fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / drivers / net / ixgb / ixgb_main.c
index 435cfa1..a083a91 100644 (file)
@@ -1,50 +1,44 @@
 /*******************************************************************************
 
-  
-  Copyright(c) 1999 - 2004 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/10GbE 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>
   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
 
 *******************************************************************************/
 
 #include "ixgb.h"
 
-/* Change Log
- * 1.0.84 10/26/04
- * - reset buffer_info->dma in Tx resource cleanup logic
- * 1.0.83 10/12/04
- * - sparse cleanup - shemminger@osdl.org
- * - fix tx resource cleanup logic
- */
-
 char ixgb_driver_name[] = "ixgb";
-char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
+static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
+
 #ifndef CONFIG_IXGB_NAPI
 #define DRIVERNAPI
 #else
 #define DRIVERNAPI "-NAPI"
 #endif
-char ixgb_driver_version[] = "1.0.87-k2"DRIVERNAPI;
-char ixgb_copyright[] = "Copyright (c) 1999-2004 Intel Corporation.";
+#define DRV_VERSION            "1.0.126-k2"DRIVERNAPI
+char ixgb_driver_version[] = DRV_VERSION;
+static char ixgb_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
 
 /* ixgb_pci_tbl - PCI Device ID Table
  *
@@ -57,6 +51,8 @@ char ixgb_copyright[] = "Copyright (c) 1999-2004 Intel Corporation.";
 static struct pci_device_id ixgb_pci_tbl[] = {
        {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
         PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
+       {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
+        PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
        {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
         PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
        {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,  
@@ -97,8 +93,9 @@ static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
 static int ixgb_set_mac(struct net_device *netdev, void *p);
-static irqreturn_t ixgb_intr(int irq, void *data, struct pt_regs *regs);
+static irqreturn_t ixgb_intr(int irq, void *data);
 static boolean_t ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
+
 #ifdef CONFIG_IXGB_NAPI
 static int ixgb_clean(struct net_device *netdev, int *budget);
 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter,
@@ -109,51 +106,57 @@ static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter);
 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter);
 void ixgb_set_ethtool_ops(struct net_device *netdev);
 static void ixgb_tx_timeout(struct net_device *dev);
-static void ixgb_tx_timeout_task(struct net_device *dev);
+static void ixgb_tx_timeout_task(struct work_struct *work);
 static void ixgb_vlan_rx_register(struct net_device *netdev,
                                  struct vlan_group *grp);
 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
 
-static int ixgb_notify_reboot(struct notifier_block *, unsigned long event,
-                             void *ptr);
-static int ixgb_suspend(struct pci_dev *pdev, uint32_t state);
-
 #ifdef CONFIG_NET_POLL_CONTROLLER
 /* for netdump / net console */
 static void ixgb_netpoll(struct net_device *dev);
 #endif
 
-struct notifier_block ixgb_notifier_reboot = {
-       .notifier_call = ixgb_notify_reboot,
-       .next = NULL,
-       .priority = 0
-};
+static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
+                            enum pci_channel_state state);
+static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
+static void ixgb_io_resume (struct pci_dev *pdev);
 
 /* Exported from other modules */
-
 extern void ixgb_check_options(struct ixgb_adapter *adapter);
 
+static struct pci_error_handlers ixgb_err_handler = {
+       .error_detected = ixgb_io_error_detected,
+       .slot_reset = ixgb_io_slot_reset,
+       .resume = ixgb_io_resume,
+};
+
 static struct pci_driver ixgb_driver = {
-       .name = ixgb_driver_name,
+       .name     = ixgb_driver_name,
        .id_table = ixgb_pci_tbl,
-       .probe = ixgb_probe,
-       .remove = __devexit_p(ixgb_remove),
-       /* Power Managment Hooks */
-       .suspend = NULL,
-       .resume = NULL
+       .probe    = ixgb_probe,
+       .remove   = __devexit_p(ixgb_remove),
+       .err_handler = &ixgb_err_handler
 };
 
 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
 MODULE_LICENSE("GPL");
+MODULE_VERSION(DRV_VERSION);
+
+#define DEFAULT_DEBUG_LEVEL_SHIFT 3
+static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
+module_param(debug, int, 0);
+MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 
 /* some defines for controlling descriptor fetches in h/w */
-#define RXDCTL_PTHRESH_DEFAULT 128     /* chip considers prefech below this */
-#define RXDCTL_HTHRESH_DEFAULT 16      /* chip will only prefetch if tail is 
-                                          pushed this many descriptors from head */
-#define RXDCTL_WTHRESH_DEFAULT 16      /* chip writes back at this many or RXT0 */
+#define RXDCTL_WTHRESH_DEFAULT 15  /* chip writes back at this many or RXT0 */
+#define RXDCTL_PTHRESH_DEFAULT 0   /* chip considers prefech below
+                                    * this */
+#define RXDCTL_HTHRESH_DEFAULT 0   /* chip will only prefetch if tail
+                                    * is pushed this many descriptors
+                                    * from head */
 
 /**
  * ixgb_init_module - Driver Registration Routine
@@ -165,17 +168,12 @@ MODULE_LICENSE("GPL");
 static int __init
 ixgb_init_module(void)
 {
-       int ret;
        printk(KERN_INFO "%s - version %s\n",
               ixgb_driver_string, ixgb_driver_version);
 
        printk(KERN_INFO "%s\n", ixgb_copyright);
 
-       ret = pci_module_init(&ixgb_driver);
-       if(ret >= 0) {
-               register_reboot_notifier(&ixgb_notifier_reboot);
-       }
-       return ret;
+       return pci_register_driver(&ixgb_driver);
 }
 
 module_init(ixgb_init_module);
@@ -190,7 +188,6 @@ module_init(ixgb_init_module);
 static void __exit
 ixgb_exit_module(void)
 {
-       unregister_reboot_notifier(&ixgb_notifier_reboot);
        pci_unregister_driver(&ixgb_driver);
 }
 
@@ -201,7 +198,7 @@ module_exit(ixgb_exit_module);
  * @adapter: board private structure
  **/
 
-static inline void
+static void
 ixgb_irq_disable(struct ixgb_adapter *adapter)
 {
        atomic_inc(&adapter->irq_sem);
@@ -215,13 +212,13 @@ ixgb_irq_disable(struct ixgb_adapter *adapter)
  * @adapter: board private structure
  **/
 
