/* pci-skeleton.c: A Linux PCI network adapter skeleton device driver. */
/*
	Written 1998-2000 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.

	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/pci-skeleton.html
*/

/* These identify the driver base version and may not be removed. */
static const char version1[] =
"pci-skeleton.c:v2.03 4/06/00  Written by Donald Becker\n";
static const char version2[] =
"  http://www.scyld.com/network/drivers.html\n";

/* 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 = 0;
/* Maximum number of multicast addresses to filter (vs. rx-all-multicast).
   Typical is a 64 element hash table based on the Ethernet CRC.  */
static int multicast_filter_limit = 32;

/* Set the copy breakpoint for the copy-only-tiny-frames scheme.
   Setting to > 1518 effectively disables this feature. */
static int rx_copybreak = 0;

/* Used to pass the media type, etc.
   Both 'options[]' and 'full_duplex[]' should exist for driver
   interoperability.
   The media type is usually passed in 'options[]'.
*/
#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};

/* Operational parameters that are set at compile time. */

/* Keep the ring sizes a power of two for compile efficiency.
   The compiler will convert <unsigned>'%'<2^N> into a bit mask.
   Making the Tx ring too large 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_LEN	10		/* Limit ring entries actually used.  */
#define RX_RING_SIZE	32

/* 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.*/

#ifndef __KERNEL__
#define __KERNEL__
#endif
#if !defined(__OPTIMIZE__)
#warning  You must compile this file with the correct options!
#warning  See the last lines of the source file.
#error You must compile this driver with "-O".
#endif

/* Include files, designed to support most kernel versions 2.0.0 and later. */
#include <linux/config.h>
#if defined(CONFIG_SMP) && ! defined(__SMP__)
#define __SMP__
#endif
#if defined(CONFIG_MODVERSIONS) && ! defined(MODVERSIONS)
#define MODVERSIONS
#endif

#include <linux/version.h>
#include <linux/module.h>
#if LINUX_VERSION_CODE < 0x20300  &&  defined(MODVERSIONS)
#include <linux/modversions.h>
#endif

#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/timer.h>
#include <linux/errno.h>
#include <linux/ioport.h>
#include <linux/malloc.h>
#include <linux/interrupt.h>
#include <linux/pci.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <asm/processor.h>		/* Processor type for cache alignment. */
#include <asm/bitops.h>
#include <asm/io.h>

#ifdef INLINE_PCISCAN
#include "k_compat.h"
#else
#include "pci-scan.h"
#include "kern_compat.h"
#endif

/* Condensed operations for readability. */
#define virt_to_le32desc(addr)  cpu_to_le32(virt_to_bus(addr))
#define le32desc_to_virt(addr)  bus_to_virt(le32_to_cpu(addr))

#if (LINUX_VERSION_CODE >= 0x20100)  &&  defined(MODULE)
char kernel_version[] = UTS_RELEASE;
#endif

MODULE_AUTHOR("Donald Becker <becker@scyld.com>");
MODULE_DESCRIPTION("PCI network skeleton Ethernet driver");
MODULE_PARM(max_interrupt_work, "i");
MODULE_PARM(mtu, "i");
MODULE_PARM(debug, "i");
MODULE_PARM(rx_copybreak, "i");
MODULE_PARM(options, "1-" __MODULE_STRING(MAX_UNITS) "i");
MODULE_PARM(full_duplex, "1-" __MODULE_STRING(MAX_UNITS) "i");

/*
				Theory of Operation

I. Board Compatibility

State the chips and boards this driver is known to work with.
Note any similar chips or boards that will not work.

This driver skeleton demonstrates the driver for an idealized
descriptor-based bus-master PCI chip.

II. Board-specific settings

No jumpers exist on most PCI boards, so this section is usually empty.

III. Driver operation

IIIa. Ring buffers

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.
Some chips explicitly use only 2^N sized rings, while others use a
'next descriptor' pointer that the driver forms into rings.

IIIb/c. Transmit/Receive Structure

This driver uses a zero-copy receive and transmit scheme.
The driver allocates full frame size skbuffs for the Rx ring buffers at
open() time and passes the skb->data field to the chip 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.  Buffers consumed this way are replaced by newly allocated
skbuffs in a later phase of receives.

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.  New boards are typically used in generously configured machines
and the underfilled buffers have negligible impact compared to the benefit of
a single allocation size, so the default value of zero results in never
copying packets.  When copying is done, the cost is usually mitigated by using
a combined copy/checksum routine.  Copying also preloads the cache, which is
most useful with small frames.

A subtle aspect of the operation is that the IP header at offset 14 in an
ethernet frame isn't longword aligned for further processing.
When unaligned buffers are permitted by the hardware (and always on copies)
frames are put into the skbuff at an offset of "+2", 16-byte aligning
the IP header.

IIId. 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 interrupt handling 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 'lp->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 'lp->tx_full' flag is set, it
clears both the tx_full and tbusy flags.

IIId. SMP semantics

The following are serialized with respect to each other via the "xmit_lock".
  dev->hard_start_xmit()	Transmit a packet
  dev->tx_timeout()			Transmit watchdog for stuck Tx
  dev->set_multicast_list()	Set the recieve filter.
Note: The Tx timeout watchdog code is implemented by the timer routine in
kernels up to 2.2.*.  In 2.4.* and later the timeout code is part of the
driver interface.

The following fall under the global kernel lock.  The module will not be
unloaded during the call, unless a call with a potential reschedule e.g.
kmalloc() is called.  No other synchronization assertion is made.
  dev->open()
  dev->do_ioctl()
  dev->get_stats()
Caution: The lock for dev->open() is commonly broken with request_irq() or
kmalloc().  It is best to avoid any lock-breaking call in do_ioctl() and
get_stats(), or additional module locking code must be implemented.

The following is self-serialized (no simultaneous entry)
  An handler registered with request_irq().

IV. Notes

There are few hard rules about writing device drivers, but I have read some
amazingly unwise code.  Bad code often stems from the mistaken belief that
this device driver is the most important code the machine is running.

Remember that this is a real OS, not DOS.  Never mess with system hardware
(the timer chip, DMA channels, IRQ mapping): use the hardware-independent
kernel services instead.

While there is a udelay() function, use it sparingly and only with tiny
delays.  It is not for having the kernel wait three seconds while
autonegotiation completes!  At boot time or module insertion time this rule
can be relaxed somewhat, but even then the total delay should be under a
timer tick (10msec).

All loops should be checked with a 'boguscnt' limit.  That includes the
interrupt handler, which should limit the work it does with a tunable
parameter.  Loops that check for hardware completion should have a typical
completion count in a comment.  An exception is traversing software
maintained lists, most of which should be designed to grow arbitrarily long.

