patch-2.4.22 linux-2.4.22/drivers/atm/he.c

Next file: linux-2.4.22/drivers/atm/he.h
Previous file: linux-2.4.22/drivers/atm/fore200e.c
Back to the patch index
Back to the overall index

diff -urN linux-2.4.21/drivers/atm/he.c linux-2.4.22/drivers/atm/he.c
@@ -0,0 +1,3121 @@
+/* $Id: he.c,v 1.18 2003/05/06 22:57:15 chas Exp $ */
+
+/*
+
+  he.c
+
+  ForeRunnerHE ATM Adapter driver for ATM on Linux
+  Copyright (C) 1999-2001  Naval Research Laboratory
+
+  This library is free software; you can redistribute it and/or
+  modify it under the terms of the GNU Lesser General Public
+  License as published by the Free Software Foundation; either
+  version 2.1 of the License, or (at your option) any later version.
+
+  This library 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
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public
+  License along with this library; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+*/
+
+/*
+
+  he.c
+
+  ForeRunnerHE ATM Adapter driver for ATM on Linux
+  Copyright (C) 1999-2001  Naval Research Laboratory
+
+  Permission to use, copy, modify and distribute this software and its
+  documentation is hereby granted, provided that both the copyright
+  notice and this permission notice appear in all copies of the software,
+  derivative works or modified versions, and any portions thereof, and
+  that both notices appear in supporting documentation.
+
+  NRL ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" CONDITION AND
+  DISCLAIMS ANY LIABILITY OF ANY KIND FOR ANY DAMAGES WHATSOEVER
+  RESULTING FROM THE USE OF THIS SOFTWARE.
+
+  This driver was written using the "Programmer's Reference Manual for
+  ForeRunnerHE(tm)", MANU0361-01 - Rev. A, 08/21/98.
+
+  AUTHORS:
+	chas williams <chas@cmf.nrl.navy.mil>
+	eric kinzie <ekinzie@cmf.nrl.navy.mil>
+
+  NOTES:
+	4096 supported 'connections'
+	group 0 is used for all traffic
+	interrupt queue 0 is used for all interrupts
+	aal0 support (based on work from ulrich.u.muller@nokia.com)
+
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/version.h>
+#include <linux/kernel.h>
+#include <linux/skbuff.h>
+#include <linux/pci.h>
+#include <linux/errno.h>
+#include <linux/types.h>
+#include <linux/string.h>
+#include <linux/delay.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/sched.h>
+#include <linux/timer.h>
+#include <linux/interrupt.h>
+#include <asm/io.h>
+#include <asm/byteorder.h>
+#include <asm/uaccess.h>
+
+#include <linux/atmdev.h>
+#include <linux/atm.h>
+#include <linux/sonet.h>
+
+#define USE_TASKLET
+#undef USE_SCATTERGATHER
+#undef USE_CHECKSUM_HW			/* still confused about this */
+#define USE_RBPS
+#undef USE_RBPS_POOL			/* if memory is tight try this */
+#undef USE_RBPL_POOL			/* if memory is tight try this */
+#define USE_TPD_POOL
+/* #undef CONFIG_ATM_HE_USE_SUNI */
+
+/* compatibility */
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,69)
+typedef void irqreturn_t;
+#define IRQ_NONE
+#define IRQ_HANDLED
+#define IRQ_RETVAL(x)
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,4,9)
+#define __devexit_p(func)		func
+#endif
+
+#ifndef MODULE_LICENSE
+#define MODULE_LICENSE(x)
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,4,3)
+#define pci_set_drvdata(pci_dev, data)	(pci_dev)->driver_data = (data)
+#define pci_get_drvdata(pci_dev)	(pci_dev)->driver_data
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,44)
+#define pci_pool_create(a, b, c, d, e)	pci_pool_create(a, b, c, d, e, SLAB_KERNEL)
+#endif
+
+#include "he.h"
+
+#include "suni.h"
+
+#include <linux/atm_he.h>
+
+#define hprintk(fmt,args...)	printk(KERN_ERR DEV_LABEL "%d: " fmt, he_dev->number , ##args)
+
+#undef DEBUG
+#ifdef DEBUG
+#define HPRINTK(fmt,args...)	printk(KERN_DEBUG DEV_LABEL "%d: " fmt, he_dev->number , ##args)
+#else
+#define HPRINTK(fmt,args...)	do { } while (0)
+#endif /* DEBUG */
+
+
+/* version definition */
+
+static char *version = "$Id: he.c,v 1.18 2003/05/06 22:57:15 chas Exp $";
+
+/* declarations */
+
+static int he_open(struct atm_vcc *vcc, short vpi, int vci);
+static void he_close(struct atm_vcc *vcc);
+static int he_send(struct atm_vcc *vcc, struct sk_buff *skb);
+static int he_sg_send(struct atm_vcc *vcc, unsigned long start, unsigned long size);
+static int he_ioctl(struct atm_dev *dev, unsigned int cmd, void *arg);
+static irqreturn_t he_irq_handler(int irq, void *dev_id, struct pt_regs *regs);
+static void he_tasklet(unsigned long data);
+static int he_proc_read(struct atm_dev *dev,loff_t *pos,char *page);
+static int he_start(struct atm_dev *dev);
+static void he_stop(struct he_dev *dev);
+static void he_phy_put(struct atm_dev *, unsigned char, unsigned long);
+static unsigned char he_phy_get(struct atm_dev *, unsigned long);
+
+static u8 read_prom_byte(struct he_dev *he_dev, int addr);
+
+/* globals */
+
+struct he_dev *he_devs = NULL;
+static short disable64 = -1;
+static short nvpibits = -1;
+static short nvcibits = -1;
+static short rx_skb_reserve = 16;
+static short irq_coalesce = 1;
+static short sdh = 1;
+
+static struct atmdev_ops he_ops =
+{
+	.open =		he_open,
+	.close =	he_close,	
+	.ioctl =	he_ioctl,	
+	.send =		he_send,
+	.sg_send =	he_sg_send,	
+	.phy_put =	he_phy_put,
+	.phy_get =	he_phy_get,
+	.proc_read =	he_proc_read,
+	.owner =	THIS_MODULE
+};
+
+#define he_writel(dev, val, reg)	do { writel(val, (dev)->membase + (reg)); wmb(); } while (0)
+#define he_readl(dev, reg)		readl((dev)->membase + (reg))
+
+/* section 2.12 connection memory access */
+
+static __inline__ void
+he_writel_internal(struct he_dev *he_dev, unsigned val, unsigned addr,
+								unsigned flags)
+{
+	he_writel(he_dev, val, CON_DAT);
+#ifdef CONFIG_IA64_SGI_SN2
+	(void) he_readl(he_dev, CON_DAT);
+#endif
+	he_writel(he_dev, flags | CON_CTL_WRITE | CON_CTL_ADDR(addr), CON_CTL);
+	while (he_readl(he_dev, CON_CTL) & CON_CTL_BUSY);
+}
+
+#define he_writel_rcm(dev, val, reg) 				\
+			he_writel_internal(dev, val, reg, CON_CTL_RCM)
+
+#define he_writel_tcm(dev, val, reg) 				\
+			he_writel_internal(dev, val, reg, CON_CTL_TCM)
+
+#define he_writel_mbox(dev, val, reg) 				\
+			he_writel_internal(dev, val, reg, CON_CTL_MBOX)
+
+static unsigned
+he_readl_internal(struct he_dev *he_dev, unsigned addr, unsigned flags)
+{
+	he_writel(he_dev, flags | CON_CTL_READ | CON_CTL_ADDR(addr), CON_CTL);
+	while (he_readl(he_dev, CON_CTL) & CON_CTL_BUSY);
+	return he_readl(he_dev, CON_DAT);
+}
+
+#define he_readl_rcm(dev, reg) \
+			he_readl_internal(dev, reg, CON_CTL_RCM)
+
+#define he_readl_tcm(dev, reg) \
+			he_readl_internal(dev, reg, CON_CTL_TCM)
+
+#define he_readl_mbox(dev, reg) \
+			he_readl_internal(dev, reg, CON_CTL_MBOX)
+
+
+/* figure 2.2 connection id */
+
+#define he_mkcid(dev, vpi, vci)		(((vpi << (dev)->vcibits) | vci) & 0x1fff)
+
+/* 2.5.1 per connection transmit state registers */
+
+#define he_writel_tsr0(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 0)
+#define he_readl_tsr0(dev, cid) \
+		he_readl_tcm(dev, CONFIG_TSRA | (cid << 3) | 0)
+
+#define he_writel_tsr1(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 1)
+
+#define he_writel_tsr2(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 2)
+
+#define he_writel_tsr3(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 3)
+
+#define he_writel_tsr4(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 4)
+
+	/* from page 2-20
+	 *
+	 * NOTE While the transmit connection is active, bits 23 through 0
+	 *      of this register must not be written by the host.  Byte
+	 *      enables should be used during normal operation when writing
+	 *      the most significant byte.
