[NET]: Add Sun Cassini driver.

Written by Adrian Sun (asun@darksunrising.com).
Ported to 2.6.x by Tom 'spot' Callaway <tcallawa@redhat.com>.
Further cleaned up and integrated by David S. Miller

Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 9745 insertions, 0 deletions

diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index 96f14ab1c1f..2a908c4690a 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -548,6 +548,14 @@ config SUNGEM
 	  Support for the Sun GEM chip, aka Sun GigabitEthernet/P 2.0.  See also
 	  <http://www.sun.com/products-n-solutions/hardware/docs/pdf/806-3985-10.pdf>.
 
+config CASSINI
+	tristate "Sun Cassini support"
+	depends on NET_ETHERNET && PCI
+	select CRC32
+	help
+	  Support for the Sun Cassini chip, aka Sun GigaSwift Ethernet. See also
+	  <http://www.sun.com/products-n-solutions/hardware/docs/pdf/817-4341-10.pdf>
+
 config NET_VENDOR_3COM
 	bool "3COM cards"
 	depends on NET_ETHERNET && (ISA || EISA || MCA || PCI)
diff --git a/drivers/net/Makefile b/drivers/net/Makefile
index 8645c843cf4..8aeec9f2495 100644
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -28,6 +28,7 @@ obj-$(CONFIG_SUNQE) += sunqe.o
 obj-$(CONFIG_SUNBMAC) += sunbmac.o
 obj-$(CONFIG_MYRI_SBUS) += myri_sbus.o
 obj-$(CONFIG_SUNGEM) += sungem.o sungem_phy.o
+obj-$(CONFIG_CASSINI) += cassini.o
 
 obj-$(CONFIG_MACE) += mace.o
 obj-$(CONFIG_BMAC) += bmac.o
diff --git a/drivers/net/cassini.c b/drivers/net/cassini.c
new file mode 100644
index 00000000000..69cb368247e
--- /dev/null
+++ b/drivers/net/cassini.c
@@ -0,0 +1,5311 @@
+/* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
+ *
+ * Copyright (C) 2004 Sun Microsystems Inc.
+ * Copyright (C) 2003 Adrian Sun (asun@darksunrising.com)
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
+ * 02111-1307, USA.
+ *
+ * This driver uses the sungem driver (c) David Miller
+ * (davem@redhat.com) as its basis.
+ *
+ * The cassini chip has a number of features that distinguish it from
+ * the gem chip:
+ *  4 transmit descriptor rings that are used for either QoS (VLAN) or
+ *      load balancing (non-VLAN mode)
+ *  batching of multiple packets
+ *  multiple CPU dispatching
+ *  page-based RX descriptor engine with separate completion rings
+ *  Gigabit support (GMII and PCS interface)
+ *  MIF link up/down detection works
+ *
+ * RX is handled by page sized buffers that are attached as fragments to
+ * the skb. here's what's done:
+ *  -- driver allocates pages at a time and keeps reference counts
+ *     on them.
+ *  -- the upper protocol layers assume that the header is in the skb
+ *     itself. as a result, cassini will copy a small amount (64 bytes)
+ *     to make them happy.
+ *  -- driver appends the rest of the data pages as frags to skbuffs
+ *     and increments the reference count
+ *  -- on page reclamation, the driver swaps the page with a spare page.
+ *     if that page is still in use, it frees its reference to that page,
+ *     and allocates a new page for use. otherwise, it just recycles the
+ *     the page. 
+ *
+ * NOTE: cassini can parse the header. however, it's not worth it
+ *       as long as the network stack requires a header copy.
+ *
+ * TX has 4 queues. currently these queues are used in a round-robin
+ * fashion for load balancing. They can also be used for QoS. for that
+ * to work, however, QoS information needs to be exposed down to the driver
+ * level so that subqueues get targetted to particular transmit rings.
+ * alternatively, the queues can be configured via use of the all-purpose
+ * ioctl.
+ *
+ * RX DATA: the rx completion ring has all the info, but the rx desc
+ * ring has all of the data. RX can conceivably come in under multiple
+ * interrupts, but the INT# assignment needs to be set up properly by
+ * the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
+ * that. also, the two descriptor rings are designed to distinguish between
+ * encrypted and non-encrypted packets, but we use them for buffering 
+ * instead.
+ *
+ * by default, the selective clear mask is set up to process rx packets.  