The device driver source code file should be self-contained, and as compact
as readability permits.  It should not be spread out over multiple source
files, and there should only be a driver.h file in special circumstances.

Finally, always support multiple devices.  That means few, if any, global
variables.  All driver variables should be 'static'.

IVb. References

http://cesdis.gsfc.nasa.gov/linux/misc/100mbps.html
http://cesdis.gsfc.nasa.gov/linux/misc/NWay.html

List the documentation used to write the driver.  Note any proprietary or
trade secret information, and the agreement you have to release the same.

IVc. Errata

Note any known bugs or limitations.
*/



/* This table drives the PCI probe routines.
   Note the matching code -- the first table entry matches only the 5678 card,
   the second all remaining 56** cards.
*/

static void *netfin_probe1(struct pci_dev *pdev, void *init_dev,
						   long ioaddr, int irq, int chip_idx, int find_cnt);
static int netdev_pwr_event(void *dev_instance, int event);
enum chip_capability_flags {CanHaveMII=1, };
#ifdef USE_IO_OPS
#define PCI_IOTYPE (PCI_USES_MASTER | PCI_USES_IO  | PCI_ADDR0)
#else
#define PCI_IOTYPE (PCI_USES_MASTER | PCI_USES_MEM | PCI_ADDR1)
#endif

static struct pci_id_info pci_id_tbl[] = {
	{"NetTechCom 5678 adapter", {0x56781234, 0xffffffff, },
	 PCI_IOTYPE, 128, CanHaveMII},
	{"NetTechCom 5600 series",  {0x56001234, 0xff00ffff, },
	 PCI_IOTYPE, 128, CanHaveMII},
	{0,},						/* 0 terminated list. */
};

struct drv_id_info netfin_drv_id = {
	"netfin", 0, PCI_CLASS_NETWORK_ETHERNET<<8, pci_id_tbl, netfin_probe1,
	netdev_pwr_event };

/* This driver was written to use PCI memory space, however x86-oriented
   hardware sometimes works only with I/O space accesses. */
#ifdef USE_IO_OPS
#undef readb
#undef readw
#undef readl
#undef writeb
#undef writew
#undef writel
#define readb inb
#define readw inw
#define readl inl
#define writeb outb
#define writew outw
#define writel outl
#endif

/* Offsets to the device registers.
   Unlike software-only systems, device drivers interact with complex hardware.
   It's not useful to define symbolic names for every register bit in the
   device.  The name can only partially document the semantics and make
   the driver longer and more difficult to read.
   In general, only the important configuration values or bits changed
   multiple times should be defined symbolically.
*/
enum register_offsets {
	ChipCmd=0x00, IntrStatus=0x04, IntrEnable=0x08,
	TxStatus=0x10, TxCmd=0x14, TxRingPtr=0x18,
	RxStatus=0x20, RxCmd=0x24, RxRingPtr=0x28,
	EECtrl=0x40, MIICtrl=0x44, LEDCtrl=0x48,
	StationAddr=0x50, RxMode=0x58, TxMode=0x5C,
	RxMissed=0x60, RxCRCErrs=0x64, MulticastFilter0=0x68,MulticastFilter1=0x6C,
	PCIBusCfg=0x70, FIFOCfg=0x74, ChipReset=0x78,
};

/* Bits in the interrupt status/mask registers. */
enum intr_status_bits {
	IntrRxDone=0x01, IntrRxEmpty=0x02, IntrRxPCIErr=0x04,
	IntrTxDone=0x10, IntrTxEmpty=0x20, IntrTxPCIErr=0x40,
	StatsMax=0x0100, LinkChange=0x0200,
	IntrNormalSummary=0x8000,	IntrAbnormalSummary=0x4000,
};

/* Bits in the RxMode register. */
enum rx_mode_bits {
	AcceptErr=0x20, AcceptRunt=0x10,
	AcceptBroadcast=0x08, AcceptMulticast=0x04,
	AcceptAllPhys=0x02, AcceptMyPhys=0x01,
};

/* The Rx and Tx buffer descriptors. */
/* Note that using only 32 bit fields simplifies conversion to big-endian
   architectures. */
struct netdev_desc {
	u16 length;			/* u16 == u_int16_t */
	u16 status;
	u32 addr;
};

/* Bits in network_desc.status */
enum desc_status_bits {
	DescOwn=0x8000, DescEndPacket=0x4000, DescEndRing=0x2000, DescIntr=0x1000,
};

#define PRIV_ALIGN	15 	/* Required alignment mask */
/* Use  __attribute__((aligned (L1_CACHE_BYTES)))  to maintain alignment
   within the structure. */
struct netdev_private {
	/* Descriptor rings first for alignment. */
	struct netdev_desc rx_ring[RX_RING_SIZE];
	struct netdev_desc tx_ring[TX_RING_SIZE];
	struct net_device *next_module;		/* Link for devices of this type. */
	void *priv_addr;					/* Unaligned address for kfree */
	const char *product_name;
	/* The addresses of receive-in-place skbuffs. */
	struct sk_buff* rx_skbuff[RX_RING_SIZE];
	/* The saved address of a sent-in-place packet/buffer, for later free(). */
	struct sk_buff* tx_skbuff[TX_RING_SIZE];
	struct net_device_stats stats;
	struct timer_list timer;	/* Media monitoring timer. */
	/* Frequently used values: keep some adjacent for cache effect. */
	int chip_id, drv_flags;
	struct pci_dev *pci_dev;
	long in_interrupt;			/* Word-long for SMP locks. */
	struct netdev_desc *rx_head_desc;
	unsigned int cur_rx, dirty_rx;		/* Producer/consumer ring indices */
	unsigned int cur_tx, dirty_tx;
	unsigned int rx_buf_sz;				/* Based on MTU+slack. */
	unsigned int tx_full:1;				/* The Tx queue is full. */
	/* These values are keep track of the transceiver/media in use. */
	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[2];				/* MII device addresses. */
};

static int  eeprom_read(long ioaddr, int location);
static int  mdio_read(struct net_device *dev, int phy_id, int location);
static void mdio_write(struct net_device *dev, int phy_id, int location, int value);
static int  netdev_open(struct net_device *dev);
static void check_duplex(struct net_device *dev);
static void netdev_timer(unsigned long data);
static void tx_timeout(struct net_device *dev);
static void init_ring(struct net_device *dev);
static int  start_tx(struct sk_buff *skb, struct net_device *dev);
static void intr_handler(int irq, void *dev_instance, struct pt_regs *regs);
static void netdev_error(struct net_device *dev, int intr_status);
static int  netdev_rx(struct net_device *dev);
static void netdev_error(struct net_device *dev, int intr_status);
static void set_rx_mode(struct net_device *dev);
static struct net_device_stats *get_stats(struct net_device *dev);
static int mii_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
static int  netdev_close(struct net_device *dev);



