Merge branch 'upstream-net26' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
authorDavid S. Miller <davem@davemloft.net>
Sun, 23 Mar 2008 01:22:42 +0000 (18:22 -0700)
committerDavid S. Miller <davem@davemloft.net>
Sun, 23 Mar 2008 01:22:42 +0000 (18:22 -0700)
Conflicts:

drivers/s390/net/qeth_main.c

1  2 
Documentation/feature-removal-schedule.txt
drivers/net/atlx/atl1.c
drivers/s390/net/qeth_l3_main.c

index 0000000000000000000000000000000000000000,240db847e1296de88d4d7077eb27c3750584603d..5586fc62468880a0df991cd72d461bf6e5ab77a3
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,3565 +1,3564 @@@
 -      u8 update_rx;
+ /*
+  * Copyright(c) 2005 - 2006 Attansic Corporation. All rights reserved.
+  * Copyright(c) 2006 - 2007 Chris Snook <csnook@redhat.com>
+  * Copyright(c) 2006 Jay Cliburn <jcliburn@gmail.com>
+  *
+  * Derived from Intel e1000 driver
+  * Copyright(c) 1999 - 2005 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
+  * 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 COPYING.
+  *
+  * Contact Information:
+  * Xiong Huang <xiong_huang@attansic.com>
+  * Attansic Technology Corp. 3F 147, Xianzheng 9th Road, Zhubei,
+  * Xinzhu  302, TAIWAN, REPUBLIC OF CHINA
+  *
+  * Chris Snook <csnook@redhat.com>
+  * Jay Cliburn <jcliburn@gmail.com>
+  *
+  * This version is adapted from the Attansic reference driver for
+  * inclusion in the Linux kernel.  It is currently under heavy development.
+  * A very incomplete list of things that need to be dealt with:
+  *
+  * TODO:
+  * Wake on LAN.
+  * Add more ethtool functions.
+  * Fix abstruse irq enable/disable condition described here:
+  *    http://marc.theaimsgroup.com/?l=linux-netdev&m=116398508500553&w=2
+  *
+  * NEEDS TESTING:
+  * VLAN
+  * multicast
+  * promiscuous mode
+  * interrupt coalescing
+  * SMP torture testing
+  */
+ #include <asm/atomic.h>
+ #include <asm/byteorder.h>
+ #include <linux/compiler.h>
+ #include <linux/crc32.h>
+ #include <linux/delay.h>
+ #include <linux/dma-mapping.h>
+ #include <linux/etherdevice.h>
+ #include <linux/hardirq.h>
+ #include <linux/if_ether.h>
+ #include <linux/if_vlan.h>
+ #include <linux/in.h>
+ #include <linux/interrupt.h>
+ #include <linux/ip.h>
+ #include <linux/irqflags.h>
+ #include <linux/irqreturn.h>
+ #include <linux/jiffies.h>
+ #include <linux/mii.h>
+ #include <linux/module.h>
+ #include <linux/moduleparam.h>
+ #include <linux/net.h>
+ #include <linux/netdevice.h>
+ #include <linux/pci.h>
+ #include <linux/pci_ids.h>
+ #include <linux/pm.h>
+ #include <linux/skbuff.h>
+ #include <linux/slab.h>
+ #include <linux/spinlock.h>
+ #include <linux/string.h>
+ #include <linux/tcp.h>
+ #include <linux/timer.h>
+ #include <linux/types.h>
+ #include <linux/workqueue.h>
+ #include <net/checksum.h>
+ #include "atl1.h"
+ /* Temporary hack for merging atl1 and atl2 */
+ #include "atlx.c"
+ /*
+  * atl1_pci_tbl - PCI Device ID Table
+  */
+ static const struct pci_device_id atl1_pci_tbl[] = {
+       {PCI_DEVICE(PCI_VENDOR_ID_ATTANSIC, PCI_DEVICE_ID_ATTANSIC_L1)},
+       /* required last entry */
+       {0,}
+ };
+ MODULE_DEVICE_TABLE(pci, atl1_pci_tbl);
+ static const u32 atl1_default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
+       NETIF_MSG_LINK | NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP;
+ static int debug = -1;
+ module_param(debug, int, 0);
+ MODULE_PARM_DESC(debug, "Message level (0=none,...,16=all)");
+ /*
+  * Reset the transmit and receive units; mask and clear all interrupts.
+  * hw - Struct containing variables accessed by shared code
+  * return : 0  or  idle status (if error)
+  */
+ static s32 atl1_reset_hw(struct atl1_hw *hw)
+ {
+       struct pci_dev *pdev = hw->back->pdev;
+       struct atl1_adapter *adapter = hw->back;
+       u32 icr;
+       int i;
+       /*
+        * Clear Interrupt mask to stop board from generating
+        * interrupts & Clear any pending interrupt events
+        */
+       /*
+        * iowrite32(0, hw->hw_addr + REG_IMR);
+        * iowrite32(0xffffffff, hw->hw_addr + REG_ISR);
+        */
+       /*
+        * Issue Soft Reset to the MAC.  This will reset the chip's
+        * transmit, receive, DMA.  It will not effect
+        * the current PCI configuration.  The global reset bit is self-
+        * clearing, and should clear within a microsecond.
+        */
+       iowrite32(MASTER_CTRL_SOFT_RST, hw->hw_addr + REG_MASTER_CTRL);
+       ioread32(hw->hw_addr + REG_MASTER_CTRL);
+       iowrite16(1, hw->hw_addr + REG_PHY_ENABLE);
+       ioread16(hw->hw_addr + REG_PHY_ENABLE);
+       /* delay about 1ms */
+       msleep(1);
+       /* Wait at least 10ms for All module to be Idle */
+       for (i = 0; i < 10; i++) {
+               icr = ioread32(hw->hw_addr + REG_IDLE_STATUS);
+               if (!icr)
+                       break;
+               /* delay 1 ms */
+               msleep(1);
+               /* FIXME: still the right way to do this? */
+               cpu_relax();
+       }
+       if (icr) {
+               if (netif_msg_hw(adapter))
+                       dev_dbg(&pdev->dev, "ICR = 0x%x\n", icr);
+               return icr;
+       }
+       return 0;
+ }
+ /* function about EEPROM
+  *
+  * check_eeprom_exist
+  * return 0 if eeprom exist
+  */
+ static int atl1_check_eeprom_exist(struct atl1_hw *hw)
+ {
+       u32 value;
+       value = ioread32(hw->hw_addr + REG_SPI_FLASH_CTRL);
+       if (value & SPI_FLASH_CTRL_EN_VPD) {
+               value &= ~SPI_FLASH_CTRL_EN_VPD;
+               iowrite32(value, hw->hw_addr + REG_SPI_FLASH_CTRL);
+       }
+       value = ioread16(hw->hw_addr + REG_PCIE_CAP_LIST);
+       return ((value & 0xFF00) == 0x6C00) ? 0 : 1;
+ }
+ static bool atl1_read_eeprom(struct atl1_hw *hw, u32 offset, u32 *p_value)
+ {
+       int i;
+       u32 control;
+       if (offset & 3)
+               /* address do not align */
+               return false;
+       iowrite32(0, hw->hw_addr + REG_VPD_DATA);
+       control = (offset & VPD_CAP_VPD_ADDR_MASK) << VPD_CAP_VPD_ADDR_SHIFT;
+       iowrite32(control, hw->hw_addr + REG_VPD_CAP);
+       ioread32(hw->hw_addr + REG_VPD_CAP);
+       for (i = 0; i < 10; i++) {
+               msleep(2);
+               control = ioread32(hw->hw_addr + REG_VPD_CAP);
+               if (control & VPD_CAP_VPD_FLAG)
+                       break;
+       }
+       if (control & VPD_CAP_VPD_FLAG) {
+               *p_value = ioread32(hw->hw_addr + REG_VPD_DATA);
+               return true;
+       }
+       /* timeout */
+       return false;
+ }
+ /*
+  * Reads the value from a PHY register
+  * hw - Struct containing variables accessed by shared code
+  * reg_addr - address of the PHY register to read
+  */
+ s32 atl1_read_phy_reg(struct atl1_hw *hw, u16 reg_addr, u16 *phy_data)
+ {
+       u32 val;
+       int i;
+       val = ((u32) (reg_addr & MDIO_REG_ADDR_MASK)) << MDIO_REG_ADDR_SHIFT |
+               MDIO_START | MDIO_SUP_PREAMBLE | MDIO_RW | MDIO_CLK_25_4 <<
+               MDIO_CLK_SEL_SHIFT;
+       iowrite32(val, hw->hw_addr + REG_MDIO_CTRL);
+       ioread32(hw->hw_addr + REG_MDIO_CTRL);
+       for (i = 0; i < MDIO_WAIT_TIMES; i++) {
+               udelay(2);
+               val = ioread32(hw->hw_addr + REG_MDIO_CTRL);
+               if (!(val & (MDIO_START | MDIO_BUSY)))
+                       break;
+       }
+       if (!(val & (MDIO_START | MDIO_BUSY))) {
+               *phy_data = (u16) val;
+               return 0;
+       }
+       return ATLX_ERR_PHY;
+ }
+ #define CUSTOM_SPI_CS_SETUP   2
+ #define CUSTOM_SPI_CLK_HI     2
+ #define CUSTOM_SPI_CLK_LO     2
+ #define CUSTOM_SPI_CS_HOLD    2
+ #define CUSTOM_SPI_CS_HI      3
+ static bool atl1_spi_read(struct atl1_hw *hw, u32 addr, u32 *buf)
+ {
+       int i;
+       u32 value;
+       iowrite32(0, hw->hw_addr + REG_SPI_DATA);
+       iowrite32(addr, hw->hw_addr + REG_SPI_ADDR);
+       value = SPI_FLASH_CTRL_WAIT_READY |
+           (CUSTOM_SPI_CS_SETUP & SPI_FLASH_CTRL_CS_SETUP_MASK) <<
+           SPI_FLASH_CTRL_CS_SETUP_SHIFT | (CUSTOM_SPI_CLK_HI &
+                                            SPI_FLASH_CTRL_CLK_HI_MASK) <<
+           SPI_FLASH_CTRL_CLK_HI_SHIFT | (CUSTOM_SPI_CLK_LO &
+                                          SPI_FLASH_CTRL_CLK_LO_MASK) <<
+           SPI_FLASH_CTRL_CLK_LO_SHIFT | (CUSTOM_SPI_CS_HOLD &
+                                          SPI_FLASH_CTRL_CS_HOLD_MASK) <<
+           SPI_FLASH_CTRL_CS_HOLD_SHIFT | (CUSTOM_SPI_CS_HI &
+                                           SPI_FLASH_CTRL_CS_HI_MASK) <<
+           SPI_FLASH_CTRL_CS_HI_SHIFT | (1 & SPI_FLASH_CTRL_INS_MASK) <<
+           SPI_FLASH_CTRL_INS_SHIFT;
+       iowrite32(value, hw->hw_addr + REG_SPI_FLASH_CTRL);
+       value |= SPI_FLASH_CTRL_START;
+       iowrite32(value, hw->hw_addr + REG_SPI_FLASH_CTRL);
+       ioread32(hw->hw_addr + REG_SPI_FLASH_CTRL);
+       for (i = 0; i < 10; i++) {
+               msleep(1);
+               value = ioread32(hw->hw_addr + REG_SPI_FLASH_CTRL);
+               if (!(value & SPI_FLASH_CTRL_START))
+                       break;
+       }
+       if (value & SPI_FLASH_CTRL_START)
+               return false;
+       *buf = ioread32(hw->hw_addr + REG_SPI_DATA);
+       return true;
+ }
+ /*
+  * get_permanent_address
+  * return 0 if get valid mac address,
+  */
+ static int atl1_get_permanent_address(struct atl1_hw *hw)
+ {
+       u32 addr[2];
+       u32 i, control;
+       u16 reg;
+       u8 eth_addr[ETH_ALEN];
+       bool key_valid;
+       if (is_valid_ether_addr(hw->perm_mac_addr))
+               return 0;
+       /* init */
+       addr[0] = addr[1] = 0;
+       if (!atl1_check_eeprom_exist(hw)) {
+               reg = 0;
+               key_valid = false;
+               /* Read out all EEPROM content */
+               i = 0;
+               while (1) {
+                       if (atl1_read_eeprom(hw, i + 0x100, &control)) {
+                               if (key_valid) {
+                                       if (reg == REG_MAC_STA_ADDR)
+                                               addr[0] = control;
+                                       else if (reg == (REG_MAC_STA_ADDR + 4))
+                                               addr[1] = control;
+                                       key_valid = false;
+                               } else if ((control & 0xff) == 0x5A) {
+                                       key_valid = true;
+                                       reg = (u16) (control >> 16);
+                               } else
+                                       break;
+                       } else
+                               /* read error */
+                               break;
+                       i += 4;
+               }
+               *(u32 *) &eth_addr[2] = swab32(addr[0]);
+               *(u16 *) &eth_addr[0] = swab16(*(u16 *) &addr[1]);
+               if (is_valid_ether_addr(eth_addr)) {
+                       memcpy(hw->perm_mac_addr, eth_addr, ETH_ALEN);
+                       return 0;
+               }
+               return 1;
+       }
+       /* see if SPI FLAGS exist ? */
+       addr[0] = addr[1] = 0;
+       reg = 0;
+       key_valid = false;
+       i = 0;
+       while (1) {
+               if (atl1_spi_read(hw, i + 0x1f000, &control)) {
+                       if (key_valid) {
+                               if (reg == REG_MAC_STA_ADDR)
+                                       addr[0] = control;
+                               else if (reg == (REG_MAC_STA_ADDR + 4))
+                                       addr[1] = control;
+                               key_valid = false;
+                       } else if ((control & 0xff) == 0x5A) {
+                               key_valid = true;
+                               reg = (u16) (control >> 16);
+                       } else
+                               /* data end */
+                               break;
+               } else
+                       /* read error */
+                       break;
+               i += 4;
+       }
+       *(u32 *) &eth_addr[2] = swab32(addr[0]);
+       *(u16 *) &eth_addr[0] = swab16(*(u16 *) &addr[1]);
+       if (is_valid_ether_addr(eth_addr)) {
+               memcpy(hw->perm_mac_addr, eth_addr, ETH_ALEN);
+               return 0;
+       }
+       /*
+        * On some motherboards, the MAC address is written by the
+        * BIOS directly to the MAC register during POST, and is
+        * not stored in eeprom.  If all else thus far has failed
+        * to fetch the permanent MAC address, try reading it directly.
+        */
+       addr[0] = ioread32(hw->hw_addr + REG_MAC_STA_ADDR);
+       addr[1] = ioread16(hw->hw_addr + (REG_MAC_STA_ADDR + 4));
+       *(u32 *) &eth_addr[2] = swab32(addr[0]);
+       *(u16 *) &eth_addr[0] = swab16(*(u16 *) &addr[1]);
+       if (is_valid_ether_addr(eth_addr)) {
+               memcpy(hw->perm_mac_addr, eth_addr, ETH_ALEN);
+               return 0;
+       }
+       return 1;
+ }
+ /*
+  * Reads the adapter's MAC address from the EEPROM
+  * hw - Struct containing variables accessed by shared code
+  */
+ s32 atl1_read_mac_addr(struct atl1_hw *hw)
+ {
+       u16 i;
+       if (atl1_get_permanent_address(hw))
+               random_ether_addr(hw->perm_mac_addr);
+       for (i = 0; i < ETH_ALEN; i++)
+               hw->mac_addr[i] = hw->perm_mac_addr[i];
+       return 0;
+ }
+ /*
+  * Hashes an address to determine its location in the multicast table
+  * hw - Struct containing variables accessed by shared code
+  * mc_addr - the multicast address to hash
+  *
+  * atl1_hash_mc_addr
+  *  purpose
+  *      set hash value for a multicast address
+  *      hash calcu processing :
+  *          1. calcu 32bit CRC for multicast address
+  *          2. reverse crc with MSB to LSB
+  */
+ u32 atl1_hash_mc_addr(struct atl1_hw *hw, u8 *mc_addr)
+ {
+       u32 crc32, value = 0;
+       int i;
+       crc32 = ether_crc_le(6, mc_addr);
+       for (i = 0; i < 32; i++)
+               value |= (((crc32 >> i) & 1) << (31 - i));
+       return value;
+ }
+ /*
+  * Sets the bit in the multicast table corresponding to the hash value.
+  * hw - Struct containing variables accessed by shared code
+  * hash_value - Multicast address hash value
+  */
+ void atl1_hash_set(struct atl1_hw *hw, u32 hash_value)
+ {
+       u32 hash_bit, hash_reg;
+       u32 mta;
+       /*
+        * The HASH Table  is a register array of 2 32-bit registers.
+        * It is treated like an array of 64 bits.  We want to set
+        * bit BitArray[hash_value]. So we figure out what register
+        * the bit is in, read it, OR in the new bit, then write
+        * back the new value.  The register is determined by the
+        * upper 7 bits of the hash value and the bit within that
+        * register are determined by the lower 5 bits of the value.
+        */
+       hash_reg = (hash_value >> 31) & 0x1;
+       hash_bit = (hash_value >> 26) & 0x1F;
+       mta = ioread32((hw->hw_addr + REG_RX_HASH_TABLE) + (hash_reg << 2));
+       mta |= (1 << hash_bit);
+       iowrite32(mta, (hw->hw_addr + REG_RX_HASH_TABLE) + (hash_reg << 2));
+ }
+ /*
+  * Writes a value to a PHY register
+  * hw - Struct containing variables accessed by shared code
+  * reg_addr - address of the PHY register to write
+  * data - data to write to the PHY
+  */
+ static s32 atl1_write_phy_reg(struct atl1_hw *hw, u32 reg_addr, u16 phy_data)
+ {
+       int i;
+       u32 val;
+       val = ((u32) (phy_data & MDIO_DATA_MASK)) << MDIO_DATA_SHIFT |
+           (reg_addr & MDIO_REG_ADDR_MASK) << MDIO_REG_ADDR_SHIFT |
+           MDIO_SUP_PREAMBLE |
+           MDIO_START | MDIO_CLK_25_4 << MDIO_CLK_SEL_SHIFT;
+       iowrite32(val, hw->hw_addr + REG_MDIO_CTRL);
+       ioread32(hw->hw_addr + REG_MDIO_CTRL);
+       for (i = 0; i < MDIO_WAIT_TIMES; i++) {
+               udelay(2);
+               val = ioread32(hw->hw_addr + REG_MDIO_CTRL);
+               if (!(val & (MDIO_START | MDIO_BUSY)))
+                       break;
+       }
+       if (!(val & (MDIO_START | MDIO_BUSY)))
+               return 0;
+       return ATLX_ERR_PHY;
+ }
+ /*
+  * Make L001's PHY out of Power Saving State (bug)
+  * hw - Struct containing variables accessed by shared code
+  * when power on, L001's PHY always on Power saving State
+  * (Gigabit Link forbidden)
+  */
+ static s32 atl1_phy_leave_power_saving(struct atl1_hw *hw)
+ {
+       s32 ret;
+       ret = atl1_write_phy_reg(hw, 29, 0x0029);
+       if (ret)
+               return ret;
+       return atl1_write_phy_reg(hw, 30, 0);
+ }
+ /*
+  *TODO: do something or get rid of this
+  */
++#ifdef CONFIG_PM
+ static s32 atl1_phy_enter_power_saving(struct atl1_hw *hw)
+ {
+ /*    s32 ret_val;
+  *    u16 phy_data;
+  */
+ /*
+     ret_val = atl1_write_phy_reg(hw, ...);
+     ret_val = atl1_write_phy_reg(hw, ...);
+     ....
+ */
+       return 0;
+ }
++#endif
+ /*
+  * Resets the PHY and make all config validate
+  * hw - Struct containing variables accessed by shared code
+  *
+  * Sets bit 15 and 12 of the MII Control regiser (for F001 bug)
+  */
+ static s32 atl1_phy_reset(struct atl1_hw *hw)
+ {
+       struct pci_dev *pdev = hw->back->pdev;
+       struct atl1_adapter *adapter = hw->back;
+       s32 ret_val;
+       u16 phy_data;
+       if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+           hw->media_type == MEDIA_TYPE_1000M_FULL)
+               phy_data = MII_CR_RESET | MII_CR_AUTO_NEG_EN;
+       else {
+               switch (hw->media_type) {
+               case MEDIA_TYPE_100M_FULL:
+                       phy_data =
+                           MII_CR_FULL_DUPLEX | MII_CR_SPEED_100 |
+                           MII_CR_RESET;
+                       break;
+               case MEDIA_TYPE_100M_HALF:
+                       phy_data = MII_CR_SPEED_100 | MII_CR_RESET;
+                       break;
+               case MEDIA_TYPE_10M_FULL:
+                       phy_data =
+                           MII_CR_FULL_DUPLEX | MII_CR_SPEED_10 | MII_CR_RESET;
+                       break;
+               default:
+                       /* MEDIA_TYPE_10M_HALF: */
+                       phy_data = MII_CR_SPEED_10 | MII_CR_RESET;
+                       break;
+               }
+       }
+       ret_val = atl1_write_phy_reg(hw, MII_BMCR, phy_data);
+       if (ret_val) {
+               u32 val;
+               int i;
+               /* pcie serdes link may be down! */
+               if (netif_msg_hw(adapter))
+                       dev_dbg(&pdev->dev, "pcie phy link down\n");
+               for (i = 0; i < 25; i++) {
+                       msleep(1);
+                       val = ioread32(hw->hw_addr + REG_MDIO_CTRL);
+                       if (!(val & (MDIO_START | MDIO_BUSY)))
+                               break;
+               }
+               if ((val & (MDIO_START | MDIO_BUSY)) != 0) {
+                       if (netif_msg_hw(adapter))
+                               dev_warn(&pdev->dev,
+                                       "pcie link down at least 25ms\n");
+                       return ret_val;
+               }
+       }
+       return 0;
+ }
+ /*
+  * Configures PHY autoneg and flow control advertisement settings
+  * hw - Struct containing variables accessed by shared code
+  */
+ static s32 atl1_phy_setup_autoneg_adv(struct atl1_hw *hw)
+ {
+       s32 ret_val;
+       s16 mii_autoneg_adv_reg;
+       s16 mii_1000t_ctrl_reg;
+       /* Read the MII Auto-Neg Advertisement Register (Address 4). */
+       mii_autoneg_adv_reg = MII_AR_DEFAULT_CAP_MASK;
+       /* Read the MII 1000Base-T Control Register (Address 9). */
+       mii_1000t_ctrl_reg = MII_ATLX_CR_1000T_DEFAULT_CAP_MASK;
+       /*
+        * First we clear all the 10/100 mb speed bits in the Auto-Neg
+        * Advertisement Register (Address 4) and the 1000 mb speed bits in
+        * the  1000Base-T Control Register (Address 9).
+        */
+       mii_autoneg_adv_reg &= ~MII_AR_SPEED_MASK;
+       mii_1000t_ctrl_reg &= ~MII_ATLX_CR_1000T_SPEED_MASK;
+       /*
+        * Need to parse media_type  and set up
+        * the appropriate PHY registers.
+        */
+       switch (hw->media_type) {
+       case MEDIA_TYPE_AUTO_SENSOR:
+               mii_autoneg_adv_reg |= (MII_AR_10T_HD_CAPS |
+                                       MII_AR_10T_FD_CAPS |
+                                       MII_AR_100TX_HD_CAPS |
+                                       MII_AR_100TX_FD_CAPS);
+               mii_1000t_ctrl_reg |= MII_ATLX_CR_1000T_FD_CAPS;
+               break;
+       case MEDIA_TYPE_1000M_FULL:
+               mii_1000t_ctrl_reg |= MII_ATLX_CR_1000T_FD_CAPS;
+               break;
+       case MEDIA_TYPE_100M_FULL:
+               mii_autoneg_adv_reg |= MII_AR_100TX_FD_CAPS;
+               break;
+       case MEDIA_TYPE_100M_HALF:
+               mii_autoneg_adv_reg |= MII_AR_100TX_HD_CAPS;
+               break;
+       case MEDIA_TYPE_10M_FULL:
+               mii_autoneg_adv_reg |= MII_AR_10T_FD_CAPS;
+               break;
+       default:
+               mii_autoneg_adv_reg |= MII_AR_10T_HD_CAPS;
+               break;
+       }
+       /* flow control fixed to enable all */
+       mii_autoneg_adv_reg |= (MII_AR_ASM_DIR | MII_AR_PAUSE);
+       hw->mii_autoneg_adv_reg = mii_autoneg_adv_reg;
+       hw->mii_1000t_ctrl_reg = mii_1000t_ctrl_reg;
+       ret_val = atl1_write_phy_reg(hw, MII_ADVERTISE, mii_autoneg_adv_reg);
+       if (ret_val)
+               return ret_val;
+       ret_val = atl1_write_phy_reg(hw, MII_ATLX_CR, mii_1000t_ctrl_reg);
+       if (ret_val)
+               return ret_val;
+       return 0;
+ }
+ /*
+  * Configures link settings.
+  * hw - Struct containing variables accessed by shared code
+  * Assumes the hardware has previously been reset and the
+  * transmitter and receiver are not enabled.
+  */
+ static s32 atl1_setup_link(struct atl1_hw *hw)
+ {
+       struct pci_dev *pdev = hw->back->pdev;
+       struct atl1_adapter *adapter = hw->back;
+       s32 ret_val;
+       /*
+        * Options:
+        *  PHY will advertise value(s) parsed from
+        *  autoneg_advertised and fc
+        *  no matter what autoneg is , We will not wait link result.
+        */
+       ret_val = atl1_phy_setup_autoneg_adv(hw);
+       if (ret_val) {
+               if (netif_msg_link(adapter))
+                       dev_dbg(&pdev->dev,
+                               "error setting up autonegotiation\n");
+               return ret_val;
+       }
+       /* SW.Reset , En-Auto-Neg if needed */
+       ret_val = atl1_phy_reset(hw);
+       if (ret_val) {
+               if (netif_msg_link(adapter))
+                       dev_dbg(&pdev->dev, "error resetting phy\n");
+               return ret_val;
+       }
+       hw->phy_configured = true;
+       return ret_val;
+ }
+ static void atl1_init_flash_opcode(struct atl1_hw *hw)
+ {
+       if (hw->flash_vendor >= ARRAY_SIZE(flash_table))
+               /* Atmel */
+               hw->flash_vendor = 0;
+       /* Init OP table */
+       iowrite8(flash_table[hw->flash_vendor].cmd_program,
+               hw->hw_addr + REG_SPI_FLASH_OP_PROGRAM);
+       iowrite8(flash_table[hw->flash_vendor].cmd_sector_erase,
+               hw->hw_addr + REG_SPI_FLASH_OP_SC_ERASE);
+       iowrite8(flash_table[hw->flash_vendor].cmd_chip_erase,
+               hw->hw_addr + REG_SPI_FLASH_OP_CHIP_ERASE);
+       iowrite8(flash_table[hw->flash_vendor].cmd_rdid,
+               hw->hw_addr + REG_SPI_FLASH_OP_RDID);
+       iowrite8(flash_table[hw->flash_vendor].cmd_wren,
+               hw->hw_addr + REG_SPI_FLASH_OP_WREN);
+       iowrite8(flash_table[hw->flash_vendor].cmd_rdsr,
+               hw->hw_addr + REG_SPI_FLASH_OP_RDSR);
+       iowrite8(flash_table[hw->flash_vendor].cmd_wrsr,
+               hw->hw_addr + REG_SPI_FLASH_OP_WRSR);
+       iowrite8(flash_table[hw->flash_vendor].cmd_read,
+               hw->hw_addr + REG_SPI_FLASH_OP_READ);
+ }
+ /*
+  * Performs basic configuration of the adapter.
+  * hw - Struct containing variables accessed by shared code
+  * Assumes that the controller has previously been reset and is in a
+  * post-reset uninitialized state. Initializes multicast table,
+  * and  Calls routines to setup link
+  * Leaves the transmit and receive units disabled and uninitialized.
+  */
+ static s32 atl1_init_hw(struct atl1_hw *hw)
+ {
+       u32 ret_val = 0;
+       /* Zero out the Multicast HASH table */
+       iowrite32(0, hw->hw_addr + REG_RX_HASH_TABLE);
+       /* clear the old settings from the multicast hash table */
+       iowrite32(0, (hw->hw_addr + REG_RX_HASH_TABLE) + (1 << 2));
+       atl1_init_flash_opcode(hw);
+       if (!hw->phy_configured) {
+               /* enable GPHY LinkChange Interrrupt */
+               ret_val = atl1_write_phy_reg(hw, 18, 0xC00);
+               if (ret_val)
+                       return ret_val;
+               /* make PHY out of power-saving state */
+               ret_val = atl1_phy_leave_power_saving(hw);
+               if (ret_val)
+                       return ret_val;
+               /* Call a subroutine to configure the link */
+               ret_val = atl1_setup_link(hw);
+       }
+       return ret_val;
+ }
+ /*
+  * Detects the current speed and duplex settings of the hardware.
+  * hw - Struct containing variables accessed by shared code
+  * speed - Speed of the connection
+  * duplex - Duplex setting of the connection
+  */
+ static s32 atl1_get_speed_and_duplex(struct atl1_hw *hw, u16 *speed, u16 *duplex)
+ {
+       struct pci_dev *pdev = hw->back->pdev;
+       struct atl1_adapter *adapter = hw->back;
+       s32 ret_val;
+       u16 phy_data;
+       /* ; --- Read   PHY Specific Status Register (17) */
+       ret_val = atl1_read_phy_reg(hw, MII_ATLX_PSSR, &phy_data);
+       if (ret_val)
+               return ret_val;
+       if (!(phy_data & MII_ATLX_PSSR_SPD_DPLX_RESOLVED))
+               return ATLX_ERR_PHY_RES;
+       switch (phy_data & MII_ATLX_PSSR_SPEED) {
+       case MII_ATLX_PSSR_1000MBS:
+               *speed = SPEED_1000;
+               break;
+       case MII_ATLX_PSSR_100MBS:
+               *speed = SPEED_100;
+               break;
+       case MII_ATLX_PSSR_10MBS:
+               *speed = SPEED_10;
+               break;
+       default:
+               if (netif_msg_hw(adapter))
+                       dev_dbg(&pdev->dev, "error getting speed\n");
+               return ATLX_ERR_PHY_SPEED;
+               break;
+       }
+       if (phy_data & MII_ATLX_PSSR_DPLX)
+               *duplex = FULL_DUPLEX;
+       else
+               *duplex = HALF_DUPLEX;
+       return 0;
+ }
+ void atl1_set_mac_addr(struct atl1_hw *hw)
+ {
+       u32 value;
+       /*
+        * 00-0B-6A-F6-00-DC
+        * 0:  6AF600DC   1: 000B
+        * low dword
+        */
+       value = (((u32) hw->mac_addr[2]) << 24) |
+           (((u32) hw->mac_addr[3]) << 16) |
+           (((u32) hw->mac_addr[4]) << 8) | (((u32) hw->mac_addr[5]));
+       iowrite32(value, hw->hw_addr + REG_MAC_STA_ADDR);
+       /* high dword */
+       value = (((u32) hw->mac_addr[0]) << 8) | (((u32) hw->mac_addr[1]));
+       iowrite32(value, (hw->hw_addr + REG_MAC_STA_ADDR) + (1 << 2));
+ }
+ /*
+  * atl1_sw_init - Initialize general software structures (struct atl1_adapter)
+  * @adapter: board private structure to initialize
+  *
+  * atl1_sw_init initializes the Adapter private data structure.
+  * Fields are initialized based on PCI device information and
+  * OS network device settings (MTU size).
+  */
+ static int __devinit atl1_sw_init(struct atl1_adapter *adapter)
+ {
+       struct atl1_hw *hw = &adapter->hw;
+       struct net_device *netdev = adapter->netdev;
+       hw->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
+       hw->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
+       adapter->wol = 0;
+       adapter->rx_buffer_len = (hw->max_frame_size + 7) & ~7;
+       adapter->ict = 50000;           /* 100ms */
+       adapter->link_speed = SPEED_0;  /* hardware init */
+       adapter->link_duplex = FULL_DUPLEX;
+       hw->phy_configured = false;
+       hw->preamble_len = 7;
+       hw->ipgt = 0x60;
+       hw->min_ifg = 0x50;
+       hw->ipgr1 = 0x40;
+       hw->ipgr2 = 0x60;
+       hw->max_retry = 0xf;
+       hw->lcol = 0x37;
+       hw->jam_ipg = 7;
+       hw->rfd_burst = 8;
+       hw->rrd_burst = 8;
+       hw->rfd_fetch_gap = 1;
+       hw->rx_jumbo_th = adapter->rx_buffer_len / 8;
+       hw->rx_jumbo_lkah = 1;
+       hw->rrd_ret_timer = 16;
+       hw->tpd_burst = 4;
+       hw->tpd_fetch_th = 16;
+       hw->txf_burst = 0x100;
+       hw->tx_jumbo_task_th = (hw->max_frame_size + 7) >> 3;
+       hw->tpd_fetch_gap = 1;
+       hw->rcb_value = atl1_rcb_64;
+       hw->dma_ord = atl1_dma_ord_enh;
+       hw->dmar_block = atl1_dma_req_256;
+       hw->dmaw_block = atl1_dma_req_256;
+       hw->cmb_rrd = 4;
+       hw->cmb_tpd = 4;
+       hw->cmb_rx_timer = 1;   /* about 2us */
+       hw->cmb_tx_timer = 1;   /* about 2us */
+       hw->smb_timer = 100000; /* about 200ms */
+       spin_lock_init(&adapter->lock);
+       spin_lock_init(&adapter->mb_lock);
+       return 0;
+ }
+ static int mdio_read(struct net_device *netdev, int phy_id, int reg_num)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       u16 result;
+       atl1_read_phy_reg(&adapter->hw, reg_num & 0x1f, &result);
+       return result;
+ }
+ static void mdio_write(struct net_device *netdev, int phy_id, int reg_num,
+       int val)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       atl1_write_phy_reg(&adapter->hw, reg_num, val);
+ }
+ /*
+  * atl1_mii_ioctl -
+  * @netdev:
+  * @ifreq:
+  * @cmd:
+  */
+ static int atl1_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       unsigned long flags;
+       int retval;
+       if (!netif_running(netdev))
+               return -EINVAL;
+       spin_lock_irqsave(&adapter->lock, flags);
+       retval = generic_mii_ioctl(&adapter->mii, if_mii(ifr), cmd, NULL);
+       spin_unlock_irqrestore(&adapter->lock, flags);
+       return retval;
+ }
+ /*
+  * atl1_setup_mem_resources - allocate Tx / RX descriptor resources
+  * @adapter: board private structure
+  *
+  * Return 0 on success, negative on failure
+  */
+ static s32 atl1_setup_ring_resources(struct atl1_adapter *adapter)
+ {
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       struct atl1_rrd_ring *rrd_ring = &adapter->rrd_ring;
+       struct atl1_ring_header *ring_header = &adapter->ring_header;
+       struct pci_dev *pdev = adapter->pdev;
+       int size;
+       u8 offset = 0;
+       size = sizeof(struct atl1_buffer) * (tpd_ring->count + rfd_ring->count);
+       tpd_ring->buffer_info = kzalloc(size, GFP_KERNEL);
+       if (unlikely(!tpd_ring->buffer_info)) {
+               if (netif_msg_drv(adapter))
+                       dev_err(&pdev->dev, "kzalloc failed , size = D%d\n",
+                               size);
+               goto err_nomem;
+       }
+       rfd_ring->buffer_info =
+               (struct atl1_buffer *)(tpd_ring->buffer_info + tpd_ring->count);
+       /*
+        * real ring DMA buffer
+        * each ring/block may need up to 8 bytes for alignment, hence the
+        * additional 40 bytes tacked onto the end.
