Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / packetengines / yellowfin.c
diff --git a/kernel/drivers/net/ethernet/packetengines/yellowfin.c b/kernel/drivers/net/ethernet/packetengines/yellowfin.c
new file mode 100644 (file)
index 0000000..fa2db41
--- /dev/null
@@ -0,0 +1,1425 @@
+/* yellowfin.c: A Packet Engines G-NIC ethernet driver for linux. */
+/*
+       Written 1997-2001 by Donald Becker.
+
+       This software may be used and distributed according to the terms of
+       the GNU General Public License (GPL), incorporated herein by reference.
+       Drivers based on or derived from this code fall under the GPL and must
+       retain the authorship, copyright and license notice.  This file is not
+       a complete program and may only be used when the entire operating
+       system is licensed under the GPL.
+
+       This driver is for the Packet Engines G-NIC PCI Gigabit Ethernet adapter.
+       It also supports the Symbios Logic version of the same chip core.
+
+       The author may be reached as becker@scyld.com, or C/O
+       Scyld Computing Corporation
+       410 Severn Ave., Suite 210
+       Annapolis MD 21403
+
+       Support and updates available at
+       http://www.scyld.com/network/yellowfin.html
+       [link no longer provides useful info -jgarzik]
+
+*/
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#define DRV_NAME       "yellowfin"
+#define DRV_VERSION    "2.1"
+#define DRV_RELDATE    "Sep 11, 2006"
+
+/* The user-configurable values.
+   These may be modified when a driver module is loaded.*/
+
+static int debug = 1;                  /* 1 normal messages, 0 quiet .. 7 verbose. */
+/* Maximum events (Rx packets, etc.) to handle at each interrupt. */
+static int max_interrupt_work = 20;
+static int mtu;
+#ifdef YF_PROTOTYPE                    /* Support for prototype hardware errata. */
+/* System-wide count of bogus-rx frames. */
+static int bogus_rx;
+static int dma_ctrl = 0x004A0263;                      /* Constrained by errata */
+static int fifo_cfg = 0x0020;                          /* Bypass external Tx FIFO. */
+#elif defined(YF_NEW)                                  /* A future perfect board :->.  */
+static int dma_ctrl = 0x00CAC277;                      /* Override when loading module! */
+static int fifo_cfg = 0x0028;
+#else
+static const int dma_ctrl = 0x004A0263;                        /* Constrained by errata */
+static const int fifo_cfg = 0x0020;                            /* Bypass external Tx FIFO. */
+#endif
+
+/* Set the copy breakpoint for the copy-only-tiny-frames scheme.
+   Setting to > 1514 effectively disables this feature. */
+static int rx_copybreak;
+
+/* Used to pass the media type, etc.
+   No media types are currently defined.  These exist for driver
+   interoperability.
+*/
+#define MAX_UNITS 8                            /* More are supported, limit only on options */
+static int options[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
+static int full_duplex[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
+
+/* Do ugly workaround for GX server chipset errata. */
+static int gx_fix;
+
+/* Operational parameters that are set at compile time. */
+
+/* Keep the ring sizes a power of two for efficiency.
+   Making the Tx ring too long decreases the effectiveness of channel
+   bonding and packet priority.
+   There are no ill effects from too-large receive rings. */
+#define TX_RING_SIZE   16
+#define TX_QUEUE_SIZE  12              /* Must be > 4 && <= TX_RING_SIZE */
+#define RX_RING_SIZE   64
+#define STATUS_TOTAL_SIZE      TX_RING_SIZE*sizeof(struct tx_status_words)
+#define TX_TOTAL_SIZE          2*TX_RING_SIZE*sizeof(struct yellowfin_desc)
+#define RX_TOTAL_SIZE          RX_RING_SIZE*sizeof(struct yellowfin_desc)
+
+/* Operational parameters that usually are not changed. */
+/* Time in jiffies before concluding the transmitter is hung. */
+#define TX_TIMEOUT  (2*HZ)
+#define PKT_BUF_SZ             1536                    /* Size of each temporary Rx buffer.*/
+
+#define yellowfin_debug debug
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/timer.h>
+#include <linux/errno.h>
+#include <linux/ioport.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/init.h>
+#include <linux/mii.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/skbuff.h>
+#include <linux/ethtool.h>
+#include <linux/crc32.h>
+#include <linux/bitops.h>
+#include <asm/uaccess.h>
+#include <asm/processor.h>             /* Processor type for cache alignment. */
+#include <asm/unaligned.h>
+#include <asm/io.h>
+
+/* These identify the driver base version and may not be removed. */
+static const char version[] =
+  KERN_INFO DRV_NAME ".c:v1.05  1/09/2001  Written by Donald Becker <becker@scyld.com>\n"
+  "  (unofficial 2.4.x port, " DRV_VERSION ", " DRV_RELDATE ")\n";
+
+MODULE_AUTHOR("Donald Becker <becker@scyld.com>");
+MODULE_DESCRIPTION("Packet Engines Yellowfin G-NIC Gigabit Ethernet driver");
+MODULE_LICENSE("GPL");
+
+module_param(max_interrupt_work, int, 0);
+module_param(mtu, int, 0);
+module_param(debug, int, 0);
+module_param(rx_copybreak, int, 0);
+module_param_array(options, int, NULL, 0);
+module_param_array(full_duplex, int, NULL, 0);
+module_param(gx_fix, int, 0);
+MODULE_PARM_DESC(max_interrupt_work, "G-NIC maximum events handled per interrupt");
+MODULE_PARM_DESC(mtu, "G-NIC MTU (all boards)");
+MODULE_PARM_DESC(debug, "G-NIC debug level (0-7)");
+MODULE_PARM_DESC(rx_copybreak, "G-NIC copy breakpoint for copy-only-tiny-frames");
+MODULE_PARM_DESC(options, "G-NIC: Bits 0-3: media type, bit 17: full duplex");
+MODULE_PARM_DESC(full_duplex, "G-NIC full duplex setting(s) (1)");
+MODULE_PARM_DESC(gx_fix, "G-NIC: enable GX server chipset bug workaround (0-1)");
+
+/*
+                               Theory of Operation
+
+I. Board Compatibility
+
+This device driver is designed for the Packet Engines "Yellowfin" Gigabit
+Ethernet adapter.  The G-NIC 64-bit PCI card is supported, as well as the
+Symbios 53C885E dual function chip.
+
+II. Board-specific settings
+
+PCI bus devices are configured by the system at boot time, so no jumpers
+need to be set on the board.  The system BIOS preferably should assign the
+PCI INTA signal to an otherwise unused system IRQ line.
+Note: Kernel versions earlier than 1.3.73 do not support shared PCI
+interrupt lines.
+
+III. Driver operation
+
+IIIa. Ring buffers
+
+The Yellowfin uses the Descriptor Based DMA Architecture specified by Apple.
+This is a descriptor list scheme similar to that used by the EEPro100 and
+Tulip.  This driver uses two statically allocated fixed-size descriptor lists
+formed into rings by a branch from the final descriptor to the beginning of
+the list.  The ring sizes are set at compile time by RX/TX_RING_SIZE.
+
+The driver allocates full frame size skbuffs for the Rx ring buffers at
+open() time and passes the skb->data field to the Yellowfin as receive data
+buffers.  When an incoming frame is less than RX_COPYBREAK bytes long,
+a fresh skbuff is allocated and the frame is copied to the new skbuff.
+When the incoming frame is larger, the skbuff is passed directly up the
+protocol stack and replaced by a newly allocated skbuff.