/* A list of our installed devices, for removing the driver module. */
static struct net_device *root_net_dev = NULL;

#ifndef MODULE
/* You *must* rename this! */
int skel_netdev_probe(struct net_device *dev)
{
	if (pci_drv_register(&netfin_drv_id, dev) < 0)
		return -ENODEV;
	printk(KERN_INFO "%s" KERN_INFO "%s", version1, version2);
	return 0;
}
#endif

static void *netfin_probe1(struct pci_dev *pdev, void *init_dev,
						   long ioaddr, int irq, int chip_idx, int card_idx)
{
	struct net_device *dev;
	struct netdev_private *np;
	void *priv_mem;
	int i, option = card_idx < MAX_UNITS ? options[card_idx] : 0;

	dev = init_etherdev(init_dev, 0);

	printk(KERN_INFO "%s: %s at 0x%lx, ",
		   dev->name, pci_id_tbl[chip_idx].name, ioaddr);

	for (i = 0; i < 6; i++)
		dev->dev_addr[i] = eeprom_read(ioaddr, i);
	for (i = 0; i < 5; i++)
			printk("%2.2x:", dev->dev_addr[i]);
	printk("%2.2x, IRQ %d.\n", dev->dev_addr[i], irq);

#if ! defined(final_version) /* Dump the EEPROM contents during development. */
	if (debug > 4)
		for (i = 0; i < 0x100; i++)
			printk("%4.4x%s",
				   eeprom_read(ioaddr, i), i % 16 != 15 ? " " : "\n");
#endif

	/* Make certain elements e.g. descriptor lists are aligned. */
	priv_mem = kmalloc(sizeof(*np) + PRIV_ALIGN, GFP_KERNEL);
	/* Check for the very unlikely case of no memory. */
	if (priv_mem == NULL)
		return NULL;

	/* Do bogusness checks before this point.
	   We do a request_region() only to register /proc/ioports info. */
#ifdef USE_IO_OPS
	request_region(ioaddr, pci_id_tbl[chip_idx].io_size, dev->name);
#endif

	/* Reset the chip to erase previous misconfiguration. */
	writel(1, ioaddr + ChipReset);

	dev->base_addr = ioaddr;
	dev->irq = irq;

	dev->priv = np = (void *)(((long)priv_mem + PRIV_ALIGN) & ~PRIV_ALIGN);
	memset(np, 0, sizeof(*np));
	np->priv_addr = priv_mem;

	np->next_module = root_net_dev;
	root_net_dev = dev;

	np->pci_dev = pdev;
	np->chip_id = chip_idx;
	np->drv_flags = pci_id_tbl[chip_idx].drv_flags;

	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 (card_idx < MAX_UNITS  &&  full_duplex[card_idx] > 0)
		np->full_duplex = 1;

	if (np->full_duplex)
		np->duplex_lock = 1;

	/* The chip-specific entries in the device structure. */
	dev->open = &netdev_open;
	dev->hard_start_xmit = &start_tx;
	dev->stop = &netdev_close;
	dev->get_stats = &get_stats;
	dev->set_multicast_list = &set_rx_mode;
	dev->do_ioctl = &mii_ioctl;

	if (mtu)
		dev->mtu = mtu;

	if (np->drv_flags & CanHaveMII) {
		int phy, phy_idx = 0;
		for (phy = 0; phy < 32 && phy_idx < 4; phy++) {
			int mii_status = mdio_read(dev, phy, 1);
			if (mii_status != 0xffff  &&  mii_status != 0x0000) {
				np->phys[phy_idx++] = phy;
				np->advertising = mdio_read(dev, phy, 4);
				printk(KERN_INFO "%s: MII PHY found at address %d, status "
					   "0x%4.4x advertising %4.4x.\n",
					   dev->name, phy, mii_status, np->advertising);
			}
		}
		np->mii_cnt = phy_idx;
	}

	return dev;
}


/* Read the EEPROM and MII Management Data I/O (MDIO) interfaces.  These are
   often serial bit streams generated by the host processor.
   The example below is for the common 93c46 EEPROM, 64 16 bit words. */

/* Delay between EEPROM clock transitions.
   No extra delay is needed with 33Mhz PCI, but future 66Mhz access may need
   a delay.  Note that pre-2.0.34 kernels had a cache-alignment bug that
   made udelay() unreliable.
   The old method of using an ISA access as a delay, __SLOW_DOWN_IO__, is
   depricated.
*/
#define eeprom_delay(ee_addr)	readl(ee_addr)

enum EEPROM_Ctrl_Bits {
	EE_ShiftClk=0x01, EE_DataBit=0x02, EE_ChipSelect=0x04, EE_DataDir=0x08,
};
#define EE_Write0 (EE_DataDir | EE_ChipSelect)
#define EE_Write1 (EE_DataDir | EE_ChipSelect | EE_DataBit)

/* The EEPROM commands include the alway-set leading bit. */
enum EEPROM_Cmds {
	EE_WriteCmd=(5 << 6), EE_ReadCmd=(6 << 6), EE_EraseCmd=(7 << 6),
};

static int eeprom_read(long addr, int location)
{
	int i;
	int retval = 0;
	int ee_addr = addr + EECtrl;
	int read_cmd = location | EE_ReadCmd;
	writel(EE_DataDir, ee_addr);

	/* Shift the read command bits out. */
	for (i = 10; i >= 0; i--) {
		short dataval = (read_cmd & (1 << i)) ? EE_Write1 : EE_Write0;
		writel(dataval, ee_addr);
		eeprom_delay(ee_addr);
		writel(dataval | EE_ShiftClk, ee_addr);
		eeprom_delay(ee_addr);
	}
	writel(EE_ChipSelect, ee_addr);

	for (i = 16; i > 0; i--) {
		writel(EE_ChipSelect | EE_ShiftClk, ee_addr);
		eeprom_delay(ee_addr);
		retval = (retval << 1) | ((readl(ee_addr) & EE_DataBit) ? 1 : 0);
		writel(EE_ChipSelect, ee_addr);
		eeprom_delay(ee_addr);
	}

	/* Terminate the EEPROM access. */
	writel(EE_DataDir, ee_addr);
	writel(0, ee_addr);
	return retval;
}

/*  MII transceiver control section.
	Read and write the MII registers using software-generated serial
	MDIO protocol.  See the MII specifications or DP83840A data sheet
	for details.

