/* 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, see <http://www.gnu.org/licenses/>.
 *
 * 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 targeted 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/interrupt.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 <linux/atomic.h>
#include <asm/io.h>
#include <asm/byteorder.h>
#include <asm/uaccess.h>

#define cas_page_map(x)      kmap_atomic((x))
#define cas_page_unmap(x)    kunmap_atomic((x))
#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[] =
        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[] = {
        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 const struct pci_device_id 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_nested(&cp->tx_lock[i], 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 {
                u32 speed = ethtool_cmd_speed(ep);
                cp->link_cntl = 0;
                if (speed == SPEED_100)
                        cp->link_cntl |= BMCR_SPEED100;
                else if (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 void 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;

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

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

        if (!cp->fw_data)
                return;

        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 = netdev_alloc_skb(cp->dev, 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;

                __skb_frag_set_page(frag, page->buffer);
                __skb_frag_ref(frag);
                frag->page_offset = off;
                skb_frag_size_set(frag, 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++;

                        __skb_frag_set_page(frag, page->buffer);
                        __skb_frag_ref(frag);
                        frag->page_offset = 0;
                        skb_frag_size_set(frag, 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 = (irq == cp->pci_irq_INTC) ? 2 : 3;
        u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));

        /* check for shared irq */
        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, 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++) {
                const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];

                len = skb_frag_size(fragp);
                mapping = skb_frag_dma_map(&cp->pdev->dev, fragp, 0, len,
                                           DMA_TO_DEVICE);

                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(skb_frag_page(fragp));
                        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;
}