+        */
+       ring_header->size = size =
+               sizeof(struct tx_packet_desc) * tpd_ring->count
+               + sizeof(struct rx_free_desc) * rfd_ring->count
+               + sizeof(struct rx_return_desc) * rrd_ring->count
+               + sizeof(struct coals_msg_block)
+               + sizeof(struct stats_msg_block)
+               + 40;
+       ring_header->desc = pci_alloc_consistent(pdev, ring_header->size,
+               &ring_header->dma);
+       if (unlikely(!ring_header->desc)) {
+               if (netif_msg_drv(adapter))
+                       dev_err(&pdev->dev, "pci_alloc_consistent failed\n");
+               goto err_nomem;
+       }
+       memset(ring_header->desc, 0, ring_header->size);
+       /* init TPD ring */
+       tpd_ring->dma = ring_header->dma;
+       offset = (tpd_ring->dma & 0x7) ? (8 - (ring_header->dma & 0x7)) : 0;
+       tpd_ring->dma += offset;
+       tpd_ring->desc = (u8 *) ring_header->desc + offset;
+       tpd_ring->size = sizeof(struct tx_packet_desc) * tpd_ring->count;
+       /* init RFD ring */
+       rfd_ring->dma = tpd_ring->dma + tpd_ring->size;
+       offset = (rfd_ring->dma & 0x7) ? (8 - (rfd_ring->dma & 0x7)) : 0;
+       rfd_ring->dma += offset;
+       rfd_ring->desc = (u8 *) tpd_ring->desc + (tpd_ring->size + offset);
+       rfd_ring->size = sizeof(struct rx_free_desc) * rfd_ring->count;
+       /* init RRD ring */
+       rrd_ring->dma = rfd_ring->dma + rfd_ring->size;
+       offset = (rrd_ring->dma & 0x7) ? (8 - (rrd_ring->dma & 0x7)) : 0;
+       rrd_ring->dma += offset;
+       rrd_ring->desc = (u8 *) rfd_ring->desc + (rfd_ring->size + offset);
+       rrd_ring->size = sizeof(struct rx_return_desc) * rrd_ring->count;
+       /* init CMB */
+       adapter->cmb.dma = rrd_ring->dma + rrd_ring->size;
+       offset = (adapter->cmb.dma & 0x7) ? (8 - (adapter->cmb.dma & 0x7)) : 0;
+       adapter->cmb.dma += offset;
+       adapter->cmb.cmb = (struct coals_msg_block *)
+               ((u8 *) rrd_ring->desc + (rrd_ring->size + offset));
+       /* init SMB */
+       adapter->smb.dma = adapter->cmb.dma + sizeof(struct coals_msg_block);
+       offset = (adapter->smb.dma & 0x7) ? (8 - (adapter->smb.dma & 0x7)) : 0;
+       adapter->smb.dma += offset;
+       adapter->smb.smb = (struct stats_msg_block *)
+               ((u8 *) adapter->cmb.cmb +
+               (sizeof(struct coals_msg_block) + offset));
+       return 0;
+ err_nomem:
+       kfree(tpd_ring->buffer_info);
+       return -ENOMEM;
+ }
+ static void atl1_init_ring_ptrs(struct atl1_adapter *adapter)
+ {
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       struct atl1_rrd_ring *rrd_ring = &adapter->rrd_ring;
+       atomic_set(&tpd_ring->next_to_use, 0);
+       atomic_set(&tpd_ring->next_to_clean, 0);
+       rfd_ring->next_to_clean = 0;
+       atomic_set(&rfd_ring->next_to_use, 0);
+       rrd_ring->next_to_use = 0;
+       atomic_set(&rrd_ring->next_to_clean, 0);
+ }
+ /*
+  * atl1_clean_rx_ring - Free RFD Buffers
+  * @adapter: board private structure
+  */
+ static void atl1_clean_rx_ring(struct atl1_adapter *adapter)
+ {
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       struct atl1_rrd_ring *rrd_ring = &adapter->rrd_ring;
+       struct atl1_buffer *buffer_info;
+       struct pci_dev *pdev = adapter->pdev;
+       unsigned long size;
+       unsigned int i;
+       /* Free all the Rx ring sk_buffs */
+       for (i = 0; i < rfd_ring->count; i++) {
+               buffer_info = &rfd_ring->buffer_info[i];
+               if (buffer_info->dma) {
+                       pci_unmap_page(pdev, buffer_info->dma,
+                               buffer_info->length, PCI_DMA_FROMDEVICE);
+                       buffer_info->dma = 0;
+               }
+               if (buffer_info->skb) {
+                       dev_kfree_skb(buffer_info->skb);
+                       buffer_info->skb = NULL;
+               }
+       }
+       size = sizeof(struct atl1_buffer) * rfd_ring->count;
+       memset(rfd_ring->buffer_info, 0, size);
+       /* Zero out the descriptor ring */
+       memset(rfd_ring->desc, 0, rfd_ring->size);
+       rfd_ring->next_to_clean = 0;
+       atomic_set(&rfd_ring->next_to_use, 0);
+       rrd_ring->next_to_use = 0;
+       atomic_set(&rrd_ring->next_to_clean, 0);
+ }
+ /*
+  * atl1_clean_tx_ring - Free Tx Buffers
+  * @adapter: board private structure
+  */
+ static void atl1_clean_tx_ring(struct atl1_adapter *adapter)
+ {
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_buffer *buffer_info;
+       struct pci_dev *pdev = adapter->pdev;
+       unsigned long size;
+       unsigned int i;
+       /* Free all the Tx ring sk_buffs */
+       for (i = 0; i < tpd_ring->count; i++) {
+               buffer_info = &tpd_ring->buffer_info[i];
+               if (buffer_info->dma) {
+                       pci_unmap_page(pdev, buffer_info->dma,
+                               buffer_info->length, PCI_DMA_TODEVICE);
+                       buffer_info->dma = 0;
+               }
+       }
+       for (i = 0; i < tpd_ring->count; i++) {
+               buffer_info = &tpd_ring->buffer_info[i];
+               if (buffer_info->skb) {
+                       dev_kfree_skb_any(buffer_info->skb);
+                       buffer_info->skb = NULL;
+               }
+       }
+       size = sizeof(struct atl1_buffer) * tpd_ring->count;
+       memset(tpd_ring->buffer_info, 0, size);
+       /* Zero out the descriptor ring */
+       memset(tpd_ring->desc, 0, tpd_ring->size);
+       atomic_set(&tpd_ring->next_to_use, 0);
+       atomic_set(&tpd_ring->next_to_clean, 0);
+ }
+ /*
+  * atl1_free_ring_resources - Free Tx / RX descriptor Resources
+  * @adapter: board private structure
+  *
+  * Free all transmit software resources
+  */
+ static void atl1_free_ring_resources(struct atl1_adapter *adapter)
+ {
+       struct pci_dev *pdev = adapter->pdev;
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       struct atl1_rrd_ring *rrd_ring = &adapter->rrd_ring;
+       struct atl1_ring_header *ring_header = &adapter->ring_header;
+       atl1_clean_tx_ring(adapter);
+       atl1_clean_rx_ring(adapter);
+       kfree(tpd_ring->buffer_info);
+       pci_free_consistent(pdev, ring_header->size, ring_header->desc,
+               ring_header->dma);
+       tpd_ring->buffer_info = NULL;
+       tpd_ring->desc = NULL;
+       tpd_ring->dma = 0;
+       rfd_ring->buffer_info = NULL;
+       rfd_ring->desc = NULL;
+       rfd_ring->dma = 0;
+       rrd_ring->desc = NULL;
+       rrd_ring->dma = 0;
+ }
+ static void atl1_setup_mac_ctrl(struct atl1_adapter *adapter)
+ {
+       u32 value;
+       struct atl1_hw *hw = &adapter->hw;
+       struct net_device *netdev = adapter->netdev;
+       /* Config MAC CTRL Register */
+       value = MAC_CTRL_TX_EN | MAC_CTRL_RX_EN;
+       /* duplex */
+       if (FULL_DUPLEX == adapter->link_duplex)
+               value |= MAC_CTRL_DUPLX;
+       /* speed */
+       value |= ((u32) ((SPEED_1000 == adapter->link_speed) ?
+                        MAC_CTRL_SPEED_1000 : MAC_CTRL_SPEED_10_100) <<
+                 MAC_CTRL_SPEED_SHIFT);
+       /* flow control */
+       value |= (MAC_CTRL_TX_FLOW | MAC_CTRL_RX_FLOW);
+       /* PAD & CRC */
+       value |= (MAC_CTRL_ADD_CRC | MAC_CTRL_PAD);
+       /* preamble length */
+       value |= (((u32) adapter->hw.preamble_len
+                  & MAC_CTRL_PRMLEN_MASK) << MAC_CTRL_PRMLEN_SHIFT);
+       /* vlan */
+       if (adapter->vlgrp)
+               value |= MAC_CTRL_RMV_VLAN;
+       /* rx checksum
+          if (adapter->rx_csum)
+          value |= MAC_CTRL_RX_CHKSUM_EN;
+        */
+       /* filter mode */
+       value |= MAC_CTRL_BC_EN;
+       if (netdev->flags & IFF_PROMISC)
+               value |= MAC_CTRL_PROMIS_EN;
+       else if (netdev->flags & IFF_ALLMULTI)
+               value |= MAC_CTRL_MC_ALL_EN;
+       /* value |= MAC_CTRL_LOOPBACK; */
+       iowrite32(value, hw->hw_addr + REG_MAC_CTRL);
+ }
+ static u32 atl1_check_link(struct atl1_adapter *adapter)
+ {
+       struct atl1_hw *hw = &adapter->hw;
+       struct net_device *netdev = adapter->netdev;
+       u32 ret_val;
+       u16 speed, duplex, phy_data;
+       int reconfig = 0;
+       /* MII_BMSR must read twice */
+       atl1_read_phy_reg(hw, MII_BMSR, &phy_data);
+       atl1_read_phy_reg(hw, MII_BMSR, &phy_data);
+       if (!(phy_data & BMSR_LSTATUS)) {
+               /* link down */
+               if (netif_carrier_ok(netdev)) {
+                       /* old link state: Up */
+                       if (netif_msg_link(adapter))
+                               dev_info(&adapter->pdev->dev, "link is down\n");
+                       adapter->link_speed = SPEED_0;
+                       netif_carrier_off(netdev);
+                       netif_stop_queue(netdev);
+               }
+               return 0;
+       }
+       /* Link Up */
+       ret_val = atl1_get_speed_and_duplex(hw, &speed, &duplex);
+       if (ret_val)
+               return ret_val;
+       switch (hw->media_type) {
+       case MEDIA_TYPE_1000M_FULL:
+               if (speed != SPEED_1000 || duplex != FULL_DUPLEX)
+                       reconfig = 1;
+               break;
+       case MEDIA_TYPE_100M_FULL:
+               if (speed != SPEED_100 || duplex != FULL_DUPLEX)
+                       reconfig = 1;
+               break;
+       case MEDIA_TYPE_100M_HALF:
+               if (speed != SPEED_100 || duplex != HALF_DUPLEX)
+                       reconfig = 1;
+               break;
+       case MEDIA_TYPE_10M_FULL:
+               if (speed != SPEED_10 || duplex != FULL_DUPLEX)
+                       reconfig = 1;
+               break;
+       case MEDIA_TYPE_10M_HALF:
+               if (speed != SPEED_10 || duplex != HALF_DUPLEX)
+                       reconfig = 1;
+               break;
+       }
+       /* link result is our setting */
+       if (!reconfig) {
+               if (adapter->link_speed != speed
+                   || adapter->link_duplex != duplex) {
+                       adapter->link_speed = speed;
+                       adapter->link_duplex = duplex;
+                       atl1_setup_mac_ctrl(adapter);
+                       if (netif_msg_link(adapter))
+                               dev_info(&adapter->pdev->dev,
+                                       "%s link is up %d Mbps %s\n",
+                                       netdev->name, adapter->link_speed,
+                                       adapter->link_duplex == FULL_DUPLEX ?
+                                       "full duplex" : "half duplex");
+               }
+               if (!netif_carrier_ok(netdev)) {
+                       /* Link down -> Up */
+                       netif_carrier_on(netdev);
+                       netif_wake_queue(netdev);
+               }
+               return 0;
+       }
+       /* change original link status */
+       if (netif_carrier_ok(netdev)) {
+               adapter->link_speed = SPEED_0;
+               netif_carrier_off(netdev);
+               netif_stop_queue(netdev);
+       }
+       if (hw->media_type != MEDIA_TYPE_AUTO_SENSOR &&
+           hw->media_type != MEDIA_TYPE_1000M_FULL) {
+               switch (hw->media_type) {
+               case MEDIA_TYPE_100M_FULL:
+                       phy_data = MII_CR_FULL_DUPLEX | MII_CR_SPEED_100 |
+                                  MII_CR_RESET;
+                       break;
+               case MEDIA_TYPE_100M_HALF:
+                       phy_data = MII_CR_SPEED_100 | MII_CR_RESET;
+                       break;
+               case MEDIA_TYPE_10M_FULL:
+                       phy_data =
+                           MII_CR_FULL_DUPLEX | MII_CR_SPEED_10 | MII_CR_RESET;
+                       break;
+               default:
+                       /* MEDIA_TYPE_10M_HALF: */
+                       phy_data = MII_CR_SPEED_10 | MII_CR_RESET;
+                       break;
+               }
+               atl1_write_phy_reg(hw, MII_BMCR, phy_data);
+               return 0;
+       }
+       /* auto-neg, insert timer to re-config phy */
+       if (!adapter->phy_timer_pending) {
+               adapter->phy_timer_pending = true;
+               mod_timer(&adapter->phy_config_timer, jiffies + 3 * HZ);
+       }
+       return 0;
+ }
+ static void set_flow_ctrl_old(struct atl1_adapter *adapter)
+ {
+       u32 hi, lo, value;
+       /* RFD Flow Control */
+       value = adapter->rfd_ring.count;
+       hi = value / 16;
+       if (hi < 2)
+               hi = 2;
+       lo = value * 7 / 8;
+       value = ((hi & RXQ_RXF_PAUSE_TH_HI_MASK) << RXQ_RXF_PAUSE_TH_HI_SHIFT) |
+               ((lo & RXQ_RXF_PAUSE_TH_LO_MASK) << RXQ_RXF_PAUSE_TH_LO_SHIFT);
+       iowrite32(value, adapter->hw.hw_addr + REG_RXQ_RXF_PAUSE_THRESH);
+       /* RRD Flow Control */
+       value = adapter->rrd_ring.count;
+       lo = value / 16;
+       hi = value * 7 / 8;
+       if (lo < 2)
+               lo = 2;
+       value = ((hi & RXQ_RRD_PAUSE_TH_HI_MASK) << RXQ_RRD_PAUSE_TH_HI_SHIFT) |
+               ((lo & RXQ_RRD_PAUSE_TH_LO_MASK) << RXQ_RRD_PAUSE_TH_LO_SHIFT);
+       iowrite32(value, adapter->hw.hw_addr + REG_RXQ_RRD_PAUSE_THRESH);
+ }
+ static void set_flow_ctrl_new(struct atl1_hw *hw)
+ {
+       u32 hi, lo, value;
+       /* RXF Flow Control */
+       value = ioread32(hw->hw_addr + REG_SRAM_RXF_LEN);
+       lo = value / 16;
+       if (lo < 192)
+               lo = 192;
+       hi = value * 7 / 8;
+       if (hi < lo)
+               hi = lo + 16;
+       value = ((hi & RXQ_RXF_PAUSE_TH_HI_MASK) << RXQ_RXF_PAUSE_TH_HI_SHIFT) |
+               ((lo & RXQ_RXF_PAUSE_TH_LO_MASK) << RXQ_RXF_PAUSE_TH_LO_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_RXQ_RXF_PAUSE_THRESH);
+       /* RRD Flow Control */
+       value = ioread32(hw->hw_addr + REG_SRAM_RRD_LEN);
+       lo = value / 8;
+       hi = value * 7 / 8;
+       if (lo < 2)
+               lo = 2;
+       if (hi < lo)
+               hi = lo + 3;
+       value = ((hi & RXQ_RRD_PAUSE_TH_HI_MASK) << RXQ_RRD_PAUSE_TH_HI_SHIFT) |
+               ((lo & RXQ_RRD_PAUSE_TH_LO_MASK) << RXQ_RRD_PAUSE_TH_LO_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_RXQ_RRD_PAUSE_THRESH);
+ }
+ /*
+  * atl1_configure - Configure Transmit&Receive Unit after Reset
+  * @adapter: board private structure
+  *
+  * Configure the Tx /Rx unit of the MAC after a reset.
+  */
+ static u32 atl1_configure(struct atl1_adapter *adapter)
+ {
+       struct atl1_hw *hw = &adapter->hw;
+       u32 value;
+       /* clear interrupt status */
+       iowrite32(0xffffffff, adapter->hw.hw_addr + REG_ISR);
+       /* set MAC Address */
+       value = (((u32) hw->mac_addr[2]) << 24) |
+               (((u32) hw->mac_addr[3]) << 16) |
+               (((u32) hw->mac_addr[4]) << 8) |
+               (((u32) hw->mac_addr[5]));
+       iowrite32(value, hw->hw_addr + REG_MAC_STA_ADDR);
+       value = (((u32) hw->mac_addr[0]) << 8) | (((u32) hw->mac_addr[1]));
+       iowrite32(value, hw->hw_addr + (REG_MAC_STA_ADDR + 4));
+       /* tx / rx ring */
+       /* HI base address */
+       iowrite32((u32) ((adapter->tpd_ring.dma & 0xffffffff00000000ULL) >> 32),
+               hw->hw_addr + REG_DESC_BASE_ADDR_HI);
+       /* LO base address */
+       iowrite32((u32) (adapter->rfd_ring.dma & 0x00000000ffffffffULL),
+               hw->hw_addr + REG_DESC_RFD_ADDR_LO);
+       iowrite32((u32) (adapter->rrd_ring.dma & 0x00000000ffffffffULL),
+               hw->hw_addr + REG_DESC_RRD_ADDR_LO);
+       iowrite32((u32) (adapter->tpd_ring.dma & 0x00000000ffffffffULL),
+               hw->hw_addr + REG_DESC_TPD_ADDR_LO);
+       iowrite32((u32) (adapter->cmb.dma & 0x00000000ffffffffULL),
+               hw->hw_addr + REG_DESC_CMB_ADDR_LO);
+       iowrite32((u32) (adapter->smb.dma & 0x00000000ffffffffULL),
+               hw->hw_addr + REG_DESC_SMB_ADDR_LO);
+       /* element count */
+       value = adapter->rrd_ring.count;
+       value <<= 16;
+       value += adapter->rfd_ring.count;
+       iowrite32(value, hw->hw_addr + REG_DESC_RFD_RRD_RING_SIZE);
+       iowrite32(adapter->tpd_ring.count, hw->hw_addr +
+               REG_DESC_TPD_RING_SIZE);
+       /* Load Ptr */
+       iowrite32(1, hw->hw_addr + REG_LOAD_PTR);
+       /* config Mailbox */
+       value = ((atomic_read(&adapter->tpd_ring.next_to_use)
+                 & MB_TPD_PROD_INDX_MASK) << MB_TPD_PROD_INDX_SHIFT) |
+               ((atomic_read(&adapter->rrd_ring.next_to_clean)
+               & MB_RRD_CONS_INDX_MASK) << MB_RRD_CONS_INDX_SHIFT) |
+               ((atomic_read(&adapter->rfd_ring.next_to_use)
+               & MB_RFD_PROD_INDX_MASK) << MB_RFD_PROD_INDX_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_MAILBOX);
+       /* config IPG/IFG */
+       value = (((u32) hw->ipgt & MAC_IPG_IFG_IPGT_MASK)
+                << MAC_IPG_IFG_IPGT_SHIFT) |
+               (((u32) hw->min_ifg & MAC_IPG_IFG_MIFG_MASK)
+               << MAC_IPG_IFG_MIFG_SHIFT) |
+               (((u32) hw->ipgr1 & MAC_IPG_IFG_IPGR1_MASK)
+               << MAC_IPG_IFG_IPGR1_SHIFT) |
+               (((u32) hw->ipgr2 & MAC_IPG_IFG_IPGR2_MASK)
+               << MAC_IPG_IFG_IPGR2_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_MAC_IPG_IFG);
+       /* config  Half-Duplex Control */
+       value = ((u32) hw->lcol & MAC_HALF_DUPLX_CTRL_LCOL_MASK) |
+               (((u32) hw->max_retry & MAC_HALF_DUPLX_CTRL_RETRY_MASK)
+               << MAC_HALF_DUPLX_CTRL_RETRY_SHIFT) |
+               MAC_HALF_DUPLX_CTRL_EXC_DEF_EN |
+               (0xa << MAC_HALF_DUPLX_CTRL_ABEBT_SHIFT) |
+               (((u32) hw->jam_ipg & MAC_HALF_DUPLX_CTRL_JAMIPG_MASK)
+               << MAC_HALF_DUPLX_CTRL_JAMIPG_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_MAC_HALF_DUPLX_CTRL);
+       /* set Interrupt Moderator Timer */
+       iowrite16(adapter->imt, hw->hw_addr + REG_IRQ_MODU_TIMER_INIT);
+       iowrite32(MASTER_CTRL_ITIMER_EN, hw->hw_addr + REG_MASTER_CTRL);
+       /* set Interrupt Clear Timer */
+       iowrite16(adapter->ict, hw->hw_addr + REG_CMBDISDMA_TIMER);
+       /* set max frame size hw will accept */
+       iowrite32(hw->max_frame_size, hw->hw_addr + REG_MTU);
+       /* jumbo size & rrd retirement timer */
+       value = (((u32) hw->rx_jumbo_th & RXQ_JMBOSZ_TH_MASK)
+                << RXQ_JMBOSZ_TH_SHIFT) |
+               (((u32) hw->rx_jumbo_lkah & RXQ_JMBO_LKAH_MASK)
+               << RXQ_JMBO_LKAH_SHIFT) |
+               (((u32) hw->rrd_ret_timer & RXQ_RRD_TIMER_MASK)
+               << RXQ_RRD_TIMER_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_RXQ_JMBOSZ_RRDTIM);
+       /* Flow Control */
+       switch (hw->dev_rev) {
+       case 0x8001:
+       case 0x9001:
+       case 0x9002:
+       case 0x9003:
+               set_flow_ctrl_old(adapter);
+               break;
+       default:
+               set_flow_ctrl_new(hw);
+               break;
+       }
+       /* config TXQ */
+       value = (((u32) hw->tpd_burst & TXQ_CTRL_TPD_BURST_NUM_MASK)
+                << TXQ_CTRL_TPD_BURST_NUM_SHIFT) |
+               (((u32) hw->txf_burst & TXQ_CTRL_TXF_BURST_NUM_MASK)
+               << TXQ_CTRL_TXF_BURST_NUM_SHIFT) |
+               (((u32) hw->tpd_fetch_th & TXQ_CTRL_TPD_FETCH_TH_MASK)
+               << TXQ_CTRL_TPD_FETCH_TH_SHIFT) | TXQ_CTRL_ENH_MODE |
+               TXQ_CTRL_EN;
+       iowrite32(value, hw->hw_addr + REG_TXQ_CTRL);
+       /* min tpd fetch gap & tx jumbo packet size threshold for taskoffload */
+       value = (((u32) hw->tx_jumbo_task_th & TX_JUMBO_TASK_TH_MASK)
+               << TX_JUMBO_TASK_TH_SHIFT) |
+               (((u32) hw->tpd_fetch_gap & TX_TPD_MIN_IPG_MASK)
+               << TX_TPD_MIN_IPG_SHIFT);
+       iowrite32(value, hw->hw_addr + REG_TX_JUMBO_TASK_TH_TPD_IPG);
+       /* config RXQ */
+       value = (((u32) hw->rfd_burst & RXQ_CTRL_RFD_BURST_NUM_MASK)
+               << RXQ_CTRL_RFD_BURST_NUM_SHIFT) |
+               (((u32) hw->rrd_burst & RXQ_CTRL_RRD_BURST_THRESH_MASK)
+               << RXQ_CTRL_RRD_BURST_THRESH_SHIFT) |
+               (((u32) hw->rfd_fetch_gap & RXQ_CTRL_RFD_PREF_MIN_IPG_MASK)
+               << RXQ_CTRL_RFD_PREF_MIN_IPG_SHIFT) | RXQ_CTRL_CUT_THRU_EN |
+               RXQ_CTRL_EN;
+       iowrite32(value, hw->hw_addr + REG_RXQ_CTRL);
+       /* config DMA Engine */
+       value = ((((u32) hw->dmar_block) & DMA_CTRL_DMAR_BURST_LEN_MASK)
+               << DMA_CTRL_DMAR_BURST_LEN_SHIFT) |
+               ((((u32) hw->dmaw_block) & DMA_CTRL_DMAW_BURST_LEN_MASK)
+               << DMA_CTRL_DMAW_BURST_LEN_SHIFT) | DMA_CTRL_DMAR_EN |
+               DMA_CTRL_DMAW_EN;
+       value |= (u32) hw->dma_ord;
+       if (atl1_rcb_128 == hw->rcb_value)
+               value |= DMA_CTRL_RCB_VALUE;
+       iowrite32(value, hw->hw_addr + REG_DMA_CTRL);
+       /* config CMB / SMB */
+       value = (hw->cmb_tpd > adapter->tpd_ring.count) ?
+               hw->cmb_tpd : adapter->tpd_ring.count;
+       value <<= 16;
+       value |= hw->cmb_rrd;
+       iowrite32(value, hw->hw_addr + REG_CMB_WRITE_TH);
+       value = hw->cmb_rx_timer | ((u32) hw->cmb_tx_timer << 16);
+       iowrite32(value, hw->hw_addr + REG_CMB_WRITE_TIMER);
+       iowrite32(hw->smb_timer, hw->hw_addr + REG_SMB_TIMER);
+       /* --- enable CMB / SMB */
+       value = CSMB_CTRL_CMB_EN | CSMB_CTRL_SMB_EN;
+       iowrite32(value, hw->hw_addr + REG_CSMB_CTRL);
+       value = ioread32(adapter->hw.hw_addr + REG_ISR);
+       if (unlikely((value & ISR_PHY_LINKDOWN) != 0))
+               value = 1;      /* config failed */
+       else
+               value = 0;
+       /* clear all interrupt status */
+       iowrite32(0x3fffffff, adapter->hw.hw_addr + REG_ISR);
+       iowrite32(0, adapter->hw.hw_addr + REG_ISR);
+       return value;
+ }
+ /*
+  * atl1_pcie_patch - Patch for PCIE module
+  */
+ static void atl1_pcie_patch(struct atl1_adapter *adapter)
+ {
+       u32 value;
+       /* much vendor magic here */
+       value = 0x6500;
+       iowrite32(value, adapter->hw.hw_addr + 0x12FC);
+       /* pcie flow control mode change */
+       value = ioread32(adapter->hw.hw_addr + 0x1008);
+       value |= 0x8000;
+       iowrite32(value, adapter->hw.hw_addr + 0x1008);
+ }
+ /*
+  * When ACPI resume on some VIA MotherBoard, the Interrupt Disable bit/0x400
+  * on PCI Command register is disable.
+  * The function enable this bit.