	The maximum data clock rate is 2.5 Mhz.  The minimum timing is usually
	met by back-to-back 33Mhz PCI cycles. */
#define mdio_delay(mdio_addr) readl(mdio_addr)

/* Set iff a MII transceiver on any interface requires mdio preamble.
   This only set with older tranceivers, so the extra
   code size of a per-interface flag is not worthwhile. */
static char mii_preamble_required = 0;

enum mii_reg_bits {
	MDIO_ShiftClk=0x0001, MDIO_Data=0x0002, MDIO_EnbOutput=0x0004,
};
#define MDIO_EnbIn  (0)
#define MDIO_WRITE0 (MDIO_EnbOutput)
#define MDIO_WRITE1 (MDIO_Data | MDIO_EnbOutput)

/* Generate the preamble required for initial synchronization and
   a few older transceivers. */
static void mdio_sync(long mdio_addr)
{
	int bits = 32;

	/* Establish sync by sending at least 32 logic ones. */
	while (--bits >= 0) {
		writel(MDIO_WRITE1, mdio_addr);
		mdio_delay(mdio_addr);
		writel(MDIO_WRITE1 | MDIO_ShiftClk, mdio_addr);
		mdio_delay(mdio_addr);
	}
}

static int mdio_read(struct net_device *dev, int phy_id, int location)
{
	long mdio_addr = dev->base_addr + MIICtrl;
	int mii_cmd = (0xf6 << 10) | (phy_id << 5) | location;
	int i, retval = 0;

	if (mii_preamble_required)
		mdio_sync(mdio_addr);

	/* Shift the read command bits out. */
	for (i = 15; i >= 0; i--) {
		int dataval = (mii_cmd & (1 << i)) ? MDIO_WRITE1 : MDIO_WRITE0;

		writel(dataval, mdio_addr);
		mdio_delay(mdio_addr);
		writel(dataval | MDIO_ShiftClk, mdio_addr);
		mdio_delay(mdio_addr);
	}
	/* Read the two transition, 16 data, and wire-idle bits. */
	for (i = 19; i > 0; i--) {
		writel(MDIO_EnbIn, mdio_addr);
		mdio_delay(mdio_addr);
		retval = (retval << 1) | ((readl(mdio_addr) & MDIO_Data) ? 1 : 0);
		writel(MDIO_EnbIn | MDIO_ShiftClk, mdio_addr);
		mdio_delay(mdio_addr);
	}
	return (retval>>1) & 0xffff;
}

static void mdio_write(struct net_device *dev, int phy_id, int location, int value)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	long mdio_addr = dev->base_addr + MIICtrl;
	int mii_cmd = (0x5002 << 16) | (phy_id << 23) | (location<<18) | value;
	int i;

	if (location == 4  &&  phy_id == np->phys[0])
		np->advertising = value;

	if (mii_preamble_required)
		mdio_sync(mdio_addr);

	/* Shift the command bits out. */
	for (i = 31; i >= 0; i--) {
		int dataval = (mii_cmd & (1 << i)) ? MDIO_WRITE1 : MDIO_WRITE0;

		writel(dataval, mdio_addr);
		mdio_delay(mdio_addr);
		writel(dataval | MDIO_ShiftClk, mdio_addr);
		mdio_delay(mdio_addr);
	}
	/* Clear out extra bits. */
	for (i = 2; i > 0; i--) {
		writel(MDIO_EnbIn, mdio_addr);
		mdio_delay(mdio_addr);
		writel(MDIO_EnbIn | MDIO_ShiftClk, mdio_addr);
		mdio_delay(mdio_addr);
	}
	return;
}


static int netdev_open(struct net_device *dev)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	long ioaddr = dev->base_addr;
	int i;

	/* Some chips may need to be reset. */

	MOD_INC_USE_COUNT;

	/* Note that both request_irq() and init_ring() call kmalloc(), which
	   break the global kernel lock protecting this routine. */
	if (request_irq(dev->irq, &intr_handler, SA_SHIRQ, dev->name, dev)) {
		MOD_DEC_USE_COUNT;
		return -EAGAIN;
	}

	if (debug > 1)
		printk(KERN_DEBUG "%s: netdev_open() irq %d.\n",
			   dev->name, dev->irq);

	init_ring(dev);

#if ADDRLEN == 64
	writel(virt_to_bus(np->rx_ring) >> 32, ioaddr + RxRingPtr + 4);
	writel(virt_to_bus(np->tx_ring) >> 32, ioaddr + TxRingPtr + 4);
#endif
	writel(virt_to_bus(np->rx_ring), ioaddr + RxRingPtr);
	writel(virt_to_bus(np->tx_ring), ioaddr + TxRingPtr);

	for (i = 0; i < 6; i++)
		writeb(dev->dev_addr[i], ioaddr + StationAddr + i);

	/* Initialize other registers. */
	/* Configure the PCI bus bursts and FIFO thresholds. */
	writel(0x0000, ioaddr + PCIBusCfg);
	writel(0x0000, ioaddr + FIFOCfg);

	if (dev->if_port == 0)
		dev->if_port = np->default_port;

	dev->tbusy = 0;
	dev->interrupt = 0;
	np->in_interrupt = 0;

	set_rx_mode(dev);

	dev->start = 1;

	/* Enable interrupts by setting the interrupt mask. */
	writel(IntrRxDone | IntrRxEmpty | IntrRxPCIErr |
		   IntrTxDone | IntrTxEmpty | IntrTxPCIErr |
		   StatsMax | LinkChange, ioaddr + IntrEnable);

	writel(1, dev->base_addr + RxCmd);

	if (debug > 2)
		printk(KERN_DEBUG "%s: Done netdev_open(), status: Rx %x Tx %x.\n",
			   dev->name, readl(ioaddr + RxStatus), readl(ioaddr + TxStatus));

	/* Set the timer to check for link beat. */
	init_timer(&np->timer);
	np->timer.expires = jiffies + 3*HZ;
	np->timer.data = (unsigned long)dev;
	np->timer.function = &netdev_timer;				/* timer handler */
	add_timer(&np->timer);

	return 0;
}

static void check_duplex(struct net_device *dev)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	long ioaddr = dev->base_addr;
	int mii_reg5 = mdio_read(dev, np->phys[0], 5);
	int negotiated = mii_reg5 & np->advertising;
	int duplex;

	if (np->duplex_lock  ||  mii_reg5 == 0xffff)
		return;
	duplex = (negotiated & 0x0100) || (negotiated & 0x01C0) == 0x0040;
	if (np->full_duplex != duplex) {
		np->full_duplex = duplex;
		if (debug)
			printk(KERN_INFO "%s: Setting %s-duplex based on MII #%d "
				   "negotiated capability %4.4x.\n", dev->name,
				   duplex ? "full" : "half", np->phys[0], negotiated);
		writew(duplex ? 1 : 0, ioaddr + TxMode);
	}
}

static void netdev_timer(unsigned long data)
{
	struct net_device *dev = (struct net_device *)data;
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	long ioaddr = dev->base_addr;
	int next_tick = 10*HZ;

