path: root/drivers/net/cassini.c

/* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
 *
 * Copyright (C) 2004 Sun Microsystems Inc.
 * Copyright (C) 2003 Adrian Sun (asun@darksunrising.com)
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation; either version 2 of the
 * License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
 * 02111-1307, USA.
 *
 * This driver uses the sungem driver (c) David Miller
 * (davem@redhat.com) as its basis.
 *
 * The cassini chip has a number of features that distinguish it from
 * the gem chip:
 *  4 transmit descriptor rings that are used for either QoS (VLAN) or
 *      load balancing (non-VLAN mode)
 *  batching of multiple packets
 *  multiple CPU dispatching
 *  page-based RX descriptor engine with separate completion rings
 *  Gigabit support (GMII and PCS interface)
 *  MIF link up/down detection works
 *
 * RX is handled by page sized buffers that are attached as fragments to
 * the skb. here's what's done:
 *  -- driver allocates pages at a time and keeps reference counts
 *     on them.
 *  -- the upper protocol layers assume that the header is in the skb
 *     itself. as a result, cassini will copy a small amount (64 bytes)
 *     to make them happy.
 *  -- driver appends the rest of the data pages as frags to skbuffs
 *     and increments the reference count
 *  -- on page reclamation, the driver swaps the page with a spare page.
 *     if that page is still in use, it frees its reference to that page,
 *     and allocates a new page for use. otherwise, it just recycles the
 *     the page.
 *
 * NOTE: cassini can parse the header. however, it's not worth it
 *       as long as the network stack requires a header copy.
 *
 * TX has 4 queues. currently these queues are used in a round-robin
 * fashion for load balancing. They can also be used for QoS. for that
 * to work, however, QoS information needs to be exposed down to the driver
 * level so that subqueues get targetted to particular transmit rings.
 * alternatively, the queues can be configured via use of the all-purpose
 * ioctl.
 *
 * RX DATA: the rx completion ring has all the info, but the rx desc
 * ring has all of the data. RX can conceivably come in under multiple
 * interrupts, but the INT# assignment needs to be set up properly by
 * the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
 * that. also, the two descriptor rings are designed to distinguish between
 * encrypted and non-encrypted packets, but we use them for buffering
 * instead.
 *
 * by default, the selective clear mask is set up to process rx packets.
 */

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/compiler.h>
#include <linux/slab.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/vmalloc.h>
#include <linux/ioport.h>
#include <linux/pci.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/list.h>
#include <linux/dma-mapping.h>

#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/ethtool.h>
#include <linux/crc32.h>
#include <linux/random.h>
#include <linux/mii.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/mutex.h>
#include <linux/firmware.h>

#include <net/checksum.h>

#include <asm/atomic.h>
#include <asm/system.h>
#include <asm/io.h>
#include <asm/byteorder.h>
#include <asm/uaccess.h>

#define cas_page_map(x)      kmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
#define cas_page_unmap(x)    kunmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
#define CAS_NCPUS            num_online_cpus()

#define cas_skb_release(x)  netif_rx(x)

/* select which firmware to use */
#define USE_HP_WORKAROUND
#define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
#define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */

#include "cassini.h"

#define USE_TX_COMPWB      /* use completion writeback registers */
#define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
#define USE_RX_BLANK       /* hw interrupt mitigation */
#undef USE_ENTROPY_DEV     /* don't test for entropy device */

/* NOTE: these aren't useable unless PCI interrupts can be assigned.
 * also, we need to make cp->lock finer-grained.
 */
#undef  USE_PCI_INTB
#undef  USE_PCI_INTC
#undef  USE_PCI_INTD
#undef  USE_QOS

#undef  USE_VPD_DEBUG       /* debug vpd information if defined */

/* rx processing options */
#define USE_PAGE_ORDER      /* specify to allocate large rx pages */
#define RX_DONT_BATCH  0    /* if 1, don't batch flows */
#define RX_COPY_ALWAYS 0    /* if 0, use frags */
#define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
#undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */

#define DRV_MODULE_NAME		"cassini"
#define DRV_MODULE_VERSION	"1.6"
#define DRV_MODULE_RELDATE	"21 May 2008"

#define CAS_DEF_MSG_ENABLE	  \
	(NETIF_MSG_DRV		| \
	 NETIF_MSG_PROBE	| \
	 NETIF_MSG_LINK		| \
	 NETIF_MSG_TIMER	| \
	 NETIF_MSG_IFDOWN	| \
	 NETIF_MSG_IFUP		| \
	 NETIF_MSG_RX_ERR	| \
	 NETIF_MSG_TX_ERR)

/* length of time before we decide the hardware is borked,
 * and dev->tx_timeout() should be called to fix the problem
 */
#define CAS_TX_TIMEOUT			(HZ)
#define CAS_LINK_TIMEOUT                (22*HZ/10)
#define CAS_LINK_FAST_TIMEOUT           (1)

/* timeout values for state changing. these specify the number
 * of 10us delays to be used before giving up.
 */
#define STOP_TRIES_PHY 1000
#define STOP_TRIES     5000

/* specify a minimum frame size to deal with some fifo issues
 * max mtu == 2 * page size - ethernet header - 64 - swivel =
 *            2 * page_size - 0x50
 */
#define CAS_MIN_FRAME			97
#define CAS_1000MB_MIN_FRAME            255
#define CAS_MIN_MTU                     60
#define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)

#if 1
/*
 * Eliminate these and use separate atomic counters for each, to
 * avoid a race condition.
 */
#else
#define CAS_RESET_MTU                   1
#define CAS_RESET_ALL                   2
#define CAS_RESET_SPARE                 3
#endif

static char version[] __devinitdata =
	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";

static int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
static int link_mode;

MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
MODULE_LICENSE("GPL");
MODULE_FIRMWARE("sun/cassini.bin");
module_param(cassini_debug, int, 0);
MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
module_param(link_mode, int, 0);
MODULE_PARM_DESC(link_mode, "default link mode");

/*
 * Work around for a PCS bug in which the link goes down due to the chip
 * being confused and never showing a link status of "up."
 */
#define DEFAULT_LINKDOWN_TIMEOUT 5
/*
 * Value in seconds, for user input.
 */
static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
module_param(linkdown_timeout, int, 0);
MODULE_PARM_DESC(linkdown_timeout,
"min reset interval in sec. for PCS linkdown issue; disabled if not positive");

/*
 * value in 'ticks' (units used by jiffies). Set when we init the
 * module because 'HZ' in actually a function call on some flavors of
 * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
 */
static int link_transition_timeout;



static u16 link_modes[] __devinitdata = {
	BMCR_ANENABLE,			 /* 0 : autoneg */
	0,				 /* 1 : 10bt half duplex */
	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
};

static DEFINE_PCI_DEVICE_TABLE(cas_pci_tbl) = {
	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
	{ 0, }
};

MODULE_DEVICE_TABLE(pci, cas_pci_tbl);

static void cas_set_link_modes(struct cas *cp);

static inline void cas_lock_tx(struct cas *cp)
{
	int i;

	for (i = 0; i < N_TX_RINGS; i++)
		spin_lock(&cp->tx_lock[i]);
}

static inline void cas_lock_all(struct cas *cp)
{
	spin_lock_irq(&cp->lock);
	cas_lock_tx(cp);
}

/* WTZ: QA was finding deadlock problems with the previous
 * versions after long test runs with multiple cards per machine.
 * See if replacing cas_lock_all with safer versions helps. The
 * symptoms QA is reporting match those we'd expect if interrupts
 * aren't being properly restored, and we fixed a previous deadlock
 * with similar symptoms by using save/restore versions in other
 * places.
 */
#define cas_lock_all_save(cp, flags) \
do { \
	struct cas *xxxcp = (cp); \
	spin_lock_irqsave(&xxxcp->lock, flags); \
	cas_lock_tx(xxxcp); \
} while (0)

static inline void cas_unlock_tx(struct cas *cp)
{
	int i;

	for (i = N_TX_RINGS; i > 0; i--)
		spin_unlock(&cp->tx_lock[i - 1]);
}

static inline void cas_unlock_all(struct cas *cp)
{
	cas_unlock_tx(cp);
	spin_unlock_irq(&cp->lock);
}

#define cas_unlock_all_restore(cp, flags) \
do { \
	struct cas *xxxcp = (cp); \
	cas_unlock_tx(xxxcp); \
	spin_unlock_irqrestore(&xxxcp->lock, flags); \
} while (0)

static void cas_disable_irq(struct cas *cp, const int ring)
{
	/* Make sure we won't get any more interrupts */
	if (ring == 0) {
		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
		return;
	}

	/* disable completion interrupts and selectively mask */
	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		switch (ring) {
#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
#ifdef USE_PCI_INTB
		case 1:
#endif
#ifdef USE_PCI_INTC
		case 2:
#endif
#ifdef USE_PCI_INTD
		case 3:
#endif
			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
			       cp->regs + REG_PLUS_INTRN_MASK(ring));
			break;
#endif
		default:
			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
			       REG_PLUS_INTRN_MASK(ring));
			break;
		}
	}
}

static inline void cas_mask_intr(struct cas *cp)
{
	int i;

	for (i = 0; i < N_RX_COMP_RINGS; i++)
		cas_disable_irq(cp, i);
}

static void cas_enable_irq(struct cas *cp, const int ring)
{
	if (ring == 0) { /* all but TX_DONE */
		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
		return;
	}

	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		switch (ring) {
#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
#ifdef USE_PCI_INTB
		case 1:
#endif
#ifdef USE_PCI_INTC
		case 2:
#endif
#ifdef USE_PCI_INTD
		case 3:
#endif
			writel(INTRN_MASK_RX_EN, cp->regs +
			       REG_PLUS_INTRN_MASK(ring));
			break;
#endif
		default:
			break;
		}
	}
}

static inline void cas_unmask_intr(struct cas *cp)
{
	int i;

	for (i = 0; i < N_RX_COMP_RINGS; i++)
		cas_enable_irq(cp, i);
}

static inline void cas_entropy_gather(struct cas *cp)
{
#ifdef USE_ENTROPY_DEV
	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
		return;

	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
			    readl(cp->regs + REG_ENTROPY_IV),
			    sizeof(uint64_t)*8);
#endif
}

static inline void cas_entropy_reset(struct cas *cp)
{
#ifdef USE_ENTROPY_DEV
	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
		return;

	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
	       cp->regs + REG_BIM_LOCAL_DEV_EN);
	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);

	/* if we read back 0x0, we don't have an entropy device */
	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
#endif
}

/* access to the phy. the following assumes that we've initialized the MIF to
 * be in frame rather than bit-bang mode
 */
static u16 cas_phy_read(struct cas *cp, int reg)
{
	u32 cmd;
	int limit = STOP_TRIES_PHY;

	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
	cmd |= MIF_FRAME_TURN_AROUND_MSB;
	writel(cmd, cp->regs + REG_MIF_FRAME);

	/* poll for completion */
	while (limit-- > 0) {
		udelay(10);
		cmd = readl(cp->regs + REG_MIF_FRAME);
		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
			return cmd & MIF_FRAME_DATA_MASK;
	}
	return 0xFFFF; /* -1 */
}

static int cas_phy_write(struct cas *cp, int reg, u16 val)
{
	int limit = STOP_TRIES_PHY;
	u32 cmd;

	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
	cmd |= MIF_FRAME_TURN_AROUND_MSB;
	cmd |= val & MIF_FRAME_DATA_MASK;
	writel(cmd, cp->regs + REG_MIF_FRAME);

	/* poll for completion */
	while (limit-- > 0) {
		udelay(10);
		cmd = readl(cp->regs + REG_MIF_FRAME);
		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
			return 0;
	}
	return -1;
}

static void cas_phy_powerup(struct cas *cp)
{
	u16 ctl = cas_phy_read(cp, MII_BMCR);

	if ((ctl & BMCR_PDOWN) == 0)
		return;
	ctl &= ~BMCR_PDOWN;
	cas_phy_write(cp, MII_BMCR, ctl);
}

static void cas_phy_powerdown(struct cas *cp)
{
	u16 ctl = cas_phy_read(cp, MII_BMCR);

	if (ctl & BMCR_PDOWN)
		return;
	ctl |= BMCR_PDOWN;
	cas_phy_write(cp, MII_BMCR, ctl);
}

/* cp->lock held. note: the last put_page will free the buffer */
static int cas_page_free(struct cas *cp, cas_page_t *page)
{
	pci_unmap_page(cp->pdev, page->dma_addr, cp->page_size,
		       PCI_DMA_FROMDEVICE);
	__free_pages(page->buffer, cp->page_order);
	kfree(page);
	return 0;
}

#ifdef RX_COUNT_BUFFERS
#define RX_USED_ADD(x, y)       ((x)->used += (y))
#define RX_USED_SET(x, y)       ((x)->used  = (y))
#else
#define RX_USED_ADD(x, y)
#define RX_USED_SET(x, y)
#endif

/* local page allocation routines for the receive buffers. jumbo pages
 * require at least 8K contiguous and 8K aligned buffers.
 */
static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
{
	cas_page_t *page;

	page = kmalloc(sizeof(cas_page_t), flags);
	if (!page)
		return NULL;

	INIT_LIST_HEAD(&page->list);
	RX_USED_SET(page, 0);
	page->buffer = alloc_pages(flags, cp->page_order);
	if (!page->buffer)
		goto page_err;
	page->dma_addr = pci_map_page(cp->pdev, page->buffer, 0,
				      cp->page_size, PCI_DMA_FROMDEVICE);
	return page;

page_err:
	kfree(page);
	return NULL;
}

/* initialize spare pool of rx buffers, but allocate during the open */
static void cas_spare_init(struct cas *cp)
{
  	spin_lock(&cp->rx_inuse_lock);
	INIT_LIST_HEAD(&cp->rx_inuse_list);
	spin_unlock(&cp->rx_inuse_lock);

	spin_lock(&cp->rx_spare_lock);
	INIT_LIST_HEAD(&cp->rx_spare_list);
	cp->rx_spares_needed = RX_SPARE_COUNT;
	spin_unlock(&cp->rx_spare_lock);
}

/* used on close. free all the spare buffers. */
static void cas_spare_free(struct cas *cp)
{
	struct list_head list, *elem, *tmp;

	/* free spare buffers */
	INIT_LIST_HEAD(&list);
	spin_lock(&cp->rx_spare_lock);
	list_splice_init(&cp->rx_spare_list, &list);
	spin_unlock(&cp->rx_spare_lock);
	list_for_each_safe(elem, tmp, &list) {
		cas_page_free(cp, list_entry(elem, cas_page_t, list));
	}

	INIT_LIST_HEAD(&list);
#if 1
	/*
	 * Looks like Adrian had protected this with a different
	 * lock than used everywhere else to manipulate this list.
	 */
	spin_lock(&cp->rx_inuse_lock);
	list_splice_init(&cp->rx_inuse_list, &list);
	spin_unlock(&cp->rx_inuse_lock);
#else
	spin_lock(&cp->rx_spare_lock);
	list_splice_init(&cp->rx_inuse_list, &list);
	spin_unlock(&cp->rx_spare_lock);
#endif
	list_for_each_safe(elem, tmp, &list) {
		cas_page_free(cp, list_entry(elem, cas_page_t, list));
	}
}

/* replenish spares if needed */
static void cas_spare_recover(struct cas *cp, const gfp_t flags)
{
	struct list_head list, *elem, *tmp;
	int needed, i;

	/* check inuse list. if we don't need any more free buffers,
	 * just free it
	 */

	/* make a local copy of the list */
	INIT_LIST_HEAD(&list);
	spin_lock(&cp->rx_inuse_lock);
	list_splice_init(&cp->rx_inuse_list, &list);
	spin_unlock(&cp->rx_inuse_lock);

	list_for_each_safe(elem, tmp, &list) {
		cas_page_t *page = list_entry(elem, cas_page_t, list);

		/*
		 * With the lockless pagecache, cassini buffering scheme gets
		 * slightly less accurate: we might find that a page has an
		 * elevated reference count here, due to a speculative ref,
		 * and skip it as in-use. Ideally we would be able to reclaim
		 * it. However this would be such a rare case, it doesn't
		 * matter too much as we should pick it up the next time round.
		 *
		 * Importantly, if we find that the page has a refcount of 1
		 * here (our refcount), then we know it is definitely not inuse
		 * so we can reuse it.
		 */
		if (page_count(page->buffer) > 1)
			continue;

		list_del(elem);
		spin_lock(&cp->rx_spare_lock);
		if (cp->rx_spares_needed > 0) {
			list_add(elem, &cp->rx_spare_list);
			cp->rx_spares_needed--;
			spin_unlock(&cp->rx_spare_lock);
		} else {
			spin_unlock(&cp->rx_spare_lock);
			cas_page_free(cp, page);
		}
	}

	/* put any inuse buffers back on the list */
	if (!list_empty(&list)) {
		spin_lock(&cp->rx_inuse_lock);
		list_splice(&list, &cp->rx_inuse_list);
		spin_unlock(&cp->rx_inuse_lock);
	}

	spin_lock(&cp->rx_spare_lock);
	needed = cp->rx_spares_needed;
	spin_unlock(&cp->rx_spare_lock);
	if (!needed)
		return;

	/* we still need spares, so try to allocate some */
	INIT_LIST_HEAD(&list);
	i = 0;
	while (i < needed) {
		cas_page_t *spare = cas_page_alloc(cp, flags);
		if (!spare)
			break;
		list_add(&spare->list, &list);
		i++;
	}

	spin_lock(&cp->rx_spare_lock);
	list_splice(&list, &cp->rx_spare_list);
	cp->rx_spares_needed -= i;
	spin_unlock(&cp->rx_spare_lock);
}

/* pull a page from the list. */
static cas_page_t *cas_page_dequeue(struct cas *cp)
{
	struct list_head *entry;
	int recover;

	spin_lock(&cp->rx_spare_lock);
	if (list_empty(&cp->rx_spare_list)) {
		/* try to do a quick recovery */
		spin_unlock(&cp->rx_spare_lock);
		cas_spare_recover(cp, GFP_ATOMIC);
		spin_lock(&cp->rx_spare_lock);
		if (list_empty(&cp->rx_spare_list)) {
			netif_err(cp, rx_err, cp->dev,
				  "no spare buffers available\n");
			spin_unlock(&cp->rx_spare_lock);
			return NULL;
		}
	}

	entry = cp->rx_spare_list.next;
	list_del(entry);
	recover = ++cp->rx_spares_needed;
	spin_unlock(&cp->rx_spare_lock);

	/* trigger the timer to do the recovery */
	if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
#if 1
		atomic_inc(&cp->reset_task_pending);
		atomic_inc(&cp->reset_task_pending_spare);
		schedule_work(&cp->reset_task);
#else
		atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
		schedule_work(&cp->reset_task);
#endif
	}
	return list_entry(entry, cas_page_t, list);
}


static void cas_mif_poll(struct cas *cp, const int enable)
{
	u32 cfg;

	cfg  = readl(cp->regs + REG_MIF_CFG);
	cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);

	if (cp->phy_type & CAS_PHY_MII_MDIO1)
		cfg |= MIF_CFG_PHY_SELECT;

	/* poll and interrupt on link status change. */
	if (enable) {
		cfg |= MIF_CFG_POLL_EN;
		cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
		cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
	}
	writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
	       cp->regs + REG_MIF_MASK);
	writel(cfg, cp->regs + REG_MIF_CFG);
}

