igb: add support for 82576NS SerDes adapter
[safe/jmp/linux-2.6] / drivers / net / igb / igb_main.c
index 8e93750..83c0837 100644 (file)
@@ -63,8 +63,10 @@ static const struct e1000_info *igb_info_tbl[] = {
 static struct pci_device_id igb_pci_tbl[] = {
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
+       { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
+       { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
@@ -93,16 +95,19 @@ static void igb_clean_all_tx_rings(struct igb_adapter *);
 static void igb_clean_all_rx_rings(struct igb_adapter *);
 static void igb_clean_tx_ring(struct igb_ring *);
 static void igb_clean_rx_ring(struct igb_ring *);
-static void igb_set_multi(struct net_device *);
+static void igb_set_rx_mode(struct net_device *);
 static void igb_update_phy_info(unsigned long);
 static void igb_watchdog(unsigned long);
 static void igb_watchdog_task(struct work_struct *);
-static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
-                                 struct igb_ring *);
-static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
+static netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *,
+                                          struct net_device *,
+                                          struct igb_ring *);
+static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
+                                     struct net_device *);
 static struct net_device_stats *igb_get_stats(struct net_device *);
 static int igb_change_mtu(struct net_device *, int);
 static int igb_set_mac(struct net_device *, void *);
+static void igb_set_uta(struct igb_adapter *adapter);
 static irqreturn_t igb_intr(int irq, void *);
 static irqreturn_t igb_intr_msi(int irq, void *);
 static irqreturn_t igb_msix_other(int irq, void *);
@@ -124,17 +129,46 @@ static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
 static void igb_vlan_rx_add_vid(struct net_device *, u16);
 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
 static void igb_restore_vlan(struct igb_adapter *);
+static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
 static void igb_ping_all_vfs(struct igb_adapter *);
 static void igb_msg_task(struct igb_adapter *);
 static int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
-static inline void igb_set_rah_pool(struct e1000_hw *, int , int);
-static void igb_set_mc_list_pools(struct igb_adapter *, int, u16);
 static void igb_vmm_control(struct igb_adapter *);
-static inline void igb_set_vmolr(struct e1000_hw *, int);
-static inline int igb_set_vf_rlpml(struct igb_adapter *, int, int);
 static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 
+static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
+{
+       u32 reg_data;
+
+       reg_data = rd32(E1000_VMOLR(vfn));
+       reg_data |= E1000_VMOLR_BAM |    /* Accept broadcast */
+                   E1000_VMOLR_ROMPE |  /* Accept packets matched in MTA */
+                   E1000_VMOLR_AUPE |   /* Accept untagged packets */
+                   E1000_VMOLR_STRVLAN; /* Strip vlan tags */
+       wr32(E1000_VMOLR(vfn), reg_data);
+}
+
+static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
+                                 int vfn)
+{
+       struct e1000_hw *hw = &adapter->hw;
+       u32 vmolr;
+
+       /* if it isn't the PF check to see if VFs are enabled and
+        * increase the size to support vlan tags */
+       if (vfn < adapter->vfs_allocated_count &&
+           adapter->vf_data[vfn].vlans_enabled)
+               size += VLAN_TAG_SIZE;
+
+       vmolr = rd32(E1000_VMOLR(vfn));
+       vmolr &= ~E1000_VMOLR_RLPML_MASK;
+       vmolr |= size | E1000_VMOLR_LPE;
+       wr32(E1000_VMOLR(vfn), vmolr);
+
+       return 0;
+}
+
 #ifdef CONFIG_PM
 static int igb_suspend(struct pci_dev *, pm_message_t);
 static int igb_resume(struct pci_dev *);
@@ -784,9 +818,11 @@ static void igb_irq_disable(struct igb_adapter *adapter)
        struct e1000_hw *hw = &adapter->hw;
 
        if (adapter->msix_entries) {
-               wr32(E1000_EIAM, 0);
-               wr32(E1000_EIMC, ~0);
-               wr32(E1000_EIAC, 0);
+               u32 regval = rd32(E1000_EIAM);
+               wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
+               wr32(E1000_EIMC, adapter->eims_enable_mask);
+               regval = rd32(E1000_EIAC);
+               wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
        }
 
