                skb->protocol = eth_type_trans(skb, netdev);
@@ -3983,6 +4359,8 @@ next_desc:
        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;
        return cleaned;
 }
 
@@ -4150,7 +4528,7 @@ e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
                }
 
                skb = netdev_alloc_skb(netdev,
-                                      adapter->rx_ps_bsize0 + NET_IP_ALIGN);
+                                      adapter->rx_ps_bsize0 + NET_IP_ALIGN);
 
                if (unlikely(!skb)) {
                        adapter->alloc_rx_buff_failed++;
@@ -4412,13 +4790,20 @@ e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
        pci_write_config_word(adapter->pdev, reg, *value);
 }
 
-#if 0
-uint32_t
-e1000_io_read(struct e1000_hw *hw, unsigned long port)
+int32_t
+e1000_read_pcie_cap_reg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
 {
-       return inl(port);
+    struct e1000_adapter *adapter = hw->back;
+    uint16_t cap_offset;
+
+    cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
+    if (!cap_offset)
+        return -E1000_ERR_CONFIG;
+
+    pci_read_config_word(adapter->pdev, cap_offset + reg, value);
+
+    return E1000_SUCCESS;
 }
-#endif  /*  0  */
 
 void
 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
@@ -4442,12 +4827,12 @@ e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
                E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
 
                if (adapter->hw.mac_type != e1000_ich8lan) {
-               /* enable VLAN receive filtering */
-               rctl = E1000_READ_REG(&adapter->hw, RCTL);
-               rctl |= E1000_RCTL_VFE;
-               rctl &= ~E1000_RCTL_CFIEN;
-               E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
-               e1000_update_mng_vlan(adapter);
+                       /* enable VLAN receive filtering */
+                       rctl = E1000_READ_REG(&adapter->hw, RCTL);
+                       rctl |= E1000_RCTL_VFE;
+                       rctl &= ~E1000_RCTL_CFIEN;
+                       E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
+                       e1000_update_mng_vlan(adapter);
                }
        } else {
                /* disable VLAN tag insert/strip */
@@ -4456,14 +4841,16 @@ e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
                E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
 
                if (adapter->hw.mac_type != e1000_ich8lan) {
-               /* disable VLAN filtering */
-               rctl = E1000_READ_REG(&adapter->hw, RCTL);
-               rctl &= ~E1000_RCTL_VFE;
-               E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
-               if (adapter->mng_vlan_id != (uint16_t)E1000_MNG_VLAN_NONE) {
-                       e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
-                       adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
-               }
+                       /* disable VLAN filtering */
+                       rctl = E1000_READ_REG(&adapter->hw, RCTL);
+                       rctl &= ~E1000_RCTL_VFE;
+                       E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
+                       if (adapter->mng_vlan_id !=
+                           (uint16_t)E1000_MNG_VLAN_NONE) {
+                               e1000_vlan_rx_kill_vid(netdev,
+                                                      adapter->mng_vlan_id);
+                               adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
+                       }
                }
        }
 
@@ -4693,10 +5080,9 @@ e1000_suspend(struct pci_dev *pdev, pm_message_t state)
                pci_enable_wake(pdev, PCI_D3cold, 0);
        }
 
-       /* FIXME: this code is incorrect for PCI Express */
        if (adapter->hw.mac_type >= e1000_82540 &&
-          adapter->hw.mac_type != e1000_ich8lan &&
-          adapter->hw.media_type == e1000_media_type_copper) {
+           adapter->hw.mac_type < e1000_82571 &&
+           adapter->hw.media_type == e1000_media_type_copper) {
                manc = E1000_READ_REG(&adapter->hw, MANC);
                if (manc & E1000_MANC_SMBUS_EN) {
                        manc |= E1000_MANC_ARP_EN;
@@ -4709,6 +5095,9 @@ e1000_suspend(struct pci_dev *pdev, pm_message_t state)
        if (adapter->hw.phy_type == e1000_phy_igp_3)
                e1000_phy_powerdown_workaround(&adapter->hw);
 
+       if (netif_running(netdev))
+               e1000_free_irq(adapter);
+
        /* Release control of h/w to f/w.  If f/w is AMT enabled, this
         * would have already happened in close and is redundant. */
        e1000_release_hw_control(adapter);
@@ -4739,6 +5128,10 @@ e1000_resume(struct pci_dev *pdev)
        pci_enable_wake(pdev, PCI_D3hot, 0);
        pci_enable_wake(pdev, PCI_D3cold, 0);
 
+       if (netif_running(netdev) && (err = e1000_request_irq(adapter)))
+               return err;
+
+       e1000_power_up_phy(adapter);
        e1000_reset(adapter);
        E1000_WRITE_REG(&adapter->hw, WUS, ~0);
 
@@ -4747,10 +5140,9 @@ e1000_resume(struct pci_dev *pdev)
 
        netif_device_attach(netdev);
 
-       /* FIXME: this code is incorrect for PCI Express */
        if (adapter->hw.mac_type >= e1000_82540 &&
-          adapter->hw.mac_type != e1000_ich8lan &&
-          adapter->hw.media_type == e1000_media_type_copper) {
+           adapter->hw.mac_type < e1000_82571 &&
+           adapter->hw.media_type == e1000_media_type_copper) {
                manc = E1000_READ_REG(&adapter->hw, MANC);
                manc &= ~(E1000_MANC_ARP_EN);
                E1000_WRITE_REG(&adapter->hw, MANC, manc);
@@ -4785,7 +5177,7 @@ e1000_netpoll(struct net_device *netdev)
        struct e1000_adapter *adapter = netdev_priv(netdev);
 
        disable_irq(adapter->pdev->irq);
-       e1000_intr(adapter->pdev->irq, netdev, NULL);
+       e1000_intr(adapter->pdev->irq, netdev);
        e1000_clean_tx_irq(adapter, adapter->tx_ring);
 #ifndef CONFIG_E1000_NAPI
        adapter->clean_rx(adapter, adapter->rx_ring);
@@ -4835,12 +5227,8 @@ static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
        }
        pci_set_master(pdev);
 
-       pci_enable_wake(pdev, 3, 0);
-       pci_enable_wake(pdev, 4, 0); /* 4 == D3 cold */
-
-       /* Perform card reset only on one instance of the card */
-       if (PCI_FUNC (pdev->devfn) != 0)
-               return PCI_ERS_RESULT_RECOVERED;
+       pci_enable_wake(pdev, PCI_D3hot, 0);
+       pci_enable_wake(pdev, PCI_D3cold, 0);
 
        e1000_reset(adapter);
        E1000_WRITE_REG(&adapter->hw, WUS, ~0);
@@ -4872,6 +5260,7 @@ static void e1000_io_resume(struct pci_dev *pdev)
        netif_device_attach(netdev);
 
        if (adapter->hw.mac_type >= e1000_82540 &&
+           adapter->hw.mac_type < e1000_82571 &&
            adapter->hw.media_type == e1000_media_type_copper) {
                manc = E1000_READ_REG(&adapter->hw, MANC);
                manc &= ~(E1000_MANC_ARP_EN);