-static inline void
+static void
 ixgb_irq_enable(struct ixgb_adapter *adapter)
 {
        if(atomic_dec_and_test(&adapter->irq_sem)) {
                IXGB_WRITE_REG(&adapter->hw, IMS,
-                          IXGB_INT_RXT0 | IXGB_INT_RXDMT0 | IXGB_INT_TXDW |
-                          IXGB_INT_RXO | IXGB_INT_LSC);
+                              IXGB_INT_RXT0 | IXGB_INT_RXDMT0 | IXGB_INT_TXDW |
+                              IXGB_INT_LSC);
                IXGB_WRITE_FLUSH(&adapter->hw);
        }
 }
@@ -236,6 +233,7 @@ ixgb_up(struct ixgb_adapter *adapter)
 
        /* hardware has been reset, we need to reload some things */
 
+       ixgb_rar_set(hw, netdev->dev_addr, 0);
        ixgb_set_multi(netdev);
 
        ixgb_restore_vlan(adapter);
@@ -245,6 +243,9 @@ ixgb_up(struct ixgb_adapter *adapter)
        ixgb_configure_rx(adapter);
        ixgb_alloc_rx_buffers(adapter);
 
+       /* disable interrupts and get the hardware into a known state */
+       IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
+
 #ifdef CONFIG_PCI_MSI
        {
        boolean_t pcix = (IXGB_READ_REG(&adapter->hw, STATUS) & 
@@ -254,7 +255,7 @@ ixgb_up(struct ixgb_adapter *adapter)
        if (!pcix)
           adapter->have_msi = FALSE;
        else if((err = pci_enable_msi(adapter->pdev))) {
-               printk (KERN_ERR
+               DPRINTK(PROBE, ERR,
                 "Unable to allocate MSI interrupt Error: %d\n", err);
                adapter->have_msi = FALSE;
                /* proceed to try to request regular interrupt */
@@ -263,12 +264,12 @@ ixgb_up(struct ixgb_adapter *adapter)
 
 #endif
        if((err = request_irq(adapter->pdev->irq, &ixgb_intr,
-                                 SA_SHIRQ | SA_SAMPLE_RANDOM,
-                                 netdev->name, netdev)))
+                                 IRQF_SHARED | IRQF_SAMPLE_RANDOM,
+                                 netdev->name, netdev))) {
+               DPRINTK(PROBE, ERR,
+                "Unable to allocate interrupt Error: %d\n", err);
                return err;
-
-       /* disable interrupts and get the hardware into a known state */
-       IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
+       }
 
        if((hw->max_frame_size != max_frame) ||
                (hw->max_frame_size !=
@@ -290,6 +291,10 @@ ixgb_up(struct ixgb_adapter *adapter)
        }
 
        mod_timer(&adapter->watchdog_timer, jiffies);
+
+#ifdef CONFIG_IXGB_NAPI
+       netif_poll_enable(netdev);
+#endif
        ixgb_irq_enable(adapter);
 
        return 0;
@@ -309,6 +314,9 @@ ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog)
 #endif
        if(kill_watchdog)
                del_timer_sync(&adapter->watchdog_timer);
+#ifdef CONFIG_IXGB_NAPI
+       netif_poll_disable(netdev);
+#endif
        adapter->link_speed = 0;
        adapter->link_duplex = 0;
        netif_carrier_off(netdev);
@@ -325,7 +333,7 @@ ixgb_reset(struct ixgb_adapter *adapter)
 
        ixgb_adapter_stop(&adapter->hw);
        if(!ixgb_init_hw(&adapter->hw))
-               IXGB_DBG("ixgb_init_hw failed.\n");
+               DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
 }
 
 /**
@@ -356,18 +364,21 @@ ixgb_probe(struct pci_dev *pdev,
        if((err = pci_enable_device(pdev)))
                return err;
 
-       if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK))) {
+       if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
+          !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
                pci_using_dac = 1;
        } else {
-               if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK))) {
-                       IXGB_ERR("No usable DMA configuration, aborting\n");
-                       return err;
+               if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
+                  (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
+                       printk(KERN_ERR
+                        "ixgb: No usable DMA configuration, aborting\n");
+                       goto err_dma_mask;
                }
                pci_using_dac = 0;
        }
 
        if((err = pci_request_regions(pdev, ixgb_driver_name)))
-               return err;
+               goto err_request_regions;
 
        pci_set_master(pdev);
 
@@ -381,10 +392,11 @@ ixgb_probe(struct pci_dev *pdev,
        SET_NETDEV_DEV(netdev, &pdev->dev);
 
        pci_set_drvdata(pdev, netdev);
-       adapter = netdev->priv;
+       adapter = netdev_priv(netdev);
        adapter->netdev = netdev;
        adapter->pdev = pdev;
        adapter->hw.back = adapter;
+       adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
 
        mmio_start = pci_resource_start(pdev, BAR_0);
        mmio_len = pci_resource_len(pdev, BAR_0);
@@ -413,7 +425,7 @@ ixgb_probe(struct pci_dev *pdev,
        netdev->change_mtu = &ixgb_change_mtu;
        ixgb_set_ethtool_ops(netdev);
        netdev->tx_timeout = &ixgb_tx_timeout;
-       netdev->watchdog_timeo = HZ;
+       netdev->watchdog_timeo = 5 * HZ;
 #ifdef CONFIG_IXGB_NAPI
        netdev->poll = &ixgb_clean;
        netdev->weight = 64;
@@ -425,6 +437,7 @@ ixgb_probe(struct pci_dev *pdev,
        netdev->poll_controller = ixgb_netpoll;
 #endif
 
+       strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
        netdev->mem_start = mmio_start;
        netdev->mem_end = mmio_start + mmio_len;
        netdev->base_addr = adapter->hw.io_base;
@@ -446,6 +459,9 @@ ixgb_probe(struct pci_dev *pdev,
 #ifdef NETIF_F_TSO
        netdev->features |= NETIF_F_TSO;
 #endif
+#ifdef NETIF_F_LLTX
+       netdev->features |= NETIF_F_LLTX;
+#endif
 
        if(pci_using_dac)
                netdev->features |= NETIF_F_HIGHDMA;
@@ -453,14 +469,16 @@ ixgb_probe(struct pci_dev *pdev,
        /* make sure the EEPROM is good */
 
        if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
-               printk(KERN_ERR "The EEPROM Checksum Is Not Valid\n");
+               DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
                err = -EIO;
                goto err_eeprom;
        }
 
        ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
+       memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
 
-       if(!is_valid_ether_addr(netdev->dev_addr)) {
+       if(!is_valid_ether_addr(netdev->perm_addr)) {
+               DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
                err = -EIO;
                goto err_eeprom;
        }
@@ -471,9 +489,9 @@ ixgb_probe(struct pci_dev *pdev,
        adapter->watchdog_timer.function = &ixgb_watchdog;
        adapter->watchdog_timer.data = (unsigned long)adapter;
 
