fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / drivers / net / r8169.c
index 0e27077..1aafa71 100644 (file)
@@ -6,25 +6,26 @@
  History:
  Feb  4 2002   - created initially by ShuChen <shuchen@realtek.com.tw>.
  May 20 2002   - Add link status force-mode and TBI mode support.
+       2004    - Massive updates. See kernel SCM system for details.
 =========================================================================
-  1. The media can be forced in 5 modes.
+  1. [DEPRECATED: use ethtool instead] The media can be forced in 5 modes.
         Command: 'insmod r8169 media = SET_MEDIA'
         Ex:      'insmod r8169 media = 0x04' will force PHY to operate in 100Mpbs Half-duplex.
-       
+
         SET_MEDIA can be:
                _10_Half        = 0x01
                _10_Full        = 0x02
                _100_Half       = 0x04
                _100_Full       = 0x08
                _1000_Full      = 0x10
-  
+
   2. Support TBI mode.
 =========================================================================
 VERSION 1.1    <2002/10/4>
 
        The bit4:0 of MII register 4 is called "selector field", and have to be
        00001b to indicate support of IEEE std 802.3 during NWay process of
-       exchanging Link Code Word (FLP). 
+       exchanging Link Code Word (FLP).
 
 VERSION 1.2    <2002/11/30>
 
@@ -33,72 +34,112 @@ VERSION 1.2        <2002/11/30>
        - Copy mc_filter setup code from 8139cp
          (includes an optimization, and avoids set_bit use)
 
-*/
+VERSION 1.6LK  <2004/04/14>
+
+       - Merge of Realtek's version 1.6
+       - Conversion to DMA API
+       - Suspend/resume
+       - Endianness
+       - Misc Rx/Tx bugs
+
+VERSION 2.2LK  <2005/01/25>
+
+       - RX csum, TX csum/SG, TSO
+       - VLAN
+       - baby (< 7200) Jumbo frames support
+       - Merge of Realtek's version 2.2 (new phy)
+ */
 
 #include <linux/module.h>
+#include <linux/moduleparam.h>
 #include <linux/pci.h>
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
 #include <linux/delay.h>
 #include <linux/ethtool.h>
+#include <linux/mii.h>
+#include <linux/if_vlan.h>
 #include <linux/crc32.h>
+#include <linux/in.h>
+#include <linux/ip.h>
+#include <linux/tcp.h>
 #include <linux/init.h>
 #include <linux/dma-mapping.h>
 
 #include <asm/io.h>
+#include <asm/irq.h>
+
+#ifdef CONFIG_R8169_NAPI
+#define NAPI_SUFFIX    "-NAPI"
+#else
+#define NAPI_SUFFIX    ""
+#endif
 
-#define RTL8169_VERSION "1.2"
+#define RTL8169_VERSION "2.2LK" NAPI_SUFFIX
 #define MODULENAME "r8169"
-#define RTL8169_DRIVER_NAME   MODULENAME " Gigabit Ethernet driver " RTL8169_VERSION
 #define PFX MODULENAME ": "
 
 #ifdef RTL8169_DEBUG
 #define assert(expr) \
-        if(!(expr)) {                                  \
-               printk( "Assertion failed! %s,%s,%s,line=%d\n", \
-               #expr,__FILE__,__FUNCTION__,__LINE__);          \
-        }
-#define dprintk(fmt, args...)  do { printk(PFX fmt, ## args) } while (0)
+       if (!(expr)) {                                  \
+               printk( "Assertion failed! %s,%s,%s,line=%d\n", \
+               #expr,__FILE__,__FUNCTION__,__LINE__);          \
+       }
+#define dprintk(fmt, args...)  do { printk(PFX fmt, ## args); } while (0)
 #else
 #define assert(expr) do {} while (0)
 #define dprintk(fmt, args...)  do {} while (0)
 #endif /* RTL8169_DEBUG */
 
+#define R8169_MSG_DEFAULT \
+       (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN)
+
+#define TX_BUFFS_AVAIL(tp) \
+       (tp->dirty_tx + NUM_TX_DESC - tp->cur_tx - 1)
+
+#ifdef CONFIG_R8169_NAPI
+#define rtl8169_rx_skb                 netif_receive_skb
+#define rtl8169_rx_hwaccel_skb         vlan_hwaccel_receive_skb
+#define rtl8169_rx_quota(count, quota) min(count, quota)
+#else
+#define rtl8169_rx_skb                 netif_rx
+#define rtl8169_rx_hwaccel_skb         vlan_hwaccel_rx
+#define rtl8169_rx_quota(count, quota) count
+#endif
+
 /* media options */
 #define MAX_UNITS 8
 static int media[MAX_UNITS] = { -1, -1, -1, -1, -1, -1, -1, -1 };
+static int num_media = 0;
 
 /* Maximum events (Rx packets, etc.) to handle at each interrupt. */
-static int max_interrupt_work = 20;
+static const int max_interrupt_work = 20;
 
 /* Maximum number of multicast addresses to filter (vs. Rx-all-multicast).
-   The RTL chips use a 64 element hash table based on the Ethernet CRC.  */
-static int multicast_filter_limit = 32;
+   The RTL chips use a 64 element hash table based on the Ethernet CRC. */
+static const int multicast_filter_limit = 32;
 
-/* MAC address length*/
+/* MAC address length */
 #define MAC_ADDR_LEN   6
 
-/* max supported gigabit ethernet frame size -- must be at least (dev->mtu+14+4).*/
-#define MAX_ETH_FRAME_SIZE     1536
-
-#define TX_FIFO_THRESH 256     /* In bytes */
-
-#define RX_FIFO_THRESH 7       /* 7 means NO threshold, Rx buffer level before first PCI xfer.  */
+#define RX_FIFO_THRESH 7       /* 7 means NO threshold, Rx buffer level before first PCI xfer. */
 #define RX_DMA_BURST   6       /* Maximum PCI burst, '6' is 1024 */
 #define TX_DMA_BURST   6       /* Maximum PCI burst, '6' is 1024 */
 #define EarlyTxThld    0x3F    /* 0x3F means NO early transmit */
-#define RxPacketMaxSize        0x0800  /* Maximum size supported is 16K-1 */
+#define RxPacketMaxSize        0x3FE8  /* 16K - 1 - ETH_HLEN - VLAN - CRC... */
+#define SafeMtu                0x1c20  /* ... actually life sucks beyond ~7k */
 #define InterFrameGap  0x03    /* 3 means InterFrameGap = the shortest one */
 
+#define R8169_REGS_SIZE                256
+#define R8169_NAPI_WEIGHT      64
 #define NUM_TX_DESC    64      /* Number of Tx descriptor registers */
-#define NUM_RX_DESC    64      /* Number of Rx descriptor registers */
+#define NUM_RX_DESC    256     /* Number of Rx descriptor registers */
 #define RX_BUF_SIZE    1536    /* Rx Buffer size */
 #define R8169_TX_RING_BYTES    (NUM_TX_DESC * sizeof(struct TxDesc))
 #define R8169_RX_RING_BYTES    (NUM_RX_DESC * sizeof(struct RxDesc))
 
-#define RTL_MIN_IO_SIZE 0x80
 #define RTL8169_TX_TIMEOUT     (6*HZ)
-#define RTL8169_PHY_TIMEOUT    (HZ) 
+#define RTL8169_PHY_TIMEOUT    (10*HZ)
 
 /* write/read MMIO register */
 #define RTL_W8(reg, val8)      writeb ((val8), ioaddr + (reg))
@@ -109,10 +150,16 @@ static int multicast_filter_limit = 32;
 #define RTL_R32(reg)           ((unsigned long) readl (ioaddr + (reg)))
 
 enum mac_version {
-       RTL_GIGA_MAC_VER_B = 0x00,
-       /* RTL_GIGA_MAC_VER_C = 0x03, */
-       RTL_GIGA_MAC_VER_D = 0x01,
-       RTL_GIGA_MAC_VER_E = 0x02
+       RTL_GIGA_MAC_VER_01 = 0x00,
+       RTL_GIGA_MAC_VER_02 = 0x01,
+       RTL_GIGA_MAC_VER_03 = 0x02,
+       RTL_GIGA_MAC_VER_04 = 0x03,
+       RTL_GIGA_MAC_VER_05 = 0x04,
+       RTL_GIGA_MAC_VER_11 = 0x0b,
+       RTL_GIGA_MAC_VER_12 = 0x0c,
+       RTL_GIGA_MAC_VER_13 = 0x0d,
+       RTL_GIGA_MAC_VER_14 = 0x0e,
+       RTL_GIGA_MAC_VER_15 = 0x0f
 };
 
 enum phy_version {
@@ -121,35 +168,72 @@ enum phy_version {
        RTL_GIGA_PHY_VER_E = 0x05, /* PHY Reg 0x03 bit0-3 == 0x0000 */
        RTL_GIGA_PHY_VER_F = 0x06, /* PHY Reg 0x03 bit0-3 == 0x0001 */
        RTL_GIGA_PHY_VER_G = 0x07, /* PHY Reg 0x03 bit0-3 == 0x0002 */
+       RTL_GIGA_PHY_VER_H = 0x08, /* PHY Reg 0x03 bit0-3 == 0x0003 */
 };
 
-
 #define _R(NAME,MAC,MASK) \
        { .name = NAME, .mac_version = MAC, .RxConfigMask = MASK }
 
-const static struct {
+static const struct {
        const char *name;
        u8 mac_version;
        u32 RxConfigMask;       /* Clears the bits supported by this chip */
 } rtl_chip_info[] = {
-       _R("RTL8169",           RTL_GIGA_MAC_VER_B, 0xff7e1880),
-       _R("RTL8169s/8110s",    RTL_GIGA_MAC_VER_D, 0xff7e1880),
-       _R("RTL8169s/8110s",    RTL_GIGA_MAC_VER_E, 0xff7e1880)
+       _R("RTL8169",           RTL_GIGA_MAC_VER_01, 0xff7e1880),
+       _R("RTL8169s/8110s",    RTL_GIGA_MAC_VER_02, 0xff7e1880),
+       _R("RTL8169s/8110s",    RTL_GIGA_MAC_VER_03, 0xff7e1880),
+       _R("RTL8169sb/8110sb",  RTL_GIGA_MAC_VER_04, 0xff7e1880),
+       _R("RTL8169sc/8110sc",  RTL_GIGA_MAC_VER_05, 0xff7e1880),
+       _R("RTL8168b/8111b",    RTL_GIGA_MAC_VER_11, 0xff7e1880), // PCI-E
+       _R("RTL8168b/8111b",    RTL_GIGA_MAC_VER_12, 0xff7e1880), // PCI-E
+       _R("RTL8101e",          RTL_GIGA_MAC_VER_13, 0xff7e1880), // PCI-E 8139
+       _R("RTL8100e",          RTL_GIGA_MAC_VER_14, 0xff7e1880), // PCI-E 8139
+       _R("RTL8100e",          RTL_GIGA_MAC_VER_15, 0xff7e1880)  // PCI-E 8139
 };
 #undef _R
 
+enum cfg_version {
+       RTL_CFG_0 = 0x00,
+       RTL_CFG_1,
+       RTL_CFG_2
+};
+
+static const struct {
+       unsigned int region;
+       unsigned int align;
+} rtl_cfg_info[] = {
+       [RTL_CFG_0] = { 1, NET_IP_ALIGN },
+       [RTL_CFG_1] = { 2, NET_IP_ALIGN },
+       [RTL_CFG_2] = { 2, 8 }
+};
+
 static struct pci_device_id rtl8169_pci_tbl[] = {
-       {0x10ec, 0x8169, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
+       { PCI_DEVICE(PCI_VENDOR_ID_REALTEK,     0x8129), 0, 0, RTL_CFG_0 },
+       { PCI_DEVICE(PCI_VENDOR_ID_REALTEK,     0x8136), 0, 0, RTL_CFG_2 },
+       { PCI_DEVICE(PCI_VENDOR_ID_REALTEK,     0x8167), 0, 0, RTL_CFG_0 },
+       { PCI_DEVICE(PCI_VENDOR_ID_REALTEK,     0x8168), 0, 0, RTL_CFG_2 },
+       { PCI_DEVICE(PCI_VENDOR_ID_REALTEK,     0x8169), 0, 0, RTL_CFG_0 },
+       { PCI_DEVICE(PCI_VENDOR_ID_DLINK,       0x4300), 0, 0, RTL_CFG_0 },
+       { PCI_DEVICE(0x1259,                    0xc107), 0, 0, RTL_CFG_0 },
+       { PCI_DEVICE(0x16ec,                    0x0116), 0, 0, RTL_CFG_0 },
+       { PCI_VENDOR_ID_LINKSYS,                0x1032,
+               PCI_ANY_ID, 0x0024, 0, 0, RTL_CFG_0 },
        {0,},
 };
 
 MODULE_DEVICE_TABLE(pci, rtl8169_pci_tbl);
 
 static int rx_copybreak = 200;
+static int use_dac;
+static struct {
+       u32 msg_enable;
+} debug = { -1 };
 
 enum RTL8169_registers {
        MAC0 = 0,               /* Ethernet hardware address. */
        MAR0 = 8,               /* Multicast filter. */
+       CounterAddrLow = 0x10,
+       CounterAddrHigh = 0x14,
        TxDescStartAddrLow = 0x20,
        TxDescStartAddrHigh = 0x24,
        TxHDescStartAddrLow = 0x28,
@@ -178,6 +262,7 @@ enum RTL8169_registers {
        PHYstatus = 0x6C,
        RxMaxSize = 0xDA,
        CPlusCmd = 0xE0,
+       IntrMitigate = 0xE2,
        RxDescAddrLow = 0xE4,
        RxDescAddrHigh = 0xE8,
        EarlyTxThres = 0xEC,
@@ -188,36 +273,37 @@ enum RTL8169_registers {
 };
 
 enum RTL8169_register_content {
-       /*InterruptStatusBits */
+       /* InterruptStatusBits */
        SYSErr = 0x8000,
        PCSTimeout = 0x4000,
        SWInt = 0x0100,
        TxDescUnavail = 0x80,
        RxFIFOOver = 0x40,
-       RxUnderrun = 0x20,
+       LinkChg = 0x20,
        RxOverflow = 0x10,
        TxErr = 0x08,
        TxOK = 0x04,
        RxErr = 0x02,
        RxOK = 0x01,
 
-       /*RxStatusDesc */
-       RxRES = 0x00200000,
-       RxCRC = 0x00080000,
-       RxRUNT = 0x00100000,
-       RxRWT = 0x00400000,
+       /* RxStatusDesc */
+       RxFOVF  = (1 << 23),
+       RxRWT   = (1 << 22),
+       RxRES   = (1 << 21),
+       RxRUNT  = (1 << 20),
+       RxCRC   = (1 << 19),
 
-       /*ChipCmdBits */
+       /* ChipCmdBits */
        CmdReset = 0x10,
        CmdRxEnb = 0x08,
        CmdTxEnb = 0x04,
        RxBufEmpty = 0x01,
 
-       /*Cfg9346Bits */
+       /* Cfg9346Bits */
        Cfg9346_Lock = 0x00,
        Cfg9346_Unlock = 0xC0,
 
-       /*rx_mode_bits */
+       /* rx_mode_bits */
        AcceptErr = 0x20,
        AcceptRunt = 0x10,
        AcceptBroadcast = 0x08,
@@ -225,21 +311,43 @@ enum RTL8169_register_content {
        AcceptMyPhys = 0x02,
        AcceptAllPhys = 0x01,
 
-       /*RxConfigBits */
+       /* RxConfigBits */
        RxCfgFIFOShift = 13,
        RxCfgDMAShift = 8,
 
-       /*TxConfigBits */
+       /* TxConfigBits */
        TxInterFrameGapShift = 24,
        TxDMAShift = 8, /* DMA burst value (0-7) is shift this many bits */
 
+       /* Config1 register p.24 */
+       PMEnable        = (1 << 0),     /* Power Management Enable */
+
+       /* Config3 register p.25 */
+       MagicPacket     = (1 << 5),     /* Wake up when receives a Magic Packet */
+       LinkUp          = (1 << 4),     /* Wake up when the cable connection is re-established */
+
+       /* Config5 register p.27 */
+       BWF             = (1 << 6),     /* Accept Broadcast wakeup frame */
+       MWF             = (1 << 5),     /* Accept Multicast wakeup frame */
+       UWF             = (1 << 4),     /* Accept Unicast wakeup frame */
+       LanWake         = (1 << 1),     /* LanWake enable/disable */
+       PMEStatus       = (1 << 0),     /* PME status can be reset by PCI RST# */
+
+       /* TBICSR p.28 */
+       TBIReset        = 0x80000000,
+       TBILoopback     = 0x40000000,
+       TBINwEnable     = 0x20000000,
+       TBINwRestart    = 0x10000000,
+       TBILinkOk       = 0x02000000,
+       TBINwComplete   = 0x01000000,
+
        /* CPlusCmd p.31 */
        RxVlan          = (1 << 6),
        RxChkSum        = (1 << 5),
        PCIDAC          = (1 << 4),
        PCIMulRW        = (1 << 3),
 