        wr32(E1000_IAM, 0);
@@ -804,8 +840,10 @@ static void igb_irq_enable(struct igb_adapter *adapter)
        struct e1000_hw *hw = &adapter->hw;
 
        if (adapter->msix_entries) {
-               wr32(E1000_EIAC, adapter->eims_enable_mask);
-               wr32(E1000_EIAM, adapter->eims_enable_mask);
+               u32 regval = rd32(E1000_EIAC);
+               wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
+               regval = rd32(E1000_EIAM);
+               wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
                wr32(E1000_EIMS, adapter->eims_enable_mask);
                if (adapter->vfs_allocated_count)
                        wr32(E1000_MBVFIMR, 0xFF);
@@ -891,7 +929,7 @@ static void igb_configure(struct igb_adapter *adapter)
        int i;
 
        igb_get_hw_control(adapter);
-       igb_set_multi(netdev);
+       igb_set_rx_mode(netdev);
 
        igb_restore_vlan(adapter);
 
@@ -935,7 +973,6 @@ int igb_up(struct igb_adapter *adapter)
                igb_configure_msix(adapter);
 
        igb_vmm_control(adapter);
-       igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
        igb_set_vmolr(hw, adapter->vfs_allocated_count);
 
        /* Clear any pending interrupts. */
@@ -1095,7 +1132,7 @@ void igb_reset(struct igb_adapter *adapter)
        }
        fc->pause_time = 0xFFFF;
        fc->send_xon = 1;
-       fc->type = fc->original_type;
+       fc->current_mode = fc->requested_mode;
 
        /* disable receive for all VFs and wait one second */
        if (adapter->vfs_allocated_count) {
@@ -1132,7 +1169,8 @@ static const struct net_device_ops igb_netdev_ops = {
        .ndo_stop               = igb_close,
        .ndo_start_xmit         = igb_xmit_frame_adv,
        .ndo_get_stats          = igb_get_stats,
-       .ndo_set_multicast_list = igb_set_multi,
+       .ndo_set_rx_mode        = igb_set_rx_mode,
+       .ndo_set_multicast_list = igb_set_rx_mode,
        .ndo_set_mac_address    = igb_set_mac,
        .ndo_change_mtu         = igb_change_mtu,
        .ndo_do_ioctl           = igb_ioctl,
@@ -1198,12 +1236,7 @@ static int __devinit igb_probe(struct pci_dev *pdev,
        if (err)
                goto err_pci_reg;
 
-       err = pci_enable_pcie_error_reporting(pdev);
-       if (err) {
-               dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
-                       "0x%x\n", err);
-               /* non-fatal, continue */
-       }
+       pci_enable_pcie_error_reporting(pdev);
 
        pci_set_master(pdev);
        pci_save_state(pdev);
@@ -1345,6 +1378,7 @@ static int __devinit igb_probe(struct pci_dev *pdev,
        netdev->vlan_features |= NETIF_F_TSO;
        netdev->vlan_features |= NETIF_F_TSO6;
        netdev->vlan_features |= NETIF_F_IP_CSUM;
+       netdev->vlan_features |= NETIF_F_IPV6_CSUM;
        netdev->vlan_features |= NETIF_F_SG;
 
        if (pci_using_dac)
@@ -1392,8 +1426,8 @@ static int __devinit igb_probe(struct pci_dev *pdev,
        hw->mac.autoneg = true;
        hw->phy.autoneg_advertised = 0x2f;
 
-       hw->fc.original_type = e1000_fc_default;
-       hw->fc.type = e1000_fc_default;
+       hw->fc.requested_mode = e1000_fc_default;
+       hw->fc.current_mode = e1000_fc_default;
 
        adapter->itr_setting = IGB_DEFAULT_ITR;
        adapter->itr = IGB_START_ITR;
@@ -1579,7 +1613,6 @@ static void __devexit igb_remove(struct pci_dev *pdev)
        struct net_device *netdev = pci_get_drvdata(pdev);
        struct igb_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
-       int err;
 