+  * Brackett, 2006/03/15
+  */
+ static void atl1_via_workaround(struct atl1_adapter *adapter)
+ {
+       unsigned long value;
+       value = ioread16(adapter->hw.hw_addr + PCI_COMMAND);
+       if (value & PCI_COMMAND_INTX_DISABLE)
+               value &= ~PCI_COMMAND_INTX_DISABLE;
+       iowrite32(value, adapter->hw.hw_addr + PCI_COMMAND);
+ }
+ static void atl1_inc_smb(struct atl1_adapter *adapter)
+ {
+       struct stats_msg_block *smb = adapter->smb.smb;
+       /* Fill out the OS statistics structure */
+       adapter->soft_stats.rx_packets += smb->rx_ok;
+       adapter->soft_stats.tx_packets += smb->tx_ok;
+       adapter->soft_stats.rx_bytes += smb->rx_byte_cnt;
+       adapter->soft_stats.tx_bytes += smb->tx_byte_cnt;
+       adapter->soft_stats.multicast += smb->rx_mcast;
+       adapter->soft_stats.collisions += (smb->tx_1_col + smb->tx_2_col * 2 +
+               smb->tx_late_col + smb->tx_abort_col * adapter->hw.max_retry);
+       /* Rx Errors */
+       adapter->soft_stats.rx_errors += (smb->rx_frag + smb->rx_fcs_err +
+               smb->rx_len_err + smb->rx_sz_ov + smb->rx_rxf_ov +
+               smb->rx_rrd_ov + smb->rx_align_err);
+       adapter->soft_stats.rx_fifo_errors += smb->rx_rxf_ov;
+       adapter->soft_stats.rx_length_errors += smb->rx_len_err;
+       adapter->soft_stats.rx_crc_errors += smb->rx_fcs_err;
+       adapter->soft_stats.rx_frame_errors += smb->rx_align_err;
+       adapter->soft_stats.rx_missed_errors += (smb->rx_rrd_ov +
+               smb->rx_rxf_ov);
+       adapter->soft_stats.rx_pause += smb->rx_pause;
+       adapter->soft_stats.rx_rrd_ov += smb->rx_rrd_ov;
+       adapter->soft_stats.rx_trunc += smb->rx_sz_ov;
+       /* Tx Errors */
+       adapter->soft_stats.tx_errors += (smb->tx_late_col +
+               smb->tx_abort_col + smb->tx_underrun + smb->tx_trunc);
+       adapter->soft_stats.tx_fifo_errors += smb->tx_underrun;
+       adapter->soft_stats.tx_aborted_errors += smb->tx_abort_col;
+       adapter->soft_stats.tx_window_errors += smb->tx_late_col;
+       adapter->soft_stats.excecol += smb->tx_abort_col;
+       adapter->soft_stats.deffer += smb->tx_defer;
+       adapter->soft_stats.scc += smb->tx_1_col;
+       adapter->soft_stats.mcc += smb->tx_2_col;
+       adapter->soft_stats.latecol += smb->tx_late_col;
+       adapter->soft_stats.tx_underun += smb->tx_underrun;
+       adapter->soft_stats.tx_trunc += smb->tx_trunc;
+       adapter->soft_stats.tx_pause += smb->tx_pause;
+       adapter->net_stats.rx_packets = adapter->soft_stats.rx_packets;
+       adapter->net_stats.tx_packets = adapter->soft_stats.tx_packets;
+       adapter->net_stats.rx_bytes = adapter->soft_stats.rx_bytes;
+       adapter->net_stats.tx_bytes = adapter->soft_stats.tx_bytes;
+       adapter->net_stats.multicast = adapter->soft_stats.multicast;
+       adapter->net_stats.collisions = adapter->soft_stats.collisions;
+       adapter->net_stats.rx_errors = adapter->soft_stats.rx_errors;
+       adapter->net_stats.rx_over_errors =
+               adapter->soft_stats.rx_missed_errors;
+       adapter->net_stats.rx_length_errors =
+               adapter->soft_stats.rx_length_errors;
+       adapter->net_stats.rx_crc_errors = adapter->soft_stats.rx_crc_errors;
+       adapter->net_stats.rx_frame_errors =
+               adapter->soft_stats.rx_frame_errors;
+       adapter->net_stats.rx_fifo_errors = adapter->soft_stats.rx_fifo_errors;
+       adapter->net_stats.rx_missed_errors =
+               adapter->soft_stats.rx_missed_errors;
+       adapter->net_stats.tx_errors = adapter->soft_stats.tx_errors;
+       adapter->net_stats.tx_fifo_errors = adapter->soft_stats.tx_fifo_errors;
+       adapter->net_stats.tx_aborted_errors =
+               adapter->soft_stats.tx_aborted_errors;
+       adapter->net_stats.tx_window_errors =
+               adapter->soft_stats.tx_window_errors;
+       adapter->net_stats.tx_carrier_errors =
+               adapter->soft_stats.tx_carrier_errors;
+ }
+ static void atl1_update_mailbox(struct atl1_adapter *adapter)
+ {
+       unsigned long flags;
+       u32 tpd_next_to_use;
+       u32 rfd_next_to_use;
+       u32 rrd_next_to_clean;
+       u32 value;
+       spin_lock_irqsave(&adapter->mb_lock, flags);
+       tpd_next_to_use = atomic_read(&adapter->tpd_ring.next_to_use);
+       rfd_next_to_use = atomic_read(&adapter->rfd_ring.next_to_use);
+       rrd_next_to_clean = atomic_read(&adapter->rrd_ring.next_to_clean);
+       value = ((rfd_next_to_use & MB_RFD_PROD_INDX_MASK) <<
+               MB_RFD_PROD_INDX_SHIFT) |
+               ((rrd_next_to_clean & MB_RRD_CONS_INDX_MASK) <<
+               MB_RRD_CONS_INDX_SHIFT) |
+               ((tpd_next_to_use & MB_TPD_PROD_INDX_MASK) <<
+               MB_TPD_PROD_INDX_SHIFT);
+       iowrite32(value, adapter->hw.hw_addr + REG_MAILBOX);
+       spin_unlock_irqrestore(&adapter->mb_lock, flags);
+ }
+ static void atl1_clean_alloc_flag(struct atl1_adapter *adapter,
+       struct rx_return_desc *rrd, u16 offset)
+ {
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       while (rfd_ring->next_to_clean != (rrd->buf_indx + offset)) {
+               rfd_ring->buffer_info[rfd_ring->next_to_clean].alloced = 0;
+               if (++rfd_ring->next_to_clean == rfd_ring->count) {
+                       rfd_ring->next_to_clean = 0;
+               }
+       }
+ }
+ static void atl1_update_rfd_index(struct atl1_adapter *adapter,
+       struct rx_return_desc *rrd)
+ {
+       u16 num_buf;
+       num_buf = (rrd->xsz.xsum_sz.pkt_size + adapter->rx_buffer_len - 1) /
+               adapter->rx_buffer_len;
+       if (rrd->num_buf == num_buf)
+               /* clean alloc flag for bad rrd */
+               atl1_clean_alloc_flag(adapter, rrd, num_buf);
+ }
+ static void atl1_rx_checksum(struct atl1_adapter *adapter,
+       struct rx_return_desc *rrd, struct sk_buff *skb)
+ {
+       struct pci_dev *pdev = adapter->pdev;
+       skb->ip_summed = CHECKSUM_NONE;
+       if (unlikely(rrd->pkt_flg & PACKET_FLAG_ERR)) {
+               if (rrd->err_flg & (ERR_FLAG_CRC | ERR_FLAG_TRUNC |
+                                       ERR_FLAG_CODE | ERR_FLAG_OV)) {
+                       adapter->hw_csum_err++;
+                       if (netif_msg_rx_err(adapter))
+                               dev_printk(KERN_DEBUG, &pdev->dev,
+                                       "rx checksum error\n");
+                       return;
+               }
+       }
+       /* not IPv4 */
+       if (!(rrd->pkt_flg & PACKET_FLAG_IPV4))
+               /* checksum is invalid, but it's not an IPv4 pkt, so ok */
+               return;
+       /* IPv4 packet */
+       if (likely(!(rrd->err_flg &
+               (ERR_FLAG_IP_CHKSUM | ERR_FLAG_L4_CHKSUM)))) {
+               skb->ip_summed = CHECKSUM_UNNECESSARY;
+               adapter->hw_csum_good++;
+               return;
+       }
+       /* IPv4, but hardware thinks its checksum is wrong */
+       if (netif_msg_rx_err(adapter))
+               dev_printk(KERN_DEBUG, &pdev->dev,
+                       "hw csum wrong, pkt_flag:%x, err_flag:%x\n",
+                       rrd->pkt_flg, rrd->err_flg);
+       skb->ip_summed = CHECKSUM_COMPLETE;
+       skb->csum = htons(rrd->xsz.xsum_sz.rx_chksum);
+       adapter->hw_csum_err++;
+       return;
+ }
+ /*
+  * atl1_alloc_rx_buffers - Replace used receive buffers
+  * @adapter: address of board private structure
+  */
+ static u16 atl1_alloc_rx_buffers(struct atl1_adapter *adapter)
+ {
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       struct pci_dev *pdev = adapter->pdev;
+       struct page *page;
+       unsigned long offset;
+       struct atl1_buffer *buffer_info, *next_info;
+       struct sk_buff *skb;
+       u16 num_alloc = 0;
+       u16 rfd_next_to_use, next_next;
+       struct rx_free_desc *rfd_desc;
+       next_next = rfd_next_to_use = atomic_read(&rfd_ring->next_to_use);
+       if (++next_next == rfd_ring->count)
+               next_next = 0;
+       buffer_info = &rfd_ring->buffer_info[rfd_next_to_use];
+       next_info = &rfd_ring->buffer_info[next_next];
+       while (!buffer_info->alloced && !next_info->alloced) {
+               if (buffer_info->skb) {
+                       buffer_info->alloced = 1;
+                       goto next;
+               }
+               rfd_desc = ATL1_RFD_DESC(rfd_ring, rfd_next_to_use);
+               skb = dev_alloc_skb(adapter->rx_buffer_len + NET_IP_ALIGN);
+               if (unlikely(!skb)) {
+                       /* Better luck next round */
+                       adapter->net_stats.rx_dropped++;
+                       break;
+               }
+               /*
+                * Make buffer alignment 2 beyond a 16 byte boundary
+                * this will result in a 16 byte aligned IP header after
+                * the 14 byte MAC header is removed
+                */
+               skb_reserve(skb, NET_IP_ALIGN);
+               buffer_info->alloced = 1;
+               buffer_info->skb = skb;
+               buffer_info->length = (u16) adapter->rx_buffer_len;
+               page = virt_to_page(skb->data);
+               offset = (unsigned long)skb->data & ~PAGE_MASK;
+               buffer_info->dma = pci_map_page(pdev, page, offset,
+                                               adapter->rx_buffer_len,
+                                               PCI_DMA_FROMDEVICE);
+               rfd_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
+               rfd_desc->buf_len = cpu_to_le16(adapter->rx_buffer_len);
+               rfd_desc->coalese = 0;
+ next:
+               rfd_next_to_use = next_next;
+               if (unlikely(++next_next == rfd_ring->count))
+                       next_next = 0;
+               buffer_info = &rfd_ring->buffer_info[rfd_next_to_use];
+               next_info = &rfd_ring->buffer_info[next_next];
+               num_alloc++;
+       }
+       if (num_alloc) {
+               /*
+                * 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();
+               atomic_set(&rfd_ring->next_to_use, (int)rfd_next_to_use);
+       }
+       return num_alloc;
+ }
+ static void atl1_intr_rx(struct atl1_adapter *adapter)
+ {
+       int i, count;
+       u16 length;
+       u16 rrd_next_to_clean;
+       u32 value;
+       struct atl1_rfd_ring *rfd_ring = &adapter->rfd_ring;
+       struct atl1_rrd_ring *rrd_ring = &adapter->rrd_ring;
+       struct atl1_buffer *buffer_info;
+       struct rx_return_desc *rrd;
+       struct sk_buff *skb;
+       count = 0;
+       rrd_next_to_clean = atomic_read(&rrd_ring->next_to_clean);
+       while (1) {
+               rrd = ATL1_RRD_DESC(rrd_ring, rrd_next_to_clean);
+               i = 1;
+               if (likely(rrd->xsz.valid)) {   /* packet valid */
+ chk_rrd:
+                       /* check rrd status */
+                       if (likely(rrd->num_buf == 1))
+                               goto rrd_ok;
+                       else if (netif_msg_rx_err(adapter)) {
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "unexpected RRD buffer count\n");
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "rx_buf_len = %d\n",
+                                       adapter->rx_buffer_len);
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "RRD num_buf = %d\n",
+                                       rrd->num_buf);
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "RRD pkt_len = %d\n",
+                                       rrd->xsz.xsum_sz.pkt_size);
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "RRD pkt_flg = 0x%08X\n",
+                                       rrd->pkt_flg);
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "RRD err_flg = 0x%08X\n",
+                                       rrd->err_flg);
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "RRD vlan_tag = 0x%08X\n",
+                                       rrd->vlan_tag);
+                       }
+                       /* rrd seems to be bad */
+                       if (unlikely(i-- > 0)) {
+                               /* rrd may not be DMAed completely */
+                               udelay(1);
+                               goto chk_rrd;
+                       }
+                       /* bad rrd */
+                       if (netif_msg_rx_err(adapter))
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "bad RRD\n");
+                       /* see if update RFD index */
+                       if (rrd->num_buf > 1)
+                               atl1_update_rfd_index(adapter, rrd);
+                       /* update rrd */
+                       rrd->xsz.valid = 0;
+                       if (++rrd_next_to_clean == rrd_ring->count)
+                               rrd_next_to_clean = 0;
+                       count++;
+                       continue;
+               } else {        /* current rrd still not be updated */
+                       break;
+               }
+ rrd_ok:
+               /* clean alloc flag for bad rrd */
+               atl1_clean_alloc_flag(adapter, rrd, 0);
+               buffer_info = &rfd_ring->buffer_info[rrd->buf_indx];
+               if (++rfd_ring->next_to_clean == rfd_ring->count)
+                       rfd_ring->next_to_clean = 0;
+               /* update rrd next to clean */
+               if (++rrd_next_to_clean == rrd_ring->count)
+                       rrd_next_to_clean = 0;
+               count++;
+               if (unlikely(rrd->pkt_flg & PACKET_FLAG_ERR)) {
+                       if (!(rrd->err_flg &
+                               (ERR_FLAG_IP_CHKSUM | ERR_FLAG_L4_CHKSUM
+                               | ERR_FLAG_LEN))) {
+                               /* packet error, don't need upstream */
+                               buffer_info->alloced = 0;
+                               rrd->xsz.valid = 0;
+                               continue;
+                       }
+               }
+               /* Good Receive */
+               pci_unmap_page(adapter->pdev, buffer_info->dma,
+                              buffer_info->length, PCI_DMA_FROMDEVICE);
+               skb = buffer_info->skb;
+               length = le16_to_cpu(rrd->xsz.xsum_sz.pkt_size);
+               skb_put(skb, length - ETH_FCS_LEN);
+               /* Receive Checksum Offload */
+               atl1_rx_checksum(adapter, rrd, skb);
+               skb->protocol = eth_type_trans(skb, adapter->netdev);
+               if (adapter->vlgrp && (rrd->pkt_flg & PACKET_FLAG_VLAN_INS)) {
+                       u16 vlan_tag = (rrd->vlan_tag >> 4) |
+                                       ((rrd->vlan_tag & 7) << 13) |
+                                       ((rrd->vlan_tag & 8) << 9);
+                       vlan_hwaccel_rx(skb, adapter->vlgrp, vlan_tag);
+               } else
+                       netif_rx(skb);
+               /* let protocol layer free skb */
+               buffer_info->skb = NULL;
+               buffer_info->alloced = 0;
+               rrd->xsz.valid = 0;
+               adapter->netdev->last_rx = jiffies;
+       }
+       atomic_set(&rrd_ring->next_to_clean, rrd_next_to_clean);
+       atl1_alloc_rx_buffers(adapter);
+       /* update mailbox ? */
+       if (count) {
+               u32 tpd_next_to_use;
+               u32 rfd_next_to_use;
+               spin_lock(&adapter->mb_lock);
+               tpd_next_to_use = atomic_read(&adapter->tpd_ring.next_to_use);
+               rfd_next_to_use =
+                   atomic_read(&adapter->rfd_ring.next_to_use);
+               rrd_next_to_clean =
+                   atomic_read(&adapter->rrd_ring.next_to_clean);
+               value = ((rfd_next_to_use & MB_RFD_PROD_INDX_MASK) <<
+                       MB_RFD_PROD_INDX_SHIFT) |
+                         ((rrd_next_to_clean & MB_RRD_CONS_INDX_MASK) <<
+                       MB_RRD_CONS_INDX_SHIFT) |
+                         ((tpd_next_to_use & MB_TPD_PROD_INDX_MASK) <<
+                       MB_TPD_PROD_INDX_SHIFT);
+               iowrite32(value, adapter->hw.hw_addr + REG_MAILBOX);
+               spin_unlock(&adapter->mb_lock);
+       }
+ }
+ static void atl1_intr_tx(struct atl1_adapter *adapter)
+ {
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_buffer *buffer_info;
+       u16 sw_tpd_next_to_clean;
+       u16 cmb_tpd_next_to_clean;
+       sw_tpd_next_to_clean = atomic_read(&tpd_ring->next_to_clean);
+       cmb_tpd_next_to_clean = le16_to_cpu(adapter->cmb.cmb->tpd_cons_idx);
+       while (cmb_tpd_next_to_clean != sw_tpd_next_to_clean) {
+               struct tx_packet_desc *tpd;
+               tpd = ATL1_TPD_DESC(tpd_ring, sw_tpd_next_to_clean);
+               buffer_info = &tpd_ring->buffer_info[sw_tpd_next_to_clean];
+               if (buffer_info->dma) {
+                       pci_unmap_page(adapter->pdev, buffer_info->dma,
+                                      buffer_info->length, PCI_DMA_TODEVICE);
+                       buffer_info->dma = 0;
+               }
+               if (buffer_info->skb) {
+                       dev_kfree_skb_irq(buffer_info->skb);
+                       buffer_info->skb = NULL;
+               }
+               if (++sw_tpd_next_to_clean == tpd_ring->count)
+                       sw_tpd_next_to_clean = 0;
+       }
+       atomic_set(&tpd_ring->next_to_clean, sw_tpd_next_to_clean);
+       if (netif_queue_stopped(adapter->netdev)
+           && netif_carrier_ok(adapter->netdev))
+               netif_wake_queue(adapter->netdev);
+ }
+ static u16 atl1_tpd_avail(struct atl1_tpd_ring *tpd_ring)
+ {
+       u16 next_to_clean = atomic_read(&tpd_ring->next_to_clean);
+       u16 next_to_use = atomic_read(&tpd_ring->next_to_use);
+       return ((next_to_clean > next_to_use) ?
+               next_to_clean - next_to_use - 1 :
+               tpd_ring->count + next_to_clean - next_to_use - 1);
+ }
+ static int atl1_tso(struct atl1_adapter *adapter, struct sk_buff *skb,
+       struct tx_packet_desc *ptpd)
+ {
+       /* spinlock held */
+       u8 hdr_len, ip_off;
+       u32 real_len;
+       int err;
+       if (skb_shinfo(skb)->gso_size) {
+               if (skb_header_cloned(skb)) {
+                       err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
+                       if (unlikely(err))
+                               return -1;
+               }
+               if (skb->protocol == ntohs(ETH_P_IP)) {
+                       struct iphdr *iph = ip_hdr(skb);
+                       real_len = (((unsigned char *)iph - skb->data) +
+                               ntohs(iph->tot_len));
+                       if (real_len < skb->len)
+                               pskb_trim(skb, real_len);
+                       hdr_len = (skb_transport_offset(skb) + tcp_hdrlen(skb));
+                       if (skb->len == hdr_len) {
+                               iph->check = 0;
+                               tcp_hdr(skb)->check =
+                                       ~csum_tcpudp_magic(iph->saddr,
+                                       iph->daddr, tcp_hdrlen(skb),
+                                       IPPROTO_TCP, 0);
+                               ptpd->word3 |= (iph->ihl & TPD_IPHL_MASK) <<
+                                       TPD_IPHL_SHIFT;
+                               ptpd->word3 |= ((tcp_hdrlen(skb) >> 2) &
+                                       TPD_TCPHDRLEN_MASK) <<
+                                       TPD_TCPHDRLEN_SHIFT;
+                               ptpd->word3 |= 1 << TPD_IP_CSUM_SHIFT;
+                               ptpd->word3 |= 1 << TPD_TCP_CSUM_SHIFT;
+                               return 1;
+                       }
+                       iph->check = 0;
+                       tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
+                                       iph->daddr, 0, IPPROTO_TCP, 0);
+                       ip_off = (unsigned char *)iph -
+                               (unsigned char *) skb_network_header(skb);
+                       if (ip_off == 8) /* 802.3-SNAP frame */
+                               ptpd->word3 |= 1 << TPD_ETHTYPE_SHIFT;
+                       else if (ip_off != 0)
+                               return -2;
+                       ptpd->word3 |= (iph->ihl & TPD_IPHL_MASK) <<
+                               TPD_IPHL_SHIFT;
+                       ptpd->word3 |= ((tcp_hdrlen(skb) >> 2) &
+                               TPD_TCPHDRLEN_MASK) << TPD_TCPHDRLEN_SHIFT;
+                       ptpd->word3 |= (skb_shinfo(skb)->gso_size &
+                               TPD_MSS_MASK) << TPD_MSS_SHIFT;
+                       ptpd->word3 |= 1 << TPD_SEGMENT_EN_SHIFT;
+                       return 3;
+               }
+       }
+       return false;
+ }
+ static int atl1_tx_csum(struct atl1_adapter *adapter, struct sk_buff *skb,
+       struct tx_packet_desc *ptpd)
+ {
+       u8 css, cso;
+       if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
+               css = (u8) (skb->csum_start - skb_headroom(skb));
+               cso = css + (u8) skb->csum_offset;
+               if (unlikely(css & 0x1)) {
+                       /* L1 hardware requires an even number here */
+                       if (netif_msg_tx_err(adapter))
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "payload offset not an even number\n");
+                       return -1;
+               }
+               ptpd->word3 |= (css & TPD_PLOADOFFSET_MASK) <<
+                       TPD_PLOADOFFSET_SHIFT;
+               ptpd->word3 |= (cso & TPD_CCSUMOFFSET_MASK) <<
+                       TPD_CCSUMOFFSET_SHIFT;
+               ptpd->word3 |= 1 << TPD_CUST_CSUM_EN_SHIFT;
+               return true;
+       }
+       return 0;
+ }
+ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
+       struct tx_packet_desc *ptpd)
+ {
+       /* spinlock held */
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_buffer *buffer_info;
+       u16 buf_len = skb->len;
+       struct page *page;
+       unsigned long offset;
+       unsigned int nr_frags;
+       unsigned int f;
+       int retval;
+       u16 next_to_use;
+       u16 data_len;
+       u8 hdr_len;
+       buf_len -= skb->data_len;
+       nr_frags = skb_shinfo(skb)->nr_frags;
+       next_to_use = atomic_read(&tpd_ring->next_to_use);
+       buffer_info = &tpd_ring->buffer_info[next_to_use];
+       if (unlikely(buffer_info->skb))
+               BUG();
+       /* put skb in last TPD */
+       buffer_info->skb = NULL;
+       retval = (ptpd->word3 >> TPD_SEGMENT_EN_SHIFT) & TPD_SEGMENT_EN_MASK;
+       if (retval) {
+               /* TSO */
+               hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
+               buffer_info->length = hdr_len;
+               page = virt_to_page(skb->data);
+               offset = (unsigned long)skb->data & ~PAGE_MASK;
+               buffer_info->dma = pci_map_page(adapter->pdev, page,
+                                               offset, hdr_len,
+                                               PCI_DMA_TODEVICE);
+               if (++next_to_use == tpd_ring->count)
+                       next_to_use = 0;
+               if (buf_len > hdr_len) {
+                       int i, nseg;
+                       data_len = buf_len - hdr_len;
+                       nseg = (data_len + ATL1_MAX_TX_BUF_LEN - 1) /
+                               ATL1_MAX_TX_BUF_LEN;
+                       for (i = 0; i < nseg; i++) {
+                               buffer_info =
+                                   &tpd_ring->buffer_info[next_to_use];
+                               buffer_info->skb = NULL;
+                               buffer_info->length =
+                                   (ATL1_MAX_TX_BUF_LEN >=
+                                    data_len) ? ATL1_MAX_TX_BUF_LEN : data_len;
+                               data_len -= buffer_info->length;
+                               page = virt_to_page(skb->data +
+                                       (hdr_len + i * ATL1_MAX_TX_BUF_LEN));
+                               offset = (unsigned long)(skb->data +
+                                       (hdr_len + i * ATL1_MAX_TX_BUF_LEN)) &
+                                       ~PAGE_MASK;
+                               buffer_info->dma = pci_map_page(adapter->pdev,
+                                       page, offset, buffer_info->length,
+                                       PCI_DMA_TODEVICE);
+                               if (++next_to_use == tpd_ring->count)
+                                       next_to_use = 0;
+                       }
+               }
+       } else {
+               /* not TSO */
+               buffer_info->length = buf_len;
+               page = virt_to_page(skb->data);
+               offset = (unsigned long)skb->data & ~PAGE_MASK;
+               buffer_info->dma = pci_map_page(adapter->pdev, page,
+                       offset, buf_len, PCI_DMA_TODEVICE);
+               if (++next_to_use == tpd_ring->count)
+                       next_to_use = 0;
+       }
+       for (f = 0; f < nr_frags; f++) {
+               struct skb_frag_struct *frag;
+               u16 i, nseg;
+               frag = &skb_shinfo(skb)->frags[f];
+               buf_len = frag->size;
+               nseg = (buf_len + ATL1_MAX_TX_BUF_LEN - 1) /
+                       ATL1_MAX_TX_BUF_LEN;
+               for (i = 0; i < nseg; i++) {
+                       buffer_info = &tpd_ring->buffer_info[next_to_use];
+                       if (unlikely(buffer_info->skb))
+                               BUG();
+                       buffer_info->skb = NULL;
+                       buffer_info->length = (buf_len > ATL1_MAX_TX_BUF_LEN) ?
+                               ATL1_MAX_TX_BUF_LEN : buf_len;
+                       buf_len -= buffer_info->length;
+                       buffer_info->dma = pci_map_page(adapter->pdev,
+                               frag->page,
+                               frag->page_offset + (i * ATL1_MAX_TX_BUF_LEN),
+                               buffer_info->length, PCI_DMA_TODEVICE);
+                       if (++next_to_use == tpd_ring->count)
+                               next_to_use = 0;
+               }
+       }
+       /* last tpd's buffer-info */
+       buffer_info->skb = skb;
+ }
+ static void atl1_tx_queue(struct atl1_adapter *adapter, u16 count,
+        struct tx_packet_desc *ptpd)
+ {
+       /* spinlock held */
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       struct atl1_buffer *buffer_info;
+       struct tx_packet_desc *tpd;
+       u16 j;
+       u32 val;
+       u16 next_to_use = (u16) atomic_read(&tpd_ring->next_to_use);
+       for (j = 0; j < count; j++) {
+               buffer_info = &tpd_ring->buffer_info[next_to_use];
+               tpd = ATL1_TPD_DESC(&adapter->tpd_ring, next_to_use);
+               if (tpd != ptpd)
+                       memcpy(tpd, ptpd, sizeof(struct tx_packet_desc));
+               tpd->buffer_addr = cpu_to_le64(buffer_info->dma);
+               tpd->word2 = (cpu_to_le16(buffer_info->length) &
+                       TPD_BUFLEN_MASK) << TPD_BUFLEN_SHIFT;
+               /*
+                * if this is the first packet in a TSO chain, set
+                * TPD_HDRFLAG, otherwise, clear it.
+                */
+               val = (tpd->word3 >> TPD_SEGMENT_EN_SHIFT) &
+                       TPD_SEGMENT_EN_MASK;
+               if (val) {
+                       if (!j)
+                               tpd->word3 |= 1 << TPD_HDRFLAG_SHIFT;
+                       else
+                               tpd->word3 &= ~(1 << TPD_HDRFLAG_SHIFT);
+               }
+               if (j == (count - 1))
+                       tpd->word3 |= 1 << TPD_EOP_SHIFT;
+               if (++next_to_use == tpd_ring->count)
+                       next_to_use = 0;
+       }
+       /*
+        * 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();
+       atomic_set(&tpd_ring->next_to_use, next_to_use);
+ }
+ static int atl1_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
+       int len = skb->len;
+       int tso;
+       int count = 1;
+       int ret_val;
+       struct tx_packet_desc *ptpd;
+       u16 frag_size;
+       u16 vlan_tag;
+       unsigned long flags;
+       unsigned int nr_frags = 0;
+       unsigned int mss = 0;
+       unsigned int f;
+       unsigned int proto_hdr_len;
+       len -= skb->data_len;
+       if (unlikely(skb->len <= 0)) {
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+       nr_frags = skb_shinfo(skb)->nr_frags;
+       for (f = 0; f < nr_frags; f++) {
+               frag_size = skb_shinfo(skb)->frags[f].size;
+               if (frag_size)
+                       count += (frag_size + ATL1_MAX_TX_BUF_LEN - 1) /
+                               ATL1_MAX_TX_BUF_LEN;
+       }
+       mss = skb_shinfo(skb)->gso_size;
+       if (mss) {
+               if (skb->protocol == ntohs(ETH_P_IP)) {
+                       proto_hdr_len = (skb_transport_offset(skb) +
+                                        tcp_hdrlen(skb));
+                       if (unlikely(proto_hdr_len > len)) {
+                               dev_kfree_skb_any(skb);
+                               return NETDEV_TX_OK;
+                       }
+                       /* need additional TPD ? */
+                       if (proto_hdr_len != len)
+                               count += (len - proto_hdr_len +
+                                       ATL1_MAX_TX_BUF_LEN - 1) /
+                                       ATL1_MAX_TX_BUF_LEN;
+               }
+       }
+       if (!spin_trylock_irqsave(&adapter->lock, flags)) {
+               /* Can't get lock - tell upper layer to requeue */
+               if (netif_msg_tx_queued(adapter))
+                       dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                               "tx locked\n");
+               return NETDEV_TX_LOCKED;
+       }
+       if (atl1_tpd_avail(&adapter->tpd_ring) < count) {
+               /* not enough descriptors */
+               netif_stop_queue(netdev);
+               spin_unlock_irqrestore(&adapter->lock, flags);
+               if (netif_msg_tx_queued(adapter))
+                       dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                               "tx busy\n");
+               return NETDEV_TX_BUSY;
+       }
+       ptpd = ATL1_TPD_DESC(tpd_ring,
+               (u16) atomic_read(&tpd_ring->next_to_use));
+       memset(ptpd, 0, sizeof(struct tx_packet_desc));
+       if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
+               vlan_tag = vlan_tx_tag_get(skb);
+               vlan_tag = (vlan_tag << 4) | (vlan_tag >> 13) |
+                       ((vlan_tag >> 9) & 0x8);
+               ptpd->word3 |= 1 << TPD_INS_VL_TAG_SHIFT;
+               ptpd->word3 |= (vlan_tag & TPD_VL_TAGGED_MASK) <<
+                       TPD_VL_TAGGED_SHIFT;
+       }
+       tso = atl1_tso(adapter, skb, ptpd);
+       if (tso < 0) {
+               spin_unlock_irqrestore(&adapter->lock, flags);
+               dev_kfree_skb_any(skb);
+               return NETDEV_TX_OK;
+       }
+       if (!tso) {
+               ret_val = atl1_tx_csum(adapter, skb, ptpd);
+               if (ret_val < 0) {
+                       spin_unlock_irqrestore(&adapter->lock, flags);
+                       dev_kfree_skb_any(skb);
+                       return NETDEV_TX_OK;
+               }
+       }
+       atl1_tx_map(adapter, skb, ptpd);
+       atl1_tx_queue(adapter, count, ptpd);
+       atl1_update_mailbox(adapter);
+       spin_unlock_irqrestore(&adapter->lock, flags);
+       netdev->trans_start = jiffies;
+       return NETDEV_TX_OK;
+ }
+ /*
+  * atl1_intr - Interrupt Handler
+  * @irq: interrupt number
+  * @data: pointer to a network interface device structure
+  * @pt_regs: CPU registers structure
+  */
+ static irqreturn_t atl1_intr(int irq, void *data)
+ {
+       struct atl1_adapter *adapter = netdev_priv(data);
+       u32 status;
 -      update_rx = 0;
 -
+       int max_ints = 10;
+       status = adapter->cmb.cmb->int_stats;
+       if (!status)
+               return IRQ_NONE;
+       do {
+               /* clear CMB interrupt status at once */
+               adapter->cmb.cmb->int_stats = 0;
+               if (status & ISR_GPHY)  /* clear phy status */
+                       atlx_clear_phy_int(adapter);
+               /* clear ISR status, and Enable CMB DMA/Disable Interrupt */
+               iowrite32(status | ISR_DIS_INT, adapter->hw.hw_addr + REG_ISR);
+               /* check if SMB intr */
+               if (status & ISR_SMB)
+                       atl1_inc_smb(adapter);
+               /* check if PCIE PHY Link down */
+               if (status & ISR_PHY_LINKDOWN) {
+                       if (netif_msg_intr(adapter))
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "pcie phy link down %x\n", status);
+                       if (netif_running(adapter->netdev)) {   /* reset MAC */
+                               iowrite32(0, adapter->hw.hw_addr + REG_IMR);
+                               schedule_work(&adapter->pcie_dma_to_rst_task);
+                               return IRQ_HANDLED;
+                       }
+               }
+               /* check if DMA read/write error ? */
+               if (status & (ISR_DMAR_TO_RST | ISR_DMAW_TO_RST)) {
+                       if (netif_msg_intr(adapter))
+                               dev_printk(KERN_DEBUG, &adapter->pdev->dev,
+                                       "pcie DMA r/w error (status = 0x%x)\n",
+                                       status);
+                       iowrite32(0, adapter->hw.hw_addr + REG_IMR);
+                       schedule_work(&adapter->pcie_dma_to_rst_task);
+                       return IRQ_HANDLED;
+               }
+               /* link event */
+               if (status & ISR_GPHY) {
+                       adapter->soft_stats.tx_carrier_errors++;
+                       atl1_check_for_link(adapter);
+               }
+               /* transmit event */
+               if (status & ISR_CMB_TX)
+                       atl1_intr_tx(adapter);
+               /* rx exception */
+               if (unlikely(status & (ISR_RXF_OV | ISR_RFD_UNRUN |
+                       ISR_RRD_OV | ISR_HOST_RFD_UNRUN |
+                       ISR_HOST_RRD_OV | ISR_CMB_RX))) {
+                       if (status & (ISR_RXF_OV | ISR_RFD_UNRUN |
+                               ISR_RRD_OV | ISR_HOST_RFD_UNRUN |
+                               ISR_HOST_RRD_OV))
+                               if (netif_msg_intr(adapter))
+                                       dev_printk(KERN_DEBUG,
+                                               &adapter->pdev->dev,
+                                               "rx exception, ISR = 0x%x\n",
+                                               status);
+                       atl1_intr_rx(adapter);
+               }
+               if (--max_ints < 0)
+                       break;
+       } while ((status = adapter->cmb.cmb->int_stats));
+       /* re-enable Interrupt */
+       iowrite32(ISR_DIS_SMB | ISR_DIS_DMA, adapter->hw.hw_addr + REG_ISR);
+       return IRQ_HANDLED;
+ }
+ /*
+  * atl1_watchdog - Timer Call-back
+  * @data: pointer to netdev cast into an unsigned long
+  */
+ static void atl1_watchdog(unsigned long data)
+ {
+       struct atl1_adapter *adapter = (struct atl1_adapter *)data;
+       /* Reset the timer */
+       mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
+ }
+ /*
+  * atl1_phy_config - Timer Call-back
+  * @data: pointer to netdev cast into an unsigned long
+  */
+ static void atl1_phy_config(unsigned long data)
+ {
+       struct atl1_adapter *adapter = (struct atl1_adapter *)data;
+       struct atl1_hw *hw = &adapter->hw;
+       unsigned long flags;
+       spin_lock_irqsave(&adapter->lock, flags);
+       adapter->phy_timer_pending = false;
+       atl1_write_phy_reg(hw, MII_ADVERTISE, hw->mii_autoneg_adv_reg);
+       atl1_write_phy_reg(hw, MII_ATLX_CR, hw->mii_1000t_ctrl_reg);
+       atl1_write_phy_reg(hw, MII_BMCR, MII_CR_RESET | MII_CR_AUTO_NEG_EN);
+       spin_unlock_irqrestore(&adapter->lock, flags);
+ }
+ /*
+  * Orphaned vendor comment left intact here:
+  * <vendor comment>
+  * If TPD Buffer size equal to 0, PCIE DMAR_TO_INT
+  * will assert. We do soft reset <0x1400=1> according
+  * with the SPEC. BUT, it seemes that PCIE or DMA
+  * state-machine will not be reset. DMAR_TO_INT will
+  * assert again and again.