-       /*rtl8169_PHYstatus */
+       /* rtl8169_PHYstatus */
        TBI_Enable = 0x80,
        TxFlowCtrl = 0x40,
        RxFlowCtrl = 0x20,
@@ -249,67 +357,77 @@ enum RTL8169_register_content {
        LinkStatus = 0x02,
        FullDup = 0x01,
 
-       /*GIGABIT_PHY_registers */
-       PHY_CTRL_REG = 0,
-       PHY_STAT_REG = 1,
-       PHY_AUTO_NEGO_REG = 4,
-       PHY_1000_CTRL_REG = 9,
-
-       /*GIGABIT_PHY_REG_BIT */
-       PHY_Restart_Auto_Nego = 0x0200,
-       PHY_Enable_Auto_Nego = 0x1000,
-
-       //PHY_STAT_REG = 1;
-       PHY_Auto_Neco_Comp = 0x0020,
-
-       //PHY_AUTO_NEGO_REG = 4;
-       PHY_Cap_10_Half = 0x0020,
-       PHY_Cap_10_Full = 0x0040,
-       PHY_Cap_100_Half = 0x0080,
-       PHY_Cap_100_Full = 0x0100,
-
-       //PHY_1000_CTRL_REG = 9;
-       PHY_Cap_1000_Full = 0x0200,
-
-       PHY_Cap_Null = 0x0,
-
-       /*_MediaType*/
+       /* _MediaType */
        _10_Half = 0x01,
        _10_Full = 0x02,
        _100_Half = 0x04,
        _100_Full = 0x08,
        _1000_Full = 0x10,
 
-       /*_TBICSRBit*/
+       /* _TBICSRBit */
        TBILinkOK = 0x02000000,
+
+       /* DumpCounterCommand */
+       CounterDump = 0x8,
 };
 
 enum _DescStatusBit {
-       OWNbit = 0x80000000,
-       EORbit = 0x40000000,
-       FSbit = 0x20000000,
-       LSbit = 0x10000000,
+       DescOwn         = (1 << 31), /* Descriptor is owned by NIC */
+       RingEnd         = (1 << 30), /* End of descriptor ring */
+       FirstFrag       = (1 << 29), /* First segment of a packet */
+       LastFrag        = (1 << 28), /* Final segment of a packet */
+
+       /* Tx private */
+       LargeSend       = (1 << 27), /* TCP Large Send Offload (TSO) */
+       MSSShift        = 16,        /* MSS value position */
+       MSSMask         = 0xfff,     /* MSS value + LargeSend bit: 12 bits */
+       IPCS            = (1 << 18), /* Calculate IP checksum */
+       UDPCS           = (1 << 17), /* Calculate UDP/IP checksum */
+       TCPCS           = (1 << 16), /* Calculate TCP/IP checksum */
+       TxVlanTag       = (1 << 17), /* Add VLAN tag */
+
+       /* Rx private */
+       PID1            = (1 << 18), /* Protocol ID bit 1/2 */
+       PID0            = (1 << 17), /* Protocol ID bit 2/2 */
+
+#define RxProtoUDP     (PID1)
+#define RxProtoTCP     (PID0)
+#define RxProtoIP      (PID1 | PID0)
+#define RxProtoMask    RxProtoIP
+
+       IPFail          = (1 << 16), /* IP checksum failed */
+       UDPFail         = (1 << 15), /* UDP/IP checksum failed */
+       TCPFail         = (1 << 14), /* TCP/IP checksum failed */
+       RxVlanTag       = (1 << 16), /* VLAN tag available */
 };
 
 #define RsvdMask       0x3fffc000
 
 struct TxDesc {
-       u32 status;
-       u32 vlan_tag;
+       u32 opts1;
+       u32 opts2;
        u64 addr;
 };
 
 struct RxDesc {
-       u32 status;
-       u32 vlan_tag;
+       u32 opts1;
+       u32 opts2;
        u64 addr;
 };
 
+struct ring_info {
+       struct sk_buff  *skb;
+       u32             len;
+       u8              __pad[sizeof(void *) - sizeof(u32)];
+};
+
 struct rtl8169_private {
-       void *mmio_addr;        /* memory map physical address */
-       struct pci_dev *pci_dev;        /* Index of PCI device  */
+       void __iomem *mmio_addr;        /* memory map physical address */
+       struct pci_dev *pci_dev;        /* Index of PCI device */
+       struct net_device *dev;
        struct net_device_stats stats;  /* statistics of net device */
-       spinlock_t lock;        /* spin lock flag */
+       spinlock_t lock;                /* spin lock flag */
+       u32 msg_enable;
        int chipset;
        int mac_version;
        int phy_version;
@@ -317,94 +435,722 @@ struct rtl8169_private {
        u32 cur_tx; /* Index into the Tx descriptor buffer of next Rx pkt. */
        u32 dirty_rx;
        u32 dirty_tx;
-       struct TxDesc *TxDescArray;     /* Index of 256-alignment Tx Descriptor buffer */
-       struct RxDesc *RxDescArray;     /* Index of 256-alignment Rx Descriptor buffer */
+       struct TxDesc *TxDescArray;     /* 256-aligned Tx descriptor ring */
+       struct RxDesc *RxDescArray;     /* 256-aligned Rx descriptor ring */
        dma_addr_t TxPhyAddr;
        dma_addr_t RxPhyAddr;
        struct sk_buff *Rx_skbuff[NUM_RX_DESC]; /* Rx data buffers */
-       struct sk_buff *Tx_skbuff[NUM_TX_DESC]; /* Index of Transmit data buffer */
+       struct ring_info tx_skb[NUM_TX_DESC];   /* Tx data buffers */
+       unsigned align;
+       unsigned rx_buf_sz;
        struct timer_list timer;
-       unsigned long phy_link_down_cnt;
        u16 cp_cmd;
+       u16 intr_mask;
+       int phy_auto_nego_reg;
+       int phy_1000_ctrl_reg;
+#ifdef CONFIG_R8169_VLAN
+       struct vlan_group *vlgrp;
+#endif
+       int (*set_speed)(struct net_device *, u8 autoneg, u16 speed, u8 duplex);
+       void (*get_settings)(struct net_device *, struct ethtool_cmd *);
+       void (*phy_reset_enable)(void __iomem *);
+       unsigned int (*phy_reset_pending)(void __iomem *);
+       unsigned int (*link_ok)(void __iomem *);
+       struct delayed_work task;
+       unsigned wol_enabled : 1;
 };
 
-MODULE_AUTHOR("Realtek");
+MODULE_AUTHOR("Realtek and the Linux r8169 crew <netdev@vger.kernel.org>");
 MODULE_DESCRIPTION("RealTek RTL-8169 Gigabit Ethernet driver");
-MODULE_PARM(media, "1-" __MODULE_STRING(MAX_UNITS) "i");
-MODULE_PARM(rx_copybreak, "i");
+module_param_array(media, int, &num_media, 0);
+MODULE_PARM_DESC(media, "force phy operation. Deprecated by ethtool (8).");
+module_param(rx_copybreak, int, 0);
+MODULE_PARM_DESC(rx_copybreak, "Copy breakpoint for copy-only-tiny-frames");
+module_param(use_dac, int, 0);
+MODULE_PARM_DESC(use_dac, "Enable PCI DAC. Unsafe on 32 bit PCI slot.");
+module_param_named(debug, debug.msg_enable, int, 0);
+MODULE_PARM_DESC(debug, "Debug verbosity level (0=none, ..., 16=all)");
 MODULE_LICENSE("GPL");
+MODULE_VERSION(RTL8169_VERSION);
 
 static int rtl8169_open(struct net_device *dev);
 static int rtl8169_start_xmit(struct sk_buff *skb, struct net_device *dev);
-static irqreturn_t rtl8169_interrupt(int irq, void *dev_instance,
-                             struct pt_regs *regs);
+static irqreturn_t rtl8169_interrupt(int irq, void *dev_instance);
 static int rtl8169_init_ring(struct net_device *dev);
 static void rtl8169_hw_start(struct net_device *dev);
 static int rtl8169_close(struct net_device *dev);
 static void rtl8169_set_rx_mode(struct net_device *dev);
 static void rtl8169_tx_timeout(struct net_device *dev);
-static struct net_device_stats *rtl8169_get_stats(struct net_device *netdev);
+static struct net_device_stats *rtl8169_get_stats(struct net_device *dev);
+static int rtl8169_rx_interrupt(struct net_device *, struct rtl8169_private *,
+                               void __iomem *);
+static int rtl8169_change_mtu(struct net_device *dev, int new_mtu);
+static void rtl8169_down(struct net_device *dev);
+
+#ifdef CONFIG_R8169_NAPI
+static int rtl8169_poll(struct net_device *dev, int *budget);
+#endif
 
 static const u16 rtl8169_intr_mask =
-    RxUnderrun | RxOverflow | RxFIFOOver | TxErr | TxOK | RxErr | RxOK;
+       SYSErr | LinkChg | RxOverflow | RxFIFOOver | TxErr | TxOK | RxErr | RxOK;
+static const u16 rtl8169_napi_event =
+       RxOK | RxOverflow | RxFIFOOver | TxOK | TxErr;
 static const unsigned int rtl8169_rx_config =
-    (RX_FIFO_THRESH << RxCfgFIFOShift) | (RX_DMA_BURST << RxCfgDMAShift);
+       (RX_FIFO_THRESH << RxCfgFIFOShift) | (RX_DMA_BURST << RxCfgDMAShift);
 
-#define PHY_Cap_10_Half_Or_Less PHY_Cap_10_Half
-#define PHY_Cap_10_Full_Or_Less PHY_Cap_10_Full | PHY_Cap_10_Half_Or_Less
-#define PHY_Cap_100_Half_Or_Less PHY_Cap_100_Half | PHY_Cap_10_Full_Or_Less
-#define PHY_Cap_100_Full_Or_Less PHY_Cap_100_Full | PHY_Cap_100_Half_Or_Less
-
-static void mdio_write(void *ioaddr, int RegAddr, int value)
+static void mdio_write(void __iomem *ioaddr, int RegAddr, int value)
 {
        int i;
 
        RTL_W32(PHYAR, 0x80000000 | (RegAddr & 0xFF) << 16 | value);
-       udelay(1000);
 
-       for (i = 2000; i > 0; i--) {
-               // Check if the RTL8169 has completed writing to the specified MII register
-               if (!(RTL_R32(PHYAR) & 0x80000000)) {
+       for (i = 20; i > 0; i--) {
+               /* Check if the RTL8169 has completed writing to the specified MII register */
+               if (!(RTL_R32(PHYAR) & 0x80000000))
                        break;
-               } else {
-                       udelay(100);
-               }
+               udelay(25);
        }
 }
 
-static int mdio_read(void *ioaddr, int RegAddr)
+static int mdio_read(void __iomem *ioaddr, int RegAddr)
 {
        int i, value = -1;
 
        RTL_W32(PHYAR, 0x0 | (RegAddr & 0xFF) << 16);
-       udelay(1000);
 