/* Must be invoked under cp->lock */
static void cas_begin_auto_negotiation(struct cas *cp, struct ethtool_cmd *ep)
{
	u16 ctl;
#if 1
	int lcntl;
	int changed = 0;
	int oldstate = cp->lstate;
	int link_was_not_down = !(oldstate == link_down);
#endif
	/* Setup link parameters */
	if (!ep)
		goto start_aneg;
	lcntl = cp->link_cntl;
	if (ep->autoneg == AUTONEG_ENABLE)
		cp->link_cntl = BMCR_ANENABLE;
	else {
		cp->link_cntl = 0;
		if (ep->speed == SPEED_100)
			cp->link_cntl |= BMCR_SPEED100;
		else if (ep->speed == SPEED_1000)
			cp->link_cntl |= CAS_BMCR_SPEED1000;
		if (ep->duplex == DUPLEX_FULL)
			cp->link_cntl |= BMCR_FULLDPLX;
	}
#if 1
	changed = (lcntl != cp->link_cntl);
#endif
start_aneg:
	if (cp->lstate == link_up) {
		netdev_info(cp->dev, "PCS link down\n");
	} else {
		if (changed) {
			netdev_info(cp->dev, "link configuration changed\n");
		}
	}
	cp->lstate = link_down;
	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
	if (!cp->hw_running)
		return;
#if 1
	/*
	 * WTZ: If the old state was link_up, we turn off the carrier
	 * to replicate everything we do elsewhere on a link-down
	 * event when we were already in a link-up state..
	 */
	if (oldstate == link_up)
		netif_carrier_off(cp->dev);
	if (changed  && link_was_not_down) {
		/*
		 * WTZ: This branch will simply schedule a full reset after
		 * we explicitly changed link modes in an ioctl. See if this
		 * fixes the link-problems we were having for forced mode.
		 */
		atomic_inc(&cp->reset_task_pending);
		atomic_inc(&cp->reset_task_pending_all);
		schedule_work(&cp->reset_task);
		cp->timer_ticks = 0;
		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
		return;
	}
#endif
	if (cp->phy_type & CAS_PHY_SERDES) {
		u32 val = readl(cp->regs + REG_PCS_MII_CTRL);

		if (cp->link_cntl & BMCR_ANENABLE) {
			val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
			cp->lstate = link_aneg;
		} else {
			if (cp->link_cntl & BMCR_FULLDPLX)
				val |= PCS_MII_CTRL_DUPLEX;
			val &= ~PCS_MII_AUTONEG_EN;
			cp->lstate = link_force_ok;
		}
		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
		writel(val, cp->regs + REG_PCS_MII_CTRL);

	} else {
		cas_mif_poll(cp, 0);
		ctl = cas_phy_read(cp, MII_BMCR);
		ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
			 CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
		ctl |= cp->link_cntl;
		if (ctl & BMCR_ANENABLE) {
			ctl |= BMCR_ANRESTART;
			cp->lstate = link_aneg;
		} else {
			cp->lstate = link_force_ok;
		}
		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
		cas_phy_write(cp, MII_BMCR, ctl);
		cas_mif_poll(cp, 1);
	}

	cp->timer_ticks = 0;
	mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
}

/* Must be invoked under cp->lock. */
static int cas_reset_mii_phy(struct cas *cp)
{
	int limit = STOP_TRIES_PHY;
	u16 val;

	cas_phy_write(cp, MII_BMCR, BMCR_RESET);
	udelay(100);
	while (--limit) {
		val = cas_phy_read(cp, MII_BMCR);
		if ((val & BMCR_RESET) == 0)
			break;
		udelay(10);
	}
	return limit <= 0;
}

static int cas_saturn_firmware_init(struct cas *cp)
{
	const struct firmware *fw;
	const char fw_name[] = "sun/cassini.bin";
	int err;

	if (PHY_NS_DP83065 != cp->phy_id)
		return 0;

	err = request_firmware(&fw, fw_name, &cp->pdev->dev);
	if (err) {
		pr_err("Failed to load firmware \"%s\"\n",
		       fw_name);
		return err;
	}
	if (fw->size < 2) {
		pr_err("bogus length %zu in \"%s\"\n",
		       fw->size, fw_name);
		err = -EINVAL;
		goto out;
	}
	cp->fw_load_addr= fw->data[1] << 8 | fw->data[0];
	cp->fw_size = fw->size - 2;
	cp->fw_data = vmalloc(cp->fw_size);
	if (!cp->fw_data) {
		err = -ENOMEM;
		pr_err("\"%s\" Failed %d\n", fw_name, err);
		goto out;
	}
	memcpy(cp->fw_data, &fw->data[2], cp->fw_size);
out:
	release_firmware(fw);
	return err;
}

static void cas_saturn_firmware_load(struct cas *cp)
{
	int i;

	cas_phy_powerdown(cp);

	/* expanded memory access mode */
	cas_phy_write(cp, DP83065_MII_MEM, 0x0);

	/* pointer configuration for new firmware */
	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
	cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
	cas_phy_write(cp, DP83065_MII_REGD, 0x82);
	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
	cas_phy_write(cp, DP83065_MII_REGD, 0x0);
	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
	cas_phy_write(cp, DP83065_MII_REGD, 0x39);

	/* download new firmware */
	cas_phy_write(cp, DP83065_MII_MEM, 0x1);
	cas_phy_write(cp, DP83065_MII_REGE, cp->fw_load_addr);
	for (i = 0; i < cp->fw_size; i++)
		cas_phy_write(cp, DP83065_MII_REGD, cp->fw_data[i]);

	/* enable firmware */
	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
	cas_phy_write(cp, DP83065_MII_REGD, 0x1);
}


/* phy initialization */
static void cas_phy_init(struct cas *cp)
{
	u16 val;

	/* if we're in MII/GMII mode, set up phy */
	if (CAS_PHY_MII(cp->phy_type)) {
		writel(PCS_DATAPATH_MODE_MII,
		       cp->regs + REG_PCS_DATAPATH_MODE);

		cas_mif_poll(cp, 0);
		cas_reset_mii_phy(cp); /* take out of isolate mode */

		if (PHY_LUCENT_B0 == cp->phy_id) {
			/* workaround link up/down issue with lucent */
			cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
			cas_phy_write(cp, MII_BMCR, 0x00f1);
			cas_phy_write(cp, LUCENT_MII_REG, 0x0);

		} else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
			/* workarounds for broadcom phy */
			cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
			cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);

		} else if (PHY_BROADCOM_5411 == cp->phy_id) {
			val = cas_phy_read(cp, BROADCOM_MII_REG4);
			val = cas_phy_read(cp, BROADCOM_MII_REG4);
			if (val & 0x0080) {
				/* link workaround */
				cas_phy_write(cp, BROADCOM_MII_REG4,
					      val & ~0x0080);
			}

		} else if (cp->cas_flags & CAS_FLAG_SATURN) {
			writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
			       SATURN_PCFG_FSI : 0x0,
			       cp->regs + REG_SATURN_PCFG);

			/* load firmware to address 10Mbps auto-negotiation
			 * issue. NOTE: this will need to be changed if the
			 * default firmware gets fixed.
			 */
			if (PHY_NS_DP83065 == cp->phy_id) {
				cas_saturn_firmware_load(cp);
			}
			cas_phy_powerup(cp);
		}

		/* advertise capabilities */
		val = cas_phy_read(cp, MII_BMCR);
		val &= ~BMCR_ANENABLE;
		cas_phy_write(cp, MII_BMCR, val);
		udelay(10);

		cas_phy_write(cp, MII_ADVERTISE,
			      cas_phy_read(cp, MII_ADVERTISE) |
			      (ADVERTISE_10HALF | ADVERTISE_10FULL |
			       ADVERTISE_100HALF | ADVERTISE_100FULL |
			       CAS_ADVERTISE_PAUSE |
			       CAS_ADVERTISE_ASYM_PAUSE));

		if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
			/* make sure that we don't advertise half
			 * duplex to avoid a chip issue
			 */
			val  = cas_phy_read(cp, CAS_MII_1000_CTRL);
			val &= ~CAS_ADVERTISE_1000HALF;
			val |= CAS_ADVERTISE_1000FULL;
			cas_phy_write(cp, CAS_MII_1000_CTRL, val);
		}

	} else {
		/* reset pcs for serdes */
		u32 val;
		int limit;

		writel(PCS_DATAPATH_MODE_SERDES,
		       cp->regs + REG_PCS_DATAPATH_MODE);

		/* enable serdes pins on saturn */
		if (cp->cas_flags & CAS_FLAG_SATURN)
			writel(0, cp->regs + REG_SATURN_PCFG);

		/* Reset PCS unit. */
		val = readl(cp->regs + REG_PCS_MII_CTRL);
		val |= PCS_MII_RESET;
		writel(val, cp->regs + REG_PCS_MII_CTRL);

		limit = STOP_TRIES;
		while (--limit > 0) {
			udelay(10);
			if ((readl(cp->regs + REG_PCS_MII_CTRL) &
			     PCS_MII_RESET) == 0)
				break;
		}
		if (limit <= 0)
			netdev_warn(cp->dev, "PCS reset bit would not clear [%08x]\n",
				    readl(cp->regs + REG_PCS_STATE_MACHINE));

		/* Make sure PCS is disabled while changing advertisement
		 * configuration.
		 */
		writel(0x0, cp->regs + REG_PCS_CFG);

		/* Advertise all capabilities except half-duplex. */
		val  = readl(cp->regs + REG_PCS_MII_ADVERT);
		val &= ~PCS_MII_ADVERT_HD;
		val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
			PCS_MII_ADVERT_ASYM_PAUSE);
		writel(val, cp->regs + REG_PCS_MII_ADVERT);

		/* enable PCS */
		writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);

		/* pcs workaround: enable sync detect */
		writel(PCS_SERDES_CTRL_SYNCD_EN,
		       cp->regs + REG_PCS_SERDES_CTRL);
	}
}


static int cas_pcs_link_check(struct cas *cp)
{
	u32 stat, state_machine;
	int retval = 0;

	/* The link status bit latches on zero, so you must
	 * read it twice in such a case to see a transition
	 * to the link being up.
	 */
	stat = readl(cp->regs + REG_PCS_MII_STATUS);
	if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
		stat = readl(cp->regs + REG_PCS_MII_STATUS);

	/* The remote-fault indication is only valid
	 * when autoneg has completed.
	 */
	if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
		     PCS_MII_STATUS_REMOTE_FAULT)) ==
	    (PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT))
		netif_info(cp, link, cp->dev, "PCS RemoteFault\n");

	/* work around link detection issue by querying the PCS state
	 * machine directly.
	 */
	state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
	if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
		stat &= ~PCS_MII_STATUS_LINK_STATUS;
	} else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
		stat |= PCS_MII_STATUS_LINK_STATUS;
	}

	if (stat & PCS_MII_STATUS_LINK_STATUS) {
		if (cp->lstate != link_up) {
			if (cp->opened) {
				cp->lstate = link_up;
				cp->link_transition = LINK_TRANSITION_LINK_UP;

				cas_set_link_modes(cp);
				netif_carrier_on(cp->dev);
			}
		}
	} else if (cp->lstate == link_up) {
		cp->lstate = link_down;
		if (link_transition_timeout != 0 &&
		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
		    !cp->link_transition_jiffies_valid) {
			/*
			 * force a reset, as a workaround for the
			 * link-failure problem. May want to move this to a
			 * point a bit earlier in the sequence. If we had
			 * generated a reset a short time ago, we'll wait for
			 * the link timer to check the status until a
			 * timer expires (link_transistion_jiffies_valid is
			 * true when the timer is running.)  Instead of using
			 * a system timer, we just do a check whenever the
			 * link timer is running - this clears the flag after
			 * a suitable delay.
			 */
			retval = 1;
			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
			cp->link_transition_jiffies = jiffies;
			cp->link_transition_jiffies_valid = 1;
		} else {
			cp->link_transition = LINK_TRANSITION_ON_FAILURE;
		}
		netif_carrier_off(cp->dev);
		if (cp->opened)
			netif_info(cp, link, cp->dev, "PCS link down\n");

		/* Cassini only: if you force a mode, there can be
		 * sync problems on link down. to fix that, the following
		 * things need to be checked:
		 * 1) read serialink state register
		 * 2) read pcs status register to verify link down.
		 * 3) if link down and serial link == 0x03, then you need
		 *    to global reset the chip.
		 */
		if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
			/* should check to see if we're in a forced mode */
			stat = readl(cp->regs + REG_PCS_SERDES_STATE);
			if (stat == 0x03)
				return 1;
		}
	} else if (cp->lstate == link_down) {
		if (link_transition_timeout != 0 &&
		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
		    !cp->link_transition_jiffies_valid) {
			/* force a reset, as a workaround for the
			 * link-failure problem.  May want to move
			 * this to a point a bit earlier in the
			 * sequence.
			 */
			retval = 1;
			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
			cp->link_transition_jiffies = jiffies;
			cp->link_transition_jiffies_valid = 1;
		} else {
			cp->link_transition = LINK_TRANSITION_STILL_FAILED;
		}
	}

	return retval;
}

static int cas_pcs_interrupt(struct net_device *dev,
			     struct cas *cp, u32 status)
{
	u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);

	if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
		return 0;
	return cas_pcs_link_check(cp);
}

static int cas_txmac_interrupt(struct net_device *dev,
			       struct cas *cp, u32 status)
{
	u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);

	if (!txmac_stat)
		return 0;

	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
		     "txmac interrupt, txmac_stat: 0x%x\n", txmac_stat);

	/* Defer timer expiration is quite normal,
	 * don't even log the event.
	 */
	if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
	    !(txmac_stat & ~MAC_TX_DEFER_TIMER))
		return 0;

	spin_lock(&cp->stat_lock[0]);
	if (txmac_stat & MAC_TX_UNDERRUN) {
		netdev_err(dev, "TX MAC xmit underrun\n");
		cp->net_stats[0].tx_fifo_errors++;
	}

	if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
		netdev_err(dev, "TX MAC max packet size error\n");
		cp->net_stats[0].tx_errors++;
	}

	/* The rest are all cases of one of the 16-bit TX
	 * counters expiring.
	 */
	if (txmac_stat & MAC_TX_COLL_NORMAL)
		cp->net_stats[0].collisions += 0x10000;

	if (txmac_stat & MAC_TX_COLL_EXCESS) {
		cp->net_stats[0].tx_aborted_errors += 0x10000;
		cp->net_stats[0].collisions += 0x10000;
	}

	if (txmac_stat & MAC_TX_COLL_LATE) {
		cp->net_stats[0].tx_aborted_errors += 0x10000;
		cp->net_stats[0].collisions += 0x10000;
	}
	spin_unlock(&cp->stat_lock[0]);

	/* We do not keep track of MAC_TX_COLL_FIRST and
	 * MAC_TX_PEAK_ATTEMPTS events.
	 */
	return 0;
}

static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
{
	cas_hp_inst_t *inst;
	u32 val;
	int i;

	i = 0;
	while ((inst = firmware) && inst->note) {
		writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);

		val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
		val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);

		val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
		val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
		val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
		val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
		val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
		val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
		val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);

		val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
		++firmware;
		++i;
	}
}

static void cas_init_rx_dma(struct cas *cp)
{
	u64 desc_dma = cp->block_dvma;
	u32 val;
	int i, size;

	/* rx free descriptors */
	val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
	val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
	val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
	if ((N_RX_DESC_RINGS > 1) &&
	    (cp->cas_flags & CAS_FLAG_REG_PLUS))  /* do desc 2 */
		val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
	writel(val, cp->regs + REG_RX_CFG);

	val = (unsigned long) cp->init_rxds[0] -
		(unsigned long) cp->init_block;
	writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);

	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		/* rx desc 2 is for IPSEC packets. however,
		 * we don't it that for that purpose.
		 */
		val = (unsigned long) cp->init_rxds[1] -
			(unsigned long) cp->init_block;
		writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
		writel((desc_dma + val) & 0xffffffff, cp->regs +
		       REG_PLUS_RX_DB1_LOW);
		writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
		       REG_PLUS_RX_KICK1);
	}

	/* rx completion registers */
	val = (unsigned long) cp->init_rxcs[0] -
		(unsigned long) cp->init_block;
	writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);

	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		/* rx comp 2-4 */
		for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
			val = (unsigned long) cp->init_rxcs[i] -
				(unsigned long) cp->init_block;
			writel((desc_dma + val) >> 32, cp->regs +
			       REG_PLUS_RX_CBN_HI(i));
			writel((desc_dma + val) & 0xffffffff, cp->regs +
			       REG_PLUS_RX_CBN_LOW(i));
		}
	}

	/* read selective clear regs to prevent spurious interrupts
	 * on reset because complete == kick.
	 * selective clear set up to prevent interrupts on resets
	 */
	readl(cp->regs + REG_INTR_STATUS_ALIAS);
	writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		for (i = 1; i < N_RX_COMP_RINGS; i++)
			readl(cp->regs + REG_PLUS_INTRN_STATUS_ALIAS(i));

		/* 2 is different from 3 and 4 */
		if (N_RX_COMP_RINGS > 1)
			writel(INTR_RX_DONE_ALT | INTR_RX_BUF_UNAVAIL_1,
			       cp->regs + REG_PLUS_ALIASN_CLEAR(1));

		for (i = 2; i < N_RX_COMP_RINGS; i++)
			writel(INTR_RX_DONE_ALT,
			       cp->regs + REG_PLUS_ALIASN_CLEAR(i));
	}

	/* set up pause thresholds */
	val  = CAS_BASE(RX_PAUSE_THRESH_OFF,
			cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
	val |= CAS_BASE(RX_PAUSE_THRESH_ON,
			cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
	writel(val, cp->regs + REG_RX_PAUSE_THRESH);

	/* zero out dma reassembly buffers */
	for (i = 0; i < 64; i++) {
		writel(i, cp->regs + REG_RX_TABLE_ADDR);
		writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
		writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
		writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
	}

	/* make sure address register is 0 for normal operation */
	writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
	writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);

	/* interrupt mitigation */
#ifdef USE_RX_BLANK
	val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
	val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
	writel(val, cp->regs + REG_RX_BLANK);
#else
	writel(0x0, cp->regs + REG_RX_BLANK);
#endif

	/* interrupt generation as a function of low water marks for
	 * free desc and completion entries. these are used to trigger
	 * housekeeping for rx descs. we don't use the free interrupt
	 * as it's not very useful
	 */
	/* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
	val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
	writel(val, cp->regs + REG_RX_AE_THRESH);
	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
		writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
	}

	/* Random early detect registers. useful for congestion avoidance.
	 * this should be tunable.
	 */
	writel(0x0, cp->regs + REG_RX_RED);

	/* receive page sizes. default == 2K (0x800) */
	val = 0;
	if (cp->page_size == 0x1000)
		val = 0x1;
	else if (cp->page_size == 0x2000)
		val = 0x2;
	else if (cp->page_size == 0x4000)
		val = 0x3;

	/* round mtu + offset. constrain to page size. */
	size = cp->dev->mtu + 64;
	if (size > cp->page_size)
		size = cp->page_size;

	if (size <= 0x400)
		i = 0x0;
	else if (size <= 0x800)
		i = 0x1;
	else if (size <= 0x1000)
		i = 0x2;
	else
		i = 0x3;

	cp->mtu_stride = 1 << (i + 10);
	val  = CAS_BASE(RX_PAGE_SIZE, val);
	val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
	val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
	val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
	writel(val, cp->regs + REG_RX_PAGE_SIZE);