+
+The RX_COPYBREAK value is chosen to trade-off the memory wasted by
+using a full-sized skbuff for small frames vs. the copying costs of larger
+frames.  For small frames the copying cost is negligible (esp. considering
+that we are pre-loading the cache with immediately useful header
+information).  For large frames the copying cost is non-trivial, and the
+larger copy might flush the cache of useful data.
+
+IIIC. Synchronization
+
+The driver runs as two independent, single-threaded flows of control.  One
+is the send-packet routine, which enforces single-threaded use by the
+dev->tbusy flag.  The other thread is the interrupt handler, which is single
+threaded by the hardware and other software.
+
+The send packet thread has partial control over the Tx ring and 'dev->tbusy'
+flag.  It sets the tbusy flag whenever it's queuing a Tx packet. If the next
+queue slot is empty, it clears the tbusy flag when finished otherwise it sets
+the 'yp->tx_full' flag.
+
+The interrupt handler has exclusive control over the Rx ring and records stats
+from the Tx ring.  After reaping the stats, it marks the Tx queue entry as
+empty by incrementing the dirty_tx mark. Iff the 'yp->tx_full' flag is set, it
+clears both the tx_full and tbusy flags.
+
+IV. Notes
+
+Thanks to Kim Stearns of Packet Engines for providing a pair of G-NIC boards.
+Thanks to Bruce Faust of Digitalscape for providing both their SYM53C885 board
+and an AlphaStation to verifty the Alpha port!
+
+IVb. References
+
+Yellowfin Engineering Design Specification, 4/23/97 Preliminary/Confidential
+Symbios SYM53C885 PCI-SCSI/Fast Ethernet Multifunction Controller Preliminary
+   Data Manual v3.0
+http://cesdis.gsfc.nasa.gov/linux/misc/NWay.html
+http://cesdis.gsfc.nasa.gov/linux/misc/100mbps.html
+
+IVc. Errata
+
+See Packet Engines confidential appendix (prototype chips only).
+*/
+
+
+
+enum capability_flags {
+       HasMII=1, FullTxStatus=2, IsGigabit=4, HasMulticastBug=8, FullRxStatus=16,
+       HasMACAddrBug=32, /* Only on early revs.  */
+       DontUseEeprom=64, /* Don't read the MAC from the EEPROm. */
+};
+
+/* The PCI I/O space extent. */
+enum {
+       YELLOWFIN_SIZE  = 0x100,
+};
+
+struct pci_id_info {
+        const char *name;
+        struct match_info {
+                int     pci, pci_mask, subsystem, subsystem_mask;
+                int revision, revision_mask;                            /* Only 8 bits. */
+        } id;
+        int drv_flags;                          /* Driver use, intended as capability flags. */
+};
+
+static const struct pci_id_info pci_id_tbl[] = {
+       {"Yellowfin G-NIC Gigabit Ethernet", { 0x07021000, 0xffffffff},
+        FullTxStatus | IsGigabit | HasMulticastBug | HasMACAddrBug | DontUseEeprom},
+       {"Symbios SYM83C885", { 0x07011000, 0xffffffff},
+         HasMII | DontUseEeprom },
+       { }
+};
+
+static const struct pci_device_id yellowfin_pci_tbl[] = {
+       { 0x1000, 0x0702, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
+       { 0x1000, 0x0701, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1 },
+       { }
+};
+MODULE_DEVICE_TABLE (pci, yellowfin_pci_tbl);
+
+
+/* Offsets to the Yellowfin registers.  Various sizes and alignments. */
+enum yellowfin_offsets {
+       TxCtrl=0x00, TxStatus=0x04, TxPtr=0x0C,
+       TxIntrSel=0x10, TxBranchSel=0x14, TxWaitSel=0x18,
+       RxCtrl=0x40, RxStatus=0x44, RxPtr=0x4C,
+       RxIntrSel=0x50, RxBranchSel=0x54, RxWaitSel=0x58,
+       EventStatus=0x80, IntrEnb=0x82, IntrClear=0x84, IntrStatus=0x86,
+       ChipRev=0x8C, DMACtrl=0x90, TxThreshold=0x94,
+       Cnfg=0xA0, FrameGap0=0xA2, FrameGap1=0xA4,
+       MII_Cmd=0xA6, MII_Addr=0xA8, MII_Wr_Data=0xAA, MII_Rd_Data=0xAC,
+       MII_Status=0xAE,
+       RxDepth=0xB8, FlowCtrl=0xBC,
+       AddrMode=0xD0, StnAddr=0xD2, HashTbl=0xD8, FIFOcfg=0xF8,
+       EEStatus=0xF0, EECtrl=0xF1, EEAddr=0xF2, EERead=0xF3, EEWrite=0xF4,
+       EEFeature=0xF5,
+};
+
+/* The Yellowfin Rx and Tx buffer descriptors.
+   Elements are written as 32 bit for endian portability. */
+struct yellowfin_desc {
+       __le32 dbdma_cmd;
+       __le32 addr;
+       __le32 branch_addr;
+       __le32 result_status;
+};
+
+struct tx_status_words {
+#ifdef __BIG_ENDIAN
+       u16 tx_errs;
+       u16 tx_cnt;
+       u16 paused;
+       u16 total_tx_cnt;
+#else  /* Little endian chips. */
+       u16 tx_cnt;
+       u16 tx_errs;
+       u16 total_tx_cnt;
+       u16 paused;
+#endif /* __BIG_ENDIAN */
+};
+
+/* Bits in yellowfin_desc.cmd */
+enum desc_cmd_bits {
+       CMD_TX_PKT=0x10000000, CMD_RX_BUF=0x20000000, CMD_TXSTATUS=0x30000000,
+       CMD_NOP=0x60000000, CMD_STOP=0x70000000,
+       BRANCH_ALWAYS=0x0C0000, INTR_ALWAYS=0x300000, WAIT_ALWAYS=0x030000,
+       BRANCH_IFTRUE=0x040000,
+};
+
+/* Bits in yellowfin_desc.status */
+enum desc_status_bits { RX_EOP=0x0040, };
+
+/* Bits in the interrupt status/mask registers. */
+enum intr_status_bits {
+       IntrRxDone=0x01, IntrRxInvalid=0x02, IntrRxPCIFault=0x04,IntrRxPCIErr=0x08,
+       IntrTxDone=0x10, IntrTxInvalid=0x20, IntrTxPCIFault=0x40,IntrTxPCIErr=0x80,
+       IntrEarlyRx=0x100, IntrWakeup=0x200, };
+
+#define PRIV_ALIGN     31      /* Required alignment mask */
+#define MII_CNT                4
+struct yellowfin_private {
+       /* Descriptor rings first for alignment.