-       INIT_WORK(&adapter->tx_timeout_task,
-                 (void (*)(void *))ixgb_tx_timeout_task, netdev);
+       INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
 
+       strcpy(netdev->name, "eth%d");
        if((err = register_netdev(netdev)))
                goto err_register;
 
@@ -482,8 +500,7 @@ ixgb_probe(struct pci_dev *pdev,
        netif_carrier_off(netdev);
        netif_stop_queue(netdev);
 
-       printk(KERN_INFO "%s: Intel(R) PRO/10GbE Network Connection\n",
-                  netdev->name);
+       DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
        ixgb_check_options(adapter);
        /* reset the hardware with the new settings */
 
@@ -500,6 +517,9 @@ err_ioremap:
        free_netdev(netdev);
 err_alloc_etherdev:
        pci_release_regions(pdev);
+err_request_regions:
+err_dma_mask:
+       pci_disable_device(pdev);
        return err;
 }
 
@@ -517,7 +537,7 @@ static void __devexit
 ixgb_remove(struct pci_dev *pdev)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
 
        unregister_netdev(netdev);
 
@@ -550,17 +570,17 @@ ixgb_sw_init(struct ixgb_adapter *adapter)
        hw->subsystem_vendor_id = pdev->subsystem_vendor;
        hw->subsystem_id = pdev->subsystem_device;
 
-       adapter->rx_buffer_len = IXGB_RXBUFFER_2048;
-
        hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
+       adapter->rx_buffer_len = hw->max_frame_size;
 
        if((hw->device_id == IXGB_DEVICE_ID_82597EX)
-          ||(hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
-          ||(hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
+          || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
+          || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
+          || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
                        hw->mac_type = ixgb_82597;
        else {
                /* should never have loaded on this device */
-               printk(KERN_ERR "ixgb: unsupported device id\n");
+               DPRINTK(PROBE, ERR, "unsupported device id\n");
        }
 
        /* enable flow control to be programmed */
@@ -588,7 +608,7 @@ ixgb_sw_init(struct ixgb_adapter *adapter)
 static int
 ixgb_open(struct net_device *netdev)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        int err;
 
        /* allocate transmit descriptors */
@@ -631,7 +651,7 @@ err_setup_tx:
 static int
 ixgb_close(struct net_device *netdev)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
 
        ixgb_down(adapter, TRUE);
 
@@ -658,6 +678,8 @@ ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
        size = sizeof(struct ixgb_buffer) * txdr->count;
        txdr->buffer_info = vmalloc(size);
        if(!txdr->buffer_info) {
+               DPRINTK(PROBE, ERR,
+                "Unable to allocate transmit descriptor ring memory\n");
                return -ENOMEM;
        }
        memset(txdr->buffer_info, 0, size);
@@ -670,6 +692,8 @@ ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
        txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
        if(!txdr->desc) {
                vfree(txdr->buffer_info);
+               DPRINTK(PROBE, ERR,
+                "Unable to allocate transmit descriptor memory\n");
                return -ENOMEM;
        }
        memset(txdr->desc, 0, txdr->size);
@@ -709,14 +733,8 @@ ixgb_configure_tx(struct ixgb_adapter *adapter)
        IXGB_WRITE_REG(hw, TDH, 0);
        IXGB_WRITE_REG(hw, TDT, 0);
 
-       /* don't set up txdctl, it induces performance problems if
-        * configured incorrectly
-        txdctl  = TXDCTL_PTHRESH_DEFAULT; // prefetch txds below this threshold
-        txdctl |= (TXDCTL_HTHRESH_DEFAULT // only prefetch if there are this many ready
-        << IXGB_TXDCTL_HTHRESH_SHIFT);
-        IXGB_WRITE_REG (hw, TXDCTL, txdctl);
-        */
-
+       /* don't set up txdctl, it induces performance problems if configured
+        * incorrectly */
        /* Set the Tx Interrupt Delay register */
 
        IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
@@ -749,6 +767,8 @@ ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
        size = sizeof(struct ixgb_buffer) * rxdr->count;
        rxdr->buffer_info = vmalloc(size);
        if(!rxdr->buffer_info) {
+               DPRINTK(PROBE, ERR,
+                "Unable to allocate receive descriptor ring\n");
                return -ENOMEM;
        }
        memset(rxdr->buffer_info, 0, size);
@@ -762,6 +782,8 @@ ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
 
        if(!rxdr->desc) {
                vfree(rxdr->buffer_info);
+               DPRINTK(PROBE, ERR,
+                "Unable to allocate receive descriptors\n");
                return -ENOMEM;
        }
        memset(rxdr->desc, 0, rxdr->size);
@@ -793,21 +815,14 @@ ixgb_setup_rctl(struct ixgb_adapter *adapter)
 
        rctl |= IXGB_RCTL_SECRC;
 
-       switch (adapter->rx_buffer_len) {
-       case IXGB_RXBUFFER_2048:
-       default:
+       if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
                rctl |= IXGB_RCTL_BSIZE_2048;
-               break;
-       case IXGB_RXBUFFER_4096:
+       else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
                rctl |= IXGB_RCTL_BSIZE_4096;
-               break;
-       case IXGB_RXBUFFER_8192:
+       else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
                rctl |= IXGB_RCTL_BSIZE_8192;
-               break;
-       case IXGB_RXBUFFER_16384:
+       else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
                rctl |= IXGB_RCTL_BSIZE_16384;
-               break;
-       }
 
        IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
 }
@@ -849,10 +864,17 @@ ixgb_configure_rx(struct ixgb_adapter *adapter)
        IXGB_WRITE_REG(hw, RDH, 0);
        IXGB_WRITE_REG(hw, RDT, 0);
 
-                                               /* burst 16 or burst when RXT0*/
-       rxdctl =  RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT 
-                       | RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT 
-                       | RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
+       /* set up pre-fetching of receive buffers so we get some before we
+        * run out (default hardware behavior is to run out before fetching
+        * more).  This sets up to fetch if HTHRESH rx descriptors are avail
+        * and the descriptors in hw cache are below PTHRESH.  This avoids
+        * the hardware behavior of fetching <=512 descriptors in a single
+        * burst that pre-empts all other activity, usually causing fifo
+        * overflows. */
+       /* use WTHRESH to burst write 16 descriptors or burst when RXT0 */
+       rxdctl = RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT |
+                RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT |
+                RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
        IXGB_WRITE_REG(hw, RXDCTL, rxdctl);
 