+ */
+
+#include <linux/config.h>
+#include <linux/version.h>
+
+#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/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 <net/checksum.h>
+
+#include <asm/atomic.h>
+#include <asm/system.h>
+#include <asm/io.h>
+#include <asm/byteorder.h>
+#include <asm/uaccess.h>
+
+#define cas_page_map(x)      kmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
+#define cas_page_unmap(x)    kunmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
+#define CAS_NCPUS            num_online_cpus()
+
+#if defined(CONFIG_CASSINI_NAPI) && defined(HAVE_NETDEV_POLL)
+#define USE_NAPI
+#define cas_skb_release(x)  netif_receive_skb(x)
+#else
+#define cas_skb_release(x)  netif_rx(x)
+#endif
+
+/* 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 PFX DRV_MODULE_NAME	": "
+#define DRV_MODULE_VERSION	"1.4"
+#define DRV_MODULE_RELDATE	"1 July 2004"
+
+#define CAS_DEF_MSG_ENABLE	  \
+	(NETIF_MSG_DRV		| \
+	 NETIF_MSG_PROBE	| \
+	 NETIF_MSG_LINK		| \
+	 NETIF_MSG_TIMER	| \
+	 NETIF_MSG_IFDOWN	| \
+	 NETIF_MSG_IFUP		| \
+	 NETIF_MSG_RX_ERR	| \
+	 NETIF_MSG_TX_ERR)
+
+/* length of time before we decide the hardware is borked,
+ * and dev->tx_timeout() should be called to fix the problem
+ */
+#define CAS_TX_TIMEOUT			(HZ)
+#define CAS_LINK_TIMEOUT                (22*HZ/10)
+#define CAS_LINK_FAST_TIMEOUT           (1)
+
+/* timeout values for state changing. these specify the number
+ * of 10us delays to be used before giving up.
+ */
+#define STOP_TRIES_PHY 1000
+#define STOP_TRIES     5000
+
+/* specify a minimum frame size to deal with some fifo issues 
+ * max mtu == 2 * page size - ethernet header - 64 - swivel =
+ *            2 * page_size - 0x50
+ */
+#define CAS_MIN_FRAME			97
+#define CAS_1000MB_MIN_FRAME            255
+#define CAS_MIN_MTU                     60
+#define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
+
+#if 1
+/*
+ * Eliminate these and use separate atomic counters for each, to
+ * avoid a race condition.
+ */
+#else
+#define CAS_RESET_MTU                   1
+#define CAS_RESET_ALL                   2
+#define CAS_RESET_SPARE                 3
+#endif
+
+static char version[] __devinitdata =
+	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
+
+MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
+MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
+MODULE_LICENSE("GPL");
+MODULE_PARM(cassini_debug, "i");
+MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
+MODULE_PARM(link_mode, "i");
+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_PARM(linkdown_timeout, "i");
+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 int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
+static int link_mode;
+
+static u16 link_modes[] __devinitdata = {
+	BMCR_ANENABLE,			 /* 0 : autoneg */
+	0,				 /* 1 : 10bt half duplex */
+	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
+	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
+	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
+	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
+};
+
+static struct pci_device_id cas_pci_tbl[] __devinitdata = {
+	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
+	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
+	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
+	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
+	{ 0, }
+};
+
+MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
+
+static void cas_set_link_modes(struct cas *cp);
+
+static inline void cas_lock_tx(struct cas *cp)
+{
+	int i;
+
+	for (i = 0; i < N_TX_RINGS; i++)  
+		spin_lock(&cp->tx_lock[i]);
+}
+
+static inline void cas_lock_all(struct cas *cp)
+{
+	spin_lock_irq(&cp->lock);
+	cas_lock_tx(cp);
+}
+
+/* WTZ: QA was finding deadlock problems with the previous
+ * versions after long test runs with multiple cards per machine.
+ * See if replacing cas_lock_all with safer versions helps. The
+ * symptoms QA is reporting match those we'd expect if interrupts
+ * aren't being properly restored, and we fixed a previous deadlock
+ * with similar symptoms by using save/restore versions in other
+ * places.