+  * </vendor comment>
+  */
+ static int atl1_reset(struct atl1_adapter *adapter)
+ {
+       int ret;
+       ret = atl1_reset_hw(&adapter->hw);
+       if (ret)
+               return ret;
+       return atl1_init_hw(&adapter->hw);
+ }
+ static s32 atl1_up(struct atl1_adapter *adapter)
+ {
+       struct net_device *netdev = adapter->netdev;
+       int err;
+       int irq_flags = IRQF_SAMPLE_RANDOM;
+       /* hardware has been reset, we need to reload some things */
+       atlx_set_multi(netdev);
+       atl1_init_ring_ptrs(adapter);
+       atlx_restore_vlan(adapter);
+       err = atl1_alloc_rx_buffers(adapter);
+       if (unlikely(!err))
+               /* no RX BUFFER allocated */
+               return -ENOMEM;
+       if (unlikely(atl1_configure(adapter))) {
+               err = -EIO;
+               goto err_up;
+       }
+       err = pci_enable_msi(adapter->pdev);
+       if (err) {
+               if (netif_msg_ifup(adapter))
+                       dev_info(&adapter->pdev->dev,
+                               "Unable to enable MSI: %d\n", err);
+               irq_flags |= IRQF_SHARED;
+       }
+       err = request_irq(adapter->pdev->irq, &atl1_intr, irq_flags,
+                       netdev->name, netdev);
+       if (unlikely(err))
+               goto err_up;
+       mod_timer(&adapter->watchdog_timer, jiffies);
+       atlx_irq_enable(adapter);
+       atl1_check_link(adapter);
+       return 0;
+ err_up:
+       pci_disable_msi(adapter->pdev);
+       /* free rx_buffers */
+       atl1_clean_rx_ring(adapter);
+       return err;
+ }
+ static void atl1_down(struct atl1_adapter *adapter)
+ {
+       struct net_device *netdev = adapter->netdev;
+       del_timer_sync(&adapter->watchdog_timer);
+       del_timer_sync(&adapter->phy_config_timer);
+       adapter->phy_timer_pending = false;
+       atlx_irq_disable(adapter);
+       free_irq(adapter->pdev->irq, netdev);
+       pci_disable_msi(adapter->pdev);
+       atl1_reset_hw(&adapter->hw);
+       adapter->cmb.cmb->int_stats = 0;
+       adapter->link_speed = SPEED_0;
+       adapter->link_duplex = -1;
+       netif_carrier_off(netdev);
+       netif_stop_queue(netdev);
+       atl1_clean_tx_ring(adapter);
+       atl1_clean_rx_ring(adapter);
+ }
+ static void atl1_tx_timeout_task(struct work_struct *work)
+ {
+       struct atl1_adapter *adapter =
+               container_of(work, struct atl1_adapter, tx_timeout_task);
+       struct net_device *netdev = adapter->netdev;
+       netif_device_detach(netdev);
+       atl1_down(adapter);
+       atl1_up(adapter);
+       netif_device_attach(netdev);
+ }
+ /*
+  * atl1_change_mtu - Change the Maximum Transfer Unit
+  * @netdev: network interface device structure
+  * @new_mtu: new value for maximum frame size
+  *
+  * Returns 0 on success, negative on failure
+  */
+ static int atl1_change_mtu(struct net_device *netdev, int new_mtu)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       int old_mtu = netdev->mtu;
+       int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
+       if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
+           (max_frame > MAX_JUMBO_FRAME_SIZE)) {
+               if (netif_msg_link(adapter))
+                       dev_warn(&adapter->pdev->dev, "invalid MTU setting\n");
+               return -EINVAL;
+       }
+       adapter->hw.max_frame_size = max_frame;
+       adapter->hw.tx_jumbo_task_th = (max_frame + 7) >> 3;
+       adapter->rx_buffer_len = (max_frame + 7) & ~7;
+       adapter->hw.rx_jumbo_th = adapter->rx_buffer_len / 8;
+       netdev->mtu = new_mtu;
+       if ((old_mtu != new_mtu) && netif_running(netdev)) {
+               atl1_down(adapter);
+               atl1_up(adapter);
+       }
+       return 0;
+ }
+ /*
+  * atl1_open - Called when a network interface is made active
+  * @netdev: network interface device structure
+  *
+  * Returns 0 on success, negative value on failure
+  *
+  * The open entry point is called when a network interface is made
+  * active by the system (IFF_UP).  At this point all resources needed
+  * for transmit and receive operations are allocated, the interrupt
+  * handler is registered with the OS, the watchdog timer is started,
+  * and the stack is notified that the interface is ready.
+  */
+ static int atl1_open(struct net_device *netdev)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       int err;
+       /* allocate transmit descriptors */
+       err = atl1_setup_ring_resources(adapter);
+       if (err)
+               return err;
+       err = atl1_up(adapter);
+       if (err)
+               goto err_up;
+       return 0;
+ err_up:
+       atl1_reset(adapter);
+       return err;
+ }
+ /*
+  * atl1_close - Disables a network interface
+  * @netdev: network interface device structure
+  *
+  * Returns 0, this is not allowed to fail
+  *
+  * The close entry point is called when an interface is de-activated
+  * by the OS.  The hardware is still under the drivers control, but
+  * needs to be disabled.  A global MAC reset is issued to stop the
+  * hardware, and all transmit and receive resources are freed.
+  */
+ static int atl1_close(struct net_device *netdev)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       atl1_down(adapter);
+       atl1_free_ring_resources(adapter);
+       return 0;
+ }
+ #ifdef CONFIG_PM
+ static int atl1_suspend(struct pci_dev *pdev, pm_message_t state)
+ {
+       struct net_device *netdev = pci_get_drvdata(pdev);
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       u32 ctrl = 0;
+       u32 wufc = adapter->wol;
+       netif_device_detach(netdev);
+       if (netif_running(netdev))
+               atl1_down(adapter);
+       atl1_read_phy_reg(hw, MII_BMSR, (u16 *) & ctrl);
+       atl1_read_phy_reg(hw, MII_BMSR, (u16 *) & ctrl);
+       if (ctrl & BMSR_LSTATUS)
+               wufc &= ~ATLX_WUFC_LNKC;
+       /* reduce speed to 10/100M */
+       if (wufc) {
+               atl1_phy_enter_power_saving(hw);
+               /* if resume, let driver to re- setup link */
+               hw->phy_configured = false;
+               atl1_set_mac_addr(hw);
+               atlx_set_multi(netdev);
+               ctrl = 0;
+               /* turn on magic packet wol */
+               if (wufc & ATLX_WUFC_MAG)
+                       ctrl = WOL_MAGIC_EN | WOL_MAGIC_PME_EN;
+               /* turn on Link change WOL */
+               if (wufc & ATLX_WUFC_LNKC)
+                       ctrl |= (WOL_LINK_CHG_EN | WOL_LINK_CHG_PME_EN);
+               iowrite32(ctrl, hw->hw_addr + REG_WOL_CTRL);
+               /* turn on all-multi mode if wake on multicast is enabled */
+               ctrl = ioread32(hw->hw_addr + REG_MAC_CTRL);
+               ctrl &= ~MAC_CTRL_DBG;
+               ctrl &= ~MAC_CTRL_PROMIS_EN;
+               if (wufc & ATLX_WUFC_MC)
+                       ctrl |= MAC_CTRL_MC_ALL_EN;
+               else
+                       ctrl &= ~MAC_CTRL_MC_ALL_EN;
+               /* turn on broadcast mode if wake on-BC is enabled */
+               if (wufc & ATLX_WUFC_BC)
+                       ctrl |= MAC_CTRL_BC_EN;
+               else
+                       ctrl &= ~MAC_CTRL_BC_EN;
+               /* enable RX */
+               ctrl |= MAC_CTRL_RX_EN;
+               iowrite32(ctrl, hw->hw_addr + REG_MAC_CTRL);
+               pci_enable_wake(pdev, PCI_D3hot, 1);
+               pci_enable_wake(pdev, PCI_D3cold, 1);
+       } else {
+               iowrite32(0, hw->hw_addr + REG_WOL_CTRL);
+               pci_enable_wake(pdev, PCI_D3hot, 0);
+               pci_enable_wake(pdev, PCI_D3cold, 0);
+       }
+       pci_save_state(pdev);
+       pci_disable_device(pdev);
+       pci_set_power_state(pdev, PCI_D3hot);
+       return 0;
+ }
+ static int atl1_resume(struct pci_dev *pdev)
+ {
+       struct net_device *netdev = pci_get_drvdata(pdev);
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       u32 err;
+       pci_set_power_state(pdev, PCI_D0);
+       pci_restore_state(pdev);
+       /* FIXME: check and handle */
+       err = pci_enable_device(pdev);
+       pci_enable_wake(pdev, PCI_D3hot, 0);
+       pci_enable_wake(pdev, PCI_D3cold, 0);
+       iowrite32(0, adapter->hw.hw_addr + REG_WOL_CTRL);
+       atl1_reset(adapter);
+       if (netif_running(netdev))
+               atl1_up(adapter);
+       netif_device_attach(netdev);
+       atl1_via_workaround(adapter);
+       return 0;
+ }
+ #else
+ #define atl1_suspend NULL
+ #define atl1_resume NULL
+ #endif
+ #ifdef CONFIG_NET_POLL_CONTROLLER
+ static void atl1_poll_controller(struct net_device *netdev)
+ {
+       disable_irq(netdev->irq);
+       atl1_intr(netdev->irq, netdev);
+       enable_irq(netdev->irq);
+ }
+ #endif
+ /*
+  * atl1_probe - Device Initialization Routine
+  * @pdev: PCI device information struct
+  * @ent: entry in atl1_pci_tbl
+  *
+  * Returns 0 on success, negative on failure
+  *
+  * atl1_probe initializes an adapter identified by a pci_dev structure.
+  * The OS initialization, configuring of the adapter private structure,
+  * and a hardware reset occur.
+  */
+ static int __devinit atl1_probe(struct pci_dev *pdev,
+       const struct pci_device_id *ent)
+ {
+       struct net_device *netdev;
+       struct atl1_adapter *adapter;
+       static int cards_found = 0;
+       int err;
+       err = pci_enable_device(pdev);
+       if (err)
+               return err;
+       /*
+        * The atl1 chip can DMA to 64-bit addresses, but it uses a single
+        * shared register for the high 32 bits, so only a single, aligned,
+        * 4 GB physical address range can be used at a time.
+        *
+        * Supporting 64-bit DMA on this hardware is more trouble than it's
+        * worth.  It is far easier to limit to 32-bit DMA than update
+        * various kernel subsystems to support the mechanics required by a
+        * fixed-high-32-bit system.
+        */
+       err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+       if (err) {
+               dev_err(&pdev->dev, "no usable DMA configuration\n");
+               goto err_dma;
+       }
+       /*
+        * Mark all PCI regions associated with PCI device
+        * pdev as being reserved by owner atl1_driver_name
+        */
+       err = pci_request_regions(pdev, ATLX_DRIVER_NAME);
+       if (err)
+               goto err_request_regions;
+       /*
+        * Enables bus-mastering on the device and calls
+        * pcibios_set_master to do the needed arch specific settings
+        */
+       pci_set_master(pdev);
+       netdev = alloc_etherdev(sizeof(struct atl1_adapter));
+       if (!netdev) {
+               err = -ENOMEM;
+               goto err_alloc_etherdev;
+       }
+       SET_NETDEV_DEV(netdev, &pdev->dev);
+       pci_set_drvdata(pdev, netdev);
+       adapter = netdev_priv(netdev);
+       adapter->netdev = netdev;
+       adapter->pdev = pdev;
+       adapter->hw.back = adapter;
+       adapter->msg_enable = netif_msg_init(debug, atl1_default_msg);
+       adapter->hw.hw_addr = pci_iomap(pdev, 0, 0);
+       if (!adapter->hw.hw_addr) {
+               err = -EIO;
+               goto err_pci_iomap;
+       }
+       /* get device revision number */
+       adapter->hw.dev_rev = ioread16(adapter->hw.hw_addr +
+               (REG_MASTER_CTRL + 2));
+       if (netif_msg_probe(adapter))
+               dev_info(&pdev->dev, "version %s\n", ATLX_DRIVER_VERSION);
+       /* set default ring resource counts */
+       adapter->rfd_ring.count = adapter->rrd_ring.count = ATL1_DEFAULT_RFD;
+       adapter->tpd_ring.count = ATL1_DEFAULT_TPD;
+       adapter->mii.dev = netdev;
+       adapter->mii.mdio_read = mdio_read;
+       adapter->mii.mdio_write = mdio_write;
+       adapter->mii.phy_id_mask = 0x1f;
+       adapter->mii.reg_num_mask = 0x1f;
+       netdev->open = &atl1_open;
+       netdev->stop = &atl1_close;
+       netdev->hard_start_xmit = &atl1_xmit_frame;
+       netdev->get_stats = &atlx_get_stats;
+       netdev->set_multicast_list = &atlx_set_multi;
+       netdev->set_mac_address = &atl1_set_mac;
+       netdev->change_mtu = &atl1_change_mtu;
+       netdev->do_ioctl = &atlx_ioctl;
+       netdev->tx_timeout = &atlx_tx_timeout;
+       netdev->watchdog_timeo = 5 * HZ;
+ #ifdef CONFIG_NET_POLL_CONTROLLER
+       netdev->poll_controller = atl1_poll_controller;
+ #endif
+       netdev->vlan_rx_register = atlx_vlan_rx_register;
+       netdev->ethtool_ops = &atl1_ethtool_ops;
+       adapter->bd_number = cards_found;
+       /* setup the private structure */
+       err = atl1_sw_init(adapter);
+       if (err)
+               goto err_common;
+       netdev->features = NETIF_F_HW_CSUM;
+       netdev->features |= NETIF_F_SG;
+       netdev->features |= (NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
+       netdev->features |= NETIF_F_TSO;
+       netdev->features |= NETIF_F_LLTX;
+       /*
+        * patch for some L1 of old version,
+        * the final version of L1 may not need these
+        * patches
+        */
+       /* atl1_pcie_patch(adapter); */
+       /* really reset GPHY core */
+       iowrite16(0, adapter->hw.hw_addr + REG_PHY_ENABLE);
+       /*
+        * reset the controller to
+        * put the device in a known good starting state
+        */
+       if (atl1_reset_hw(&adapter->hw)) {
+               err = -EIO;
+               goto err_common;
+       }
+       /* copy the MAC address out of the EEPROM */
+       atl1_read_mac_addr(&adapter->hw);
+       memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
+       if (!is_valid_ether_addr(netdev->dev_addr)) {
+               err = -EIO;
+               goto err_common;
+       }
+       atl1_check_options(adapter);
+       /* pre-init the MAC, and setup link */
+       err = atl1_init_hw(&adapter->hw);
+       if (err) {
+               err = -EIO;
+               goto err_common;
+       }
+       atl1_pcie_patch(adapter);
+       /* assume we have no link for now */
+       netif_carrier_off(netdev);
+       netif_stop_queue(netdev);
+       init_timer(&adapter->watchdog_timer);
+       adapter->watchdog_timer.function = &atl1_watchdog;
+       adapter->watchdog_timer.data = (unsigned long)adapter;
+       init_timer(&adapter->phy_config_timer);
+       adapter->phy_config_timer.function = &atl1_phy_config;
+       adapter->phy_config_timer.data = (unsigned long)adapter;
+       adapter->phy_timer_pending = false;
+       INIT_WORK(&adapter->tx_timeout_task, atl1_tx_timeout_task);
+       INIT_WORK(&adapter->link_chg_task, atlx_link_chg_task);
+       INIT_WORK(&adapter->pcie_dma_to_rst_task, atl1_tx_timeout_task);
+       err = register_netdev(netdev);
+       if (err)
+               goto err_common;
+       cards_found++;
+       atl1_via_workaround(adapter);
+       return 0;
+ err_common:
+       pci_iounmap(pdev, adapter->hw.hw_addr);
+ err_pci_iomap:
+       free_netdev(netdev);
+ err_alloc_etherdev:
+       pci_release_regions(pdev);
+ err_dma:
+ err_request_regions:
+       pci_disable_device(pdev);
+       return err;
+ }
+ /*
+  * atl1_remove - Device Removal Routine
+  * @pdev: PCI device information struct
+  *
+  * atl1_remove is called by the PCI subsystem to alert the driver
+  * that it should release a PCI device.  The could be caused by a
+  * Hot-Plug event, or because the driver is going to be removed from
+  * memory.
+  */
+ static void __devexit atl1_remove(struct pci_dev *pdev)
+ {
+       struct net_device *netdev = pci_get_drvdata(pdev);
+       struct atl1_adapter *adapter;
+       /* Device not available. Return. */
+       if (!netdev)
+               return;
+       adapter = netdev_priv(netdev);
+       /*
+        * Some atl1 boards lack persistent storage for their MAC, and get it
+        * from the BIOS during POST.  If we've been messing with the MAC
+        * address, we need to save the permanent one.
+        */
+       if (memcmp(adapter->hw.mac_addr, adapter->hw.perm_mac_addr, ETH_ALEN)) {
+               memcpy(adapter->hw.mac_addr, adapter->hw.perm_mac_addr,
+                       ETH_ALEN);
+               atl1_set_mac_addr(&adapter->hw);
+       }
+       iowrite16(0, adapter->hw.hw_addr + REG_PHY_ENABLE);
+       unregister_netdev(netdev);
+       pci_iounmap(pdev, adapter->hw.hw_addr);
+       pci_release_regions(pdev);
+       free_netdev(netdev);
+       pci_disable_device(pdev);
+ }
+ static struct pci_driver atl1_driver = {
+       .name = ATLX_DRIVER_NAME,
+       .id_table = atl1_pci_tbl,
+       .probe = atl1_probe,
+       .remove = __devexit_p(atl1_remove),
+       .suspend = atl1_suspend,
+       .resume = atl1_resume
+ };
+ /*
+  * atl1_exit_module - Driver Exit Cleanup Routine
+  *
+  * atl1_exit_module is called just before the driver is removed
+  * from memory.
+  */
+ static void __exit atl1_exit_module(void)
+ {
+       pci_unregister_driver(&atl1_driver);
+ }
+ /*
+  * atl1_init_module - Driver Registration Routine
+  *
+  * atl1_init_module is the first routine called when the driver is
+  * loaded. All it does is register with the PCI subsystem.
+  */
+ static int __init atl1_init_module(void)
+ {
+       return pci_register_driver(&atl1_driver);
+ }
+ module_init(atl1_init_module);
+ module_exit(atl1_exit_module);
+ struct atl1_stats {
+       char stat_string[ETH_GSTRING_LEN];
+       int sizeof_stat;
+       int stat_offset;
+ };
+ #define ATL1_STAT(m) \
+       sizeof(((struct atl1_adapter *)0)->m), offsetof(struct atl1_adapter, m)
+ static struct atl1_stats atl1_gstrings_stats[] = {
+       {"rx_packets", ATL1_STAT(soft_stats.rx_packets)},
+       {"tx_packets", ATL1_STAT(soft_stats.tx_packets)},
+       {"rx_bytes", ATL1_STAT(soft_stats.rx_bytes)},
+       {"tx_bytes", ATL1_STAT(soft_stats.tx_bytes)},
+       {"rx_errors", ATL1_STAT(soft_stats.rx_errors)},
+       {"tx_errors", ATL1_STAT(soft_stats.tx_errors)},
+       {"rx_dropped", ATL1_STAT(net_stats.rx_dropped)},
+       {"tx_dropped", ATL1_STAT(net_stats.tx_dropped)},
+       {"multicast", ATL1_STAT(soft_stats.multicast)},
+       {"collisions", ATL1_STAT(soft_stats.collisions)},
+       {"rx_length_errors", ATL1_STAT(soft_stats.rx_length_errors)},
+       {"rx_over_errors", ATL1_STAT(soft_stats.rx_missed_errors)},
+       {"rx_crc_errors", ATL1_STAT(soft_stats.rx_crc_errors)},
+       {"rx_frame_errors", ATL1_STAT(soft_stats.rx_frame_errors)},
+       {"rx_fifo_errors", ATL1_STAT(soft_stats.rx_fifo_errors)},
+       {"rx_missed_errors", ATL1_STAT(soft_stats.rx_missed_errors)},
+       {"tx_aborted_errors", ATL1_STAT(soft_stats.tx_aborted_errors)},
+       {"tx_carrier_errors", ATL1_STAT(soft_stats.tx_carrier_errors)},
+       {"tx_fifo_errors", ATL1_STAT(soft_stats.tx_fifo_errors)},
+       {"tx_window_errors", ATL1_STAT(soft_stats.tx_window_errors)},
+       {"tx_abort_exce_coll", ATL1_STAT(soft_stats.excecol)},
+       {"tx_abort_late_coll", ATL1_STAT(soft_stats.latecol)},
+       {"tx_deferred_ok", ATL1_STAT(soft_stats.deffer)},
+       {"tx_single_coll_ok", ATL1_STAT(soft_stats.scc)},
+       {"tx_multi_coll_ok", ATL1_STAT(soft_stats.mcc)},
+       {"tx_underun", ATL1_STAT(soft_stats.tx_underun)},
+       {"tx_trunc", ATL1_STAT(soft_stats.tx_trunc)},
+       {"tx_pause", ATL1_STAT(soft_stats.tx_pause)},
+       {"rx_pause", ATL1_STAT(soft_stats.rx_pause)},
+       {"rx_rrd_ov", ATL1_STAT(soft_stats.rx_rrd_ov)},
+       {"rx_trunc", ATL1_STAT(soft_stats.rx_trunc)}
+ };
+ static void atl1_get_ethtool_stats(struct net_device *netdev,
+       struct ethtool_stats *stats, u64 *data)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       int i;
+       char *p;
+       for (i = 0; i < ARRAY_SIZE(atl1_gstrings_stats); i++) {
+               p = (char *)adapter+atl1_gstrings_stats[i].stat_offset;
+               data[i] = (atl1_gstrings_stats[i].sizeof_stat ==
+                       sizeof(u64)) ? *(u64 *)p : *(u32 *)p;
+       }
+ }
+ static int atl1_get_sset_count(struct net_device *netdev, int sset)
+ {
+       switch (sset) {
+       case ETH_SS_STATS:
+               return ARRAY_SIZE(atl1_gstrings_stats);
+       default:
+               return -EOPNOTSUPP;
+       }
+ }
+ static int atl1_get_settings(struct net_device *netdev,
+       struct ethtool_cmd *ecmd)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       ecmd->supported = (SUPPORTED_10baseT_Half |
+                          SUPPORTED_10baseT_Full |
+                          SUPPORTED_100baseT_Half |
+                          SUPPORTED_100baseT_Full |
+                          SUPPORTED_1000baseT_Full |
+                          SUPPORTED_Autoneg | SUPPORTED_TP);
+       ecmd->advertising = ADVERTISED_TP;
+       if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+           hw->media_type == MEDIA_TYPE_1000M_FULL) {
+               ecmd->advertising |= ADVERTISED_Autoneg;
+               if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR) {
+                       ecmd->advertising |= ADVERTISED_Autoneg;
+                       ecmd->advertising |=
+                           (ADVERTISED_10baseT_Half |
+                            ADVERTISED_10baseT_Full |
+                            ADVERTISED_100baseT_Half |
+                            ADVERTISED_100baseT_Full |
+                            ADVERTISED_1000baseT_Full);
+               } else
+                       ecmd->advertising |= (ADVERTISED_1000baseT_Full);
+       }
+       ecmd->port = PORT_TP;
+       ecmd->phy_address = 0;
+       ecmd->transceiver = XCVR_INTERNAL;
+       if (netif_carrier_ok(adapter->netdev)) {
+               u16 link_speed, link_duplex;
+               atl1_get_speed_and_duplex(hw, &link_speed, &link_duplex);
+               ecmd->speed = link_speed;
+               if (link_duplex == FULL_DUPLEX)
+                       ecmd->duplex = DUPLEX_FULL;
+               else
+                       ecmd->duplex = DUPLEX_HALF;
+       } else {
+               ecmd->speed = -1;
+               ecmd->duplex = -1;
+       }
+       if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+           hw->media_type == MEDIA_TYPE_1000M_FULL)
+               ecmd->autoneg = AUTONEG_ENABLE;
+       else
+               ecmd->autoneg = AUTONEG_DISABLE;
+       return 0;
+ }
+ static int atl1_set_settings(struct net_device *netdev,
+       struct ethtool_cmd *ecmd)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       u16 phy_data;
+       int ret_val = 0;
+       u16 old_media_type = hw->media_type;
+       if (netif_running(adapter->netdev)) {
+               if (netif_msg_link(adapter))
+                       dev_dbg(&adapter->pdev->dev,
+                               "ethtool shutting down adapter\n");
+               atl1_down(adapter);
+       }
+       if (ecmd->autoneg == AUTONEG_ENABLE)
+               hw->media_type = MEDIA_TYPE_AUTO_SENSOR;
+       else {
+               if (ecmd->speed == SPEED_1000) {
+                       if (ecmd->duplex != DUPLEX_FULL) {
+                               if (netif_msg_link(adapter))
+                                       dev_warn(&adapter->pdev->dev,
+                                               "1000M half is invalid\n");
+                               ret_val = -EINVAL;
+                               goto exit_sset;
+                       }
+                       hw->media_type = MEDIA_TYPE_1000M_FULL;
+               } else if (ecmd->speed == SPEED_100) {
+                       if (ecmd->duplex == DUPLEX_FULL)
+                               hw->media_type = MEDIA_TYPE_100M_FULL;
+                       else
+                               hw->media_type = MEDIA_TYPE_100M_HALF;
+               } else {
+                       if (ecmd->duplex == DUPLEX_FULL)
+                               hw->media_type = MEDIA_TYPE_10M_FULL;
+                       else
+                               hw->media_type = MEDIA_TYPE_10M_HALF;
+               }
+       }
+       switch (hw->media_type) {
+       case MEDIA_TYPE_AUTO_SENSOR:
+               ecmd->advertising =
+                   ADVERTISED_10baseT_Half |
+                   ADVERTISED_10baseT_Full |
+                   ADVERTISED_100baseT_Half |
+                   ADVERTISED_100baseT_Full |
+                   ADVERTISED_1000baseT_Full |
+                   ADVERTISED_Autoneg | ADVERTISED_TP;
+               break;
+       case MEDIA_TYPE_1000M_FULL:
+               ecmd->advertising =
+                   ADVERTISED_1000baseT_Full |
+                   ADVERTISED_Autoneg | ADVERTISED_TP;
+               break;
+       default:
+               ecmd->advertising = 0;
+               break;
+       }
+       if (atl1_phy_setup_autoneg_adv(hw)) {
+               ret_val = -EINVAL;
+               if (netif_msg_link(adapter))
+                       dev_warn(&adapter->pdev->dev,
+                               "invalid ethtool speed/duplex setting\n");
+               goto exit_sset;
+       }
+       if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+           hw->media_type == MEDIA_TYPE_1000M_FULL)
+               phy_data = MII_CR_RESET | MII_CR_AUTO_NEG_EN;
+       else {
+               switch (hw->media_type) {
+               case MEDIA_TYPE_100M_FULL:
+                       phy_data =
+                           MII_CR_FULL_DUPLEX | MII_CR_SPEED_100 |
+                           MII_CR_RESET;
+                       break;
+               case MEDIA_TYPE_100M_HALF:
+                       phy_data = MII_CR_SPEED_100 | MII_CR_RESET;
+                       break;
+               case MEDIA_TYPE_10M_FULL:
+                       phy_data =
+                           MII_CR_FULL_DUPLEX | MII_CR_SPEED_10 | MII_CR_RESET;
+                       break;
+               default:
+                       /* MEDIA_TYPE_10M_HALF: */
+                       phy_data = MII_CR_SPEED_10 | MII_CR_RESET;
+                       break;
+               }
+       }
+       atl1_write_phy_reg(hw, MII_BMCR, phy_data);
+ exit_sset:
+       if (ret_val)
+               hw->media_type = old_media_type;
+       if (netif_running(adapter->netdev)) {
+               if (netif_msg_link(adapter))
+                       dev_dbg(&adapter->pdev->dev,
+                               "ethtool starting adapter\n");
+               atl1_up(adapter);
+       } else if (!ret_val) {
+               if (netif_msg_link(adapter))
+                       dev_dbg(&adapter->pdev->dev,
+                               "ethtool resetting adapter\n");
+               atl1_reset(adapter);
+       }
+       return ret_val;
+ }
+ static void atl1_get_drvinfo(struct net_device *netdev,
+       struct ethtool_drvinfo *drvinfo)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       strncpy(drvinfo->driver, ATLX_DRIVER_NAME, sizeof(drvinfo->driver));
+       strncpy(drvinfo->version, ATLX_DRIVER_VERSION,
+               sizeof(drvinfo->version));
+       strncpy(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
+       strncpy(drvinfo->bus_info, pci_name(adapter->pdev),
+               sizeof(drvinfo->bus_info));
+       drvinfo->eedump_len = ATL1_EEDUMP_LEN;
+ }
+ static void atl1_get_wol(struct net_device *netdev,
+       struct ethtool_wolinfo *wol)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       wol->supported = WAKE_UCAST | WAKE_MCAST | WAKE_BCAST | WAKE_MAGIC;
+       wol->wolopts = 0;
+       if (adapter->wol & ATLX_WUFC_EX)
+               wol->wolopts |= WAKE_UCAST;
+       if (adapter->wol & ATLX_WUFC_MC)
+               wol->wolopts |= WAKE_MCAST;
+       if (adapter->wol & ATLX_WUFC_BC)
+               wol->wolopts |= WAKE_BCAST;
+       if (adapter->wol & ATLX_WUFC_MAG)
+               wol->wolopts |= WAKE_MAGIC;
+       return;
+ }
+ static int atl1_set_wol(struct net_device *netdev,
+       struct ethtool_wolinfo *wol)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       if (wol->wolopts & (WAKE_PHY | WAKE_ARP | WAKE_MAGICSECURE))
+               return -EOPNOTSUPP;
+       adapter->wol = 0;
+       if (wol->wolopts & WAKE_UCAST)
+               adapter->wol |= ATLX_WUFC_EX;
+       if (wol->wolopts & WAKE_MCAST)
+               adapter->wol |= ATLX_WUFC_MC;
+       if (wol->wolopts & WAKE_BCAST)
+               adapter->wol |= ATLX_WUFC_BC;
+       if (wol->wolopts & WAKE_MAGIC)
+               adapter->wol |= ATLX_WUFC_MAG;
+       return 0;
+ }
+ static u32 atl1_get_msglevel(struct net_device *netdev)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       return adapter->msg_enable;
+ }
+ static void atl1_set_msglevel(struct net_device *netdev, u32 value)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       adapter->msg_enable = value;
+ }
+ static int atl1_get_regs_len(struct net_device *netdev)
+ {
+       return ATL1_REG_COUNT * sizeof(u32);
+ }
+ static void atl1_get_regs(struct net_device *netdev, struct ethtool_regs *regs,
+       void *p)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       unsigned int i;
+       u32 *regbuf = p;
+       for (i = 0; i < ATL1_REG_COUNT; i++) {
+               /*
+                * This switch statement avoids reserved regions
+                * of register space.