        /* Enable Receive Checksum Offload for TCP and UDP */
@@ -890,22 +912,25 @@ ixgb_free_tx_resources(struct ixgb_adapter *adapter)
        adapter->tx_ring.desc = NULL;
 }
 
-static inline void
+static void
 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
                                        struct ixgb_buffer *buffer_info)
 {
        struct pci_dev *pdev = adapter->pdev;
-       if(buffer_info->dma) {
-               pci_unmap_page(pdev,
-                          buffer_info->dma,
-                          buffer_info->length,
-                          PCI_DMA_TODEVICE);
-               buffer_info->dma = 0;
-       }
-       if(buffer_info->skb) {
+
+       if (buffer_info->dma)
+               pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
+                              PCI_DMA_TODEVICE);
+
+       if (buffer_info->skb)
                dev_kfree_skb_any(buffer_info->skb);
-               buffer_info->skb = NULL;
-       }
+
+       buffer_info->skb = NULL;
+       buffer_info->dma = 0;
+       buffer_info->time_stamp = 0;
+       /* these fields must always be initialized in tx
+        * buffer_info->length = 0;
+        * buffer_info->next_to_watch = 0; */
 }
 
 /**
@@ -1021,7 +1046,7 @@ ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
 static int
 ixgb_set_mac(struct net_device *netdev, void *p)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        struct sockaddr *addr = p;
 
        if(!is_valid_ether_addr(addr->sa_data))
@@ -1047,7 +1072,7 @@ ixgb_set_mac(struct net_device *netdev, void *p)
 static void
 ixgb_set_multi(struct net_device *netdev)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        struct ixgb_hw *hw = &adapter->hw;
        struct dev_mc_list *mc_ptr;
        uint32_t rctl;
@@ -1094,7 +1119,6 @@ ixgb_watchdog(unsigned long data)
        struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
        struct net_device *netdev = adapter->netdev;
        struct ixgb_desc_ring *txdr = &adapter->tx_ring;
-       unsigned int i;
 
        ixgb_check_for_link(&adapter->hw);
 
@@ -1105,8 +1129,8 @@ ixgb_watchdog(unsigned long data)
 
        if(adapter->hw.link_up) {
                if(!netif_carrier_ok(netdev)) {
-                       printk(KERN_INFO "ixgb: %s NIC Link is Up %d Mbps %s\n",
-                                  netdev->name, 10000, "Full Duplex");
+                       DPRINTK(LINK, INFO,
+                               "NIC Link is Up 10000 Mbps Full Duplex\n");
                        adapter->link_speed = 10000;
                        adapter->link_duplex = FULL_DUPLEX;
                        netif_carrier_on(netdev);
@@ -1116,9 +1140,7 @@ ixgb_watchdog(unsigned long data)
                if(netif_carrier_ok(netdev)) {
                        adapter->link_speed = 0;
                        adapter->link_duplex = 0;
-                       printk(KERN_INFO
-                                  "ixgb: %s NIC Link is Down\n",
-                                  netdev->name);
+                       DPRINTK(LINK, INFO, "NIC Link is Down\n");
                        netif_carrier_off(netdev);
                        netif_stop_queue(netdev);
 
@@ -1137,12 +1159,8 @@ ixgb_watchdog(unsigned long data)
                }
        }
 
-       /* Early detection of hung controller */
-       i = txdr->next_to_clean;
-       if(txdr->buffer_info[i].dma &&
-          time_after(jiffies, txdr->buffer_info[i].time_stamp + HZ) &&
-          !(IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_TXOFF))
-               netif_stop_queue(netdev);
+       /* Force detection of hung controller every watchdog period */
+       adapter->detect_tx_hung = TRUE;
 
        /* generate an interrupt to force clean up of any stragglers */
        IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
@@ -1155,7 +1173,7 @@ ixgb_watchdog(unsigned long data)
 #define IXGB_TX_FLAGS_VLAN             0x00000002
 #define IXGB_TX_FLAGS_TSO              0x00000004
 
-static inline boolean_t
+static int
 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
 {
 #ifdef NETIF_F_TSO
@@ -1163,10 +1181,18 @@ ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
        unsigned int i;
        uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
        uint16_t ipcse, tucse, mss;
+       int err;
+
+       if (likely(skb_is_gso(skb))) {
+               struct ixgb_buffer *buffer_info;
+               if (skb_header_cloned(skb)) {
+                       err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
+                       if (err)
+                               return err;
+               }
 
-       if(likely(skb_shinfo(skb)->tso_size)) {
                hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
-               mss = skb_shinfo(skb)->tso_size;
+               mss = skb_shinfo(skb)->gso_size;
                skb->nh.iph->tot_len = 0;
                skb->nh.iph->check = 0;
                skb->h.th->check = ~csum_tcpudp_magic(skb->nh.iph->saddr,
@@ -1181,6 +1207,8 @@ ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
 
                i = adapter->tx_ring.next_to_use;
                context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
+               buffer_info = &adapter->tx_ring.buffer_info[i];
+               WARN_ON(buffer_info->dma != 0);
 
                context_desc->ipcss = ipcss;
                context_desc->ipcso = ipcso;
@@ -1196,33 +1224,36 @@ ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
                                                | IXGB_CONTEXT_DESC_CMD_TSE
                                                | IXGB_CONTEXT_DESC_CMD_IP
                                                | IXGB_CONTEXT_DESC_CMD_TCP
-                                               | IXGB_CONTEXT_DESC_CMD_RS
                                                | IXGB_CONTEXT_DESC_CMD_IDE
                                                | (skb->len - (hdr_len)));
 
+
                if(++i == adapter->tx_ring.count) i = 0;
                adapter->tx_ring.next_to_use = i;
 
-               return TRUE;
+               return 1;
        }
 #endif
 
-       return FALSE;
+       return 0;
 }
 
-static inline boolean_t
+static boolean_t
 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
 {
        struct ixgb_context_desc *context_desc;
        unsigned int i;
        uint8_t css, cso;
 