        /* flush_scheduled work may reschedule our watchdog task, so
         * explicitly disable watchdog tasks from being rescheduled  */
@@ -1633,10 +1666,7 @@ static void __devexit igb_remove(struct pci_dev *pdev)
 
        free_netdev(netdev);
 
-       err = pci_disable_pcie_error_reporting(pdev);
-       if (err)
-               dev_err(&pdev->dev,
-                       "pci_disable_pcie_error_reporting failed 0x%x\n", err);
+       pci_disable_pcie_error_reporting(pdev);
 
        pci_disable_device(pdev);
 }
@@ -1729,7 +1759,6 @@ static int igb_open(struct net_device *netdev)
        igb_configure(adapter);
 
        igb_vmm_control(adapter);
-       igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
        igb_set_vmolr(hw, adapter->vfs_allocated_count);
 
        err = igb_request_irq(adapter);
@@ -2258,6 +2287,13 @@ static void igb_configure_rx(struct igb_adapter *adapter)
        /* Set the default pool for the PF's first queue */
        igb_configure_vt_default_pool(adapter);
 
+       /* set UTA to appropriate mode */
+       igb_set_uta(adapter);
+
+       /* set the correct pool for the PF default MAC address in entry 0 */
+       igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
+                        adapter->vfs_allocated_count);
+
        igb_rlpml_set(adapter);
 
        /* Enable Receives */
@@ -2481,57 +2517,44 @@ static int igb_set_mac(struct net_device *netdev, void *p)
        memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
        memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
 
-       hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
-
-       igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
+       /* set the correct pool for the new PF MAC address in entry 0 */
+       igb_rar_set_qsel(adapter, hw->mac.addr, 0,
+                        adapter->vfs_allocated_count);
 
        return 0;
 }
 
 /**
- * igb_set_multi - Multicast and Promiscuous mode set
+ * igb_write_mc_addr_list - write multicast addresses to MTA
  * @netdev: network interface device structure
  *
- * The set_multi entry point is called whenever the multicast address
- * list or the network interface flags are updated.  This routine is
- * responsible for configuring the hardware for proper multicast,
- * promiscuous mode, and all-multi behavior.
+ * Writes multicast address list to the MTA hash table.
+ * Returns: -ENOMEM on failure
+ *                0 on no addresses written
+ *                X on writing X addresses to MTA
  **/
-static void igb_set_multi(struct net_device *netdev)
+static int igb_write_mc_addr_list(struct net_device *netdev)
 {
        struct igb_adapter *adapter = netdev_priv(netdev);
        struct e1000_hw *hw = &adapter->hw;
-       struct e1000_mac_info *mac = &hw->mac;
-       struct dev_mc_list *mc_ptr;
-       u8  *mta_list = NULL;
-       u32 rctl;
+       struct dev_mc_list *mc_ptr = netdev->mc_list;
+       u8  *mta_list;
+       u32 vmolr = 0;
        int i;
 
-       /* Check for Promiscuous and All Multicast modes */
-
-       rctl = rd32(E1000_RCTL);
-
-       if (netdev->flags & IFF_PROMISC) {
-               rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
-               rctl &= ~E1000_RCTL_VFE;
-       } else {
-               if (netdev->flags & IFF_ALLMULTI) {
-                       rctl |= E1000_RCTL_MPE;
-                       rctl &= ~E1000_RCTL_UPE;
-               } else
-                       rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
-               rctl |= E1000_RCTL_VFE;
+       if (!netdev->mc_count) {
+               /* nothing to program, so clear mc list */
+               igb_update_mc_addr_list(hw, NULL, 0);
+               igb_restore_vf_multicasts(adapter);
+               return 0;
        }
-       wr32(E1000_RCTL, rctl);
 
-       if (netdev->mc_count) {
-               mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
-               if (!mta_list) {
-                       dev_err(&adapter->pdev->dev,
-                               "failed to allocate multicast filter list\n");
-                       return;
-               }
-       }
+       mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
+       if (!mta_list)
+               return -ENOMEM;
+
+       /* set vmolr receive overflow multicast bit */
+       vmolr |= E1000_VMOLR_ROMPE;
 