+                */
+               switch (i) {
+               case 6 ... 9:
+               case 14:
+               case 29 ... 31:
+               case 34 ... 63:
+               case 75 ... 127:
+               case 136 ... 1023:
+               case 1027 ... 1087:
+               case 1091 ... 1151:
+               case 1194 ... 1195:
+               case 1200 ... 1201:
+               case 1206 ... 1213:
+               case 1216 ... 1279:
+               case 1290 ... 1311:
+               case 1323 ... 1343:
+               case 1358 ... 1359:
+               case 1368 ... 1375:
+               case 1378 ... 1383:
+               case 1388 ... 1391:
+               case 1393 ... 1395:
+               case 1402 ... 1403:
+               case 1410 ... 1471:
+               case 1522 ... 1535:
+                       /* reserved region; don't read it */
+                       regbuf[i] = 0;
+                       break;
+               default:
+                       /* unreserved region */
+                       regbuf[i] = ioread32(hw->hw_addr + (i * sizeof(u32)));
+               }
+       }
+ }
+ static void atl1_get_ringparam(struct net_device *netdev,
+       struct ethtool_ringparam *ring)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_tpd_ring *txdr = &adapter->tpd_ring;
+       struct atl1_rfd_ring *rxdr = &adapter->rfd_ring;
+       ring->rx_max_pending = ATL1_MAX_RFD;
+       ring->tx_max_pending = ATL1_MAX_TPD;
+       ring->rx_mini_max_pending = 0;
+       ring->rx_jumbo_max_pending = 0;
+       ring->rx_pending = rxdr->count;
+       ring->tx_pending = txdr->count;
+       ring->rx_mini_pending = 0;
+       ring->rx_jumbo_pending = 0;
+ }
+ static int atl1_set_ringparam(struct net_device *netdev,
+       struct ethtool_ringparam *ring)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_tpd_ring *tpdr = &adapter->tpd_ring;
+       struct atl1_rrd_ring *rrdr = &adapter->rrd_ring;
+       struct atl1_rfd_ring *rfdr = &adapter->rfd_ring;
+       struct atl1_tpd_ring tpd_old, tpd_new;
+       struct atl1_rfd_ring rfd_old, rfd_new;
+       struct atl1_rrd_ring rrd_old, rrd_new;
+       struct atl1_ring_header rhdr_old, rhdr_new;
+       int err;
+       tpd_old = adapter->tpd_ring;
+       rfd_old = adapter->rfd_ring;
+       rrd_old = adapter->rrd_ring;
+       rhdr_old = adapter->ring_header;
+       if (netif_running(adapter->netdev))
+               atl1_down(adapter);
+       rfdr->count = (u16) max(ring->rx_pending, (u32) ATL1_MIN_RFD);
+       rfdr->count = rfdr->count > ATL1_MAX_RFD ? ATL1_MAX_RFD :
+                       rfdr->count;
+       rfdr->count = (rfdr->count + 3) & ~3;
+       rrdr->count = rfdr->count;
+       tpdr->count = (u16) max(ring->tx_pending, (u32) ATL1_MIN_TPD);
+       tpdr->count = tpdr->count > ATL1_MAX_TPD ? ATL1_MAX_TPD :
+                       tpdr->count;
+       tpdr->count = (tpdr->count + 3) & ~3;
+       if (netif_running(adapter->netdev)) {
+               /* try to get new resources before deleting old */
+               err = atl1_setup_ring_resources(adapter);
+               if (err)
+                       goto err_setup_ring;
+               /*
+                * save the new, restore the old in order to free it,
+                * then restore the new back again
+                */
+               rfd_new = adapter->rfd_ring;
+               rrd_new = adapter->rrd_ring;
+               tpd_new = adapter->tpd_ring;
+               rhdr_new = adapter->ring_header;
+               adapter->rfd_ring = rfd_old;
+               adapter->rrd_ring = rrd_old;
+               adapter->tpd_ring = tpd_old;
+               adapter->ring_header = rhdr_old;
+               atl1_free_ring_resources(adapter);
+               adapter->rfd_ring = rfd_new;
+               adapter->rrd_ring = rrd_new;
+               adapter->tpd_ring = tpd_new;
+               adapter->ring_header = rhdr_new;
+               err = atl1_up(adapter);
+               if (err)
+                       return err;
+       }
+       return 0;
+ err_setup_ring:
+       adapter->rfd_ring = rfd_old;
+       adapter->rrd_ring = rrd_old;
+       adapter->tpd_ring = tpd_old;
+       adapter->ring_header = rhdr_old;
+       atl1_up(adapter);
+       return err;
+ }
+ static void atl1_get_pauseparam(struct net_device *netdev,
+       struct ethtool_pauseparam *epause)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+           hw->media_type == MEDIA_TYPE_1000M_FULL) {
+               epause->autoneg = AUTONEG_ENABLE;
+       } else {
+               epause->autoneg = AUTONEG_DISABLE;
+       }
+       epause->rx_pause = 1;
+       epause->tx_pause = 1;
+ }
+ static int atl1_set_pauseparam(struct net_device *netdev,
+       struct ethtool_pauseparam *epause)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+           hw->media_type == MEDIA_TYPE_1000M_FULL) {
+               epause->autoneg = AUTONEG_ENABLE;
+       } else {
+               epause->autoneg = AUTONEG_DISABLE;
+       }
+       epause->rx_pause = 1;
+       epause->tx_pause = 1;
+       return 0;
+ }
+ /* FIXME: is this right? -- CHS */
+ static u32 atl1_get_rx_csum(struct net_device *netdev)
+ {
+       return 1;
+ }
+ static void atl1_get_strings(struct net_device *netdev, u32 stringset,
+       u8 *data)
+ {
+       u8 *p = data;
+       int i;
+       switch (stringset) {
+       case ETH_SS_STATS:
+               for (i = 0; i < ARRAY_SIZE(atl1_gstrings_stats); i++) {
+                       memcpy(p, atl1_gstrings_stats[i].stat_string,
+                               ETH_GSTRING_LEN);
+                       p += ETH_GSTRING_LEN;
+               }
+               break;
+       }
+ }
+ static int atl1_nway_reset(struct net_device *netdev)
+ {
+       struct atl1_adapter *adapter = netdev_priv(netdev);
+       struct atl1_hw *hw = &adapter->hw;
+       if (netif_running(netdev)) {
+               u16 phy_data;
+               atl1_down(adapter);
+               if (hw->media_type == MEDIA_TYPE_AUTO_SENSOR ||
+                       hw->media_type == MEDIA_TYPE_1000M_FULL) {
+                       phy_data = MII_CR_RESET | MII_CR_AUTO_NEG_EN;
+               } else {
+                       switch (hw->media_type) {
+                       case MEDIA_TYPE_100M_FULL:
+                               phy_data = MII_CR_FULL_DUPLEX |
+                                       MII_CR_SPEED_100 | MII_CR_RESET;
+                               break;
+                       case MEDIA_TYPE_100M_HALF:
+                               phy_data = MII_CR_SPEED_100 | MII_CR_RESET;
+                               break;
+                       case MEDIA_TYPE_10M_FULL:
+                               phy_data = MII_CR_FULL_DUPLEX |
+                                       MII_CR_SPEED_10 | MII_CR_RESET;
+                               break;
+                       default:
+                               /* MEDIA_TYPE_10M_HALF */
+                               phy_data = MII_CR_SPEED_10 | MII_CR_RESET;
+                       }
+               }
+               atl1_write_phy_reg(hw, MII_BMCR, phy_data);
+               atl1_up(adapter);
+       }
+       return 0;
+ }
+ const struct ethtool_ops atl1_ethtool_ops = {
+       .get_settings           = atl1_get_settings,
+       .set_settings           = atl1_set_settings,
+       .get_drvinfo            = atl1_get_drvinfo,
+       .get_wol                = atl1_get_wol,
+       .set_wol                = atl1_set_wol,
+       .get_msglevel           = atl1_get_msglevel,
+       .set_msglevel           = atl1_set_msglevel,
+       .get_regs_len           = atl1_get_regs_len,
+       .get_regs               = atl1_get_regs,
+       .get_ringparam          = atl1_get_ringparam,
+       .set_ringparam          = atl1_set_ringparam,
+       .get_pauseparam         = atl1_get_pauseparam,
+       .set_pauseparam         = atl1_set_pauseparam,
+       .get_rx_csum            = atl1_get_rx_csum,
+       .set_tx_csum            = ethtool_op_set_tx_hw_csum,
+       .get_link               = ethtool_op_get_link,
+       .set_sg                 = ethtool_op_set_sg,
+       .get_strings            = atl1_get_strings,
+       .nway_reset             = atl1_nway_reset,
+       .get_ethtool_stats      = atl1_get_ethtool_stats,
+       .get_sset_count         = atl1_get_sset_count,
+       .set_tso                = ethtool_op_set_tso,
+ };
index 0000000000000000000000000000000000000000,cfa199a9df000e30c896313677afd8787be7f2d5..a856cb47fc7839d35518b7b599a2b5042ead0aa7
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,3388 +1,3391 @@@
 -                              unsigned long event, void *ptr)
+ /*
+  *  drivers/s390/net/qeth_l3_main.c
+  *
+  *    Copyright IBM Corp. 2007
+  *    Author(s): Utz Bacher <utz.bacher@de.ibm.com>,
+  *             Frank Pavlic <fpavlic@de.ibm.com>,
+  *             Thomas Spatzier <tspat@de.ibm.com>,
+  *             Frank Blaschka <frank.blaschka@de.ibm.com>
+  */
+ #include <linux/module.h>
+ #include <linux/moduleparam.h>
+ #include <linux/string.h>
+ #include <linux/errno.h>
+ #include <linux/kernel.h>
+ #include <linux/etherdevice.h>
+ #include <linux/mii.h>
+ #include <linux/ip.h>
+ #include <linux/reboot.h>
+ #include <linux/inetdevice.h>
+ #include <linux/igmp.h>
+ #include <net/ip.h>
+ #include <net/arp.h>
+ #include <asm/s390_rdev.h>
+ #include "qeth_l3.h"
+ #include "qeth_core_offl.h"
+ DEFINE_PER_CPU(char[256], qeth_l3_dbf_txt_buf);
+ static int qeth_l3_set_offline(struct ccwgroup_device *);
+ static int qeth_l3_recover(void *);
+ static int qeth_l3_stop(struct net_device *);
+ static void qeth_l3_set_multicast_list(struct net_device *);
+ static int qeth_l3_neigh_setup(struct net_device *, struct neigh_parms *);
+ static int qeth_l3_register_addr_entry(struct qeth_card *,
+               struct qeth_ipaddr *);
+ static int qeth_l3_deregister_addr_entry(struct qeth_card *,
+               struct qeth_ipaddr *);
+ static int __qeth_l3_set_online(struct ccwgroup_device *, int);
+ static int __qeth_l3_set_offline(struct ccwgroup_device *, int);
+ static int qeth_l3_isxdigit(char *buf)
+ {
+       while (*buf) {
+               if (!isxdigit(*buf++))
+                       return 0;
+       }
+       return 1;
+ }
+ void qeth_l3_ipaddr4_to_string(const __u8 *addr, char *buf)
+ {
+       sprintf(buf, "%i.%i.%i.%i", addr[0], addr[1], addr[2], addr[3]);
+ }
+ int qeth_l3_string_to_ipaddr4(const char *buf, __u8 *addr)
+ {
+       int count = 0, rc = 0;
+       int in[4];
+       char c;
+       rc = sscanf(buf, "%u.%u.%u.%u%c",
+                   &in[0], &in[1], &in[2], &in[3], &c);
+       if (rc != 4 && (rc != 5 || c != '\n'))
+               return -EINVAL;
+       for (count = 0; count < 4; count++) {
+               if (in[count] > 255)
+                       return -EINVAL;
+               addr[count] = in[count];
+       }
+       return 0;
+ }
+ void qeth_l3_ipaddr6_to_string(const __u8 *addr, char *buf)
+ {
+       sprintf(buf, "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
+                    ":%02x%02x:%02x%02x:%02x%02x:%02x%02x",
+                    addr[0], addr[1], addr[2], addr[3],
+                    addr[4], addr[5], addr[6], addr[7],
+                    addr[8], addr[9], addr[10], addr[11],
+                    addr[12], addr[13], addr[14], addr[15]);
+ }
+ int qeth_l3_string_to_ipaddr6(const char *buf, __u8 *addr)
+ {
+       const char *end, *end_tmp, *start;
+       __u16 *in;
+       char num[5];
+       int num2, cnt, out, found, save_cnt;
+       unsigned short in_tmp[8] = {0, };
+       cnt = out = found = save_cnt = num2 = 0;
+       end = start = buf;
+       in = (__u16 *) addr;
+       memset(in, 0, 16);
+       while (*end) {
+               end = strchr(start, ':');
+               if (end == NULL) {
+                       end = buf + strlen(buf);
+                       end_tmp = strchr(start, '\n');
+                       if (end_tmp != NULL)
+                               end = end_tmp;
+                       out = 1;
+               }
+               if ((end - start)) {
+                       memset(num, 0, 5);
+                       if ((end - start) > 4)
+                               return -EINVAL;
+                       memcpy(num, start, end - start);
+                       if (!qeth_l3_isxdigit(num))
+                               return -EINVAL;
+                       sscanf(start, "%x", &num2);
+                       if (found)
+                               in_tmp[save_cnt++] = num2;
+                       else
+                               in[cnt++] = num2;
+                       if (out)
+                               break;
+               } else {
+                       if (found)
+                               return -EINVAL;
+                       found = 1;
+               }
+               start = ++end;
+       }
+       if (cnt + save_cnt > 8)
+               return -EINVAL;
+       cnt = 7;
+       while (save_cnt)
+               in[cnt--] = in_tmp[--save_cnt];
+       return 0;
+ }
+ void qeth_l3_ipaddr_to_string(enum qeth_prot_versions proto, const __u8 *addr,
+                               char *buf)
+ {
+       if (proto == QETH_PROT_IPV4)
+               qeth_l3_ipaddr4_to_string(addr, buf);
+       else if (proto == QETH_PROT_IPV6)
+               qeth_l3_ipaddr6_to_string(addr, buf);
+ }
+ int qeth_l3_string_to_ipaddr(const char *buf, enum qeth_prot_versions proto,
+                               __u8 *addr)
+ {
+       if (proto == QETH_PROT_IPV4)
+               return qeth_l3_string_to_ipaddr4(buf, addr);
+       else if (proto == QETH_PROT_IPV6)
+               return qeth_l3_string_to_ipaddr6(buf, addr);
+       else
+               return -EINVAL;
+ }
+ static void qeth_l3_convert_addr_to_bits(u8 *addr, u8 *bits, int len)
+ {
+       int i, j;
+       u8 octet;
+       for (i = 0; i < len; ++i) {
+               octet = addr[i];
+               for (j = 7; j >= 0; --j) {
+                       bits[i*8 + j] = octet & 1;
+                       octet >>= 1;
+               }
+       }
+ }
+ static int qeth_l3_is_addr_covered_by_ipato(struct qeth_card *card,
+                                               struct qeth_ipaddr *addr)
+ {
+       struct qeth_ipato_entry *ipatoe;
+       u8 addr_bits[128] = {0, };
+       u8 ipatoe_bits[128] = {0, };
+       int rc = 0;
+       if (!card->ipato.enabled)
+               return 0;
+       qeth_l3_convert_addr_to_bits((u8 *) &addr->u, addr_bits,
+                                 (addr->proto == QETH_PROT_IPV4)? 4:16);
+       list_for_each_entry(ipatoe, &card->ipato.entries, entry) {
+               if (addr->proto != ipatoe->proto)
+                       continue;
+               qeth_l3_convert_addr_to_bits(ipatoe->addr, ipatoe_bits,
+                                         (ipatoe->proto == QETH_PROT_IPV4) ?
+                                         4 : 16);
+               if (addr->proto == QETH_PROT_IPV4)
+                       rc = !memcmp(addr_bits, ipatoe_bits,
+                                    min(32, ipatoe->mask_bits));
+               else
+                       rc = !memcmp(addr_bits, ipatoe_bits,
+                                    min(128, ipatoe->mask_bits));
+               if (rc)
+                       break;
+       }
+       /* invert? */
+       if ((addr->proto == QETH_PROT_IPV4) && card->ipato.invert4)
+               rc = !rc;
+       else if ((addr->proto == QETH_PROT_IPV6) && card->ipato.invert6)
+               rc = !rc;
+       return rc;
+ }
+ /*
+  * Add IP to be added to todo list. If there is already an "add todo"
+  * in this list we just incremenent the reference count.
+  * Returns 0 if we  just incremented reference count.
+  */
+ static int __qeth_l3_insert_ip_todo(struct qeth_card *card,
+                                       struct qeth_ipaddr *addr, int add)
+ {
+       struct qeth_ipaddr *tmp, *t;
+       int found = 0;
+       list_for_each_entry_safe(tmp, t, card->ip_tbd_list, entry) {
+               if ((addr->type == QETH_IP_TYPE_DEL_ALL_MC) &&
+                   (tmp->type == QETH_IP_TYPE_DEL_ALL_MC))
+                       return 0;
+               if ((tmp->proto        == QETH_PROT_IPV4)     &&
+                   (addr->proto       == QETH_PROT_IPV4)     &&
+                   (tmp->type         == addr->type)         &&
+                   (tmp->is_multicast == addr->is_multicast) &&
+                   (tmp->u.a4.addr    == addr->u.a4.addr)    &&
+                   (tmp->u.a4.mask    == addr->u.a4.mask)) {
+                       found = 1;
+                       break;
+               }
+               if ((tmp->proto        == QETH_PROT_IPV6)      &&
+                   (addr->proto       == QETH_PROT_IPV6)      &&
+                   (tmp->type         == addr->type)          &&
+                   (tmp->is_multicast == addr->is_multicast)  &&
+                   (tmp->u.a6.pfxlen  == addr->u.a6.pfxlen)   &&
+                   (memcmp(&tmp->u.a6.addr, &addr->u.a6.addr,
+                           sizeof(struct in6_addr)) == 0)) {
+                       found = 1;
+                       break;
+               }
+       }
+       if (found) {
+               if (addr->users != 0)
+                       tmp->users += addr->users;
+               else
+                       tmp->users += add ? 1 : -1;
+               if (tmp->users == 0) {
+                       list_del(&tmp->entry);
+                       kfree(tmp);
+               }
+               return 0;
+       } else {
+               if (addr->type == QETH_IP_TYPE_DEL_ALL_MC)
+                       list_add(&addr->entry, card->ip_tbd_list);
+               else {
+                       if (addr->users == 0)
+                               addr->users += add ? 1 : -1;
+                       if (add && (addr->type == QETH_IP_TYPE_NORMAL) &&
+                           qeth_l3_is_addr_covered_by_ipato(card, addr)) {
+                               QETH_DBF_TEXT(trace, 2, "tkovaddr");
+                               addr->set_flags |= QETH_IPA_SETIP_TAKEOVER_FLAG;
+                       }
+                       list_add_tail(&addr->entry, card->ip_tbd_list);
+               }
+               return 1;
+       }
+ }
+ static int qeth_l3_delete_ip(struct qeth_card *card, struct qeth_ipaddr *addr)
+ {
+       unsigned long flags;
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 4, "delip");
+       if (addr->proto == QETH_PROT_IPV4)
+               QETH_DBF_HEX(trace, 4, &addr->u.a4.addr, 4);
+       else {
+               QETH_DBF_HEX(trace, 4, &addr->u.a6.addr, 8);
+               QETH_DBF_HEX(trace, 4, ((char *)&addr->u.a6.addr) + 8, 8);
+       }
+       spin_lock_irqsave(&card->ip_lock, flags);
+       rc = __qeth_l3_insert_ip_todo(card, addr, 0);
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+       return rc;
+ }
+ static int qeth_l3_add_ip(struct qeth_card *card, struct qeth_ipaddr *addr)
+ {
+       unsigned long flags;
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 4, "addip");
+       if (addr->proto == QETH_PROT_IPV4)
+               QETH_DBF_HEX(trace, 4, &addr->u.a4.addr, 4);
+       else {
+               QETH_DBF_HEX(trace, 4, &addr->u.a6.addr, 8);
+               QETH_DBF_HEX(trace, 4, ((char *)&addr->u.a6.addr) + 8, 8);
+       }
+       spin_lock_irqsave(&card->ip_lock, flags);
+       rc = __qeth_l3_insert_ip_todo(card, addr, 1);
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+       return rc;
+ }
+ static struct qeth_ipaddr *qeth_l3_get_addr_buffer(
+                               enum qeth_prot_versions prot)
+ {
+       struct qeth_ipaddr *addr;
+       addr = kzalloc(sizeof(struct qeth_ipaddr), GFP_ATOMIC);
+       if (addr == NULL) {
+               PRINT_WARN("Not enough memory to add address\n");
+               return NULL;
+       }
+       addr->type = QETH_IP_TYPE_NORMAL;
+       addr->proto = prot;
+       return addr;
+ }
+ static void qeth_l3_delete_mc_addresses(struct qeth_card *card)
+ {
+       struct qeth_ipaddr *iptodo;
+       unsigned long flags;
+       QETH_DBF_TEXT(trace, 4, "delmc");
+       iptodo = qeth_l3_get_addr_buffer(QETH_PROT_IPV4);
+       if (!iptodo) {
+               QETH_DBF_TEXT(trace, 2, "dmcnomem");
+               return;
+       }
+       iptodo->type = QETH_IP_TYPE_DEL_ALL_MC;
+       spin_lock_irqsave(&card->ip_lock, flags);
+       if (!__qeth_l3_insert_ip_todo(card, iptodo, 0))
+               kfree(iptodo);
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+ }
+ /*
+  * Add/remove address to/from card's ip list, i.e. try to add or remove
+  * reference to/from an IP address that is already registered on the card.
+  * Returns:
+  *    0  address was on card and its reference count has been adjusted,
+  *       but is still > 0, so nothing has to be done
+  *       also returns 0 if card was not on card and the todo was to delete
+  *       the address -> there is also nothing to be done
+  *    1  address was not on card and the todo is to add it to the card's ip
+  *       list
+  *    -1 address was on card and its reference count has been decremented
+  *       to <= 0 by the todo -> address must be removed from card
+  */
+ static int __qeth_l3_ref_ip_on_card(struct qeth_card *card,
+               struct qeth_ipaddr *todo, struct qeth_ipaddr **__addr)
+ {
+       struct qeth_ipaddr *addr;
+       int found = 0;
+       list_for_each_entry(addr, &card->ip_list, entry) {
+               if ((addr->proto == QETH_PROT_IPV4) &&
+                   (todo->proto == QETH_PROT_IPV4) &&
+                   (addr->type == todo->type) &&
+                   (addr->u.a4.addr == todo->u.a4.addr) &&
+                   (addr->u.a4.mask == todo->u.a4.mask)) {
+                       found = 1;
+                       break;
+               }
+               if ((addr->proto == QETH_PROT_IPV6) &&
+                   (todo->proto == QETH_PROT_IPV6) &&
+                   (addr->type == todo->type) &&
+                   (addr->u.a6.pfxlen == todo->u.a6.pfxlen) &&
+                   (memcmp(&addr->u.a6.addr, &todo->u.a6.addr,
+                           sizeof(struct in6_addr)) == 0)) {
+                       found = 1;
+                       break;
+               }
+       }
+       if (found) {
+               addr->users += todo->users;
+               if (addr->users <= 0) {
+                       *__addr = addr;
+                       return -1;
+               } else {
+                       /* for VIPA and RXIP limit refcount to 1 */
+                       if (addr->type != QETH_IP_TYPE_NORMAL)
+                               addr->users = 1;
+                       return 0;
+               }
+       }
+       if (todo->users > 0) {
+               /* for VIPA and RXIP limit refcount to 1 */
+               if (todo->type != QETH_IP_TYPE_NORMAL)
+                       todo->users = 1;
+               return 1;
+       } else
+               return 0;
+ }
+ static void __qeth_l3_delete_all_mc(struct qeth_card *card,
+                                       unsigned long *flags)
+ {
+       struct qeth_ipaddr *addr, *tmp;
+       int rc;
+ again:
+       list_for_each_entry_safe(addr, tmp, &card->ip_list, entry) {
+               if (addr->is_multicast) {
+                       list_del(&addr->entry);
+                       spin_unlock_irqrestore(&card->ip_lock, *flags);
+                       rc = qeth_l3_deregister_addr_entry(card, addr);
+                       spin_lock_irqsave(&card->ip_lock, *flags);
+                       if (!rc) {
+                               kfree(addr);
+                               goto again;
+                       } else
+                               list_add(&addr->entry, &card->ip_list);
+               }
+       }
+ }
+ static void qeth_l3_set_ip_addr_list(struct qeth_card *card)
+ {
+       struct list_head *tbd_list;
+       struct qeth_ipaddr *todo, *addr;
+       unsigned long flags;
+       int rc;
+       QETH_DBF_TEXT(trace, 2, "sdiplist");
+       QETH_DBF_HEX(trace, 2, &card, sizeof(void *));
+       spin_lock_irqsave(&card->ip_lock, flags);
+       tbd_list = card->ip_tbd_list;
+       card->ip_tbd_list = kmalloc(sizeof(struct list_head), GFP_ATOMIC);
+       if (!card->ip_tbd_list) {
+               QETH_DBF_TEXT(trace, 0, "silnomem");
+               card->ip_tbd_list = tbd_list;
+               spin_unlock_irqrestore(&card->ip_lock, flags);
+               return;
+       } else
+               INIT_LIST_HEAD(card->ip_tbd_list);
+       while (!list_empty(tbd_list)) {
+               todo = list_entry(tbd_list->next, struct qeth_ipaddr, entry);
+               list_del(&todo->entry);
+               if (todo->type == QETH_IP_TYPE_DEL_ALL_MC) {
+                       __qeth_l3_delete_all_mc(card, &flags);
+                       kfree(todo);
+                       continue;
+               }
+               rc = __qeth_l3_ref_ip_on_card(card, todo, &addr);
+               if (rc == 0) {
+                       /* nothing to be done; only adjusted refcount */
+                       kfree(todo);
+               } else if (rc == 1) {
+                       /* new entry to be added to on-card list */
+                       spin_unlock_irqrestore(&card->ip_lock, flags);
+                       rc = qeth_l3_register_addr_entry(card, todo);
+                       spin_lock_irqsave(&card->ip_lock, flags);
+                       if (!rc)
+                               list_add_tail(&todo->entry, &card->ip_list);
+                       else
+                               kfree(todo);
+               } else if (rc == -1) {
+                       /* on-card entry to be removed */
+                       list_del_init(&addr->entry);
+                       spin_unlock_irqrestore(&card->ip_lock, flags);
+                       rc = qeth_l3_deregister_addr_entry(card, addr);
+                       spin_lock_irqsave(&card->ip_lock, flags);
+                       if (!rc)
+                               kfree(addr);
+                       else
+                               list_add_tail(&addr->entry, &card->ip_list);
+                       kfree(todo);
+               }
+       }
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+       kfree(tbd_list);
+ }
+ static void qeth_l3_clear_ip_list(struct qeth_card *card, int clean,
+                                       int recover)
+ {
+       struct qeth_ipaddr *addr, *tmp;
+       unsigned long flags;
+       QETH_DBF_TEXT(trace, 4, "clearip");
+       spin_lock_irqsave(&card->ip_lock, flags);
+       /* clear todo list */
+       list_for_each_entry_safe(addr, tmp, card->ip_tbd_list, entry) {
+               list_del(&addr->entry);
+               kfree(addr);
+       }
+       while (!list_empty(&card->ip_list)) {
+               addr = list_entry(card->ip_list.next,
+                                 struct qeth_ipaddr, entry);
+               list_del_init(&addr->entry);
+               if (clean) {
+                       spin_unlock_irqrestore(&card->ip_lock, flags);
+                       qeth_l3_deregister_addr_entry(card, addr);
+                       spin_lock_irqsave(&card->ip_lock, flags);
+               }
+               if (!recover || addr->is_multicast) {
+                       kfree(addr);
+                       continue;
+               }
+               list_add_tail(&addr->entry, card->ip_tbd_list);
+       }
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+ }
+ static int qeth_l3_address_exists_in_list(struct list_head *list,
+               struct qeth_ipaddr *addr, int same_type)
+ {
+       struct qeth_ipaddr *tmp;
+       list_for_each_entry(tmp, list, entry) {
+               if ((tmp->proto == QETH_PROT_IPV4) &&
+                   (addr->proto == QETH_PROT_IPV4) &&
+                   ((same_type && (tmp->type == addr->type)) ||
+                   (!same_type && (tmp->type != addr->type))) &&
+                   (tmp->u.a4.addr == addr->u.a4.addr))
+                       return 1;
+               if ((tmp->proto == QETH_PROT_IPV6) &&
+                   (addr->proto == QETH_PROT_IPV6) &&
+                   ((same_type && (tmp->type == addr->type)) ||
+                   (!same_type && (tmp->type != addr->type))) &&
+                   (memcmp(&tmp->u.a6.addr, &addr->u.a6.addr,
+                           sizeof(struct in6_addr)) == 0))
+                       return 1;
+       }
+       return 0;
+ }
+ static int qeth_l3_send_setdelmc(struct qeth_card *card,
+                       struct qeth_ipaddr *addr, int ipacmd)
+ {
+       int rc;
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(trace, 4, "setdelmc");
+       iob = qeth_get_ipacmd_buffer(card, ipacmd, addr->proto);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       memcpy(&cmd->data.setdelipm.mac, addr->mac, OSA_ADDR_LEN);
+       if (addr->proto == QETH_PROT_IPV6)
+               memcpy(cmd->data.setdelipm.ip6, &addr->u.a6.addr,
+                      sizeof(struct in6_addr));
+       else
+               memcpy(&cmd->data.setdelipm.ip4, &addr->u.a4.addr, 4);
+       rc = qeth_send_ipa_cmd(card, iob, NULL, NULL);
+       return rc;
+ }
+ static void qeth_l3_fill_netmask(u8 *netmask, unsigned int len)
+ {
+       int i, j;
+       for (i = 0; i < 16; i++) {
+               j = (len) - (i * 8);
+               if (j >= 8)
+                       netmask[i] = 0xff;
+               else if (j > 0)
+                       netmask[i] = (u8)(0xFF00 >> j);
+               else
+                       netmask[i] = 0;
+       }
+ }
+ static int qeth_l3_send_setdelip(struct qeth_card *card,
+               struct qeth_ipaddr *addr, int ipacmd, unsigned int flags)
+ {
+       int rc;
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       __u8 netmask[16];
+       QETH_DBF_TEXT(trace, 4, "setdelip");
+       QETH_DBF_TEXT_(trace, 4, "flags%02X", flags);
+       iob = qeth_get_ipacmd_buffer(card, ipacmd, addr->proto);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       if (addr->proto == QETH_PROT_IPV6) {
+               memcpy(cmd->data.setdelip6.ip_addr, &addr->u.a6.addr,
+                      sizeof(struct in6_addr));
+               qeth_l3_fill_netmask(netmask, addr->u.a6.pfxlen);
+               memcpy(cmd->data.setdelip6.mask, netmask,
+                      sizeof(struct in6_addr));
+               cmd->data.setdelip6.flags = flags;
+       } else {
+               memcpy(cmd->data.setdelip4.ip_addr, &addr->u.a4.addr, 4);
+               memcpy(cmd->data.setdelip4.mask, &addr->u.a4.mask, 4);
+               cmd->data.setdelip4.flags = flags;
+       }
+       rc = qeth_send_ipa_cmd(card, iob, NULL, NULL);
+       return rc;
+ }
+ static int qeth_l3_send_setrouting(struct qeth_card *card,
+       enum qeth_routing_types type, enum qeth_prot_versions prot)
+ {
+       int rc;
+       struct qeth_ipa_cmd *cmd;
+       struct qeth_cmd_buffer *iob;
+       QETH_DBF_TEXT(trace, 4, "setroutg");
+       iob = qeth_get_ipacmd_buffer(card, IPA_CMD_SETRTG, prot);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       cmd->data.setrtg.type = (type);
+       rc = qeth_send_ipa_cmd(card, iob, NULL, NULL);
+       return rc;
+ }
+ static void qeth_l3_correct_routing_type(struct qeth_card *card,
+               enum qeth_routing_types *type, enum qeth_prot_versions prot)
+ {
+       if (card->info.type == QETH_CARD_TYPE_IQD) {
+               switch (*type) {
+               case NO_ROUTER:
+               case PRIMARY_CONNECTOR:
+               case SECONDARY_CONNECTOR:
+               case MULTICAST_ROUTER:
+                       return;
+               default:
+                       goto out_inval;
+               }
+       } else {
+               switch (*type) {
+               case NO_ROUTER:
+               case PRIMARY_ROUTER:
+               case SECONDARY_ROUTER:
+                       return;
+               case MULTICAST_ROUTER:
+                       if (qeth_is_ipafunc_supported(card, prot,
+                                                     IPA_OSA_MC_ROUTER))
+                               return;
+               default:
+                       goto out_inval;
+               }
+       }
+ out_inval:
+       PRINT_WARN("Routing type '%s' not supported for interface %s.\n"
+                  "Router status set to 'no router'.\n",
+                  ((*type == PRIMARY_ROUTER)? "primary router" :
+                   (*type == SECONDARY_ROUTER)? "secondary router" :
+                   (*type == PRIMARY_CONNECTOR)? "primary connector" :
+                   (*type == SECONDARY_CONNECTOR)? "secondary connector" :
+                   (*type == MULTICAST_ROUTER)? "multicast router" :
+                   "unknown"),
+                  card->dev->name);
+       *type = NO_ROUTER;
+ }
+ int qeth_l3_setrouting_v4(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "setrtg4");
+       qeth_l3_correct_routing_type(card, &card->options.route4.type,
+                                 QETH_PROT_IPV4);
+       rc = qeth_l3_send_setrouting(card, card->options.route4.type,
+                                 QETH_PROT_IPV4);
+       if (rc) {
+               card->options.route4.type = NO_ROUTER;
+               PRINT_WARN("Error (0x%04x) while setting routing type on %s. "
+                          "Type set to 'no router'.\n",
+                          rc, QETH_CARD_IFNAME(card));
+       }
+       return rc;
+ }
+ int qeth_l3_setrouting_v6(struct qeth_card *card)
+ {
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 3, "setrtg6");
+ #ifdef CONFIG_QETH_IPV6
+       if (!qeth_is_supported(card, IPA_IPV6))
+               return 0;
+       qeth_l3_correct_routing_type(card, &card->options.route6.type,
+                                 QETH_PROT_IPV6);
+       rc = qeth_l3_send_setrouting(card, card->options.route6.type,
+                                 QETH_PROT_IPV6);
+       if (rc) {
+               card->options.route6.type = NO_ROUTER;
+               PRINT_WARN("Error (0x%04x) while setting routing type on %s. "
+                          "Type set to 'no router'.\n",
+                          rc, QETH_CARD_IFNAME(card));
+       }
+ #endif
+       return rc;
+ }
+ /*
+  * IP address takeover related functions
+  */
+ static void qeth_l3_clear_ipato_list(struct qeth_card *card)
+ {
+       struct qeth_ipato_entry *ipatoe, *tmp;
+       unsigned long flags;
+       spin_lock_irqsave(&card->ip_lock, flags);
+       list_for_each_entry_safe(ipatoe, tmp, &card->ipato.entries, entry) {
+               list_del(&ipatoe->entry);
+               kfree(ipatoe);
+       }
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+ }
+ int qeth_l3_add_ipato_entry(struct qeth_card *card,
+                               struct qeth_ipato_entry *new)
+ {
+       struct qeth_ipato_entry *ipatoe;
+       unsigned long flags;
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 2, "addipato");
+       spin_lock_irqsave(&card->ip_lock, flags);
+       list_for_each_entry(ipatoe, &card->ipato.entries, entry) {
+               if (ipatoe->proto != new->proto)
+                       continue;
+               if (!memcmp(ipatoe->addr, new->addr,
+                           (ipatoe->proto == QETH_PROT_IPV4)? 4:16) &&
+                   (ipatoe->mask_bits == new->mask_bits)) {
+                       PRINT_WARN("ipato entry already exists!\n");
+                       rc = -EEXIST;
+                       break;
+               }
+       }
+       if (!rc)
+               list_add_tail(&new->entry, &card->ipato.entries);
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+       return rc;
+ }
+ void qeth_l3_del_ipato_entry(struct qeth_card *card,
+               enum qeth_prot_versions proto, u8 *addr, int mask_bits)
+ {
+       struct qeth_ipato_entry *ipatoe, *tmp;
+       unsigned long flags;
+       QETH_DBF_TEXT(trace, 2, "delipato");
+       spin_lock_irqsave(&card->ip_lock, flags);
+       list_for_each_entry_safe(ipatoe, tmp, &card->ipato.entries, entry) {
+               if (ipatoe->proto != proto)
+                       continue;
+               if (!memcmp(ipatoe->addr, addr,
+                           (proto == QETH_PROT_IPV4)? 4:16) &&
+                   (ipatoe->mask_bits == mask_bits)) {
+                       list_del(&ipatoe->entry);
+                       kfree(ipatoe);
+               }
+       }
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+ }
+ /*
+  * VIPA related functions
+  */
+ int qeth_l3_add_vipa(struct qeth_card *card, enum qeth_prot_versions proto,
+             const u8 *addr)
+ {
+       struct qeth_ipaddr *ipaddr;
+       unsigned long flags;
+       int rc = 0;
+       ipaddr = qeth_l3_get_addr_buffer(proto);
+       if (ipaddr) {
+               if (proto == QETH_PROT_IPV4) {
+                       QETH_DBF_TEXT(trace, 2, "addvipa4");
+                       memcpy(&ipaddr->u.a4.addr, addr, 4);
+                       ipaddr->u.a4.mask = 0;
+               } else if (proto == QETH_PROT_IPV6) {
+                       QETH_DBF_TEXT(trace, 2, "addvipa6");
+                       memcpy(&ipaddr->u.a6.addr, addr, 16);
+                       ipaddr->u.a6.pfxlen = 0;
+               }
+               ipaddr->type = QETH_IP_TYPE_VIPA;
+               ipaddr->set_flags = QETH_IPA_SETIP_VIPA_FLAG;
+               ipaddr->del_flags = QETH_IPA_DELIP_VIPA_FLAG;
+       } else
+               return -ENOMEM;
+       spin_lock_irqsave(&card->ip_lock, flags);
+       if (qeth_l3_address_exists_in_list(&card->ip_list, ipaddr, 0) ||
+           qeth_l3_address_exists_in_list(card->ip_tbd_list, ipaddr, 0))
+               rc = -EEXIST;
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+       if (rc) {
+               PRINT_WARN("Cannot add VIPA. Address already exists!\n");
+               return rc;
+       }
+       if (!qeth_l3_add_ip(card, ipaddr))
+               kfree(ipaddr);
+       qeth_l3_set_ip_addr_list(card);
+       return rc;
+ }
+ void qeth_l3_del_vipa(struct qeth_card *card, enum qeth_prot_versions proto,
+             const u8 *addr)
+ {
+       struct qeth_ipaddr *ipaddr;
+       ipaddr = qeth_l3_get_addr_buffer(proto);
+       if (ipaddr) {
+               if (proto == QETH_PROT_IPV4) {
+                       QETH_DBF_TEXT(trace, 2, "delvipa4");
+                       memcpy(&ipaddr->u.a4.addr, addr, 4);
+                       ipaddr->u.a4.mask = 0;