+	 */
+
+#define he_writel_tsr4_upper(dev, val, cid) \
+		he_writel_internal(dev, val, CONFIG_TSRA | (cid << 3) | 4, \
+							CON_CTL_TCM \
+							| CON_BYTE_DISABLE_2 \
+							| CON_BYTE_DISABLE_1 \
+							| CON_BYTE_DISABLE_0)
+
+#define he_readl_tsr4(dev, cid) \
+		he_readl_tcm(dev, CONFIG_TSRA | (cid << 3) | 4)
+
+#define he_writel_tsr5(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 5)
+
+#define he_writel_tsr6(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 6)
+
+#define he_writel_tsr7(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRA | (cid << 3) | 7)
+
+
+#define he_writel_tsr8(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRB | (cid << 2) | 0)
+
+#define he_writel_tsr9(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRB | (cid << 2) | 1)
+
+#define he_writel_tsr10(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRB | (cid << 2) | 2)
+
+#define he_writel_tsr11(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRB | (cid << 2) | 3)
+
+
+#define he_writel_tsr12(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRC | (cid << 1) | 0)
+
+#define he_writel_tsr13(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRC | (cid << 1) | 1)
+
+
+#define he_writel_tsr14(dev, val, cid) \
+		he_writel_tcm(dev, val, CONFIG_TSRD | cid)
+
+#define he_writel_tsr14_upper(dev, val, cid) \
+		he_writel_internal(dev, val, CONFIG_TSRD | cid, \
+							CON_CTL_TCM \
+							| CON_BYTE_DISABLE_2 \
+							| CON_BYTE_DISABLE_1 \
+							| CON_BYTE_DISABLE_0)
+
+/* 2.7.1 per connection receive state registers */
+
+#define he_writel_rsr0(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 0)
+#define he_readl_rsr0(dev, cid) \
+		he_readl_rcm(dev, 0x00000 | (cid << 3) | 0)
+
+#define he_writel_rsr1(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 1)
+
+#define he_writel_rsr2(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 2)
+
+#define he_writel_rsr3(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 3)
+
+#define he_writel_rsr4(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 4)
+
+#define he_writel_rsr5(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 5)
+
+#define he_writel_rsr6(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 6)
+
+#define he_writel_rsr7(dev, val, cid) \
+		he_writel_rcm(dev, val, 0x00000 | (cid << 3) | 7)
+
+static __inline__ struct atm_vcc*
+he_find_vcc(struct he_dev *he_dev, unsigned cid)
+{
+	unsigned long flags;
+	struct atm_vcc *vcc;
+	short vpi;
+	int vci;
+
+	vpi = cid >> he_dev->vcibits;
+	vci = cid & ((1 << he_dev->vcibits) - 1);
+
+	spin_lock_irqsave(&he_dev->atm_dev->lock, flags);
+	for (vcc = he_dev->atm_dev->vccs; vcc; vcc = vcc->next)
+		if (vcc->vci == vci && vcc->vpi == vpi
+			&& vcc->qos.rxtp.traffic_class != ATM_NONE) {
+				spin_unlock_irqrestore(&he_dev->atm_dev->lock, flags);
+				return vcc;
+			}
+
+	spin_unlock_irqrestore(&he_dev->atm_dev->lock, flags);
+	return NULL;
+}
+
+static int __devinit
+he_init_one(struct pci_dev *pci_dev, const struct pci_device_id *pci_ent)
+{
+	struct atm_dev *atm_dev = NULL;
+	struct he_dev *he_dev = NULL;
+	int err = 0;
+
+	printk(KERN_INFO "he: %s\n", version);
+
+	if (pci_enable_device(pci_dev))
+		return -EIO;
+	if (pci_set_dma_mask(pci_dev, HE_DMA_MASK) != 0) {
+		printk(KERN_WARNING "he: no suitable dma available\n");
+		err = -EIO;
+		goto init_one_failure;
+	}
+
+	atm_dev = atm_dev_register(DEV_LABEL, &he_ops, -1, 0);
+	if (!atm_dev) {
+		err = -ENODEV;
+		goto init_one_failure;
+	}
+	pci_set_drvdata(pci_dev, atm_dev);
+
+	he_dev = (struct he_dev *) kmalloc(sizeof(struct he_dev),
+							GFP_KERNEL);
+	if (!he_dev) {
+		err = -ENOMEM;
+		goto init_one_failure;
+	}
+	memset(he_dev, 0, sizeof(struct he_dev));
+
+	he_dev->pci_dev = pci_dev;
+	he_dev->atm_dev = atm_dev;
+	he_dev->atm_dev->dev_data = he_dev;
+	HE_DEV(atm_dev) = he_dev;
+	he_dev->number = atm_dev->number;
+	if (he_start(atm_dev)) {
+		he_stop(he_dev);
+		err = -ENODEV;
+		goto init_one_failure;
+	}
+	he_dev->next = NULL;
+	if (he_devs)
+		he_dev->next = he_devs;
+	he_devs = he_dev;
+	return 0;
+
+init_one_failure:
+	if (atm_dev)
+		atm_dev_deregister(atm_dev);
+	if (he_dev)
+		kfree(he_dev);
+	pci_disable_device(pci_dev);
+	return err;
+}
+
+static void __devexit
+he_remove_one (struct pci_dev *pci_dev)
+{
+	struct atm_dev *atm_dev;
+	struct he_dev *he_dev;
+
+	atm_dev = pci_get_drvdata(pci_dev);
+	he_dev = HE_DEV(atm_dev);
+
+	/* need to remove from he_devs */