-       if(likely(skb->ip_summed == CHECKSUM_HW)) {
+       if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
+               struct ixgb_buffer *buffer_info;
                css = skb->h.raw - skb->data;
-               cso = (skb->h.raw + skb->csum) - skb->data;
+               cso = css + skb->csum_offset;
 
                i = adapter->tx_ring.next_to_use;
                context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
+               buffer_info = &adapter->tx_ring.buffer_info[i];
+               WARN_ON(buffer_info->dma != 0);
 
                context_desc->tucss = css;
                context_desc->tucso = cso;
@@ -1234,8 +1265,7 @@ ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
                context_desc->mss = 0;
                context_desc->cmd_type_len =
                        cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
-                                       | IXGB_TX_DESC_CMD_RS 
-                                       | IXGB_TX_DESC_CMD_IDE);
+                                   | IXGB_TX_DESC_CMD_IDE);
 
                if(++i == adapter->tx_ring.count) i = 0;
                adapter->tx_ring.next_to_use = i;
@@ -1249,7 +1279,7 @@ ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
 #define IXGB_MAX_TXD_PWR       14
 #define IXGB_MAX_DATA_PER_TXD  (1<<IXGB_MAX_TXD_PWR)
 
-static inline int
+static int
 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
            unsigned int first)
 {
@@ -1257,23 +1287,32 @@ ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
        struct ixgb_buffer *buffer_info;
        int len = skb->len;
        unsigned int offset = 0, size, count = 0, i;
+       unsigned int mss = skb_shinfo(skb)->gso_size;
 
        unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
        unsigned int f;
+
        len -= skb->data_len;
 
        i = tx_ring->next_to_use;
 
        while(len) {
                buffer_info = &tx_ring->buffer_info[i];
-               size = min(len, IXGB_MAX_JUMBO_FRAME_SIZE);
+               size = min(len, IXGB_MAX_DATA_PER_TXD);
+               /* Workaround for premature desc write-backs
+                * in TSO mode.  Append 4-byte sentinel desc */
+               if (unlikely(mss && !nr_frags && size == len && size > 8))
+                       size -= 4;
+
                buffer_info->length = size;
+               WARN_ON(buffer_info->dma != 0);
                buffer_info->dma =
                        pci_map_single(adapter->pdev,
                                skb->data + offset,
                                size,
                                PCI_DMA_TODEVICE);
                buffer_info->time_stamp = jiffies;
+               buffer_info->next_to_watch = 0;
 
                len -= size;
                offset += size;
@@ -1290,7 +1329,14 @@ ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
 
                while(len) {
                        buffer_info = &tx_ring->buffer_info[i];
-                       size = min(len, IXGB_MAX_JUMBO_FRAME_SIZE);
+                       size = min(len, IXGB_MAX_DATA_PER_TXD);
+
+                       /* Workaround for premature desc write-backs
+                        * in TSO mode.  Append 4-byte sentinel desc */
+                       if (unlikely(mss && !nr_frags && size == len
+                                    && size > 8))
+                               size -= 4;
+
                        buffer_info->length = size;
                        buffer_info->dma =
                                pci_map_page(adapter->pdev,
@@ -1299,6 +1345,7 @@ ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
                                        size,
                                        PCI_DMA_TODEVICE);
                        buffer_info->time_stamp = jiffies;
+                       buffer_info->next_to_watch = 0;
 
                        len -= size;
                        offset += size;
@@ -1313,7 +1360,7 @@ ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
        return count;
 }
 
-static inline void
+static void
 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
 {
        struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
@@ -1364,33 +1411,80 @@ ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
        IXGB_WRITE_REG(&adapter->hw, TDT, i);
 }
 
+static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
+{
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
+       struct ixgb_desc_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(IXGB_DESC_UNUSED(tx_ring) < size))
+               return -EBUSY;
+
+       /* A reprieve! */
+       netif_start_queue(netdev);
+       ++adapter->restart_queue;
+       return 0;
+}
+
+static int ixgb_maybe_stop_tx(struct net_device *netdev,
+                              struct ixgb_desc_ring *tx_ring, int size)
+{
+       if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
+               return 0;
+       return __ixgb_maybe_stop_tx(netdev, size);
+}
+
+
 /* Tx Descriptors needed, worst case */
 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
                         (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
-#define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) + \
-       MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1
+#define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
+       MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
+       + 1 /* one more needed for sentinel TSO workaround */
 
 static int
 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        unsigned int first;
        unsigned int tx_flags = 0;
        unsigned long flags;
        int vlan_id = 0;
+       int tso;
 
        if(skb->len <= 0) {
                dev_kfree_skb_any(skb);
                return 0;
        }
 
+#ifdef NETIF_F_LLTX
+       local_irq_save(flags);
+       if (!spin_trylock(&adapter->tx_lock)) {
+               /* Collision - tell upper layer to requeue */
+               local_irq_restore(flags);
+               return NETDEV_TX_LOCKED;
+       }
+#else
        spin_lock_irqsave(&adapter->tx_lock, flags);
-       if(unlikely(IXGB_DESC_UNUSED(&adapter->tx_ring) < DESC_NEEDED)) {
+#endif
+
+       if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
+                     DESC_NEEDED))) {
                netif_stop_queue(netdev);
                spin_unlock_irqrestore(&adapter->tx_lock, flags);
-               return 1;
+               return NETDEV_TX_BUSY;
        }
+
+#ifndef NETIF_F_LLTX
        spin_unlock_irqrestore(&adapter->tx_lock, flags);
+#endif
 
        if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
                tx_flags |= IXGB_TX_FLAGS_VLAN;
@@ -1399,7 +1493,16 @@ ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 
        first = adapter->tx_ring.next_to_use;
        
-       if(ixgb_tso(adapter, skb))
+       tso = ixgb_tso(adapter, skb);
+       if (tso < 0) {
+               dev_kfree_skb_any(skb);
+#ifdef NETIF_F_LLTX
+               spin_unlock_irqrestore(&adapter->tx_lock, flags);
+#endif
+               return NETDEV_TX_OK;
+       }
+
+       if (likely(tso))
                tx_flags |= IXGB_TX_FLAGS_TSO;
        else if(ixgb_tx_csum(adapter, skb))
                tx_flags |= IXGB_TX_FLAGS_CSUM;
@@ -1409,7 +1512,14 @@ ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 
        netdev->trans_start = jiffies;
 
-       return 0;
+#ifdef NETIF_F_LLTX
+       /* Make sure there is space in the ring for the next send. */
+       ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
+
+       spin_unlock_irqrestore(&adapter->tx_lock, flags);
+
+#endif
+       return NETDEV_TX_OK;
 }
 
 /**
@@ -1420,17 +1530,19 @@ ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 static void
 ixgb_tx_timeout(struct net_device *netdev)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
 
        /* Do the reset outside of interrupt context */
        schedule_work(&adapter->tx_timeout_task);
 }
 
 static void
-ixgb_tx_timeout_task(struct net_device *netdev)
+ixgb_tx_timeout_task(struct work_struct *work)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter =
+               container_of(work, struct ixgb_adapter, tx_timeout_task);
 