	if (debug > 3) {
		printk(KERN_DEBUG "%s: Media selection timer tick, status %8.8x, "
			   "Tx %x Rx %x.\n",
			   dev->name, readl(ioaddr + IntrStatus),
			   readl(ioaddr + TxStatus), readl(ioaddr + RxStatus));
	}
	/* This will either have a small false-trigger window or will not catch
	   tbusy incorrectly set when the queue is empty. */
	if (test_bit(0, (void*)&dev->tbusy) &&
		np->cur_tx - np->dirty_tx > 1  &&
		(jiffies - dev->trans_start) > TX_TIMEOUT) {
		tx_timeout(dev);
	}
	check_duplex(dev);
	np->timer.expires = jiffies + next_tick;
	add_timer(&np->timer);
}

static void tx_timeout(struct net_device *dev)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	long ioaddr = dev->base_addr;

	printk(KERN_WARNING "%s: Transmit timed out, status %8.8x,"
		   " resetting...\n", dev->name, readl(ioaddr + TxStatus));

#ifndef __alpha__
	{
		int i;
		printk(KERN_DEBUG "  Rx ring %8.8x: ", (int)np->rx_ring);
		for (i = 0; i < RX_RING_SIZE; i++)
			printk(" %8.8x", (unsigned int)np->rx_ring[i].status);
		printk("\n"KERN_DEBUG"  Tx ring %8.8x: ", (int)np->tx_ring);
		for (i = 0; i < TX_RING_SIZE; i++)
			printk(" %4.4x", np->tx_ring[i].status);
		printk("\n");
	}
#endif

	/* Perhaps we should reinitialize the hardware here. */
	dev->if_port = 0;
	/* Stop and restart the chip's Tx processes . */

	/* Trigger an immediate transmit demand. */

	dev->trans_start = jiffies;
	np->stats.tx_errors++;
	return;
}


/* Initialize the Rx and Tx rings, along with various 'dev' bits. */
static void init_ring(struct net_device *dev)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	int i;

	np->tx_full = 0;
	np->cur_rx = np->cur_tx = 0;
	np->dirty_rx = np->dirty_tx = 0;

	np->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
	np->rx_head_desc = &np->rx_ring[0];

	/* Initialize all Rx descriptors. */
	for (i = 0; i < RX_RING_SIZE; i++) {
		np->rx_ring[i].length = cpu_to_le32(np->rx_buf_sz);
		np->rx_ring[i].status = 0;
		/* np->rx_ring[i].next_desc = virt_to_le32desc(&np->rx_ring[i+1]);*/
		np->rx_skbuff[i] = 0;
	}
	/* Mark the last entry as wrapping the ring. */
	np->rx_ring[i-1].status |= cpu_to_le32(DescEndRing);
	/* Or np->rx_ring[i-1].next_desc = virt_to_le32desc(&np->rx_ring[0]);*/

	/* Fill in the Rx buffers.  Handle allocation failure gracefully. */
	for (i = 0; i < RX_RING_SIZE; i++) {
		struct sk_buff *skb = dev_alloc_skb(np->rx_buf_sz);
		np->rx_skbuff[i] = skb;
		if (skb == NULL)
			break;
		skb->dev = dev;			/* Mark as being used by this device. */
		skb_reserve(skb, 2);	/* 16 byte align the IP header. */
		np->rx_ring[i].addr = virt_to_le32desc(skb->tail);
		np->rx_ring[i].status = cpu_to_le32(DescOwn | DescIntr);
	}
	np->dirty_rx = (unsigned int)(i - RX_RING_SIZE);

	for (i = 0; i < TX_RING_SIZE; i++) {
		np->tx_skbuff[i] = 0;
		np->tx_ring[i].status = 0;
		/* Or np->tx_ring[i].next_desc = virt_to_le32desc(&np->tx_ring[i+1]);*/
	}
	/* Or np->tx_ring[i-1].next_desc = virt_to_le32desc(&np->tx_ring[0]); */
	return;
}

static int start_tx(struct sk_buff *skb, struct net_device *dev)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	unsigned entry;

	/* Block a timer-based transmit from overlapping.  This could better be
	   done with atomic_swap(1, dev->tbusy), but set_bit() works as well. */
	if (test_and_set_bit(0, (void*)&dev->tbusy) != 0) {
		/* This watchdog code is redundant with the media monitor timer. */
		if (jiffies - dev->trans_start > TX_TIMEOUT)
			tx_timeout(dev);
		return 1;
	}

	/* 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 = np->cur_tx % TX_RING_SIZE;

	np->tx_skbuff[entry] = skb;

	np->tx_ring[entry].addr = virt_to_le32desc(skb->data);
	np->tx_ring[entry].length = cpu_to_le32(skb->len);
	if (entry >= TX_RING_SIZE-1)		 /* Wrap ring */
		np->tx_ring[entry].status =
			cpu_to_le32(DescOwn|DescEndPacket|DescEndRing);
	else
		np->tx_ring[entry].status = cpu_to_le32(DescOwn|DescEndPacket);
	np->cur_tx++;

	/* On some architectures: explicitly flush cache lines here. */

	if (np->cur_tx - np->dirty_tx < TX_QUEUE_LEN - 1)
		clear_bit(0, (void*)&dev->tbusy);		/* Typical path */
	else
		np->tx_full = 1;
	/* Wake the potentially-idle transmit channel. */
	writel(1, dev->base_addr + TxCmd);

	dev->trans_start = jiffies;

	if (debug > 4) {
		printk(KERN_DEBUG "%s: Transmit frame #%d queued in slot %d.\n",
			   dev->name, np->cur_tx, entry);
	}
	return 0;
}

/* The interrupt handler does all of the Rx thread work and cleans up
   after the Tx thread. */
static void intr_handler(int irq, void *dev_instance, struct pt_regs *rgs)
{
	struct net_device *dev = (struct net_device *)dev_instance;
	struct netdev_private *np;
	long ioaddr;
	int boguscnt = max_interrupt_work;

#ifndef final_version			/* Can never occur. */
	if (dev == NULL) {
		printk (KERN_ERR "Netdev interrupt handler(): IRQ %d for unknown "
				"device.\n", irq);
		return;
	}
#endif

	ioaddr = dev->base_addr;
	np = (struct netdev_private *)dev->priv;
#if defined(__i386__)
	/* A lock to prevent simultaneous entry bug on Intel SMP machines. */
	if (test_and_set_bit(0, (void*)&dev->interrupt)) {
		printk(KERN_ERR"%s: SMP simultaneous entry of an interrupt handler.\n",
			   dev->name);
		dev->interrupt = 0;	/* Avoid halting machine. */
		return;
	}
#else
	if (dev->interrupt) {
		printk(KERN_ERR "%s: Re-entering the interrupt handler.\n", dev->name);
		return;
	}
	dev->interrupt = 1;
#endif

	do {
		u32 intr_status = readl(ioaddr + IntrStatus);