        /* The shared function expects a packed array of only addresses. */
        mc_ptr = netdev->mc_list;
@@ -2542,14 +2565,125 @@ static void igb_set_multi(struct net_device *netdev)
                memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
                mc_ptr = mc_ptr->next;
        }
-       igb_update_mc_addr_list(hw, mta_list, i,
-                               adapter->vfs_allocated_count + 1,
-                               mac->rar_entry_count);
+       igb_update_mc_addr_list(hw, mta_list, i);
+       kfree(mta_list);
 
-       igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
-       igb_restore_vf_multicasts(adapter);
+       return netdev->mc_count;
+}
 
-       kfree(mta_list);
+/**
+ * igb_write_uc_addr_list - write unicast addresses to RAR table
+ * @netdev: network interface device structure
+ *
+ * Writes unicast address list to the RAR table.
+ * Returns: -ENOMEM on failure/insufficient address space
+ *                0 on no addresses written
+ *                X on writing X addresses to the RAR table
+ **/
+static int igb_write_uc_addr_list(struct net_device *netdev)
+{
+       struct igb_adapter *adapter = netdev_priv(netdev);
+       struct e1000_hw *hw = &adapter->hw;
+       unsigned int vfn = adapter->vfs_allocated_count;
+       unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
+       int count = 0;
+
+       /* return ENOMEM indicating insufficient memory for addresses */
+       if (netdev->uc.count > rar_entries)
+               return -ENOMEM;
+
+       if (netdev->uc.count && rar_entries) {
+               struct netdev_hw_addr *ha;
+               list_for_each_entry(ha, &netdev->uc.list, list) {
+                       if (!rar_entries)
+                               break;
+                       igb_rar_set_qsel(adapter, ha->addr,
+                                        rar_entries--,
+                                        vfn);
+                       count++;
+               }
+       }
+       /* write the addresses in reverse order to avoid write combining */
+       for (; rar_entries > 0 ; rar_entries--) {
+               wr32(E1000_RAH(rar_entries), 0);
+               wr32(E1000_RAL(rar_entries), 0);
+       }
+       wrfl();
+
+       return count;
+}
+
+/**
+ * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
+ * @netdev: network interface device structure
+ *
+ * The set_rx_mode entry point is called whenever the unicast or multicast
+ * address lists or the network interface flags are updated.  This routine is
+ * responsible for configuring the hardware for proper unicast, multicast,
+ * promiscuous mode, and all-multi behavior.
+ **/
+static void igb_set_rx_mode(struct net_device *netdev)
+{
+       struct igb_adapter *adapter = netdev_priv(netdev);
+       struct e1000_hw *hw = &adapter->hw;
+       unsigned int vfn = adapter->vfs_allocated_count;
+       u32 rctl, vmolr = 0;
+       int count;
+
+       /* Check for Promiscuous and All Multicast modes */
+       rctl = rd32(E1000_RCTL);
+
+       /* clear the effected bits */
+       rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
+
+       if (netdev->flags & IFF_PROMISC) {
+               rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
+               vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
+       } else {
+               if (netdev->flags & IFF_ALLMULTI) {
+                       rctl |= E1000_RCTL_MPE;
+                       vmolr |= E1000_VMOLR_MPME;
+               } else {
+                       /*
+                        * Write addresses to the MTA, if the attempt fails
+                        * then we should just turn on promiscous mode so
+                        * that we can at least receive multicast traffic
+                        */
+                       count = igb_write_mc_addr_list(netdev);
+                       if (count < 0) {
+                               rctl |= E1000_RCTL_MPE;
+                               vmolr |= E1000_VMOLR_MPME;
+                       } else if (count) {
+                               vmolr |= E1000_VMOLR_ROMPE;
+                       }
+               }
+               /*
+                * Write addresses to available RAR registers, if there is not
+                * sufficient space to store all the addresses then enable
+                * unicast promiscous mode
+                */
+               count = igb_write_uc_addr_list(netdev);
+               if (count < 0) {
+                       rctl |= E1000_RCTL_UPE;
+                       vmolr |= E1000_VMOLR_ROPE;
+               }
+               rctl |= E1000_RCTL_VFE;
+       }
+       wr32(E1000_RCTL, rctl);
+
+       /*
+        * In order to support SR-IOV and eventually VMDq it is necessary to set
+        * the VMOLR to enable the appropriate modes.  Without this workaround
+        * we will have issues with VLAN tag stripping not being done for frames
+        * that are only arriving because we are the default pool
+        */
+       if (hw->mac.type < e1000_82576)
+               return;
+
+       vmolr |= rd32(E1000_VMOLR(vfn)) &
+                ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
+       wr32(E1000_VMOLR(vfn), vmolr);
+       igb_restore_vf_multicasts(adapter);
 }
 