+       adapter->tx_timeout_count++;
        ixgb_down(adapter, TRUE);
        ixgb_up(adapter);
 }
@@ -1446,7 +1558,7 @@ ixgb_tx_timeout_task(struct net_device *netdev)
 static struct net_device_stats *
 ixgb_get_stats(struct net_device *netdev)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
 
        return &adapter->net_stats;
 }
@@ -1462,35 +1574,22 @@ ixgb_get_stats(struct net_device *netdev)
 static int
 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
        int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 
 
        if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
           || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
-               IXGB_ERR("Invalid MTU setting\n");
+               DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
                return -EINVAL;
        }
 
-       if((max_frame <= IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
-          || (max_frame <= IXGB_RXBUFFER_2048)) {
-               adapter->rx_buffer_len = IXGB_RXBUFFER_2048;
-
-       } else if(max_frame <= IXGB_RXBUFFER_4096) {
-               adapter->rx_buffer_len = IXGB_RXBUFFER_4096;
-
-       } else if(max_frame <= IXGB_RXBUFFER_8192) {
-               adapter->rx_buffer_len = IXGB_RXBUFFER_8192;
-
-       } else {
-               adapter->rx_buffer_len = IXGB_RXBUFFER_16384;
-       }
+       adapter->rx_buffer_len = max_frame;
 
        netdev->mtu = new_mtu;
 
-       if(old_max_frame != max_frame && netif_running(netdev)) {
-
+       if ((old_max_frame != max_frame) && netif_running(netdev)) {
                ixgb_down(adapter, TRUE);
                ixgb_up(adapter);
        }
@@ -1506,14 +1605,39 @@ ixgb_change_mtu(struct net_device *netdev, int new_mtu)
 void
 ixgb_update_stats(struct ixgb_adapter *adapter)
 {
+       struct net_device *netdev = adapter->netdev;
+       struct pci_dev *pdev = adapter->pdev;
+
+       /* Prevent stats update while adapter is being reset */
+       if (pdev->error_state && pdev->error_state != pci_channel_io_normal)
+               return;
+
+       if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
+          (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
+               u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
+               u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
+               u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
+               u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
+
+               multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
+               /* fix up multicast stats by removing broadcasts */
+               if(multi >= bcast)
+                       multi -= bcast;
+               
+               adapter->stats.mprcl += (multi & 0xFFFFFFFF);
+               adapter->stats.mprch += (multi >> 32);
+               adapter->stats.bprcl += bcast_l; 
+               adapter->stats.bprch += bcast_h;
+       } else {
+               adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
+               adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
+               adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
+               adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
+       }
        adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
        adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
        adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
        adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
-       adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
-       adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
-       adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
-       adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
        adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
        adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
        adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
@@ -1585,8 +1709,6 @@ ixgb_update_stats(struct ixgb_adapter *adapter)
            adapter->stats.icbc +
            adapter->stats.ecbc + adapter->stats.mpc;
 
-       adapter->net_stats.rx_dropped = adapter->stats.mpc;
-
        /* see above
         * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
         */
@@ -1610,14 +1732,13 @@ ixgb_update_stats(struct ixgb_adapter *adapter)
  * ixgb_intr - Interrupt Handler
  * @irq: interrupt number
  * @data: pointer to a network interface device structure
- * @pt_regs: CPU registers structure
  **/
 
 static irqreturn_t
-ixgb_intr(int irq, void *data, struct pt_regs *regs)
+ixgb_intr(int irq, void *data)
 {
        struct net_device *netdev = data;
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        struct ixgb_hw *hw = &adapter->hw;
        uint32_t icr = IXGB_READ_REG(hw, ICR);
 #ifndef CONFIG_IXGB_NAPI
@@ -1664,24 +1785,20 @@ ixgb_intr(int irq, void *data, struct pt_regs *regs)
 static int
 ixgb_clean(struct net_device *netdev, int *budget)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        int work_to_do = min(*budget, netdev->quota);
        int tx_cleaned;
        int work_done = 0;
-       
-       if (!netif_carrier_ok(netdev))
-               goto quit_polling;
 
        tx_cleaned = ixgb_clean_tx_irq(adapter);
        ixgb_clean_rx_irq(adapter, &work_done, work_to_do);
 
        *budget -= work_done;
        netdev->quota -= work_done;
-       
-       /* if no Tx cleanup and not enough Rx work done, exit the polling mode */
-       if((!tx_cleaned && (work_done < work_to_do)) || 
-                               !netif_running(netdev)) {
-quit_polling:  netif_rx_complete(netdev);
+
+       /* if no Tx and not enough Rx work done, exit the polling mode */
+       if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev)) {
+               netif_rx_complete(netdev);
                ixgb_irq_enable(adapter);
                return 0;
        }
@@ -1734,13 +1851,44 @@ ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
 
        tx_ring->next_to_clean = i;
 
-       spin_lock(&adapter->tx_lock);
-       if(cleaned && netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
-          (IXGB_DESC_UNUSED(tx_ring) > IXGB_TX_QUEUE_WAKE)) {
+       if (unlikely(netif_queue_stopped(netdev))) {
+               spin_lock(&adapter->tx_lock);
+               if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
+                   (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
+                       netif_wake_queue(netdev);
+               spin_unlock(&adapter->tx_lock);
+       }
 
-               netif_wake_queue(netdev);
+       if(adapter->detect_tx_hung) {
+               /* 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 + HZ)
+                  && !(IXGB_READ_REG(&adapter->hw, STATUS) &
+                       IXGB_STATUS_TXOFF)) {
+                       /* detected Tx unit hang */
+                       DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
+                                       "  TDH                  <%x>\n"
+                                       "  TDT                  <%x>\n"
+                                       "  next_to_use          <%x>\n"
+                                       "  next_to_clean        <%x>\n"
+                                       "buffer_info[next_to_clean]\n"
+                                       "  time_stamp           <%lx>\n"
+                                       "  next_to_watch        <%x>\n"
+                                       "  jiffies              <%lx>\n"
+                                       "  next_to_watch.status <%x>\n",
+                               IXGB_READ_REG(&adapter->hw, TDH),
+                               IXGB_READ_REG(&adapter->hw, TDT),
+                               tx_ring->next_to_use,
+                               tx_ring->next_to_clean,
+                               tx_ring->buffer_info[eop].time_stamp,
+                               eop,
+                               jiffies,
+                               eop_desc->status);
+                       netif_stop_queue(netdev);
+               }
        }
-       spin_unlock(&adapter->tx_lock);
 
        return cleaned;
 }
@@ -1752,7 +1900,7 @@ ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
  * @sk_buff: socket buffer with received data
  **/
 