-       for (i = 2000; i > 0; i--) {
-               // Check if the RTL8169 has completed retrieving data from the specified MII register
+       for (i = 20; i > 0; i--) {
+               /* Check if the RTL8169 has completed retrieving data from the specified MII register */
                if (RTL_R32(PHYAR) & 0x80000000) {
                        value = (int) (RTL_R32(PHYAR) & 0xFFFF);
                        break;
-               }
-               udelay(100);
+               }
+               udelay(25);
+       }
+       return value;
+}
+
+static void rtl8169_irq_mask_and_ack(void __iomem *ioaddr)
+{
+       RTL_W16(IntrMask, 0x0000);
+
+       RTL_W16(IntrStatus, 0xffff);
+}
+
+static void rtl8169_asic_down(void __iomem *ioaddr)
+{
+       RTL_W8(ChipCmd, 0x00);
+       rtl8169_irq_mask_and_ack(ioaddr);
+       RTL_R16(CPlusCmd);
+}
+
+static unsigned int rtl8169_tbi_reset_pending(void __iomem *ioaddr)
+{
+       return RTL_R32(TBICSR) & TBIReset;
+}
+
+static unsigned int rtl8169_xmii_reset_pending(void __iomem *ioaddr)
+{
+       return mdio_read(ioaddr, MII_BMCR) & BMCR_RESET;
+}
+
+static unsigned int rtl8169_tbi_link_ok(void __iomem *ioaddr)
+{
+       return RTL_R32(TBICSR) & TBILinkOk;
+}
+
+static unsigned int rtl8169_xmii_link_ok(void __iomem *ioaddr)
+{
+       return RTL_R8(PHYstatus) & LinkStatus;
+}
+
+static void rtl8169_tbi_reset_enable(void __iomem *ioaddr)
+{
+       RTL_W32(TBICSR, RTL_R32(TBICSR) | TBIReset);
+}
+
+static void rtl8169_xmii_reset_enable(void __iomem *ioaddr)
+{
+       unsigned int val;
+
+       mdio_write(ioaddr, MII_BMCR, BMCR_RESET);
+       val = mdio_read(ioaddr, MII_BMCR);
+}
+
+static void rtl8169_check_link_status(struct net_device *dev,
+                                     struct rtl8169_private *tp, void __iomem *ioaddr)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&tp->lock, flags);
+       if (tp->link_ok(ioaddr)) {
+               netif_carrier_on(dev);
+               if (netif_msg_ifup(tp))
+                       printk(KERN_INFO PFX "%s: link up\n", dev->name);
+       } else {
+               if (netif_msg_ifdown(tp))
+                       printk(KERN_INFO PFX "%s: link down\n", dev->name);
+               netif_carrier_off(dev);
+       }
+       spin_unlock_irqrestore(&tp->lock, flags);
+}
+
+static void rtl8169_link_option(int idx, u8 *autoneg, u16 *speed, u8 *duplex)
+{
+       struct {
+               u16 speed;
+               u8 duplex;
+               u8 autoneg;
+               u8 media;
+       } link_settings[] = {
+               { SPEED_10,     DUPLEX_HALF, AUTONEG_DISABLE,   _10_Half },
+               { SPEED_10,     DUPLEX_FULL, AUTONEG_DISABLE,   _10_Full },
+               { SPEED_100,    DUPLEX_HALF, AUTONEG_DISABLE,   _100_Half },
+               { SPEED_100,    DUPLEX_FULL, AUTONEG_DISABLE,   _100_Full },
+               { SPEED_1000,   DUPLEX_FULL, AUTONEG_DISABLE,   _1000_Full },
+               /* Make TBI happy */
+               { SPEED_1000,   DUPLEX_FULL, AUTONEG_ENABLE,    0xff }
+       }, *p;
+       unsigned char option;
+
+       option = ((idx < MAX_UNITS) && (idx >= 0)) ? media[idx] : 0xff;
+
+       if ((option != 0xff) && !idx && netif_msg_drv(&debug))
+               printk(KERN_WARNING PFX "media option is deprecated.\n");
+
+       for (p = link_settings; p->media != 0xff; p++) {
+               if (p->media == option)
+                       break;
+       }
+       *autoneg = p->autoneg;
+       *speed = p->speed;
+       *duplex = p->duplex;
+}
+
+static void rtl8169_get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       u8 options;
+
+       wol->wolopts = 0;
+
+#define WAKE_ANY (WAKE_PHY | WAKE_MAGIC | WAKE_UCAST | WAKE_BCAST | WAKE_MCAST)
+       wol->supported = WAKE_ANY;
+
+       spin_lock_irq(&tp->lock);
+
+       options = RTL_R8(Config1);
+       if (!(options & PMEnable))
+               goto out_unlock;
+
+       options = RTL_R8(Config3);
+       if (options & LinkUp)
+               wol->wolopts |= WAKE_PHY;
+       if (options & MagicPacket)
+               wol->wolopts |= WAKE_MAGIC;
+
+       options = RTL_R8(Config5);
+       if (options & UWF)
+               wol->wolopts |= WAKE_UCAST;
+       if (options & BWF)
+               wol->wolopts |= WAKE_BCAST;
+       if (options & MWF)
+               wol->wolopts |= WAKE_MCAST;
+
+out_unlock:
+       spin_unlock_irq(&tp->lock);
+}
+
+static int rtl8169_set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       int i;
+       static struct {
+               u32 opt;
+               u16 reg;
+               u8  mask;
+       } cfg[] = {
+               { WAKE_ANY,   Config1, PMEnable },
+               { WAKE_PHY,   Config3, LinkUp },
+               { WAKE_MAGIC, Config3, MagicPacket },
+               { WAKE_UCAST, Config5, UWF },
+               { WAKE_BCAST, Config5, BWF },
+               { WAKE_MCAST, Config5, MWF },
+               { WAKE_ANY,   Config5, LanWake }
+       };
+
+       spin_lock_irq(&tp->lock);
+
+       RTL_W8(Cfg9346, Cfg9346_Unlock);
+
+       for (i = 0; i < ARRAY_SIZE(cfg); i++) {
+               u8 options = RTL_R8(cfg[i].reg) & ~cfg[i].mask;
+               if (wol->wolopts & cfg[i].opt)
+                       options |= cfg[i].mask;
+               RTL_W8(cfg[i].reg, options);
+       }
+
+       RTL_W8(Cfg9346, Cfg9346_Lock);
+
+       tp->wol_enabled = (wol->wolopts) ? 1 : 0;
+
+       spin_unlock_irq(&tp->lock);
+
+       return 0;
+}
+
+static void rtl8169_get_drvinfo(struct net_device *dev,
+                               struct ethtool_drvinfo *info)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+
+       strcpy(info->driver, MODULENAME);
+       strcpy(info->version, RTL8169_VERSION);
+       strcpy(info->bus_info, pci_name(tp->pci_dev));
+}
+
+static int rtl8169_get_regs_len(struct net_device *dev)
+{
+       return R8169_REGS_SIZE;
+}
+
+static int rtl8169_set_speed_tbi(struct net_device *dev,
+                                u8 autoneg, u16 speed, u8 duplex)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       int ret = 0;
+       u32 reg;
+
+       reg = RTL_R32(TBICSR);
+       if ((autoneg == AUTONEG_DISABLE) && (speed == SPEED_1000) &&
+           (duplex == DUPLEX_FULL)) {
+               RTL_W32(TBICSR, reg & ~(TBINwEnable | TBINwRestart));
+       } else if (autoneg == AUTONEG_ENABLE)
+               RTL_W32(TBICSR, reg | TBINwEnable | TBINwRestart);
+       else {
+               if (netif_msg_link(tp)) {
+                       printk(KERN_WARNING "%s: "
+                              "incorrect speed setting refused in TBI mode\n",
+                              dev->name);
+               }
+               ret = -EOPNOTSUPP;
+       }
+
+       return ret;
+}
+
+static int rtl8169_set_speed_xmii(struct net_device *dev,
+                                 u8 autoneg, u16 speed, u8 duplex)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       int auto_nego, giga_ctrl;
+
+       auto_nego = mdio_read(ioaddr, MII_ADVERTISE);
+       auto_nego &= ~(ADVERTISE_10HALF | ADVERTISE_10FULL |
+                      ADVERTISE_100HALF | ADVERTISE_100FULL);
+       giga_ctrl = mdio_read(ioaddr, MII_CTRL1000);
+       giga_ctrl &= ~(ADVERTISE_1000FULL | ADVERTISE_1000HALF);
+
+       if (autoneg == AUTONEG_ENABLE) {
+               auto_nego |= (ADVERTISE_10HALF | ADVERTISE_10FULL |
+                             ADVERTISE_100HALF | ADVERTISE_100FULL);
+               giga_ctrl |= ADVERTISE_1000FULL | ADVERTISE_1000HALF;
+       } else {
+               if (speed == SPEED_10)
+                       auto_nego |= ADVERTISE_10HALF | ADVERTISE_10FULL;
+               else if (speed == SPEED_100)
+                       auto_nego |= ADVERTISE_100HALF | ADVERTISE_100FULL;
+               else if (speed == SPEED_1000)
+                       giga_ctrl |= ADVERTISE_1000FULL | ADVERTISE_1000HALF;
+
+               if (duplex == DUPLEX_HALF)
+                       auto_nego &= ~(ADVERTISE_10FULL | ADVERTISE_100FULL);
+
+               if (duplex == DUPLEX_FULL)
+                       auto_nego &= ~(ADVERTISE_10HALF | ADVERTISE_100HALF);
+
+               /* This tweak comes straight from Realtek's driver. */
+               if ((speed == SPEED_100) && (duplex == DUPLEX_HALF) &&
+                   (tp->mac_version == RTL_GIGA_MAC_VER_13)) {
+                       auto_nego = ADVERTISE_100HALF | ADVERTISE_CSMA;
+               }
+       }
+
+       /* The 8100e/8101e do Fast Ethernet only. */
+       if ((tp->mac_version == RTL_GIGA_MAC_VER_13) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_14) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_15)) {
+               if ((giga_ctrl & (ADVERTISE_1000FULL | ADVERTISE_1000HALF)) &&
+                   netif_msg_link(tp)) {
+                       printk(KERN_INFO "%s: PHY does not support 1000Mbps.\n",
+                              dev->name);
+               }
+               giga_ctrl &= ~(ADVERTISE_1000FULL | ADVERTISE_1000HALF);
+       }
+
+       auto_nego |= ADVERTISE_PAUSE_CAP | ADVERTISE_PAUSE_ASYM;
+
+       tp->phy_auto_nego_reg = auto_nego;
+       tp->phy_1000_ctrl_reg = giga_ctrl;
+
+       mdio_write(ioaddr, MII_ADVERTISE, auto_nego);
+       mdio_write(ioaddr, MII_CTRL1000, giga_ctrl);
+       mdio_write(ioaddr, MII_BMCR, BMCR_ANENABLE | BMCR_ANRESTART);
+       return 0;
+}
+
+static int rtl8169_set_speed(struct net_device *dev,
+                            u8 autoneg, u16 speed, u8 duplex)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       int ret;
+
+       ret = tp->set_speed(dev, autoneg, speed, duplex);
+
+       if (netif_running(dev) && (tp->phy_1000_ctrl_reg & ADVERTISE_1000FULL))
+               mod_timer(&tp->timer, jiffies + RTL8169_PHY_TIMEOUT);
+
+       return ret;
+}
+
+static int rtl8169_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       unsigned long flags;
+       int ret;
+
+       spin_lock_irqsave(&tp->lock, flags);
+       ret = rtl8169_set_speed(dev, cmd->autoneg, cmd->speed, cmd->duplex);
+       spin_unlock_irqrestore(&tp->lock, flags);
+
+       return ret;
+}
+
+static u32 rtl8169_get_rx_csum(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+
+       return tp->cp_cmd & RxChkSum;
+}
+
+static int rtl8169_set_rx_csum(struct net_device *dev, u32 data)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tp->lock, flags);
+
+       if (data)
+               tp->cp_cmd |= RxChkSum;
+       else
+               tp->cp_cmd &= ~RxChkSum;
+
+       RTL_W16(CPlusCmd, tp->cp_cmd);
+       RTL_R16(CPlusCmd);
+
+       spin_unlock_irqrestore(&tp->lock, flags);
+
+       return 0;
+}
+
+#ifdef CONFIG_R8169_VLAN
+
+static inline u32 rtl8169_tx_vlan_tag(struct rtl8169_private *tp,
+                                     struct sk_buff *skb)
+{
+       return (tp->vlgrp && vlan_tx_tag_present(skb)) ?
+               TxVlanTag | swab16(vlan_tx_tag_get(skb)) : 0x00;
+}
+
+static void rtl8169_vlan_rx_register(struct net_device *dev,
+                                    struct vlan_group *grp)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned long flags;
+
+       spin_lock_irqsave(&tp->lock, flags);
+       tp->vlgrp = grp;
+       if (tp->vlgrp)
+               tp->cp_cmd |= RxVlan;
+       else
+               tp->cp_cmd &= ~RxVlan;
+       RTL_W16(CPlusCmd, tp->cp_cmd);
+       RTL_R16(CPlusCmd);
+       spin_unlock_irqrestore(&tp->lock, flags);
+}
+
+static void rtl8169_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       unsigned long flags;
+
+       spin_lock_irqsave(&tp->lock, flags);
+       if (tp->vlgrp)
+               tp->vlgrp->vlan_devices[vid] = NULL;
+       spin_unlock_irqrestore(&tp->lock, flags);
+}
+
+static int rtl8169_rx_vlan_skb(struct rtl8169_private *tp, struct RxDesc *desc,
+                              struct sk_buff *skb)
+{
+       u32 opts2 = le32_to_cpu(desc->opts2);
+       int ret;
+
+       if (tp->vlgrp && (opts2 & RxVlanTag)) {
+               rtl8169_rx_hwaccel_skb(skb, tp->vlgrp,
+                                      swab16(opts2 & 0xffff));
+               ret = 0;
+       } else
+               ret = -1;
+       desc->opts2 = 0;
+       return ret;
+}
+
+#else /* !CONFIG_R8169_VLAN */
+
+static inline u32 rtl8169_tx_vlan_tag(struct rtl8169_private *tp,
+                                     struct sk_buff *skb)
+{
+       return 0;
+}
+
+static int rtl8169_rx_vlan_skb(struct rtl8169_private *tp, struct RxDesc *desc,
+                              struct sk_buff *skb)
+{
+       return -1;
+}
+
+#endif
+
+static void rtl8169_gset_tbi(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       u32 status;
+
+       cmd->supported =
+               SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE;
+       cmd->port = PORT_FIBRE;
+       cmd->transceiver = XCVR_INTERNAL;
+
+       status = RTL_R32(TBICSR);
+       cmd->advertising = (status & TBINwEnable) ?  ADVERTISED_Autoneg : 0;
+       cmd->autoneg = !!(status & TBINwEnable);
+
+       cmd->speed = SPEED_1000;
+       cmd->duplex = DUPLEX_FULL; /* Always set */
+}
+
+static void rtl8169_gset_xmii(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       u8 status;
+
+       cmd->supported = SUPPORTED_10baseT_Half |
+                        SUPPORTED_10baseT_Full |
+                        SUPPORTED_100baseT_Half |
+                        SUPPORTED_100baseT_Full |
+                        SUPPORTED_1000baseT_Full |
+                        SUPPORTED_Autoneg |
+                        SUPPORTED_TP;
+
+       cmd->autoneg = 1;
+       cmd->advertising = ADVERTISED_TP | ADVERTISED_Autoneg;
+
+       if (tp->phy_auto_nego_reg & ADVERTISE_10HALF)
+               cmd->advertising |= ADVERTISED_10baseT_Half;
+       if (tp->phy_auto_nego_reg & ADVERTISE_10FULL)
+               cmd->advertising |= ADVERTISED_10baseT_Full;
+       if (tp->phy_auto_nego_reg & ADVERTISE_100HALF)
+               cmd->advertising |= ADVERTISED_100baseT_Half;
+       if (tp->phy_auto_nego_reg & ADVERTISE_100FULL)
+               cmd->advertising |= ADVERTISED_100baseT_Full;
+       if (tp->phy_1000_ctrl_reg & ADVERTISE_1000FULL)
+               cmd->advertising |= ADVERTISED_1000baseT_Full;
+
+       status = RTL_R8(PHYstatus);
+
+       if (status & _1000bpsF)
+               cmd->speed = SPEED_1000;
+       else if (status & _100bps)
+               cmd->speed = SPEED_100;
+       else if (status & _10bps)
+               cmd->speed = SPEED_10;
+
+       if (status & TxFlowCtrl)
+               cmd->advertising |= ADVERTISED_Asym_Pause;
+       if (status & RxFlowCtrl)
+               cmd->advertising |= ADVERTISED_Pause;
+
+       cmd->duplex = ((status & _1000bpsF) || (status & FullDup)) ?
+                     DUPLEX_FULL : DUPLEX_HALF;
+}
+
+static int rtl8169_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       unsigned long flags;
+
+       spin_lock_irqsave(&tp->lock, flags);
+
+       tp->get_settings(dev, cmd);
+
+       spin_unlock_irqrestore(&tp->lock, flags);
+       return 0;
+}
+
+static void rtl8169_get_regs(struct net_device *dev, struct ethtool_regs *regs,
+                            void *p)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       unsigned long flags;
+
+       if (regs->len > R8169_REGS_SIZE)
+               regs->len = R8169_REGS_SIZE;
+
+       spin_lock_irqsave(&tp->lock, flags);
+       memcpy_fromio(p, tp->mmio_addr, regs->len);
+       spin_unlock_irqrestore(&tp->lock, flags);
+}
+
+static u32 rtl8169_get_msglevel(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+
+       return tp->msg_enable;
+}
+
+static void rtl8169_set_msglevel(struct net_device *dev, u32 value)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+
+       tp->msg_enable = value;
+}
+
+static const char rtl8169_gstrings[][ETH_GSTRING_LEN] = {
+       "tx_packets",
+       "rx_packets",
+       "tx_errors",
+       "rx_errors",
+       "rx_missed",
+       "align_errors",
+       "tx_single_collisions",
+       "tx_multi_collisions",
+       "unicast",
+       "broadcast",
+       "multicast",
+       "tx_aborted",
+       "tx_underrun",
+};
+
+struct rtl8169_counters {
+       u64     tx_packets;
+       u64     rx_packets;
+       u64     tx_errors;
+       u32     rx_errors;
+       u16     rx_missed;
+       u16     align_errors;
+       u32     tx_one_collision;
+       u32     tx_multi_collision;
+       u64     rx_unicast;
+       u64     rx_broadcast;
+       u32     rx_multicast;
+       u16     tx_aborted;
+       u16     tx_underun;
+};
+
+static int rtl8169_get_stats_count(struct net_device *dev)
+{
+       return ARRAY_SIZE(rtl8169_gstrings);
+}
+
+static void rtl8169_get_ethtool_stats(struct net_device *dev,
+                                     struct ethtool_stats *stats, u64 *data)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       struct rtl8169_counters *counters;
+       dma_addr_t paddr;
+       u32 cmd;
+
+       ASSERT_RTNL();
+
+       counters = pci_alloc_consistent(tp->pci_dev, sizeof(*counters), &paddr);
+       if (!counters)
+               return;
+
+       RTL_W32(CounterAddrHigh, (u64)paddr >> 32);
+       cmd = (u64)paddr & DMA_32BIT_MASK;
+       RTL_W32(CounterAddrLow, cmd);
+       RTL_W32(CounterAddrLow, cmd | CounterDump);
+
+       while (RTL_R32(CounterAddrLow) & CounterDump) {
+               if (msleep_interruptible(1))
+                       break;
        }
-       return value;
+
+       RTL_W32(CounterAddrLow, 0);
+       RTL_W32(CounterAddrHigh, 0);
+
+       data[0] = le64_to_cpu(counters->tx_packets);
+       data[1] = le64_to_cpu(counters->rx_packets);
+       data[2] = le64_to_cpu(counters->tx_errors);
+       data[3] = le32_to_cpu(counters->rx_errors);
+       data[4] = le16_to_cpu(counters->rx_missed);
+       data[5] = le16_to_cpu(counters->align_errors);
+       data[6] = le32_to_cpu(counters->tx_one_collision);
+       data[7] = le32_to_cpu(counters->tx_multi_collision);
+       data[8] = le64_to_cpu(counters->rx_unicast);
+       data[9] = le64_to_cpu(counters->rx_broadcast);
+       data[10] = le32_to_cpu(counters->rx_multicast);
+       data[11] = le16_to_cpu(counters->tx_aborted);
+       data[12] = le16_to_cpu(counters->tx_underun);
+
+       pci_free_consistent(tp->pci_dev, sizeof(*counters), counters, paddr);
 }
 
-static void rtl8169_get_drvinfo(struct net_device *dev,
-                               struct ethtool_drvinfo *info)
+static void rtl8169_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 {
-       struct rtl8169_private *tp = dev->priv;
-
-       strcpy(info->driver, RTL8169_DRIVER_NAME);
-       strcpy(info->version, RTL8169_VERSION );
-       strcpy(info->bus_info, pci_name(tp->pci_dev));
+       switch(stringset) {
+       case ETH_SS_STATS:
+               memcpy(data, *rtl8169_gstrings, sizeof(rtl8169_gstrings));
+               break;
+       }
 }
 
-static struct ethtool_ops rtl8169_ethtool_ops = {
+
+static const struct ethtool_ops rtl8169_ethtool_ops = {
        .get_drvinfo            = rtl8169_get_drvinfo,
+       .get_regs_len           = rtl8169_get_regs_len,
+       .get_link               = ethtool_op_get_link,
+       .get_settings           = rtl8169_get_settings,
+       .set_settings           = rtl8169_set_settings,
+       .get_msglevel           = rtl8169_get_msglevel,
+       .set_msglevel           = rtl8169_set_msglevel,
+       .get_rx_csum            = rtl8169_get_rx_csum,
+       .set_rx_csum            = rtl8169_set_rx_csum,
+       .get_tx_csum            = ethtool_op_get_tx_csum,
+       .set_tx_csum            = ethtool_op_set_tx_csum,
+       .get_sg                 = ethtool_op_get_sg,
+       .set_sg                 = ethtool_op_set_sg,
+       .get_tso                = ethtool_op_get_tso,
+       .set_tso                = ethtool_op_set_tso,
+       .get_regs               = rtl8169_get_regs,
+       .get_wol                = rtl8169_get_wol,
+       .set_wol                = rtl8169_set_wol,
+       .get_strings            = rtl8169_get_strings,
+       .get_stats_count        = rtl8169_get_stats_count,
+       .get_ethtool_stats      = rtl8169_get_ethtool_stats,
+       .get_perm_addr          = ethtool_op_get_perm_addr,
 };
 