+          Tx requires a second descriptor for status. */
+       struct yellowfin_desc *rx_ring;
+       struct yellowfin_desc *tx_ring;
+       struct sk_buff* rx_skbuff[RX_RING_SIZE];
+       struct sk_buff* tx_skbuff[TX_RING_SIZE];
+       dma_addr_t rx_ring_dma;
+       dma_addr_t tx_ring_dma;
+
+       struct tx_status_words *tx_status;
+       dma_addr_t tx_status_dma;
+
+       struct timer_list timer;        /* Media selection timer. */
+       /* Frequently used and paired value: keep adjacent for cache effect. */
+       int chip_id, drv_flags;
+       struct pci_dev *pci_dev;
+       unsigned int cur_rx, dirty_rx;          /* Producer/consumer ring indices */
+       unsigned int rx_buf_sz;                         /* Based on MTU+slack. */
+       struct tx_status_words *tx_tail_desc;
+       unsigned int cur_tx, dirty_tx;
+       int tx_threshold;
+       unsigned int tx_full:1;                         /* The Tx queue is full. */
+       unsigned int full_duplex:1;                     /* Full-duplex operation requested. */
+       unsigned int duplex_lock:1;
+       unsigned int medialock:1;                       /* Do not sense media. */
+       unsigned int default_port:4;            /* Last dev->if_port value. */
+       /* MII transceiver section. */
+       int mii_cnt;                                            /* MII device addresses. */
+       u16 advertising;                                        /* NWay media advertisement */
+       unsigned char phys[MII_CNT];            /* MII device addresses, only first one used */
+       spinlock_t lock;
+       void __iomem *base;
+};
+
+static int read_eeprom(void __iomem *ioaddr, int location);
+static int mdio_read(void __iomem *ioaddr, int phy_id, int location);
+static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value);
+static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
+static int yellowfin_open(struct net_device *dev);
+static void yellowfin_timer(unsigned long data);
+static void yellowfin_tx_timeout(struct net_device *dev);
+static int yellowfin_init_ring(struct net_device *dev);
+static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb,
+                                       struct net_device *dev);
+static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance);
+static int yellowfin_rx(struct net_device *dev);
+static void yellowfin_error(struct net_device *dev, int intr_status);
+static int yellowfin_close(struct net_device *dev);
+static void set_rx_mode(struct net_device *dev);
+static const struct ethtool_ops ethtool_ops;
+
+static const struct net_device_ops netdev_ops = {
+       .ndo_open               = yellowfin_open,
+       .ndo_stop               = yellowfin_close,
+       .ndo_start_xmit         = yellowfin_start_xmit,
+       .ndo_set_rx_mode        = set_rx_mode,
+       .ndo_change_mtu         = eth_change_mtu,
+       .ndo_validate_addr      = eth_validate_addr,
+       .ndo_set_mac_address    = eth_mac_addr,
+       .ndo_do_ioctl           = netdev_ioctl,
+       .ndo_tx_timeout         = yellowfin_tx_timeout,
+};
+
+static int yellowfin_init_one(struct pci_dev *pdev,
+                             const struct pci_device_id *ent)
+{
+       struct net_device *dev;
+       struct yellowfin_private *np;
+       int irq;
+       int chip_idx = ent->driver_data;
+       static int find_cnt;
+       void __iomem *ioaddr;
+       int i, option = find_cnt < MAX_UNITS ? options[find_cnt] : 0;
+       int drv_flags = pci_id_tbl[chip_idx].drv_flags;
+        void *ring_space;
+        dma_addr_t ring_dma;
+#ifdef USE_IO_OPS
+       int bar = 0;
+#else
+       int bar = 1;
+#endif
+
+/* when built into the kernel, we only print version if device is found */
+#ifndef MODULE
+       static int printed_version;
+       if (!printed_version++)
+               printk(version);
+#endif
+
+       i = pci_enable_device(pdev);
+       if (i) return i;
+
+       dev = alloc_etherdev(sizeof(*np));
+       if (!dev)
+               return -ENOMEM;
+
+       SET_NETDEV_DEV(dev, &pdev->dev);
+
+       np = netdev_priv(dev);
+
+       if (pci_request_regions(pdev, DRV_NAME))
+               goto err_out_free_netdev;
+
+       pci_set_master (pdev);
+
+       ioaddr = pci_iomap(pdev, bar, YELLOWFIN_SIZE);
+       if (!ioaddr)
+               goto err_out_free_res;
+
+       irq = pdev->irq;
+
+       if (drv_flags & DontUseEeprom)
+               for (i = 0; i < 6; i++)
+                       dev->dev_addr[i] = ioread8(ioaddr + StnAddr + i);
+       else {
+               int ee_offset = (read_eeprom(ioaddr, 6) == 0xff ? 0x100 : 0);
+               for (i = 0; i < 6; i++)
+                       dev->dev_addr[i] = read_eeprom(ioaddr, ee_offset + i);
+       }
+
+       /* Reset the chip. */
+       iowrite32(0x80000000, ioaddr + DMACtrl);
+
+       pci_set_drvdata(pdev, dev);
+       spin_lock_init(&np->lock);
+
+       np->pci_dev = pdev;
+       np->chip_id = chip_idx;
+       np->drv_flags = drv_flags;
+       np->base = ioaddr;
+
+       ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma);
+       if (!ring_space)
+               goto err_out_cleardev;
+       np->tx_ring = ring_space;
+       np->tx_ring_dma = ring_dma;
+
+       ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma);
+       if (!ring_space)
+               goto err_out_unmap_tx;
+       np->rx_ring = ring_space;
+       np->rx_ring_dma = ring_dma;
+
+       ring_space = pci_alloc_consistent(pdev, STATUS_TOTAL_SIZE, &ring_dma);
+       if (!ring_space)
+               goto err_out_unmap_rx;
+       np->tx_status = ring_space;
+       np->tx_status_dma = ring_dma;
+
+       if (dev->mem_start)
+               option = dev->mem_start;
+
+       /* The lower four bits are the media type. */
+       if (option > 0) {
+               if (option & 0x200)
+                       np->full_duplex = 1;
+               np->default_port = option & 15;
+               if (np->default_port)
+                       np->medialock = 1;
+       }
+       if (find_cnt < MAX_UNITS  &&  full_duplex[find_cnt] > 0)
+               np->full_duplex = 1;
+
+       if (np->full_duplex)
+               np->duplex_lock = 1;
+
+       /* The Yellowfin-specific entries in the device structure. */
+       dev->netdev_ops = &netdev_ops;
+       dev->ethtool_ops = &ethtool_ops;
+       dev->watchdog_timeo = TX_TIMEOUT;
+
+       if (mtu)
+               dev->mtu = mtu;
+
+       i = register_netdev(dev);
+       if (i)
+               goto err_out_unmap_status;
+
+       netdev_info(dev, "%s type %8x at %p, %pM, IRQ %d\n",
+                   pci_id_tbl[chip_idx].name,
+                   ioread32(ioaddr + ChipRev), ioaddr,
+                   dev->dev_addr, irq);
+
+       if (np->drv_flags & HasMII) {
+               int phy, phy_idx = 0;
+               for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) {
+                       int mii_status = mdio_read(ioaddr, phy, 1);
+                       if (mii_status != 0xffff  &&  mii_status != 0x0000) {
+                               np->phys[phy_idx++] = phy;
+                               np->advertising = mdio_read(ioaddr, phy, 4);
+                               netdev_info(dev, "MII PHY found at address %d, status 0x%04x advertising %04x\n",
+                                           phy, mii_status, np->advertising);
+                       }
+               }
+               np->mii_cnt = phy_idx;
+       }
+
+       find_cnt++;
+
+       return 0;
+
+err_out_unmap_status:
+        pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status,
+               np->tx_status_dma);
+err_out_unmap_rx:
+        pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma);
+err_out_unmap_tx:
+        pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma);
+err_out_cleardev:
+       pci_iounmap(pdev, ioaddr);
+err_out_free_res:
+       pci_release_regions(pdev);
+err_out_free_netdev:
+       free_netdev (dev);
+       return -ENODEV;
+}
+
+static int read_eeprom(void __iomem *ioaddr, int location)
+{
+       int bogus_cnt = 10000;          /* Typical 33Mhz: 1050 ticks */
+
+       iowrite8(location, ioaddr + EEAddr);
+       iowrite8(0x30 | ((location >> 8) & 7), ioaddr + EECtrl);
+       while ((ioread8(ioaddr + EEStatus) & 0x80)  &&  --bogus_cnt > 0)
+               ;
+       return ioread8(ioaddr + EERead);
+}
+
+/* MII Managemen Data I/O accesses.