+
+	he_stop(he_dev);
+	atm_dev_deregister(atm_dev);
+	kfree(he_dev);
+
+	pci_set_drvdata(pci_dev, NULL);
+	pci_disable_device(pci_dev);
+}
+
+
+static unsigned
+rate_to_atmf(unsigned rate)		/* cps to atm forum format */
+{
+#define NONZERO (1 << 14)
+
+	unsigned exp = 0;
+
+	if (rate == 0)
+		return 0;
+
+	rate <<= 9;
+	while (rate > 0x3ff) {
+		++exp;
+		rate >>= 1;
+	}
+
+	return (NONZERO | (exp << 9) | (rate & 0x1ff));
+}
+
+static void __init
+he_init_rx_lbfp0(struct he_dev *he_dev)
+{
+	unsigned i, lbm_offset, lbufd_index, lbuf_addr, lbuf_count;
+	unsigned lbufs_per_row = he_dev->cells_per_row / he_dev->cells_per_lbuf;
+	unsigned lbuf_bufsize = he_dev->cells_per_lbuf * ATM_CELL_PAYLOAD;
+	unsigned row_offset = he_dev->r0_startrow * he_dev->bytes_per_row;
+	
+	lbufd_index = 0;
+	lbm_offset = he_readl(he_dev, RCMLBM_BA);
+
+	he_writel(he_dev, lbufd_index, RLBF0_H);
+
+	for (i = 0, lbuf_count = 0; i < he_dev->r0_numbuffs; ++i) {
+		lbufd_index += 2;
+		lbuf_addr = (row_offset + (lbuf_count * lbuf_bufsize)) / 32;
+
+		he_writel_rcm(he_dev, lbuf_addr, lbm_offset);
+		he_writel_rcm(he_dev, lbufd_index, lbm_offset + 1);
+
+		if (++lbuf_count == lbufs_per_row) {
+			lbuf_count = 0;
+			row_offset += he_dev->bytes_per_row;
+		}
+		lbm_offset += 4;
+	}
+		
+	he_writel(he_dev, lbufd_index - 2, RLBF0_T);
+	he_writel(he_dev, he_dev->r0_numbuffs, RLBF0_C);
+}
+
+static void __init
+he_init_rx_lbfp1(struct he_dev *he_dev)
+{
+	unsigned i, lbm_offset, lbufd_index, lbuf_addr, lbuf_count;
+	unsigned lbufs_per_row = he_dev->cells_per_row / he_dev->cells_per_lbuf;
+	unsigned lbuf_bufsize = he_dev->cells_per_lbuf * ATM_CELL_PAYLOAD;
+	unsigned row_offset = he_dev->r1_startrow * he_dev->bytes_per_row;
+	
+	lbufd_index = 1;
+	lbm_offset = he_readl(he_dev, RCMLBM_BA) + (2 * lbufd_index);
+
+	he_writel(he_dev, lbufd_index, RLBF1_H);
+
+	for (i = 0, lbuf_count = 0; i < he_dev->r1_numbuffs; ++i) {
+		lbufd_index += 2;
+		lbuf_addr = (row_offset + (lbuf_count * lbuf_bufsize)) / 32;
+
+		he_writel_rcm(he_dev, lbuf_addr, lbm_offset);
+		he_writel_rcm(he_dev, lbufd_index, lbm_offset + 1);
+
+		if (++lbuf_count == lbufs_per_row) {
+			lbuf_count = 0;
+			row_offset += he_dev->bytes_per_row;
+		}
+		lbm_offset += 4;
+	}
+		
+	he_writel(he_dev, lbufd_index - 2, RLBF1_T);
+	he_writel(he_dev, he_dev->r1_numbuffs, RLBF1_C);
+}
+
+static void __init
+he_init_tx_lbfp(struct he_dev *he_dev)
+{
+	unsigned i, lbm_offset, lbufd_index, lbuf_addr, lbuf_count;
+	unsigned lbufs_per_row = he_dev->cells_per_row / he_dev->cells_per_lbuf;
+	unsigned lbuf_bufsize = he_dev->cells_per_lbuf * ATM_CELL_PAYLOAD;
+	unsigned row_offset = he_dev->tx_startrow * he_dev->bytes_per_row;
+	
+	lbufd_index = he_dev->r0_numbuffs + he_dev->r1_numbuffs;
+	lbm_offset = he_readl(he_dev, RCMLBM_BA) + (2 * lbufd_index);
+
+	he_writel(he_dev, lbufd_index, TLBF_H);
+
+	for (i = 0, lbuf_count = 0; i < he_dev->tx_numbuffs; ++i) {
+		lbufd_index += 1;
+		lbuf_addr = (row_offset + (lbuf_count * lbuf_bufsize)) / 32;
+
+		he_writel_rcm(he_dev, lbuf_addr, lbm_offset);
+		he_writel_rcm(he_dev, lbufd_index, lbm_offset + 1);
+
+		if (++lbuf_count == lbufs_per_row) {
+			lbuf_count = 0;
+			row_offset += he_dev->bytes_per_row;
+		}
+		lbm_offset += 2;
+	}
+		
+	he_writel(he_dev, lbufd_index - 1, TLBF_T);
+}
+
+static int __init
+he_init_tpdrq(struct he_dev *he_dev)
+{
+	he_dev->tpdrq_base = pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_TPDRQ_SIZE * sizeof(struct he_tpdrq), &he_dev->tpdrq_phys);
+	if (he_dev->tpdrq_base == NULL) {
+		hprintk("failed to alloc tpdrq\n");
+		return -ENOMEM;
+	}
+	memset(he_dev->tpdrq_base, 0,
+				CONFIG_TPDRQ_SIZE * sizeof(struct he_tpdrq));
+
+	he_dev->tpdrq_tail = he_dev->tpdrq_base;
+	he_dev->tpdrq_head = he_dev->tpdrq_base;
+
+	he_writel(he_dev, he_dev->tpdrq_phys, TPDRQ_B_H);
+	he_writel(he_dev, 0, TPDRQ_T);	
+	he_writel(he_dev, CONFIG_TPDRQ_SIZE - 1, TPDRQ_S);
+
+	return 0;
+}
+
+static void __init
+he_init_cs_block(struct he_dev *he_dev)
+{
+	unsigned clock, rate, delta;
+	int reg;
+
+	/* 5.1.7 cs block initialization */
+
+	for (reg = 0; reg < 0x20; ++reg)
+		he_writel_mbox(he_dev, 0x0, CS_STTIM0 + reg);
+
+	/* rate grid timer reload values */
+
+	clock = he_is622(he_dev) ? 66667000 : 50000000;
+	rate = he_dev->atm_dev->link_rate;
+	delta = rate / 16 / 2;
+
+	for (reg = 0; reg < 0x10; ++reg) {
+		/* 2.4 internal transmit function
+		 *
+	 	 * we initialize the first row in the rate grid.