-static void rtl8169_write_gmii_reg_bit(void *ioaddr, int reg, int bitnum,
+static void rtl8169_write_gmii_reg_bit(void __iomem *ioaddr, int reg, int bitnum,
                                       int bitval)
 {
        int val;
@@ -412,18 +1158,25 @@ static void rtl8169_write_gmii_reg_bit(void *ioaddr, int reg, int bitnum,
        val = mdio_read(ioaddr, reg);
        val = (bitval == 1) ?
                val | (bitval << bitnum) :  val & ~(0x0001 << bitnum);
-       mdio_write(ioaddr, reg, val & 0xffff); 
+       mdio_write(ioaddr, reg, val & 0xffff);
 }
 
-static void rtl8169_get_mac_version(struct rtl8169_private *tp, void *ioaddr)
+static void rtl8169_get_mac_version(struct rtl8169_private *tp, void __iomem *ioaddr)
 {
        const struct {
                u32 mask;
                int mac_version;
        } mac_info[] = {
-               { 0x1 << 26,    RTL_GIGA_MAC_VER_E },
-               { 0x1 << 23,    RTL_GIGA_MAC_VER_D }, 
-               { 0x00000000,   RTL_GIGA_MAC_VER_B } /* Catch-all */
+               { 0x38800000,   RTL_GIGA_MAC_VER_15 },
+               { 0x38000000,   RTL_GIGA_MAC_VER_12 },
+               { 0x34000000,   RTL_GIGA_MAC_VER_13 },
+               { 0x30800000,   RTL_GIGA_MAC_VER_14 },
+               { 0x30000000,   RTL_GIGA_MAC_VER_11 },
+               { 0x18000000,   RTL_GIGA_MAC_VER_05 },
+               { 0x10000000,   RTL_GIGA_MAC_VER_04 },
+               { 0x04000000,   RTL_GIGA_MAC_VER_03 },
+               { 0x00800000,   RTL_GIGA_MAC_VER_02 },
+               { 0x00000000,   RTL_GIGA_MAC_VER_01 }   /* Catch-all */
        }, *p = mac_info;
        u32 reg;
 
@@ -435,27 +1188,10 @@ static void rtl8169_get_mac_version(struct rtl8169_private *tp, void *ioaddr)
 
 static void rtl8169_print_mac_version(struct rtl8169_private *tp)
 {
-       struct {
-               int version;
-               char *msg;
-       } mac_print[] = {
-               { RTL_GIGA_MAC_VER_E, "RTL_GIGA_MAC_VER_E" },
-               { RTL_GIGA_MAC_VER_D, "RTL_GIGA_MAC_VER_D" },
-               { RTL_GIGA_MAC_VER_B, "RTL_GIGA_MAC_VER_B" },
-               { 0, NULL }
-       }, *p;
-
-       for (p = mac_print; p->msg; p++) {
-               if (tp->mac_version == p->version) {
-                       dprintk("mac_version == %s (%04d)\n", p->msg,
-                                 p->version);
-                       return;
-               }
-       }
-       dprintk("mac_version == Unknown\n");
+       dprintk("mac_version = 0x%02x\n", tp->mac_version);
 }
 
-static void rtl8169_get_phy_version(struct rtl8169_private *tp, void *ioaddr)
+static void rtl8169_get_phy_version(struct rtl8169_private *tp, void __iomem *ioaddr)
 {
        const struct {
                u16 mask;
@@ -469,7 +1205,7 @@ static void rtl8169_get_phy_version(struct rtl8169_private *tp, void *ioaddr)
        }, *p = phy_info;
        u16 reg;
 
-       reg = mdio_read(ioaddr, 3) & 0xffff;
+       reg = mdio_read(ioaddr, MII_PHYSID2) & 0xffff;
        while ((reg & p->mask) != p->set)
                p++;
        tp->phy_version = p->phy_version;
@@ -500,8 +1236,8 @@ static void rtl8169_print_phy_version(struct rtl8169_private *tp)
 
 static void rtl8169_hw_phy_config(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
        struct {
                u16 regs[5]; /* Beware of bit-sign propagation */
        } phy_magic[5] = { {
@@ -537,9 +1273,9 @@ static void rtl8169_hw_phy_config(struct net_device *dev)
        rtl8169_print_mac_version(tp);
        rtl8169_print_phy_version(tp);
 
-       if (tp->mac_version <= RTL_GIGA_MAC_VER_B)
+       if (tp->mac_version <= RTL_GIGA_MAC_VER_01)
                return;
-       if (tp->phy_version >= RTL_GIGA_PHY_VER_F) 
+       if (tp->phy_version >= RTL_GIGA_PHY_VER_H)
                return;
 
        dprintk("MAC version != 0 && PHY version == 0 or 1\n");
@@ -547,7 +1283,14 @@ static void rtl8169_hw_phy_config(struct net_device *dev)
 
        /* Shazam ! */
 
-       // phy config for RTL8169s mac_version C chip
+       if (tp->mac_version == RTL_GIGA_MAC_VER_04) {
+               mdio_write(ioaddr, 31, 0x0002);
+               mdio_write(ioaddr,  1, 0x90d0);
+               mdio_write(ioaddr, 31, 0x0000);
+               return;
+       }
+
+       /* phy config for RTL8169s mac_version C chip */
        mdio_write(ioaddr, 31, 0x0001);                 //w 31 2 0 1
        mdio_write(ioaddr, 21, 0x1000);                 //w 21 15 0 1000
        mdio_write(ioaddr, 24, 0x65c7);                 //w 24 15 0 65c7
@@ -566,198 +1309,298 @@ static void rtl8169_hw_phy_config(struct net_device *dev)
        mdio_write(ioaddr, 31, 0x0000); //w 31 2 0 0
 }
 
-static void rtl8169_hw_phy_reset(struct net_device *dev)
-{
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
-       int i, val;
-
-       printk(KERN_WARNING PFX "%s: Reset RTL8169s PHY\n", dev->name);
-
-       val = (mdio_read(ioaddr, 0) | 0x8000) & 0xffff;
-       mdio_write(ioaddr, 0, val);
-
-       for (i = 50; i >= 0; i--) {
-               if (!(mdio_read(ioaddr, 0) & 0x8000))
-                       break;
-               udelay(100); /* Gross */
-       }
-
-       if (i < 0) {
-               printk(KERN_WARNING PFX "%s: no PHY Reset ack. Giving up.\n",
-                      dev->name);
-       }
-}
-
 static void rtl8169_phy_timer(unsigned long __opaque)
 {
        struct net_device *dev = (struct net_device *)__opaque;
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
        struct timer_list *timer = &tp->timer;
-       void *ioaddr = tp->mmio_addr;
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned long timeout = RTL8169_PHY_TIMEOUT;
 
-       assert(tp->mac_version > RTL_GIGA_MAC_VER_B);
-       assert(tp->phy_version < RTL_GIGA_PHY_VER_G);
+       assert(tp->mac_version > RTL_GIGA_MAC_VER_01);
+       assert(tp->phy_version < RTL_GIGA_PHY_VER_H);
 
-       if (RTL_R8(PHYstatus) & LinkStatus)
-               tp->phy_link_down_cnt = 0;
-       else {
-               tp->phy_link_down_cnt++;
-               if (tp->phy_link_down_cnt >= 12) {
-                       int reg;
+       if (!(tp->phy_1000_ctrl_reg & ADVERTISE_1000FULL))
+               return;
 
-                       // If link on 1000, perform phy reset.
-                       reg = mdio_read(ioaddr, PHY_1000_CTRL_REG);
-                       if (reg & PHY_Cap_1000_Full) 
-                               rtl8169_hw_phy_reset(dev);
+       spin_lock_irq(&tp->lock);
 
-                       tp->phy_link_down_cnt = 0;
-               }
+       if (tp->phy_reset_pending(ioaddr)) {
+               /*
+                * A busy loop could burn quite a few cycles on nowadays CPU.
+                * Let's delay the execution of the timer for a few ticks.
+                */
+               timeout = HZ/10;
+               goto out_mod_timer;
        }
 
-       mod_timer(timer, jiffies + RTL8169_PHY_TIMEOUT);
+       if (tp->link_ok(ioaddr))
+               goto out_unlock;
+
+       if (netif_msg_link(tp))
+               printk(KERN_WARNING "%s: PHY reset until link up\n", dev->name);
+
+       tp->phy_reset_enable(ioaddr);
+
+out_mod_timer:
+       mod_timer(timer, jiffies + timeout);
+out_unlock:
+       spin_unlock_irq(&tp->lock);
 }
 
 static inline void rtl8169_delete_timer(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
        struct timer_list *timer = &tp->timer;
 
-       if ((tp->mac_version <= RTL_GIGA_MAC_VER_B) ||
-           (tp->phy_version >= RTL_GIGA_PHY_VER_G))
+       if ((tp->mac_version <= RTL_GIGA_MAC_VER_01) ||
+           (tp->phy_version >= RTL_GIGA_PHY_VER_H))
                return;
 
        del_timer_sync(timer);
-
-       tp->phy_link_down_cnt = 0;
 }
 
 static inline void rtl8169_request_timer(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
        struct timer_list *timer = &tp->timer;
 
-       if ((tp->mac_version <= RTL_GIGA_MAC_VER_B) ||
-           (tp->phy_version >= RTL_GIGA_PHY_VER_G))
+       if ((tp->mac_version <= RTL_GIGA_MAC_VER_01) ||
+           (tp->phy_version >= RTL_GIGA_PHY_VER_H))
                return;
 
-       tp->phy_link_down_cnt = 0;
+       mod_timer(timer, jiffies + RTL8169_PHY_TIMEOUT);
+}
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+/*
+ * Polling 'interrupt' - used by things like netconsole to send skbs
+ * without having to re-enable interrupts. It's not called while
+ * the interrupt routine is executing.
+ */
+static void rtl8169_netpoll(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       struct pci_dev *pdev = tp->pci_dev;
+
+       disable_irq(pdev->irq);
+       rtl8169_interrupt(pdev->irq, dev);
+       enable_irq(pdev->irq);
+}
+#endif
 
-       init_timer(timer);
-       timer->expires = jiffies + RTL8169_PHY_TIMEOUT;
-       timer->data = (unsigned long)(dev);
-       timer->function = rtl8169_phy_timer;
-       add_timer(timer);
+static void rtl8169_release_board(struct pci_dev *pdev, struct net_device *dev,
+                                 void __iomem *ioaddr)
+{
+       iounmap(ioaddr);
+       pci_release_regions(pdev);
+       pci_disable_device(pdev);
+       free_netdev(dev);
 }
 
-static int __devinit
-rtl8169_init_board(struct pci_dev *pdev, struct net_device **dev_out,
-                  void **ioaddr_out)
+static void rtl8169_phy_reset(struct net_device *dev,
+                             struct rtl8169_private *tp)
 {
-       void *ioaddr = NULL;
-       struct net_device *dev;
-       struct rtl8169_private *tp;
-       unsigned long mmio_start, mmio_end, mmio_flags, mmio_len;
-       int rc, i, acpi_idle_state = 0, pm_cap;
+       void __iomem *ioaddr = tp->mmio_addr;
+       int i;
+
+       tp->phy_reset_enable(ioaddr);
+       for (i = 0; i < 100; i++) {
+               if (!tp->phy_reset_pending(ioaddr))
+                       return;
+               msleep(1);
+       }
+       if (netif_msg_link(tp))
+               printk(KERN_ERR "%s: PHY reset failed.\n", dev->name);
+}
 
+static void rtl8169_init_phy(struct net_device *dev, struct rtl8169_private *tp)
+{
+       void __iomem *ioaddr = tp->mmio_addr;
+       static int board_idx = -1;
+       u8 autoneg, duplex;
+       u16 speed;
+
+       board_idx++;
+
+       rtl8169_hw_phy_config(dev);
+
+       dprintk("Set MAC Reg C+CR Offset 0x82h = 0x01h\n");
+       RTL_W8(0x82, 0x01);
+
+       if (tp->mac_version < RTL_GIGA_MAC_VER_03) {
+               dprintk("Set PCI Latency=0x40\n");
+               pci_write_config_byte(tp->pci_dev, PCI_LATENCY_TIMER, 0x40);
+       }
+
+       if (tp->mac_version == RTL_GIGA_MAC_VER_02) {
+               dprintk("Set MAC Reg C+CR Offset 0x82h = 0x01h\n");
+               RTL_W8(0x82, 0x01);
+               dprintk("Set PHY Reg 0x0bh = 0x00h\n");
+               mdio_write(ioaddr, 0x0b, 0x0000); //w 0x0b 15 0 0
+       }
+
+       rtl8169_link_option(board_idx, &autoneg, &speed, &duplex);
+
+       rtl8169_phy_reset(dev, tp);
+
+       rtl8169_set_speed(dev, autoneg, speed, duplex);
+
+       if ((RTL_R8(PHYstatus) & TBI_Enable) && netif_msg_link(tp))
+               printk(KERN_INFO PFX "%s: TBI auto-negotiating\n", dev->name);
+}
+
+static int rtl8169_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       struct mii_ioctl_data *data = if_mii(ifr);
+
+       if (!netif_running(dev))
+               return -ENODEV;
+
+       switch (cmd) {
+       case SIOCGMIIPHY:
+               data->phy_id = 32; /* Internal PHY */
+               return 0;
+
+       case SIOCGMIIREG:
+               data->val_out = mdio_read(tp->mmio_addr, data->reg_num & 0x1f);
+               return 0;
+
+       case SIOCSMIIREG:
+               if (!capable(CAP_NET_ADMIN))
+                       return -EPERM;
+               mdio_write(tp->mmio_addr, data->reg_num & 0x1f, data->val_in);
+               return 0;
+       }
+       return -EOPNOTSUPP;
+}
 
-       assert(pdev != NULL);
-       assert(ioaddr_out != NULL);
+static int __devinit
+rtl8169_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+       const unsigned int region = rtl_cfg_info[ent->driver_data].region;
+       struct rtl8169_private *tp;
+       struct net_device *dev;
+       void __iomem *ioaddr;
+       unsigned int pm_cap;
+       int i, rc;
 
-       *ioaddr_out = NULL;
-       *dev_out = NULL;
+       if (netif_msg_drv(&debug)) {
+               printk(KERN_INFO "%s Gigabit Ethernet driver %s loaded\n",
+                      MODULENAME, RTL8169_VERSION);
+       }
 
-       // dev zeroed in alloc_etherdev 
        dev = alloc_etherdev(sizeof (*tp));
-       if (dev == NULL) {
-               printk(KERN_ERR PFX "unable to alloc new ethernet\n");
-               return -ENOMEM;
+       if (!dev) {
+               if (netif_msg_drv(&debug))
+                       dev_err(&pdev->dev, "unable to alloc new ethernet\n");
+               rc = -ENOMEM;
+               goto out;
        }
 
        SET_MODULE_OWNER(dev);
        SET_NETDEV_DEV(dev, &pdev->dev);
-       tp = dev->priv;
+       tp = netdev_priv(dev);
+       tp->dev = dev;
+       tp->msg_enable = netif_msg_init(debug.msg_enable, R8169_MSG_DEFAULT);
 
-       // enable device (incl. PCI PM wakeup and hotplug setup)
+       /* enable device (incl. PCI PM wakeup and hotplug setup) */
        rc = pci_enable_device(pdev);
-       if (rc) {
-               printk(KERN_ERR PFX "%s: unable to enable device\n", pdev->slot_name);
-               goto err_out;
+       if (rc < 0) {
+               if (netif_msg_probe(tp))
+                       dev_err(&pdev->dev, "enable failure\n");
+               goto err_out_free_dev_1;
        }
 
+       rc = pci_set_mwi(pdev);
+       if (rc < 0)
+               goto err_out_disable_2;
+
        /* save power state before pci_enable_device overwrites it */
        pm_cap = pci_find_capability(pdev, PCI_CAP_ID_PM);
        if (pm_cap) {
-               u16 pwr_command;
+               u16 pwr_command, acpi_idle_state;
 
                pci_read_config_word(pdev, pm_cap + PCI_PM_CTRL, &pwr_command);
                acpi_idle_state = pwr_command & PCI_PM_CTRL_STATE_MASK;
        } else {
-               printk(KERN_ERR PFX "Cannot find PowerManagement capability, aborting.\n");
-               goto err_out_free_res;
+               if (netif_msg_probe(tp)) {
+                       dev_err(&pdev->dev,
+                               "PowerManagement capability not found.\n");
+               }
        }
 