+   These routines assume the MDIO controller is idle, and do not exit until
+   the command is finished. */
+
+static int mdio_read(void __iomem *ioaddr, int phy_id, int location)
+{
+       int i;
+
+       iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
+       iowrite16(1, ioaddr + MII_Cmd);
+       for (i = 10000; i >= 0; i--)
+               if ((ioread16(ioaddr + MII_Status) & 1) == 0)
+                       break;
+       return ioread16(ioaddr + MII_Rd_Data);
+}
+
+static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value)
+{
+       int i;
+
+       iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
+       iowrite16(value, ioaddr + MII_Wr_Data);
+
+       /* Wait for the command to finish. */
+       for (i = 10000; i >= 0; i--)
+               if ((ioread16(ioaddr + MII_Status) & 1) == 0)
+                       break;
+}
+
+
+static int yellowfin_open(struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       const int irq = yp->pci_dev->irq;
+       void __iomem *ioaddr = yp->base;
+       int i, rc;
+
+       /* Reset the chip. */
+       iowrite32(0x80000000, ioaddr + DMACtrl);
+
+       rc = request_irq(irq, yellowfin_interrupt, IRQF_SHARED, dev->name, dev);
+       if (rc)
+               return rc;
+
+       rc = yellowfin_init_ring(dev);
+       if (rc < 0)
+               goto err_free_irq;
+
+       iowrite32(yp->rx_ring_dma, ioaddr + RxPtr);
+       iowrite32(yp->tx_ring_dma, ioaddr + TxPtr);
+
+       for (i = 0; i < 6; i++)
+               iowrite8(dev->dev_addr[i], ioaddr + StnAddr + i);
+
+       /* Set up various condition 'select' registers.
+          There are no options here. */
+       iowrite32(0x00800080, ioaddr + TxIntrSel);      /* Interrupt on Tx abort */
+       iowrite32(0x00800080, ioaddr + TxBranchSel);    /* Branch on Tx abort */
+       iowrite32(0x00400040, ioaddr + TxWaitSel);      /* Wait on Tx status */
+       iowrite32(0x00400040, ioaddr + RxIntrSel);      /* Interrupt on Rx done */
+       iowrite32(0x00400040, ioaddr + RxBranchSel);    /* Branch on Rx error */
+       iowrite32(0x00400040, ioaddr + RxWaitSel);      /* Wait on Rx done */
+
+       /* Initialize other registers: with so many this eventually this will
+          converted to an offset/value list. */
+       iowrite32(dma_ctrl, ioaddr + DMACtrl);
+       iowrite16(fifo_cfg, ioaddr + FIFOcfg);
+       /* Enable automatic generation of flow control frames, period 0xffff. */
+       iowrite32(0x0030FFFF, ioaddr + FlowCtrl);
+
+       yp->tx_threshold = 32;
+       iowrite32(yp->tx_threshold, ioaddr + TxThreshold);
+
+       if (dev->if_port == 0)
+               dev->if_port = yp->default_port;
+
+       netif_start_queue(dev);
+
+       /* Setting the Rx mode will start the Rx process. */
+       if (yp->drv_flags & IsGigabit) {
+               /* We are always in full-duplex mode with gigabit! */
+               yp->full_duplex = 1;
+               iowrite16(0x01CF, ioaddr + Cnfg);
+       } else {
+               iowrite16(0x0018, ioaddr + FrameGap0); /* 0060/4060 for non-MII 10baseT */
+               iowrite16(0x1018, ioaddr + FrameGap1);
+               iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
+       }
+       set_rx_mode(dev);
+
+       /* Enable interrupts by setting the interrupt mask. */
+       iowrite16(0x81ff, ioaddr + IntrEnb);                    /* See enum intr_status_bits */
+       iowrite16(0x0000, ioaddr + EventStatus);                /* Clear non-interrupting events */
+       iowrite32(0x80008000, ioaddr + RxCtrl);         /* Start Rx and Tx channels. */
+       iowrite32(0x80008000, ioaddr + TxCtrl);
+
+       if (yellowfin_debug > 2) {
+               netdev_printk(KERN_DEBUG, dev, "Done %s()\n", __func__);
+       }
+
+       /* Set the timer to check for link beat. */
+       init_timer(&yp->timer);
+       yp->timer.expires = jiffies + 3*HZ;
+       yp->timer.data = (unsigned long)dev;
+       yp->timer.function = yellowfin_timer;                           /* timer handler */
+       add_timer(&yp->timer);
+out:
+       return rc;
+
+err_free_irq:
+       free_irq(irq, dev);
+       goto out;
+}
+
+static void yellowfin_timer(unsigned long data)
+{
+       struct net_device *dev = (struct net_device *)data;
+       struct yellowfin_private *yp = netdev_priv(dev);
+       void __iomem *ioaddr = yp->base;
+       int next_tick = 60*HZ;
+
+       if (yellowfin_debug > 3) {
+               netdev_printk(KERN_DEBUG, dev, "Yellowfin timer tick, status %08x\n",
+                             ioread16(ioaddr + IntrStatus));
+       }
+
+       if (yp->mii_cnt) {
+               int bmsr = mdio_read(ioaddr, yp->phys[0], MII_BMSR);
+               int lpa = mdio_read(ioaddr, yp->phys[0], MII_LPA);
+               int negotiated = lpa & yp->advertising;
+               if (yellowfin_debug > 1)
+                       netdev_printk(KERN_DEBUG, dev, "MII #%d status register is %04x, link partner capability %04x\n",
+                                     yp->phys[0], bmsr, lpa);
+
+               yp->full_duplex = mii_duplex(yp->duplex_lock, negotiated);
+
+               iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
+
+               if (bmsr & BMSR_LSTATUS)
+                       next_tick = 60*HZ;
+               else
+                       next_tick = 3*HZ;
+       }
+
+       yp->timer.expires = jiffies + next_tick;
+       add_timer(&yp->timer);
+}
+
+static void yellowfin_tx_timeout(struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       void __iomem *ioaddr = yp->base;
+
+       netdev_warn(dev, "Yellowfin transmit timed out at %d/%d Tx status %04x, Rx status %04x, resetting...\n",
+                   yp->cur_tx, yp->dirty_tx,
+                   ioread32(ioaddr + TxStatus),
+                   ioread32(ioaddr + RxStatus));
+
+       /* Note: these should be KERN_DEBUG. */
+       if (yellowfin_debug) {
+               int i;
+               pr_warn("  Rx ring %p: ", yp->rx_ring);
+               for (i = 0; i < RX_RING_SIZE; i++)
+                       pr_cont(" %08x", yp->rx_ring[i].result_status);
+               pr_cont("\n");
+               pr_warn("  Tx ring %p: ", yp->tx_ring);
+               for (i = 0; i < TX_RING_SIZE; i++)
+                       pr_cont(" %04x /%08x",
+                              yp->tx_status[i].tx_errs,
+                              yp->tx_ring[i].result_status);
+               pr_cont("\n");
+       }
+
+       /* If the hardware is found to hang regularly, we will update the code
+          to reinitialize the chip here. */
+       dev->if_port = 0;
+
+       /* Wake the potentially-idle transmit channel. */
+       iowrite32(0x10001000, yp->base + TxCtrl);
+       if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
+               netif_wake_queue (dev);         /* Typical path */
+
+       dev->trans_start = jiffies; /* prevent tx timeout */
+       dev->stats.tx_errors++;
+}
+
+/* Initialize the Rx and Tx rings, along with various 'dev' bits. */
+static int yellowfin_init_ring(struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       int i, j;
+
+       yp->tx_full = 0;
+       yp->cur_rx = yp->cur_tx = 0;
+       yp->dirty_tx = 0;
+
+       yp->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
+
+       for (i = 0; i < RX_RING_SIZE; i++) {
+               yp->rx_ring[i].dbdma_cmd =
+                       cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
+               yp->rx_ring[i].branch_addr = cpu_to_le32(yp->rx_ring_dma +
+                       ((i+1)%RX_RING_SIZE)*sizeof(struct yellowfin_desc));
+       }
+
+       for (i = 0; i < RX_RING_SIZE; i++) {
+               struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
+               yp->rx_skbuff[i] = skb;
+               if (skb == NULL)