-static inline void
+static void
 ixgb_rx_checksum(struct ixgb_adapter *adapter,
                 struct ixgb_rx_desc *rx_desc,
                 struct sk_buff *skb)
@@ -1796,7 +1944,6 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
        struct pci_dev *pdev = adapter->pdev;
        struct ixgb_rx_desc *rx_desc, *next_rxd;
        struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
-       struct sk_buff *skb, *next_skb;
        uint32_t length;
        unsigned int i, j;
        boolean_t cleaned = FALSE;
@@ -1806,6 +1953,8 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
        buffer_info = &rx_ring->buffer_info[i];
 
        while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
+               struct sk_buff *skb, *next_skb;
+               u8 status;
 
 #ifdef CONFIG_IXGB_NAPI
                if(*work_done >= work_to_do)
@@ -1813,7 +1962,10 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
 
                (*work_done)++;
 #endif
+               status = rx_desc->status;
                skb = buffer_info->skb;
+               buffer_info->skb = NULL;
+
                prefetch(skb->data);
 
                if(++i == rx_ring->count) i = 0;
@@ -1828,7 +1980,6 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
                next_skb = next_buffer->skb;
                prefetch(next_skb);
 
-
                cleaned = TRUE;
 
                pci_unmap_single(pdev,
@@ -1838,7 +1989,7 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
 
                length = le16_to_cpu(rx_desc->length);
 
-               if(unlikely(!(rx_desc->status & IXGB_RX_DESC_STATUS_EOP))) {
+               if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
 
                        /* All receives must fit into a single buffer */
 
@@ -1846,12 +1997,7 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
                                         "length<%x>\n", length);
 
                        dev_kfree_skb_irq(skb);
-                       rx_desc->status = 0;
-                       buffer_info->skb = NULL;
-
-                       rx_desc = next_rxd;
-                       buffer_info = next_buffer;
-                       continue;
+                       goto rxdesc_done;
                }
 
                if (unlikely(rx_desc->errors
@@ -1860,13 +2006,27 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
                                IXGB_RX_DESC_ERRORS_RXE))) {
 
                        dev_kfree_skb_irq(skb);
-                       rx_desc->status = 0;
-                       buffer_info->skb = NULL;
+                       goto rxdesc_done;
+               }
 
-                       rx_desc = next_rxd;
-                       buffer_info = next_buffer;
-                       continue;
+               /* code added for copybreak, this should improve
+                * performance for small packets with large amounts
+                * of reassembly being done in the stack */
+#define IXGB_CB_LENGTH 256
+               if (length < IXGB_CB_LENGTH) {
+                       struct sk_buff *new_skb =
+                           netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
+                       if (new_skb) {
+                               skb_reserve(new_skb, NET_IP_ALIGN);
+                               memcpy(new_skb->data - NET_IP_ALIGN,
+                                      skb->data - NET_IP_ALIGN,
+                                      length + NET_IP_ALIGN);
+                               /* save the skb in buffer_info as good */
+                               buffer_info->skb = skb;
+                               skb = new_skb;
+                       }
                }
+               /* end copybreak code */
 
                /* Good Receive */
                skb_put(skb, length);
@@ -1876,7 +2036,7 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
 
                skb->protocol = eth_type_trans(skb, netdev);
 #ifdef CONFIG_IXGB_NAPI
-               if(adapter->vlgrp && (rx_desc->status & IXGB_RX_DESC_STATUS_VP)) {
+               if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
                        vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
                                le16_to_cpu(rx_desc->special) &
                                        IXGB_RX_DESC_SPECIAL_VLAN_MASK);
@@ -1884,7 +2044,7 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
                        netif_receive_skb(skb);
                }
 #else /* CONFIG_IXGB_NAPI */
-               if(adapter->vlgrp && (rx_desc->status & IXGB_RX_DESC_STATUS_VP)) {
+               if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
                        vlan_hwaccel_rx(skb, adapter->vlgrp,
                                le16_to_cpu(rx_desc->special) &
                                        IXGB_RX_DESC_SPECIAL_VLAN_MASK);
@@ -1894,9 +2054,11 @@ ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
 #endif /* CONFIG_IXGB_NAPI */
                netdev->last_rx = jiffies;
 
+rxdesc_done:
+               /* clean up descriptor, might be written over by hw */
                rx_desc->status = 0;
-               buffer_info->skb = NULL;
 
+               /* use prefetched values */
                rx_desc = next_rxd;
                buffer_info = next_buffer;
        }
@@ -1932,14 +2094,20 @@ ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
 
        num_group_tail_writes = IXGB_RX_BUFFER_WRITE;
 
-       /* leave one descriptor unused */
-       while(--cleancount > 0) {
-               rx_desc = IXGB_RX_DESC(*rx_ring, i);
-
-               skb = dev_alloc_skb(adapter->rx_buffer_len + NET_IP_ALIGN);
+       /* leave three descriptors unused */
+       while(--cleancount > 2) {
+               /* recycle! its good for you */
+               skb = buffer_info->skb;
+               if (skb) {
+                       skb_trim(skb, 0);
+                       goto map_skb;
+               }
 
-               if(unlikely(!skb)) {
+               skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
+                                      + NET_IP_ALIGN);
+               if (unlikely(!skb)) {
                        /* Better luck next round */
+                       adapter->alloc_rx_buff_failed++;
                        break;
                }
 
@@ -1949,33 +2117,38 @@ ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
                 */
                skb_reserve(skb, NET_IP_ALIGN);
 
-               skb->dev = netdev;
-
                buffer_info->skb = skb;
                buffer_info->length = adapter->rx_buffer_len;
-               buffer_info->dma =
-                       pci_map_single(pdev,
-                                  skb->data,
-                                  adapter->rx_buffer_len,
-                                  PCI_DMA_FROMDEVICE);
+map_skb:
+               buffer_info->dma = pci_map_single(pdev,
+                                                 skb->data,
+                                                 adapter->rx_buffer_len,
+                                                 PCI_DMA_FROMDEVICE);
 
+               rx_desc = IXGB_RX_DESC(*rx_ring, i);
                rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
+               /* guarantee DD bit not set now before h/w gets descriptor
+                * this is the rest of the workaround for h/w double 
+                * writeback. */
+               rx_desc->status = 0;
 