-       mmio_start = pci_resource_start(pdev, 1);
-       mmio_end = pci_resource_end(pdev, 1);
-       mmio_flags = pci_resource_flags(pdev, 1);
-       mmio_len = pci_resource_len(pdev, 1);
-
-       // make sure PCI base addr 1 is MMIO
-       if (!(mmio_flags & IORESOURCE_MEM)) {
-               printk(KERN_ERR PFX
-                      "region #1 not an MMIO resource, aborting\n");
+       /* make sure PCI base addr 1 is MMIO */
+       if (!(pci_resource_flags(pdev, region) & IORESOURCE_MEM)) {
+               if (netif_msg_probe(tp)) {
+                       dev_err(&pdev->dev,
+                               "region #%d not an MMIO resource, aborting\n",
+                               region);
+               }
                rc = -ENODEV;
-               goto err_out_disable;
+               goto err_out_mwi_3;
        }
-       // check for weird/broken PCI region reporting
-       if (mmio_len < RTL_MIN_IO_SIZE) {
-               printk(KERN_ERR PFX "Invalid PCI region size(s), aborting\n");
+
+       /* check for weird/broken PCI region reporting */
+       if (pci_resource_len(pdev, region) < R8169_REGS_SIZE) {
+               if (netif_msg_probe(tp)) {
+                       dev_err(&pdev->dev,
+                               "Invalid PCI region size(s), aborting\n");
+               }
                rc = -ENODEV;
-               goto err_out_disable;
+               goto err_out_mwi_3;
        }
 
-       rc = pci_request_regions(pdev, dev->name);
-       if (rc) {
-               printk(KERN_ERR PFX "%s: Could not request regions.\n", pdev->slot_name);
-               goto err_out_disable;
+       rc = pci_request_regions(pdev, MODULENAME);
+       if (rc < 0) {
+               if (netif_msg_probe(tp))
+                       dev_err(&pdev->dev, "could not request regions.\n");
+               goto err_out_mwi_3;
        }
 
        tp->cp_cmd = PCIMulRW | RxChkSum;
 
-       if ((sizeof(dma_addr_t) > 32) &&
-           !pci_set_dma_mask(pdev, DMA_64BIT_MASK))
+       if ((sizeof(dma_addr_t) > 4) &&
+           !pci_set_dma_mask(pdev, DMA_64BIT_MASK) && use_dac) {
                tp->cp_cmd |= PCIDAC;
-       else {
+               dev->features |= NETIF_F_HIGHDMA;
+       } else {
                rc = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
                if (rc < 0) {
-                       printk(KERN_ERR PFX "DMA configuration failed.\n");
-                       goto err_out_free_res;
+                       if (netif_msg_probe(tp)) {
+                               dev_err(&pdev->dev,
+                                       "DMA configuration failed.\n");
+                       }
+                       goto err_out_free_res_4;
                }
        }
 
-
-       // enable PCI bus-mastering
        pci_set_master(pdev);
 
-       // ioremap MMIO region 
-       ioaddr = ioremap(mmio_start, mmio_len);
-       if (ioaddr == NULL) {
-               printk(KERN_ERR PFX "cannot remap MMIO, aborting\n");
+       /* ioremap MMIO region */
+       ioaddr = ioremap(pci_resource_start(pdev, region), R8169_REGS_SIZE);
+       if (!ioaddr) {
+               if (netif_msg_probe(tp))
+                       dev_err(&pdev->dev, "cannot remap MMIO, aborting\n");
                rc = -EIO;
-               goto err_out_free_res;
+               goto err_out_free_res_4;
        }
 
-       // Soft reset the chip. 
+       /* Unneeded ? Don't mess with Mrs. Murphy. */
+       rtl8169_irq_mask_and_ack(ioaddr);
+
+       /* Soft reset the chip. */
        RTL_W8(ChipCmd, CmdReset);
 
-       // Check that the chip has finished the reset.
-       for (i = 1000; i > 0; i--) {
+       /* Check that the chip has finished the reset. */
+       for (i = 100; i > 0; i--) {
                if ((RTL_R8(ChipCmd) & CmdReset) == 0)
                        break;
-               udelay(10);
+               msleep_interruptible(1);
        }
 
-       // Identify chip attached to board
+       /* Identify chip attached to board */
        rtl8169_get_mac_version(tp, ioaddr);
        rtl8169_get_phy_version(tp, ioaddr);
 
@@ -770,284 +1613,149 @@ rtl8169_init_board(struct pci_dev *pdev, struct net_device **dev_out,
        }
        if (i < 0) {
                /* Unknown chip: assume array element #0, original RTL-8169 */
-               printk(KERN_DEBUG PFX
-                      "PCI device %s: unknown chip version, assuming %s\n",
-                      pci_name(pdev), rtl_chip_info[0].name);
+               if (netif_msg_probe(tp)) {
+                       dev_printk(KERN_DEBUG, &pdev->dev,
+                               "unknown chip version, assuming %s\n",
+                               rtl_chip_info[0].name);
+               }
                i++;
        }
        tp->chipset = i;
 
-       *ioaddr_out = ioaddr;
-       *dev_out = dev;
-       return 0;
-
-err_out_free_res:
-       pci_release_regions(pdev);
-
-err_out_disable:
-       pci_disable_device(pdev);
-
-err_out:
-       free_netdev(dev);
-       return rc;
-}
-
-static int __devinit
-rtl8169_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
-{
-       struct net_device *dev = NULL;
-       struct rtl8169_private *tp = NULL;
-       void *ioaddr = NULL;
-       static int board_idx = -1;
-       static int printed_version = 0;
-       int i, rc;
-       int option = -1, Cap10_100 = 0, Cap1000 = 0;
+       RTL_W8(Cfg9346, Cfg9346_Unlock);
+       RTL_W8(Config1, RTL_R8(Config1) | PMEnable);
+       RTL_W8(Config5, RTL_R8(Config5) & PMEStatus);
+       RTL_W8(Cfg9346, Cfg9346_Lock);
 
-       assert(pdev != NULL);
-       assert(ent != NULL);
+       if (RTL_R8(PHYstatus) & TBI_Enable) {
+               tp->set_speed = rtl8169_set_speed_tbi;
+               tp->get_settings = rtl8169_gset_tbi;
+               tp->phy_reset_enable = rtl8169_tbi_reset_enable;
+               tp->phy_reset_pending = rtl8169_tbi_reset_pending;
+               tp->link_ok = rtl8169_tbi_link_ok;
 
-       board_idx++;
+               tp->phy_1000_ctrl_reg = ADVERTISE_1000FULL; /* Implied by TBI */
+       } else {
+               tp->set_speed = rtl8169_set_speed_xmii;
+               tp->get_settings = rtl8169_gset_xmii;
+               tp->phy_reset_enable = rtl8169_xmii_reset_enable;
+               tp->phy_reset_pending = rtl8169_xmii_reset_pending;
+               tp->link_ok = rtl8169_xmii_link_ok;
 
-       if (!printed_version) {
-               printk(KERN_INFO RTL8169_DRIVER_NAME " loaded\n");
-               printed_version = 1;
+               dev->do_ioctl = rtl8169_ioctl;
        }
 
-       rc = rtl8169_init_board(pdev, &dev, &ioaddr);
-       if (rc)
-               return rc;
-
-       tp = dev->priv;
-       assert(ioaddr != NULL);
-       assert(dev != NULL);
-       assert(tp != NULL);
-
-       // Get MAC address.  FIXME: read EEPROM
+       /* Get MAC address.  FIXME: read EEPROM */
        for (i = 0; i < MAC_ADDR_LEN; i++)
                dev->dev_addr[i] = RTL_R8(MAC0 + i);
+       memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
 
        dev->open = rtl8169_open;
        dev->hard_start_xmit = rtl8169_start_xmit;
        dev->get_stats = rtl8169_get_stats;
-       dev->ethtool_ops = &rtl8169_ethtool_ops;
+       SET_ETHTOOL_OPS(dev, &rtl8169_ethtool_ops);
        dev->stop = rtl8169_close;
        dev->tx_timeout = rtl8169_tx_timeout;
        dev->set_multicast_list = rtl8169_set_rx_mode;
        dev->watchdog_timeo = RTL8169_TX_TIMEOUT;
        dev->irq = pdev->irq;
        dev->base_addr = (unsigned long) ioaddr;
-//      dev->do_ioctl           = mii_ioctl;
+       dev->change_mtu = rtl8169_change_mtu;
+
+#ifdef CONFIG_R8169_NAPI
+       dev->poll = rtl8169_poll;
+       dev->weight = R8169_NAPI_WEIGHT;
+#endif
+
+#ifdef CONFIG_R8169_VLAN
+       dev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
+       dev->vlan_rx_register = rtl8169_vlan_rx_register;
+       dev->vlan_rx_kill_vid = rtl8169_vlan_rx_kill_vid;
+#endif
+
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       dev->poll_controller = rtl8169_netpoll;
+#endif
 
-       tp = dev->priv;         // private data //
+       tp->intr_mask = 0xffff;
        tp->pci_dev = pdev;
        tp->mmio_addr = ioaddr;
+       tp->align = rtl_cfg_info[ent->driver_data].align;
+
+       init_timer(&tp->timer);
+       tp->timer.data = (unsigned long) dev;
+       tp->timer.function = rtl8169_phy_timer;
 
        spin_lock_init(&tp->lock);
 
        rc = register_netdev(dev);
-       if (rc) {
-               iounmap(ioaddr);
-               pci_release_regions(pdev);
-               pci_disable_device(pdev);
-               free_netdev(dev);
-               return rc;
-       }
-
-       printk(KERN_DEBUG "%s: Identified chip type is '%s'.\n", dev->name,
-              rtl_chip_info[tp->chipset].name);
+       if (rc < 0)
+               goto err_out_unmap_5;
 
        pci_set_drvdata(pdev, dev);
 
-       printk(KERN_INFO "%s: %s at 0x%lx, "
-              "%2.2x:%2.2x:%2.2x:%2.2x:%2.2x:%2.2x, "
-              "IRQ %d\n",
-              dev->name,
-              rtl_chip_info[ent->driver_data].name,
-              dev->base_addr,
-              dev->dev_addr[0], dev->dev_addr[1],
-              dev->dev_addr[2], dev->dev_addr[3],
-              dev->dev_addr[4], dev->dev_addr[5], dev->irq);
-
-       rtl8169_hw_phy_config(dev);
-
-       dprintk("Set MAC Reg C+CR Offset 0x82h = 0x01h\n");
-       RTL_W8(0x82, 0x01);
-
-       if (tp->mac_version < RTL_GIGA_MAC_VER_E) {
-               dprintk("Set PCI Latency=0x40\n");
-               pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0x40);
-       }
-
-       if (tp->mac_version == RTL_GIGA_MAC_VER_D) {
-               dprintk("Set MAC Reg C+CR Offset 0x82h = 0x01h\n");
-               RTL_W8(0x82, 0x01);
-               dprintk("Set PHY Reg 0x0bh = 0x00h\n");
-               mdio_write(ioaddr, 0x0b, 0x0000); //w 0x0b 15 0 0
+       if (netif_msg_probe(tp)) {
+               printk(KERN_INFO "%s: %s at 0x%lx, "
+                      "%2.2x:%2.2x:%2.2x:%2.2x:%2.2x:%2.2x, "
+                      "IRQ %d\n",
+                      dev->name,
+                      rtl_chip_info[tp->chipset].name,
+                      dev->base_addr,
+                      dev->dev_addr[0], dev->dev_addr[1],
+                      dev->dev_addr[2], dev->dev_addr[3],
+                      dev->dev_addr[4], dev->dev_addr[5], dev->irq);
        }
 
-       // if TBI is not endbled
-       if (!(RTL_R8(PHYstatus) & TBI_Enable)) {
-               int val = mdio_read(ioaddr, PHY_AUTO_NEGO_REG);
-
-               option = (board_idx >= MAX_UNITS) ? 0 : media[board_idx];
-               // Force RTL8169 in 10/100/1000 Full/Half mode.
-               if (option > 0) {
-                       printk(KERN_INFO "%s: Force-mode Enabled.\n",
-                              dev->name);
-                       Cap10_100 = 0, Cap1000 = 0;
-                       switch (option) {
-                       case _10_Half:
-                               Cap10_100 = PHY_Cap_10_Half_Or_Less;
-                               Cap1000 = PHY_Cap_Null;
-                               break;
-                       case _10_Full:
-                               Cap10_100 = PHY_Cap_10_Full_Or_Less;
-                               Cap1000 = PHY_Cap_Null;
-                               break;
-                       case _100_Half:
-                               Cap10_100 = PHY_Cap_100_Half_Or_Less;
-                               Cap1000 = PHY_Cap_Null;
-                               break;
-                       case _100_Full:
-                               Cap10_100 = PHY_Cap_100_Full_Or_Less;
-                               Cap1000 = PHY_Cap_Null;
-                               break;
-                       case _1000_Full:
-                               Cap10_100 = PHY_Cap_100_Full_Or_Less;
-                               Cap1000 = PHY_Cap_1000_Full;
-                               break;
-                       default:
-                               break;
-                       }
-                       mdio_write(ioaddr, PHY_AUTO_NEGO_REG, Cap10_100 | (val & 0x1F));        //leave PHY_AUTO_NEGO_REG bit4:0 unchanged
-                       mdio_write(ioaddr, PHY_1000_CTRL_REG, Cap1000);
-               } else {
-                       printk(KERN_INFO "%s: Auto-negotiation Enabled.\n",
-                              dev->name);
-
-                       // enable 10/100 Full/Half Mode, leave PHY_AUTO_NEGO_REG bit4:0 unchanged
-                       mdio_write(ioaddr, PHY_AUTO_NEGO_REG,
-                                  PHY_Cap_100_Full_Or_Less | (val & 0x1f));
+       rtl8169_init_phy(dev, tp);
 
-                       // enable 1000 Full Mode
-                       mdio_write(ioaddr, PHY_1000_CTRL_REG,
-                                  PHY_Cap_1000_Full);
-
-               }
-
-               // Enable auto-negotiation and restart auto-nigotiation
-               mdio_write(ioaddr, PHY_CTRL_REG,
-                          PHY_Enable_Auto_Nego | PHY_Restart_Auto_Nego);
-               udelay(100);
-
-               // wait for auto-negotiation process
-               for (i = 10000; i > 0; i--) {
-                       //check if auto-negotiation complete
-                       if (mdio_read(ioaddr, PHY_STAT_REG) &
-                           PHY_Auto_Neco_Comp) {
-                               udelay(100);
-                               option = RTL_R8(PHYstatus);
-                               if (option & _1000bpsF) {
-                                       printk(KERN_INFO
-                                              "%s: 1000Mbps Full-duplex operation.\n",
-                                              dev->name);
-                               } else {
-                                       printk(KERN_INFO
-                                              "%s: %sMbps %s-duplex operation.\n",
-                                              dev->name,
-                                              (option & _100bps) ? "100" :
-                                              "10",
-                                              (option & FullDup) ? "Full" :
-                                              "Half");
-                               }
-                               break;
-                       } else {
-                               udelay(100);
-                       }
-               }               // end for-loop to wait for auto-negotiation process
-
-       } else {
-               udelay(100);
-               printk(KERN_INFO
-                      "%s: 1000Mbps Full-duplex operation, TBI Link %s!\n",
-                      dev->name,
-                      (RTL_R32(TBICSR) & TBILinkOK) ? "OK" : "Failed");
-
-       }
+out:
+       return rc;
 
-       return 0;
+err_out_unmap_5:
+       iounmap(ioaddr);
+err_out_free_res_4:
+       pci_release_regions(pdev);
+err_out_mwi_3:
+       pci_clear_mwi(pdev);
+err_out_disable_2:
+       pci_disable_device(pdev);
+err_out_free_dev_1:
+       free_netdev(dev);
+       goto out;
 }
 
 static void __devexit
 rtl8169_remove_one(struct pci_dev *pdev)
 {
        struct net_device *dev = pci_get_drvdata(pdev);
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
 
        assert(dev != NULL);
        assert(tp != NULL);
 
        unregister_netdev(dev);
-       iounmap(tp->mmio_addr);
-       pci_release_regions(pdev);
-
-       pci_disable_device(pdev);
-       free_netdev(dev);
+       rtl8169_release_board(pdev, dev, tp->mmio_addr);
        pci_set_drvdata(pdev, NULL);
 }
 