+                       break;
+               skb_reserve(skb, 2);    /* 16 byte align the IP header. */
+               yp->rx_ring[i].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
+                       skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE));
+       }
+       if (i != RX_RING_SIZE) {
+               for (j = 0; j < i; j++)
+                       dev_kfree_skb(yp->rx_skbuff[j]);
+               return -ENOMEM;
+       }
+       yp->rx_ring[i-1].dbdma_cmd = cpu_to_le32(CMD_STOP);
+       yp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
+
+#define NO_TXSTATS
+#ifdef NO_TXSTATS
+       /* In this mode the Tx ring needs only a single descriptor. */
+       for (i = 0; i < TX_RING_SIZE; i++) {
+               yp->tx_skbuff[i] = NULL;
+               yp->tx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
+               yp->tx_ring[i].branch_addr = cpu_to_le32(yp->tx_ring_dma +
+                       ((i+1)%TX_RING_SIZE)*sizeof(struct yellowfin_desc));
+       }
+       /* Wrap ring */
+       yp->tx_ring[--i].dbdma_cmd = cpu_to_le32(CMD_STOP | BRANCH_ALWAYS);
+#else
+{
+       /* Tx ring needs a pair of descriptors, the second for the status. */
+       for (i = 0; i < TX_RING_SIZE; i++) {
+               j = 2*i;
+               yp->tx_skbuff[i] = 0;
+               /* Branch on Tx error. */
+               yp->tx_ring[j].dbdma_cmd = cpu_to_le32(CMD_STOP);
+               yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
+                       (j+1)*sizeof(struct yellowfin_desc));
+               j++;
+               if (yp->flags & FullTxStatus) {
+                       yp->tx_ring[j].dbdma_cmd =
+                               cpu_to_le32(CMD_TXSTATUS | sizeof(*yp->tx_status));
+                       yp->tx_ring[j].request_cnt = sizeof(*yp->tx_status);
+                       yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
+                               i*sizeof(struct tx_status_words));
+               } else {
+                       /* Symbios chips write only tx_errs word. */
+                       yp->tx_ring[j].dbdma_cmd =
+                               cpu_to_le32(CMD_TXSTATUS | INTR_ALWAYS | 2);
+                       yp->tx_ring[j].request_cnt = 2;
+                       /* Om pade ummmmm... */
+                       yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
+                               i*sizeof(struct tx_status_words) +
+                               &(yp->tx_status[0].tx_errs) -
+                               &(yp->tx_status[0]));
+               }
+               yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
+                       ((j+1)%(2*TX_RING_SIZE))*sizeof(struct yellowfin_desc));
+       }
+       /* Wrap ring */
+       yp->tx_ring[++j].dbdma_cmd |= cpu_to_le32(BRANCH_ALWAYS | INTR_ALWAYS);
+}
+#endif
+       yp->tx_tail_desc = &yp->tx_status[0];
+       return 0;
+}
+
+static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb,
+                                       struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       unsigned entry;
+       int len = skb->len;
+
+       netif_stop_queue (dev);
+
+       /* Note: Ordering is important here, set the field with the
+          "ownership" bit last, and only then increment cur_tx. */
+
+       /* Calculate the next Tx descriptor entry. */
+       entry = yp->cur_tx % TX_RING_SIZE;
+
+       if (gx_fix) {   /* Note: only works for paddable protocols e.g.  IP. */
+               int cacheline_end = ((unsigned long)skb->data + skb->len) % 32;
+               /* Fix GX chipset errata. */
+               if (cacheline_end > 24  || cacheline_end == 0) {
+                       len = skb->len + 32 - cacheline_end + 1;
+                       if (skb_padto(skb, len)) {
+                               yp->tx_skbuff[entry] = NULL;
+                               netif_wake_queue(dev);
+                               return NETDEV_TX_OK;
+                       }
+               }
+       }
+       yp->tx_skbuff[entry] = skb;
+
+#ifdef NO_TXSTATS
+       yp->tx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
+               skb->data, len, PCI_DMA_TODEVICE));
+       yp->tx_ring[entry].result_status = 0;
+       if (entry >= TX_RING_SIZE-1) {
+               /* New stop command. */
+               yp->tx_ring[0].dbdma_cmd = cpu_to_le32(CMD_STOP);
+               yp->tx_ring[TX_RING_SIZE-1].dbdma_cmd =
+                       cpu_to_le32(CMD_TX_PKT|BRANCH_ALWAYS | len);
+       } else {
+               yp->tx_ring[entry+1].dbdma_cmd = cpu_to_le32(CMD_STOP);
+               yp->tx_ring[entry].dbdma_cmd =
+                       cpu_to_le32(CMD_TX_PKT | BRANCH_IFTRUE | len);
+       }
+       yp->cur_tx++;
+#else
+       yp->tx_ring[entry<<1].request_cnt = len;
+       yp->tx_ring[entry<<1].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
+               skb->data, len, PCI_DMA_TODEVICE));
+       /* The input_last (status-write) command is constant, but we must
+          rewrite the subsequent 'stop' command. */
+
+       yp->cur_tx++;
+       {
+               unsigned next_entry = yp->cur_tx % TX_RING_SIZE;
+               yp->tx_ring[next_entry<<1].dbdma_cmd = cpu_to_le32(CMD_STOP);
+       }
+       /* Final step -- overwrite the old 'stop' command. */
+
+       yp->tx_ring[entry<<1].dbdma_cmd =
+               cpu_to_le32( ((entry % 6) == 0 ? CMD_TX_PKT|INTR_ALWAYS|BRANCH_IFTRUE :
+                                         CMD_TX_PKT | BRANCH_IFTRUE) | len);
+#endif
+
+       /* Non-x86 Todo: explicitly flush cache lines here. */
+
+       /* Wake the potentially-idle transmit channel. */
+       iowrite32(0x10001000, yp->base + TxCtrl);
+
+       if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
+               netif_start_queue (dev);                /* Typical path */
+       else
+               yp->tx_full = 1;
+
+       if (yellowfin_debug > 4) {
+               netdev_printk(KERN_DEBUG, dev, "Yellowfin transmit frame #%d queued in slot %d\n",
+                             yp->cur_tx, entry);
+       }
+       return NETDEV_TX_OK;
+}
+
+/* The interrupt handler does all of the Rx thread work and cleans up
+   after the Tx thread. */
+static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance)
+{
+       struct net_device *dev = dev_instance;
+       struct yellowfin_private *yp;
+       void __iomem *ioaddr;
+       int boguscnt = max_interrupt_work;
+       unsigned int handled = 0;
+
+       yp = netdev_priv(dev);
+       ioaddr = yp->base;
+
+       spin_lock (&yp->lock);
+
+       do {
+               u16 intr_status = ioread16(ioaddr + IntrClear);
+
+               if (yellowfin_debug > 4)
+                       netdev_printk(KERN_DEBUG, dev, "Yellowfin interrupt, status %04x\n",
+                                     intr_status);
+
+               if (intr_status == 0)
+                       break;
+               handled = 1;
+
+               if (intr_status & (IntrRxDone | IntrEarlyRx)) {
+                       yellowfin_rx(dev);
+                       iowrite32(0x10001000, ioaddr + RxCtrl);         /* Wake Rx engine. */
+               }
+
+#ifdef NO_TXSTATS
+               for (; yp->cur_tx - yp->dirty_tx > 0; yp->dirty_tx++) {
+                       int entry = yp->dirty_tx % TX_RING_SIZE;
+                       struct sk_buff *skb;
+
+                       if (yp->tx_ring[entry].result_status == 0)
+                               break;
+                       skb = yp->tx_skbuff[entry];
+                       dev->stats.tx_packets++;
+                       dev->stats.tx_bytes += skb->len;
+                       /* Free the original skb. */
+                       pci_unmap_single(yp->pci_dev, le32_to_cpu(yp->tx_ring[entry].addr),
+                               skb->len, PCI_DMA_TODEVICE);