+               } else if (proto == QETH_PROT_IPV6) {
+                       QETH_DBF_TEXT(trace, 2, "delvipa6");
+                       memcpy(&ipaddr->u.a6.addr, addr, 16);
+                       ipaddr->u.a6.pfxlen = 0;
+               }
+               ipaddr->type = QETH_IP_TYPE_VIPA;
+       } else
+               return;
+       if (!qeth_l3_delete_ip(card, ipaddr))
+               kfree(ipaddr);
+       qeth_l3_set_ip_addr_list(card);
+ }
+ /*
+  * proxy ARP related functions
+  */
+ int qeth_l3_add_rxip(struct qeth_card *card, enum qeth_prot_versions proto,
+             const u8 *addr)
+ {
+       struct qeth_ipaddr *ipaddr;
+       unsigned long flags;
+       int rc = 0;
+       ipaddr = qeth_l3_get_addr_buffer(proto);
+       if (ipaddr) {
+               if (proto == QETH_PROT_IPV4) {
+                       QETH_DBF_TEXT(trace, 2, "addrxip4");
+                       memcpy(&ipaddr->u.a4.addr, addr, 4);
+                       ipaddr->u.a4.mask = 0;
+               } else if (proto == QETH_PROT_IPV6) {
+                       QETH_DBF_TEXT(trace, 2, "addrxip6");
+                       memcpy(&ipaddr->u.a6.addr, addr, 16);
+                       ipaddr->u.a6.pfxlen = 0;
+               }
+               ipaddr->type = QETH_IP_TYPE_RXIP;
+               ipaddr->set_flags = QETH_IPA_SETIP_TAKEOVER_FLAG;
+               ipaddr->del_flags = 0;
+       } else
+               return -ENOMEM;
+       spin_lock_irqsave(&card->ip_lock, flags);
+       if (qeth_l3_address_exists_in_list(&card->ip_list, ipaddr, 0) ||
+           qeth_l3_address_exists_in_list(card->ip_tbd_list, ipaddr, 0))
+               rc = -EEXIST;
+       spin_unlock_irqrestore(&card->ip_lock, flags);
+       if (rc) {
+               PRINT_WARN("Cannot add RXIP. Address already exists!\n");
+               return rc;
+       }
+       if (!qeth_l3_add_ip(card, ipaddr))
+               kfree(ipaddr);
+       qeth_l3_set_ip_addr_list(card);
+       return 0;
+ }
+ void qeth_l3_del_rxip(struct qeth_card *card, enum qeth_prot_versions proto,
+                       const u8 *addr)
+ {
+       struct qeth_ipaddr *ipaddr;
+       ipaddr = qeth_l3_get_addr_buffer(proto);
+       if (ipaddr) {
+               if (proto == QETH_PROT_IPV4) {
+                       QETH_DBF_TEXT(trace, 2, "addrxip4");
+                       memcpy(&ipaddr->u.a4.addr, addr, 4);
+                       ipaddr->u.a4.mask = 0;
+               } else if (proto == QETH_PROT_IPV6) {
+                       QETH_DBF_TEXT(trace, 2, "addrxip6");
+                       memcpy(&ipaddr->u.a6.addr, addr, 16);
+                       ipaddr->u.a6.pfxlen = 0;
+               }
+               ipaddr->type = QETH_IP_TYPE_RXIP;
+       } else
+               return;
+       if (!qeth_l3_delete_ip(card, ipaddr))
+               kfree(ipaddr);
+       qeth_l3_set_ip_addr_list(card);
+ }
+ static int qeth_l3_register_addr_entry(struct qeth_card *card,
+                               struct qeth_ipaddr *addr)
+ {
+       char buf[50];
+       int rc = 0;
+       int cnt = 3;
+       if (addr->proto == QETH_PROT_IPV4) {
+               QETH_DBF_TEXT(trace, 2, "setaddr4");
+               QETH_DBF_HEX(trace, 3, &addr->u.a4.addr, sizeof(int));
+       } else if (addr->proto == QETH_PROT_IPV6) {
+               QETH_DBF_TEXT(trace, 2, "setaddr6");
+               QETH_DBF_HEX(trace, 3, &addr->u.a6.addr, 8);
+               QETH_DBF_HEX(trace, 3, ((char *)&addr->u.a6.addr) + 8, 8);
+       } else {
+               QETH_DBF_TEXT(trace, 2, "setaddr?");
+               QETH_DBF_HEX(trace, 3, addr, sizeof(struct qeth_ipaddr));
+       }
+       do {
+               if (addr->is_multicast)
+                       rc =  qeth_l3_send_setdelmc(card, addr, IPA_CMD_SETIPM);
+               else
+                       rc = qeth_l3_send_setdelip(card, addr, IPA_CMD_SETIP,
+                                       addr->set_flags);
+               if (rc)
+                       QETH_DBF_TEXT(trace, 2, "failed");
+       } while ((--cnt > 0) && rc);
+       if (rc) {
+               QETH_DBF_TEXT(trace, 2, "FAILED");
+               qeth_l3_ipaddr_to_string(addr->proto, (u8 *)&addr->u, buf);
+               PRINT_WARN("Could not register IP address %s (rc=0x%x/%d)\n",
+                          buf, rc, rc);
+       }
+       return rc;
+ }
+ static int qeth_l3_deregister_addr_entry(struct qeth_card *card,
+                                               struct qeth_ipaddr *addr)
+ {
+       int rc = 0;
+       if (addr->proto == QETH_PROT_IPV4) {
+               QETH_DBF_TEXT(trace, 2, "deladdr4");
+               QETH_DBF_HEX(trace, 3, &addr->u.a4.addr, sizeof(int));
+       } else if (addr->proto == QETH_PROT_IPV6) {
+               QETH_DBF_TEXT(trace, 2, "deladdr6");
+               QETH_DBF_HEX(trace, 3, &addr->u.a6.addr, 8);
+               QETH_DBF_HEX(trace, 3, ((char *)&addr->u.a6.addr) + 8, 8);
+       } else {
+               QETH_DBF_TEXT(trace, 2, "deladdr?");
+               QETH_DBF_HEX(trace, 3, addr, sizeof(struct qeth_ipaddr));
+       }
+       if (addr->is_multicast)
+               rc = qeth_l3_send_setdelmc(card, addr, IPA_CMD_DELIPM);
+       else
+               rc = qeth_l3_send_setdelip(card, addr, IPA_CMD_DELIP,
+                                       addr->del_flags);
+       if (rc) {
+               QETH_DBF_TEXT(trace, 2, "failed");
+               /* TODO: re-activate this warning as soon as we have a
+                * clean mirco code
+               qeth_ipaddr_to_string(addr->proto, (u8 *)&addr->u, buf);
+               PRINT_WARN("Could not deregister IP address %s (rc=%x)\n",
+                          buf, rc);
+               */
+       }
+       return rc;
+ }
+ static inline u8 qeth_l3_get_qeth_hdr_flags4(int cast_type)
+ {
+       if (cast_type == RTN_MULTICAST)
+               return QETH_CAST_MULTICAST;
+       if (cast_type == RTN_BROADCAST)
+               return QETH_CAST_BROADCAST;
+       return QETH_CAST_UNICAST;
+ }
+ static inline u8 qeth_l3_get_qeth_hdr_flags6(int cast_type)
+ {
+       u8 ct = QETH_HDR_PASSTHRU | QETH_HDR_IPV6;
+       if (cast_type == RTN_MULTICAST)
+               return ct | QETH_CAST_MULTICAST;
+       if (cast_type == RTN_ANYCAST)
+               return ct | QETH_CAST_ANYCAST;
+       if (cast_type == RTN_BROADCAST)
+               return ct | QETH_CAST_BROADCAST;
+       return ct | QETH_CAST_UNICAST;
+ }
+ static int qeth_l3_send_setadp_mode(struct qeth_card *card, __u32 command,
+                                       __u32 mode)
+ {
+       int rc;
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(trace, 4, "adpmode");
+       iob = qeth_get_adapter_cmd(card, command,
+                                  sizeof(struct qeth_ipacmd_setadpparms));
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       cmd->data.setadapterparms.data.mode = mode;
+       rc = qeth_send_ipa_cmd(card, iob, qeth_default_setadapterparms_cb,
+                              NULL);
+       return rc;
+ }
+ static int qeth_l3_setadapter_hstr(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 4, "adphstr");
+       if (qeth_adp_supported(card, IPA_SETADP_SET_BROADCAST_MODE)) {
+               rc = qeth_l3_send_setadp_mode(card,
+                                       IPA_SETADP_SET_BROADCAST_MODE,
+                                       card->options.broadcast_mode);
+               if (rc)
+                       PRINT_WARN("couldn't set broadcast mode on "
+                                  "device %s: x%x\n",
+                                  CARD_BUS_ID(card), rc);
+               rc = qeth_l3_send_setadp_mode(card,
+                                       IPA_SETADP_ALTER_MAC_ADDRESS,
+                                       card->options.macaddr_mode);
+               if (rc)
+                       PRINT_WARN("couldn't set macaddr mode on "
+                                  "device %s: x%x\n", CARD_BUS_ID(card), rc);
+               return rc;
+       }
+       if (card->options.broadcast_mode == QETH_TR_BROADCAST_LOCAL)
+               PRINT_WARN("set adapter parameters not available "
+                          "to set broadcast mode, using ALLRINGS "
+                          "on device %s:\n", CARD_BUS_ID(card));
+       if (card->options.macaddr_mode == QETH_TR_MACADDR_CANONICAL)
+               PRINT_WARN("set adapter parameters not available "
+                          "to set macaddr mode, using NONCANONICAL "
+                          "on device %s:\n", CARD_BUS_ID(card));
+       return 0;
+ }
+ static int qeth_l3_setadapter_parms(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(setup, 2, "setadprm");
+       if (!qeth_is_supported(card, IPA_SETADAPTERPARMS)) {
+               PRINT_WARN("set adapter parameters not supported "
+                          "on device %s.\n",
+                          CARD_BUS_ID(card));
+               QETH_DBF_TEXT(setup, 2, " notsupp");
+               return 0;
+       }
+       rc = qeth_query_setadapterparms(card);
+       if (rc) {
+               PRINT_WARN("couldn't set adapter parameters on device %s: "
+                          "x%x\n", CARD_BUS_ID(card), rc);
+               return rc;
+       }
+       if (qeth_adp_supported(card, IPA_SETADP_ALTER_MAC_ADDRESS)) {
+               rc = qeth_setadpparms_change_macaddr(card);
+               if (rc)
+                       PRINT_WARN("couldn't get MAC address on "
+                                  "device %s: x%x\n",
+                                  CARD_BUS_ID(card), rc);
+       }
+       if ((card->info.link_type == QETH_LINK_TYPE_HSTR) ||
+           (card->info.link_type == QETH_LINK_TYPE_LANE_TR))
+               rc = qeth_l3_setadapter_hstr(card);
+       return rc;
+ }
+ static int qeth_l3_default_setassparms_cb(struct qeth_card *card,
+                       struct qeth_reply *reply, unsigned long data)
+ {
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(trace, 4, "defadpcb");
+       cmd = (struct qeth_ipa_cmd *) data;
+       if (cmd->hdr.return_code == 0) {
+               cmd->hdr.return_code = cmd->data.setassparms.hdr.return_code;
+               if (cmd->hdr.prot_version == QETH_PROT_IPV4)
+                       card->options.ipa4.enabled_funcs = cmd->hdr.ipa_enabled;
+               if (cmd->hdr.prot_version == QETH_PROT_IPV6)
+                       card->options.ipa6.enabled_funcs = cmd->hdr.ipa_enabled;
+       }
+       if (cmd->data.setassparms.hdr.assist_no == IPA_INBOUND_CHECKSUM &&
+           cmd->data.setassparms.hdr.command_code == IPA_CMD_ASS_START) {
+               card->info.csum_mask = cmd->data.setassparms.data.flags_32bit;
+               QETH_DBF_TEXT_(trace, 3, "csum:%d", card->info.csum_mask);
+       }
+       return 0;
+ }
+ static struct qeth_cmd_buffer *qeth_l3_get_setassparms_cmd(
+       struct qeth_card *card, enum qeth_ipa_funcs ipa_func, __u16 cmd_code,
+       __u16 len, enum qeth_prot_versions prot)
+ {
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(trace, 4, "getasscm");
+       iob = qeth_get_ipacmd_buffer(card, IPA_CMD_SETASSPARMS, prot);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       cmd->data.setassparms.hdr.assist_no = ipa_func;
+       cmd->data.setassparms.hdr.length = 8 + len;
+       cmd->data.setassparms.hdr.command_code = cmd_code;
+       cmd->data.setassparms.hdr.return_code = 0;
+       cmd->data.setassparms.hdr.seq_no = 0;
+       return iob;
+ }
+ static int qeth_l3_send_setassparms(struct qeth_card *card,
+       struct qeth_cmd_buffer *iob, __u16 len, long data,
+       int (*reply_cb)(struct qeth_card *, struct qeth_reply *,
+               unsigned long),
+       void *reply_param)
+ {
+       int rc;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(trace, 4, "sendassp");
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       if (len <= sizeof(__u32))
+               cmd->data.setassparms.data.flags_32bit = (__u32) data;
+       else   /* (len > sizeof(__u32)) */
+               memcpy(&cmd->data.setassparms.data, (void *) data, len);
+       rc = qeth_send_ipa_cmd(card, iob, reply_cb, reply_param);
+       return rc;
+ }
+ #ifdef CONFIG_QETH_IPV6
+ static int qeth_l3_send_simple_setassparms_ipv6(struct qeth_card *card,
+               enum qeth_ipa_funcs ipa_func, __u16 cmd_code)
+ {
+       int rc;
+       struct qeth_cmd_buffer *iob;
+       QETH_DBF_TEXT(trace, 4, "simassp6");
+       iob = qeth_l3_get_setassparms_cmd(card, ipa_func, cmd_code,
+                                      0, QETH_PROT_IPV6);
+       rc = qeth_l3_send_setassparms(card, iob, 0, 0,
+                                  qeth_l3_default_setassparms_cb, NULL);
+       return rc;
+ }
+ #endif
+ static int qeth_l3_send_simple_setassparms(struct qeth_card *card,
+               enum qeth_ipa_funcs ipa_func, __u16 cmd_code, long data)
+ {
+       int rc;
+       int length = 0;
+       struct qeth_cmd_buffer *iob;
+       QETH_DBF_TEXT(trace, 4, "simassp4");
+       if (data)
+               length = sizeof(__u32);
+       iob = qeth_l3_get_setassparms_cmd(card, ipa_func, cmd_code,
+                                      length, QETH_PROT_IPV4);
+       rc = qeth_l3_send_setassparms(card, iob, length, data,
+                                  qeth_l3_default_setassparms_cb, NULL);
+       return rc;
+ }
+ static int qeth_l3_start_ipa_arp_processing(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "ipaarp");
+       if (!qeth_is_supported(card, IPA_ARP_PROCESSING)) {
+               PRINT_WARN("ARP processing not supported "
+                          "on %s!\n", QETH_CARD_IFNAME(card));
+               return 0;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_ARP_PROCESSING,
+                                       IPA_CMD_ASS_START, 0);
+       if (rc) {
+               PRINT_WARN("Could not start ARP processing "
+                          "assist on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+       }
+       return rc;
+ }
+ static int qeth_l3_start_ipa_ip_fragmentation(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "ipaipfrg");
+       if (!qeth_is_supported(card, IPA_IP_FRAGMENTATION)) {
+               PRINT_INFO("Hardware IP fragmentation not supported on %s\n",
+                          QETH_CARD_IFNAME(card));
+               return  -EOPNOTSUPP;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_IP_FRAGMENTATION,
+                                         IPA_CMD_ASS_START, 0);
+       if (rc) {
+               PRINT_WARN("Could not start Hardware IP fragmentation "
+                          "assist on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+       } else
+               PRINT_INFO("Hardware IP fragmentation enabled \n");
+       return rc;
+ }
+ static int qeth_l3_start_ipa_source_mac(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "stsrcmac");
+       if (!card->options.fake_ll)
+               return -EOPNOTSUPP;
+       if (!qeth_is_supported(card, IPA_SOURCE_MAC)) {
+               PRINT_INFO("Inbound source address not "
+                          "supported on %s\n", QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_SOURCE_MAC,
+                                         IPA_CMD_ASS_START, 0);
+       if (rc)
+               PRINT_WARN("Could not start inbound source "
+                          "assist on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+       return rc;
+ }
+ static int qeth_l3_start_ipa_vlan(struct qeth_card *card)
+ {
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 3, "strtvlan");
+       if (!qeth_is_supported(card, IPA_FULL_VLAN)) {
+               PRINT_WARN("VLAN not supported on %s\n",
+                               QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_VLAN_PRIO,
+                                         IPA_CMD_ASS_START, 0);
+       if (rc) {
+               PRINT_WARN("Could not start vlan "
+                          "assist on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+       } else {
+               PRINT_INFO("VLAN enabled \n");
+       }
+       return rc;
+ }
+ static int qeth_l3_start_ipa_multicast(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "stmcast");
+       if (!qeth_is_supported(card, IPA_MULTICASTING)) {
+               PRINT_WARN("Multicast not supported on %s\n",
+                          QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_MULTICASTING,
+                                         IPA_CMD_ASS_START, 0);
+       if (rc) {
+               PRINT_WARN("Could not start multicast "
+                          "assist on %s: rc=%i\n",
+                          QETH_CARD_IFNAME(card), rc);
+       } else {
+               PRINT_INFO("Multicast enabled\n");
+               card->dev->flags |= IFF_MULTICAST;
+       }
+       return rc;
+ }
+ static int qeth_l3_query_ipassists_cb(struct qeth_card *card,
+               struct qeth_reply *reply, unsigned long data)
+ {
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(setup, 2, "qipasscb");
+       cmd = (struct qeth_ipa_cmd *) data;
+       if (cmd->hdr.prot_version == QETH_PROT_IPV4) {
+               card->options.ipa4.supported_funcs = cmd->hdr.ipa_supported;
+               card->options.ipa4.enabled_funcs = cmd->hdr.ipa_enabled;
+       } else {
+               card->options.ipa6.supported_funcs = cmd->hdr.ipa_supported;
+               card->options.ipa6.enabled_funcs = cmd->hdr.ipa_enabled;
+       }
+       QETH_DBF_TEXT(setup, 2, "suppenbl");
+       QETH_DBF_TEXT_(setup, 2, "%x", cmd->hdr.ipa_supported);
+       QETH_DBF_TEXT_(setup, 2, "%x", cmd->hdr.ipa_enabled);
+       return 0;
+ }
+ static int qeth_l3_query_ipassists(struct qeth_card *card,
+                               enum qeth_prot_versions prot)
+ {
+       int rc;
+       struct qeth_cmd_buffer *iob;
+       QETH_DBF_TEXT_(setup, 2, "qipassi%i", prot);
+       iob = qeth_get_ipacmd_buffer(card, IPA_CMD_QIPASSIST, prot);
+       rc = qeth_send_ipa_cmd(card, iob, qeth_l3_query_ipassists_cb, NULL);
+       return rc;
+ }
+ #ifdef CONFIG_QETH_IPV6
+ static int qeth_l3_softsetup_ipv6(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "softipv6");
+       if (card->info.type == QETH_CARD_TYPE_IQD)
+               goto out;
+       rc = qeth_l3_query_ipassists(card, QETH_PROT_IPV6);
+       if (rc) {
+               PRINT_ERR("IPv6 query ipassist failed on %s\n",
+                         QETH_CARD_IFNAME(card));
+               return rc;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_IPV6,
+                                         IPA_CMD_ASS_START, 3);
+       if (rc) {
+               PRINT_WARN("IPv6 start assist (version 4) failed "
+                          "on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+               return rc;
+       }
+       rc = qeth_l3_send_simple_setassparms_ipv6(card, IPA_IPV6,
+                                              IPA_CMD_ASS_START);
+       if (rc) {
+               PRINT_WARN("IPV6 start assist (version 6) failed  "
+                          "on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+               return rc;
+       }
+       rc = qeth_l3_send_simple_setassparms_ipv6(card, IPA_PASSTHRU,
+                                              IPA_CMD_ASS_START);
+       if (rc) {
+               PRINT_WARN("Could not enable passthrough "
+                          "on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+               return rc;
+       }
+ out:
+       PRINT_INFO("IPV6 enabled \n");
+       return 0;
+ }
+ #endif
+ static int qeth_l3_start_ipa_ipv6(struct qeth_card *card)
+ {
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 3, "strtipv6");
+       if (!qeth_is_supported(card, IPA_IPV6)) {
+               PRINT_WARN("IPv6 not supported on %s\n",
+                          QETH_CARD_IFNAME(card));
+               return 0;
+       }
+ #ifdef CONFIG_QETH_IPV6
+       rc = qeth_l3_softsetup_ipv6(card);
+ #endif
+       return rc ;
+ }
+ static int qeth_l3_start_ipa_broadcast(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "stbrdcst");
+       card->info.broadcast_capable = 0;
+       if (!qeth_is_supported(card, IPA_FILTERING)) {
+               PRINT_WARN("Broadcast not supported on %s\n",
+                          QETH_CARD_IFNAME(card));
+               rc = -EOPNOTSUPP;
+               goto out;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_FILTERING,
+                                         IPA_CMD_ASS_START, 0);
+       if (rc) {
+               PRINT_WARN("Could not enable broadcasting filtering "
+                          "on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+               goto out;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_FILTERING,
+                                         IPA_CMD_ASS_CONFIGURE, 1);
+       if (rc) {
+               PRINT_WARN("Could not set up broadcast filtering on %s: 0x%x\n",
+                          QETH_CARD_IFNAME(card), rc);
+               goto out;
+       }
+       card->info.broadcast_capable = QETH_BROADCAST_WITH_ECHO;
+       PRINT_INFO("Broadcast enabled \n");
+       rc = qeth_l3_send_simple_setassparms(card, IPA_FILTERING,
+                                         IPA_CMD_ASS_ENABLE, 1);
+       if (rc) {
+               PRINT_WARN("Could not set up broadcast echo filtering on "
+                          "%s: 0x%x\n", QETH_CARD_IFNAME(card), rc);
+               goto out;
+       }
+       card->info.broadcast_capable = QETH_BROADCAST_WITHOUT_ECHO;
+ out:
+       if (card->info.broadcast_capable)
+               card->dev->flags |= IFF_BROADCAST;
+       else
+               card->dev->flags &= ~IFF_BROADCAST;
+       return rc;
+ }
+ static int qeth_l3_send_checksum_command(struct qeth_card *card)
+ {
+       int rc;
+       rc = qeth_l3_send_simple_setassparms(card, IPA_INBOUND_CHECKSUM,
+                                         IPA_CMD_ASS_START, 0);
+       if (rc) {
+               PRINT_WARN("Starting Inbound HW Checksumming failed on %s: "
+                          "0x%x,\ncontinuing using Inbound SW Checksumming\n",
+                          QETH_CARD_IFNAME(card), rc);
+               return rc;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_INBOUND_CHECKSUM,
+                                         IPA_CMD_ASS_ENABLE,
+                                         card->info.csum_mask);
+       if (rc) {
+               PRINT_WARN("Enabling Inbound HW Checksumming failed on %s: "
+                          "0x%x,\ncontinuing using Inbound SW Checksumming\n",
+                          QETH_CARD_IFNAME(card), rc);
+               return rc;
+       }
+       return 0;
+ }
+ static int qeth_l3_start_ipa_checksum(struct qeth_card *card)
+ {
+       int rc = 0;
+       QETH_DBF_TEXT(trace, 3, "strtcsum");
+       if (card->options.checksum_type == NO_CHECKSUMMING) {
+               PRINT_WARN("Using no checksumming on %s.\n",
+                          QETH_CARD_IFNAME(card));
+               return 0;
+       }
+       if (card->options.checksum_type == SW_CHECKSUMMING) {
+               PRINT_WARN("Using SW checksumming on %s.\n",
+                          QETH_CARD_IFNAME(card));
+               return 0;
+       }
+       if (!qeth_is_supported(card, IPA_INBOUND_CHECKSUM)) {
+               PRINT_WARN("Inbound HW Checksumming not "
+                          "supported on %s,\ncontinuing "
+                          "using Inbound SW Checksumming\n",
+                          QETH_CARD_IFNAME(card));
+               card->options.checksum_type = SW_CHECKSUMMING;
+               return 0;
+       }
+       rc = qeth_l3_send_checksum_command(card);
+       if (!rc)
+               PRINT_INFO("HW Checksumming (inbound) enabled \n");
+       return rc;
+ }
+ static int qeth_l3_start_ipa_tso(struct qeth_card *card)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "sttso");
+       if (!qeth_is_supported(card, IPA_OUTBOUND_TSO)) {
+               PRINT_WARN("Outbound TSO not supported on %s\n",
+                          QETH_CARD_IFNAME(card));
+               rc = -EOPNOTSUPP;
+       } else {
+               rc = qeth_l3_send_simple_setassparms(card, IPA_OUTBOUND_TSO,
+                                               IPA_CMD_ASS_START, 0);
+               if (rc)
+                       PRINT_WARN("Could not start outbound TSO "
+                                  "assist on %s: rc=%i\n",
+                                  QETH_CARD_IFNAME(card), rc);
+               else
+                       PRINT_INFO("Outbound TSO enabled\n");
+       }
+       if (rc && (card->options.large_send == QETH_LARGE_SEND_TSO)) {
+               card->options.large_send = QETH_LARGE_SEND_NO;
+               card->dev->features &= ~(NETIF_F_TSO | NETIF_F_SG);
+       }
+       return rc;
+ }
+ static int qeth_l3_start_ipassists(struct qeth_card *card)
+ {
+       QETH_DBF_TEXT(trace, 3, "strtipas");
+       qeth_l3_start_ipa_arp_processing(card); /* go on*/
+       qeth_l3_start_ipa_ip_fragmentation(card);       /* go on*/
+       qeth_l3_start_ipa_source_mac(card);     /* go on*/
+       qeth_l3_start_ipa_vlan(card);           /* go on*/
+       qeth_l3_start_ipa_multicast(card);              /* go on*/
+       qeth_l3_start_ipa_ipv6(card);           /* go on*/
+       qeth_l3_start_ipa_broadcast(card);              /* go on*/
+       qeth_l3_start_ipa_checksum(card);               /* go on*/
+       qeth_l3_start_ipa_tso(card);            /* go on*/
+       return 0;
+ }
+ static int qeth_l3_put_unique_id(struct qeth_card *card)
+ {
+       int rc = 0;
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(trace, 2, "puniqeid");
+       if ((card->info.unique_id & UNIQUE_ID_NOT_BY_CARD) ==
+               UNIQUE_ID_NOT_BY_CARD)
+               return -1;
+       iob = qeth_get_ipacmd_buffer(card, IPA_CMD_DESTROY_ADDR,
+                                    QETH_PROT_IPV6);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       *((__u16 *) &cmd->data.create_destroy_addr.unique_id[6]) =
+                               card->info.unique_id;
+       memcpy(&cmd->data.create_destroy_addr.unique_id[0],
+              card->dev->dev_addr, OSA_ADDR_LEN);
+       rc = qeth_send_ipa_cmd(card, iob, NULL, NULL);
+       return rc;
+ }
+ static int qeth_l3_iqd_read_initial_mac_cb(struct qeth_card *card,
+               struct qeth_reply *reply, unsigned long data)
+ {
+       struct qeth_ipa_cmd *cmd;
+       cmd = (struct qeth_ipa_cmd *) data;
+       if (cmd->hdr.return_code == 0)
+               memcpy(card->dev->dev_addr,
+                       cmd->data.create_destroy_addr.unique_id, ETH_ALEN);
+       else
+               random_ether_addr(card->dev->dev_addr);
+       return 0;
+ }
+ static int qeth_l3_iqd_read_initial_mac(struct qeth_card *card)
+ {
+       int rc = 0;
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(setup, 2, "hsrmac");
+       iob = qeth_get_ipacmd_buffer(card, IPA_CMD_CREATE_ADDR,
+                                    QETH_PROT_IPV6);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       *((__u16 *) &cmd->data.create_destroy_addr.unique_id[6]) =
+                       card->info.unique_id;
+       rc = qeth_send_ipa_cmd(card, iob, qeth_l3_iqd_read_initial_mac_cb,
+                               NULL);
+       return rc;
+ }
+ static int qeth_l3_get_unique_id_cb(struct qeth_card *card,
+               struct qeth_reply *reply, unsigned long data)
+ {
+       struct qeth_ipa_cmd *cmd;
+       cmd = (struct qeth_ipa_cmd *) data;
+       if (cmd->hdr.return_code == 0)
+               card->info.unique_id = *((__u16 *)
+                               &cmd->data.create_destroy_addr.unique_id[6]);
+       else {
+               card->info.unique_id =  UNIQUE_ID_IF_CREATE_ADDR_FAILED |
+                                       UNIQUE_ID_NOT_BY_CARD;
+               PRINT_WARN("couldn't get a unique id from the card on device "
+                          "%s (result=x%x), using default id. ipv6 "
+                          "autoconfig on other lpars may lead to duplicate "
+                          "ip addresses. please use manually "
+                          "configured ones.\n",
+                          CARD_BUS_ID(card), cmd->hdr.return_code);
+       }
+       return 0;
+ }
+ static int qeth_l3_get_unique_id(struct qeth_card *card)
+ {
+       int rc = 0;
+       struct qeth_cmd_buffer *iob;
+       struct qeth_ipa_cmd *cmd;
+       QETH_DBF_TEXT(setup, 2, "guniqeid");
+       if (!qeth_is_supported(card, IPA_IPV6)) {
+               card->info.unique_id =  UNIQUE_ID_IF_CREATE_ADDR_FAILED |
+                                       UNIQUE_ID_NOT_BY_CARD;
+               return 0;
+       }
+       iob = qeth_get_ipacmd_buffer(card, IPA_CMD_CREATE_ADDR,
+                                    QETH_PROT_IPV6);
+       cmd = (struct qeth_ipa_cmd *)(iob->data+IPA_PDU_HEADER_SIZE);
+       *((__u16 *) &cmd->data.create_destroy_addr.unique_id[6]) =
+                       card->info.unique_id;
+       rc = qeth_send_ipa_cmd(card, iob, qeth_l3_get_unique_id_cb, NULL);
+       return rc;
+ }
+ static void qeth_l3_get_mac_for_ipm(__u32 ipm, char *mac,
+                               struct net_device *dev)
+ {
+       if (dev->type == ARPHRD_IEEE802_TR)
+               ip_tr_mc_map(ipm, mac);
+       else
+               ip_eth_mc_map(ipm, mac);
+ }
+ static void qeth_l3_add_mc(struct qeth_card *card, struct in_device *in4_dev)
+ {
+       struct qeth_ipaddr *ipm;
+       struct ip_mc_list *im4;
+       char buf[MAX_ADDR_LEN];
+       QETH_DBF_TEXT(trace, 4, "addmc");
+       for (im4 = in4_dev->mc_list; im4; im4 = im4->next) {
+               qeth_l3_get_mac_for_ipm(im4->multiaddr, buf, in4_dev->dev);
+               ipm = qeth_l3_get_addr_buffer(QETH_PROT_IPV4);
+               if (!ipm)
+                       continue;
+               ipm->u.a4.addr = im4->multiaddr;
+               memcpy(ipm->mac, buf, OSA_ADDR_LEN);
+               ipm->is_multicast = 1;
+               if (!qeth_l3_add_ip(card, ipm))
+                       kfree(ipm);
+       }
+ }
+ static void qeth_l3_add_vlan_mc(struct qeth_card *card)
+ {
+       struct in_device *in_dev;
+       struct vlan_group *vg;
+       int i;
+       QETH_DBF_TEXT(trace, 4, "addmcvl");
+       if (!qeth_is_supported(card, IPA_FULL_VLAN) || (card->vlangrp == NULL))
+               return;
+       vg = card->vlangrp;
+       for (i = 0; i < VLAN_GROUP_ARRAY_LEN; i++) {
+               struct net_device *netdev = vlan_group_get_device(vg, i);
+               if (netdev == NULL ||
+                   !(netdev->flags & IFF_UP))
+                       continue;
+               in_dev = in_dev_get(netdev);
+               if (!in_dev)
+                       continue;
+               read_lock(&in_dev->mc_list_lock);
+               qeth_l3_add_mc(card, in_dev);
+               read_unlock(&in_dev->mc_list_lock);
+               in_dev_put(in_dev);
+       }
+ }
+ static void qeth_l3_add_multicast_ipv4(struct qeth_card *card)
+ {
+       struct in_device *in4_dev;
+       QETH_DBF_TEXT(trace, 4, "chkmcv4");
+       in4_dev = in_dev_get(card->dev);
+       if (in4_dev == NULL)
+               return;
+       read_lock(&in4_dev->mc_list_lock);
+       qeth_l3_add_mc(card, in4_dev);
+       qeth_l3_add_vlan_mc(card);
+       read_unlock(&in4_dev->mc_list_lock);
+       in_dev_put(in4_dev);
+ }
+ #ifdef CONFIG_QETH_IPV6
+ static void qeth_l3_add_mc6(struct qeth_card *card, struct inet6_dev *in6_dev)
+ {
+       struct qeth_ipaddr *ipm;
+       struct ifmcaddr6 *im6;
+       char buf[MAX_ADDR_LEN];
+       QETH_DBF_TEXT(trace, 4, "addmc6");
+       for (im6 = in6_dev->mc_list; im6 != NULL; im6 = im6->next) {
+               ndisc_mc_map(&im6->mca_addr, buf, in6_dev->dev, 0);
+               ipm = qeth_l3_get_addr_buffer(QETH_PROT_IPV6);
+               if (!ipm)
+                       continue;
+               ipm->is_multicast = 1;
+               memcpy(ipm->mac, buf, OSA_ADDR_LEN);
+               memcpy(&ipm->u.a6.addr, &im6->mca_addr.s6_addr,
+                      sizeof(struct in6_addr));
+               if (!qeth_l3_add_ip(card, ipm))
+                       kfree(ipm);
+       }
+ }
+ static void qeth_l3_add_vlan_mc6(struct qeth_card *card)
+ {
+       struct inet6_dev *in_dev;
+       struct vlan_group *vg;
+       int i;
+       QETH_DBF_TEXT(trace, 4, "admc6vl");
+       if (!qeth_is_supported(card, IPA_FULL_VLAN) || (card->vlangrp == NULL))
+               return;
+       vg = card->vlangrp;
+       for (i = 0; i < VLAN_GROUP_ARRAY_LEN; i++) {
+               struct net_device *netdev = vlan_group_get_device(vg, i);
+               if (netdev == NULL ||
+                   !(netdev->flags & IFF_UP))
+                       continue;
+               in_dev = in6_dev_get(netdev);
+               if (!in_dev)
+                       continue;
+               read_lock_bh(&in_dev->lock);
+               qeth_l3_add_mc6(card, in_dev);
+               read_unlock_bh(&in_dev->lock);
+               in6_dev_put(in_dev);
+       }
+ }
+ static void qeth_l3_add_multicast_ipv6(struct qeth_card *card)
+ {
+       struct inet6_dev *in6_dev;
+       QETH_DBF_TEXT(trace, 4, "chkmcv6");
+       if (!qeth_is_supported(card, IPA_IPV6))
+               return ;
+       in6_dev = in6_dev_get(card->dev);
+       if (in6_dev == NULL)
+               return;
+       read_lock_bh(&in6_dev->lock);
+       qeth_l3_add_mc6(card, in6_dev);
+       qeth_l3_add_vlan_mc6(card);
+       read_unlock_bh(&in6_dev->lock);
+       in6_dev_put(in6_dev);
+ }
+ #endif /* CONFIG_QETH_IPV6 */
+ static void qeth_l3_free_vlan_addresses4(struct qeth_card *card,
+                       unsigned short vid)
+ {
+       struct in_device *in_dev;
+       struct in_ifaddr *ifa;
+       struct qeth_ipaddr *addr;
+       QETH_DBF_TEXT(trace, 4, "frvaddr4");
+       in_dev = in_dev_get(vlan_group_get_device(card->vlangrp, vid));
+       if (!in_dev)
+               return;
+       for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
+               addr = qeth_l3_get_addr_buffer(QETH_PROT_IPV4);
+               if (addr) {
+                       addr->u.a4.addr = ifa->ifa_address;
+                       addr->u.a4.mask = ifa->ifa_mask;
+                       addr->type = QETH_IP_TYPE_NORMAL;
+                       if (!qeth_l3_delete_ip(card, addr))
+                               kfree(addr);
+               }
+       }
+       in_dev_put(in_dev);
+ }
+ static void qeth_l3_free_vlan_addresses6(struct qeth_card *card,
+                       unsigned short vid)
+ {
+ #ifdef CONFIG_QETH_IPV6
+       struct inet6_dev *in6_dev;
+       struct inet6_ifaddr *ifa;
+       struct qeth_ipaddr *addr;
+       QETH_DBF_TEXT(trace, 4, "frvaddr6");
+       in6_dev = in6_dev_get(vlan_group_get_device(card->vlangrp, vid));
+       if (!in6_dev)
+               return;
+       for (ifa = in6_dev->addr_list; ifa; ifa = ifa->lst_next) {
+               addr = qeth_l3_get_addr_buffer(QETH_PROT_IPV6);
+               if (addr) {
+                       memcpy(&addr->u.a6.addr, &ifa->addr,
+                              sizeof(struct in6_addr));
+                       addr->u.a6.pfxlen = ifa->prefix_len;
+                       addr->type = QETH_IP_TYPE_NORMAL;
+                       if (!qeth_l3_delete_ip(card, addr))
+                               kfree(addr);
+               }
+       }
+       in6_dev_put(in6_dev);
+ #endif /* CONFIG_QETH_IPV6 */
+ }
+ static void qeth_l3_free_vlan_addresses(struct qeth_card *card,
+                       unsigned short vid)
+ {
+       if (!card->vlangrp)
+               return;
+       qeth_l3_free_vlan_addresses4(card, vid);
+       qeth_l3_free_vlan_addresses6(card, vid);
+ }
+ static void qeth_l3_vlan_rx_register(struct net_device *dev,
+                       struct vlan_group *grp)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       unsigned long flags;
+       QETH_DBF_TEXT(trace, 4, "vlanreg");
+       spin_lock_irqsave(&card->vlanlock, flags);
+       card->vlangrp = grp;
+       spin_unlock_irqrestore(&card->vlanlock, flags);
+ }
+ static void qeth_l3_vlan_rx_add_vid(struct net_device *dev, unsigned short vid)
+ {
+       struct net_device *vlandev;
+       struct qeth_card *card = (struct qeth_card *) dev->priv;
+       struct in_device *in_dev;
+       if (card->info.type == QETH_CARD_TYPE_IQD)
+               return;
+       vlandev = vlan_group_get_device(card->vlangrp, vid);
+       vlandev->neigh_setup = qeth_l3_neigh_setup;
+       in_dev = in_dev_get(vlandev);
+ #ifdef CONFIG_SYSCTL
+       neigh_sysctl_unregister(in_dev->arp_parms);
+ #endif
+       neigh_parms_release(&arp_tbl, in_dev->arp_parms);
+       in_dev->arp_parms = neigh_parms_alloc(vlandev, &arp_tbl);
+ #ifdef CONFIG_SYSCTL
+       neigh_sysctl_register(vlandev, in_dev->arp_parms, NET_IPV4,
+                             NET_IPV4_NEIGH, "ipv4", NULL, NULL);
+ #endif
+       in_dev_put(in_dev);
+       return;
+ }
+ static void qeth_l3_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       unsigned long flags;
+       QETH_DBF_TEXT_(trace, 4, "kid:%d", vid);
+       spin_lock_irqsave(&card->vlanlock, flags);
+       /* unregister IP addresses of vlan device */
+       qeth_l3_free_vlan_addresses(card, vid);
+       vlan_group_set_device(card->vlangrp, vid, NULL);
+       spin_unlock_irqrestore(&card->vlanlock, flags);
+       qeth_l3_set_multicast_list(card->dev);
+ }
+ static inline __u16 qeth_l3_rebuild_skb(struct qeth_card *card,
+                       struct sk_buff *skb, struct qeth_hdr *hdr)
+ {
+       unsigned short vlan_id = 0;
+       __be16 prot;
+       struct iphdr *ip_hdr;
+       unsigned char tg_addr[MAX_ADDR_LEN];
+       if (!(hdr->hdr.l3.flags & QETH_HDR_PASSTHRU)) {
+               prot = htons((hdr->hdr.l3.flags & QETH_HDR_IPV6)? ETH_P_IPV6 :
+                             ETH_P_IP);
+               switch (hdr->hdr.l3.flags & QETH_HDR_CAST_MASK) {
+               case QETH_CAST_MULTICAST:
+                       switch (prot) {
+ #ifdef CONFIG_QETH_IPV6
+                       case __constant_htons(ETH_P_IPV6):
+                               ndisc_mc_map((struct in6_addr *)
+                                    skb->data + 24,
+                                    tg_addr, card->dev, 0);
+                               break;
+ #endif
+                       case __constant_htons(ETH_P_IP):
+                               ip_hdr = (struct iphdr *)skb->data;
+                               (card->dev->type == ARPHRD_IEEE802_TR) ?