+		 * values are period (in clock cycles) of timer
+		 */
+		unsigned period = clock / rate;
+
+		he_writel_mbox(he_dev, period, CS_TGRLD0 + reg);
+		rate -= delta;
+	}
+
+	if (he_is622(he_dev)) {
+		/* table 5.2 (4 cells per lbuf) */
+		he_writel_mbox(he_dev, 0x000800fa, CS_ERTHR0);
+		he_writel_mbox(he_dev, 0x000c33cb, CS_ERTHR1);
+		he_writel_mbox(he_dev, 0x0010101b, CS_ERTHR2);
+		he_writel_mbox(he_dev, 0x00181dac, CS_ERTHR3);
+		he_writel_mbox(he_dev, 0x00280600, CS_ERTHR4);
+
+		/* table 5.3, 5.4, 5.5, 5.6, 5.7 */
+		he_writel_mbox(he_dev, 0x023de8b3, CS_ERCTL0);
+		he_writel_mbox(he_dev, 0x1801, CS_ERCTL1);
+		he_writel_mbox(he_dev, 0x68b3, CS_ERCTL2);
+		he_writel_mbox(he_dev, 0x1280, CS_ERSTAT0);
+		he_writel_mbox(he_dev, 0x68b3, CS_ERSTAT1);
+		he_writel_mbox(he_dev, 0x14585, CS_RTFWR);
+
+		he_writel_mbox(he_dev, 0x4680, CS_RTATR);
+
+		/* table 5.8 */
+		he_writel_mbox(he_dev, 0x00159ece, CS_TFBSET);
+		he_writel_mbox(he_dev, 0x68b3, CS_WCRMAX);
+		he_writel_mbox(he_dev, 0x5eb3, CS_WCRMIN);
+		he_writel_mbox(he_dev, 0xe8b3, CS_WCRINC);
+		he_writel_mbox(he_dev, 0xdeb3, CS_WCRDEC);
+		he_writel_mbox(he_dev, 0x68b3, CS_WCRCEIL);
+
+		/* table 5.9 */
+		he_writel_mbox(he_dev, 0x5, CS_OTPPER);
+		he_writel_mbox(he_dev, 0x14, CS_OTWPER);
+	} else {
+		/* table 5.1 (4 cells per lbuf) */
+		he_writel_mbox(he_dev, 0x000400ea, CS_ERTHR0);
+		he_writel_mbox(he_dev, 0x00063388, CS_ERTHR1);
+		he_writel_mbox(he_dev, 0x00081018, CS_ERTHR2);
+		he_writel_mbox(he_dev, 0x000c1dac, CS_ERTHR3);
+		he_writel_mbox(he_dev, 0x0014051a, CS_ERTHR4);
+
+		/* table 5.3, 5.4, 5.5, 5.6, 5.7 */
+		he_writel_mbox(he_dev, 0x0235e4b1, CS_ERCTL0);
+		he_writel_mbox(he_dev, 0x4701, CS_ERCTL1);
+		he_writel_mbox(he_dev, 0x64b1, CS_ERCTL2);
+		he_writel_mbox(he_dev, 0x1280, CS_ERSTAT0);
+		he_writel_mbox(he_dev, 0x64b1, CS_ERSTAT1);
+		he_writel_mbox(he_dev, 0xf424, CS_RTFWR);
+
+		he_writel_mbox(he_dev, 0x4680, CS_RTATR);
+
+		/* table 5.8 */
+		he_writel_mbox(he_dev, 0x000563b7, CS_TFBSET);
+		he_writel_mbox(he_dev, 0x64b1, CS_WCRMAX);
+		he_writel_mbox(he_dev, 0x5ab1, CS_WCRMIN);
+		he_writel_mbox(he_dev, 0xe4b1, CS_WCRINC);
+		he_writel_mbox(he_dev, 0xdab1, CS_WCRDEC);
+		he_writel_mbox(he_dev, 0x64b1, CS_WCRCEIL);
+
+		/* table 5.9 */
+		he_writel_mbox(he_dev, 0x6, CS_OTPPER);
+		he_writel_mbox(he_dev, 0x1e, CS_OTWPER);
+	}
+
+	he_writel_mbox(he_dev, 0x8, CS_OTTLIM);
+
+	for (reg = 0; reg < 0x8; ++reg)
+		he_writel_mbox(he_dev, 0x0, CS_HGRRT0 + reg);
+
+}
+
+static int __init
+he_init_cs_block_rcm(struct he_dev *he_dev)
+{
+	unsigned (*rategrid)[16][16];
+	unsigned rate, delta;
+	int i, j, reg;
+
+	unsigned rate_atmf, exp, man;
+	unsigned long long rate_cps;
+	int mult, buf, buf_limit = 4;
+
+	rategrid = kmalloc( sizeof(unsigned) * 16 * 16, GFP_KERNEL);
+	if (!rategrid)
+		return -ENOMEM;
+
+	/* initialize rate grid group table */
+
+	for (reg = 0x0; reg < 0xff; ++reg)
+		he_writel_rcm(he_dev, 0x0, CONFIG_RCMABR + reg);
+
+	/* initialize rate controller groups */
+
+	for (reg = 0x100; reg < 0x1ff; ++reg)
+		he_writel_rcm(he_dev, 0x0, CONFIG_RCMABR + reg);
+	
+	/* initialize tNrm lookup table */
+
+	/* the manual makes reference to a routine in a sample driver
+	   for proper configuration; fortunately, we only need this
+	   in order to support abr connection */
+	
+	/* initialize rate to group table */
+
+	rate = he_dev->atm_dev->link_rate;
+	delta = rate / 32;
+
+	/*
+	 * 2.4 transmit internal functions
+	 * 
+	 * we construct a copy of the rate grid used by the scheduler
+	 * in order to construct the rate to group table below
+	 */
+
+	for (j = 0; j < 16; j++) {
+		(*rategrid)[0][j] = rate;
+		rate -= delta;
+	}
+
+	for (i = 1; i < 16; i++)
+		for (j = 0; j < 16; j++)
+			if (i > 14)
+				(*rategrid)[i][j] = (*rategrid)[i - 1][j] / 4;
+			else
+				(*rategrid)[i][j] = (*rategrid)[i - 1][j] / 2;
+
+	/*
+	 * 2.4 transmit internal function
+	 *
+	 * this table maps the upper 5 bits of exponent and mantissa
+	 * of the atm forum representation of the rate into an index
+	 * on rate grid  
+	 */
+
+	rate_atmf = 0;
+	while (rate_atmf < 0x400) {
+		man = (rate_atmf & 0x1f) << 4;
+		exp = rate_atmf >> 5;
+
+		/* 
+			instead of '/ 512', use '>> 9' to prevent a call
+			to divdu3 on x86 platforms
+		*/
+		rate_cps = (unsigned long long) (1 << exp) * (man + 512) >> 9;
+
+		if (rate_cps < 10)
+			rate_cps = 10;	/* 2.2.1 minimum payload rate is 10 cps */
+
+		for (i = 255; i > 0; i--)
+			if ((*rategrid)[i/16][i%16] >= rate_cps)
+				break;	 /* pick nearest rate instead? */
+
+		/*
+		 * each table entry is 16 bits: (rate grid index (8 bits)
+		 * and a buffer limit (8 bits)
+		 * there are two table entries in each 32-bit register
+		 */
+
+#ifdef notdef
+		buf = rate_cps * he_dev->tx_numbuffs /
+				(he_dev->atm_dev->link_rate * 2);
+#else
+		/* this is pretty, but avoids _divdu3 and is mostly correct */
+		mult = he_dev->atm_dev->link_rate / ATM_OC3_PCR;
+		if (rate_cps > (272 * mult))
+			buf = 4;
+		else if (rate_cps > (204 * mult))
+			buf = 3;
+		else if (rate_cps > (136 * mult))
+			buf = 2;
+		else if (rate_cps > (68 * mult))
+			buf = 1;
+		else
+			buf = 0;
+#endif
+		if (buf > buf_limit)
+			buf = buf_limit;
+		reg = (reg << 16) | ((i << 8) | buf);
+
+#define RTGTBL_OFFSET 0x400
+	  
+		if (rate_atmf & 0x1)
+			he_writel_rcm(he_dev, reg,
+				CONFIG_RCMABR + RTGTBL_OFFSET + (rate_atmf >> 1));
+
+		++rate_atmf;
+	}
+
+	kfree(rategrid);
+	return 0;
+}
+
+static int __init
+he_init_group(struct he_dev *he_dev, int group)
+{
+	int i;
+
+#ifdef USE_RBPS