+                       dev_kfree_skb_irq(skb);
+                       yp->tx_skbuff[entry] = NULL;
+               }
+               if (yp->tx_full &&
+                   yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE - 4) {
+                       /* The ring is no longer full, clear tbusy. */
+                       yp->tx_full = 0;
+                       netif_wake_queue(dev);
+               }
+#else
+               if ((intr_status & IntrTxDone) || (yp->tx_tail_desc->tx_errs)) {
+                       unsigned dirty_tx = yp->dirty_tx;
+
+                       for (dirty_tx = yp->dirty_tx; yp->cur_tx - dirty_tx > 0;
+                                dirty_tx++) {
+                               /* Todo: optimize this. */
+                               int entry = dirty_tx % TX_RING_SIZE;
+                               u16 tx_errs = yp->tx_status[entry].tx_errs;
+                               struct sk_buff *skb;
+
+#ifndef final_version
+                               if (yellowfin_debug > 5)
+                                       netdev_printk(KERN_DEBUG, dev, "Tx queue %d check, Tx status %04x %04x %04x %04x\n",
+                                                     entry,
+                                                     yp->tx_status[entry].tx_cnt,
+                                                     yp->tx_status[entry].tx_errs,
+                                                     yp->tx_status[entry].total_tx_cnt,
+                                                     yp->tx_status[entry].paused);
+#endif
+                               if (tx_errs == 0)
+                                       break;  /* It still hasn't been Txed */
+                               skb = yp->tx_skbuff[entry];
+                               if (tx_errs & 0xF810) {
+                                       /* There was an major error, log it. */
+#ifndef final_version
+                                       if (yellowfin_debug > 1)
+                                               netdev_printk(KERN_DEBUG, dev, "Transmit error, Tx status %04x\n",
+                                                             tx_errs);
+#endif
+                                       dev->stats.tx_errors++;
+                                       if (tx_errs & 0xF800) dev->stats.tx_aborted_errors++;
+                                       if (tx_errs & 0x0800) dev->stats.tx_carrier_errors++;
+                                       if (tx_errs & 0x2000) dev->stats.tx_window_errors++;
+                                       if (tx_errs & 0x8000) dev->stats.tx_fifo_errors++;
+                               } else {
+#ifndef final_version
+                                       if (yellowfin_debug > 4)
+                                               netdev_printk(KERN_DEBUG, dev, "Normal transmit, Tx status %04x\n",
+                                                             tx_errs);
+#endif
+                                       dev->stats.tx_bytes += skb->len;
+                                       dev->stats.collisions += tx_errs & 15;
+                                       dev->stats.tx_packets++;
+                               }
+                               /* Free the original skb. */
+                               pci_unmap_single(yp->pci_dev,
+                                       yp->tx_ring[entry<<1].addr, skb->len,
+                                       PCI_DMA_TODEVICE);
+                               dev_kfree_skb_irq(skb);
+                               yp->tx_skbuff[entry] = 0;
+                               /* Mark status as empty. */
+                               yp->tx_status[entry].tx_errs = 0;
+                       }
+
+#ifndef final_version
+                       if (yp->cur_tx - dirty_tx > TX_RING_SIZE) {
+                               netdev_err(dev, "Out-of-sync dirty pointer, %d vs. %d, full=%d\n",
+                                          dirty_tx, yp->cur_tx, yp->tx_full);
+                               dirty_tx += TX_RING_SIZE;
+                       }
+#endif
+
+                       if (yp->tx_full &&
+                           yp->cur_tx - dirty_tx < TX_QUEUE_SIZE - 2) {
+                               /* The ring is no longer full, clear tbusy. */
+                               yp->tx_full = 0;
+                               netif_wake_queue(dev);
+                       }
+
+                       yp->dirty_tx = dirty_tx;
+                       yp->tx_tail_desc = &yp->tx_status[dirty_tx % TX_RING_SIZE];
+               }
+#endif
+
+               /* Log errors and other uncommon events. */
+               if (intr_status & 0x2ee)        /* Abnormal error summary. */
+                       yellowfin_error(dev, intr_status);
+
+               if (--boguscnt < 0) {
+                       netdev_warn(dev, "Too much work at interrupt, status=%#04x\n",
+                                   intr_status);
+                       break;
+               }
+       } while (1);
+
+       if (yellowfin_debug > 3)
+               netdev_printk(KERN_DEBUG, dev, "exiting interrupt, status=%#04x\n",
+                             ioread16(ioaddr + IntrStatus));
+
+       spin_unlock (&yp->lock);
+       return IRQ_RETVAL(handled);
+}
+
+/* This routine is logically part of the interrupt handler, but separated
+   for clarity and better register allocation. */
+static int yellowfin_rx(struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       int entry = yp->cur_rx % RX_RING_SIZE;
+       int boguscnt = yp->dirty_rx + RX_RING_SIZE - yp->cur_rx;
+
+       if (yellowfin_debug > 4) {
+               printk(KERN_DEBUG " In yellowfin_rx(), entry %d status %08x\n",
+                          entry, yp->rx_ring[entry].result_status);
+               printk(KERN_DEBUG "   #%d desc. %08x %08x %08x\n",
+                          entry, yp->rx_ring[entry].dbdma_cmd, yp->rx_ring[entry].addr,
+                          yp->rx_ring[entry].result_status);
+       }
+
+       /* If EOP is set on the next entry, it's a new packet. Send it up. */
+       while (1) {
+               struct yellowfin_desc *desc = &yp->rx_ring[entry];
+               struct sk_buff *rx_skb = yp->rx_skbuff[entry];
+               s16 frame_status;
+               u16 desc_status;
+               int data_size, yf_size;
+               u8 *buf_addr;
+
+               if(!desc->result_status)
+                       break;
+               pci_dma_sync_single_for_cpu(yp->pci_dev, le32_to_cpu(desc->addr),
+                       yp->rx_buf_sz, PCI_DMA_FROMDEVICE);
+               desc_status = le32_to_cpu(desc->result_status) >> 16;
+               buf_addr = rx_skb->data;
+               data_size = (le32_to_cpu(desc->dbdma_cmd) -
+                       le32_to_cpu(desc->result_status)) & 0xffff;
+               frame_status = get_unaligned_le16(&(buf_addr[data_size - 2]));
+               if (yellowfin_debug > 4)
+                       printk(KERN_DEBUG "  %s() status was %04x\n",
+                              __func__, frame_status);
+               if (--boguscnt < 0)
+                       break;
+
+               yf_size = sizeof(struct yellowfin_desc);
+
+               if ( ! (desc_status & RX_EOP)) {
+                       if (data_size != 0)
+                               netdev_warn(dev, "Oversized Ethernet frame spanned multiple buffers, status %04x, data_size %d!\n",
+                                           desc_status, data_size);
+                       dev->stats.rx_length_errors++;
+               } else if ((yp->drv_flags & IsGigabit)  &&  (frame_status & 0x0038)) {
+                       /* There was a error. */
+                       if (yellowfin_debug > 3)
+                               printk(KERN_DEBUG "  %s() Rx error was %04x\n",