-#ifdef CONFIG_PM
-
-static int rtl8169_suspend(struct pci_dev *pdev, u32 state)
-{
-       struct net_device *dev = pci_get_drvdata(pdev);
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
-       unsigned long flags;
-
-       if (!netif_running(dev))
-               return 0;
-       
-       netif_device_detach(dev);
-       netif_stop_queue(dev);
-       spin_lock_irqsave(&tp->lock, flags);
-
-       /* Disable interrupts, stop Rx and Tx */
-       RTL_W16(IntrMask, 0);
-       RTL_W8(ChipCmd, 0);
-               
-       /* Update the error counts. */
-       tp->stats.rx_missed_errors += RTL_R32(RxMissed);
-       RTL_W32(RxMissed, 0);
-       spin_unlock_irqrestore(&tp->lock, flags);
-       
-       return 0;
-}
-
-static int rtl8169_resume(struct pci_dev *pdev)
+static void rtl8169_set_rxbufsize(struct rtl8169_private *tp,
+                                 struct net_device *dev)
 {
-       struct net_device *dev = pci_get_drvdata(pdev);
-
-       if (!netif_running(dev))
-           return 0;
-
-       netif_device_attach(dev);
-       rtl8169_hw_start(dev);
+       unsigned int mtu = dev->mtu;
 
-       return 0;
+       tp->rx_buf_sz = (mtu > RX_BUF_SIZE) ? mtu + ETH_HLEN + 8 : RX_BUF_SIZE;
 }
-                                                                                
-#endif /* CONFIG_PM */
 
-static int
-rtl8169_open(struct net_device *dev)
+static int rtl8169_open(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
        struct pci_dev *pdev = tp->pci_dev;
        int retval;
 
+       rtl8169_set_rxbufsize(tp, dev);
+
        retval =
-           request_irq(dev->irq, rtl8169_interrupt, SA_SHIRQ, dev->name, dev);
+           request_irq(dev->irq, rtl8169_interrupt, IRQF_SHARED, dev->name, dev);
        if (retval < 0)
                goto out;
 
@@ -1071,9 +1779,13 @@ rtl8169_open(struct net_device *dev)
        if (retval < 0)
                goto err_free_rx;
 
+       INIT_DELAYED_WORK(&tp->task, NULL);
+
        rtl8169_hw_start(dev);
 
        rtl8169_request_timer(dev);
+
+       rtl8169_check_link_status(dev, tp, tp->mmio_addr);
 out:
        return retval;
 
@@ -1088,57 +1800,132 @@ err_free_irq:
        goto out;
 }
 
-static void
-rtl8169_hw_start(struct net_device *dev)
+static void rtl8169_hw_reset(void __iomem *ioaddr)
+{
+       /* Disable interrupts */
+       rtl8169_irq_mask_and_ack(ioaddr);
+
+       /* Reset the chipset */
+       RTL_W8(ChipCmd, CmdReset);
+
+       /* PCI commit */
+       RTL_R8(ChipCmd);
+}
+
+static void rtl8169_set_rx_tx_config_registers(struct rtl8169_private *tp)
+{
+       void __iomem *ioaddr = tp->mmio_addr;
+       u32 cfg = rtl8169_rx_config;
+
+       cfg |= (RTL_R32(RxConfig) & rtl_chip_info[tp->chipset].RxConfigMask);
+       RTL_W32(RxConfig, cfg);
+
+       /* Set DMA burst size and Interframe Gap Time */
+       RTL_W32(TxConfig, (TX_DMA_BURST << TxDMAShift) |
+               (InterFrameGap << TxInterFrameGapShift));
+}
+
+static void rtl8169_hw_start(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       struct pci_dev *pdev = tp->pci_dev;
+       u16 cmd;
        u32 i;
 
        /* Soft reset the chip. */
        RTL_W8(ChipCmd, CmdReset);
 
        /* Check that the chip has finished the reset. */
-       for (i = 1000; i > 0; i--) {
+       for (i = 100; i > 0; i--) {
                if ((RTL_R8(ChipCmd) & CmdReset) == 0)
                        break;
-               else
-                       udelay(10);
+               msleep_interruptible(1);
+       }
+
+       if (tp->mac_version == RTL_GIGA_MAC_VER_05) {
+               RTL_W16(CPlusCmd, RTL_R16(CPlusCmd) | PCIMulRW);
+               pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);
+       }
+
+       if (tp->mac_version == RTL_GIGA_MAC_VER_13) {
+               pci_write_config_word(pdev, 0x68, 0x00);
+               pci_write_config_word(pdev, 0x69, 0x08);
+       }
+
+       /* Undocumented stuff. */
+       if (tp->mac_version == RTL_GIGA_MAC_VER_05) {
+               /* Realtek's r1000_n.c driver uses '&& 0x01' here. Well... */
+               if ((RTL_R8(Config2) & 0x07) & 0x01)
+                       RTL_W32(0x7c, 0x0007ffff);
+
+               RTL_W32(0x7c, 0x0007ff00);
+
+               pci_read_config_word(pdev, PCI_COMMAND, &cmd);
+               cmd = cmd & 0xef;
+               pci_write_config_word(pdev, PCI_COMMAND, cmd);
        }
 
        RTL_W8(Cfg9346, Cfg9346_Unlock);
-       RTL_W8(ChipCmd, CmdTxEnb | CmdRxEnb);
+       if ((tp->mac_version == RTL_GIGA_MAC_VER_01) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_02) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_03) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_04))
+               RTL_W8(ChipCmd, CmdTxEnb | CmdRxEnb);
+
        RTL_W8(EarlyTxThres, EarlyTxThld);
 
-       // For gigabit rtl8169
-       RTL_W16(RxMaxSize, RxPacketMaxSize);
+       /* Low hurts. Let's disable the filtering. */
+       RTL_W16(RxMaxSize, 16383);
 
-       // Set Rx Config register
-       i = rtl8169_rx_config | (RTL_R32(RxConfig) & rtl_chip_info[tp->chipset].
-                                RxConfigMask);
-       RTL_W32(RxConfig, i);
+       if ((tp->mac_version == RTL_GIGA_MAC_VER_01) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_02) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_03) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_04))
+               rtl8169_set_rx_tx_config_registers(tp);
 
-       /* Set DMA burst size and Interframe Gap Time */
-       RTL_W32(TxConfig,
-               (TX_DMA_BURST << TxDMAShift) | (InterFrameGap <<
-                                               TxInterFrameGapShift));
-       tp->cp_cmd |= RTL_R16(CPlusCmd);
-       RTL_W16(CPlusCmd, tp->cp_cmd);
+       cmd = RTL_R16(CPlusCmd);
+       RTL_W16(CPlusCmd, cmd);
 
-       if (tp->mac_version == RTL_GIGA_MAC_VER_D) {
-               dprintk(KERN_INFO PFX "Set MAC Reg C+CR Offset 0xE0: bit-3 and bit-14 MUST be 1\n");
-               tp->cp_cmd |= (1 << 14) | PCIMulRW;
-               RTL_W16(CPlusCmd, tp->cp_cmd);
+       tp->cp_cmd |= cmd | PCIMulRW;
+
+       if ((tp->mac_version == RTL_GIGA_MAC_VER_02) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_03)) {
+               dprintk(KERN_INFO PFX "Set MAC Reg C+CR Offset 0xE0. "
+                       "Bit-3 and bit-14 MUST be 1\n");
+               tp->cp_cmd |= (1 << 14);
        }
 
-       tp->cur_rx = 0;
+       RTL_W16(CPlusCmd, tp->cp_cmd);
+
+       /*
+        * Undocumented corner. Supposedly:
+        * (TxTimer << 12) | (TxPackets << 8) | (RxTimer << 4) | RxPackets
+        */
+       RTL_W16(IntrMitigate, 0x0000);
 
-       RTL_W32(TxDescStartAddrLow, ((u64) tp->TxPhyAddr & DMA_32BIT_MASK));
+       /*
+        * Magic spell: some iop3xx ARM board needs the TxDescAddrHigh
+        * register to be written before TxDescAddrLow to work.
+        * Switching from MMIO to I/O access fixes the issue as well.
+        */
        RTL_W32(TxDescStartAddrHigh, ((u64) tp->TxPhyAddr >> 32));
-       RTL_W32(RxDescAddrLow, ((u64) tp->RxPhyAddr & DMA_32BIT_MASK));
+       RTL_W32(TxDescStartAddrLow, ((u64) tp->TxPhyAddr & DMA_32BIT_MASK));
        RTL_W32(RxDescAddrHigh, ((u64) tp->RxPhyAddr >> 32));
+       RTL_W32(RxDescAddrLow, ((u64) tp->RxPhyAddr & DMA_32BIT_MASK));
+
+       if ((tp->mac_version != RTL_GIGA_MAC_VER_01) &&
+           (tp->mac_version != RTL_GIGA_MAC_VER_02) &&
+           (tp->mac_version != RTL_GIGA_MAC_VER_03) &&
+           (tp->mac_version != RTL_GIGA_MAC_VER_04)) {
+               RTL_W8(ChipCmd, CmdTxEnb | CmdRxEnb);
+               rtl8169_set_rx_tx_config_registers(tp);
+       }
+
        RTL_W8(Cfg9346, Cfg9346_Lock);
-       udelay(10);
+
+       /* Initially a 10 us delay. Turned it into a PCI commit. - FR */
+       RTL_R8(IntrMask);
 
        RTL_W32(RxMissed, 0);
 
@@ -1151,55 +1938,91 @@ rtl8169_hw_start(struct net_device *dev)
        RTL_W16(IntrMask, rtl8169_intr_mask);
 
        netif_start_queue(dev);
+}
+
+static int rtl8169_change_mtu(struct net_device *dev, int new_mtu)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       int ret = 0;
+
+       if (new_mtu < ETH_ZLEN || new_mtu > SafeMtu)
+               return -EINVAL;
+
+       dev->mtu = new_mtu;
+
+       if (!netif_running(dev))
+               goto out;
+
+       rtl8169_down(dev);
+
+       rtl8169_set_rxbufsize(tp, dev);
 
+       ret = rtl8169_init_ring(dev);
+       if (ret < 0)
+               goto out;
+
+       netif_poll_enable(dev);
+
+       rtl8169_hw_start(dev);
+
+       rtl8169_request_timer(dev);
+
+out:
+       return ret;
 }
 
 static inline void rtl8169_make_unusable_by_asic(struct RxDesc *desc)
 {
        desc->addr = 0x0badbadbadbadbadull;
-       desc->status &= ~cpu_to_le32(OWNbit | RsvdMask);
+       desc->opts1 &= ~cpu_to_le32(DescOwn | RsvdMask);
 }
 
-static void rtl8169_free_rx_skb(struct pci_dev *pdev, struct sk_buff **sk_buff,
-                               struct RxDesc *desc)
+static void rtl8169_free_rx_skb(struct rtl8169_private *tp,
+                               struct sk_buff **sk_buff, struct RxDesc *desc)
 {
-       pci_unmap_single(pdev, le64_to_cpu(desc->addr), RX_BUF_SIZE,
+       struct pci_dev *pdev = tp->pci_dev;
+
+       pci_unmap_single(pdev, le64_to_cpu(desc->addr), tp->rx_buf_sz,
                         PCI_DMA_FROMDEVICE);
        dev_kfree_skb(*sk_buff);
        *sk_buff = NULL;
        rtl8169_make_unusable_by_asic(desc);
 }
 
-static inline void rtl8169_return_to_asic(struct RxDesc *desc)
+static inline void rtl8169_mark_to_asic(struct RxDesc *desc, u32 rx_buf_sz)
 {
-       desc->status |= cpu_to_le32(OWNbit + RX_BUF_SIZE);
+       u32 eor = le32_to_cpu(desc->opts1) & RingEnd;
+
+       desc->opts1 = cpu_to_le32(DescOwn | eor | rx_buf_sz);
 }
 
-static inline void rtl8169_give_to_asic(struct RxDesc *desc, dma_addr_t mapping)
+static inline void rtl8169_map_to_asic(struct RxDesc *desc, dma_addr_t mapping,
+                                      u32 rx_buf_sz)
 {
        desc->addr = cpu_to_le64(mapping);
-       desc->status |= cpu_to_le32(OWNbit + RX_BUF_SIZE);
+       wmb();
+       rtl8169_mark_to_asic(desc, rx_buf_sz);
 }
 
-static int rtl8169_alloc_rx_skb(struct pci_dev *pdev, struct net_device *dev,
-                               struct sk_buff **sk_buff, struct RxDesc *desc)
+static int rtl8169_alloc_rx_skb(struct pci_dev *pdev, struct sk_buff **sk_buff,
+                               struct RxDesc *desc, int rx_buf_sz,
+                               unsigned int align)
 {
        struct sk_buff *skb;
        dma_addr_t mapping;
        int ret = 0;
 
-       skb = dev_alloc_skb(RX_BUF_SIZE);
+       skb = dev_alloc_skb(rx_buf_sz + align);
        if (!skb)
                goto err_out;
 
-       skb->dev = dev;
-       skb_reserve(skb, 2);
+       skb_reserve(skb, (align - 1) & (u32)skb->data);
        *sk_buff = skb;
 
-       mapping = pci_map_single(pdev, skb->tail, RX_BUF_SIZE,
+       mapping = pci_map_single(pdev, skb->data, rx_buf_sz,
                                 PCI_DMA_FROMDEVICE);
 
-       rtl8169_give_to_asic(desc, mapping);
+       rtl8169_map_to_asic(desc, mapping, rx_buf_sz);
 
 out:
        return ret;
@@ -1216,7 +2039,7 @@ static void rtl8169_rx_clear(struct rtl8169_private *tp)
 
        for (i = 0; i < NUM_RX_DESC; i++) {
                if (tp->Rx_skbuff[i]) {
-                       rtl8169_free_rx_skb(tp->pci_dev, tp->Rx_skbuff + i,
+                       rtl8169_free_rx_skb(tp, tp->Rx_skbuff + i,
                                            tp->RxDescArray + i);
                }
        }
@@ -1226,15 +2049,15 @@ static u32 rtl8169_rx_fill(struct rtl8169_private *tp, struct net_device *dev,
                           u32 start, u32 end)
 {
        u32 cur;
-       
+
        for (cur = start; end - cur > 0; cur++) {
                int ret, i = cur % NUM_RX_DESC;
 
                if (tp->Rx_skbuff[i])
                        continue;
-                       
-               ret = rtl8169_alloc_rx_skb(tp->pci_dev, dev, tp->Rx_skbuff + i,
-                                          tp->RxDescArray + i);
+
+               ret = rtl8169_alloc_rx_skb(tp->pci_dev, tp->Rx_skbuff + i,
+                       tp->RxDescArray + i, tp->rx_buf_sz, tp->align);
                if (ret < 0)
                        break;
        }
@@ -1243,19 +2066,21 @@ static u32 rtl8169_rx_fill(struct rtl8169_private *tp, struct net_device *dev,
 
 static inline void rtl8169_mark_as_last_descriptor(struct RxDesc *desc)
 {
-       desc->status |= cpu_to_le32(EORbit);
+       desc->opts1 |= cpu_to_le32(RingEnd);
+}
+
+static void rtl8169_init_ring_indexes(struct rtl8169_private *tp)
+{
+       tp->dirty_tx = tp->dirty_rx = tp->cur_tx = tp->cur_rx = 0;
 }
 
 static int rtl8169_init_ring(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
 
-       tp->cur_rx = tp->dirty_rx = 0;
-       tp->cur_tx = tp->dirty_tx = 0;
-       memset(tp->TxDescArray, 0x0, NUM_TX_DESC * sizeof (struct TxDesc));
-       memset(tp->RxDescArray, 0x0, NUM_RX_DESC * sizeof (struct RxDesc));
+       rtl8169_init_ring_indexes(tp);
 
-       memset(tp->Tx_skbuff, 0x0, NUM_TX_DESC * sizeof(struct sk_buff *));
+       memset(tp->tx_skb, 0x0, NUM_TX_DESC * sizeof(struct ring_info));
        memset(tp->Rx_skbuff, 0x0, NUM_RX_DESC * sizeof(struct sk_buff *));
 
        if (rtl8169_rx_fill(tp, dev, 0, NUM_RX_DESC) != NUM_RX_DESC)
@@ -1270,184 +2095,414 @@ err_out:
        return -ENOMEM;
 }
 
-static void rtl8169_unmap_tx_skb(struct pci_dev *pdev, struct sk_buff **sk_buff,
+static void rtl8169_unmap_tx_skb(struct pci_dev *pdev, struct ring_info *tx_skb,
                                 struct TxDesc *desc)
 {
-       u32 len = sk_buff[0]->len;
+       unsigned int len = tx_skb->len;
 
-       pci_unmap_single(pdev, le64_to_cpu(desc->addr),
-                        len < ETH_ZLEN ? ETH_ZLEN : len, PCI_DMA_TODEVICE);
+       pci_unmap_single(pdev, le64_to_cpu(desc->addr), len, PCI_DMA_TODEVICE);
+       desc->opts1 = 0x00;
+       desc->opts2 = 0x00;
        desc->addr = 0x00;
-       *sk_buff = NULL;
+       tx_skb->len = 0;
 }
 