+	/* small buffer pool */
+#ifdef USE_RBPS_POOL
+	he_dev->rbps_pool = pci_pool_create("rbps", he_dev->pci_dev,
+			CONFIG_RBPS_BUFSIZE, 8, 0);
+	if (he_dev->rbps_pool == NULL) {
+		hprintk("unable to create rbps pages\n");
+		return -ENOMEM;
+	}
+#else /* !USE_RBPS_POOL */
+	he_dev->rbps_pages = pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_RBPS_SIZE * CONFIG_RBPS_BUFSIZE, &he_dev->rbps_pages_phys);
+	if (he_dev->rbps_pages == NULL) {
+		hprintk("unable to create rbps page pool\n");
+		return -ENOMEM;
+	}
+#endif /* USE_RBPS_POOL */
+
+	he_dev->rbps_base = pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_RBPS_SIZE * sizeof(struct he_rbp), &he_dev->rbps_phys);
+	if (he_dev->rbps_base == NULL) {
+		hprintk("failed to alloc rbps\n");
+		return -ENOMEM;
+	}
+	memset(he_dev->rbps_base, 0, CONFIG_RBPS_SIZE * sizeof(struct he_rbp));
+	he_dev->rbps_virt = kmalloc(CONFIG_RBPS_SIZE * sizeof(struct he_virt), GFP_KERNEL);
+
+	for (i = 0; i < CONFIG_RBPS_SIZE; ++i) {
+		dma_addr_t dma_handle;
+		void *cpuaddr;
+
+#ifdef USE_RBPS_POOL 
+		cpuaddr = pci_pool_alloc(he_dev->rbps_pool, SLAB_KERNEL|SLAB_DMA, &dma_handle);
+		if (cpuaddr == NULL)
+			return -ENOMEM;
+#else
+		cpuaddr = he_dev->rbps_pages + (i * CONFIG_RBPS_BUFSIZE);
+		dma_handle = he_dev->rbps_pages_phys + (i * CONFIG_RBPS_BUFSIZE);
+#endif
+
+		he_dev->rbps_virt[i].virt = cpuaddr;
+		he_dev->rbps_base[i].status = RBP_LOANED | RBP_SMALLBUF | (i << RBP_INDEX_OFF);
+		he_dev->rbps_base[i].phys = dma_handle;
+
+	}
+	he_dev->rbps_tail = &he_dev->rbps_base[CONFIG_RBPS_SIZE - 1];
+
+	he_writel(he_dev, he_dev->rbps_phys, G0_RBPS_S + (group * 32));
+	he_writel(he_dev, RBPS_MASK(he_dev->rbps_tail),
+						G0_RBPS_T + (group * 32));
+	he_writel(he_dev, CONFIG_RBPS_BUFSIZE/4,
+						G0_RBPS_BS + (group * 32));
+	he_writel(he_dev,
+			RBP_THRESH(CONFIG_RBPS_THRESH) |
+			RBP_QSIZE(CONFIG_RBPS_SIZE - 1) |
+			RBP_INT_ENB,
+						G0_RBPS_QI + (group * 32));
+#else /* !USE_RBPS */
+	he_writel(he_dev, 0x0, G0_RBPS_S + (group * 32));
+	he_writel(he_dev, 0x0, G0_RBPS_T + (group * 32));
+	he_writel(he_dev, 0x0, G0_RBPS_QI + (group * 32));
+	he_writel(he_dev, RBP_THRESH(0x1) | RBP_QSIZE(0x0),
+						G0_RBPS_BS + (group * 32));
+#endif /* USE_RBPS */
+
+	/* large buffer pool */
+#ifdef USE_RBPL_POOL
+	he_dev->rbpl_pool = pci_pool_create("rbpl", he_dev->pci_dev,
+			CONFIG_RBPL_BUFSIZE, 8, 0);
+	if (he_dev->rbpl_pool == NULL) {
+		hprintk("unable to create rbpl pool\n");
+		return -ENOMEM;
+	}
+#else /* !USE_RBPL_POOL */
+	he_dev->rbpl_pages = (void *) pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_RBPL_SIZE * CONFIG_RBPL_BUFSIZE, &he_dev->rbpl_pages_phys);
+	if (he_dev->rbpl_pages == NULL) {
+		hprintk("unable to create rbpl pages\n");
+		return -ENOMEM;
+	}
+#endif /* USE_RBPL_POOL */
+
+	he_dev->rbpl_base = pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_RBPL_SIZE * sizeof(struct he_rbp), &he_dev->rbpl_phys);
+	if (he_dev->rbpl_base == NULL) {
+		hprintk("failed to alloc rbpl\n");
+		return -ENOMEM;
+	}
+	memset(he_dev->rbpl_base, 0, CONFIG_RBPL_SIZE * sizeof(struct he_rbp));
+	he_dev->rbpl_virt = kmalloc(CONFIG_RBPL_SIZE * sizeof(struct he_virt), GFP_KERNEL);
+
+	for (i = 0; i < CONFIG_RBPL_SIZE; ++i) {
+		dma_addr_t dma_handle;
+		void *cpuaddr;
+
+#ifdef USE_RBPL_POOL
+		cpuaddr = pci_pool_alloc(he_dev->rbpl_pool, SLAB_KERNEL|SLAB_DMA, &dma_handle);
+		if (cpuaddr == NULL)
+			return -ENOMEM;
+#else
+		cpuaddr = he_dev->rbpl_pages + (i * CONFIG_RBPL_BUFSIZE);
+		dma_handle = he_dev->rbpl_pages_phys + (i * CONFIG_RBPL_BUFSIZE);
+#endif
+
+		he_dev->rbpl_virt[i].virt = cpuaddr;
+		he_dev->rbpl_base[i].status = RBP_LOANED | (i << RBP_INDEX_OFF);
+		he_dev->rbpl_base[i].phys = dma_handle;
+	}
+	he_dev->rbpl_tail = &he_dev->rbpl_base[CONFIG_RBPL_SIZE - 1];
+
+	he_writel(he_dev, he_dev->rbpl_phys, G0_RBPL_S + (group * 32));
+	he_writel(he_dev, RBPL_MASK(he_dev->rbpl_tail),
+						G0_RBPL_T + (group * 32));
+	he_writel(he_dev, CONFIG_RBPL_BUFSIZE/4,
+						G0_RBPL_BS + (group * 32));
+	he_writel(he_dev,
+			RBP_THRESH(CONFIG_RBPL_THRESH) |
+			RBP_QSIZE(CONFIG_RBPL_SIZE - 1) |
+			RBP_INT_ENB,
+						G0_RBPL_QI + (group * 32));
+
+	/* rx buffer ready queue */
+
+	he_dev->rbrq_base = pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_RBRQ_SIZE * sizeof(struct he_rbrq), &he_dev->rbrq_phys);
+	if (he_dev->rbrq_base == NULL) {
+		hprintk("failed to allocate rbrq\n");
+		return -ENOMEM;
+	}
+	memset(he_dev->rbrq_base, 0, CONFIG_RBRQ_SIZE * sizeof(struct he_rbrq));
+
+	he_dev->rbrq_head = he_dev->rbrq_base;
+	he_writel(he_dev, he_dev->rbrq_phys, G0_RBRQ_ST + (group * 16));
+	he_writel(he_dev, 0, G0_RBRQ_H + (group * 16));
+	he_writel(he_dev,
+		RBRQ_THRESH(CONFIG_RBRQ_THRESH) | RBRQ_SIZE(CONFIG_RBRQ_SIZE - 1),
+						G0_RBRQ_Q + (group * 16));
+	if (irq_coalesce) {
+		hprintk("coalescing interrupts\n");
+		he_writel(he_dev, RBRQ_TIME(768) | RBRQ_COUNT(7),
+						G0_RBRQ_I + (group * 16));
+	} else
+		he_writel(he_dev, RBRQ_TIME(0) | RBRQ_COUNT(1),
+						G0_RBRQ_I + (group * 16));
+
+	/* tx buffer ready queue */
+
+	he_dev->tbrq_base = pci_alloc_consistent(he_dev->pci_dev,
+		CONFIG_TBRQ_SIZE * sizeof(struct he_tbrq), &he_dev->tbrq_phys);
+	if (he_dev->tbrq_base == NULL) {
+		hprintk("failed to allocate tbrq\n");
+		return -ENOMEM;
+	}
+	memset(he_dev->tbrq_base, 0, CONFIG_TBRQ_SIZE * sizeof(struct he_tbrq));
+
+	he_dev->tbrq_head = he_dev->tbrq_base;
+
+	he_writel(he_dev, he_dev->tbrq_phys, G0_TBRQ_B_T + (group * 16));
+	he_writel(he_dev, 0, G0_TBRQ_H + (group * 16));