+                                      __func__, frame_status);
+                       dev->stats.rx_errors++;
+                       if (frame_status & 0x0060) dev->stats.rx_length_errors++;
+                       if (frame_status & 0x0008) dev->stats.rx_frame_errors++;
+                       if (frame_status & 0x0010) dev->stats.rx_crc_errors++;
+                       if (frame_status < 0) dev->stats.rx_dropped++;
+               } else if ( !(yp->drv_flags & IsGigabit)  &&
+                                  ((buf_addr[data_size-1] & 0x85) || buf_addr[data_size-2] & 0xC0)) {
+                       u8 status1 = buf_addr[data_size-2];
+                       u8 status2 = buf_addr[data_size-1];
+                       dev->stats.rx_errors++;
+                       if (status1 & 0xC0) dev->stats.rx_length_errors++;
+                       if (status2 & 0x03) dev->stats.rx_frame_errors++;
+                       if (status2 & 0x04) dev->stats.rx_crc_errors++;
+                       if (status2 & 0x80) dev->stats.rx_dropped++;
+#ifdef YF_PROTOTYPE            /* Support for prototype hardware errata. */
+               } else if ((yp->flags & HasMACAddrBug)  &&
+                       !ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
+                                                     entry * yf_size),
+                                         dev->dev_addr) &&
+                       !ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
+                                                     entry * yf_size),
+                                         "\377\377\377\377\377\377")) {
+                       if (bogus_rx++ == 0)
+                               netdev_warn(dev, "Bad frame to %pM\n",
+                                           buf_addr);
+#endif
+               } else {
+                       struct sk_buff *skb;
+                       int pkt_len = data_size -
+                               (yp->chip_id ? 7 : 8 + buf_addr[data_size - 8]);
+                       /* To verify: Yellowfin Length should omit the CRC! */
+
+#ifndef final_version
+                       if (yellowfin_debug > 4)
+                               printk(KERN_DEBUG "  %s() normal Rx pkt length %d of %d, bogus_cnt %d\n",
+                                      __func__, pkt_len, data_size, boguscnt);
+#endif
+                       /* Check if the packet is long enough to just pass up the skbuff
+                          without copying to a properly sized skbuff. */
+                       if (pkt_len > rx_copybreak) {
+                               skb_put(skb = rx_skb, pkt_len);
+                               pci_unmap_single(yp->pci_dev,
+                                       le32_to_cpu(yp->rx_ring[entry].addr),
+                                       yp->rx_buf_sz,
+                                       PCI_DMA_FROMDEVICE);
+                               yp->rx_skbuff[entry] = NULL;
+                       } else {
+                               skb = netdev_alloc_skb(dev, pkt_len + 2);
+                               if (skb == NULL)
+                                       break;
+                               skb_reserve(skb, 2);    /* 16 byte align the IP header */
+                               skb_copy_to_linear_data(skb, rx_skb->data, pkt_len);
+                               skb_put(skb, pkt_len);
+                               pci_dma_sync_single_for_device(yp->pci_dev,
+                                                               le32_to_cpu(desc->addr),
+                                                               yp->rx_buf_sz,
+                                                               PCI_DMA_FROMDEVICE);
+                       }
+                       skb->protocol = eth_type_trans(skb, dev);
+                       netif_rx(skb);
+                       dev->stats.rx_packets++;
+                       dev->stats.rx_bytes += pkt_len;
+               }
+               entry = (++yp->cur_rx) % RX_RING_SIZE;
+       }
+
+       /* Refill the Rx ring buffers. */
+       for (; yp->cur_rx - yp->dirty_rx > 0; yp->dirty_rx++) {
+               entry = yp->dirty_rx % RX_RING_SIZE;
+               if (yp->rx_skbuff[entry] == NULL) {
+                       struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
+                       if (skb == NULL)
+                               break;                          /* Better luck next round. */
+                       yp->rx_skbuff[entry] = skb;
+                       skb_reserve(skb, 2);    /* Align IP on 16 byte boundaries */
+                       yp->rx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
+                               skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE));
+               }
+               yp->rx_ring[entry].dbdma_cmd = cpu_to_le32(CMD_STOP);
+               yp->rx_ring[entry].result_status = 0;   /* Clear complete bit. */
+               if (entry != 0)
+                       yp->rx_ring[entry - 1].dbdma_cmd =
+                               cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
+               else
+                       yp->rx_ring[RX_RING_SIZE - 1].dbdma_cmd =
+                               cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | BRANCH_ALWAYS
+                                                       | yp->rx_buf_sz);
+       }
+
+       return 0;
+}
+
+static void yellowfin_error(struct net_device *dev, int intr_status)
+{
+       netdev_err(dev, "Something Wicked happened! %04x\n", intr_status);
+       /* Hmmmmm, it's not clear what to do here. */
+       if (intr_status & (IntrTxPCIErr | IntrTxPCIFault))
+               dev->stats.tx_errors++;
+       if (intr_status & (IntrRxPCIErr | IntrRxPCIFault))
+               dev->stats.rx_errors++;
+}
+
+static int yellowfin_close(struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       void __iomem *ioaddr = yp->base;
+       int i;
+
+       netif_stop_queue (dev);
+
+       if (yellowfin_debug > 1) {
+               netdev_printk(KERN_DEBUG, dev, "Shutting down ethercard, status was Tx %04x Rx %04x Int %02x\n",
+                             ioread16(ioaddr + TxStatus),
+                             ioread16(ioaddr + RxStatus),
+                             ioread16(ioaddr + IntrStatus));
+               netdev_printk(KERN_DEBUG, dev, "Queue pointers were Tx %d / %d,  Rx %d / %d\n",
+                             yp->cur_tx, yp->dirty_tx,
+                             yp->cur_rx, yp->dirty_rx);
+       }
+
+       /* Disable interrupts by clearing the interrupt mask. */
+       iowrite16(0x0000, ioaddr + IntrEnb);
+
+       /* Stop the chip's Tx and Rx processes. */
+       iowrite32(0x80000000, ioaddr + RxCtrl);
+       iowrite32(0x80000000, ioaddr + TxCtrl);
+
+       del_timer(&yp->timer);
+
+#if defined(__i386__)
+       if (yellowfin_debug > 2) {
+               printk(KERN_DEBUG "  Tx ring at %08llx:\n",
+                               (unsigned long long)yp->tx_ring_dma);
+               for (i = 0; i < TX_RING_SIZE*2; i++)
+                       printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x %08x\n",
+                                  ioread32(ioaddr + TxPtr) == (long)&yp->tx_ring[i] ? '>' : ' ',
+                                  i, yp->tx_ring[i].dbdma_cmd, yp->tx_ring[i].addr,
+                                  yp->tx_ring[i].branch_addr, yp->tx_ring[i].result_status);
+               printk(KERN_DEBUG "  Tx status %p:\n", yp->tx_status);
+               for (i = 0; i < TX_RING_SIZE; i++)
+                       printk(KERN_DEBUG "   #%d status %04x %04x %04x %04x\n",
+                                  i, yp->tx_status[i].tx_cnt, yp->tx_status[i].tx_errs,
+                                  yp->tx_status[i].total_tx_cnt, yp->tx_status[i].paused);