	/* enable the header parser if desired */
	if (CAS_HP_FIRMWARE == cas_prog_null)
		return;

	val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
	val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
	val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
	writel(val, cp->regs + REG_HP_CFG);
}

static inline void cas_rxc_init(struct cas_rx_comp *rxc)
{
	memset(rxc, 0, sizeof(*rxc));
	rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
}

/* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
 * flipping is protected by the fact that the chip will not
 * hand back the same page index while it's being processed.
 */
static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
{
	cas_page_t *page = cp->rx_pages[1][index];
	cas_page_t *new;

	if (page_count(page->buffer) == 1)
		return page;

	new = cas_page_dequeue(cp);
	if (new) {
		spin_lock(&cp->rx_inuse_lock);
		list_add(&page->list, &cp->rx_inuse_list);
		spin_unlock(&cp->rx_inuse_lock);
	}
	return new;
}

/* this needs to be changed if we actually use the ENC RX DESC ring */
static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
				 const int index)
{
	cas_page_t **page0 = cp->rx_pages[0];
	cas_page_t **page1 = cp->rx_pages[1];

	/* swap if buffer is in use */
	if (page_count(page0[index]->buffer) > 1) {
		cas_page_t *new = cas_page_spare(cp, index);
		if (new) {
			page1[index] = page0[index];
			page0[index] = new;
		}
	}
	RX_USED_SET(page0[index], 0);
	return page0[index];
}

static void cas_clean_rxds(struct cas *cp)
{
	/* only clean ring 0 as ring 1 is used for spare buffers */
        struct cas_rx_desc *rxd = cp->init_rxds[0];
	int i, size;

	/* release all rx flows */
	for (i = 0; i < N_RX_FLOWS; i++) {
		struct sk_buff *skb;
		while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
			cas_skb_release(skb);
		}
	}

	/* initialize descriptors */
	size = RX_DESC_RINGN_SIZE(0);
	for (i = 0; i < size; i++) {
		cas_page_t *page = cas_page_swap(cp, 0, i);
		rxd[i].buffer = cpu_to_le64(page->dma_addr);
		rxd[i].index  = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
					    CAS_BASE(RX_INDEX_RING, 0));
	}

	cp->rx_old[0]  = RX_DESC_RINGN_SIZE(0) - 4;
	cp->rx_last[0] = 0;
	cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
}

static void cas_clean_rxcs(struct cas *cp)
{
	int i, j;

	/* take ownership of rx comp descriptors */
	memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
	memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
	for (i = 0; i < N_RX_COMP_RINGS; i++) {
		struct cas_rx_comp *rxc = cp->init_rxcs[i];
		for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
			cas_rxc_init(rxc + j);
		}
	}
}

#if 0
/* When we get a RX fifo overflow, the RX unit is probably hung
 * so we do the following.
 *
 * If any part of the reset goes wrong, we return 1 and that causes the
 * whole chip to be reset.
 */
static int cas_rxmac_reset(struct cas *cp)
{
	struct net_device *dev = cp->dev;
	int limit;
	u32 val;

	/* First, reset MAC RX. */
	writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
	for (limit = 0; limit < STOP_TRIES; limit++) {
		if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
			break;
		udelay(10);
	}
	if (limit == STOP_TRIES) {
		netdev_err(dev, "RX MAC will not disable, resetting whole chip\n");
		return 1;
	}

	/* Second, disable RX DMA. */
	writel(0, cp->regs + REG_RX_CFG);
	for (limit = 0; limit < STOP_TRIES; limit++) {
		if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
			break;
		udelay(10);
	}
	if (limit == STOP_TRIES) {
		netdev_err(dev, "RX DMA will not disable, resetting whole chip\n");
		return 1;
	}

	mdelay(5);

	/* Execute RX reset command. */
	writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
	for (limit = 0; limit < STOP_TRIES; limit++) {
		if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
			break;
		udelay(10);
	}
	if (limit == STOP_TRIES) {
		netdev_err(dev, "RX reset command will not execute, resetting whole chip\n");
		return 1;
	}

	/* reset driver rx state */
	cas_clean_rxds(cp);
	cas_clean_rxcs(cp);

	/* Now, reprogram the rest of RX unit. */
	cas_init_rx_dma(cp);

	/* re-enable */
	val = readl(cp->regs + REG_RX_CFG);
	writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
	val = readl(cp->regs + REG_MAC_RX_CFG);
	writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
	return 0;
}
#endif

static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
			       u32 status)
{
	u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);

	if (!stat)
		return 0;

	netif_dbg(cp, intr, cp->dev, "rxmac interrupt, stat: 0x%x\n", stat);

	/* these are all rollovers */
	spin_lock(&cp->stat_lock[0]);
	if (stat & MAC_RX_ALIGN_ERR)
		cp->net_stats[0].rx_frame_errors += 0x10000;

	if (stat & MAC_RX_CRC_ERR)
		cp->net_stats[0].rx_crc_errors += 0x10000;

	if (stat & MAC_RX_LEN_ERR)
		cp->net_stats[0].rx_length_errors += 0x10000;

	if (stat & MAC_RX_OVERFLOW) {
		cp->net_stats[0].rx_over_errors++;
		cp->net_stats[0].rx_fifo_errors++;
	}

	/* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
	 * events.
	 */
	spin_unlock(&cp->stat_lock[0]);
	return 0;
}

static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
			     u32 status)
{
	u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);

	if (!stat)
		return 0;

	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
		     "mac interrupt, stat: 0x%x\n", stat);

	/* This interrupt is just for pause frame and pause
	 * tracking.  It is useful for diagnostics and debug
	 * but probably by default we will mask these events.
	 */
	if (stat & MAC_CTRL_PAUSE_STATE)
		cp->pause_entered++;

	if (stat & MAC_CTRL_PAUSE_RECEIVED)
		cp->pause_last_time_recvd = (stat >> 16);

	return 0;
}


/* Must be invoked under cp->lock. */
static inline int cas_mdio_link_not_up(struct cas *cp)
{
	u16 val;

	switch (cp->lstate) {
	case link_force_ret:
		netif_info(cp, link, cp->dev, "Autoneg failed again, keeping forced mode\n");
		cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
		cp->timer_ticks = 5;
		cp->lstate = link_force_ok;
		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
		break;

	case link_aneg:
		val = cas_phy_read(cp, MII_BMCR);

		/* Try forced modes. we try things in the following order:
		 * 1000 full -> 100 full/half -> 10 half
		 */
		val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
		val |= BMCR_FULLDPLX;
		val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
			CAS_BMCR_SPEED1000 : BMCR_SPEED100;
		cas_phy_write(cp, MII_BMCR, val);
		cp->timer_ticks = 5;
		cp->lstate = link_force_try;
		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
		break;

	case link_force_try:
		/* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
		val = cas_phy_read(cp, MII_BMCR);
		cp->timer_ticks = 5;
		if (val & CAS_BMCR_SPEED1000) { /* gigabit */
			val &= ~CAS_BMCR_SPEED1000;
			val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
			cas_phy_write(cp, MII_BMCR, val);
			break;
		}

		if (val & BMCR_SPEED100) {
			if (val & BMCR_FULLDPLX) /* fd failed */
				val &= ~BMCR_FULLDPLX;
			else { /* 100Mbps failed */
				val &= ~BMCR_SPEED100;
			}
			cas_phy_write(cp, MII_BMCR, val);
			break;
		}
	default:
		break;
	}
	return 0;
}


/* must be invoked with cp->lock held */
static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
{
	int restart;

	if (bmsr & BMSR_LSTATUS) {
		/* Ok, here we got a link. If we had it due to a forced
		 * fallback, and we were configured for autoneg, we
		 * retry a short autoneg pass. If you know your hub is
		 * broken, use ethtool ;)
		 */
		if ((cp->lstate == link_force_try) &&
		    (cp->link_cntl & BMCR_ANENABLE)) {
			cp->lstate = link_force_ret;
			cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
			cas_mif_poll(cp, 0);
			cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
			cp->timer_ticks = 5;
			if (cp->opened)
				netif_info(cp, link, cp->dev,
					   "Got link after fallback, retrying autoneg once...\n");
			cas_phy_write(cp, MII_BMCR,
				      cp->link_fcntl | BMCR_ANENABLE |
				      BMCR_ANRESTART);
			cas_mif_poll(cp, 1);

		} else if (cp->lstate != link_up) {
			cp->lstate = link_up;
			cp->link_transition = LINK_TRANSITION_LINK_UP;

			if (cp->opened) {
				cas_set_link_modes(cp);
				netif_carrier_on(cp->dev);
			}
		}
		return 0;
	}

	/* link not up. if the link was previously up, we restart the
	 * whole process
	 */
	restart = 0;
	if (cp->lstate == link_up) {
		cp->lstate = link_down;
		cp->link_transition = LINK_TRANSITION_LINK_DOWN;

		netif_carrier_off(cp->dev);
		if (cp->opened)
			netif_info(cp, link, cp->dev, "Link down\n");
		restart = 1;

	} else if (++cp->timer_ticks > 10)
		cas_mdio_link_not_up(cp);

	return restart;
}

static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
			     u32 status)
{
	u32 stat = readl(cp->regs + REG_MIF_STATUS);
	u16 bmsr;

	/* check for a link change */
	if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
		return 0;

	bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
	return cas_mii_link_check(cp, bmsr);
}

static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
			     u32 status)
{
	u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);

	if (!stat)
		return 0;

	netdev_err(dev, "PCI error [%04x:%04x]",
		   stat, readl(cp->regs + REG_BIM_DIAG));

	/* cassini+ has this reserved */
	if ((stat & PCI_ERR_BADACK) &&
	    ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
		pr_cont(" <No ACK64# during ABS64 cycle>");

	if (stat & PCI_ERR_DTRTO)
		pr_cont(" <Delayed transaction timeout>");
	if (stat & PCI_ERR_OTHER)
		pr_cont(" <other>");
	if (stat & PCI_ERR_BIM_DMA_WRITE)
		pr_cont(" <BIM DMA 0 write req>");
	if (stat & PCI_ERR_BIM_DMA_READ)
		pr_cont(" <BIM DMA 0 read req>");
	pr_cont("\n");

	if (stat & PCI_ERR_OTHER) {
		u16 cfg;

		/* Interrogate PCI config space for the
		 * true cause.
		 */
		pci_read_config_word(cp->pdev, PCI_STATUS, &cfg);
		netdev_err(dev, "Read PCI cfg space status [%04x]\n", cfg);
		if (cfg & PCI_STATUS_PARITY)
			netdev_err(dev, "PCI parity error detected\n");
		if (cfg & PCI_STATUS_SIG_TARGET_ABORT)
			netdev_err(dev, "PCI target abort\n");
		if (cfg & PCI_STATUS_REC_TARGET_ABORT)
			netdev_err(dev, "PCI master acks target abort\n");
		if (cfg & PCI_STATUS_REC_MASTER_ABORT)
			netdev_err(dev, "PCI master abort\n");
		if (cfg & PCI_STATUS_SIG_SYSTEM_ERROR)
			netdev_err(dev, "PCI system error SERR#\n");
		if (cfg & PCI_STATUS_DETECTED_PARITY)
			netdev_err(dev, "PCI parity error\n");

		/* Write the error bits back to clear them. */
		cfg &= (PCI_STATUS_PARITY |
			PCI_STATUS_SIG_TARGET_ABORT |
			PCI_STATUS_REC_TARGET_ABORT |
			PCI_STATUS_REC_MASTER_ABORT |
			PCI_STATUS_SIG_SYSTEM_ERROR |
			PCI_STATUS_DETECTED_PARITY);
		pci_write_config_word(cp->pdev, PCI_STATUS, cfg);
	}

	/* For all PCI errors, we should reset the chip. */
	return 1;
}

/* All non-normal interrupt conditions get serviced here.
 * Returns non-zero if we should just exit the interrupt
 * handler right now (ie. if we reset the card which invalidates
 * all of the other original irq status bits).
 */
static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
			    u32 status)
{
	if (status & INTR_RX_TAG_ERROR) {
		/* corrupt RX tag framing */
		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
			     "corrupt rx tag framing\n");
		spin_lock(&cp->stat_lock[0]);
		cp->net_stats[0].rx_errors++;
		spin_unlock(&cp->stat_lock[0]);
		goto do_reset;
	}

	if (status & INTR_RX_LEN_MISMATCH) {
		/* length mismatch. */
		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
			     "length mismatch for rx frame\n");
		spin_lock(&cp->stat_lock[0]);
		cp->net_stats[0].rx_errors++;
		spin_unlock(&cp->stat_lock[0]);
		goto do_reset;
	}

	if (status & INTR_PCS_STATUS) {
		if (cas_pcs_interrupt(dev, cp, status))
			goto do_reset;
	}

	if (status & INTR_TX_MAC_STATUS) {
		if (cas_txmac_interrupt(dev, cp, status))
			goto do_reset;
	}

	if (status & INTR_RX_MAC_STATUS) {
		if (cas_rxmac_interrupt(dev, cp, status))
			goto do_reset;
	}

	if (status & INTR_MAC_CTRL_STATUS) {
		if (cas_mac_interrupt(dev, cp, status))
			goto do_reset;
	}

	if (status & INTR_MIF_STATUS) {
		if (cas_mif_interrupt(dev, cp, status))
			goto do_reset;
	}

	if (status & INTR_PCI_ERROR_STATUS) {
		if (cas_pci_interrupt(dev, cp, status))
			goto do_reset;
	}
	return 0;

do_reset:
#if 1
	atomic_inc(&cp->reset_task_pending);
	atomic_inc(&cp->reset_task_pending_all);
	netdev_err(dev, "reset called in cas_abnormal_irq [0x%x]\n", status);
	schedule_work(&cp->reset_task);
#else
	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
	netdev_err(dev, "reset called in cas_abnormal_irq\n");
	schedule_work(&cp->reset_task);
#endif
	return 1;
}

/* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
 *       determining whether to do a netif_stop/wakeup
 */
#define CAS_TABORT(x)      (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
#define CAS_ROUND_PAGE(x)  (((x) + PAGE_SIZE - 1) & PAGE_MASK)
static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
				  const int len)
{
	unsigned long off = addr + len;

	if (CAS_TABORT(cp) == 1)
		return 0;
	if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
		return 0;
	return TX_TARGET_ABORT_LEN;
}

static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
{
	struct cas_tx_desc *txds;
	struct sk_buff **skbs;
	struct net_device *dev = cp->dev;
	int entry, count;

	spin_lock(&cp->tx_lock[ring]);
	txds = cp->init_txds[ring];
	skbs = cp->tx_skbs[ring];
	entry = cp->tx_old[ring];

	count = TX_BUFF_COUNT(ring, entry, limit);
	while (entry != limit) {
		struct sk_buff *skb = skbs[entry];
		dma_addr_t daddr;
		u32 dlen;
		int frag;

		if (!skb) {
			/* this should never occur */
			entry = TX_DESC_NEXT(ring, entry);
			continue;
		}

		/* however, we might get only a partial skb release. */
		count -= skb_shinfo(skb)->nr_frags +
			+ cp->tx_tiny_use[ring][entry].nbufs + 1;
		if (count < 0)
			break;

		netif_printk(cp, tx_done, KERN_DEBUG, cp->dev,
			     "tx[%d] done, slot %d\n", ring, entry);

		skbs[entry] = NULL;
		cp->tx_tiny_use[ring][entry].nbufs = 0;

		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
			struct cas_tx_desc *txd = txds + entry;

			daddr = le64_to_cpu(txd->buffer);
			dlen = CAS_VAL(TX_DESC_BUFLEN,
				       le64_to_cpu(txd->control));
			pci_unmap_page(cp->pdev, daddr, dlen,
				       PCI_DMA_TODEVICE);
			entry = TX_DESC_NEXT(ring, entry);

			/* tiny buffer may follow */
			if (cp->tx_tiny_use[ring][entry].used) {
				cp->tx_tiny_use[ring][entry].used = 0;
				entry = TX_DESC_NEXT(ring, entry);
			}
		}

		spin_lock(&cp->stat_lock[ring]);
		cp->net_stats[ring].tx_packets++;
		cp->net_stats[ring].tx_bytes += skb->len;
		spin_unlock(&cp->stat_lock[ring]);
		dev_kfree_skb_irq(skb);
	}
	cp->tx_old[ring] = entry;

	/* this is wrong for multiple tx rings. the net device needs
	 * multiple queues for this to do the right thing.  we wait
	 * for 2*packets to be available when using tiny buffers
	 */
	if (netif_queue_stopped(dev) &&
	    (TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
		netif_wake_queue(dev);
	spin_unlock(&cp->tx_lock[ring]);
}

static void cas_tx(struct net_device *dev, struct cas *cp,
		   u32 status)
{
        int limit, ring;
#ifdef USE_TX_COMPWB
	u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
#endif
	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
		     "tx interrupt, status: 0x%x, %llx\n",
		     status, (unsigned long long)compwb);
	/* process all the rings */
	for (ring = 0; ring < N_TX_RINGS; ring++) {
#ifdef USE_TX_COMPWB
		/* use the completion writeback registers */
		limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
			CAS_VAL(TX_COMPWB_LSB, compwb);
		compwb = TX_COMPWB_NEXT(compwb);
#else
		limit = readl(cp->regs + REG_TX_COMPN(ring));
#endif
		if (cp->tx_old[ring] != limit)
			cas_tx_ringN(cp, ring, limit);
	}
}


static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
			      int entry, const u64 *words,
			      struct sk_buff **skbref)
{
	int dlen, hlen, len, i, alloclen;
	int off, swivel = RX_SWIVEL_OFF_VAL;
	struct cas_page *page;
	struct sk_buff *skb;
	void *addr, *crcaddr;
	__sum16 csum;
	char *p;

	hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
	dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
	len  = hlen + dlen;

	if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
		alloclen = len;
	else
		alloclen = max(hlen, RX_COPY_MIN);

	skb = dev_alloc_skb(alloclen + swivel + cp->crc_size);
	if (skb == NULL)
		return -1;

	*skbref = skb;
	skb_reserve(skb, swivel);

	p = skb->data;
	addr = crcaddr = NULL;
	if (hlen) { /* always copy header pages */
		i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
		off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
			swivel;

		i = hlen;
		if (!dlen) /* attach FCS */
			i += cp->crc_size;
		pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
				    PCI_DMA_FROMDEVICE);
		addr = cas_page_map(page->buffer);
		memcpy(p, addr + off, i);
		pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
				    PCI_DMA_FROMDEVICE);
		cas_page_unmap(addr);
		RX_USED_ADD(page, 0x100);
		p += hlen;
		swivel = 0;
	}


	if (alloclen < (hlen + dlen)) {
		skb_frag_t *frag = skb_shinfo(skb)->frags;

		/* normal or jumbo packets. we use frags */
		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;

		hlen = min(cp->page_size - off, dlen);
		if (hlen < 0) {
			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
				     "rx page overflow: %d\n", hlen);
			dev_kfree_skb_irq(skb);
			return -1;
		}
		i = hlen;
		if (i == dlen)  /* attach FCS */
			i += cp->crc_size;
		pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
				    PCI_DMA_FROMDEVICE);