 /* Need to wait a few seconds after link up to get diagnostic information from
@@ -2584,10 +2718,6 @@ static bool igb_has_link(struct igb_adapter *adapter)
                        link_active = true;
                }
                break;
-       case e1000_media_type_fiber:
-               ret_val = hw->mac.ops.check_for_link(hw);
-               link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
-               break;
        case e1000_media_type_internal_serdes:
                ret_val = hw->mac.ops.check_for_link(hw);
                link_active = hw->mac.serdes_has_link;
@@ -3139,8 +3269,7 @@ static inline int igb_tx_map_adv(struct igb_adapter *adapter,
        /* set time_stamp *before* dma to help avoid a possible race */
        buffer_info->time_stamp = jiffies;
        buffer_info->next_to_watch = i;
-       buffer_info->dma = map[count];
-       count++;
+       buffer_info->dma = skb_shinfo(skb)->dma_head;
 
        for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
                struct skb_frag_struct *frag;
@@ -3164,7 +3293,7 @@ static inline int igb_tx_map_adv(struct igb_adapter *adapter,
        tx_ring->buffer_info[i].skb = skb;
        tx_ring->buffer_info[first].next_to_watch = i;
 
-       return count;
+       return count + 1;
 }
 
 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
@@ -3265,9 +3394,9 @@ static int igb_maybe_stop_tx(struct net_device *netdev,
        return __igb_maybe_stop_tx(netdev, tx_ring, size);
 }
 
-static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
-                                  struct net_device *netdev,
-                                  struct igb_ring *tx_ring)
+static netdev_tx_t igb_xmit_frame_ring_adv(struct sk_buff *skb,
+                                          struct net_device *netdev,
+                                          struct igb_ring *tx_ring)
 {
        struct igb_adapter *adapter = netdev_priv(netdev);
        unsigned int first;
@@ -3344,7 +3473,6 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
        if (count) {
                igb_tx_queue_adv(adapter, tx_ring, tx_flags, count,
                                 skb->len, hdr_len);
-               netdev->trans_start = jiffies;
                /* Make sure there is space in the ring for the next send. */
                igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
        } else {
@@ -3356,7 +3484,8 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
        return NETDEV_TX_OK;
 }
 
-static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
+static netdev_tx_t igb_xmit_frame_adv(struct sk_buff *skb,
+                                     struct net_device *netdev)
 {
        struct igb_adapter *adapter = netdev_priv(netdev);
        struct igb_ring *tx_ring;
@@ -3369,7 +3498,7 @@ static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
         * 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. */
-       return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
+       return igb_xmit_frame_ring_adv(skb, netdev, tx_ring);
 }
 
 /**
@@ -3903,10 +4032,10 @@ static int igb_set_vf_multicasts(struct igb_adapter *adapter,
        /* VFs are limited to using the MTA hash table for their multicast
         * addresses */
        for (i = 0; i < n; i++)
-               vf_data->vf_mc_hashes[i] = hash_list[i];;
+               vf_data->vf_mc_hashes[i] = hash_list[i];
 
        /* Flush and reset the mta with the new values */
-       igb_set_multi(adapter->netdev);
+       igb_set_rx_mode(adapter->netdev);
 
        return 0;
 }
@@ -3949,6 +4078,8 @@ static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
 
                wr32(E1000_VLVF(i), reg);
        }
+
+       adapter->vf_data[vf].vlans_enabled = 0;
 }
 
 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
@@ -3997,6 +4128,22 @@ static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
                        reg |= vid;
 