+                               ip_tr_mc_map(ip_hdr->daddr, tg_addr):
+                               ip_eth_mc_map(ip_hdr->daddr, tg_addr);
+                               break;
+                       default:
+                               memcpy(tg_addr, card->dev->broadcast,
+                                       card->dev->addr_len);
+                       }
+                       card->stats.multicast++;
+                       skb->pkt_type = PACKET_MULTICAST;
+                       break;
+               case QETH_CAST_BROADCAST:
+                       memcpy(tg_addr, card->dev->broadcast,
+                               card->dev->addr_len);
+                       card->stats.multicast++;
+                       skb->pkt_type = PACKET_BROADCAST;
+                       break;
+               case QETH_CAST_UNICAST:
+               case QETH_CAST_ANYCAST:
+               case QETH_CAST_NOCAST:
+               default:
+                       skb->pkt_type = PACKET_HOST;
+                       memcpy(tg_addr, card->dev->dev_addr,
+                               card->dev->addr_len);
+               }
+               card->dev->header_ops->create(skb, card->dev, prot, tg_addr,
+                                             "FAKELL", card->dev->addr_len);
+       }
+ #ifdef CONFIG_TR
+       if (card->dev->type == ARPHRD_IEEE802_TR)
+               skb->protocol = tr_type_trans(skb, card->dev);
+       else
+ #endif
+               skb->protocol = eth_type_trans(skb, card->dev);
+       if (hdr->hdr.l3.ext_flags &
+           (QETH_HDR_EXT_VLAN_FRAME | QETH_HDR_EXT_INCLUDE_VLAN_TAG)) {
+               vlan_id = (hdr->hdr.l3.ext_flags & QETH_HDR_EXT_VLAN_FRAME)?
+                hdr->hdr.l3.vlan_id : *((u16 *)&hdr->hdr.l3.dest_addr[12]);
+       }
+       skb->ip_summed = card->options.checksum_type;
+       if (card->options.checksum_type == HW_CHECKSUMMING) {
+               if ((hdr->hdr.l3.ext_flags &
+                     (QETH_HDR_EXT_CSUM_HDR_REQ |
+                      QETH_HDR_EXT_CSUM_TRANSP_REQ)) ==
+                    (QETH_HDR_EXT_CSUM_HDR_REQ |
+                     QETH_HDR_EXT_CSUM_TRANSP_REQ))
+                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+               else
+                       skb->ip_summed = SW_CHECKSUMMING;
+       }
+       return vlan_id;
+ }
+ static void qeth_l3_process_inbound_buffer(struct qeth_card *card,
+                           struct qeth_qdio_buffer *buf, int index)
+ {
+       struct qdio_buffer_element *element;
+       struct sk_buff *skb;
+       struct qeth_hdr *hdr;
+       int offset;
+       __u16 vlan_tag = 0;
+       unsigned int len;
+       /* get first element of current buffer */
+       element = (struct qdio_buffer_element *)&buf->buffer->element[0];
+       offset = 0;
+       if (card->options.performance_stats)
+               card->perf_stats.bufs_rec++;
+       while ((skb = qeth_core_get_next_skb(card, buf->buffer, &element,
+                                      &offset, &hdr))) {
+               skb->dev = card->dev;
+               /* is device UP ? */
+               if (!(card->dev->flags & IFF_UP)) {
+                       dev_kfree_skb_any(skb);
+                       continue;
+               }
+               switch (hdr->hdr.l3.id) {
+               case QETH_HEADER_TYPE_LAYER3:
+                       vlan_tag = qeth_l3_rebuild_skb(card, skb, hdr);
+                       len = skb->len;
+                       if (vlan_tag)
+                               if (card->vlangrp)
+                                       vlan_hwaccel_rx(skb, card->vlangrp,
+                                               vlan_tag);
+                               else {
+                                       dev_kfree_skb_any(skb);
+                                       continue;
+                               }
+                       else
+                               netif_rx(skb);
+                       break;
+               default:
+                       dev_kfree_skb_any(skb);
+                       QETH_DBF_TEXT(trace, 3, "inbunkno");
+                       QETH_DBF_HEX(control, 3, hdr, QETH_DBF_CONTROL_LEN);
+                       continue;
+               }
+               card->dev->last_rx = jiffies;
+               card->stats.rx_packets++;
+               card->stats.rx_bytes += len;
+       }
+ }
+ static int qeth_l3_verify_vlan_dev(struct net_device *dev,
+                       struct qeth_card *card)
+ {
+       int rc = 0;
+       struct vlan_group *vg;
+       int i;
+       vg = card->vlangrp;
+       if (!vg)
+               return rc;
+       for (i = 0; i < VLAN_GROUP_ARRAY_LEN; i++) {
+               if (vlan_group_get_device(vg, i) == dev) {
+                       rc = QETH_VLAN_CARD;
+                       break;
+               }
+       }
+       if (rc && !(netdev_priv(vlan_dev_info(dev)->real_dev) == (void *)card))
+               return 0;
+       return rc;
+ }
+ static int qeth_l3_verify_dev(struct net_device *dev)
+ {
+       struct qeth_card *card;
+       unsigned long flags;
+       int rc = 0;
+       read_lock_irqsave(&qeth_core_card_list.rwlock, flags);
+       list_for_each_entry(card, &qeth_core_card_list.list, list) {
+               if (card->dev == dev) {
+                       rc = QETH_REAL_CARD;
+                       break;
+               }
+               rc = qeth_l3_verify_vlan_dev(dev, card);
+               if (rc)
+                       break;
+       }
+       read_unlock_irqrestore(&qeth_core_card_list.rwlock, flags);
+       return rc;
+ }
+ static struct qeth_card *qeth_l3_get_card_from_dev(struct net_device *dev)
+ {
+       struct qeth_card *card = NULL;
+       int rc;
+       rc = qeth_l3_verify_dev(dev);
+       if (rc == QETH_REAL_CARD)
+               card = netdev_priv(dev);
+       else if (rc == QETH_VLAN_CARD)
+               card = netdev_priv(vlan_dev_info(dev)->real_dev);
+       if (card->options.layer2)
+               card = NULL;
+       QETH_DBF_TEXT_(trace, 4, "%d", rc);
+       return card ;
+ }
+ static int qeth_l3_stop_card(struct qeth_card *card, int recovery_mode)
+ {
+       int rc = 0;
+       QETH_DBF_TEXT(setup, 2, "stopcard");
+       QETH_DBF_HEX(setup, 2, &card, sizeof(void *));
+       qeth_set_allowed_threads(card, 0, 1);
+       if (qeth_wait_for_threads(card, ~QETH_RECOVER_THREAD))
+               return -ERESTARTSYS;
+       if (card->read.state == CH_STATE_UP &&
+           card->write.state == CH_STATE_UP &&
+           (card->state == CARD_STATE_UP)) {
+               if (recovery_mode)
+                       qeth_l3_stop(card->dev);
+               if (!card->use_hard_stop) {
+                       rc = qeth_send_stoplan(card);
+                       if (rc)
+                               QETH_DBF_TEXT_(setup, 2, "1err%d", rc);
+               }
+               card->state = CARD_STATE_SOFTSETUP;
+       }
+       if (card->state == CARD_STATE_SOFTSETUP) {
+               qeth_l3_clear_ip_list(card, !card->use_hard_stop, 1);
+               qeth_clear_ipacmd_list(card);
+               card->state = CARD_STATE_HARDSETUP;
+       }
+       if (card->state == CARD_STATE_HARDSETUP) {
+               if (!card->use_hard_stop &&
+                   (card->info.type != QETH_CARD_TYPE_IQD)) {
+                       rc = qeth_l3_put_unique_id(card);
+                       if (rc)
+                               QETH_DBF_TEXT_(setup, 2, "2err%d", rc);
+               }
+               qeth_qdio_clear_card(card, 0);
+               qeth_clear_qdio_buffers(card);
+               qeth_clear_working_pool_list(card);
+               card->state = CARD_STATE_DOWN;
+       }
+       if (card->state == CARD_STATE_DOWN) {
+               qeth_clear_cmd_buffers(&card->read);
+               qeth_clear_cmd_buffers(&card->write);
+       }
+       card->use_hard_stop = 0;
+       return rc;
+ }
+ static void qeth_l3_set_multicast_list(struct net_device *dev)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       QETH_DBF_TEXT(trace, 3, "setmulti");
+       qeth_l3_delete_mc_addresses(card);
+       qeth_l3_add_multicast_ipv4(card);
+ #ifdef CONFIG_QETH_IPV6
+       qeth_l3_add_multicast_ipv6(card);
+ #endif
+       qeth_l3_set_ip_addr_list(card);
+       if (!qeth_adp_supported(card, IPA_SETADP_SET_PROMISC_MODE))
+               return;
+       qeth_setadp_promisc_mode(card);
+ }
+ static const char *qeth_l3_arp_get_error_cause(int *rc)
+ {
+       switch (*rc) {
+       case QETH_IPA_ARP_RC_FAILED:
+               *rc = -EIO;
+               return "operation failed";
+       case QETH_IPA_ARP_RC_NOTSUPP:
+               *rc = -EOPNOTSUPP;
+               return "operation not supported";
+       case QETH_IPA_ARP_RC_OUT_OF_RANGE:
+               *rc = -EINVAL;
+               return "argument out of range";
+       case QETH_IPA_ARP_RC_Q_NOTSUPP:
+               *rc = -EOPNOTSUPP;
+               return "query operation not supported";
+       case QETH_IPA_ARP_RC_Q_NO_DATA:
+               *rc = -ENOENT;
+               return "no query data available";
+       default:
+               return "unknown error";
+       }
+ }
+ static int qeth_l3_arp_set_no_entries(struct qeth_card *card, int no_entries)
+ {
+       int tmp;
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "arpstnoe");
+       /*
+        * currently GuestLAN only supports the ARP assist function
+        * IPA_CMD_ASS_ARP_QUERY_INFO, but not IPA_CMD_ASS_ARP_SET_NO_ENTRIES;
+        * thus we say EOPNOTSUPP for this ARP function
+        */
+       if (card->info.guestlan)
+               return -EOPNOTSUPP;
+       if (!qeth_is_supported(card, IPA_ARP_PROCESSING)) {
+               PRINT_WARN("ARP processing not supported "
+                          "on %s!\n", QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_ARP_PROCESSING,
+                                         IPA_CMD_ASS_ARP_SET_NO_ENTRIES,
+                                         no_entries);
+       if (rc) {
+               tmp = rc;
+               PRINT_WARN("Could not set number of ARP entries on %s: "
+                       "%s (0x%x/%d)\n", QETH_CARD_IFNAME(card),
+                       qeth_l3_arp_get_error_cause(&rc), tmp, tmp);
+       }
+       return rc;
+ }
+ static void qeth_l3_copy_arp_entries_stripped(struct qeth_arp_query_info *qinfo,
+               struct qeth_arp_query_data *qdata, int entry_size,
+               int uentry_size)
+ {
+       char *entry_ptr;
+       char *uentry_ptr;
+       int i;
+       entry_ptr = (char *)&qdata->data;
+       uentry_ptr = (char *)(qinfo->udata + qinfo->udata_offset);
+       for (i = 0; i < qdata->no_entries; ++i) {
+               /* strip off 32 bytes "media specific information" */
+               memcpy(uentry_ptr, (entry_ptr + 32), entry_size - 32);
+               entry_ptr += entry_size;
+               uentry_ptr += uentry_size;
+       }
+ }
+ static int qeth_l3_arp_query_cb(struct qeth_card *card,
+               struct qeth_reply *reply, unsigned long data)
+ {
+       struct qeth_ipa_cmd *cmd;
+       struct qeth_arp_query_data *qdata;
+       struct qeth_arp_query_info *qinfo;
+       int entry_size;
+       int uentry_size;
+       int i;
+       QETH_DBF_TEXT(trace, 4, "arpquecb");
+       qinfo = (struct qeth_arp_query_info *) reply->param;
+       cmd = (struct qeth_ipa_cmd *) data;
+       if (cmd->hdr.return_code) {
+               QETH_DBF_TEXT_(trace, 4, "qaer1%i", cmd->hdr.return_code);
+               return 0;
+       }
+       if (cmd->data.setassparms.hdr.return_code) {
+               cmd->hdr.return_code = cmd->data.setassparms.hdr.return_code;
+               QETH_DBF_TEXT_(trace, 4, "qaer2%i", cmd->hdr.return_code);
+               return 0;
+       }
+       qdata = &cmd->data.setassparms.data.query_arp;
+       switch (qdata->reply_bits) {
+       case 5:
+               uentry_size = entry_size = sizeof(struct qeth_arp_qi_entry5);
+               if (qinfo->mask_bits & QETH_QARP_STRIP_ENTRIES)
+                       uentry_size = sizeof(struct qeth_arp_qi_entry5_short);
+               break;
+       case 7:
+               /* fall through to default */
+       default:
+               /* tr is the same as eth -> entry7 */
+               uentry_size = entry_size = sizeof(struct qeth_arp_qi_entry7);
+               if (qinfo->mask_bits & QETH_QARP_STRIP_ENTRIES)
+                       uentry_size = sizeof(struct qeth_arp_qi_entry7_short);
+               break;
+       }
+       /* check if there is enough room in userspace */
+       if ((qinfo->udata_len - qinfo->udata_offset) <
+                       qdata->no_entries * uentry_size){
+               QETH_DBF_TEXT_(trace, 4, "qaer3%i", -ENOMEM);
+               cmd->hdr.return_code = -ENOMEM;
+               PRINT_WARN("query ARP user space buffer is too small for "
+                          "the returned number of ARP entries. "
+                          "Aborting query!\n");
+               goto out_error;
+       }
+       QETH_DBF_TEXT_(trace, 4, "anore%i",
+                      cmd->data.setassparms.hdr.number_of_replies);
+       QETH_DBF_TEXT_(trace, 4, "aseqn%i", cmd->data.setassparms.hdr.seq_no);
+       QETH_DBF_TEXT_(trace, 4, "anoen%i", qdata->no_entries);
+       if (qinfo->mask_bits & QETH_QARP_STRIP_ENTRIES) {
+               /* strip off "media specific information" */
+               qeth_l3_copy_arp_entries_stripped(qinfo, qdata, entry_size,
+                                              uentry_size);
+       } else
+               /*copy entries to user buffer*/
+               memcpy(qinfo->udata + qinfo->udata_offset,
+                      (char *)&qdata->data, qdata->no_entries*uentry_size);
+       qinfo->no_entries += qdata->no_entries;
+       qinfo->udata_offset += (qdata->no_entries*uentry_size);
+       /* check if all replies received ... */
+       if (cmd->data.setassparms.hdr.seq_no <
+           cmd->data.setassparms.hdr.number_of_replies)
+               return 1;
+       memcpy(qinfo->udata, &qinfo->no_entries, 4);
+       /* keep STRIP_ENTRIES flag so the user program can distinguish
+        * stripped entries from normal ones */
+       if (qinfo->mask_bits & QETH_QARP_STRIP_ENTRIES)
+               qdata->reply_bits |= QETH_QARP_STRIP_ENTRIES;
+       memcpy(qinfo->udata + QETH_QARP_MASK_OFFSET, &qdata->reply_bits, 2);
+       return 0;
+ out_error:
+       i = 0;
+       memcpy(qinfo->udata, &i, 4);
+       return 0;
+ }
+ static int qeth_l3_send_ipa_arp_cmd(struct qeth_card *card,
+               struct qeth_cmd_buffer *iob, int len,
+               int (*reply_cb)(struct qeth_card *, struct qeth_reply *,
+                       unsigned long),
+               void *reply_param)
+ {
+       QETH_DBF_TEXT(trace, 4, "sendarp");
+       memcpy(iob->data, IPA_PDU_HEADER, IPA_PDU_HEADER_SIZE);
+       memcpy(QETH_IPA_CMD_DEST_ADDR(iob->data),
+              &card->token.ulp_connection_r, QETH_MPC_TOKEN_LENGTH);
+       return qeth_send_control_data(card, IPA_PDU_HEADER_SIZE + len, iob,
+                                     reply_cb, reply_param);
+ }
+ static int qeth_l3_arp_query(struct qeth_card *card, char __user *udata)
+ {
+       struct qeth_cmd_buffer *iob;
+       struct qeth_arp_query_info qinfo = {0, };
+       int tmp;
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "arpquery");
+       if (!qeth_is_supported(card,/*IPA_QUERY_ARP_ADDR_INFO*/
+                              IPA_ARP_PROCESSING)) {
+               PRINT_WARN("ARP processing not supported "
+                          "on %s!\n", QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       /* get size of userspace buffer and mask_bits -> 6 bytes */
+       if (copy_from_user(&qinfo, udata, 6))
+               return -EFAULT;
+       qinfo.udata = kzalloc(qinfo.udata_len, GFP_KERNEL);
+       if (!qinfo.udata)
+               return -ENOMEM;
+       qinfo.udata_offset = QETH_QARP_ENTRIES_OFFSET;
+       iob = qeth_l3_get_setassparms_cmd(card, IPA_ARP_PROCESSING,
+                                      IPA_CMD_ASS_ARP_QUERY_INFO,
+                                      sizeof(int), QETH_PROT_IPV4);
+       rc = qeth_l3_send_ipa_arp_cmd(card, iob,
+                                  QETH_SETASS_BASE_LEN+QETH_ARP_CMD_LEN,
+                                  qeth_l3_arp_query_cb, (void *)&qinfo);
+       if (rc) {
+               tmp = rc;
+               PRINT_WARN("Error while querying ARP cache on %s: %s "
+                       "(0x%x/%d)\n", QETH_CARD_IFNAME(card),
+                       qeth_l3_arp_get_error_cause(&rc), tmp, tmp);
+               if (copy_to_user(udata, qinfo.udata, 4))
+                       rc = -EFAULT;
+       } else {
+               if (copy_to_user(udata, qinfo.udata, qinfo.udata_len))
+                       rc = -EFAULT;
+       }
+       kfree(qinfo.udata);
+       return rc;
+ }
+ static int qeth_l3_arp_add_entry(struct qeth_card *card,
+                               struct qeth_arp_cache_entry *entry)
+ {
+       struct qeth_cmd_buffer *iob;
+       char buf[16];
+       int tmp;
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "arpadent");
+       /*
+        * currently GuestLAN only supports the ARP assist function
+        * IPA_CMD_ASS_ARP_QUERY_INFO, but not IPA_CMD_ASS_ARP_ADD_ENTRY;
+        * thus we say EOPNOTSUPP for this ARP function
+        */
+       if (card->info.guestlan)
+               return -EOPNOTSUPP;
+       if (!qeth_is_supported(card, IPA_ARP_PROCESSING)) {
+               PRINT_WARN("ARP processing not supported "
+                          "on %s!\n", QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       iob = qeth_l3_get_setassparms_cmd(card, IPA_ARP_PROCESSING,
+                                      IPA_CMD_ASS_ARP_ADD_ENTRY,
+                                      sizeof(struct qeth_arp_cache_entry),
+                                      QETH_PROT_IPV4);
+       rc = qeth_l3_send_setassparms(card, iob,
+                                  sizeof(struct qeth_arp_cache_entry),
+                                  (unsigned long) entry,
+                                  qeth_l3_default_setassparms_cb, NULL);
+       if (rc) {
+               tmp = rc;
+               qeth_l3_ipaddr4_to_string((u8 *)entry->ipaddr, buf);
+               PRINT_WARN("Could not add ARP entry for address %s on %s: "
+                          "%s (0x%x/%d)\n",
+                          buf, QETH_CARD_IFNAME(card),
+                          qeth_l3_arp_get_error_cause(&rc), tmp, tmp);
+       }
+       return rc;
+ }
+ static int qeth_l3_arp_remove_entry(struct qeth_card *card,
+                               struct qeth_arp_cache_entry *entry)
+ {
+       struct qeth_cmd_buffer *iob;
+       char buf[16] = {0, };
+       int tmp;
+       int rc;
+       QETH_DBF_TEXT(trace, 3, "arprment");
+       /*
+        * currently GuestLAN only supports the ARP assist function
+        * IPA_CMD_ASS_ARP_QUERY_INFO, but not IPA_CMD_ASS_ARP_REMOVE_ENTRY;
+        * thus we say EOPNOTSUPP for this ARP function
+        */
+       if (card->info.guestlan)
+               return -EOPNOTSUPP;
+       if (!qeth_is_supported(card, IPA_ARP_PROCESSING)) {
+               PRINT_WARN("ARP processing not supported "
+                          "on %s!\n", QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       memcpy(buf, entry, 12);
+       iob = qeth_l3_get_setassparms_cmd(card, IPA_ARP_PROCESSING,
+                                      IPA_CMD_ASS_ARP_REMOVE_ENTRY,
+                                      12,
+                                      QETH_PROT_IPV4);
+       rc = qeth_l3_send_setassparms(card, iob,
+                                  12, (unsigned long)buf,
+                                  qeth_l3_default_setassparms_cb, NULL);
+       if (rc) {
+               tmp = rc;
+               memset(buf, 0, 16);
+               qeth_l3_ipaddr4_to_string((u8 *)entry->ipaddr, buf);
+               PRINT_WARN("Could not delete ARP entry for address %s on %s: "
+                          "%s (0x%x/%d)\n",
+                          buf, QETH_CARD_IFNAME(card),
+                          qeth_l3_arp_get_error_cause(&rc), tmp, tmp);
+       }
+       return rc;
+ }
+ static int qeth_l3_arp_flush_cache(struct qeth_card *card)
+ {
+       int rc;
+       int tmp;
+       QETH_DBF_TEXT(trace, 3, "arpflush");
+       /*
+        * currently GuestLAN only supports the ARP assist function
+        * IPA_CMD_ASS_ARP_QUERY_INFO, but not IPA_CMD_ASS_ARP_FLUSH_CACHE;
+        * thus we say EOPNOTSUPP for this ARP function
+       */
+       if (card->info.guestlan || (card->info.type == QETH_CARD_TYPE_IQD))
+               return -EOPNOTSUPP;
+       if (!qeth_is_supported(card, IPA_ARP_PROCESSING)) {
+               PRINT_WARN("ARP processing not supported "
+                          "on %s!\n", QETH_CARD_IFNAME(card));
+               return -EOPNOTSUPP;
+       }
+       rc = qeth_l3_send_simple_setassparms(card, IPA_ARP_PROCESSING,
+                                         IPA_CMD_ASS_ARP_FLUSH_CACHE, 0);
+       if (rc) {
+               tmp = rc;
+               PRINT_WARN("Could not flush ARP cache on %s: %s (0x%x/%d)\n",
+                       QETH_CARD_IFNAME(card),
+                       qeth_l3_arp_get_error_cause(&rc), tmp, tmp);
+       }
+       return rc;
+ }
+ static int qeth_l3_do_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       struct qeth_arp_cache_entry arp_entry;
+       struct mii_ioctl_data *mii_data;
+       int rc = 0;
+       if (!card)
+               return -ENODEV;
+       if ((card->state != CARD_STATE_UP) &&
+               (card->state != CARD_STATE_SOFTSETUP))
+               return -ENODEV;
+       switch (cmd) {
+       case SIOC_QETH_ARP_SET_NO_ENTRIES:
+               if (!capable(CAP_NET_ADMIN)) {
+                       rc = -EPERM;
+                       break;
+               }
+               rc = qeth_l3_arp_set_no_entries(card, rq->ifr_ifru.ifru_ivalue);
+               break;
+       case SIOC_QETH_ARP_QUERY_INFO:
+               if (!capable(CAP_NET_ADMIN)) {
+                       rc = -EPERM;
+                       break;
+               }
+               rc = qeth_l3_arp_query(card, rq->ifr_ifru.ifru_data);
+               break;
+       case SIOC_QETH_ARP_ADD_ENTRY:
+               if (!capable(CAP_NET_ADMIN)) {
+                       rc = -EPERM;
+                       break;
+               }
+               if (copy_from_user(&arp_entry, rq->ifr_ifru.ifru_data,
+                                  sizeof(struct qeth_arp_cache_entry)))
+                       rc = -EFAULT;
+               else
+                       rc = qeth_l3_arp_add_entry(card, &arp_entry);
+               break;
+       case SIOC_QETH_ARP_REMOVE_ENTRY:
+               if (!capable(CAP_NET_ADMIN)) {
+                       rc = -EPERM;
+                       break;
+               }
+               if (copy_from_user(&arp_entry, rq->ifr_ifru.ifru_data,
+                                  sizeof(struct qeth_arp_cache_entry)))
+                       rc = -EFAULT;
+               else
+                       rc = qeth_l3_arp_remove_entry(card, &arp_entry);
+               break;
+       case SIOC_QETH_ARP_FLUSH_CACHE:
+               if (!capable(CAP_NET_ADMIN)) {
+                       rc = -EPERM;
+                       break;
+               }
+               rc = qeth_l3_arp_flush_cache(card);
+               break;
+       case SIOC_QETH_ADP_SET_SNMP_CONTROL:
+               rc = qeth_snmp_command(card, rq->ifr_ifru.ifru_data);
+               break;
+       case SIOC_QETH_GET_CARD_TYPE:
+               if ((card->info.type == QETH_CARD_TYPE_OSAE) &&
+                   !card->info.guestlan)
+                       return 1;
+               return 0;
+               break;
+       case SIOCGMIIPHY:
+               mii_data = if_mii(rq);
+               mii_data->phy_id = 0;
+               break;
+       case SIOCGMIIREG:
+               mii_data = if_mii(rq);
+               if (mii_data->phy_id != 0)
+                       rc = -EINVAL;
+               else
+                       mii_data->val_out = qeth_mdio_read(dev,
+                                                       mii_data->phy_id,
+                                                       mii_data->reg_num);
+               break;
+       default:
+               rc = -EOPNOTSUPP;
+       }
+       if (rc)
+               QETH_DBF_TEXT_(trace, 2, "ioce%d", rc);
+       return rc;
+ }
+ static void qeth_l3_fill_header(struct qeth_card *card, struct qeth_hdr *hdr,
+               struct sk_buff *skb, int ipv, int cast_type)
+ {
+       QETH_DBF_TEXT(trace, 6, "fillhdr");
+       memset(hdr, 0, sizeof(struct qeth_hdr));
+       hdr->hdr.l3.id = QETH_HEADER_TYPE_LAYER3;
+       hdr->hdr.l3.ext_flags = 0;
+       /*
+        * before we're going to overwrite this location with next hop ip.