		/* make sure we always copy a header */
		swivel = 0;
		if (p == (char *) skb->data) { /* not split */
			addr = cas_page_map(page->buffer);
			memcpy(p, addr + off, RX_COPY_MIN);
			pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
					PCI_DMA_FROMDEVICE);
			cas_page_unmap(addr);
			off += RX_COPY_MIN;
			swivel = RX_COPY_MIN;
			RX_USED_ADD(page, cp->mtu_stride);
		} else {
			RX_USED_ADD(page, hlen);
		}
		skb_put(skb, alloclen);

		skb_shinfo(skb)->nr_frags++;
		skb->data_len += hlen - swivel;
		skb->truesize += hlen - swivel;
		skb->len      += hlen - swivel;

		get_page(page->buffer);
		frag->page = page->buffer;
		frag->page_offset = off;
		frag->size = hlen - swivel;

		/* any more data? */
		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
			hlen = dlen;
			off = 0;

			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
			pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr,
					    hlen + cp->crc_size,
					    PCI_DMA_FROMDEVICE);
			pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
					    hlen + cp->crc_size,
					    PCI_DMA_FROMDEVICE);

			skb_shinfo(skb)->nr_frags++;
			skb->data_len += hlen;
			skb->len      += hlen;
			frag++;

			get_page(page->buffer);
			frag->page = page->buffer;
			frag->page_offset = 0;
			frag->size = hlen;
			RX_USED_ADD(page, hlen + cp->crc_size);
		}

		if (cp->crc_size) {
			addr = cas_page_map(page->buffer);
			crcaddr  = addr + off + hlen;
		}

	} else {
		/* copying packet */
		if (!dlen)
			goto end_copy_pkt;

		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
		hlen = min(cp->page_size - off, dlen);
		if (hlen < 0) {
			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
				     "rx page overflow: %d\n", hlen);
			dev_kfree_skb_irq(skb);
			return -1;
		}
		i = hlen;
		if (i == dlen) /* attach FCS */
			i += cp->crc_size;
		pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
				    PCI_DMA_FROMDEVICE);
		addr = cas_page_map(page->buffer);
		memcpy(p, addr + off, i);
		pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
				    PCI_DMA_FROMDEVICE);
		cas_page_unmap(addr);
		if (p == (char *) skb->data) /* not split */
			RX_USED_ADD(page, cp->mtu_stride);
		else
			RX_USED_ADD(page, i);

		/* any more data? */
		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
			p += hlen;
			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
			pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr,
					    dlen + cp->crc_size,
					    PCI_DMA_FROMDEVICE);
			addr = cas_page_map(page->buffer);
			memcpy(p, addr, dlen + cp->crc_size);
			pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
					    dlen + cp->crc_size,
					    PCI_DMA_FROMDEVICE);
			cas_page_unmap(addr);
			RX_USED_ADD(page, dlen + cp->crc_size);
		}
end_copy_pkt:
		if (cp->crc_size) {
			addr    = NULL;
			crcaddr = skb->data + alloclen;
		}
		skb_put(skb, alloclen);
	}

	csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
	if (cp->crc_size) {
		/* checksum includes FCS. strip it out. */
		csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
					      csum_unfold(csum)));
		if (addr)
			cas_page_unmap(addr);
	}
	skb->protocol = eth_type_trans(skb, cp->dev);
	if (skb->protocol == htons(ETH_P_IP)) {
		skb->csum = csum_unfold(~csum);
		skb->ip_summed = CHECKSUM_COMPLETE;
	} else
		skb_checksum_none_assert(skb);
	return len;
}


/* we can handle up to 64 rx flows at a time. we do the same thing
 * as nonreassm except that we batch up the buffers.
 * NOTE: we currently just treat each flow as a bunch of packets that
 *       we pass up. a better way would be to coalesce the packets
 *       into a jumbo packet. to do that, we need to do the following:
 *       1) the first packet will have a clean split between header and
 *          data. save both.
 *       2) each time the next flow packet comes in, extend the
 *          data length and merge the checksums.
 *       3) on flow release, fix up the header.
 *       4) make sure the higher layer doesn't care.
 * because packets get coalesced, we shouldn't run into fragment count
 * issues.
 */
static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
				   struct sk_buff *skb)
{
	int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
	struct sk_buff_head *flow = &cp->rx_flows[flowid];

	/* this is protected at a higher layer, so no need to
	 * do any additional locking here. stick the buffer
	 * at the end.
	 */
	__skb_queue_tail(flow, skb);
	if (words[0] & RX_COMP1_RELEASE_FLOW) {
		while ((skb = __skb_dequeue(flow))) {
			cas_skb_release(skb);
		}
	}
}

/* put rx descriptor back on ring. if a buffer is in use by a higher
 * layer, this will need to put in a replacement.
 */
static void cas_post_page(struct cas *cp, const int ring, const int index)
{
	cas_page_t *new;
	int entry;

	entry = cp->rx_old[ring];

	new = cas_page_swap(cp, ring, index);
	cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
	cp->init_rxds[ring][entry].index  =
		cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
			    CAS_BASE(RX_INDEX_RING, ring));

	entry = RX_DESC_ENTRY(ring, entry + 1);
	cp->rx_old[ring] = entry;

	if (entry % 4)
		return;

	if (ring == 0)
		writel(entry, cp->regs + REG_RX_KICK);
	else if ((N_RX_DESC_RINGS > 1) &&
		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
		writel(entry, cp->regs + REG_PLUS_RX_KICK1);
}


/* only when things are bad */
static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
{
	unsigned int entry, last, count, released;
	int cluster;
	cas_page_t **page = cp->rx_pages[ring];

	entry = cp->rx_old[ring];

	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
		     "rxd[%d] interrupt, done: %d\n", ring, entry);

	cluster = -1;
	count = entry & 0x3;
	last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
	released = 0;
	while (entry != last) {
		/* make a new buffer if it's still in use */
		if (page_count(page[entry]->buffer) > 1) {
			cas_page_t *new = cas_page_dequeue(cp);
			if (!new) {
				/* let the timer know that we need to
				 * do this again
				 */
				cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
				if (!timer_pending(&cp->link_timer))
					mod_timer(&cp->link_timer, jiffies +
						  CAS_LINK_FAST_TIMEOUT);
				cp->rx_old[ring]  = entry;
				cp->rx_last[ring] = num ? num - released : 0;
				return -ENOMEM;
			}
			spin_lock(&cp->rx_inuse_lock);
			list_add(&page[entry]->list, &cp->rx_inuse_list);
			spin_unlock(&cp->rx_inuse_lock);
			cp->init_rxds[ring][entry].buffer =
				cpu_to_le64(new->dma_addr);
			page[entry] = new;

		}

		if (++count == 4) {
			cluster = entry;
			count = 0;
		}
		released++;
		entry = RX_DESC_ENTRY(ring, entry + 1);
	}
	cp->rx_old[ring] = entry;

	if (cluster < 0)
		return 0;

	if (ring == 0)
		writel(cluster, cp->regs + REG_RX_KICK);
	else if ((N_RX_DESC_RINGS > 1) &&
		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
		writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
	return 0;
}


/* process a completion ring. packets are set up in three basic ways:
 * small packets: should be copied header + data in single buffer.
 * large packets: header and data in a single buffer.
 * split packets: header in a separate buffer from data.
 *                data may be in multiple pages. data may be > 256
 *                bytes but in a single page.
 *
 * NOTE: RX page posting is done in this routine as well. while there's
 *       the capability of using multiple RX completion rings, it isn't
 *       really worthwhile due to the fact that the page posting will
 *       force serialization on the single descriptor ring.
 */
static int cas_rx_ringN(struct cas *cp, int ring, int budget)
{
	struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
	int entry, drops;
	int npackets = 0;

	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
		     "rx[%d] interrupt, done: %d/%d\n",
		     ring,
		     readl(cp->regs + REG_RX_COMP_HEAD), cp->rx_new[ring]);

	entry = cp->rx_new[ring];
	drops = 0;
	while (1) {
		struct cas_rx_comp *rxc = rxcs + entry;
		struct sk_buff *uninitialized_var(skb);
		int type, len;
		u64 words[4];
		int i, dring;

		words[0] = le64_to_cpu(rxc->word1);
		words[1] = le64_to_cpu(rxc->word2);
		words[2] = le64_to_cpu(rxc->word3);
		words[3] = le64_to_cpu(rxc->word4);

		/* don't touch if still owned by hw */
		type = CAS_VAL(RX_COMP1_TYPE, words[0]);
		if (type == 0)
			break;

		/* hw hasn't cleared the zero bit yet */
		if (words[3] & RX_COMP4_ZERO) {
			break;
		}

		/* get info on the packet */
		if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
			spin_lock(&cp->stat_lock[ring]);
			cp->net_stats[ring].rx_errors++;
			if (words[3] & RX_COMP4_LEN_MISMATCH)
				cp->net_stats[ring].rx_length_errors++;
			if (words[3] & RX_COMP4_BAD)
				cp->net_stats[ring].rx_crc_errors++;
			spin_unlock(&cp->stat_lock[ring]);

			/* We'll just return it to Cassini. */
		drop_it:
			spin_lock(&cp->stat_lock[ring]);
			++cp->net_stats[ring].rx_dropped;
			spin_unlock(&cp->stat_lock[ring]);
			goto next;
		}

		len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
		if (len < 0) {
			++drops;
			goto drop_it;
		}

		/* see if it's a flow re-assembly or not. the driver
		 * itself handles release back up.
		 */
		if (RX_DONT_BATCH || (type == 0x2)) {
			/* non-reassm: these always get released */
			cas_skb_release(skb);
		} else {
			cas_rx_flow_pkt(cp, words, skb);
		}

		spin_lock(&cp->stat_lock[ring]);
		cp->net_stats[ring].rx_packets++;
		cp->net_stats[ring].rx_bytes += len;
		spin_unlock(&cp->stat_lock[ring]);

	next:
		npackets++;

		/* should it be released? */
		if (words[0] & RX_COMP1_RELEASE_HDR) {
			i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
			dring = CAS_VAL(RX_INDEX_RING, i);
			i = CAS_VAL(RX_INDEX_NUM, i);
			cas_post_page(cp, dring, i);
		}

		if (words[0] & RX_COMP1_RELEASE_DATA) {
			i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
			dring = CAS_VAL(RX_INDEX_RING, i);
			i = CAS_VAL(RX_INDEX_NUM, i);
			cas_post_page(cp, dring, i);
		}

		if (words[0] & RX_COMP1_RELEASE_NEXT) {
			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
			dring = CAS_VAL(RX_INDEX_RING, i);
			i = CAS_VAL(RX_INDEX_NUM, i);
			cas_post_page(cp, dring, i);
		}

		/* skip to the next entry */
		entry = RX_COMP_ENTRY(ring, entry + 1 +
				      CAS_VAL(RX_COMP1_SKIP, words[0]));
#ifdef USE_NAPI
		if (budget && (npackets >= budget))
			break;
#endif
	}
	cp->rx_new[ring] = entry;

	if (drops)
		netdev_info(cp->dev, "Memory squeeze, deferring packet\n");
	return npackets;
}


/* put completion entries back on the ring */
static void cas_post_rxcs_ringN(struct net_device *dev,
				struct cas *cp, int ring)
{
	struct cas_rx_comp *rxc = cp->init_rxcs[ring];
	int last, entry;

	last = cp->rx_cur[ring];
	entry = cp->rx_new[ring];
	netif_printk(cp, intr, KERN_DEBUG, dev,
		     "rxc[%d] interrupt, done: %d/%d\n",
		     ring, readl(cp->regs + REG_RX_COMP_HEAD), entry);

	/* zero and re-mark descriptors */
	while (last != entry) {
		cas_rxc_init(rxc + last);
		last = RX_COMP_ENTRY(ring, last + 1);
	}
	cp->rx_cur[ring] = last;

	if (ring == 0)
		writel(last, cp->regs + REG_RX_COMP_TAIL);
	else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
		writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
}



/* cassini can use all four PCI interrupts for the completion ring.
 * rings 3 and 4 are identical
 */
#if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
static inline void cas_handle_irqN(struct net_device *dev,
				   struct cas *cp, const u32 status,
				   const int ring)
{
	if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
		cas_post_rxcs_ringN(dev, cp, ring);
}

static irqreturn_t cas_interruptN(int irq, void *dev_id)
{
	struct net_device *dev = dev_id;
	struct cas *cp = netdev_priv(dev);
	unsigned long flags;
	int ring;
	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));

	/* check for shared irq */
	if (status == 0)
		return IRQ_NONE;

	ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
	spin_lock_irqsave(&cp->lock, flags);
	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
#ifdef USE_NAPI
		cas_mask_intr(cp);
		napi_schedule(&cp->napi);
#else
		cas_rx_ringN(cp, ring, 0);
#endif
		status &= ~INTR_RX_DONE_ALT;
	}

	if (status)
		cas_handle_irqN(dev, cp, status, ring);
	spin_unlock_irqrestore(&cp->lock, flags);
	return IRQ_HANDLED;
}
#endif

#ifdef USE_PCI_INTB
/* everything but rx packets */
static inline void cas_handle_irq1(struct cas *cp, const u32 status)
{
	if (status & INTR_RX_BUF_UNAVAIL_1) {
		/* Frame arrived, no free RX buffers available.
		 * NOTE: we can get this on a link transition. */
		cas_post_rxds_ringN(cp, 1, 0);
		spin_lock(&cp->stat_lock[1]);
		cp->net_stats[1].rx_dropped++;
		spin_unlock(&cp->stat_lock[1]);
	}

	if (status & INTR_RX_BUF_AE_1)
		cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
				    RX_AE_FREEN_VAL(1));

	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
		cas_post_rxcs_ringN(cp, 1);
}

/* ring 2 handles a few more events than 3 and 4 */
static irqreturn_t cas_interrupt1(int irq, void *dev_id)
{
	struct net_device *dev = dev_id;
	struct cas *cp = netdev_priv(dev);
	unsigned long flags;
	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));

	/* check for shared interrupt */
	if (status == 0)
		return IRQ_NONE;

	spin_lock_irqsave(&cp->lock, flags);
	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
#ifdef USE_NAPI
		cas_mask_intr(cp);
		napi_schedule(&cp->napi);
#else
		cas_rx_ringN(cp, 1, 0);
#endif
		status &= ~INTR_RX_DONE_ALT;
	}
	if (status)
		cas_handle_irq1(cp, status);
	spin_unlock_irqrestore(&cp->lock, flags);
	return IRQ_HANDLED;
}
#endif

static inline void cas_handle_irq(struct net_device *dev,
				  struct cas *cp, const u32 status)
{
	/* housekeeping interrupts */
	if (status & INTR_ERROR_MASK)
		cas_abnormal_irq(dev, cp, status);

	if (status & INTR_RX_BUF_UNAVAIL) {
		/* Frame arrived, no free RX buffers available.
		 * NOTE: we can get this on a link transition.
		 */
		cas_post_rxds_ringN(cp, 0, 0);
		spin_lock(&cp->stat_lock[0]);
		cp->net_stats[0].rx_dropped++;
		spin_unlock(&cp->stat_lock[0]);
	} else if (status & INTR_RX_BUF_AE) {
		cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
				    RX_AE_FREEN_VAL(0));
	}

	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
		cas_post_rxcs_ringN(dev, cp, 0);
}

static irqreturn_t cas_interrupt(int irq, void *dev_id)
{
	struct net_device *dev = dev_id;
	struct cas *cp = netdev_priv(dev);
	unsigned long flags;
	u32 status = readl(cp->regs + REG_INTR_STATUS);

	if (status == 0)
		return IRQ_NONE;

	spin_lock_irqsave(&cp->lock, flags);
	if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
		cas_tx(dev, cp, status);
		status &= ~(INTR_TX_ALL | INTR_TX_INTME);
	}

	if (status & INTR_RX_DONE) {
#ifdef USE_NAPI
		cas_mask_intr(cp);
		napi_schedule(&cp->napi);
#else
		cas_rx_ringN(cp, 0, 0);
#endif
		status &= ~INTR_RX_DONE;
	}

	if (status)
		cas_handle_irq(dev, cp, status);
	spin_unlock_irqrestore(&cp->lock, flags);
	return IRQ_HANDLED;
}


#ifdef USE_NAPI
static int cas_poll(struct napi_struct *napi, int budget)
{
	struct cas *cp = container_of(napi, struct cas, napi);
	struct net_device *dev = cp->dev;
	int i, enable_intr, credits;
	u32 status = readl(cp->regs + REG_INTR_STATUS);
	unsigned long flags;

	spin_lock_irqsave(&cp->lock, flags);
	cas_tx(dev, cp, status);
	spin_unlock_irqrestore(&cp->lock, flags);

	/* NAPI rx packets. we spread the credits across all of the
	 * rxc rings
	 *
	 * to make sure we're fair with the work we loop through each
	 * ring N_RX_COMP_RING times with a request of
	 * budget / N_RX_COMP_RINGS
	 */
	enable_intr = 1;
	credits = 0;
	for (i = 0; i < N_RX_COMP_RINGS; i++) {
		int j;
		for (j = 0; j < N_RX_COMP_RINGS; j++) {
			credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
			if (credits >= budget) {
				enable_intr = 0;
				goto rx_comp;
			}
		}
	}

rx_comp:
	/* final rx completion */
	spin_lock_irqsave(&cp->lock, flags);
	if (status)
		cas_handle_irq(dev, cp, status);

#ifdef USE_PCI_INTB
	if (N_RX_COMP_RINGS > 1) {
		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
		if (status)
			cas_handle_irq1(dev, cp, status);
	}
#endif

#ifdef USE_PCI_INTC
	if (N_RX_COMP_RINGS > 2) {
		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
		if (status)
			cas_handle_irqN(dev, cp, status, 2);
	}
#endif

#ifdef USE_PCI_INTD
	if (N_RX_COMP_RINGS > 3) {
		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
		if (status)
			cas_handle_irqN(dev, cp, status, 3);
	}
#endif
	spin_unlock_irqrestore(&cp->lock, flags);
	if (enable_intr) {
		napi_complete(napi);
		cas_unmask_intr(cp);
	}
	return credits;
}
#endif

#ifdef CONFIG_NET_POLL_CONTROLLER
static void cas_netpoll(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);

	cas_disable_irq(cp, 0);
	cas_interrupt(cp->pdev->irq, dev);
	cas_enable_irq(cp, 0);

#ifdef USE_PCI_INTB
	if (N_RX_COMP_RINGS > 1) {
		/* cas_interrupt1(); */
	}
#endif
#ifdef USE_PCI_INTC
	if (N_RX_COMP_RINGS > 2) {
		/* cas_interruptN(); */
	}
#endif
#ifdef USE_PCI_INTD
	if (N_RX_COMP_RINGS > 3) {
		/* cas_interruptN(); */
	}
#endif
}
#endif

static void cas_tx_timeout(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);

	netdev_err(dev, "transmit timed out, resetting\n");
	if (!cp->hw_running) {
		netdev_err(dev, "hrm.. hw not running!\n");
		return;
	}

	netdev_err(dev, "MIF_STATE[%08x]\n",
		   readl(cp->regs + REG_MIF_STATE_MACHINE));

	netdev_err(dev, "MAC_STATE[%08x]\n",
		   readl(cp->regs + REG_MAC_STATE_MACHINE));

	netdev_err(dev, "TX_STATE[%08x:%08x:%08x] FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
		   readl(cp->regs + REG_TX_CFG),
		   readl(cp->regs + REG_MAC_TX_STATUS),
		   readl(cp->regs + REG_MAC_TX_CFG),
		   readl(cp->regs + REG_TX_FIFO_PKT_CNT),
		   readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
		   readl(cp->regs + REG_TX_FIFO_READ_PTR),
		   readl(cp->regs + REG_TX_SM_1),
		   readl(cp->regs + REG_TX_SM_2));

	netdev_err(dev, "RX_STATE[%08x:%08x:%08x]\n",
		   readl(cp->regs + REG_RX_CFG),
		   readl(cp->regs + REG_MAC_RX_STATUS),
		   readl(cp->regs + REG_MAC_RX_CFG));

	netdev_err(dev, "HP_STATE[%08x:%08x:%08x:%08x]\n",
		   readl(cp->regs + REG_HP_STATE_MACHINE),
		   readl(cp->regs + REG_HP_STATUS0),
		   readl(cp->regs + REG_HP_STATUS1),
		   readl(cp->regs + REG_HP_STATUS2));