                        wr32(E1000_VLVF(i), reg);
+
+                       /* do not modify RLPML for PF devices */
+                       if (vf >= adapter->vfs_allocated_count)
+                               return 0;
+
+                       if (!adapter->vf_data[vf].vlans_enabled) {
+                               u32 size;
+                               reg = rd32(E1000_VMOLR(vf));
+                               size = reg & E1000_VMOLR_RLPML_MASK;
+                               size += 4;
+                               reg &= ~E1000_VMOLR_RLPML_MASK;
+                               reg |= size;
+                               wr32(E1000_VMOLR(vf), reg);
+                       }
+                       adapter->vf_data[vf].vlans_enabled++;
+
                        return 0;
                }
        } else {
@@ -4009,6 +4156,21 @@ static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
                                igb_vfta_set(hw, vid, false);
                        }
                        wr32(E1000_VLVF(i), reg);
+
+                       /* do not modify RLPML for PF devices */
+                       if (vf >= adapter->vfs_allocated_count)
+                               return 0;
+
+                       adapter->vf_data[vf].vlans_enabled--;
+                       if (!adapter->vf_data[vf].vlans_enabled) {
+                               u32 size;
+                               reg = rd32(E1000_VMOLR(vf));
+                               size = reg & E1000_VMOLR_RLPML_MASK;
+                               size -= 4;
+                               reg &= ~E1000_VMOLR_RLPML_MASK;
+                               reg |= size;
+                               wr32(E1000_VMOLR(vf), reg);
+                       }
                        return 0;
                }
        }
@@ -4040,13 +4202,14 @@ static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
        adapter->vf_data[vf].num_vf_mc_hashes = 0;
 
        /* Flush and reset the mta with the new values */
-       igb_set_multi(adapter->netdev);
+       igb_set_rx_mode(adapter->netdev);
 }
 
 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
 {
        struct e1000_hw *hw = &adapter->hw;
        unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
+       int rar_entry = hw->mac.rar_entry_count - (vf + 1);
        u32 reg, msgbuf[3];
        u8 *addr = (u8 *)(&msgbuf[1]);
 
@@ -4054,8 +4217,7 @@ static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
        igb_vf_reset_event(adapter, vf);
 
        /* set vf mac address */
-       igb_rar_set(hw, vf_mac, vf + 1);
-       igb_set_rah_pool(hw, vf, vf + 1);
+       igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
 
        /* enable transmit and receive for vf */
        reg = rd32(E1000_VFTE);
@@ -4185,6 +4347,33 @@ static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
 }
 
 /**
+ *  igb_set_uta - Set unicast filter table address
+ *  @adapter: board private structure
+ *
+ *  The unicast table address is a register array of 32-bit registers.
+ *  The table is meant to be used in a way similar to how the MTA is used
+ *  however due to certain limitations in the hardware it is necessary to
+ *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscous
+ *  enable bit to allow vlan tag stripping when promiscous mode is enabled
+ **/
+static void igb_set_uta(struct igb_adapter *adapter)
+{
+       struct e1000_hw *hw = &adapter->hw;
+       int i;
+
+       /* The UTA table only exists on 82576 hardware and newer */
+       if (hw->mac.type < e1000_82576)
+               return;
+
+       /* we only need to do this if VMDq is enabled */
+       if (!adapter->vfs_allocated_count)
+               return;
+
+       for (i = 0; i < hw->mac.uta_reg_count; i++)
+               array_wr32(E1000_UTA, i, ~0);
+}
+
+/**
  * igb_intr_msi - Interrupt Handler
  * @irq: interrupt number
  * @data: pointer to a network interface device structure
@@ -4510,6 +4699,20 @@ static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
        adapter->hw_csum_good++;
 }
 