+ */
+#define cas_lock_all_save(cp, flags) \
+do { \
+	struct cas *xxxcp = (cp); \
+	spin_lock_irqsave(&xxxcp->lock, flags); \
+	cas_lock_tx(xxxcp); \
+} while (0)
+
+static inline void cas_unlock_tx(struct cas *cp)
+{
+	int i;
+
+	for (i = N_TX_RINGS; i > 0; i--)  
+		spin_unlock(&cp->tx_lock[i - 1]);  
+}
+
+static inline void cas_unlock_all(struct cas *cp)
+{
+	cas_unlock_tx(cp);
+	spin_unlock_irq(&cp->lock);
+}
+
+#define cas_unlock_all_restore(cp, flags) \
+do { \
+	struct cas *xxxcp = (cp); \
+	cas_unlock_tx(xxxcp); \
+	spin_unlock_irqrestore(&xxxcp->lock, flags); \
+} while (0)
+
+static void cas_disable_irq(struct cas *cp, const int ring)
+{
+	/* Make sure we won't get any more interrupts */
+	if (ring == 0) {
+		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
+		return;
+	}
+
+	/* disable completion interrupts and selectively mask */
+	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
+		switch (ring) {
+#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
+#ifdef USE_PCI_INTB
+		case 1:
+#endif
+#ifdef USE_PCI_INTC
+		case 2:
+#endif
+#ifdef USE_PCI_INTD
+		case 3:
+#endif
+			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN, 
+			       cp->regs + REG_PLUS_INTRN_MASK(ring));
+			break;
+#endif
+		default:
+			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
+			       REG_PLUS_INTRN_MASK(ring));
+			break;
+		}
+	}
+}
+
+static inline void cas_mask_intr(struct cas *cp)
+{
+	int i;
+
+	for (i = 0; i < N_RX_COMP_RINGS; i++)
+		cas_disable_irq(cp, i);
+}
+
+static void cas_enable_irq(struct cas *cp, const int ring)
+{
+	if (ring == 0) { /* all but TX_DONE */
+		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
+		return;
+	}
+
+	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
+		switch (ring) {
+#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
+#ifdef USE_PCI_INTB
+		case 1:
+#endif
+#ifdef USE_PCI_INTC
+		case 2:
+#endif
+#ifdef USE_PCI_INTD
+		case 3:
+#endif
+			writel(INTRN_MASK_RX_EN, cp->regs +
+			       REG_PLUS_INTRN_MASK(ring));
+			break;
+#endif
+		default:
+			break;
+		}
+	}
+}
+
+static inline void cas_unmask_intr(struct cas *cp)
+{
+	int i;
+
+	for (i = 0; i < N_RX_COMP_RINGS; i++)
+		cas_enable_irq(cp, i);
+}
+
+static inline void cas_entropy_gather(struct cas *cp)
+{
+#ifdef USE_ENTROPY_DEV
+	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
+		return;
+
+	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
+			    readl(cp->regs + REG_ENTROPY_IV),
+			    sizeof(uint64_t)*8);
+#endif
+}
+
+static inline void cas_entropy_reset(struct cas *cp)
+{
+#ifdef USE_ENTROPY_DEV
+	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
+		return;
+
+	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT, 
+	       cp->regs + REG_BIM_LOCAL_DEV_EN);
+	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
+	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
+
+	/* if we read back 0x0, we don't have an entropy device */
+	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
+		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
+#endif
+}
+
+/* access to the phy. the following assumes that we've initialized the MIF to 
+ * be in frame rather than bit-bang mode
+ */
+static u16 cas_phy_read(struct cas *cp, int reg)
+{
+	u32 cmd;
+	int limit = STOP_TRIES_PHY;
+
+	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
+	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
+	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
+	cmd |= MIF_FRAME_TURN_AROUND_MSB;
+	writel(cmd, cp->regs + REG_MIF_FRAME);
+	
+	/* poll for completion */
+	while (limit-- > 0) {
+		udelay(10);
+		cmd = readl(cp->regs + REG_MIF_FRAME);
+		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
+			return (cmd & MIF_FRAME_DATA_MASK);
+	}
+	return 0xFFFF; /* -1 */
+}
+
+static int cas_phy_write(struct cas *cp, int reg, u16 val)
+{
+	int limit = STOP_TRIES_PHY;
+	u32 cmd;
+
+	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
+	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
+	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
+	cmd |= MIF_FRAME_TURN_AROUND_MSB;
+	cmd |= val & MIF_FRAME_DATA_MASK;
+	writel(cmd, cp->regs + REG_MIF_FRAME);
+	
+	/* poll for completion */
+	while (limit-- > 0) {
+		udelay(10);
+		cmd = readl(cp->regs + REG_MIF_FRAME);
+		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
+			return 0;
+	}
+	return -1;
+}
+
+static void cas_phy_powerup(struct cas *cp)
+{
+	u16 ctl = cas_phy_read(cp, MII_BMCR);	
+
+	if ((ctl & BMCR_PDOWN) == 0)
+		return;
+	ctl &= ~BMCR_PDOWN;
+	cas_phy_write(cp, MII_BMCR, ctl);
+}
+
+static void cas_phy_powerdown(struct cas *cp)
+{
+	u16 ctl = cas_phy_read(cp, MII_BMCR);	
+
+	if (ctl & BMCR_PDOWN)
+		return;
+	ctl |= BMCR_PDOWN;
+	cas_phy_write(cp, MII_BMCR, ctl);
+}
+
+/* cp->lock held. note: the last put_page will free the buffer */
+static int cas_page_free(struct cas *cp, cas_page_t *page)
+{
+	pci_unmap_page(cp->pdev, page->dma_addr, cp->page_size, 
+		       PCI_DMA_FROMDEVICE);
+	__free_pages(page->buffer, cp->page_order);
+	kfree(page);
+	return 0;
+}
+
+#ifdef RX_COUNT_BUFFERS
+#define RX_USED_ADD(x, y)       ((x)->used += (y))
+#define RX_USED_SET(x, y)       ((x)->used  = (y))
+#else
+#define RX_USED_ADD(x, y) 
+#define RX_USED_SET(x, y)
+#endif
+
+/* local page allocation routines for the receive buffers. jumbo pages
+ * require at least 8K contiguous and 8K aligned buffers.