+        * v6 uses passthrough, v4 sets the tag in the QDIO header.
+        */
+       if (card->vlangrp && vlan_tx_tag_present(skb)) {
+               hdr->hdr.l3.ext_flags = (ipv == 4) ?
+                       QETH_HDR_EXT_VLAN_FRAME :
+                       QETH_HDR_EXT_INCLUDE_VLAN_TAG;
+               hdr->hdr.l3.vlan_id = vlan_tx_tag_get(skb);
+       }
+       hdr->hdr.l3.length = skb->len - sizeof(struct qeth_hdr);
+       if (ipv == 4) {
+               /* IPv4 */
+               hdr->hdr.l3.flags = qeth_l3_get_qeth_hdr_flags4(cast_type);
+               memset(hdr->hdr.l3.dest_addr, 0, 12);
+               if ((skb->dst) && (skb->dst->neighbour)) {
+                       *((u32 *) (&hdr->hdr.l3.dest_addr[12])) =
+                           *((u32 *) skb->dst->neighbour->primary_key);
+               } else {
+                       /* fill in destination address used in ip header */
+                       *((u32 *) (&hdr->hdr.l3.dest_addr[12])) =
+                                                       ip_hdr(skb)->daddr;
+               }
+       } else if (ipv == 6) {
+               /* IPv6 */
+               hdr->hdr.l3.flags = qeth_l3_get_qeth_hdr_flags6(cast_type);
+               if (card->info.type == QETH_CARD_TYPE_IQD)
+                       hdr->hdr.l3.flags &= ~QETH_HDR_PASSTHRU;
+               if ((skb->dst) && (skb->dst->neighbour)) {
+                       memcpy(hdr->hdr.l3.dest_addr,
+                              skb->dst->neighbour->primary_key, 16);
+               } else {
+                       /* fill in destination address used in ip header */
+                       memcpy(hdr->hdr.l3.dest_addr,
+                              &ipv6_hdr(skb)->daddr, 16);
+               }
+       } else {
+               /* passthrough */
+               if ((skb->dev->type == ARPHRD_IEEE802_TR) &&
+                       !memcmp(skb->data + sizeof(struct qeth_hdr) +
+                       sizeof(__u16), skb->dev->broadcast, 6)) {
+                       hdr->hdr.l3.flags = QETH_CAST_BROADCAST |
+                                               QETH_HDR_PASSTHRU;
+               } else if (!memcmp(skb->data + sizeof(struct qeth_hdr),
+                           skb->dev->broadcast, 6)) {
+                       /* broadcast? */
+                       hdr->hdr.l3.flags = QETH_CAST_BROADCAST |
+                                               QETH_HDR_PASSTHRU;
+               } else {
+                       hdr->hdr.l3.flags = (cast_type == RTN_MULTICAST) ?
+                               QETH_CAST_MULTICAST | QETH_HDR_PASSTHRU :
+                               QETH_CAST_UNICAST | QETH_HDR_PASSTHRU;
+               }
+       }
+ }
+ static int qeth_l3_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
+ {
+       int rc;
+       u16 *tag;
+       struct qeth_hdr *hdr = NULL;
+       int elements_needed = 0;
+       struct qeth_card *card = netdev_priv(dev);
+       struct sk_buff *new_skb = NULL;
+       int ipv = qeth_get_ip_version(skb);
+       int cast_type = qeth_get_cast_type(card, skb);
+       struct qeth_qdio_out_q *queue = card->qdio.out_qs
+               [qeth_get_priority_queue(card, skb, ipv, cast_type)];
+       int tx_bytes = skb->len;
+       enum qeth_large_send_types large_send = QETH_LARGE_SEND_NO;
+       struct qeth_eddp_context *ctx = NULL;
+       QETH_DBF_TEXT(trace, 6, "l3xmit");
+       if ((card->info.type == QETH_CARD_TYPE_IQD) &&
+           (skb->protocol != htons(ETH_P_IPV6)) &&
+           (skb->protocol != htons(ETH_P_IP)))
+                       goto tx_drop;
+       if ((card->state != CARD_STATE_UP) || !card->lan_online) {
+               card->stats.tx_carrier_errors++;
+               goto tx_drop;
+       }
+       if ((cast_type == RTN_BROADCAST) &&
+           (card->info.broadcast_capable == 0))
+               goto tx_drop;
+       if (card->options.performance_stats) {
+               card->perf_stats.outbound_cnt++;
+               card->perf_stats.outbound_start_time = qeth_get_micros();
+       }
+       /* create a clone with writeable headroom */
+       new_skb = skb_realloc_headroom(skb, sizeof(struct qeth_hdr_tso) +
+                                       VLAN_HLEN);
+       if (!new_skb)
+               goto tx_drop;
+       if (card->info.type == QETH_CARD_TYPE_IQD) {
+               skb_pull(new_skb, ETH_HLEN);
+       } else {
+               if (new_skb->protocol == htons(ETH_P_IP)) {
+                       if (card->dev->type == ARPHRD_IEEE802_TR)
+                               skb_pull(new_skb, TR_HLEN);
+                       else
+                               skb_pull(new_skb, ETH_HLEN);
+               }
+               if (new_skb->protocol == ETH_P_IPV6 && card->vlangrp &&
+                               vlan_tx_tag_present(new_skb)) {
+                       skb_push(new_skb, VLAN_HLEN);
+                       skb_copy_to_linear_data(new_skb, new_skb->data + 4, 4);
+                       skb_copy_to_linear_data_offset(new_skb, 4,
+                               new_skb->data + 8, 4);
+                       skb_copy_to_linear_data_offset(new_skb, 8,
+                               new_skb->data + 12, 4);
+                       tag = (u16 *)(new_skb->data + 12);
+                       *tag = __constant_htons(ETH_P_8021Q);
+                       *(tag + 1) = htons(vlan_tx_tag_get(new_skb));
+                       VLAN_TX_SKB_CB(new_skb)->magic = 0;
+               }
+       }
+       netif_stop_queue(dev);
+       if (skb_is_gso(new_skb))
+               large_send = card->options.large_send;
+       /* fix hardware limitation: as long as we do not have sbal
+        * chaining we can not send long frag lists so we temporary
+        * switch to EDDP
+        */
+       if ((large_send == QETH_LARGE_SEND_TSO) &&
+               ((skb_shinfo(new_skb)->nr_frags + 2) > 16))
+               large_send = QETH_LARGE_SEND_EDDP;
+       if ((large_send == QETH_LARGE_SEND_TSO) &&
+           (cast_type == RTN_UNSPEC)) {
+               hdr = (struct qeth_hdr *)skb_push(new_skb,
+                                               sizeof(struct qeth_hdr_tso));
+               memset(hdr, 0, sizeof(struct qeth_hdr_tso));
+               qeth_l3_fill_header(card, hdr, new_skb, ipv, cast_type);
+               qeth_tso_fill_header(card, hdr, new_skb);
+               elements_needed++;
+       } else {
+               hdr = (struct qeth_hdr *)skb_push(new_skb,
+                                               sizeof(struct qeth_hdr));
+               qeth_l3_fill_header(card, hdr, new_skb, ipv, cast_type);
+       }
+       if (large_send == QETH_LARGE_SEND_EDDP) {
+               /* new_skb is not owned by a socket so we use skb to get
+                * the protocol
+                */
+               ctx = qeth_eddp_create_context(card, new_skb, hdr,
+                                               skb->sk->sk_protocol);
+               if (ctx == NULL) {
+                       PRINT_WARN("could not create eddp context\n");
+                       goto tx_drop;
+               }
+       } else {
+               int elems = qeth_get_elements_no(card, (void *)hdr, new_skb,
+                                                elements_needed);
+               if (!elems)
+                       goto tx_drop;
+               elements_needed += elems;
+       }
+       if ((large_send == QETH_LARGE_SEND_NO) &&
+           (new_skb->ip_summed == CHECKSUM_PARTIAL))
+               qeth_tx_csum(new_skb);
+       if (card->info.type != QETH_CARD_TYPE_IQD)
+               rc = qeth_do_send_packet(card, queue, new_skb, hdr,
+                                        elements_needed, ctx);
+       else
+               rc = qeth_do_send_packet_fast(card, queue, new_skb, hdr,
+                                             elements_needed, ctx);
+       if (!rc) {
+               card->stats.tx_packets++;
+               card->stats.tx_bytes += tx_bytes;
+               if (new_skb != skb)
+                       dev_kfree_skb_any(skb);
+               if (card->options.performance_stats) {
+                       if (large_send != QETH_LARGE_SEND_NO) {
+                               card->perf_stats.large_send_bytes += tx_bytes;
+                               card->perf_stats.large_send_cnt++;
+                       }
+                       if (skb_shinfo(new_skb)->nr_frags > 0) {
+                               card->perf_stats.sg_skbs_sent++;
+                               /* nr_frags + skb->data */
+                               card->perf_stats.sg_frags_sent +=
+                                       skb_shinfo(new_skb)->nr_frags + 1;
+                       }
+               }
+               if (ctx != NULL) {
+                       qeth_eddp_put_context(ctx);
+                       dev_kfree_skb_any(new_skb);
+               }
+       } else {
+               if (ctx != NULL)
+                       qeth_eddp_put_context(ctx);
+               if (rc == -EBUSY) {
+                       if (new_skb != skb)
+                               dev_kfree_skb_any(new_skb);
+                       return NETDEV_TX_BUSY;
+               } else
+                       goto tx_drop;
+       }
+       netif_wake_queue(dev);
+       if (card->options.performance_stats)
+               card->perf_stats.outbound_time += qeth_get_micros() -
+                       card->perf_stats.outbound_start_time;
+       return rc;
+ tx_drop:
+       card->stats.tx_dropped++;
+       card->stats.tx_errors++;
+       if ((new_skb != skb) && new_skb)
+               dev_kfree_skb_any(new_skb);
+       dev_kfree_skb_any(skb);
+       return NETDEV_TX_OK;
+ }
+ static int qeth_l3_open(struct net_device *dev)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       QETH_DBF_TEXT(trace, 4, "qethopen");
+       if (card->state != CARD_STATE_SOFTSETUP)
+               return -ENODEV;
+       card->data.state = CH_STATE_UP;
+       card->state = CARD_STATE_UP;
+       card->dev->flags |= IFF_UP;
+       netif_start_queue(dev);
+       if (!card->lan_online && netif_carrier_ok(dev))
+               netif_carrier_off(dev);
+       return 0;
+ }
+ static int qeth_l3_stop(struct net_device *dev)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       QETH_DBF_TEXT(trace, 4, "qethstop");
+       netif_tx_disable(dev);
+       card->dev->flags &= ~IFF_UP;
+       if (card->state == CARD_STATE_UP)
+               card->state = CARD_STATE_SOFTSETUP;
+       return 0;
+ }
+ static u32 qeth_l3_ethtool_get_rx_csum(struct net_device *dev)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       return (card->options.checksum_type == HW_CHECKSUMMING);
+ }
+ static int qeth_l3_ethtool_set_rx_csum(struct net_device *dev, u32 data)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       enum qeth_card_states old_state;
+       enum qeth_checksum_types csum_type;
+       if ((card->state != CARD_STATE_UP) &&
+           (card->state != CARD_STATE_DOWN))
+               return -EPERM;
+       if (data)
+               csum_type = HW_CHECKSUMMING;
+       else
+               csum_type = SW_CHECKSUMMING;
+       if (card->options.checksum_type != csum_type) {
+               old_state = card->state;
+               if (card->state == CARD_STATE_UP)
+                       __qeth_l3_set_offline(card->gdev, 1);
+               card->options.checksum_type = csum_type;
+               if (old_state == CARD_STATE_UP)
+                       __qeth_l3_set_online(card->gdev, 1);
+       }
+       return 0;
+ }
+ static int qeth_l3_ethtool_set_tso(struct net_device *dev, u32 data)
+ {
+       struct qeth_card *card = netdev_priv(dev);
+       if (data) {
+               if (card->options.large_send == QETH_LARGE_SEND_NO) {
+                       if (card->info.type == QETH_CARD_TYPE_IQD)
+                               card->options.large_send = QETH_LARGE_SEND_EDDP;
+                       else
+                               card->options.large_send = QETH_LARGE_SEND_TSO;
+                       dev->features |= NETIF_F_TSO;
+               }
+       } else {
+               dev->features &= ~NETIF_F_TSO;
+               card->options.large_send = QETH_LARGE_SEND_NO;
+       }
+       return 0;
+ }
+ static struct ethtool_ops qeth_l3_ethtool_ops = {
+       .get_link = ethtool_op_get_link,
+       .get_tx_csum = ethtool_op_get_tx_csum,
+       .set_tx_csum = ethtool_op_set_tx_hw_csum,
+       .get_rx_csum = qeth_l3_ethtool_get_rx_csum,
+       .set_rx_csum = qeth_l3_ethtool_set_rx_csum,
+       .get_sg      = ethtool_op_get_sg,
+       .set_sg      = ethtool_op_set_sg,
+       .get_tso     = ethtool_op_get_tso,
+       .set_tso     = qeth_l3_ethtool_set_tso,
+       .get_strings = qeth_core_get_strings,
+       .get_ethtool_stats = qeth_core_get_ethtool_stats,
+       .get_stats_count = qeth_core_get_stats_count,
+       .get_drvinfo = qeth_core_get_drvinfo,
+ };
+ /*
+  * we need NOARP for IPv4 but we want neighbor solicitation for IPv6. Setting
+  * NOARP on the netdevice is no option because it also turns off neighbor
+  * solicitation. For IPv4 we install a neighbor_setup function. We don't want
+  * arp resolution but we want the hard header (packet socket will work
+  * e.g. tcpdump)
+  */
+ static int qeth_l3_neigh_setup_noarp(struct neighbour *n)
+ {
+       n->nud_state = NUD_NOARP;
+       memcpy(n->ha, "FAKELL", 6);
+       n->output = n->ops->connected_output;
+       return 0;
+ }
+ static int
+ qeth_l3_neigh_setup(struct net_device *dev, struct neigh_parms *np)
+ {
+       if (np->tbl->family == AF_INET)
+               np->neigh_setup = qeth_l3_neigh_setup_noarp;
+       return 0;
+ }
+ static int qeth_l3_setup_netdev(struct qeth_card *card)
+ {
+       if (card->info.type == QETH_CARD_TYPE_OSAE) {
+               if ((card->info.link_type == QETH_LINK_TYPE_LANE_TR) ||
+                   (card->info.link_type == QETH_LINK_TYPE_HSTR)) {
+ #ifdef CONFIG_TR
+                       card->dev = alloc_trdev(0);
+ #endif
+                       if (!card->dev)
+                               return -ENODEV;
+               } else {
+                       card->dev = alloc_etherdev(0);
+                       if (!card->dev)
+                               return -ENODEV;
+                       card->dev->neigh_setup = qeth_l3_neigh_setup;
+                       /*IPv6 address autoconfiguration stuff*/
+                       qeth_l3_get_unique_id(card);
+                       if (!(card->info.unique_id & UNIQUE_ID_NOT_BY_CARD))
+                               card->dev->dev_id = card->info.unique_id &
+                                                        0xffff;
+               }
+       } else if (card->info.type == QETH_CARD_TYPE_IQD) {
+               card->dev = alloc_netdev(0, "hsi%d", ether_setup);
+               if (!card->dev)
+                       return -ENODEV;
+               card->dev->flags |= IFF_NOARP;
+               qeth_l3_iqd_read_initial_mac(card);
+       } else
+               return -ENODEV;
+       card->dev->hard_start_xmit = qeth_l3_hard_start_xmit;
+       card->dev->priv = card;
+       card->dev->tx_timeout = &qeth_tx_timeout;
+       card->dev->watchdog_timeo = QETH_TX_TIMEOUT;
+       card->dev->open = qeth_l3_open;
+       card->dev->stop = qeth_l3_stop;
+       card->dev->do_ioctl = qeth_l3_do_ioctl;
+       card->dev->get_stats = qeth_get_stats;
+       card->dev->change_mtu = qeth_change_mtu;
+       card->dev->set_multicast_list = qeth_l3_set_multicast_list;
+       card->dev->vlan_rx_register = qeth_l3_vlan_rx_register;
+       card->dev->vlan_rx_add_vid = qeth_l3_vlan_rx_add_vid;
+       card->dev->vlan_rx_kill_vid = qeth_l3_vlan_rx_kill_vid;
+       card->dev->mtu = card->info.initial_mtu;
+       SET_ETHTOOL_OPS(card->dev, &qeth_l3_ethtool_ops);
+       card->dev->features |=  NETIF_F_HW_VLAN_TX |
+                               NETIF_F_HW_VLAN_RX |
+                               NETIF_F_HW_VLAN_FILTER;
+       SET_NETDEV_DEV(card->dev, &card->gdev->dev);
+       return register_netdev(card->dev);
+ }
+ static void qeth_l3_qdio_input_handler(struct ccw_device *ccwdev,
+               unsigned int status, unsigned int qdio_err,
+               unsigned int siga_err, unsigned int queue, int first_element,
+               int count, unsigned long card_ptr)
+ {
+       struct net_device *net_dev;
+       struct qeth_card *card;
+       struct qeth_qdio_buffer *buffer;
+       int index;
+       int i;
+       QETH_DBF_TEXT(trace, 6, "qdinput");
+       card = (struct qeth_card *) card_ptr;
+       net_dev = card->dev;
+       if (card->options.performance_stats) {
+               card->perf_stats.inbound_cnt++;
+               card->perf_stats.inbound_start_time = qeth_get_micros();
+       }
+       if (status & QDIO_STATUS_LOOK_FOR_ERROR) {
+               if (status & QDIO_STATUS_ACTIVATE_CHECK_CONDITION) {
+                       QETH_DBF_TEXT(trace, 1, "qdinchk");
+                       QETH_DBF_TEXT_(trace, 1, "%s", CARD_BUS_ID(card));
+                       QETH_DBF_TEXT_(trace, 1, "%04X%04X",
+                                       first_element, count);
+                       QETH_DBF_TEXT_(trace, 1, "%04X%04X", queue, status);
+                       qeth_schedule_recovery(card);
+                       return;
+               }
+       }
+       for (i = first_element; i < (first_element + count); ++i) {
+               index = i % QDIO_MAX_BUFFERS_PER_Q;
+               buffer = &card->qdio.in_q->bufs[index];
+               if (!((status & QDIO_STATUS_LOOK_FOR_ERROR) &&
+                     qeth_check_qdio_errors(buffer->buffer,
+                                            qdio_err, siga_err, "qinerr")))
+                       qeth_l3_process_inbound_buffer(card, buffer, index);
+               /* clear buffer and give back to hardware */
+               qeth_put_buffer_pool_entry(card, buffer->pool_entry);
+               qeth_queue_input_buffer(card, index);
+       }
+       if (card->options.performance_stats)
+               card->perf_stats.inbound_time += qeth_get_micros() -
+                       card->perf_stats.inbound_start_time;
+ }
+ static int qeth_l3_probe_device(struct ccwgroup_device *gdev)
+ {
+       struct qeth_card *card = dev_get_drvdata(&gdev->dev);
+       qeth_l3_create_device_attributes(&gdev->dev);
+       card->options.layer2 = 0;
+       card->discipline.input_handler = (qdio_handler_t *)
+               qeth_l3_qdio_input_handler;
+       card->discipline.output_handler = (qdio_handler_t *)
+               qeth_qdio_output_handler;
+       card->discipline.recover = qeth_l3_recover;
+       return 0;
+ }
+ static void qeth_l3_remove_device(struct ccwgroup_device *cgdev)
+ {
+       struct qeth_card *card = dev_get_drvdata(&cgdev->dev);
+       wait_event(card->wait_q, qeth_threads_running(card, 0xffffffff) == 0);
+       if (cgdev->state == CCWGROUP_ONLINE) {
+               card->use_hard_stop = 1;
+               qeth_l3_set_offline(cgdev);
+       }
+       if (card->dev) {
+               unregister_netdev(card->dev);
+               card->dev = NULL;
+       }
+       qeth_l3_remove_device_attributes(&cgdev->dev);
+       qeth_l3_clear_ip_list(card, 0, 0);
+       qeth_l3_clear_ipato_list(card);
+       return;
+ }
+ static int __qeth_l3_set_online(struct ccwgroup_device *gdev, int recovery_mode)
+ {
+       struct qeth_card *card = dev_get_drvdata(&gdev->dev);
+       int rc = 0;
+       enum qeth_card_states recover_flag;
+       BUG_ON(!card);
+       QETH_DBF_TEXT(setup, 2, "setonlin");
+       QETH_DBF_HEX(setup, 2, &card, sizeof(void *));
+       qeth_set_allowed_threads(card, QETH_RECOVER_THREAD, 1);
+       if (qeth_wait_for_threads(card, ~QETH_RECOVER_THREAD)) {
+               PRINT_WARN("set_online of card %s interrupted by user!\n",
+                          CARD_BUS_ID(card));
+               return -ERESTARTSYS;
+       }
+       recover_flag = card->state;
+       rc = ccw_device_set_online(CARD_RDEV(card));
+       if (rc) {
+               QETH_DBF_TEXT_(setup, 2, "1err%d", rc);
+               return -EIO;
+       }
+       rc = ccw_device_set_online(CARD_WDEV(card));
+       if (rc) {
+               QETH_DBF_TEXT_(setup, 2, "1err%d", rc);
+               return -EIO;
+       }
+       rc = ccw_device_set_online(CARD_DDEV(card));
+       if (rc) {
+               QETH_DBF_TEXT_(setup, 2, "1err%d", rc);
+               return -EIO;
+       }
+       rc = qeth_core_hardsetup_card(card);
+       if (rc) {
+               QETH_DBF_TEXT_(setup, 2, "2err%d", rc);
+               goto out_remove;
+       }
+       qeth_l3_query_ipassists(card, QETH_PROT_IPV4);
+       if (!card->dev && qeth_l3_setup_netdev(card))
+               goto out_remove;
+       card->state = CARD_STATE_HARDSETUP;
+       qeth_print_status_message(card);
+       /* softsetup */
+       QETH_DBF_TEXT(setup, 2, "softsetp");
+       rc = qeth_send_startlan(card);
+       if (rc) {
+               QETH_DBF_TEXT_(setup, 2, "1err%d", rc);
+               if (rc == 0xe080) {
+                       PRINT_WARN("LAN on card %s if offline! "
+                                  "Waiting for STARTLAN from card.\n",
+                                  CARD_BUS_ID(card));
+                       card->lan_online = 0;
+               }
+               return rc;
+       } else
+               card->lan_online = 1;
+       qeth_set_large_send(card, card->options.large_send);
+       rc = qeth_l3_setadapter_parms(card);
+       if (rc)
+               QETH_DBF_TEXT_(setup, 2, "2err%d", rc);
+       rc = qeth_l3_start_ipassists(card);
+       if (rc)
+               QETH_DBF_TEXT_(setup, 2, "3err%d", rc);
+       rc = qeth_l3_setrouting_v4(card);
+       if (rc)
+               QETH_DBF_TEXT_(setup, 2, "4err%d", rc);
+       rc = qeth_l3_setrouting_v6(card);
+       if (rc)
+               QETH_DBF_TEXT_(setup, 2, "5err%d", rc);
+       netif_tx_disable(card->dev);
+       rc = qeth_init_qdio_queues(card);
+       if (rc) {
+               QETH_DBF_TEXT_(setup, 2, "6err%d", rc);
+               goto out_remove;
+       }
+       card->state = CARD_STATE_SOFTSETUP;
+       netif_carrier_on(card->dev);
+       qeth_set_allowed_threads(card, 0xffffffff, 0);
+       if ((recover_flag == CARD_STATE_RECOVER) && recovery_mode) {
+                       qeth_l3_open(card->dev);
+                       qeth_l3_set_multicast_list(card->dev);
+       }
+       /* let user_space know that device is online */
+       kobject_uevent(&gdev->dev.kobj, KOBJ_CHANGE);
+       return 0;
+ out_remove:
+       card->use_hard_stop = 1;
+       qeth_l3_stop_card(card, 0);
+       ccw_device_set_offline(CARD_DDEV(card));
+       ccw_device_set_offline(CARD_WDEV(card));
+       ccw_device_set_offline(CARD_RDEV(card));
+       if (recover_flag == CARD_STATE_RECOVER)
+               card->state = CARD_STATE_RECOVER;
+       else
+               card->state = CARD_STATE_DOWN;
+       return -ENODEV;
+ }
+ static int qeth_l3_set_online(struct ccwgroup_device *gdev)
+ {
+       return __qeth_l3_set_online(gdev, 0);
+ }
+ static int __qeth_l3_set_offline(struct ccwgroup_device *cgdev,
+                       int recovery_mode)
+ {
+       struct qeth_card *card = dev_get_drvdata(&cgdev->dev);
+       int rc = 0, rc2 = 0, rc3 = 0;
+       enum qeth_card_states recover_flag;
+       QETH_DBF_TEXT(setup, 3, "setoffl");
+       QETH_DBF_HEX(setup, 3, &card, sizeof(void *));
+       if (card->dev && netif_carrier_ok(card->dev))
+               netif_carrier_off(card->dev);
+       recover_flag = card->state;
+       if (qeth_l3_stop_card(card, recovery_mode) == -ERESTARTSYS) {
+               PRINT_WARN("Stopping card %s interrupted by user!\n",
+                          CARD_BUS_ID(card));
+               return -ERESTARTSYS;
+       }
+       rc  = ccw_device_set_offline(CARD_DDEV(card));
+       rc2 = ccw_device_set_offline(CARD_WDEV(card));
+       rc3 = ccw_device_set_offline(CARD_RDEV(card));
+       if (!rc)
+               rc = (rc2) ? rc2 : rc3;
+       if (rc)
+               QETH_DBF_TEXT_(setup, 2, "1err%d", rc);
+       if (recover_flag == CARD_STATE_UP)
+               card->state = CARD_STATE_RECOVER;
+       /* let user_space know that device is offline */
+       kobject_uevent(&cgdev->dev.kobj, KOBJ_CHANGE);
+       return 0;
+ }
+ static int qeth_l3_set_offline(struct ccwgroup_device *cgdev)
+ {
+       return __qeth_l3_set_offline(cgdev, 0);
+ }
+ static int qeth_l3_recover(void *ptr)
+ {
+       struct qeth_card *card;
+       int rc = 0;
+       card = (struct qeth_card *) ptr;
+       QETH_DBF_TEXT(trace, 2, "recover1");
+       QETH_DBF_HEX(trace, 2, &card, sizeof(void *));
+       if (!qeth_do_run_thread(card, QETH_RECOVER_THREAD))
+               return 0;
+       QETH_DBF_TEXT(trace, 2, "recover2");
+       PRINT_WARN("Recovery of device %s started ...\n",
+                  CARD_BUS_ID(card));
+       card->use_hard_stop = 1;
+       __qeth_l3_set_offline(card->gdev, 1);
+       rc = __qeth_l3_set_online(card->gdev, 1);
+       /* don't run another scheduled recovery */
+       qeth_clear_thread_start_bit(card, QETH_RECOVER_THREAD);
+       qeth_clear_thread_running_bit(card, QETH_RECOVER_THREAD);
+       if (!rc)
+               PRINT_INFO("Device %s successfully recovered!\n",
+                          CARD_BUS_ID(card));
+       else
+               PRINT_INFO("Device %s could not be recovered!\n",
+                          CARD_BUS_ID(card));
+       return 0;
+ }
+ static void qeth_l3_shutdown(struct ccwgroup_device *gdev)
+ {
+       struct qeth_card *card = dev_get_drvdata(&gdev->dev);
+       qeth_l3_clear_ip_list(card, 0, 0);
+       qeth_qdio_clear_card(card, 0);
+       qeth_clear_qdio_buffers(card);
+ }
+ struct ccwgroup_driver qeth_l3_ccwgroup_driver = {
+       .probe = qeth_l3_probe_device,
+       .remove = qeth_l3_remove_device,
+       .set_online = qeth_l3_set_online,
+       .set_offline = qeth_l3_set_offline,
+       .shutdown = qeth_l3_shutdown,
+ };
+ EXPORT_SYMBOL_GPL(qeth_l3_ccwgroup_driver);
+ static int qeth_l3_ip_event(struct notifier_block *this,
 -                              unsigned long event, void *ptr)
++                          unsigned long event, void *ptr)
+ {
+       struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
+       struct net_device *dev = (struct net_device *)ifa->ifa_dev->dev;
+       struct qeth_ipaddr *addr;
+       struct qeth_card *card;
++      if (dev->nd_net != &init_net)
++              return NOTIFY_DONE;
++
+       QETH_DBF_TEXT(trace, 3, "ipevent");
+       card = qeth_l3_get_card_from_dev(dev);
+       if (!card)
+               return NOTIFY_DONE;
+       addr = qeth_l3_get_addr_buffer(QETH_PROT_IPV4);
+       if (addr != NULL) {
+               addr->u.a4.addr = ifa->ifa_address;
+               addr->u.a4.mask = ifa->ifa_mask;
+               addr->type = QETH_IP_TYPE_NORMAL;
+       } else
+               goto out;
+       switch (event) {
+       case NETDEV_UP:
+               if (!qeth_l3_add_ip(card, addr))
+                       kfree(addr);
+               break;
+       case NETDEV_DOWN:
+               if (!qeth_l3_delete_ip(card, addr))
+                       kfree(addr);
+               break;
+       default:
+               break;
+       }
+       qeth_l3_set_ip_addr_list(card);
+ out:
+       return NOTIFY_DONE;
+ }
+ static struct notifier_block qeth_l3_ip_notifier = {
+       qeth_l3_ip_event,
+       NULL,
+ };
+ #ifdef CONFIG_QETH_IPV6
+ /**
+  * IPv6 event handler
+  */
+ static int qeth_l3_ip6_event(struct notifier_block *this,
++                           unsigned long event, void *ptr)
+ {
+       struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
+       struct net_device *dev = (struct net_device *)ifa->idev->dev;
+       struct qeth_ipaddr *addr;
+       struct qeth_card *card;
+       QETH_DBF_TEXT(trace, 3, "ip6event");
+       card = qeth_l3_get_card_from_dev(dev);
+       if (!card)
+               return NOTIFY_DONE;
+       if (!qeth_is_supported(card, IPA_IPV6))
+               return NOTIFY_DONE;
+       addr = qeth_l3_get_addr_buffer(QETH_PROT_IPV6);
+       if (addr != NULL) {
+               memcpy(&addr->u.a6.addr, &ifa->addr, sizeof(struct in6_addr));
+               addr->u.a6.pfxlen = ifa->prefix_len;
+               addr->type = QETH_IP_TYPE_NORMAL;
+       } else
+               goto out;
+       switch (event) {
+       case NETDEV_UP:
+               if (!qeth_l3_add_ip(card, addr))
+                       kfree(addr);
+               break;
+       case NETDEV_DOWN:
+               if (!qeth_l3_delete_ip(card, addr))
+                       kfree(addr);
+               break;
+       default:
+               break;
+       }
+       qeth_l3_set_ip_addr_list(card);
+ out:
+       return NOTIFY_DONE;
+ }
+ static struct notifier_block qeth_l3_ip6_notifier = {
+       qeth_l3_ip6_event,
+       NULL,
+ };
+ #endif
+ static int qeth_l3_register_notifiers(void)
+ {
+       int rc;
+       QETH_DBF_TEXT(trace, 5, "regnotif");
+       rc = register_inetaddr_notifier(&qeth_l3_ip_notifier);
+       if (rc)
+               return rc;
+ #ifdef CONFIG_QETH_IPV6
+       rc = register_inet6addr_notifier(&qeth_l3_ip6_notifier);
+       if (rc) {
+               unregister_inetaddr_notifier(&qeth_l3_ip_notifier);
+               return rc;
+       }
+ #else
+       PRINT_WARN("layer 3 discipline no IPv6 support\n");
+ #endif
+       return 0;
+ }
+ static void qeth_l3_unregister_notifiers(void)
+ {
+       QETH_DBF_TEXT(trace, 5, "unregnot");
+       BUG_ON(unregister_inetaddr_notifier(&qeth_l3_ip_notifier));
+ #ifdef CONFIG_QETH_IPV6
+       BUG_ON(unregister_inet6addr_notifier(&qeth_l3_ip6_notifier));
+ #endif /* QETH_IPV6 */
+ }
+ static int __init qeth_l3_init(void)
+ {
+       int rc = 0;
+       PRINT_INFO("register layer 3 discipline\n");
+       rc = qeth_l3_register_notifiers();
+       return rc;
+ }
+ static void __exit qeth_l3_exit(void)
+ {
+       qeth_l3_unregister_notifiers();
+       PRINT_INFO("unregister layer 3 discipline\n");
+ }
+ module_init(qeth_l3_init);
+ module_exit(qeth_l3_exit);
+ MODULE_AUTHOR("Frank Blaschka <frank.blaschka@de.ibm.com>");
+ MODULE_DESCRIPTION("qeth layer 3 discipline");
+ MODULE_LICENSE("GPL");