+	he_writel(he_dev, CONFIG_TBRQ_SIZE - 1, G0_TBRQ_S + (group * 16));
+	he_writel(he_dev, CONFIG_TBRQ_THRESH, G0_TBRQ_THRESH + (group * 16));
+
+	return 0;
+}
+
+static int __init
+he_init_irq(struct he_dev *he_dev)
+{
+	int i;
+
+	/* 2.9.3.5  tail offset for each interrupt queue is located after the
+		    end of the interrupt queue */
+
+	he_dev->irq_base = pci_alloc_consistent(he_dev->pci_dev,
+			(CONFIG_IRQ_SIZE+1) * sizeof(struct he_irq), &he_dev->irq_phys);
+	if (he_dev->irq_base == NULL) {
+		hprintk("failed to allocate irq\n");
+		return -ENOMEM;
+	}
+	he_dev->irq_tailoffset = (unsigned *)
+					&he_dev->irq_base[CONFIG_IRQ_SIZE];
+	*he_dev->irq_tailoffset = 0;
+	he_dev->irq_head = he_dev->irq_base;
+	he_dev->irq_tail = he_dev->irq_base;
+
+	for (i = 0; i < CONFIG_IRQ_SIZE; ++i)
+		he_dev->irq_base[i].isw = ITYPE_INVALID;
+
+	he_writel(he_dev, he_dev->irq_phys, IRQ0_BASE);
+	he_writel(he_dev,
+		IRQ_SIZE(CONFIG_IRQ_SIZE) | IRQ_THRESH(CONFIG_IRQ_THRESH),
+								IRQ0_HEAD);
+	he_writel(he_dev, IRQ_INT_A | IRQ_TYPE_LINE, IRQ0_CNTL);
+	he_writel(he_dev, 0x0, IRQ0_DATA);
+
+	he_writel(he_dev, 0x0, IRQ1_BASE);
+	he_writel(he_dev, 0x0, IRQ1_HEAD);
+	he_writel(he_dev, 0x0, IRQ1_CNTL);
+	he_writel(he_dev, 0x0, IRQ1_DATA);
+
+	he_writel(he_dev, 0x0, IRQ2_BASE);
+	he_writel(he_dev, 0x0, IRQ2_HEAD);
+	he_writel(he_dev, 0x0, IRQ2_CNTL);
+	he_writel(he_dev, 0x0, IRQ2_DATA);
+
+	he_writel(he_dev, 0x0, IRQ3_BASE);
+	he_writel(he_dev, 0x0, IRQ3_HEAD);
+	he_writel(he_dev, 0x0, IRQ3_CNTL);
+	he_writel(he_dev, 0x0, IRQ3_DATA);
+
+	/* 2.9.3.2 interrupt queue mapping registers */
+
+	he_writel(he_dev, 0x0, GRP_10_MAP);
+	he_writel(he_dev, 0x0, GRP_32_MAP);
+	he_writel(he_dev, 0x0, GRP_54_MAP);
+	he_writel(he_dev, 0x0, GRP_76_MAP);
+
+	if (request_irq(he_dev->pci_dev->irq, he_irq_handler, SA_INTERRUPT|SA_SHIRQ, DEV_LABEL, he_dev)) {
+		hprintk("irq %d already in use\n", he_dev->pci_dev->irq);
+		return -EINVAL;
+	}   
+
+	he_dev->irq = he_dev->pci_dev->irq;
+
+	return 0;
+}
+
+static int __init
+he_start(struct atm_dev *dev)
+{
+	struct he_dev *he_dev;
+	struct pci_dev *pci_dev;
+
+	u16 command;
+	u32 gen_cntl_0, host_cntl, lb_swap;
+	u8 cache_size, timer;
+	
+	unsigned err;
+	unsigned int status, reg;
+	int i, group;
+
+	he_dev = HE_DEV(dev);
+	pci_dev = he_dev->pci_dev;
+
+	he_dev->membase = pci_dev->resource[0].start;
+	HPRINTK("membase = 0x%lx  irq = %d.\n", he_dev->membase, pci_dev->irq);
+
+	/*
+	 * pci bus controller initialization 
+	 */
+
+	/* 4.3 pci bus controller-specific initialization */
+	if (pci_read_config_dword(pci_dev, GEN_CNTL_0, &gen_cntl_0) != 0) {
+		hprintk("can't read GEN_CNTL_0\n");
+		return -EINVAL;
+	}
+	gen_cntl_0 |= (MRL_ENB | MRM_ENB | IGNORE_TIMEOUT);
+	if (pci_write_config_dword(pci_dev, GEN_CNTL_0, gen_cntl_0) != 0) {
+		hprintk("can't write GEN_CNTL_0.\n");
+		return -EINVAL;
+	}
+
+	if (pci_read_config_word(pci_dev, PCI_COMMAND, &command) != 0) {
+		hprintk("can't read PCI_COMMAND.\n");
+		return -EINVAL;
+	}
+
+	command |= (PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER | PCI_COMMAND_INVALIDATE);
+	if (pci_write_config_word(pci_dev, PCI_COMMAND, command) != 0) {
+		hprintk("can't enable memory.\n");
+		return -EINVAL;
+	}
+
+	if (pci_read_config_byte(pci_dev, PCI_CACHE_LINE_SIZE, &cache_size)) {
+		hprintk("can't read cache line size?\n");
+		return -EINVAL;
+	}
+
+	if (cache_size < 16) {
+		cache_size = 16;
+		if (pci_write_config_byte(pci_dev, PCI_CACHE_LINE_SIZE, cache_size))
+			hprintk("can't set cache line size to %d\n", cache_size);
+	}
+
+	if (pci_read_config_byte(pci_dev, PCI_LATENCY_TIMER, &timer)) {
+		hprintk("can't read latency timer?\n");
+		return -EINVAL;
+	}
+
+	/* from table 3.9
+	 *
+	 * LAT_TIMER = 1 + AVG_LAT + BURST_SIZE/BUS_SIZE
+	 * 
+	 * AVG_LAT: The average first data read/write latency [maximum 16 clock cycles]
+	 * BURST_SIZE: 1536 bytes (read) for 622, 768 bytes (read) for 155 [192 clock cycles]
+	 *
+	 */ 
+#define LAT_TIMER 209
+	if (timer < LAT_TIMER) {
+		HPRINTK("latency timer was %d, setting to %d\n", timer, LAT_TIMER);
+		timer = LAT_TIMER;
+		if (pci_write_config_byte(pci_dev, PCI_LATENCY_TIMER, timer))
+			hprintk("can't set latency timer to %d\n", timer);
+	}
+
+	if (!(he_dev->membase = (unsigned long) ioremap(he_dev->membase, HE_REGMAP_SIZE))) {
+		hprintk("can't set up page mapping\n");
+		return -EINVAL;
+	}
+
+	/* 4.4 card reset */
+	he_writel(he_dev, 0x0, RESET_CNTL);
+	he_writel(he_dev, 0xff, RESET_CNTL);
+
+	udelay(16*1000);	/* 16 ms */
+	status = he_readl(he_dev, RESET_CNTL);
+	if ((status & BOARD_RST_STATUS) == 0) {
+		hprintk("reset failed\n");
+		return -EINVAL;
+	}
+
+	/* 4.5 set bus width */
+	host_cntl = he_readl(he_dev, HOST_CNTL);
+	if (host_cntl & PCI_BUS_SIZE64)
+		gen_cntl_0 |= ENBL_64;
+	else
+		gen_cntl_0 &= ~ENBL_64;
+
+	if (disable64 == 1) {
+		hprintk("disabling 64-bit pci bus transfers\n");
+		gen_cntl_0 &= ~ENBL_64;
+	}
+
+	if (gen_cntl_0 & ENBL_64)
+		hprintk("64-bit transfers enabled\n");
+
+	pci_write_config_dword(pci_dev, GEN_CNTL_0, gen_cntl_0);
+
+	/* 4.7 read prom contents */
+	for (i = 0; i < PROD_ID_LEN; ++i)
+		he_dev->prod_id[i] = read_prom_byte(he_dev, PROD_ID + i);
+
+	he_dev->media = read_prom_byte(he_dev, MEDIA);
+
+	for (i = 0; i < 6; ++i)
+		dev->esi[i] = read_prom_byte(he_dev, MAC_ADDR + i);
+
+	hprintk("%s%s, %x:%x:%x:%x:%x:%x\n",
+				he_dev->prod_id,
+					he_dev->media & 0x40 ? "SM" : "MM",
+						dev->esi[0],
+						dev->esi[1],
+						dev->esi[2],
+						dev->esi[3],
+						dev->esi[4],
+						dev->esi[5]);
+	he_dev->atm_dev->link_rate = he_is622(he_dev) ?