+ */
+static cas_page_t *cas_page_alloc(struct cas *cp, const int 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(&cp->rx_spare_list, &list);
+	INIT_LIST_HEAD(&cp->rx_spare_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(&cp->rx_inuse_list, &list);
+	INIT_LIST_HEAD(&cp->rx_inuse_list);
+	spin_unlock(&cp->rx_inuse_lock);
+#else
+	spin_lock(&cp->rx_spare_lock);
+	list_splice(&cp->rx_inuse_list, &list);
+	INIT_LIST_HEAD(&cp->rx_inuse_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 int 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(&cp->rx_inuse_list, &list);
+	INIT_LIST_HEAD(&cp->rx_inuse_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);
+
+		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)) {
+			if (netif_msg_rx_err(cp))
+				printk(KERN_ERR "%s: no spare buffers "
+				       "available.\n", cp->dev->name);
+			spin_unlock(&cp->rx_spare_lock);
+			return NULL;
+		}
+	}
+
+	entry = cp->rx_spare_list.next;
+	list_del(entry);
+	recover = ++cp->rx_spares_needed;
+	spin_unlock(&cp->rx_spare_lock);
+
+	/* trigger the timer to do the recovery */
+	if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
+#if 1
+		atomic_inc(&cp->reset_task_pending);
+		atomic_inc(&cp->reset_task_pending_spare);
+		schedule_work(&cp->reset_task);
+#else
+		atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
+		schedule_work(&cp->reset_task);
+#endif
+	}
+	return list_entry(entry, cas_page_t, list);
+}
+
+
+static void cas_mif_poll(struct cas *cp, const int enable)
+{
+	u32 cfg;
+	
+	cfg  = readl(cp->regs + REG_MIF_CFG); 
+	cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
+
+	if (cp->phy_type & CAS_PHY_MII_MDIO1)
+		cfg |= MIF_CFG_PHY_SELECT; 
+
+	/* poll and interrupt on link status change. */
+	if (enable) {
+		cfg |= MIF_CFG_POLL_EN;
+		cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
+		cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
+	}
+	writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF, 
+	       cp->regs + REG_MIF_MASK); 
+	writel(cfg, cp->regs + REG_MIF_CFG);
+}
+
+/* Must be invoked under cp->lock */
+static void cas_begin_auto_negotiation(struct cas *cp, struct ethtool_cmd *ep)
+{
+	u16 ctl;
+#if 1
+	int lcntl;
+	int changed = 0;
+	int oldstate = cp->lstate;
+	int link_was_not_down = !(oldstate == link_down);
+#endif
+	/* Setup link parameters */
+	if (!ep)
+		goto start_aneg;
+	lcntl = cp->link_cntl;
+	if (ep->autoneg == AUTONEG_ENABLE)
+		cp->link_cntl = BMCR_ANENABLE;
+	else {
+		cp->link_cntl = 0;
+		if (ep->speed == SPEED_100)
+			cp->link_cntl |= BMCR_SPEED100;
+		else if (ep->speed == SPEED_1000)
+			cp->link_cntl |= CAS_BMCR_SPEED1000;
+		if (ep->duplex == DUPLEX_FULL)
+			cp->link_cntl |= BMCR_FULLDPLX;
+	}
+#if 1
+	changed = (lcntl != cp->link_cntl);
+#endif
+start_aneg:
+	if (cp->lstate == link_up) {
+		printk(KERN_INFO "%s: PCS link down.\n",
+		       cp->dev->name);
+	} else {
+		if (changed) {
+			printk(KERN_INFO "%s: link configuration changed\n",
+			       cp->dev->name);
+		}
+	}
+	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_load(struct cas *cp)
+{
+	cas_saturn_patch_t *patch = cas_saturn_patch;
+
+	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, patch->addr);
+	while (patch->addr) {
+		cas_phy_write(cp, DP83065_MII_REGD, patch->val);
+		patch++;
+	}
+
+	/* 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)
+			printk(KERN_WARNING "%s: PCS reset bit would not "
+			       "clear [%08x].\n", cp->dev->name,
+			       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)) {
+		if (netif_msg_link(cp))
+			printk(KERN_INFO "%s: PCS RemoteFault\n", 
+			       cp->dev->name);
+	}
+
+	/* 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_msg_link(cp)) {
+			printk(KERN_INFO "%s: PCS link down.\n",
+			       cp->dev->name);
+		}
+
+		/* 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 globa