#if 1
	atomic_inc(&cp->reset_task_pending);
	atomic_inc(&cp->reset_task_pending_all);
	schedule_work(&cp->reset_task);
#else
	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
	schedule_work(&cp->reset_task);
#endif
}

static inline int cas_intme(int ring, int entry)
{
	/* Algorithm: IRQ every 1/2 of descriptors. */
	if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
		return 1;
	return 0;
}


static void cas_write_txd(struct cas *cp, int ring, int entry,
			  dma_addr_t mapping, int len, u64 ctrl, int last)
{
	struct cas_tx_desc *txd = cp->init_txds[ring] + entry;

	ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
	if (cas_intme(ring, entry))
		ctrl |= TX_DESC_INTME;
	if (last)
		ctrl |= TX_DESC_EOF;
	txd->control = cpu_to_le64(ctrl);
	txd->buffer = cpu_to_le64(mapping);
}

static inline void *tx_tiny_buf(struct cas *cp, const int ring,
				const int entry)
{
	return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
}

static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
				     const int entry, const int tentry)
{
	cp->tx_tiny_use[ring][tentry].nbufs++;
	cp->tx_tiny_use[ring][entry].used = 1;
	return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
}

static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
				    struct sk_buff *skb)
{
	struct net_device *dev = cp->dev;
	int entry, nr_frags, frag, tabort, tentry;
	dma_addr_t mapping;
	unsigned long flags;
	u64 ctrl;
	u32 len;

	spin_lock_irqsave(&cp->tx_lock[ring], flags);

	/* This is a hard error, log it. */
	if (TX_BUFFS_AVAIL(cp, ring) <=
	    CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
		netif_stop_queue(dev);
		spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
		netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
		return 1;
	}

	ctrl = 0;
	if (skb->ip_summed == CHECKSUM_PARTIAL) {
		const u64 csum_start_off = skb_checksum_start_offset(skb);
		const u64 csum_stuff_off = csum_start_off + skb->csum_offset;

		ctrl =  TX_DESC_CSUM_EN |
			CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
			CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
	}

	entry = cp->tx_new[ring];
	cp->tx_skbs[ring][entry] = skb;

	nr_frags = skb_shinfo(skb)->nr_frags;
	len = skb_headlen(skb);
	mapping = pci_map_page(cp->pdev, virt_to_page(skb->data),
			       offset_in_page(skb->data), len,
			       PCI_DMA_TODEVICE);

	tentry = entry;
	tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
	if (unlikely(tabort)) {
		/* NOTE: len is always >  tabort */
		cas_write_txd(cp, ring, entry, mapping, len - tabort,
			      ctrl | TX_DESC_SOF, 0);
		entry = TX_DESC_NEXT(ring, entry);

		skb_copy_from_linear_data_offset(skb, len - tabort,
			      tx_tiny_buf(cp, ring, entry), tabort);
		mapping = tx_tiny_map(cp, ring, entry, tentry);
		cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
			      (nr_frags == 0));
	} else {
		cas_write_txd(cp, ring, entry, mapping, len, ctrl |
			      TX_DESC_SOF, (nr_frags == 0));
	}
	entry = TX_DESC_NEXT(ring, entry);

	for (frag = 0; frag < nr_frags; frag++) {
		skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];

		len = fragp->size;
		mapping = pci_map_page(cp->pdev, fragp->page,
				       fragp->page_offset, len,
				       PCI_DMA_TODEVICE);

		tabort = cas_calc_tabort(cp, fragp->page_offset, len);
		if (unlikely(tabort)) {
			void *addr;

			/* NOTE: len is always > tabort */
			cas_write_txd(cp, ring, entry, mapping, len - tabort,
				      ctrl, 0);
			entry = TX_DESC_NEXT(ring, entry);

			addr = cas_page_map(fragp->page);
			memcpy(tx_tiny_buf(cp, ring, entry),
			       addr + fragp->page_offset + len - tabort,
			       tabort);
			cas_page_unmap(addr);
			mapping = tx_tiny_map(cp, ring, entry, tentry);
			len     = tabort;
		}

		cas_write_txd(cp, ring, entry, mapping, len, ctrl,
			      (frag + 1 == nr_frags));
		entry = TX_DESC_NEXT(ring, entry);
	}

	cp->tx_new[ring] = entry;
	if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
		netif_stop_queue(dev);

	netif_printk(cp, tx_queued, KERN_DEBUG, dev,
		     "tx[%d] queued, slot %d, skblen %d, avail %d\n",
		     ring, entry, skb->len, TX_BUFFS_AVAIL(cp, ring));
	writel(entry, cp->regs + REG_TX_KICKN(ring));
	spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
	return 0;
}

static netdev_tx_t cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);

	/* this is only used as a load-balancing hint, so it doesn't
	 * need to be SMP safe
	 */
	static int ring;

	if (skb_padto(skb, cp->min_frame_size))
		return NETDEV_TX_OK;

	/* XXX: we need some higher-level QoS hooks to steer packets to
	 *      individual queues.
	 */
	if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
		return NETDEV_TX_BUSY;
	return NETDEV_TX_OK;
}

static void cas_init_tx_dma(struct cas *cp)
{
	u64 desc_dma = cp->block_dvma;
	unsigned long off;
	u32 val;
	int i;

	/* set up tx completion writeback registers. must be 8-byte aligned */
#ifdef USE_TX_COMPWB
	off = offsetof(struct cas_init_block, tx_compwb);
	writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
	writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
#endif

	/* enable completion writebacks, enable paced mode,
	 * disable read pipe, and disable pre-interrupt compwbs
	 */
	val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
		TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
		TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
		TX_CFG_INTR_COMPWB_DIS;

	/* write out tx ring info and tx desc bases */
	for (i = 0; i < MAX_TX_RINGS; i++) {
		off = (unsigned long) cp->init_txds[i] -
			(unsigned long) cp->init_block;

		val |= CAS_TX_RINGN_BASE(i);
		writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
		writel((desc_dma + off) & 0xffffffff, cp->regs +
		       REG_TX_DBN_LOW(i));
		/* don't zero out the kick register here as the system
		 * will wedge
		 */
	}
	writel(val, cp->regs + REG_TX_CFG);

	/* program max burst sizes. these numbers should be different
	 * if doing QoS.
	 */
#ifdef USE_QOS
	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
	writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
	writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
	writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
#else
	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
	writel(0x800, cp->regs + REG_TX_MAXBURST_1);
	writel(0x800, cp->regs + REG_TX_MAXBURST_2);
	writel(0x800, cp->regs + REG_TX_MAXBURST_3);
#endif
}

/* Must be invoked under cp->lock. */
static inline void cas_init_dma(struct cas *cp)
{
	cas_init_tx_dma(cp);
	cas_init_rx_dma(cp);
}

static void cas_process_mc_list(struct cas *cp)
{
	u16 hash_table[16];
	u32 crc;
	struct netdev_hw_addr *ha;
	int i = 1;

	memset(hash_table, 0, sizeof(hash_table));
	netdev_for_each_mc_addr(ha, cp->dev) {
		if (i <= CAS_MC_EXACT_MATCH_SIZE) {
			/* use the alternate mac address registers for the
			 * first 15 multicast addresses
			 */
			writel((ha->addr[4] << 8) | ha->addr[5],
			       cp->regs + REG_MAC_ADDRN(i*3 + 0));
			writel((ha->addr[2] << 8) | ha->addr[3],
			       cp->regs + REG_MAC_ADDRN(i*3 + 1));
			writel((ha->addr[0] << 8) | ha->addr[1],
			       cp->regs + REG_MAC_ADDRN(i*3 + 2));
			i++;
		}
		else {
			/* use hw hash table for the next series of
			 * multicast addresses
			 */
			crc = ether_crc_le(ETH_ALEN, ha->addr);
			crc >>= 24;
			hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
		}
	}
	for (i = 0; i < 16; i++)
		writel(hash_table[i], cp->regs + REG_MAC_HASH_TABLEN(i));
}

/* Must be invoked under cp->lock. */
static u32 cas_setup_multicast(struct cas *cp)
{
	u32 rxcfg = 0;
	int i;

	if (cp->dev->flags & IFF_PROMISC) {
		rxcfg |= MAC_RX_CFG_PROMISC_EN;

	} else if (cp->dev->flags & IFF_ALLMULTI) {
	    	for (i=0; i < 16; i++)
			writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;

	} else {
		cas_process_mc_list(cp);
		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
	}

	return rxcfg;
}

/* must be invoked under cp->stat_lock[N_TX_RINGS] */
static void cas_clear_mac_err(struct cas *cp)
{
	writel(0, cp->regs + REG_MAC_COLL_NORMAL);
	writel(0, cp->regs + REG_MAC_COLL_FIRST);
	writel(0, cp->regs + REG_MAC_COLL_EXCESS);
	writel(0, cp->regs + REG_MAC_COLL_LATE);
	writel(0, cp->regs + REG_MAC_TIMER_DEFER);
	writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
	writel(0, cp->regs + REG_MAC_RECV_FRAME);
	writel(0, cp->regs + REG_MAC_LEN_ERR);
	writel(0, cp->regs + REG_MAC_ALIGN_ERR);
	writel(0, cp->regs + REG_MAC_FCS_ERR);
	writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
}


static void cas_mac_reset(struct cas *cp)
{
	int i;

	/* do both TX and RX reset */
	writel(0x1, cp->regs + REG_MAC_TX_RESET);
	writel(0x1, cp->regs + REG_MAC_RX_RESET);

	/* wait for TX */
	i = STOP_TRIES;
	while (i-- > 0) {
		if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
			break;
		udelay(10);
	}

	/* wait for RX */
	i = STOP_TRIES;
	while (i-- > 0) {
		if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
			break;
		udelay(10);
	}

	if (readl(cp->regs + REG_MAC_TX_RESET) |
	    readl(cp->regs + REG_MAC_RX_RESET))
		netdev_err(cp->dev, "mac tx[%d]/rx[%d] reset failed [%08x]\n",
			   readl(cp->regs + REG_MAC_TX_RESET),
			   readl(cp->regs + REG_MAC_RX_RESET),
			   readl(cp->regs + REG_MAC_STATE_MACHINE));
}


/* Must be invoked under cp->lock. */
static void cas_init_mac(struct cas *cp)
{
	unsigned char *e = &cp->dev->dev_addr[0];
	int i;
	cas_mac_reset(cp);

	/* setup core arbitration weight register */
	writel(CAWR_RR_DIS, cp->regs + REG_CAWR);

	/* XXX Use pci_dma_burst_advice() */
#if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
	/* set the infinite burst register for chips that don't have
	 * pci issues.
	 */
	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
		writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
#endif

	writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);

	writel(0x00, cp->regs + REG_MAC_IPG0);
	writel(0x08, cp->regs + REG_MAC_IPG1);
	writel(0x04, cp->regs + REG_MAC_IPG2);

	/* change later for 802.3z */
	writel(0x40, cp->regs + REG_MAC_SLOT_TIME);

	/* min frame + FCS */
	writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);

	/* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
	 * specify the maximum frame size to prevent RX tag errors on
	 * oversized frames.
	 */
	writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
	       CAS_BASE(MAC_FRAMESIZE_MAX_FRAME,
			(CAS_MAX_MTU + ETH_HLEN + 4 + 4)),
	       cp->regs + REG_MAC_FRAMESIZE_MAX);

	/* NOTE: crc_size is used as a surrogate for half-duplex.
	 * workaround saturn half-duplex issue by increasing preamble
	 * size to 65 bytes.
	 */
	if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
		writel(0x41, cp->regs + REG_MAC_PA_SIZE);
	else
		writel(0x07, cp->regs + REG_MAC_PA_SIZE);
	writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
	writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
	writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);

	writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);

	writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
	writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
	writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
	writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
	writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);

	/* setup mac address in perfect filter array */
	for (i = 0; i < 45; i++)
		writel(0x0, cp->regs + REG_MAC_ADDRN(i));

	writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
	writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
	writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));

	writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
	writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
	writel(0x0180, cp->regs + REG_MAC_ADDRN(44));

	cp->mac_rx_cfg = cas_setup_multicast(cp);

	spin_lock(&cp->stat_lock[N_TX_RINGS]);
	cas_clear_mac_err(cp);
	spin_unlock(&cp->stat_lock[N_TX_RINGS]);

	/* Setup MAC interrupts.  We want to get all of the interesting
	 * counter expiration events, but we do not want to hear about
	 * normal rx/tx as the DMA engine tells us that.
	 */
	writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);

	/* Don't enable even the PAUSE interrupts for now, we
	 * make no use of those events other than to record them.
	 */
	writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
}

/* Must be invoked under cp->lock. */
static void cas_init_pause_thresholds(struct cas *cp)
{
	/* Calculate pause thresholds.  Setting the OFF threshold to the
	 * full RX fifo size effectively disables PAUSE generation
	 */
	if (cp->rx_fifo_size <= (2 * 1024)) {
		cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
	} else {
		int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
		if (max_frame * 3 > cp->rx_fifo_size) {
			cp->rx_pause_off = 7104;
			cp->rx_pause_on  = 960;
		} else {
			int off = (cp->rx_fifo_size - (max_frame * 2));
			int on = off - max_frame;
			cp->rx_pause_off = off;
			cp->rx_pause_on = on;
		}
	}
}

static int cas_vpd_match(const void __iomem *p, const char *str)
{
	int len = strlen(str) + 1;
	int i;

	for (i = 0; i < len; i++) {
		if (readb(p + i) != str[i])
			return 0;
	}
	return 1;
}


/* get the mac address by reading the vpd information in the rom.
 * also get the phy type and determine if there's an entropy generator.
 * NOTE: this is a bit convoluted for the following reasons:
 *  1) vpd info has order-dependent mac addresses for multinic cards
 *  2) the only way to determine the nic order is to use the slot
 *     number.
 *  3) fiber cards don't have bridges, so their slot numbers don't
 *     mean anything.
 *  4) we don't actually know we have a fiber card until after
 *     the mac addresses are parsed.
 */
static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
			    const int offset)
{
	void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
	void __iomem *base, *kstart;
	int i, len;
	int found = 0;
#define VPD_FOUND_MAC        0x01
#define VPD_FOUND_PHY        0x02

	int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
	int mac_off  = 0;

	/* give us access to the PROM */
	writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
	       cp->regs + REG_BIM_LOCAL_DEV_EN);

	/* check for an expansion rom */
	if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
		goto use_random_mac_addr;

	/* search for beginning of vpd */
	base = NULL;
	for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
		/* check for PCIR */
		if ((readb(p + i + 0) == 0x50) &&
		    (readb(p + i + 1) == 0x43) &&
		    (readb(p + i + 2) == 0x49) &&
		    (readb(p + i + 3) == 0x52)) {
			base = p + (readb(p + i + 8) |
				    (readb(p + i + 9) << 8));
			break;
		}
	}

	if (!base || (readb(base) != 0x82))
		goto use_random_mac_addr;

	i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
	while (i < EXPANSION_ROM_SIZE) {
		if (readb(base + i) != 0x90) /* no vpd found */
			goto use_random_mac_addr;

		/* found a vpd field */
		len = readb(base + i + 1) | (readb(base + i + 2) << 8);

		/* extract keywords */
		kstart = base + i + 3;
		p = kstart;
		while ((p - kstart) < len) {
			int klen = readb(p + 2);
			int j;
			char type;

			p += 3;

			/* look for the following things:
			 * -- correct length == 29
			 * 3 (type) + 2 (size) +
			 * 18 (strlen("local-mac-address") + 1) +
			 * 6 (mac addr)
			 * -- VPD Instance 'I'
			 * -- VPD Type Bytes 'B'
			 * -- VPD data length == 6
			 * -- property string == local-mac-address
			 *
			 * -- correct length == 24
			 * 3 (type) + 2 (size) +
			 * 12 (strlen("entropy-dev") + 1) +
			 * 7 (strlen("vms110") + 1)
			 * -- VPD Instance 'I'
			 * -- VPD Type String 'B'
			 * -- VPD data length == 7
			 * -- property string == entropy-dev
			 *
			 * -- correct length == 18
			 * 3 (type) + 2 (size) +
			 * 9 (strlen("phy-type") + 1) +
			 * 4 (strlen("pcs") + 1)
			 * -- VPD Instance 'I'
			 * -- VPD Type String 'S'
			 * -- VPD data length == 4
			 * -- property string == phy-type
			 *
			 * -- correct length == 23
			 * 3 (type) + 2 (size) +
			 * 14 (strlen("phy-interface") + 1) +
			 * 4 (strlen("pcs") + 1)
			 * -- VPD Instance 'I'
			 * -- VPD Type String 'S'
			 * -- VPD data length == 4
			 * -- property string == phy-interface
			 */
			if (readb(p) != 'I')
				goto next;

			/* finally, check string and length */
			type = readb(p + 3);
			if (type == 'B') {
				if ((klen == 29) && readb(p + 4) == 6 &&
				    cas_vpd_match(p + 5,
						  "local-mac-address")) {
					if (mac_off++ > offset)
						goto next;

					/* set mac address */
					for (j = 0; j < 6; j++)
						dev_addr[j] =
							readb(p + 23 + j);
					goto found_mac;
				}
			}

			if (type != 'S')
				goto next;

#ifdef USE_ENTROPY_DEV
			if ((klen == 24) &&
			    cas_vpd_match(p + 5, "entropy-dev") &&
			    cas_vpd_match(p + 17, "vms110")) {
				cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
				goto next;
			}
#endif

			if (found & VPD_FOUND_PHY)
				goto next;

			if ((klen == 18) && readb(p + 4) == 4 &&
			    cas_vpd_match(p + 5, "phy-type")) {
				if (cas_vpd_match(p + 14, "pcs")) {
					phy_type = CAS_PHY_SERDES;
					goto found_phy;
				}
			}

			if ((klen == 23) && readb(p + 4) == 4 &&
			    cas_vpd_match(p + 5, "phy-interface")) {
				if (cas_vpd_match(p + 19, "pcs")) {
					phy_type = CAS_PHY_SERDES;
					goto found_phy;
				}
			}
found_mac:
			found |= VPD_FOUND_MAC;
			goto next;

found_phy:
			found |= VPD_FOUND_PHY;

next:
			p += klen;
		}
		i += len + 3;
	}

use_random_mac_addr:
	if (found & VPD_FOUND_MAC)
		goto done;