-static void
-rtl8169_tx_clear(struct rtl8169_private *tp)
+static void rtl8169_tx_clear(struct rtl8169_private *tp)
 {
-       int i;
+       unsigned int i;
 
-       tp->cur_tx = 0;
-       for (i = 0; i < NUM_TX_DESC; i++) {
-               struct sk_buff *skb = tp->Tx_skbuff[i];
+       for (i = tp->dirty_tx; i < tp->dirty_tx + NUM_TX_DESC; i++) {
+               unsigned int entry = i % NUM_TX_DESC;
+               struct ring_info *tx_skb = tp->tx_skb + entry;
+               unsigned int len = tx_skb->len;
 
-               if (skb) {
-                       rtl8169_unmap_tx_skb(tp->pci_dev, tp->Tx_skbuff + i,
-                                            tp->TxDescArray + i);
-                       dev_kfree_skb(skb);
+               if (len) {
+                       struct sk_buff *skb = tx_skb->skb;
+
+                       rtl8169_unmap_tx_skb(tp->pci_dev, tx_skb,
+                                            tp->TxDescArray + entry);
+                       if (skb) {
+                               dev_kfree_skb(skb);
+                               tx_skb->skb = NULL;
+                       }
                        tp->stats.tx_dropped++;
                }
        }
+       tp->cur_tx = tp->dirty_tx = 0;
 }
 
-static void
-rtl8169_tx_timeout(struct net_device *dev)
+static void rtl8169_schedule_work(struct net_device *dev, work_func_t task)
 {
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
-       u8 tmp8;
+       struct rtl8169_private *tp = netdev_priv(dev);
 
-       /* disable Tx, if not already */
-       tmp8 = RTL_R8(ChipCmd);
-       if (tmp8 & CmdTxEnb)
-               RTL_W8(ChipCmd, tmp8 & ~CmdTxEnb);
+       PREPARE_DELAYED_WORK(&tp->task, task);
+       schedule_delayed_work(&tp->task, 4);
+}
 
-       /* Disable interrupts by clearing the interrupt mask. */
-       RTL_W16(IntrMask, 0x0000);
+static void rtl8169_wait_for_quiescence(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
 
-       /* Stop a shared interrupt from scavenging while we are. */
-       spin_lock_irq(&tp->lock);
-       rtl8169_tx_clear(tp);
-       spin_unlock_irq(&tp->lock);
+       synchronize_irq(dev->irq);
 
-       /* ...and finally, reset everything */
-       rtl8169_hw_start(dev);
+       /* Wait for any pending NAPI task to complete */
+       netif_poll_disable(dev);
 
-       netif_wake_queue(dev);
+       rtl8169_irq_mask_and_ack(ioaddr);
+
+       netif_poll_enable(dev);
 }
 
-static int
-rtl8169_start_xmit(struct sk_buff *skb, struct net_device *dev)
+static void rtl8169_reinit_task(struct work_struct *work)
 {
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
-       int entry = tp->cur_tx % NUM_TX_DESC;
-       u32 len = skb->len;
+       struct rtl8169_private *tp =
+               container_of(work, struct rtl8169_private, task.work);
+       struct net_device *dev = tp->dev;
+       int ret;
 
-       if (unlikely(skb->len < ETH_ZLEN)) {
-               skb = skb_padto(skb, ETH_ZLEN);
-               if (!skb)
-                       goto err_update_stats;
-               len = ETH_ZLEN;
+       if (netif_running(dev)) {
+               rtl8169_wait_for_quiescence(dev);
+               rtl8169_close(dev);
        }
-       
-       spin_lock_irq(&tp->lock);
 
-       if (!(le32_to_cpu(tp->TxDescArray[entry].status) & OWNbit)) {
+       ret = rtl8169_open(dev);
+       if (unlikely(ret < 0)) {
+               if (net_ratelimit()) {
+                       struct rtl8169_private *tp = netdev_priv(dev);
+
+                       if (netif_msg_drv(tp)) {
+                               printk(PFX KERN_ERR
+                                      "%s: reinit failure (status = %d)."
+                                      " Rescheduling.\n", dev->name, ret);
+                       }
+               }
+               rtl8169_schedule_work(dev, rtl8169_reinit_task);
+       }
+}
+
+static void rtl8169_reset_task(struct work_struct *work)
+{
+       struct rtl8169_private *tp =
+               container_of(work, struct rtl8169_private, task.work);
+       struct net_device *dev = tp->dev;
+
+       if (!netif_running(dev))
+               return;
+
+       rtl8169_wait_for_quiescence(dev);
+
+       rtl8169_rx_interrupt(dev, tp, tp->mmio_addr);
+       rtl8169_tx_clear(tp);
+
+       if (tp->dirty_rx == tp->cur_rx) {
+               rtl8169_init_ring_indexes(tp);
+               rtl8169_hw_start(dev);
+               netif_wake_queue(dev);
+       } else {
+               if (net_ratelimit()) {
+                       struct rtl8169_private *tp = netdev_priv(dev);
+
+                       if (netif_msg_intr(tp)) {
+                               printk(PFX KERN_EMERG
+                                      "%s: Rx buffers shortage\n", dev->name);
+                       }
+               }
+               rtl8169_schedule_work(dev, rtl8169_reset_task);
+       }
+}
+
+static void rtl8169_tx_timeout(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+
+       rtl8169_hw_reset(tp->mmio_addr);
+
+       /* Let's wait a bit while any (async) irq lands on */
+       rtl8169_schedule_work(dev, rtl8169_reset_task);
+}
+
+static int rtl8169_xmit_frags(struct rtl8169_private *tp, struct sk_buff *skb,
+                             u32 opts1)
+{
+       struct skb_shared_info *info = skb_shinfo(skb);
+       unsigned int cur_frag, entry;
+       struct TxDesc *txd;
+
+       entry = tp->cur_tx;
+       for (cur_frag = 0; cur_frag < info->nr_frags; cur_frag++) {
+               skb_frag_t *frag = info->frags + cur_frag;
                dma_addr_t mapping;
+               u32 status, len;
+               void *addr;
 
-               mapping = pci_map_single(tp->pci_dev, skb->data, len,
-                                        PCI_DMA_TODEVICE);
+               entry = (entry + 1) % NUM_TX_DESC;
 
-               tp->Tx_skbuff[entry] = skb;
-               tp->TxDescArray[entry].addr = cpu_to_le64(mapping);
+               txd = tp->TxDescArray + entry;
+               len = frag->size;
+               addr = ((void *) page_address(frag->page)) + frag->page_offset;
+               mapping = pci_map_single(tp->pci_dev, addr, len, PCI_DMA_TODEVICE);
 
-               tp->TxDescArray[entry].status = cpu_to_le32(OWNbit | FSbit |
-                       LSbit | len | (EORbit * !((entry + 1) % NUM_TX_DESC)));
-                       
-               RTL_W8(TxPoll, 0x40);   //set polling bit
+               /* anti gcc 2.95.3 bugware (sic) */
+               status = opts1 | len | (RingEnd * !((entry + 1) % NUM_TX_DESC));
 
-               dev->trans_start = jiffies;
+               txd->opts1 = cpu_to_le32(status);
+               txd->addr = cpu_to_le64(mapping);
 
-               tp->cur_tx++;
-       } else
-               goto err_drop;
+               tp->tx_skb[entry].len = len;
+       }
+
+       if (cur_frag) {
+               tp->tx_skb[entry].skb = skb;
+               txd->opts1 |= cpu_to_le32(LastFrag);
+       }
 
+       return cur_frag;
+}
+
+static inline u32 rtl8169_tso_csum(struct sk_buff *skb, struct net_device *dev)
+{
+       if (dev->features & NETIF_F_TSO) {
+               u32 mss = skb_shinfo(skb)->gso_size;
+
+               if (mss)
+                       return LargeSend | ((mss & MSSMask) << MSSShift);
+       }
+       if (skb->ip_summed == CHECKSUM_PARTIAL) {
+               const struct iphdr *ip = skb->nh.iph;
+
+               if (ip->protocol == IPPROTO_TCP)
+                       return IPCS | TCPCS;
+               else if (ip->protocol == IPPROTO_UDP)
+                       return IPCS | UDPCS;
+               WARN_ON(1);     /* we need a WARN() */
+       }
+       return 0;
+}
+
+static int rtl8169_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       unsigned int frags, entry = tp->cur_tx % NUM_TX_DESC;
+       struct TxDesc *txd = tp->TxDescArray + entry;
+       void __iomem *ioaddr = tp->mmio_addr;
+       dma_addr_t mapping;
+       u32 status, len;
+       u32 opts1;
+       int ret = NETDEV_TX_OK;
+
+       if (unlikely(TX_BUFFS_AVAIL(tp) < skb_shinfo(skb)->nr_frags)) {
+               if (netif_msg_drv(tp)) {
+                       printk(KERN_ERR
+                              "%s: BUG! Tx Ring full when queue awake!\n",
+                              dev->name);
+               }
+               goto err_stop;
+       }
+
+       if (unlikely(le32_to_cpu(txd->opts1) & DescOwn))
+               goto err_stop;
+
+       opts1 = DescOwn | rtl8169_tso_csum(skb, dev);
 
-       if ((tp->cur_tx - NUM_TX_DESC) == tp->dirty_tx) {
+       frags = rtl8169_xmit_frags(tp, skb, opts1);
+       if (frags) {
+               len = skb_headlen(skb);
+               opts1 |= FirstFrag;
+       } else {
+               len = skb->len;
+
+               if (unlikely(len < ETH_ZLEN)) {
+                       if (skb_padto(skb, ETH_ZLEN))
+                               goto err_update_stats;
+                       len = ETH_ZLEN;
+               }
+
+               opts1 |= FirstFrag | LastFrag;
+               tp->tx_skb[entry].skb = skb;
+       }
+
+       mapping = pci_map_single(tp->pci_dev, skb->data, len, PCI_DMA_TODEVICE);
+
+       tp->tx_skb[entry].len = len;
+       txd->addr = cpu_to_le64(mapping);
+       txd->opts2 = cpu_to_le32(rtl8169_tx_vlan_tag(tp, skb));
+
+       wmb();
+
+       /* anti gcc 2.95.3 bugware (sic) */
+       status = opts1 | len | (RingEnd * !((entry + 1) % NUM_TX_DESC));
+       txd->opts1 = cpu_to_le32(status);
+
+       dev->trans_start = jiffies;
+
+       tp->cur_tx += frags + 1;
+
+       smp_wmb();
+
+       RTL_W8(TxPoll, 0x40);   /* set polling bit */
+
+       if (TX_BUFFS_AVAIL(tp) < MAX_SKB_FRAGS) {
                netif_stop_queue(dev);
+               smp_rmb();
+               if (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)
+                       netif_wake_queue(dev);
        }
-out:
-       spin_unlock_irq(&tp->lock);
 
-       return 0;
+out:
+       return ret;
 
-err_drop:
-       dev_kfree_skb(skb);
+err_stop:
+       netif_stop_queue(dev);
+       ret = NETDEV_TX_BUSY;
 err_update_stats:
        tp->stats.tx_dropped++;
        goto out;
 }
 
+static void rtl8169_pcierr_interrupt(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       struct pci_dev *pdev = tp->pci_dev;
+       void __iomem *ioaddr = tp->mmio_addr;
+       u16 pci_status, pci_cmd;
+
+       pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
+       pci_read_config_word(pdev, PCI_STATUS, &pci_status);
+
+       if (netif_msg_intr(tp)) {
+               printk(KERN_ERR
+                      "%s: PCI error (cmd = 0x%04x, status = 0x%04x).\n",
+                      dev->name, pci_cmd, pci_status);
+       }
+
+       /*
+        * The recovery sequence below admits a very elaborated explanation:
+        * - it seems to work;
+        * - I did not see what else could be done;
+        * - it makes iop3xx happy.
+        *
+        * Feel free to adjust to your needs.
+        */
+       if (pdev->broken_parity_status)
+               pci_cmd &= ~PCI_COMMAND_PARITY;
+       else
+               pci_cmd |= PCI_COMMAND_SERR | PCI_COMMAND_PARITY;
+
+       pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
+
+       pci_write_config_word(pdev, PCI_STATUS,
+               pci_status & (PCI_STATUS_DETECTED_PARITY |
+               PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_REC_MASTER_ABORT |
+               PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_SIG_TARGET_ABORT));
+
+       /* The infamous DAC f*ckup only happens at boot time */
+       if ((tp->cp_cmd & PCIDAC) && !tp->dirty_rx && !tp->cur_rx) {
+               if (netif_msg_intr(tp))
+                       printk(KERN_INFO "%s: disabling PCI DAC.\n", dev->name);
+               tp->cp_cmd &= ~PCIDAC;
+               RTL_W16(CPlusCmd, tp->cp_cmd);
+               dev->features &= ~NETIF_F_HIGHDMA;
+       }
+
+       rtl8169_hw_reset(ioaddr);
+
+       rtl8169_schedule_work(dev, rtl8169_reinit_task);
+}
+
 static void
 rtl8169_tx_interrupt(struct net_device *dev, struct rtl8169_private *tp,
-                    void *ioaddr)
+                    void __iomem *ioaddr)
 {
-       unsigned long dirty_tx, tx_left;
+       unsigned int dirty_tx, tx_left;
 
        assert(dev != NULL);
        assert(tp != NULL);
        assert(ioaddr != NULL);
 
        dirty_tx = tp->dirty_tx;
+       smp_rmb();
        tx_left = tp->cur_tx - dirty_tx;
 
        while (tx_left > 0) {
-               int entry = dirty_tx % NUM_TX_DESC;
-               struct sk_buff *skb = tp->Tx_skbuff[entry];
+               unsigned int entry = dirty_tx % NUM_TX_DESC;
+               struct ring_info *tx_skb = tp->tx_skb + entry;
+               u32 len = tx_skb->len;
                u32 status;
 
                rmb();
-               status = le32_to_cpu(tp->TxDescArray[entry].status);
-               if (status & OWNbit)
+               status = le32_to_cpu(tp->TxDescArray[entry].opts1);
+               if (status & DescOwn)
                        break;
 
-               /* FIXME: is it really accurate for TxErr ? */
-               tp->stats.tx_bytes += skb->len >= ETH_ZLEN ?
-                                     skb->len : ETH_ZLEN;
+               tp->stats.tx_bytes += len;
                tp->stats.tx_packets++;
-               rtl8169_unmap_tx_skb(tp->pci_dev, tp->Tx_skbuff + entry,
-                                    tp->TxDescArray + entry);
-               dev_kfree_skb_irq(skb);
-               tp->Tx_skbuff[entry] = NULL;
+
+               rtl8169_unmap_tx_skb(tp->pci_dev, tx_skb, tp->TxDescArray + entry);
+
+               if (status & LastFrag) {
+                       dev_kfree_skb_irq(tx_skb->skb);
+                       tx_skb->skb = NULL;
+               }
                dirty_tx++;
                tx_left--;
        }
 
        if (tp->dirty_tx != dirty_tx) {
                tp->dirty_tx = dirty_tx;
-               if (netif_queue_stopped(dev))
+               smp_wmb();
+               if (netif_queue_stopped(dev) &&
+                   (TX_BUFFS_AVAIL(tp) >= MAX_SKB_FRAGS)) {
                        netif_wake_queue(dev);
+               }
        }
 }
 
+static inline int rtl8169_fragmented_frame(u32 status)
+{
+       return (status & (FirstFrag | LastFrag)) != (FirstFrag | LastFrag);
+}
+
+static inline void rtl8169_rx_csum(struct sk_buff *skb, struct RxDesc *desc)
+{
+       u32 opts1 = le32_to_cpu(desc->opts1);
+       u32 status = opts1 & RxProtoMask;
+
+       if (((status == RxProtoTCP) && !(opts1 & TCPFail)) ||
+           ((status == RxProtoUDP) && !(opts1 & UDPFail)) ||
+           ((status == RxProtoIP) && !(opts1 & IPFail)))
+               skb->ip_summed = CHECKSUM_UNNECESSARY;
+       else
+               skb->ip_summed = CHECKSUM_NONE;
+}
+
 static inline int rtl8169_try_rx_copy(struct sk_buff **sk_buff, int pkt_size,
-                                     struct RxDesc *desc,
-                                     struct net_device *dev)
+                                     struct RxDesc *desc, int rx_buf_sz,
+                                     unsigned int align)
 {
        int ret = -1;
 
        if (pkt_size < rx_copybreak) {
                struct sk_buff *skb;
 
-               skb = dev_alloc_skb(pkt_size + 2);
+               skb = dev_alloc_skb(pkt_size + align);
                if (skb) {
-                       skb->dev = dev;
-                       skb_reserve(skb, 2);
-                       eth_copy_and_sum(skb, sk_buff[0]->tail, pkt_size, 0);
+                       skb_reserve(skb, (align - 1) & (u32)skb->data);
+                       eth_copy_and_sum(skb, sk_buff[0]->data, pkt_size, 0);
                        *sk_buff = skb;
-                       rtl8169_return_to_asic(desc);
+                       rtl8169_mark_to_asic(desc, rx_buf_sz);
                        ret = 0;
                }
        }
        return ret;
 }
 