+						ATM_OC12_PCR : ATM_OC3_PCR;
+
+	/* 4.6 set host endianess */
+	lb_swap = he_readl(he_dev, LB_SWAP);
+	if (he_is622(he_dev))
+		lb_swap &= ~XFER_SIZE;		/* 4 cells */
+	else
+		lb_swap |= XFER_SIZE;		/* 8 cells */
+#ifdef __BIG_ENDIAN
+	lb_swap |= DESC_WR_SWAP | INTR_SWAP | BIG_ENDIAN_HOST;
+#else
+	lb_swap &= ~(DESC_WR_SWAP | INTR_SWAP | BIG_ENDIAN_HOST |
+			DATA_WR_SWAP | DATA_RD_SWAP | DESC_RD_SWAP);
+#endif /* __BIG_ENDIAN */
+	he_writel(he_dev, lb_swap, LB_SWAP);
+
+	/* 4.8 sdram controller initialization */
+	he_writel(he_dev, he_is622(he_dev) ? LB_64_ENB : 0x0, SDRAM_CTL);
+
+	/* 4.9 initialize rnum value */
+	lb_swap |= SWAP_RNUM_MAX(0xf);
+	he_writel(he_dev, lb_swap, LB_SWAP);
+
+	/* 4.10 initialize the interrupt queues */
+	if ((err = he_init_irq(he_dev)) != 0)
+		return err;
+
+#ifdef USE_TASKLET
+	tasklet_init(&he_dev->tasklet, he_tasklet, (unsigned long) he_dev);
+#endif
+	spin_lock_init(&he_dev->global_lock);
+
+	/* 4.11 enable pci bus controller state machines */
+	host_cntl |= (OUTFF_ENB | CMDFF_ENB |
+				QUICK_RD_RETRY | QUICK_WR_RETRY | PERR_INT_ENB);
+	he_writel(he_dev, host_cntl, HOST_CNTL);
+
+	gen_cntl_0 |= INT_PROC_ENBL|INIT_ENB;
+	pci_write_config_dword(pci_dev, GEN_CNTL_0, gen_cntl_0);
+
+	/*
+	 * atm network controller initialization
+	 */
+
+	/* 5.1.1 generic configuration state */
+
+	/*
+	 *		local (cell) buffer memory map
+	 *                    
+	 *             HE155                          HE622
+	 *                                                      
+	 *        0 ____________1023 bytes  0 _______________________2047 bytes
+	 *         |            |            |                   |   |
+	 *         |  utility   |            |        rx0        |   |
+	 *        5|____________|         255|___________________| u |
+	 *        6|            |         256|                   | t |
+	 *         |            |            |                   | i |
+	 *         |    rx0     |     row    |        tx         | l |
+	 *         |            |            |                   | i |
+	 *         |            |         767|___________________| t |
+	 *      517|____________|         768|                   | y |
+	 * row  518|            |            |        rx1        |   |
+	 *         |            |        1023|___________________|___|
+	 *         |            |
+	 *         |    tx      |
+	 *         |            |
+	 *         |            |
+	 *     1535|____________|
+	 *     1536|            |
+	 *         |    rx1     |
+	 *     2047|____________|
+	 *
+	 */
+
+	/* total 4096 connections */
+	he_dev->vcibits = CONFIG_DEFAULT_VCIBITS;
+	he_dev->vpibits = CONFIG_DEFAULT_VPIBITS;
+
+	if (nvpibits != -1 && nvcibits != -1 && nvpibits+nvcibits != HE_MAXCIDBITS) {
+		hprintk("nvpibits + nvcibits != %d\n", HE_MAXCIDBITS);
+		return -ENODEV;
+	}
+
+	if (nvpibits != -1) {
+		he_dev->vpibits = nvpibits;
+		he_dev->vcibits = HE_MAXCIDBITS - nvpibits;
+	}
+
+	if (nvcibits != -1) {
+		he_dev->vcibits = nvcibits;
+		he_dev->vpibits = HE_MAXCIDBITS - nvcibits;
+	}
+
+
+	if (he_is622(he_dev)) {
+		he_dev->cells_per_row = 40;
+		he_dev->bytes_per_row = 2048;
+		he_dev->r0_numrows = 256;
+		he_dev->tx_numrows = 512;
+		he_dev->r1_numrows = 256;
+		he_dev->r0_startrow = 0;
+		he_dev->tx_startrow = 256;
+		he_dev->r1_startrow = 768;
+	} else {
+		he_dev->cells_per_row = 20;
+		he_dev->bytes_per_row = 1024;
+		he_dev->r0_numrows = 512;
+		he_dev->tx_numrows = 1018;
+		he_dev->r1_numrows = 512;
+		he_dev->r0_startrow = 6;
+		he_dev->tx_startrow = 518;
+		he_dev->r1_startrow = 1536;
+	}
+
+	he_dev->cells_per_lbuf = 4;
+	he_dev->buffer_limit = 4;
+	he_dev->r0_numbuffs = he_dev->r0_numrows *
+				he_dev->cells_per_row / he_dev->cells_per_lbuf;
+	if (he_dev->r0_numbuffs > 2560)
+		he_dev->r0_numbuffs = 2560;
+
+	he_dev->r1_numbuffs = he_dev->r1_numrows *
+				he_dev->cells_per_row / he_dev->cells_per_lbuf;
+	if (he_dev->r1_numbuffs > 2560)
+		he_dev->r1_numbuffs = 2560;
+
+	he_dev->tx_numbuffs = he_dev->tx_numrows *
+				he_dev->cells_per_row / he_dev->cells_per_lbuf;
+	if (he_dev->tx_numbuffs > 5120)
+		he_dev->tx_numbuffs = 5120;
+
+	/* 5.1.2 configure hardware dependent registers */
+
+	he_writel(he_dev, 
+		SLICE_X(0x2) | ARB_RNUM_MAX(0xf) | TH_PRTY(0x3) |
+		RH_PRTY(0x3) | TL_PRTY(0x2) | RL_PRTY(0x1) |
+		(he_is622(he_dev) ? BUS_MULTI(0x28) : BUS_MULTI(0x46)) |
+		(he_is622(he_dev) ? NET_PREF(0x50) : NET_PREF(0x8c)),
+								LBARB);
+
+	he_writel(he_dev, BANK_ON |
+		(he_is622(he_dev) ? (REF_RATE(0x384) | WIDE_DATA) : REF_RATE(0x150)),
+								SDRAMCON);
+
+	he_writel(he_dev,
+		(he_is622(he_dev) ? RM_BANK_WAIT(1) : RM_BANK_WAIT(0)) |
+						RM_RW_WAIT(1), RCMCONFIG);
+	he_writel(he_dev,
+		(he_is622(he_dev) ? TM_BANK_WAIT(2) : TM_BANK_WAIT(1)) |
+						TM_RW_WAIT(1), TCMCONFIG);
+
+	he_writel(he_dev, he_dev->cells_per_lbuf * ATM_CELL_PAYLOAD, LB_CONFIG);
+
+	he_writel(he_dev, 