	/* Sun MAC prefix then 3 random bytes. */
	pr_info("MAC address not found in ROM VPD\n");
	dev_addr[0] = 0x08;
	dev_addr[1] = 0x00;
	dev_addr[2] = 0x20;
	get_random_bytes(dev_addr + 3, 3);

done:
	writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
	return phy_type;
}

/* check pci invariants */
static void cas_check_pci_invariants(struct cas *cp)
{
	struct pci_dev *pdev = cp->pdev;

	cp->cas_flags = 0;
	if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
	    (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
		if (pdev->revision >= CAS_ID_REVPLUS)
			cp->cas_flags |= CAS_FLAG_REG_PLUS;
		if (pdev->revision < CAS_ID_REVPLUS02u)
			cp->cas_flags |= CAS_FLAG_TARGET_ABORT;

		/* Original Cassini supports HW CSUM, but it's not
		 * enabled by default as it can trigger TX hangs.
		 */
		if (pdev->revision < CAS_ID_REV2)
			cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
	} else {
		/* Only sun has original cassini chips.  */
		cp->cas_flags |= CAS_FLAG_REG_PLUS;

		/* We use a flag because the same phy might be externally
		 * connected.
		 */
		if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
		    (pdev->device == PCI_DEVICE_ID_NS_SATURN))
			cp->cas_flags |= CAS_FLAG_SATURN;
	}
}


static int cas_check_invariants(struct cas *cp)
{
	struct pci_dev *pdev = cp->pdev;
	u32 cfg;
	int i;

	/* get page size for rx buffers. */
	cp->page_order = 0;
#ifdef USE_PAGE_ORDER
	if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
		/* see if we can allocate larger pages */
		struct page *page = alloc_pages(GFP_ATOMIC,
						CAS_JUMBO_PAGE_SHIFT -
						PAGE_SHIFT);
		if (page) {
			__free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
			cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
		} else {
			printk("MTU limited to %d bytes\n", CAS_MAX_MTU);
		}
	}
#endif
	cp->page_size = (PAGE_SIZE << cp->page_order);

	/* Fetch the FIFO configurations. */
	cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
	cp->rx_fifo_size = RX_FIFO_SIZE;

	/* finish phy determination. MDIO1 takes precedence over MDIO0 if
	 * they're both connected.
	 */
	cp->phy_type = cas_get_vpd_info(cp, cp->dev->dev_addr,
					PCI_SLOT(pdev->devfn));
	if (cp->phy_type & CAS_PHY_SERDES) {
		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
		return 0; /* no more checking needed */
	}

	/* MII */
	cfg = readl(cp->regs + REG_MIF_CFG);
	if (cfg & MIF_CFG_MDIO_1) {
		cp->phy_type = CAS_PHY_MII_MDIO1;
	} else if (cfg & MIF_CFG_MDIO_0) {
		cp->phy_type = CAS_PHY_MII_MDIO0;
	}

	cas_mif_poll(cp, 0);
	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);

	for (i = 0; i < 32; i++) {
		u32 phy_id;
		int j;

		for (j = 0; j < 3; j++) {
			cp->phy_addr = i;
			phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
			phy_id |= cas_phy_read(cp, MII_PHYSID2);
			if (phy_id && (phy_id != 0xFFFFFFFF)) {
				cp->phy_id = phy_id;
				goto done;
			}
		}
	}
	pr_err("MII phy did not respond [%08x]\n",
	       readl(cp->regs + REG_MIF_STATE_MACHINE));
	return -1;

done:
	/* see if we can do gigabit */
	cfg = cas_phy_read(cp, MII_BMSR);
	if ((cfg & CAS_BMSR_1000_EXTEND) &&
	    cas_phy_read(cp, CAS_MII_1000_EXTEND))
		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
	return 0;
}

/* Must be invoked under cp->lock. */
static inline void cas_start_dma(struct cas *cp)
{
	int i;
	u32 val;
	int txfailed = 0;

	/* enable dma */
	val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
	writel(val, cp->regs + REG_TX_CFG);
	val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
	writel(val, cp->regs + REG_RX_CFG);

	/* enable the mac */
	val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
	writel(val, cp->regs + REG_MAC_TX_CFG);
	val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
	writel(val, cp->regs + REG_MAC_RX_CFG);

	i = STOP_TRIES;
	while (i-- > 0) {
		val = readl(cp->regs + REG_MAC_TX_CFG);
		if ((val & MAC_TX_CFG_EN))
			break;
		udelay(10);
	}
	if (i < 0) txfailed = 1;
	i = STOP_TRIES;
	while (i-- > 0) {
		val = readl(cp->regs + REG_MAC_RX_CFG);
		if ((val & MAC_RX_CFG_EN)) {
			if (txfailed) {
				netdev_err(cp->dev,
					   "enabling mac failed [tx:%08x:%08x]\n",
					   readl(cp->regs + REG_MIF_STATE_MACHINE),
					   readl(cp->regs + REG_MAC_STATE_MACHINE));
			}
			goto enable_rx_done;
		}
		udelay(10);
	}
	netdev_err(cp->dev, "enabling mac failed [%s:%08x:%08x]\n",
		   (txfailed ? "tx,rx" : "rx"),
		   readl(cp->regs + REG_MIF_STATE_MACHINE),
		   readl(cp->regs + REG_MAC_STATE_MACHINE));

enable_rx_done:
	cas_unmask_intr(cp); /* enable interrupts */
	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
	writel(0, cp->regs + REG_RX_COMP_TAIL);

	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
		if (N_RX_DESC_RINGS > 1)
			writel(RX_DESC_RINGN_SIZE(1) - 4,
			       cp->regs + REG_PLUS_RX_KICK1);

		for (i = 1; i < N_RX_COMP_RINGS; i++)
			writel(0, cp->regs + REG_PLUS_RX_COMPN_TAIL(i));
	}
}

/* Must be invoked under cp->lock. */
static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
				   int *pause)
{
	u32 val = readl(cp->regs + REG_PCS_MII_LPA);
	*fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
	*pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
	if (val & PCS_MII_LPA_ASYM_PAUSE)
		*pause |= 0x10;
	*spd = 1000;
}

/* Must be invoked under cp->lock. */
static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
				   int *pause)
{
	u32 val;

	*fd = 0;
	*spd = 10;
	*pause = 0;

	/* use GMII registers */
	val = cas_phy_read(cp, MII_LPA);
	if (val & CAS_LPA_PAUSE)
		*pause = 0x01;

	if (val & CAS_LPA_ASYM_PAUSE)
		*pause |= 0x10;

	if (val & LPA_DUPLEX)
		*fd = 1;
	if (val & LPA_100)
		*spd = 100;

	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
		val = cas_phy_read(cp, CAS_MII_1000_STATUS);
		if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
			*spd = 1000;
		if (val & CAS_LPA_1000FULL)
			*fd = 1;
	}
}

/* A link-up condition has occurred, initialize and enable the
 * rest of the chip.
 *
 * Must be invoked under cp->lock.
 */
static void cas_set_link_modes(struct cas *cp)
{
	u32 val;
	int full_duplex, speed, pause;

	full_duplex = 0;
	speed = 10;
	pause = 0;

	if (CAS_PHY_MII(cp->phy_type)) {
		cas_mif_poll(cp, 0);
		val = cas_phy_read(cp, MII_BMCR);
		if (val & BMCR_ANENABLE) {
			cas_read_mii_link_mode(cp, &full_duplex, &speed,
					       &pause);
		} else {
			if (val & BMCR_FULLDPLX)
				full_duplex = 1;

			if (val & BMCR_SPEED100)
				speed = 100;
			else if (val & CAS_BMCR_SPEED1000)
				speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
					1000 : 100;
		}
		cas_mif_poll(cp, 1);

	} else {
		val = readl(cp->regs + REG_PCS_MII_CTRL);
		cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
		if ((val & PCS_MII_AUTONEG_EN) == 0) {
			if (val & PCS_MII_CTRL_DUPLEX)
				full_duplex = 1;
		}
	}

	netif_info(cp, link, cp->dev, "Link up at %d Mbps, %s-duplex\n",
		   speed, full_duplex ? "full" : "half");

	val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
	if (CAS_PHY_MII(cp->phy_type)) {
		val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
		if (!full_duplex)
			val |= MAC_XIF_DISABLE_ECHO;
	}
	if (full_duplex)
		val |= MAC_XIF_FDPLX_LED;
	if (speed == 1000)
		val |= MAC_XIF_GMII_MODE;
	writel(val, cp->regs + REG_MAC_XIF_CFG);

	/* deal with carrier and collision detect. */
	val = MAC_TX_CFG_IPG_EN;
	if (full_duplex) {
		val |= MAC_TX_CFG_IGNORE_CARRIER;
		val |= MAC_TX_CFG_IGNORE_COLL;
	} else {
#ifndef USE_CSMA_CD_PROTO
		val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
		val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
#endif
	}
	/* val now set up for REG_MAC_TX_CFG */

	/* If gigabit and half-duplex, enable carrier extension
	 * mode.  increase slot time to 512 bytes as well.
	 * else, disable it and make sure slot time is 64 bytes.
	 * also activate checksum bug workaround
	 */
	if ((speed == 1000) && !full_duplex) {
		writel(val | MAC_TX_CFG_CARRIER_EXTEND,
		       cp->regs + REG_MAC_TX_CFG);

		val = readl(cp->regs + REG_MAC_RX_CFG);
		val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
		writel(val | MAC_RX_CFG_CARRIER_EXTEND,
		       cp->regs + REG_MAC_RX_CFG);

		writel(0x200, cp->regs + REG_MAC_SLOT_TIME);

		cp->crc_size = 4;
		/* minimum size gigabit frame at half duplex */
		cp->min_frame_size = CAS_1000MB_MIN_FRAME;

	} else {
		writel(val, cp->regs + REG_MAC_TX_CFG);

		/* checksum bug workaround. don't strip FCS when in
		 * half-duplex mode
		 */
		val = readl(cp->regs + REG_MAC_RX_CFG);
		if (full_duplex) {
			val |= MAC_RX_CFG_STRIP_FCS;
			cp->crc_size = 0;
			cp->min_frame_size = CAS_MIN_MTU;
		} else {
			val &= ~MAC_RX_CFG_STRIP_FCS;
			cp->crc_size = 4;
			cp->min_frame_size = CAS_MIN_FRAME;
		}
		writel(val & ~MAC_RX_CFG_CARRIER_EXTEND,
		       cp->regs + REG_MAC_RX_CFG);
		writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
	}

	if (netif_msg_link(cp)) {
		if (pause & 0x01) {
			netdev_info(cp->dev, "Pause is enabled (rxfifo: %d off: %d on: %d)\n",
				    cp->rx_fifo_size,
				    cp->rx_pause_off,
				    cp->rx_pause_on);
		} else if (pause & 0x10) {
			netdev_info(cp->dev, "TX pause enabled\n");
		} else {
			netdev_info(cp->dev, "Pause is disabled\n");
		}
	}

	val = readl(cp->regs + REG_MAC_CTRL_CFG);
	val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
	if (pause) { /* symmetric or asymmetric pause */
		val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
		if (pause & 0x01) { /* symmetric pause */
			val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
		}
	}
	writel(val, cp->regs + REG_MAC_CTRL_CFG);
	cas_start_dma(cp);
}

/* Must be invoked under cp->lock. */
static void cas_init_hw(struct cas *cp, int restart_link)
{
	if (restart_link)
		cas_phy_init(cp);

	cas_init_pause_thresholds(cp);
	cas_init_mac(cp);
	cas_init_dma(cp);

	if (restart_link) {
		/* Default aneg parameters */
		cp->timer_ticks = 0;
		cas_begin_auto_negotiation(cp, NULL);
	} else if (cp->lstate == link_up) {
		cas_set_link_modes(cp);
		netif_carrier_on(cp->dev);
	}
}

/* Must be invoked under cp->lock. on earlier cassini boards,
 * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
 * let it settle out, and then restore pci state.
 */
static void cas_hard_reset(struct cas *cp)
{
	writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN);
	udelay(20);
	pci_restore_state(cp->pdev);
}


static void cas_global_reset(struct cas *cp, int blkflag)
{
	int limit;

	/* issue a global reset. don't use RSTOUT. */
	if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
		/* For PCS, when the blkflag is set, we should set the
		 * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
		 * the last autonegotiation from being cleared.  We'll
		 * need some special handling if the chip is set into a
		 * loopback mode.
		 */
		writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK),
		       cp->regs + REG_SW_RESET);
	} else {
		writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
	}

	/* need to wait at least 3ms before polling register */
	mdelay(3);

	limit = STOP_TRIES;
	while (limit-- > 0) {
		u32 val = readl(cp->regs + REG_SW_RESET);
		if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
			goto done;
		udelay(10);
	}
	netdev_err(cp->dev, "sw reset failed\n");

done:
	/* enable various BIM interrupts */
	writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE |
	       BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);

	/* clear out pci error status mask for handled errors.
	 * we don't deal with DMA counter overflows as they happen
	 * all the time.
	 */
	writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO |
			       PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE |
			       PCI_ERR_BIM_DMA_READ), cp->regs +
	       REG_PCI_ERR_STATUS_MASK);

	/* set up for MII by default to address mac rx reset timeout
	 * issue
	 */
	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
}

static void cas_reset(struct cas *cp, int blkflag)
{
	u32 val;

	cas_mask_intr(cp);
	cas_global_reset(cp, blkflag);
	cas_mac_reset(cp);
	cas_entropy_reset(cp);

	/* disable dma engines. */
	val = readl(cp->regs + REG_TX_CFG);
	val &= ~TX_CFG_DMA_EN;
	writel(val, cp->regs + REG_TX_CFG);

	val = readl(cp->regs + REG_RX_CFG);
	val &= ~RX_CFG_DMA_EN;
	writel(val, cp->regs + REG_RX_CFG);

	/* program header parser */
	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
	    (CAS_HP_ALT_FIRMWARE == cas_prog_null)) {
		cas_load_firmware(cp, CAS_HP_FIRMWARE);
	} else {
		cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
	}

	/* clear out error registers */
	spin_lock(&cp->stat_lock[N_TX_RINGS]);
	cas_clear_mac_err(cp);
	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
}

/* Shut down the chip, must be called with pm_mutex held.  */
static void cas_shutdown(struct cas *cp)
{
	unsigned long flags;

	/* Make us not-running to avoid timers respawning */
	cp->hw_running = 0;

	del_timer_sync(&cp->link_timer);

	/* Stop the reset task */
#if 0
	while (atomic_read(&cp->reset_task_pending_mtu) ||
	       atomic_read(&cp->reset_task_pending_spare) ||
	       atomic_read(&cp->reset_task_pending_all))
		schedule();

#else
	while (atomic_read(&cp->reset_task_pending))
		schedule();
#endif
	/* Actually stop the chip */
	cas_lock_all_save(cp, flags);
	cas_reset(cp, 0);
	if (cp->cas_flags & CAS_FLAG_SATURN)
		cas_phy_powerdown(cp);
	cas_unlock_all_restore(cp, flags);
}

static int cas_change_mtu(struct net_device *dev, int new_mtu)
{
	struct cas *cp = netdev_priv(dev);

	if (new_mtu < CAS_MIN_MTU || new_mtu > CAS_MAX_MTU)
		return -EINVAL;

	dev->mtu = new_mtu;
	if (!netif_running(dev) || !netif_device_present(dev))
		return 0;

	/* let the reset task handle it */
#if 1
	atomic_inc(&cp->reset_task_pending);
	if ((cp->phy_type & CAS_PHY_SERDES)) {
		atomic_inc(&cp->reset_task_pending_all);
	} else {
		atomic_inc(&cp->reset_task_pending_mtu);
	}
	schedule_work(&cp->reset_task);
#else
	atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ?
		   CAS_RESET_ALL : CAS_RESET_MTU);
	pr_err("reset called in cas_change_mtu\n");
	schedule_work(&cp->reset_task);
#endif

	flush_work_sync(&cp->reset_task);
	return 0;
}

static void cas_clean_txd(struct cas *cp, int ring)
{
	struct cas_tx_desc *txd = cp->init_txds[ring];
	struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
	u64 daddr, dlen;
	int i, size;

	size = TX_DESC_RINGN_SIZE(ring);
	for (i = 0; i < size; i++) {
		int frag;

		if (skbs[i] == NULL)
			continue;

		skb = skbs[i];
		skbs[i] = NULL;

		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
			int ent = i & (size - 1);

			/* first buffer is never a tiny buffer and so
			 * needs to be unmapped.
			 */
			daddr = le64_to_cpu(txd[ent].buffer);
			dlen  =  CAS_VAL(TX_DESC_BUFLEN,
					 le64_to_cpu(txd[ent].control));
			pci_unmap_page(cp->pdev, daddr, dlen,
				       PCI_DMA_TODEVICE);

			if (frag != skb_shinfo(skb)->nr_frags) {
				i++;

				/* next buffer might by a tiny buffer.
				 * skip past it.
				 */
				ent = i & (size - 1);
				if (cp->tx_tiny_use[ring][ent].used)
					i++;
			}
		}
		dev_kfree_skb_any(skb);
	}

	/* zero out tiny buf usage */
	memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
}

/* freed on close */
static inline void cas_free_rx_desc(struct cas *cp, int ring)
{
	cas_page_t **page = cp->rx_pages[ring];
	int i, size;

	size = RX_DESC_RINGN_SIZE(ring);
	for (i = 0; i < size; i++) {
		if (page[i]) {
			cas_page_free(cp, page[i]);
			page[i] = NULL;
		}
	}
}

static void cas_free_rxds(struct cas *cp)
{
	int i;

	for (i = 0; i < N_RX_DESC_RINGS; i++)
		cas_free_rx_desc(cp, i);
}

/* Must be invoked under cp->lock. */
static void cas_clean_rings(struct cas *cp)
{
	int i;

	/* need to clean all tx rings */
	memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
	memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
	for (i = 0; i < N_TX_RINGS; i++)
		cas_clean_txd(cp, i);

	/* zero out init block */
	memset(cp->init_block, 0, sizeof(struct cas_init_block));
	cas_clean_rxds(cp);
	cas_clean_rxcs(cp);
}

/* allocated on open */
static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
{
	cas_page_t **page = cp->rx_pages[ring];
	int size, i = 0;

	size = RX_DESC_RINGN_SIZE(ring);
	for (i = 0; i < size; i++) {
		if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL)
			return -1;
	}
	return 0;
}

static int cas_alloc_rxds(struct cas *cp)
{
	int i;

	for (i = 0; i < N_RX_DESC_RINGS; i++) {
		if (cas_alloc_rx_desc(cp, i) < 0) {
			cas_free_rxds(cp);
			return -1;
		}
	}
	return 0;
}

static void cas_reset_task(struct work_struct *work)
{
	struct cas *cp = container_of(work, struct cas, reset_task);
#if 0
	int pending = atomic_read(&cp->reset_task_pending);
#else
	int pending_all = atomic_read(&cp->reset_task_pending_all);
	int pending_spare = atomic_read(&cp->reset_task_pending_spare);
	int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);

	if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
		/* We can have more tasks scheduled than actually
		 * needed.
		 */
		atomic_dec(&cp->reset_task_pending);
		return;
	}
#endif
	/* The link went down, we reset the ring, but keep
	 * DMA stopped. Use this function for reset
	 * on error as well.
	 */
	if (cp->hw_running) {
		unsigned long flags;