+static inline u16 igb_get_hlen(struct igb_adapter *adapter,
+                               union e1000_adv_rx_desc *rx_desc)
+{
+       /* HW will not DMA in data larger than the given buffer, even if it
+        * parses the (NFS, of course) header to be larger.  In that case, it
+        * fills the header buffer and spills the rest into the page.
+        */
+       u16 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
+                  E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
+       if (hlen > adapter->rx_ps_hdr_size)
+               hlen = adapter->rx_ps_hdr_size;
+       return hlen;
+}
+
 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
                                 int *work_done, int budget)
 {
@@ -4524,7 +4727,8 @@ static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
        int cleaned_count = 0;
        unsigned int total_bytes = 0, total_packets = 0;
        unsigned int i;
-       u32 length, hlen, staterr;
+       u32 staterr;
+       u16 length;
 
        i = rx_ring->next_to_clean;
        buffer_info = &rx_ring->buffer_info[i];
@@ -4551,29 +4755,22 @@ static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
                cleaned = true;
                cleaned_count++;
 
+               /* this is the fast path for the non-packet split case */
                if (!adapter->rx_ps_hdr_size) {
                        pci_unmap_single(pdev, buffer_info->dma,
-                                        adapter->rx_buffer_len +
-                                          NET_IP_ALIGN,
+                                        adapter->rx_buffer_len,
                                         PCI_DMA_FROMDEVICE);
+                       buffer_info->dma = 0;
                        skb_put(skb, length);
                        goto send_up;
                }
 
-               /* HW will not DMA in data larger than the given buffer, even
-                * if it parses the (NFS, of course) header to be larger.  In
-                * that case, it fills the header buffer and spills the rest
-                * into the page.
-                */
-               hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
-                 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
-               if (hlen > adapter->rx_ps_hdr_size)
-                       hlen = adapter->rx_ps_hdr_size;
-
-               if (!skb_shinfo(skb)->nr_frags) {
+               if (buffer_info->dma) {
+                       u16 hlen = igb_get_hlen(adapter, rx_desc);
                        pci_unmap_single(pdev, buffer_info->dma,
-                                        adapter->rx_ps_hdr_size + NET_IP_ALIGN,
+                                        adapter->rx_ps_hdr_size,
                                         PCI_DMA_FROMDEVICE);
+                       buffer_info->dma = 0;
                        skb_put(skb, hlen);
                }
 
@@ -4715,7 +4912,6 @@ static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
                bufsz = adapter->rx_ps_hdr_size;
        else
                bufsz = adapter->rx_buffer_len;
-       bufsz += NET_IP_ALIGN;
 
        while (cleaned_count--) {
                rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
@@ -4739,7 +4935,7 @@ static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
                }
 
                if (!buffer_info->skb) {
-                       skb = netdev_alloc_skb(netdev, bufsz);
+                       skb = netdev_alloc_skb(netdev, bufsz + NET_IP_ALIGN);
                        if (!skb) {
                                adapter->alloc_rx_buff_failed++;
                                goto no_buffers;
@@ -4810,8 +5006,6 @@ static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
                data->phy_id = adapter->hw.phy.addr;
                break;
        case SIOCGMIIREG:
-               if (!capable(CAP_NET_ADMIN))
-                       return -EPERM;
                if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
                                     &data->val_out))
                        return -EIO;
@@ -5000,6 +5194,34 @@ static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
        }
 }
 
+s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
+{
+       struct igb_adapter *adapter = hw->back;
+       u16 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 0;
+}
+
+s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
+{
+       struct igb_adapter *adapter = hw->back;
+       u16 cap_offset;
+
+       cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
+       if (!cap_offset)
+               return -E1000_ERR_CONFIG;
+
+       pci_write_config_word(adapter->pdev, cap_offset + reg, *value);
+
+       return 0;
+}
+
 static void igb_vlan_rx_register(struct net_device *netdev,
                                 struct vlan_group *grp)
 {
@@ -5103,14 +5325,6 @@ int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
 
        mac->autoneg = 0;
 
-       /* Fiber NICs only allow 1000 gbps Full duplex */
-       if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
-               spddplx != (SPEED_1000 + DUPLEX_FULL)) {
-               dev_err(&adapter->pdev->dev,
-                       "Unsupported Speed/Duplex configuration\n");
-               return -EINVAL;
-       }
-
        switch (spddplx) {
        case SPEED_10 + DUPLEX_HALF:
                mac->forced_speed_duplex = ADVERTISE_10_HALF;
@@ -5169,7 +5383,7 @@ static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
 
        if (wufc) {
                igb_setup_rctl(adapter);
-               igb_set_multi(netdev);
+               igb_set_rx_mode(netdev);
 