		/* Acknowledge all of the current interrupt sources ASAP. */
		writel(intr_status & 0x000ffff, ioaddr + IntrStatus);

		if (debug > 4)
			printk(KERN_DEBUG "%s: Interrupt, status %4.4x.\n",
				   dev->name, intr_status);

		if (intr_status == 0)
			break;

		if (intr_status & IntrRxDone)
			netdev_rx(dev);

		for (; np->cur_tx - np->dirty_tx > 0; np->dirty_tx++) {
			int entry = np->dirty_tx % TX_RING_SIZE;
			if (np->tx_ring[entry].status == 0)
				break;
			np->stats.tx_packets++;
#if LINUX_VERSION_CODE > 0x20127
			np->stats.tx_bytes += np->tx_skbuff[entry]->len;
#endif
			/* Free the original skb. */
			dev_free_skb(np->tx_skbuff[entry]);
			np->tx_skbuff[entry] = 0;
		}
		/* Note the 4 slot hysteresis in mark the queue non-full. */
		if (np->tx_full && dev->tbusy
			&& np->cur_tx - np->dirty_tx < TX_QUEUE_LEN - 4) {
			/* The ring is no longer full, clear tbusy. */
			np->tx_full = 0;
			clear_bit(0, (void*)&dev->tbusy);
			netif_wake_queue(dev);
		}

		/* Abnormal error summary/uncommon events handlers. */
		if (intr_status & (IntrTxPCIErr | IntrRxPCIErr | LinkChange | StatsMax))
			netdev_error(dev, intr_status);

		if (--boguscnt < 0) {
			printk(KERN_WARNING "%s: Too much work at interrupt, "
				   "status=0x%4.4x.\n",
				   dev->name, intr_status);
			break;
		}
	} while (1);

	if (debug > 3)
		printk(KERN_DEBUG "%s: exiting interrupt, status=%#4.4x.\n",
			   dev->name, readl(ioaddr + IntrStatus));

#ifndef final_version
	/* Code that should never be run!  Perhaps remove after testing.. */
	{
		static int stopit = 10;
		if (dev->start == 0  &&  --stopit < 0) {
			printk(KERN_ERR "%s: Emergency stop, looping startup interrupt.\n",
				   dev->name);
			free_irq(irq, dev);
		}
	}
#endif

#if defined(__i386__)
	clear_bit(0, (void*)&dev->interrupt);
#else
	dev->interrupt = 0;
#endif
	return;
}

/* This routine is logically part of the interrupt handler, but separated
   for clarity and better register allocation. */
static int netdev_rx(struct net_device *dev)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	int entry = np->cur_rx % RX_RING_SIZE;
	int boguscnt = np->dirty_rx + RX_RING_SIZE - np->cur_rx;

	if (debug > 4) {
		printk(KERN_DEBUG " In netdev_rx(), entry %d status %4.4x.\n",
			   entry, np->rx_ring[entry].status);
	}

	/* If EOP is set on the next entry, it's a new packet. Send it up. */
	while ( ! (np->rx_head_desc->status & cpu_to_le32(DescOwn))) {
		struct netdev_desc *desc = np->rx_head_desc;
		u32 desc_status = le32_to_cpu(desc->status);
		int data_size = le32_to_cpu(desc->length);

		if (debug > 4)
			printk(KERN_DEBUG "  netdev_rx() status was %8.8x.\n",
				   desc_status);
		if (--boguscnt < 0)
			break;
		if ( ! (desc_status & DescEndPacket)) {
			/* Select a message. */
			printk(KERN_WARNING "%s: Oversized Ethernet frame spanned "
				   "multiple buffers, entry %#x length %d status %4.4x!\n",
				   dev->name, np->cur_rx, data_size, desc_status);
			printk(KERN_WARNING "%s: Oversized Ethernet frame %p vs %p.\n",
				   dev->name, np->rx_head_desc,
				   &np->rx_ring[np->cur_rx % RX_RING_SIZE]);
			printk(KERN_WARNING "%s: Oversized Ethernet frame -- next status %x/%x last status %x.\n",
				   dev->name,
				   np->rx_ring[(np->cur_rx+1) % RX_RING_SIZE].status,
				   np->rx_ring[(np->cur_rx+1) % RX_RING_SIZE].length,
				   np->rx_ring[(np->cur_rx-1) % RX_RING_SIZE].status);
			np->stats.rx_length_errors++;
		} else if (desc_status & 0x0038) {
			/* There was a error. */
			if (debug > 2)
				printk(KERN_DEBUG "  netdev_rx() Rx error was %8.8x.\n",
					   desc_status);
			np->stats.rx_errors++;
			if (desc_status & 0x0060) np->stats.rx_length_errors++;
			if (desc_status & 0x0008) np->stats.rx_frame_errors++;
			if (desc_status & 0x0010) np->stats.rx_crc_errors++;
			if (desc_status < 0) np->stats.rx_dropped++;
		} else {
			struct sk_buff *skb;
			/* Reported length should omit the CRC. */
			u16 pkt_len = data_size - 4;

#ifndef final_version
			if (debug > 4)
				printk(KERN_DEBUG "  netdev_rx() normal Rx pkt length %d"
					   " of %d, bogus_cnt %d.\n",
					   pkt_len, data_size, boguscnt);
#endif
			/* Check if the packet is long enough to accept without copying
			   to a minimally-sized skbuff. */
			if (pkt_len < rx_copybreak
				&& (skb = dev_alloc_skb(pkt_len + 2)) != NULL) {
				skb->dev = dev;
				skb_reserve(skb, 2);	/* 16 byte align the IP header */
#if HAS_IP_COPYSUM			/* Call copy + cksum if available. */
				eth_copy_and_sum(skb, np->rx_skbuff[entry]->tail, pkt_len, 0);
				skb_put(skb, pkt_len);
#else
				memcpy(skb_put(skb, pkt_len), np->rx_skbuff[entry]->tail,
					   pkt_len);
#endif
			} else {
				char *temp = skb_put(skb = np->rx_skbuff[entry], pkt_len);
				np->rx_skbuff[entry] = NULL;
#ifndef final_version				/* Remove after testing. */
				if (le32desc_to_virt(np->rx_ring[entry].addr) != temp)
					printk(KERN_ERR "%s: Internal fault: The skbuff addresses "
						   "do not match in netdev_rx: %p vs. %p / %p.\n",
						   dev->name,
						   le32desc_to_virt(np->rx_ring[entry].addr),
						   skb->head, temp);
#endif
			}
#ifndef final_version				/* Remove after testing. */
			/* You will want this info for the initial debug. */
			if (debug > 5)
				printk(KERN_DEBUG "  Rx data %2.2x:%2.2x:%2.2x:%2.2x:%2.2x:"
					   "%2.2x %2.2x:%2.2x:%2.2x:%2.2x:%2.2x:%2.2x %2.2x%2.2x "
					   "%d.%d.%d.%d.\n",
					   skb->data[0], skb->data[1], skb->data[2], skb->data[3],
					   skb->data[4], skb->data[5], skb->data[6], skb->data[7],
					   skb->data[8], skb->data[9], skb->data[10],
					   skb->data[11], skb->data[12], skb->data[13],
					   skb->data[14], skb->data[15], skb->data[16],
					   skb->data[17]);
#endif
			skb->protocol = eth_type_trans(skb, dev);
			/* Note: checksum -> skb->ip_summed = CHECKSUM_UNNECESSARY; */
			netif_rx(skb);
			dev->last_rx = jiffies;
			np->stats.rx_packets++;
#if LINUX_VERSION_CODE > 0x20127
			np->stats.rx_bytes += pkt_len;
#endif
		}
		entry = (++np->cur_rx) % RX_RING_SIZE;
		np->rx_head_desc = &np->rx_ring[entry];
	}