		/* Make sure we don't get interrupts or tx packets */
		netif_device_detach(cp->dev);
		cas_lock_all_save(cp, flags);

		if (cp->opened) {
			/* We call cas_spare_recover when we call cas_open.
			 * but we do not initialize the lists cas_spare_recover
			 * uses until cas_open is called.
			 */
			cas_spare_recover(cp, GFP_ATOMIC);
		}
#if 1
		/* test => only pending_spare set */
		if (!pending_all && !pending_mtu)
			goto done;
#else
		if (pending == CAS_RESET_SPARE)
			goto done;
#endif
		/* when pending == CAS_RESET_ALL, the following
		 * call to cas_init_hw will restart auto negotiation.
		 * Setting the second argument of cas_reset to
		 * !(pending == CAS_RESET_ALL) will set this argument
		 * to 1 (avoiding reinitializing the PHY for the normal
		 * PCS case) when auto negotiation is not restarted.
		 */
#if 1
		cas_reset(cp, !(pending_all > 0));
		if (cp->opened)
			cas_clean_rings(cp);
		cas_init_hw(cp, (pending_all > 0));
#else
		cas_reset(cp, !(pending == CAS_RESET_ALL));
		if (cp->opened)
			cas_clean_rings(cp);
		cas_init_hw(cp, pending == CAS_RESET_ALL);
#endif

done:
		cas_unlock_all_restore(cp, flags);
		netif_device_attach(cp->dev);
	}
#if 1
	atomic_sub(pending_all, &cp->reset_task_pending_all);
	atomic_sub(pending_spare, &cp->reset_task_pending_spare);
	atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
	atomic_dec(&cp->reset_task_pending);
#else
	atomic_set(&cp->reset_task_pending, 0);
#endif
}

static void cas_link_timer(unsigned long data)
{
	struct cas *cp = (struct cas *) data;
	int mask, pending = 0, reset = 0;
	unsigned long flags;

	if (link_transition_timeout != 0 &&
	    cp->link_transition_jiffies_valid &&
	    ((jiffies - cp->link_transition_jiffies) >
	      (link_transition_timeout))) {
		/* One-second counter so link-down workaround doesn't
		 * cause resets to occur so fast as to fool the switch
		 * into thinking the link is down.
		 */
		cp->link_transition_jiffies_valid = 0;
	}

	if (!cp->hw_running)
		return;

	spin_lock_irqsave(&cp->lock, flags);
	cas_lock_tx(cp);
	cas_entropy_gather(cp);

	/* If the link task is still pending, we just
	 * reschedule the link timer
	 */
#if 1
	if (atomic_read(&cp->reset_task_pending_all) ||
	    atomic_read(&cp->reset_task_pending_spare) ||
	    atomic_read(&cp->reset_task_pending_mtu))
		goto done;
#else
	if (atomic_read(&cp->reset_task_pending))
		goto done;
#endif

	/* check for rx cleaning */
	if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
		int i, rmask;

		for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
			rmask = CAS_FLAG_RXD_POST(i);
			if ((mask & rmask) == 0)
				continue;

			/* post_rxds will do a mod_timer */
			if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
				pending = 1;
				continue;
			}
			cp->cas_flags &= ~rmask;
		}
	}

	if (CAS_PHY_MII(cp->phy_type)) {
		u16 bmsr;
		cas_mif_poll(cp, 0);
		bmsr = cas_phy_read(cp, MII_BMSR);
		/* WTZ: Solaris driver reads this twice, but that
		 * may be due to the PCS case and the use of a
		 * common implementation. Read it twice here to be
		 * safe.
		 */
		bmsr = cas_phy_read(cp, MII_BMSR);
		cas_mif_poll(cp, 1);
		readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
		reset = cas_mii_link_check(cp, bmsr);
	} else {
		reset = cas_pcs_link_check(cp);
	}

	if (reset)
		goto done;

	/* check for tx state machine confusion */
	if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
		u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
		u32 wptr, rptr;
		int tlm  = CAS_VAL(MAC_SM_TLM, val);

		if (((tlm == 0x5) || (tlm == 0x3)) &&
		    (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
				     "tx err: MAC_STATE[%08x]\n", val);
			reset = 1;
			goto done;
		}

		val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
		wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
		rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
		if ((val == 0) && (wptr != rptr)) {
			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
				     "tx err: TX_FIFO[%08x:%08x:%08x]\n",
				     val, wptr, rptr);
			reset = 1;
		}

		if (reset)
			cas_hard_reset(cp);
	}

done:
	if (reset) {
#if 1
		atomic_inc(&cp->reset_task_pending);
		atomic_inc(&cp->reset_task_pending_all);
		schedule_work(&cp->reset_task);
#else
		atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
		pr_err("reset called in cas_link_timer\n");
		schedule_work(&cp->reset_task);
#endif
	}

	if (!pending)
		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
	cas_unlock_tx(cp);
	spin_unlock_irqrestore(&cp->lock, flags);
}

/* tiny buffers are used to avoid target abort issues with
 * older cassini's
 */
static void cas_tx_tiny_free(struct cas *cp)
{
	struct pci_dev *pdev = cp->pdev;
	int i;

	for (i = 0; i < N_TX_RINGS; i++) {
		if (!cp->tx_tiny_bufs[i])
			continue;

		pci_free_consistent(pdev, TX_TINY_BUF_BLOCK,
				    cp->tx_tiny_bufs[i],
				    cp->tx_tiny_dvma[i]);
		cp->tx_tiny_bufs[i] = NULL;
	}
}

static int cas_tx_tiny_alloc(struct cas *cp)
{
	struct pci_dev *pdev = cp->pdev;
	int i;

	for (i = 0; i < N_TX_RINGS; i++) {
		cp->tx_tiny_bufs[i] =
			pci_alloc_consistent(pdev, TX_TINY_BUF_BLOCK,
					     &cp->tx_tiny_dvma[i]);
		if (!cp->tx_tiny_bufs[i]) {
			cas_tx_tiny_free(cp);
			return -1;
		}
	}
	return 0;
}


static int cas_open(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	int hw_was_up, err;
	unsigned long flags;

	mutex_lock(&cp->pm_mutex);

	hw_was_up = cp->hw_running;

	/* The power-management mutex protects the hw_running
	 * etc. state so it is safe to do this bit without cp->lock
	 */
	if (!cp->hw_running) {
		/* Reset the chip */
		cas_lock_all_save(cp, flags);
		/* We set the second arg to cas_reset to zero
		 * because cas_init_hw below will have its second
		 * argument set to non-zero, which will force
		 * autonegotiation to start.
		 */
		cas_reset(cp, 0);
		cp->hw_running = 1;
		cas_unlock_all_restore(cp, flags);
	}

	err = -ENOMEM;
	if (cas_tx_tiny_alloc(cp) < 0)
		goto err_unlock;

	/* alloc rx descriptors */
	if (cas_alloc_rxds(cp) < 0)
		goto err_tx_tiny;

	/* allocate spares */
	cas_spare_init(cp);
	cas_spare_recover(cp, GFP_KERNEL);

	/* We can now request the interrupt as we know it's masked
	 * on the controller. cassini+ has up to 4 interrupts
	 * that can be used, but you need to do explicit pci interrupt
	 * mapping to expose them
	 */
	if (request_irq(cp->pdev->irq, cas_interrupt,
			IRQF_SHARED, dev->name, (void *) dev)) {
		netdev_err(cp->dev, "failed to request irq !\n");
		err = -EAGAIN;
		goto err_spare;
	}

#ifdef USE_NAPI
	napi_enable(&cp->napi);
#endif
	/* init hw */
	cas_lock_all_save(cp, flags);
	cas_clean_rings(cp);
	cas_init_hw(cp, !hw_was_up);
	cp->opened = 1;
	cas_unlock_all_restore(cp, flags);

	netif_start_queue(dev);
	mutex_unlock(&cp->pm_mutex);
	return 0;

err_spare:
	cas_spare_free(cp);
	cas_free_rxds(cp);
err_tx_tiny:
	cas_tx_tiny_free(cp);
err_unlock:
	mutex_unlock(&cp->pm_mutex);
	return err;
}

static int cas_close(struct net_device *dev)
{
	unsigned long flags;
	struct cas *cp = netdev_priv(dev);

#ifdef USE_NAPI
	napi_disable(&cp->napi);
#endif
	/* Make sure we don't get distracted by suspend/resume */
	mutex_lock(&cp->pm_mutex);

	netif_stop_queue(dev);

	/* Stop traffic, mark us closed */
	cas_lock_all_save(cp, flags);
	cp->opened = 0;
	cas_reset(cp, 0);
	cas_phy_init(cp);
	cas_begin_auto_negotiation(cp, NULL);
	cas_clean_rings(cp);
	cas_unlock_all_restore(cp, flags);

	free_irq(cp->pdev->irq, (void *) dev);
	cas_spare_free(cp);
	cas_free_rxds(cp);
	cas_tx_tiny_free(cp);
	mutex_unlock(&cp->pm_mutex);
	return 0;
}

static struct {
	const char name[ETH_GSTRING_LEN];
} ethtool_cassini_statnames[] = {
	{"collisions"},
	{"rx_bytes"},
	{"rx_crc_errors"},
	{"rx_dropped"},
	{"rx_errors"},
	{"rx_fifo_errors"},
	{"rx_frame_errors"},
	{"rx_length_errors"},
	{"rx_over_errors"},
	{"rx_packets"},
	{"tx_aborted_errors"},
	{"tx_bytes"},
	{"tx_dropped"},
	{"tx_errors"},
	{"tx_fifo_errors"},
	{"tx_packets"}
};
#define CAS_NUM_STAT_KEYS ARRAY_SIZE(ethtool_cassini_statnames)

static struct {
	const int offsets;	/* neg. values for 2nd arg to cas_read_phy */
} ethtool_register_table[] = {
	{-MII_BMSR},
	{-MII_BMCR},
	{REG_CAWR},
	{REG_INF_BURST},
	{REG_BIM_CFG},
	{REG_RX_CFG},
	{REG_HP_CFG},
	{REG_MAC_TX_CFG},
	{REG_MAC_RX_CFG},
	{REG_MAC_CTRL_CFG},
	{REG_MAC_XIF_CFG},
	{REG_MIF_CFG},
	{REG_PCS_CFG},
	{REG_SATURN_PCFG},
	{REG_PCS_MII_STATUS},
	{REG_PCS_STATE_MACHINE},
	{REG_MAC_COLL_EXCESS},
	{REG_MAC_COLL_LATE}
};
#define CAS_REG_LEN 	ARRAY_SIZE(ethtool_register_table)
#define CAS_MAX_REGS 	(sizeof (u32)*CAS_REG_LEN)

static void cas_read_regs(struct cas *cp, u8 *ptr, int len)
{
	u8 *p;
	int i;
	unsigned long flags;

	spin_lock_irqsave(&cp->lock, flags);
	for (i = 0, p = ptr; i < len ; i ++, p += sizeof(u32)) {
		u16 hval;
		u32 val;
		if (ethtool_register_table[i].offsets < 0) {
			hval = cas_phy_read(cp,
				    -ethtool_register_table[i].offsets);
			val = hval;
		} else {
			val= readl(cp->regs+ethtool_register_table[i].offsets);
		}
		memcpy(p, (u8 *)&val, sizeof(u32));
	}
	spin_unlock_irqrestore(&cp->lock, flags);
}

static struct net_device_stats *cas_get_stats(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	struct net_device_stats *stats = cp->net_stats;
	unsigned long flags;
	int i;
	unsigned long tmp;

	/* we collate all of the stats into net_stats[N_TX_RING] */
	if (!cp->hw_running)
		return stats + N_TX_RINGS;

	/* collect outstanding stats */
	/* WTZ: the Cassini spec gives these as 16 bit counters but
	 * stored in 32-bit words.  Added a mask of 0xffff to be safe,
	 * in case the chip somehow puts any garbage in the other bits.
	 * Also, counter usage didn't seem to mach what Adrian did
	 * in the parts of the code that set these quantities. Made
	 * that consistent.
	 */
	spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
	stats[N_TX_RINGS].rx_crc_errors +=
	  readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
	stats[N_TX_RINGS].rx_frame_errors +=
		readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
	stats[N_TX_RINGS].rx_length_errors +=
		readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
#if 1
	tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
		(readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
	stats[N_TX_RINGS].tx_aborted_errors += tmp;
	stats[N_TX_RINGS].collisions +=
	  tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
#else
	stats[N_TX_RINGS].tx_aborted_errors +=
		readl(cp->regs + REG_MAC_COLL_EXCESS);
	stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
		readl(cp->regs + REG_MAC_COLL_LATE);
#endif
	cas_clear_mac_err(cp);

	/* saved bits that are unique to ring 0 */
	spin_lock(&cp->stat_lock[0]);
	stats[N_TX_RINGS].collisions        += stats[0].collisions;
	stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
	stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
	stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
	stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
	stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
	spin_unlock(&cp->stat_lock[0]);

	for (i = 0; i < N_TX_RINGS; i++) {
		spin_lock(&cp->stat_lock[i]);
		stats[N_TX_RINGS].rx_length_errors +=
			stats[i].rx_length_errors;
		stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
		stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
		stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
		stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
		stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
		stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
		stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
		stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
		stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
		memset(stats + i, 0, sizeof(struct net_device_stats));
		spin_unlock(&cp->stat_lock[i]);
	}
	spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
	return stats + N_TX_RINGS;
}


static void cas_set_multicast(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	u32 rxcfg, rxcfg_new;
	unsigned long flags;
	int limit = STOP_TRIES;

	if (!cp->hw_running)
		return;

	spin_lock_irqsave(&cp->lock, flags);
	rxcfg = readl(cp->regs + REG_MAC_RX_CFG);

	/* disable RX MAC and wait for completion */
	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
		if (!limit--)
			break;
		udelay(10);
	}

	/* disable hash filter and wait for completion */
	limit = STOP_TRIES;
	rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
		if (!limit--)
			break;
		udelay(10);
	}

	/* program hash filters */
	cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
	rxcfg |= rxcfg_new;
	writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
	spin_unlock_irqrestore(&cp->lock, flags);
}

static void cas_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
{
	struct cas *cp = netdev_priv(dev);
	strncpy(info->driver, DRV_MODULE_NAME, ETHTOOL_BUSINFO_LEN);
	strncpy(info->version, DRV_MODULE_VERSION, ETHTOOL_BUSINFO_LEN);
	info->fw_version[0] = '\0';
	strncpy(info->bus_info, pci_name(cp->pdev), ETHTOOL_BUSINFO_LEN);
	info->regdump_len = cp->casreg_len < CAS_MAX_REGS ?
		cp->casreg_len : CAS_MAX_REGS;
	info->n_stats = CAS_NUM_STAT_KEYS;
}

static int cas_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
{
	struct cas *cp = netdev_priv(dev);
	u16 bmcr;
	int full_duplex, speed, pause;
	unsigned long flags;
	enum link_state linkstate = link_up;

	cmd->advertising = 0;
	cmd->supported = SUPPORTED_Autoneg;
	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
		cmd->supported |= SUPPORTED_1000baseT_Full;
		cmd->advertising |= ADVERTISED_1000baseT_Full;
	}

	/* Record PHY settings if HW is on. */
	spin_lock_irqsave(&cp->lock, flags);
	bmcr = 0;
	linkstate = cp->lstate;
	if (CAS_PHY_MII(cp->phy_type)) {
		cmd->port = PORT_MII;
		cmd->transceiver = (cp->cas_flags & CAS_FLAG_SATURN) ?
			XCVR_INTERNAL : XCVR_EXTERNAL;
		cmd->phy_address = cp->phy_addr;
		cmd->advertising |= ADVERTISED_TP | ADVERTISED_MII |
			ADVERTISED_10baseT_Half |
			ADVERTISED_10baseT_Full |
			ADVERTISED_100baseT_Half |
			ADVERTISED_100baseT_Full;

		cmd->supported |=
			(SUPPORTED_10baseT_Half |
			 SUPPORTED_10baseT_Full |
			 SUPPORTED_100baseT_Half |
			 SUPPORTED_100baseT_Full |
			 SUPPORTED_TP | SUPPORTED_MII);

		if (cp->hw_running) {
			cas_mif_poll(cp, 0);
			bmcr = cas_phy_read(cp, MII_BMCR);
			cas_read_mii_link_mode(cp, &full_duplex,
					       &speed, &pause);
			cas_mif_poll(cp, 1);
		}

	} else {
		cmd->port = PORT_FIBRE;
		cmd->transceiver = XCVR_INTERNAL;
		cmd->phy_address = 0;
		cmd->supported   |= SUPPORTED_FIBRE;
		cmd->advertising |= ADVERTISED_FIBRE;

		if (cp->hw_running) {
			/* pcs uses the same bits as mii */
			bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
			cas_read_pcs_link_mode(cp, &full_duplex,
					       &speed, &pause);
		}
	}
	spin_unlock_irqrestore(&cp->lock, flags);

	if (bmcr & BMCR_ANENABLE) {
		cmd->advertising |= ADVERTISED_Autoneg;
		cmd->autoneg = AUTONEG_ENABLE;
		cmd->speed = ((speed == 10) ?
			      SPEED_10 :
			      ((speed == 1000) ?
			       SPEED_1000 : SPEED_100));
		cmd->duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
	} else {
		cmd->autoneg = AUTONEG_DISABLE;
		cmd->speed =
			(bmcr & CAS_BMCR_SPEED1000) ?
			SPEED_1000 :
			((bmcr & BMCR_SPEED100) ? SPEED_100:
			 SPEED_10);
		cmd->duplex =
			(bmcr & BMCR_FULLDPLX) ?
			DUPLEX_FULL : DUPLEX_HALF;
	}
	if (linkstate != link_up) {
		/* Force these to "unknown" if the link is not up and
		 * autonogotiation in enabled. We can set the link
		 * speed to 0, but not cmd->duplex,
		 * because its legal values are 0 and 1.  Ethtool will
		 * print the value reported in parentheses after the
		 * word "Unknown" for unrecognized values.
		 *
		 * If in forced mode, we report the speed and duplex
		 * settings that we configured.
		 */
		if (cp->link_cntl & BMCR_ANENABLE) {
			cmd->speed = 0;
			cmd->duplex = 0xff;
		} else {
			cmd->speed = SPEED_10;
			if (cp->link_cntl & BMCR_SPEED100) {
				cmd->speed = SPEED_100;
			} else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
				cmd->speed = SPEED_1000;
			}
			cmd->duplex = (cp->link_cntl & BMCR_FULLDPLX)?
				DUPLEX_FULL : DUPLEX_HALF;
		}
	}
	return 0;
}

static int cas_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
{
	struct cas *cp = netdev_priv(dev);
	unsigned long flags;

	/* Verify the settings we care about. */
	if (cmd->autoneg != AUTONEG_ENABLE &&
	    cmd->autoneg != AUTONEG_DISABLE)
		return -EINVAL;

	if (cmd->autoneg == AUTONEG_DISABLE &&
	    ((cmd->speed != SPEED_1000 &&
	      cmd->speed != SPEED_100 &&
	      cmd->speed != SPEED_10) ||
	     (cmd->duplex != DUPLEX_HALF &&
	      cmd->duplex != DUPLEX_FULL)))
		return -EINVAL;

	/* Apply settings and restart link process. */
	spin_lock_irqsave(&cp->lock, flags);
	cas_begin_auto_negotiation(cp, cmd);
	spin_unlock_irqrestore(&cp->lock, flags);
	return 0;
}

static int cas_nway_reset(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	unsigned long flags;

	if ((cp->link_cntl & BMCR_ANENABLE) == 0)
		return -EINVAL;