                /* turn on all-multi mode if wake on multicast is enabled */
                if (wufc & E1000_WUFC_MC) {
@@ -5198,7 +5412,7 @@ static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake)
 
        *enable_wake = wufc || adapter->en_mng_pt;
        if (!*enable_wake)
-               igb_shutdown_fiber_serdes_link_82575(hw);
+               igb_shutdown_serdes_link_82575(hw);
 
        /* Release control of h/w to f/w.  If f/w is AMT enabled, this
         * would have already happened in close and is redundant. */
@@ -5340,6 +5554,9 @@ static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
 
        netif_device_detach(netdev);
 
+       if (state == pci_channel_io_perm_failure)
+               return PCI_ERS_RESULT_DISCONNECT;
+
        if (netif_running(netdev))
                igb_down(adapter);
        pci_disable_device(pdev);
@@ -5416,67 +5633,44 @@ static void igb_io_resume(struct pci_dev *pdev)
        igb_get_hw_control(adapter);
 }
 
-static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
-{
-       u32 reg_data;
-
-       reg_data = rd32(E1000_VMOLR(vfn));
-       reg_data |= E1000_VMOLR_BAM |    /* Accept broadcast */
-                   E1000_VMOLR_ROPE |   /* Accept packets matched in UTA */
-                   E1000_VMOLR_ROMPE |  /* Accept packets matched in MTA */
-                   E1000_VMOLR_AUPE |   /* Accept untagged packets */
-                   E1000_VMOLR_STRVLAN; /* Strip vlan tags */
-       wr32(E1000_VMOLR(vfn), reg_data);
-}
-
-static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
-                                 int vfn)
+static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
+                             u8 qsel)
 {
+       u32 rar_low, rar_high;
        struct e1000_hw *hw = &adapter->hw;
-       u32 vmolr;
 
-       vmolr = rd32(E1000_VMOLR(vfn));
-       vmolr &= ~E1000_VMOLR_RLPML_MASK;
-       vmolr |= size | E1000_VMOLR_LPE;
-       wr32(E1000_VMOLR(vfn), vmolr);
-
-       return 0;
-}
-
-static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
-{
-       u32 reg_data;
-
-       reg_data = rd32(E1000_RAH(entry));
-       reg_data &= ~E1000_RAH_POOL_MASK;
-       reg_data |= E1000_RAH_POOL_1 << pool;;
-       wr32(E1000_RAH(entry), reg_data);
-}
+       /* HW expects these in little endian so we reverse the byte order
+        * from network order (big endian) to little endian
+        */
+       rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
+                 ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
+       rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
 
-static void igb_set_mc_list_pools(struct igb_adapter *adapter,
-                                 int entry_count, u16 total_rar_filters)
-{
-       struct e1000_hw *hw = &adapter->hw;
-       int i = adapter->vfs_allocated_count + 1;
+       /* Indicate to hardware the Address is Valid. */
+       rar_high |= E1000_RAH_AV;
 
-       if ((i + entry_count) < total_rar_filters)
-               total_rar_filters = i + entry_count;
+       if (hw->mac.type == e1000_82575)
+               rar_high |= E1000_RAH_POOL_1 * qsel;
+       else
+               rar_high |= E1000_RAH_POOL_1 << qsel;
 
-       for (; i < total_rar_filters; i++)
-               igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
+       wr32(E1000_RAL(index), rar_low);
+       wrfl();
+       wr32(E1000_RAH(index), rar_high);
+       wrfl();
 }
 
 static int igb_set_vf_mac(struct igb_adapter *adapter,
                           int vf, unsigned char *mac_addr)
 {
        struct e1000_hw *hw = &adapter->hw;
-       int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
-
-       igb_rar_set(hw, mac_addr, rar_entry);
+       /* VF MAC addresses start at end of receive addresses and moves
+        * torwards the first, as a result a collision should not be possible */
+       int rar_entry = hw->mac.rar_entry_count - (vf + 1);
 
        memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
 
-       igb_set_rah_pool(hw, vf, rar_entry);
+       igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
 
        return 0;
 }