	/* Refill the Rx ring buffers. */
	for (; np->cur_rx - np->dirty_rx > 0; np->dirty_rx++) {
		struct sk_buff *skb;
		entry = np->dirty_rx % RX_RING_SIZE;
		if (np->rx_skbuff[entry] == NULL) {
			skb = dev_alloc_skb(np->rx_buf_sz);
			np->rx_skbuff[entry] = skb;
			if (skb == NULL)
				break;				/* Better luck next round. */
			skb->dev = dev;			/* Mark as being used by this device. */
			skb_reserve(skb, 2);	/* Align IP on 16 byte boundaries */
			np->rx_ring[entry].addr = virt_to_le32desc(skb->tail);
		}
		np->rx_ring[entry].length = cpu_to_le32(np->rx_buf_sz);
		np->rx_ring[entry].status = (entry == RX_RING_SIZE - 1)
			? cpu_to_le32(DescOwn | DescEndPacket | DescEndRing | DescIntr)
			: cpu_to_le32(DescOwn | DescEndPacket | DescIntr);
	}

	/* Restart Rx engine if stopped. */
	writel(1, dev->base_addr + RxCmd);
	return 0;
}

static void netdev_error(struct net_device *dev, int intr_status)
{
	struct netdev_private *np = (struct netdev_private *)dev->priv;

	if (intr_status & LinkChange) {
		printk(KERN_ERR "%s: Link changed: Autonegotiation advertising"
			   " %4.4x  partner %4.4x.\n", dev->name,
			   mdio_read(dev, np->phys[0], 4),
			   mdio_read(dev, np->phys[0], 5));
		check_duplex(dev);
	}
	if (intr_status & StatsMax) {
		get_stats(dev);
	}
	if ((intr_status & ~(LinkChange|StatsMax)) && debug)
		printk(KERN_ERR "%s: Something Wicked happened! %4.4x.\n",
			   dev->name, intr_status);
	/* Hmmmmm, it's not clear how to recover from PCI faults. */
	if (intr_status & IntrTxPCIErr)
		np->stats.tx_fifo_errors++;
	if (intr_status & IntrRxPCIErr)
		np->stats.rx_fifo_errors++;
}

static struct net_device_stats *get_stats(struct net_device *dev)
{
	long ioaddr = dev->base_addr;
	struct netdev_private *np = (struct netdev_private *)dev->priv;

	/* We should lock this segment of code for SMP eventually, although
	   the vulnerability window is very small and statistics are
	   non-critical. */
	/* The chip only need report frame silently dropped. */
	np->stats.rx_crc_errors	+= readl(ioaddr + RxCRCErrs);
	np->stats.rx_missed_errors	+= readl(ioaddr + RxMissed);

	return &np->stats;
}

/* The little-endian AUTODIN II ethernet CRC calculations.
   A big-endian version is also available.
   This is slow but compact code.  Do not use this routine for bulk data,
   use a table-based routine instead.
   This is common code and should be moved to net/core/crc.c.
   Chips may use the upper or lower CRC bits, and may reverse and/or invert
   them.  Select the endian-ness that results in minimal calculations.
*/
static unsigned const ethernet_polynomial_le = 0xedb88320U;
static inline unsigned ether_crc_le(int length, unsigned char *data)
{
	unsigned int crc = 0xffffffff;	/* Initial value. */
	while(--length >= 0) {
		unsigned char current_octet = *data++;
		int bit;
		for (bit = 8; --bit >= 0; current_octet >>= 1) {
			if ((crc ^ current_octet) & 1) {
				crc >>= 1;
				crc ^= ethernet_polynomial_le;
			} else
				crc >>= 1;
		}
	}
	return crc;
}

static void set_rx_mode(struct net_device *dev)
{
	long ioaddr = dev->base_addr;
	u32 mc_filter[2];			/* Multicast hash filter */
	u32 rx_mode;

	if (dev->flags & IFF_PROMISC) {			/* Set promiscuous. */
		/* Unconditionally log net taps. */
		printk(KERN_NOTICE "%s: Promiscuous mode enabled.\n", dev->name);
		memset(mc_filter, 0xff, sizeof(mc_filter));
		rx_mode = AcceptBroadcast | AcceptMulticast | AcceptAllPhys
			| AcceptMyPhys;
	} else if ((dev->mc_count > multicast_filter_limit)
			   ||  (dev->flags & IFF_ALLMULTI)) {
		/* Too many to match, or accept all multicasts. */
		memset(mc_filter, 0xff, sizeof(mc_filter));
		rx_mode = AcceptBroadcast | AcceptMulticast | AcceptMyPhys;
	} else {
		struct dev_mc_list *mclist;
		int i;
		memset(mc_filter, 0, sizeof(mc_filter));
		for (i = 0, mclist = dev->mc_list; mclist && i < dev->mc_count;
			 i++, mclist = mclist->next) {
			set_bit(ether_crc_le(ETH_ALEN, mclist->dmi_addr) & 0x3f,
					mc_filter);
		}
		rx_mode = AcceptBroadcast | AcceptMulticast | AcceptMyPhys;
	}
	writel(mc_filter[0], ioaddr + MulticastFilter0);
	writel(mc_filter[1], ioaddr + MulticastFilter1);
	writel(rx_mode, ioaddr + RxMode);
}

static int mii_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
{
	u16 *data = (u16 *)&rq->ifr_data;