	/* Restart link process. */
	spin_lock_irqsave(&cp->lock, flags);
	cas_begin_auto_negotiation(cp, NULL);
	spin_unlock_irqrestore(&cp->lock, flags);

	return 0;
}

static u32 cas_get_link(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	return cp->lstate == link_up;
}

static u32 cas_get_msglevel(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	return cp->msg_enable;
}

static void cas_set_msglevel(struct net_device *dev, u32 value)
{
	struct cas *cp = netdev_priv(dev);
	cp->msg_enable = value;
}

static int cas_get_regs_len(struct net_device *dev)
{
	struct cas *cp = netdev_priv(dev);
	return cp->casreg_len < CAS_MAX_REGS ? cp->casreg_len: CAS_MAX_REGS;
}

static void cas_get_regs(struct net_device *dev, struct ethtool_regs *regs,
			     void *p)
{
	struct cas *cp = netdev_priv(dev);
	regs->version = 0;
	/* cas_read_regs handles locks (cp->lock).  */
	cas_read_regs(cp, p, regs->len / sizeof(u32));
}

static int cas_get_sset_count(struct net_device *dev, int sset)
{
	switch (sset) {
	case ETH_SS_STATS:
		return CAS_NUM_STAT_KEYS;
	default:
		return -EOPNOTSUPP;
	}
}

static void cas_get_strings(struct net_device *dev, u32 stringset, u8 *data)
{
	 memcpy(data, &ethtool_cassini_statnames,
					 CAS_NUM_STAT_KEYS * ETH_GSTRING_LEN);
}

static void cas_get_ethtool_stats(struct net_device *dev,
				      struct ethtool_stats *estats, u64 *data)
{
	struct cas *cp = netdev_priv(dev);
	struct net_device_stats *stats = cas_get_stats(cp->dev);
	int i = 0;
	data[i++] = stats->collisions;
	data[i++] = stats->rx_bytes;
	data[i++] = stats->rx_crc_errors;
	data[i++] = stats->rx_dropped;
	data[i++] = stats->rx_errors;
	data[i++] = stats->rx_fifo_errors;
	data[i++] = stats->rx_frame_errors;
	data[i++] = stats->rx_length_errors;
	data[i++] = stats->rx_over_errors;
	data[i++] = stats->rx_packets;
	data[i++] = stats->tx_aborted_errors;
	data[i++] = stats->tx_bytes;
	data[i++] = stats->tx_dropped;
	data[i++] = stats->tx_errors;
	data[i++] = stats->tx_fifo_errors;
	data[i++] = stats->tx_packets;
	BUG_ON(i != CAS_NUM_STAT_KEYS);
}

static const struct ethtool_ops cas_ethtool_ops = {
	.get_drvinfo		= cas_get_drvinfo,
	.get_settings		= cas_get_settings,
	.set_settings		= cas_set_settings,
	.nway_reset		= cas_nway_reset,
	.get_link		= cas_get_link,
	.get_msglevel		= cas_get_msglevel,
	.set_msglevel		= cas_set_msglevel,
	.get_regs_len		= cas_get_regs_len,
	.get_regs		= cas_get_regs,
	.get_sset_count		= cas_get_sset_count,
	.get_strings		= cas_get_strings,
	.get_ethtool_stats	= cas_get_ethtool_stats,
};

static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
{
	struct cas *cp = netdev_priv(dev);
	struct mii_ioctl_data *data = if_mii(ifr);
	unsigned long flags;
	int rc = -EOPNOTSUPP;

	/* Hold the PM mutex while doing ioctl's or we may collide
	 * with open/close and power management and oops.
	 */
	mutex_lock(&cp->pm_mutex);
	switch (cmd) {
	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
		data->phy_id = cp->phy_addr;
		/* Fallthrough... */

	case SIOCGMIIREG:		/* Read MII PHY register. */
		spin_lock_irqsave(&cp->lock, flags);
		cas_mif_poll(cp, 0);
		data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
		cas_mif_poll(cp, 1);
		spin_unlock_irqrestore(&cp->lock, flags);
		rc = 0;
		break;

	case SIOCSMIIREG:		/* Write MII PHY register. */
		spin_lock_irqsave(&cp->lock, flags);
		cas_mif_poll(cp, 0);
		rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
		cas_mif_poll(cp, 1);
		spin_unlock_irqrestore(&cp->lock, flags);
		break;
	default:
		break;
	}

	mutex_unlock(&cp->pm_mutex);
	return rc;
}

/* When this chip sits underneath an Intel 31154 bridge, it is the
 * only subordinate device and we can tweak the bridge settings to
 * reflect that fact.
 */
static void __devinit cas_program_bridge(struct pci_dev *cas_pdev)
{
	struct pci_dev *pdev = cas_pdev->bus->self;
	u32 val;

	if (!pdev)
		return;

	if (pdev->vendor != 0x8086 || pdev->device != 0x537c)
		return;

	/* Clear bit 10 (Bus Parking Control) in the Secondary
	 * Arbiter Control/Status Register which lives at offset
	 * 0x41.  Using a 32-bit word read/modify/write at 0x40
	 * is much simpler so that's how we do this.
	 */
	pci_read_config_dword(pdev, 0x40, &val);
	val &= ~0x00040000;
	pci_write_config_dword(pdev, 0x40, val);

	/* Max out the Multi-Transaction Timer settings since
	 * Cassini is the only device present.
	 *
	 * The register is 16-bit and lives at 0x50.  When the
	 * settings are enabled, it extends the GRANT# signal
	 * for a requestor after a transaction is complete.  This
	 * allows the next request to run without first needing
	 * to negotiate the GRANT# signal back.
	 *
	 * Bits 12:10 define the grant duration:
	 *
	 *	1	--	16 clocks
	 *	2	--	32 clocks
	 *	3	--	64 clocks
	 *	4	--	128 clocks
	 *	5	--	256 clocks
	 *
	 * All other values are illegal.
	 *
	 * Bits 09:00 define which REQ/GNT signal pairs get the
	 * GRANT# signal treatment.  We set them all.
	 */
	pci_write_config_word(pdev, 0x50, (5 << 10) | 0x3ff);

	/* The Read Prefecth Policy register is 16-bit and sits at
	 * offset 0x52.  It enables a "smart" pre-fetch policy.  We
	 * enable it and max out all of the settings since only one
	 * device is sitting underneath and thus bandwidth sharing is
	 * not an issue.
	 *
	 * The register has several 3 bit fields, which indicates a
	 * multiplier applied to the base amount of prefetching the
	 * chip would do.  These fields are at:
	 *
	 *	15:13	---	ReRead Primary Bus
	 *	12:10	---	FirstRead Primary Bus
	 *	09:07	---	ReRead Secondary Bus
	 *	06:04	---	FirstRead Secondary Bus
	 *
	 * Bits 03:00 control which REQ/GNT pairs the prefetch settings
	 * get enabled on.  Bit 3 is a grouped enabler which controls
	 * all of the REQ/GNT pairs from [8:3].  Bits 2 to 0 control
	 * the individual REQ/GNT pairs [2:0].
	 */
	pci_write_config_word(pdev, 0x52,
			      (0x7 << 13) |
			      (0x7 << 10) |
			      (0x7 <<  7) |
			      (0x7 <<  4) |
			      (0xf <<  0));

	/* Force cacheline size to 0x8 */
	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);

	/* Force latency timer to maximum setting so Cassini can
	 * sit on the bus as long as it likes.
	 */
	pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0xff);
}

static const struct net_device_ops cas_netdev_ops = {
	.ndo_open		= cas_open,
	.ndo_stop		= cas_close,
	.ndo_start_xmit		= cas_start_xmit,
	.ndo_get_stats 		= cas_get_stats,
	.ndo_set_multicast_list = cas_set_multicast,
	.ndo_do_ioctl		= cas_ioctl,
	.ndo_tx_timeout		= cas_tx_timeout,
	.ndo_change_mtu		= cas_change_mtu,
	.ndo_set_mac_address	= eth_mac_addr,
	.ndo_validate_addr	= eth_validate_addr,
#ifdef CONFIG_NET_POLL_CONTROLLER
	.ndo_poll_controller	= cas_netpoll,
#endif
};

static int __devinit cas_init_one(struct pci_dev *pdev,
				  const struct pci_device_id *ent)
{
	static int cas_version_printed = 0;
	unsigned long casreg_len;
	struct net_device *dev;
	struct cas *cp;
	int i, err, pci_using_dac;
	u16 pci_cmd;
	u8 orig_cacheline_size = 0, cas_cacheline_size = 0;

	if (cas_version_printed++ == 0)
		pr_info("%s", version);

	err = pci_enable_device(pdev);
	if (err) {
		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
		return err;
	}

	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
		dev_err(&pdev->dev, "Cannot find proper PCI device "
		       "base address, aborting\n");
		err = -ENODEV;
		goto err_out_disable_pdev;
	}

	dev = alloc_etherdev(sizeof(*cp));
	if (!dev) {
		dev_err(&pdev->dev, "Etherdev alloc failed, aborting\n");
		err = -ENOMEM;
		goto err_out_disable_pdev;
	}
	SET_NETDEV_DEV(dev, &pdev->dev);

	err = pci_request_regions(pdev, dev->name);
	if (err) {
		dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting\n");
		goto err_out_free_netdev;
	}
	pci_set_master(pdev);

	/* we must always turn on parity response or else parity
	 * doesn't get generated properly. disable SERR/PERR as well.
	 * in addition, we want to turn MWI on.
	 */
	pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
	pci_cmd &= ~PCI_COMMAND_SERR;
	pci_cmd |= PCI_COMMAND_PARITY;
	pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
	if (pci_try_set_mwi(pdev))
		pr_warning("Could not enable MWI for %s\n", pci_name(pdev));

	cas_program_bridge(pdev);

	/*
	 * On some architectures, the default cache line size set
	 * by pci_try_set_mwi reduces perforamnce.  We have to increase
	 * it for this case.  To start, we'll print some configuration
	 * data.
	 */
#if 1
	pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
			     &orig_cacheline_size);
	if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
		cas_cacheline_size =
			(CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ?
			CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
		if (pci_write_config_byte(pdev,
					  PCI_CACHE_LINE_SIZE,
					  cas_cacheline_size)) {
			dev_err(&pdev->dev, "Could not set PCI cache "
			       "line size\n");
			goto err_write_cacheline;
		}
	}
#endif


	/* Configure DMA attributes. */
	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
		pci_using_dac = 1;
		err = pci_set_consistent_dma_mask(pdev,
						  DMA_BIT_MASK(64));
		if (err < 0) {
			dev_err(&pdev->dev, "Unable to obtain 64-bit DMA "
			       "for consistent allocations\n");
			goto err_out_free_res;
		}

	} else {
		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
		if (err) {
			dev_err(&pdev->dev, "No usable DMA configuration, "
			       "aborting\n");
			goto err_out_free_res;
		}
		pci_using_dac = 0;
	}

	casreg_len = pci_resource_len(pdev, 0);

	cp = netdev_priv(dev);
	cp->pdev = pdev;
#if 1
	/* A value of 0 indicates we never explicitly set it */
	cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
#endif
	cp->dev = dev;
	cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE :
	  cassini_debug;

	cp->link_transition = LINK_TRANSITION_UNKNOWN;
	cp->link_transition_jiffies_valid = 0;

	spin_lock_init(&cp->lock);
	spin_lock_init(&cp->rx_inuse_lock);
	spin_lock_init(&cp->rx_spare_lock);
	for (i = 0; i < N_TX_RINGS; i++) {
		spin_lock_init(&cp->stat_lock[i]);
		spin_lock_init(&cp->tx_lock[i]);
	}
	spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
	mutex_init(&cp->pm_mutex);

	init_timer(&cp->link_timer);
	cp->link_timer.function = cas_link_timer;
	cp->link_timer.data = (unsigned long) cp;

#if 1
	/* Just in case the implementation of atomic operations
	 * change so that an explicit initialization is necessary.
	 */
	atomic_set(&cp->reset_task_pending, 0);
	atomic_set(&cp->reset_task_pending_all, 0);
	atomic_set(&cp->reset_task_pending_spare, 0);
	atomic_set(&cp->reset_task_pending_mtu, 0);
#endif
	INIT_WORK(&cp->reset_task, cas_reset_task);

	/* Default link parameters */
	if (link_mode >= 0 && link_mode < 6)
		cp->link_cntl = link_modes[link_mode];
	else
		cp->link_cntl = BMCR_ANENABLE;
	cp->lstate = link_down;
	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
	netif_carrier_off(cp->dev);
	cp->timer_ticks = 0;

	/* give us access to cassini registers */
	cp->regs = pci_iomap(pdev, 0, casreg_len);
	if (!cp->regs) {
		dev_err(&pdev->dev, "Cannot map device registers, aborting\n");
		goto err_out_free_res;
	}
	cp->casreg_len = casreg_len;

	pci_save_state(pdev);
	cas_check_pci_invariants(cp);
	cas_hard_reset(cp);
	cas_reset(cp, 0);
	if (cas_check_invariants(cp))
		goto err_out_iounmap;
	if (cp->cas_flags & CAS_FLAG_SATURN)
		if (cas_saturn_firmware_init(cp))
			goto err_out_iounmap;

	cp->init_block = (struct cas_init_block *)
		pci_alloc_consistent(pdev, sizeof(struct cas_init_block),
				     &cp->block_dvma);
	if (!cp->init_block) {
		dev_err(&pdev->dev, "Cannot allocate init block, aborting\n");
		goto err_out_iounmap;
	}

	for (i = 0; i < N_TX_RINGS; i++)
		cp->init_txds[i] = cp->init_block->txds[i];

	for (i = 0; i < N_RX_DESC_RINGS; i++)
		cp->init_rxds[i] = cp->init_block->rxds[i];

	for (i = 0; i < N_RX_COMP_RINGS; i++)
		cp->init_rxcs[i] = cp->init_block->rxcs[i];

	for (i = 0; i < N_RX_FLOWS; i++)
		skb_queue_head_init(&cp->rx_flows[i]);

	dev->netdev_ops = &cas_netdev_ops;
	dev->ethtool_ops = &cas_ethtool_ops;
	dev->watchdog_timeo = CAS_TX_TIMEOUT;

#ifdef USE_NAPI
	netif_napi_add(dev, &cp->napi, cas_poll, 64);
#endif
	dev->irq = pdev->irq;
	dev->dma = 0;

	/* Cassini features. */
	if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
		dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;

	if (pci_using_dac)
		dev->features |= NETIF_F_HIGHDMA;

	if (register_netdev(dev)) {
		dev_err(&pdev->dev, "Cannot register net device, aborting\n");
		goto err_out_free_consistent;
	}

	i = readl(cp->regs + REG_BIM_CFG);
	netdev_info(dev, "Sun Cassini%s (%sbit/%sMHz PCI/%s) Ethernet[%d] %pM\n",
		    (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "",
		    (i & BIM_CFG_32BIT) ? "32" : "64",
		    (i & BIM_CFG_66MHZ) ? "66" : "33",
		    (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq,
		    dev->dev_addr);

	pci_set_drvdata(pdev, dev);
	cp->hw_running = 1;
	cas_entropy_reset(cp);
	cas_phy_init(cp);
	cas_begin_auto_negotiation(cp, NULL);
	return 0;

err_out_free_consistent:
	pci_free_consistent(pdev, sizeof(struct cas_init_block),
			    cp->init_block, cp->block_dvma);

err_out_iounmap:
	mutex_lock(&cp->pm_mutex);
	if (cp->hw_running)
		cas_shutdown(cp);
	mutex_unlock(&cp->pm_mutex);

	pci_iounmap(pdev, cp->regs);


err_out_free_res:
	pci_release_regions(pdev);

err_write_cacheline:
	/* Try to restore it in case the error occured after we
	 * set it.
	 */
	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);

err_out_free_netdev:
	free_netdev(dev);

err_out_disable_pdev:
	pci_disable_device(pdev);
	pci_set_drvdata(pdev, NULL);
	return -ENODEV;
}

static void __devexit cas_remove_one(struct pci_dev *pdev)
{
	struct net_device *dev = pci_get_drvdata(pdev);
	struct cas *cp;
	if (!dev)
		return;

	cp = netdev_priv(dev);
	unregister_netdev(dev);

	if (cp->fw_data)
		vfree(cp->fw_data);

	mutex_lock(&cp->pm_mutex);
	cancel_work_sync(&cp->reset_task);
	if (cp->hw_running)
		cas_shutdown(cp);
	mutex_unlock(&cp->pm_mutex);

#if 1
	if (cp->orig_cacheline_size) {
		/* Restore the cache line size if we had modified
		 * it.
		 */
		pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
				      cp->orig_cacheline_size);
	}
#endif
	pci_free_consistent(pdev, sizeof(struct cas_init_block),
			    cp->init_block, cp->block_dvma);
	pci_iounmap(pdev, cp->regs);
	free_netdev(dev);
	pci_release_regions(pdev);
	pci_disable_device(pdev);
	pci_set_drvdata(pdev, NULL);
}

#ifdef CONFIG_PM
static int cas_suspend(struct pci_dev *pdev, pm_message_t state)
{
	struct net_device *dev = pci_get_drvdata(pdev);
	struct cas *cp = netdev_priv(dev);
	unsigned long flags;

	mutex_lock(&cp->pm_mutex);

	/* If the driver is opened, we stop the DMA */
	if (cp->opened) {
		netif_device_detach(dev);

		cas_lock_all_save(cp, flags);

		/* We can set the second arg of cas_reset to 0
		 * because on resume, we'll call cas_init_hw with
		 * its second arg set so that autonegotiation is
		 * restarted.
		 */
		cas_reset(cp, 0);
		cas_clean_rings(cp);
		cas_unlock_all_restore(cp, flags);
	}

	if (cp->hw_running)
		cas_shutdown(cp);
	mutex_unlock(&cp->pm_mutex);

	return 0;
}

static int cas_resume(struct pci_dev *pdev)
{
	struct net_device *dev = pci_get_drvdata(pdev);
	struct cas *cp = netdev_priv(dev);

	netdev_info(dev, "resuming\n");

	mutex_lock(&cp->pm_mutex);
	cas_hard_reset(cp);
	if (cp->opened) {
		unsigned long flags;
		cas_lock_all_save(cp, flags);
		cas_reset(cp, 0);
		cp->hw_running = 1;
		cas_clean_rings(cp);
		cas_init_hw(cp, 1);
		cas_unlock_all_restore(cp, flags);

		netif_device_attach(dev);
	}
	mutex_unlock(&cp->pm_mutex);
	return 0;
}
#endif /* CONFIG_PM */

static struct pci_driver cas_driver = {
	.name		= DRV_MODULE_NAME,
	.id_table	= cas_pci_tbl,
	.probe		= cas_init_one,
	.remove		= __devexit_p(cas_remove_one),
#ifdef CONFIG_PM
	.suspend	= cas_suspend,
	.resume		= cas_resume
#endif
};

static int __init cas_init(void)
{
	if (linkdown_timeout > 0)
		link_transition_timeout = linkdown_timeout * HZ;
	else
		link_transition_timeout = 0;

	return pci_register_driver(&cas_driver);
}

static void __exit cas_cleanup(void)
{
	pci_unregister_driver(&cas_driver);
}

module_init(cas_init);
module_exit(cas_cleanup);