-static void
+static int
 rtl8169_rx_interrupt(struct net_device *dev, struct rtl8169_private *tp,
-                    void *ioaddr)
+                    void __iomem *ioaddr)
 {
-       unsigned long cur_rx, rx_left;
-       int delta;
+       unsigned int cur_rx, rx_left;
+       unsigned int delta, count;
 
        assert(dev != NULL);
        assert(tp != NULL);
@@ -1455,63 +2510,87 @@ rtl8169_rx_interrupt(struct net_device *dev, struct rtl8169_private *tp,
 
        cur_rx = tp->cur_rx;
        rx_left = NUM_RX_DESC + tp->dirty_rx - cur_rx;
+       rx_left = rtl8169_rx_quota(rx_left, (u32) dev->quota);
 
-       while (rx_left > 0) {
-               int entry = cur_rx % NUM_RX_DESC;
+       for (; rx_left > 0; rx_left--, cur_rx++) {
+               unsigned int entry = cur_rx % NUM_RX_DESC;
+               struct RxDesc *desc = tp->RxDescArray + entry;
                u32 status;
 
                rmb();
-               status = le32_to_cpu(tp->RxDescArray[entry].status);
+               status = le32_to_cpu(desc->opts1);
 
-               if (status & OWNbit)
+               if (status & DescOwn)
                        break;
-               if (status & RxRES) {
-                       printk(KERN_INFO "%s: Rx ERROR!!!\n", dev->name);
+               if (unlikely(status & RxRES)) {
+                       if (netif_msg_rx_err(tp)) {
+                               printk(KERN_INFO
+                                      "%s: Rx ERROR. status = %08x\n",
+                                      dev->name, status);
+                       }
                        tp->stats.rx_errors++;
                        if (status & (RxRWT | RxRUNT))
                                tp->stats.rx_length_errors++;
                        if (status & RxCRC)
                                tp->stats.rx_crc_errors++;
+                       if (status & RxFOVF) {
+                               rtl8169_schedule_work(dev, rtl8169_reset_task);
+                               tp->stats.rx_fifo_errors++;
+                       }
+                       rtl8169_mark_to_asic(desc, tp->rx_buf_sz);
                } else {
-                       struct RxDesc *desc = tp->RxDescArray + entry;
                        struct sk_buff *skb = tp->Rx_skbuff[entry];
                        int pkt_size = (status & 0x00001FFF) - 4;
                        void (*pci_action)(struct pci_dev *, dma_addr_t,
                                size_t, int) = pci_dma_sync_single_for_device;
 
+                       /*
+                        * The driver does not support incoming fragmented
+                        * frames. They are seen as a symptom of over-mtu
+                        * sized frames.
+                        */
+                       if (unlikely(rtl8169_fragmented_frame(status))) {
+                               tp->stats.rx_dropped++;
+                               tp->stats.rx_length_errors++;
+                               rtl8169_mark_to_asic(desc, tp->rx_buf_sz);
+                               continue;
+                       }
+
+                       rtl8169_rx_csum(skb, desc);
 
                        pci_dma_sync_single_for_cpu(tp->pci_dev,
-                               le64_to_cpu(desc->addr), RX_BUF_SIZE,
+                               le64_to_cpu(desc->addr), tp->rx_buf_sz,
                                PCI_DMA_FROMDEVICE);
 
-                       if (rtl8169_try_rx_copy(&skb, pkt_size, desc, dev)) {
+                       if (rtl8169_try_rx_copy(&skb, pkt_size, desc,
+                                               tp->rx_buf_sz, tp->align)) {
                                pci_action = pci_unmap_single;
                                tp->Rx_skbuff[entry] = NULL;
                        }
 
                        pci_action(tp->pci_dev, le64_to_cpu(desc->addr),
-                                  RX_BUF_SIZE, PCI_DMA_FROMDEVICE);
+                                  tp->rx_buf_sz, PCI_DMA_FROMDEVICE);
 
+                       skb->dev = dev;
                        skb_put(skb, pkt_size);
                        skb->protocol = eth_type_trans(skb, dev);
-                       netif_rx(skb);
+
+                       if (rtl8169_rx_vlan_skb(tp, desc, skb) < 0)
+                               rtl8169_rx_skb(skb);
 
                        dev->last_rx = jiffies;
                        tp->stats.rx_bytes += pkt_size;
                        tp->stats.rx_packets++;
                }
-               
-               cur_rx++; 
-               rx_left--;
        }
 
+       count = cur_rx - tp->cur_rx;
        tp->cur_rx = cur_rx;
 
        delta = rtl8169_rx_fill(tp, dev, tp->dirty_rx, tp->cur_rx);
-       if (delta > 0)
-               tp->dirty_rx += delta;
-       else if (delta < 0)
+       if (!delta && count && netif_msg_intr(tp))
                printk(KERN_INFO "%s: no Rx buffer allocated\n", dev->name);
+       tp->dirty_rx += delta;
 
        /*
         * FIXME: until there is periodic timer to try and refill the ring,
@@ -1520,19 +2599,21 @@ rtl8169_rx_interrupt(struct net_device *dev, struct rtl8169_private *tp,
         *   after refill ?
         * - how do others driver handle this condition (Uh oh...).
         */
-       if (tp->dirty_rx + NUM_RX_DESC == tp->cur_rx)
+       if ((tp->dirty_rx + NUM_RX_DESC == tp->cur_rx) && netif_msg_intr(tp))
                printk(KERN_EMERG "%s: Rx buffers exhausted\n", dev->name);
+
+       return count;
 }
 
 /* The interrupt handler does all of the Rx thread work and cleans up after the Tx thread. */
 static irqreturn_t
-rtl8169_interrupt(int irq, void *dev_instance, struct pt_regs *regs)
+rtl8169_interrupt(int irq, void *dev_instance)
 {
        struct net_device *dev = (struct net_device *) dev_instance;
-       struct rtl8169_private *tp = dev->priv;
+       struct rtl8169_private *tp = netdev_priv(dev);
        int boguscnt = max_interrupt_work;
-       void *ioaddr = tp->mmio_addr;
-       int status = 0;
+       void __iomem *ioaddr = tp->mmio_addr;
+       int status;
        int handled = 0;
 
        do {
@@ -1543,57 +2624,110 @@ rtl8169_interrupt(int irq, void *dev_instance, struct pt_regs *regs)
                        break;
 
                handled = 1;
-/*
-               if (status & RxUnderrun)
-                       link_changed = RTL_R16 (CSCR) & CSCR_LinkChangeBit;
-*/
+
+               if (unlikely(!netif_running(dev))) {
+                       rtl8169_asic_down(ioaddr);
+                       goto out;
+               }
+
+               status &= tp->intr_mask;
                RTL_W16(IntrStatus,
                        (status & RxFIFOOver) ? (status | RxOverflow) : status);
 
                if (!(status & rtl8169_intr_mask))
                        break;
 
-               // Rx interrupt 
-               if (status & (RxOK | RxUnderrun | RxOverflow | RxFIFOOver)) {
+               if (unlikely(status & SYSErr)) {
+                       rtl8169_pcierr_interrupt(dev);
+                       break;
+               }
+
+               if (status & LinkChg)
+                       rtl8169_check_link_status(dev, tp, ioaddr);
+
+#ifdef CONFIG_R8169_NAPI
+               RTL_W16(IntrMask, rtl8169_intr_mask & ~rtl8169_napi_event);
+               tp->intr_mask = ~rtl8169_napi_event;
+
+               if (likely(netif_rx_schedule_prep(dev)))
+                       __netif_rx_schedule(dev);
+               else if (netif_msg_intr(tp)) {
+                       printk(KERN_INFO "%s: interrupt %04x taken in poll\n",
+                              dev->name, status);
+               }
+               break;
+#else
+               /* Rx interrupt */
+               if (status & (RxOK | RxOverflow | RxFIFOOver)) {
                        rtl8169_rx_interrupt(dev, tp, ioaddr);
                }
-               // Tx interrupt
-               if (status & (TxOK | TxErr)) {
-                       spin_lock(&tp->lock);
+               /* Tx interrupt */
+               if (status & (TxOK | TxErr))
                        rtl8169_tx_interrupt(dev, tp, ioaddr);
-                       spin_unlock(&tp->lock);
-               }
+#endif
 
                boguscnt--;
        } while (boguscnt > 0);
 
        if (boguscnt <= 0) {
-               printk(KERN_WARNING "%s: Too much work at interrupt!\n",
-                      dev->name);
+               if (netif_msg_intr(tp) && net_ratelimit() ) {
+                       printk(KERN_WARNING
+                              "%s: Too much work at interrupt!\n", dev->name);
+               }
                /* Clear all interrupt sources. */
                RTL_W16(IntrStatus, 0xffff);
        }
+out:
        return IRQ_RETVAL(handled);
 }
 
-static int
-rtl8169_close(struct net_device *dev)
+#ifdef CONFIG_R8169_NAPI
+static int rtl8169_poll(struct net_device *dev, int *budget)
 {
-       struct rtl8169_private *tp = dev->priv;
-       struct pci_dev *pdev = tp->pci_dev;
-       void *ioaddr = tp->mmio_addr;
+       unsigned int work_done, work_to_do = min(*budget, dev->quota);
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+
+       work_done = rtl8169_rx_interrupt(dev, tp, ioaddr);
+       rtl8169_tx_interrupt(dev, tp, ioaddr);
+
+       *budget -= work_done;
+       dev->quota -= work_done;
+
+       if (work_done < work_to_do) {
+               netif_rx_complete(dev);
+               tp->intr_mask = 0xffff;
+               /*
+                * 20040426: the barrier is not strictly required but the
+                * behavior of the irq handler could be less predictable
+                * without it. Btw, the lack of flush for the posted pci
+                * write is safe - FR
+                */
+               smp_wmb();
+               RTL_W16(IntrMask, rtl8169_intr_mask);
+       }
 
-       netif_stop_queue(dev);
+       return (work_done >= work_to_do);
+}
+#endif
+
+static void rtl8169_down(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+       unsigned int poll_locked = 0;
+       unsigned int intrmask;
 
        rtl8169_delete_timer(dev);
 
-       spin_lock_irq(&tp->lock);
+       netif_stop_queue(dev);
 
-       /* Stop the chip's Tx and Rx DMA processes. */
-       RTL_W8(ChipCmd, 0x00);
+       flush_scheduled_work();
 
-       /* Disable interrupts by clearing the interrupt mask. */
-       RTL_W16(IntrMask, 0x0000);
+core_down:
+       spin_lock_irq(&tp->lock);
+
+       rtl8169_asic_down(ioaddr);
 
        /* Update the error counts. */
        tp->stats.rx_missed_errors += RTL_R32(RxMissed);
@@ -1602,11 +2736,47 @@ rtl8169_close(struct net_device *dev)
        spin_unlock_irq(&tp->lock);
 
        synchronize_irq(dev->irq);
-       free_irq(dev->irq, dev);
+
+       if (!poll_locked) {
+               netif_poll_disable(dev);
+               poll_locked++;
+       }
+
+       /* Give a racing hard_start_xmit a few cycles to complete. */
+       synchronize_sched();  /* FIXME: should this be synchronize_irq()? */
+
+       /*
+        * And now for the 50k$ question: are IRQ disabled or not ?
+        *
+        * Two paths lead here:
+        * 1) dev->close
+        *    -> netif_running() is available to sync the current code and the
+        *       IRQ handler. See rtl8169_interrupt for details.
+        * 2) dev->change_mtu
+        *    -> rtl8169_poll can not be issued again and re-enable the
+        *       interruptions. Let's simply issue the IRQ down sequence again.
+        *
+        * No loop if hotpluged or major error (0xffff).
+        */
+       intrmask = RTL_R16(IntrMask);
+       if (intrmask && (intrmask != 0xffff))
+               goto core_down;
 
        rtl8169_tx_clear(tp);
 
        rtl8169_rx_clear(tp);
+}
+
+static int rtl8169_close(struct net_device *dev)
+{
+       struct rtl8169_private *tp = netdev_priv(dev);
+       struct pci_dev *pdev = tp->pci_dev;
+
+       rtl8169_down(dev);
+
+       free_irq(dev->irq, dev);
+
+       netif_poll_enable(dev);
 
        pci_free_consistent(pdev, R8169_RX_RING_BYTES, tp->RxDescArray,
                            tp->RxPhyAddr);
@@ -1621,8 +2791,8 @@ rtl8169_close(struct net_device *dev)
 static void
 rtl8169_set_rx_mode(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
        unsigned long flags;
        u32 mc_filter[2];       /* Multicast hash filter */
        int i, rx_mode;
@@ -1630,8 +2800,10 @@ rtl8169_set_rx_mode(struct net_device *dev)
 
        if (dev->flags & IFF_PROMISC) {
                /* Unconditionally log net taps. */
-               printk(KERN_NOTICE "%s: Promiscuous mode enabled.\n",
-                      dev->name);
+               if (netif_msg_link(tp)) {
+                       printk(KERN_NOTICE "%s: Promiscuous mode enabled.\n",
+                              dev->name);
+               }
                rx_mode =
                    AcceptBroadcast | AcceptMulticast | AcceptMyPhys |
                    AcceptAllPhys;
@@ -1655,10 +2827,17 @@ rtl8169_set_rx_mode(struct net_device *dev)
 
        spin_lock_irqsave(&tp->lock, flags);
 
-       tmp =
-           rtl8169_rx_config | rx_mode | (RTL_R32(RxConfig) &
-                                          rtl_chip_info[tp->chipset].
-                                          RxConfigMask);
+       tmp = rtl8169_rx_config | rx_mode |
+             (RTL_R32(RxConfig) & rtl_chip_info[tp->chipset].RxConfigMask);
+
+       if ((tp->mac_version == RTL_GIGA_MAC_VER_11) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_12) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_13) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_14) ||
+           (tp->mac_version == RTL_GIGA_MAC_VER_15)) {
+               mc_filter[0] = 0xffffffff;
+               mc_filter[1] = 0xffffffff;
+       }
 
        RTL_W32(RxConfig, tmp);
        RTL_W32(MAR0 + 0, mc_filter[0]);
@@ -1675,8 +2854,8 @@ rtl8169_set_rx_mode(struct net_device *dev)
  */
 static struct net_device_stats *rtl8169_get_stats(struct net_device *dev)
 {
-       struct rtl8169_private *tp = dev->priv;
-       void *ioaddr = tp->mmio_addr;
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
        unsigned long flags;
 
        if (netif_running(dev)) {
@@ -1685,10 +2864,60 @@ static struct net_device_stats *rtl8169_get_stats(struct net_device *dev)
                RTL_W32(RxMissed, 0);
                spin_unlock_irqrestore(&tp->lock, flags);
        }
-               
+
        return &tp->stats;
 }
 
+#ifdef CONFIG_PM
+
+static int rtl8169_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+       struct rtl8169_private *tp = netdev_priv(dev);
+       void __iomem *ioaddr = tp->mmio_addr;
+
+       if (!netif_running(dev))
+               goto out;
+
+       netif_device_detach(dev);
+       netif_stop_queue(dev);
+
+       spin_lock_irq(&tp->lock);
+
+       rtl8169_asic_down(ioaddr);
+
+       tp->stats.rx_missed_errors += RTL_R32(RxMissed);
+       RTL_W32(RxMissed, 0);
+
+       spin_unlock_irq(&tp->lock);
+
+       pci_save_state(pdev);
+       pci_enable_wake(pdev, pci_choose_state(pdev, state), tp->wol_enabled);
+       pci_set_power_state(pdev, pci_choose_state(pdev, state));
+out:
+       return 0;
+}
+
+static int rtl8169_resume(struct pci_dev *pdev)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+
+       if (!netif_running(dev))
+               goto out;
+
+       netif_device_attach(dev);
+
+       pci_set_power_state(pdev, PCI_D0);
+       pci_restore_state(pdev);
+       pci_enable_wake(pdev, PCI_D0, 0);
+
+       rtl8169_schedule_work(dev, rtl8169_reset_task);
+out:
+       return 0;
+}
+
+#endif /* CONFIG_PM */
+
 static struct pci_driver rtl8169_pci_driver = {
        .name           = MODULENAME,
        .id_table       = rtl8169_pci_tbl,
@@ -1703,7 +2932,7 @@ static struct pci_driver rtl8169_pci_driver = {
 static int __init
 rtl8169_init_module(void)
 {
-       return pci_module_init(&rtl8169_pci_driver);
+       return pci_register_driver(&rtl8169_pci_driver);
 }
 
 static void __exit