	switch(cmd) {
	case SIOCDEVPRIVATE:		/* Get the address of the PHY in use. */
		data[0] = ((struct netdev_private *)dev->priv)->phys[0] & 0x1f;
		/* Fall Through */
	case SIOCDEVPRIVATE+1:		/* Read the specified MII register. */
		data[3] = mdio_read(dev, data[0] & 0x1f, data[1] & 0x1f);
		return 0;
	case SIOCDEVPRIVATE+2:		/* Write the specified MII register */
		if (!capable(CAP_NET_ADMIN))
			return -EPERM;
		mdio_write(dev, data[0] & 0x1f, data[1] & 0x1f, data[2]);
		return 0;
	default:
		return -EOPNOTSUPP;
	}
}

static int netdev_close(struct net_device *dev)
{
	long ioaddr = dev->base_addr;
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	int i;

	dev->start = 0;
	dev->tbusy = 1;

	if (debug > 1) {
		printk(KERN_DEBUG "%s: Shutting down ethercard, status was Tx %4.4x "
			   "Rx %4.4x Int %2.2x.\n",
			   dev->name, readl(ioaddr + TxStatus),
			   readl(ioaddr + RxStatus), readl(ioaddr + IntrStatus));
		printk(KERN_DEBUG "%s: Queue pointers were Tx %d / %d,  Rx %d / %d.\n",
			   dev->name, np->cur_tx, np->dirty_tx, np->cur_rx, np->dirty_rx);
	}

	/* Disable interrupts by clearing the interrupt mask. */
	writel(0x0000, ioaddr + IntrEnable);

	/* Stop the chip's Tx and Rx processes. */
	writel(2, ioaddr + RxCmd);
	writew(2, ioaddr + TxCmd);

	del_timer(&np->timer);

#ifdef __i386__
	if (debug > 2) {
		printk("\n"KERN_DEBUG"  Tx ring at %8.8x:\n",
			   (int)virt_to_bus(np->tx_ring));
		for (i = 0; i < TX_RING_SIZE; i++)
			printk(" #%d desc. %4.4x %4.4x %8.8x.\n",
				   i, np->tx_ring[i].length,
				   np->tx_ring[i].status, np->tx_ring[i].addr);
		printk("\n"KERN_DEBUG "  Rx ring %8.8x:\n",
			   (int)virt_to_bus(np->rx_ring));
		for (i = 0; i < RX_RING_SIZE; i++) {
			printk(KERN_DEBUG " #%d desc. %4.4x %4.4x %8.8x\n",
				   i, np->rx_ring[i].length,
				   np->rx_ring[i].status, np->rx_ring[i].addr);
			if (debug > 6) {
				if (*(u8*)np->rx_ring[i].addr != 0x69) {
					int j;
					for (j = 0; j < 0x50; j++)
						printk(" %4.4x", ((u16*)np->rx_ring[i].addr)[j]);
					printk("\n");
				}
			}
		}
	}
#endif /* __i386__ debugging only */

	free_irq(dev->irq, dev);

	/* Free all the skbuffs in the Rx queue. */
	for (i = 0; i < RX_RING_SIZE; i++) {
		np->rx_ring[i].status = 0;
		np->rx_ring[i].addr = 0xBADF00D0; /* An invalid address. */
		if (np->rx_skbuff[i]) {
#if LINUX_VERSION_CODE < 0x20100
			np->rx_skbuff[i]->free = 1;
#endif
			dev_free_skb(np->rx_skbuff[i]);
		}
		np->rx_skbuff[i] = 0;
	}
	for (i = 0; i < TX_RING_SIZE; i++) {
		if (np->tx_skbuff[i])
			dev_free_skb(np->tx_skbuff[i]);
		np->tx_skbuff[i] = 0;
	}

	writel(0x00, ioaddr + LEDCtrl);

	MOD_DEC_USE_COUNT;

	return 0;
}

static int netdev_pwr_event(void *dev_instance, int event)
{
	struct net_device *dev = dev_instance;
	struct netdev_private *np = (struct netdev_private *)dev->priv;
	long ioaddr = dev->base_addr;

	if (debug > 1)
		printk("%s: Handling power event %d.\n", dev->name, event);
	switch(event) {
	case DRV_ATTACH:
		MOD_INC_USE_COUNT;
		break;
	case DRV_SUSPEND:
		/* Disable interrupts, stop Tx and Rx. */
		writel(0x0000, ioaddr + IntrEnable);
		writel(2, ioaddr + RxCmd);
		writew(2, ioaddr + TxCmd);
		break;
	case DRV_RESUME:
		/* This is incomplete: the actions are very chip specific. */
		set_rx_mode(dev);
		break;
	case DRV_DETACH: {
		struct net_device **devp, **next;
		if (dev->flags & IFF_UP) {
			/* Some, but not all, kernel versions close automatically. */
			dev_close(dev);
			dev->flags &= ~(IFF_UP|IFF_RUNNING);
		}
		unregister_netdev(dev);
		release_region(dev->base_addr, pci_id_tbl[np->chip_id].io_size);
#ifndef USE_IO_OPS
		iounmap((char *)dev->base_addr);
#endif
		for (devp = &root_net_dev; *devp; devp = next) {
			next = &((struct netdev_private *)(*devp)->priv)->next_module;
			if (*devp == dev) {
				*devp = *next;
				break;
			}
		}
		if (np->priv_addr)
			kfree(np->priv_addr);
		kfree(dev);
		MOD_DEC_USE_COUNT;
		break;
	}
	}

	return 0;
}


#ifdef MODULE
int init_module(void)
{
	if (debug)					/* Emit version even if no cards detected. */
		printk(KERN_INFO "%s" KERN_INFO "%s", version1, version2);
	return pci_drv_register(&netfin_drv_id, NULL);
}

void cleanup_module(void)
{
	struct net_device *next_dev;

	pci_drv_unregister(&netfin_drv_id);

	/* No need to check MOD_IN_USE, as sys_delete_module() checks. */
	while (root_net_dev) {
		struct netdev_private *np = (void *)(root_net_dev->priv);
		unregister_netdev(root_net_dev);
#ifdef USE_IO_OPS
		release_region(root_net_dev->base_addr,
					   pci_id_tbl[np->chip_id].io_size);
#else
		iounmap((char *)(root_net_dev->base_addr));
#endif
		next_dev = np->next_module;
		if (np->priv_addr)
			kfree(np->priv_addr);
		kfree(root_net_dev);
		root_net_dev = next_dev;
	}
}

#endif  /* MODULE */

/*
 * Local variables:
 *  compile-command: "gcc -DMODULE -Wall -Wstrict-prototypes -O6 -c pci-skeleton.c"
 *  simple-compile-command: "gcc -DMODULE -O6 -c pci-skeleton.c"
 *  c-indent-level: 4
 *  c-basic-offset: 4
 *  tab-width: 4
 * End:
 */