-               if((i & ~(num_group_tail_writes- 1)) == i) {
-                       /* Force memory writes to complete before letting h/w
-                        * know there are new descriptors to fetch.  (Only
-                        * applicable for weak-ordered memory model archs,
-                        * such as IA-64). */
-                       wmb();
-
-                       IXGB_WRITE_REG(&adapter->hw, RDT, i);
-               }
 
                if(++i == rx_ring->count) i = 0;
                buffer_info = &rx_ring->buffer_info[i];
        }
 
-       rx_ring->next_to_use = i;
+       if (likely(rx_ring->next_to_use != i)) {
+               rx_ring->next_to_use = i;
+               if (unlikely(i-- == 0))
+                       i = (rx_ring->count - 1);
+
+               /* Force memory writes to complete before letting h/w
+                * know there are new descriptors to fetch.  (Only
+                * applicable for weak-ordered memory model archs, such
+                * as IA-64). */
+               wmb();
+               IXGB_WRITE_REG(&adapter->hw, RDT, i);
+       }
 }
 
 /**
@@ -1987,7 +2160,7 @@ ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
 static void
 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        uint32_t ctrl, rctl;
 
        ixgb_irq_disable(adapter);
@@ -2025,7 +2198,7 @@ ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
 static void
 ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        uint32_t vfta, index;
 
        /* add VID to filter table */
@@ -2039,7 +2212,7 @@ ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
 static void
 ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
 {
-       struct ixgb_adapter *adapter = netdev->priv;
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
        uint32_t vfta, index;
 
        ixgb_irq_disable(adapter);
@@ -2072,68 +2245,115 @@ ixgb_restore_vlan(struct ixgb_adapter *adapter)
        }
 }
 
-/**
- * ixgb_notify_reboot - handles OS notification of reboot event.
- * @param nb notifier block, unused
- * @param event Event being passed to driver to act upon
- * @param p A pointer to our net device
- **/
-static int
-ixgb_notify_reboot(struct notifier_block *nb, unsigned long event, void *p)
+#ifdef CONFIG_NET_POLL_CONTROLLER
+/*
+ * Polling 'interrupt' - used by things like netconsole to send skbs
+ * without having to re-enable interrupts. It's not called while
+ * the interrupt routine is executing.
+ */
+
+static void ixgb_netpoll(struct net_device *dev)
 {
-       struct pci_dev *pdev = NULL;
-
-       switch(event) {
-       case SYS_DOWN:
-       case SYS_HALT:
-       case SYS_POWER_OFF:
-               while ((pdev = pci_find_device(PCI_ANY_ID, PCI_ANY_ID, pdev))) {
-                       if (pci_dev_driver(pdev) == &ixgb_driver)
-                               ixgb_suspend(pdev, 3);
-               }
-       }
-       return NOTIFY_DONE;
+       struct ixgb_adapter *adapter = netdev_priv(dev);
+
+       disable_irq(adapter->pdev->irq);
+       ixgb_intr(adapter->pdev->irq, dev);
+       enable_irq(adapter->pdev->irq);
 }
+#endif
 
 /**
- * ixgb_suspend - driver suspend function called from notify.
- * @param pdev pci driver structure used for passing to
- * @param state power state to enter 
- **/
-static int
-ixgb_suspend(struct pci_dev *pdev, uint32_t state)
+ * ixgb_io_error_detected() - called when PCI error is detected
+ * @pdev    pointer to pci device with error
+ * @state   pci channel state after error
+ *
+ * This callback is called by the PCI subsystem whenever
+ * a PCI bus error is detected.
+ */
+static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
+                                    enum pci_channel_state state)
 {
        struct net_device *netdev = pci_get_drvdata(pdev);
-       struct ixgb_adapter *adapter = netdev->priv;
-
-       netif_device_detach(netdev);
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
 
        if(netif_running(netdev))
                ixgb_down(adapter, TRUE);
 
-       pci_save_state(pdev);
+       pci_disable_device(pdev);
 
-       state = (state > 0) ? 3 : 0;
-       pci_set_power_state(pdev, state);
-       msec_delay(200);
-
-       return 0;
+       /* Request a slot reset. */
+       return PCI_ERS_RESULT_NEED_RESET;
 }
 
-#ifdef CONFIG_NET_POLL_CONTROLLER
-/*
- * Polling 'interrupt' - used by things like netconsole to send skbs
- * without having to re-enable interrupts. It's not called while
- * the interrupt routine is executing.
+/**
+ * ixgb_io_slot_reset - called after the pci bus has been reset.
+ * @pdev    pointer to pci device with error
+ *
+ * This callback is called after the PCI buss has been reset.
+ * Basically, this tries to restart the card from scratch.
+ * This is a shortened version of the device probe/discovery code,
+ * it resembles the first-half of the ixgb_probe() routine.
  */
+static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev)
+{
+       struct net_device *netdev = pci_get_drvdata(pdev);
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
 
-static void ixgb_netpoll(struct net_device *dev)
+       if(pci_enable_device(pdev)) {
+               DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+
+       /* Perform card reset only on one instance of the card */
+       if (0 != PCI_FUNC (pdev->devfn))
+               return PCI_ERS_RESULT_RECOVERED;
+
+       pci_set_master(pdev);
+
+       netif_carrier_off(netdev);
+       netif_stop_queue(netdev);
+       ixgb_reset(adapter);
+
+       /* Make sure the EEPROM is good */
+       if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
+               DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+       ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
+       memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
+
+       if(!is_valid_ether_addr(netdev->perm_addr)) {
+               DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+
+       return PCI_ERS_RESULT_RECOVERED;
+}
+
+/**
+ * ixgb_io_resume - called when its OK to resume normal operations
+ * @pdev    pointer to pci device with error
+ *
+ * The error recovery driver tells us that its OK to resume
+ * normal operation. Implementation resembles the second-half
+ * of the ixgb_probe() routine.
+ */
+static void ixgb_io_resume (struct pci_dev *pdev)
 {
-       struct ixgb_adapter *adapter = dev->priv;
-       disable_irq(adapter->pdev->irq);
-       ixgb_intr(adapter->pdev->irq, dev, NULL);
-       enable_irq(adapter->pdev->irq);
+       struct net_device *netdev = pci_get_drvdata(pdev);
+       struct ixgb_adapter *adapter = netdev_priv(netdev);
+
+       pci_set_master(pdev);
+
+       if(netif_running(netdev)) {
+               if(ixgb_up(adapter)) {
+                       printk ("ixgb: can't bring device back up after reset\n");
+                       return;
+               }
+       }
+
+       netif_device_attach(netdev);
+       mod_timer(&adapter->watchdog_timer, jiffies);
 }
-#endif
 
 /* ixgb_main.c */