+
+               printk(KERN_DEBUG "  Rx ring %08llx:\n",
+                               (unsigned long long)yp->rx_ring_dma);
+               for (i = 0; i < RX_RING_SIZE; i++) {
+                       printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x\n",
+                                  ioread32(ioaddr + RxPtr) == (long)&yp->rx_ring[i] ? '>' : ' ',
+                                  i, yp->rx_ring[i].dbdma_cmd, yp->rx_ring[i].addr,
+                                  yp->rx_ring[i].result_status);
+                       if (yellowfin_debug > 6) {
+                               if (get_unaligned((u8*)yp->rx_ring[i].addr) != 0x69) {
+                                       int j;
+
+                                       printk(KERN_DEBUG);
+                                       for (j = 0; j < 0x50; j++)
+                                               pr_cont(" %04x",
+                                                       get_unaligned(((u16*)yp->rx_ring[i].addr) + j));
+                                       pr_cont("\n");
+                               }
+                       }
+               }
+       }
+#endif /* __i386__ debugging only */
+
+       free_irq(yp->pci_dev->irq, dev);
+
+       /* Free all the skbuffs in the Rx queue. */
+       for (i = 0; i < RX_RING_SIZE; i++) {
+               yp->rx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
+               yp->rx_ring[i].addr = cpu_to_le32(0xBADF00D0); /* An invalid address. */
+               if (yp->rx_skbuff[i]) {
+                       dev_kfree_skb(yp->rx_skbuff[i]);
+               }
+               yp->rx_skbuff[i] = NULL;
+       }
+       for (i = 0; i < TX_RING_SIZE; i++) {
+               if (yp->tx_skbuff[i])
+                       dev_kfree_skb(yp->tx_skbuff[i]);
+               yp->tx_skbuff[i] = NULL;
+       }
+
+#ifdef YF_PROTOTYPE                    /* Support for prototype hardware errata. */
+       if (yellowfin_debug > 0) {
+               netdev_printk(KERN_DEBUG, dev, "Received %d frames that we should not have\n",
+                             bogus_rx);
+       }
+#endif
+
+       return 0;
+}
+
+/* Set or clear the multicast filter for this adaptor. */
+
+static void set_rx_mode(struct net_device *dev)
+{
+       struct yellowfin_private *yp = netdev_priv(dev);
+       void __iomem *ioaddr = yp->base;
+       u16 cfg_value = ioread16(ioaddr + Cnfg);
+
+       /* Stop the Rx process to change any value. */
+       iowrite16(cfg_value & ~0x1000, ioaddr + Cnfg);
+       if (dev->flags & IFF_PROMISC) {                 /* Set promiscuous. */
+               iowrite16(0x000F, ioaddr + AddrMode);
+       } else if ((netdev_mc_count(dev) > 64) ||
+                  (dev->flags & IFF_ALLMULTI)) {
+               /* Too many to filter well, or accept all multicasts. */
+               iowrite16(0x000B, ioaddr + AddrMode);
+       } else if (!netdev_mc_empty(dev)) { /* Must use the multicast hash table. */
+               struct netdev_hw_addr *ha;
+               u16 hash_table[4];
+               int i;
+
+               memset(hash_table, 0, sizeof(hash_table));
+               netdev_for_each_mc_addr(ha, dev) {
+                       unsigned int bit;
+
+                       /* Due to a bug in the early chip versions, multiple filter
+                          slots must be set for each address. */
+                       if (yp->drv_flags & HasMulticastBug) {
+                               bit = (ether_crc_le(3, ha->addr) >> 3) & 0x3f;
+                               hash_table[bit >> 4] |= (1 << bit);
+                               bit = (ether_crc_le(4, ha->addr) >> 3) & 0x3f;
+                               hash_table[bit >> 4] |= (1 << bit);
+                               bit = (ether_crc_le(5, ha->addr) >> 3) & 0x3f;
+                               hash_table[bit >> 4] |= (1 << bit);
+                       }
+                       bit = (ether_crc_le(6, ha->addr) >> 3) & 0x3f;
+                       hash_table[bit >> 4] |= (1 << bit);
+               }
+               /* Copy the hash table to the chip. */
+               for (i = 0; i < 4; i++)
+                       iowrite16(hash_table[i], ioaddr + HashTbl + i*2);
+               iowrite16(0x0003, ioaddr + AddrMode);
+       } else {                                        /* Normal, unicast/broadcast-only mode. */
+               iowrite16(0x0001, ioaddr + AddrMode);
+       }
+       /* Restart the Rx process. */
+       iowrite16(cfg_value | 0x1000, ioaddr + Cnfg);
+}
+
+static void yellowfin_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
+{
+       struct yellowfin_private *np = netdev_priv(dev);
+
+       strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
+       strlcpy(info->version, DRV_VERSION, sizeof(info->version));
+       strlcpy(info->bus_info, pci_name(np->pci_dev), sizeof(info->bus_info));
+}
+
+static const struct ethtool_ops ethtool_ops = {
+       .get_drvinfo = yellowfin_get_drvinfo
+};
+
+static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
+{
+       struct yellowfin_private *np = netdev_priv(dev);
+       void __iomem *ioaddr = np->base;
+       struct mii_ioctl_data *data = if_mii(rq);
+
+       switch(cmd) {
+       case SIOCGMIIPHY:               /* Get address of MII PHY in use. */
+               data->phy_id = np->phys[0] & 0x1f;
+               /* Fall Through */
+
+       case SIOCGMIIREG:               /* Read MII PHY register. */
+               data->val_out = mdio_read(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f);
+               return 0;
+
+       case SIOCSMIIREG:               /* Write MII PHY register. */
+               if (data->phy_id == np->phys[0]) {
+                       u16 value = data->val_in;
+                       switch (data->reg_num) {
+                       case 0:
+                               /* Check for autonegotiation on or reset. */
+                               np->medialock = (value & 0x9000) ? 0 : 1;
+                               if (np->medialock)
+                                       np->full_duplex = (value & 0x0100) ? 1 : 0;
+                               break;
+                       case 4: np->advertising = value; break;
+                       }
+                       /* Perhaps check_duplex(dev), depending on chip semantics. */
+               }
+               mdio_write(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f, data->val_in);
+               return 0;
+       default:
+               return -EOPNOTSUPP;
+       }
+}
+
+
+static void yellowfin_remove_one(struct pci_dev *pdev)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+       struct yellowfin_private *np;
+
+       BUG_ON(!dev);
+       np = netdev_priv(dev);
+
+        pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status,
+               np->tx_status_dma);
+       pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma);
+       pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma);
+       unregister_netdev (dev);
+
+       pci_iounmap(pdev, np->base);
+
+       pci_release_regions (pdev);
+
+       free_netdev (dev);
+}
+
+
+static struct pci_driver yellowfin_driver = {
+       .name           = DRV_NAME,
+       .id_table       = yellowfin_pci_tbl,
+       .probe          = yellowfin_init_one,
+       .remove         = yellowfin_remove_one,
+};
+
+
+static int __init yellowfin_init (void)
+{
+/* when a module, this is printed whether or not devices are found in probe */
+#ifdef MODULE
+       printk(version);
+#endif
+       return pci_register_driver(&yellowfin_driver);
+}
+
+
+static void __exit yellowfin_cleanup (void)
+{
+       pci_unregister_driver (&yellowfin_driver);
+}
+
+
+module_init(yellowfin_init);
+module_exit(yellowfin_cleanup);