+		(he_is622(he_dev) ? UT_RD_DELAY(8) : UT_RD_DELAY(0)) |
+		(he_is622(he_dev) ? RC_UT_MODE(0) : RC_UT_MODE(1)) |
+		RX_VALVP(he_dev->vpibits) |
+		RX_VALVC(he_dev->vcibits),			 RC_CONFIG);
+
+	he_writel(he_dev, DRF_THRESH(0x20) |
+		(he_is622(he_dev) ? TX_UT_MODE(0) : TX_UT_MODE(1)) |
+		TX_VCI_MASK(he_dev->vcibits) |
+		LBFREE_CNT(he_dev->tx_numbuffs), 		TX_CONFIG);
+
+	he_writel(he_dev, 0x0, TXAAL5_PROTO);
+
+	he_writel(he_dev, PHY_INT_ENB |
+		(he_is622(he_dev) ? PTMR_PRE(67 - 1) : PTMR_PRE(50 - 1)),
+								RH_CONFIG);
+
+	/* 5.1.3 initialize connection memory */
+
+	for (i = 0; i < TCM_MEM_SIZE; ++i)
+		he_writel_tcm(he_dev, 0, i);
+
+	for (i = 0; i < RCM_MEM_SIZE; ++i)
+		he_writel_rcm(he_dev, 0, i);
+
+	/*
+	 *	transmit connection memory map
+	 *
+	 *                  tx memory
+	 *          0x0 ___________________
+	 *             |                   |
+	 *             |                   |
+	 *             |       TSRa        |
+	 *             |                   |
+	 *             |                   |
+	 *       0x8000|___________________|
+	 *             |                   |
+	 *             |       TSRb        |
+	 *       0xc000|___________________|
+	 *             |                   |
+	 *             |       TSRc        |
+	 *       0xe000|___________________|
+	 *             |       TSRd        |
+	 *       0xf000|___________________|
+	 *             |       tmABR       |
+	 *      0x10000|___________________|
+	 *             |                   |
+	 *             |       tmTPD       |
+	 *             |___________________|
+	 *             |                   |
+	 *                      ....
+	 *      0x1ffff|___________________|
+	 *
+	 *
+	 */
+
+	he_writel(he_dev, CONFIG_TSRB, TSRB_BA);
+	he_writel(he_dev, CONFIG_TSRC, TSRC_BA);
+	he_writel(he_dev, CONFIG_TSRD, TSRD_BA);
+	he_writel(he_dev, CONFIG_TMABR, TMABR_BA);
+	he_writel(he_dev, CONFIG_TPDBA, TPD_BA);
+
+
+	/*
+	 *	receive connection memory map
+	 *
+	 *          0x0 ___________________
+	 *             |                   |
+	 *             |                   |
+	 *             |       RSRa        |
+	 *             |                   |
+	 *             |                   |
+	 *       0x8000|___________________|
+	 *             |                   |
+	 *             |             rx0/1 |
+	 *             |       LBM         |   link lists of local
+	 *             |             tx    |   buffer memory 
+	 *             |                   |
+	 *       0xd000|___________________|
+	 *             |                   |
+	 *             |      rmABR        |
+	 *       0xe000|___________________|
+	 *             |                   |
+	 *             |       RSRb        |
+	 *             |___________________|
+	 *             |                   |
+	 *                      ....
+	 *       0xffff|___________________|
+	 */
+
+	he_writel(he_dev, 0x08000, RCMLBM_BA);
+	he_writel(he_dev, 0x0e000, RCMRSRB_BA);
+	he_writel(he_dev, 0x0d800, RCMABR_BA);
+
+	/* 5.1.4 initialize local buffer free pools linked lists */
+
+	he_init_rx_lbfp0(he_dev);
+	he_init_rx_lbfp1(he_dev);
+
+	he_writel(he_dev, 0x0, RLBC_H);
+	he_writel(he_dev, 0x0, RLBC_T);
+	he_writel(he_dev, 0x0, RLBC_H2);
+
+	he_writel(he_dev, 512, RXTHRSH);	/* 10% of r0+r1 buffers */
+	he_writel(he_dev, 256, LITHRSH); 	/* 5% of r0+r1 buffers */
+
+	he_init_tx_lbfp(he_dev);
+
+	he_writel(he_dev, he_is622(he_dev) ? 0x104780 : 0x800, UBUFF_BA);
+
+	/* 5.1.5 initialize intermediate receive queues */
+
+	if (he_is622(he_dev)) {
+		he_writel(he_dev, 0x000f, G0_INMQ_S);
+		he_writel(he_dev, 0x200f, G0_INMQ_L);
+
+		he_writel(he_dev, 0x001f, G1_INMQ_S);
+		he_writel(he_dev, 0x201f, G1_INMQ_L);
+
+		he_writel(he_dev, 0x002f, G2_INMQ_S);
+		he_writel(he_dev, 0x202f, G2_INMQ_L);
+
+		he_writel(he_dev, 0x003f, G3_INMQ_S);
+		he_writel(he_dev, 0x203f, G3_INMQ_L);
+
+		he_writel(he_dev, 0x004f, G4_INMQ_S);
+		he_writel(he_dev, 0x204f, G4_INMQ_L);
+
+		he_writel(he_dev, 0x005f, G5_INMQ_S);
+		he_writel(he_dev, 0x205f, G5_INMQ_L);
+
+		he_writel(he_dev, 0x006f, G6_INMQ_S);
+		he_writel(he_dev, 0x206f, G6_INMQ_L);
+
+		he_writel(he_dev, 0x007f, G7_INMQ_S);
+		he_writel(he_dev, 0x207f, G7_INMQ_L);
+	} else {
+		he_writel(he_dev, 0x0000, G0_INMQ_S);
+		he_writel(he_dev, 0x0008, G0_INMQ_L);
+
+		he_writel(he_dev, 0x0001, G1_INMQ_S);
+		he_writel(he_dev, 0x0009, G1_INMQ_L);
+
+		he_writel(he_dev, 0x0002, G2_INMQ_S);
+		he_writel(he_dev, 0x000a, G2_INMQ_L);
+
+		he_writel(he_dev, 0x0003, G3_INMQ_S);
+		he_writel(he_dev, 0x000b, G3_INMQ_L);
+
+		he_writel(he_dev, 0x0004, G4_INMQ_S);
+		he_writel(he_dev, 0x000c, G4_INMQ_L);
+
+		he_writel(he_dev, 0x0005, G5_INMQ_S);
+		he_writel(he_dev, 0x000d, G5_INMQ_L);
+
+		he_writel(he_dev, 0x0006, G6_INMQ_S);
+		he_writel(he_dev, 0x000e, G6_INMQ_L);
+
+		he_writel(he_dev, 0x0007, G7_INMQ_S);
+		he_writel(he_dev, 0x000f, G7_INMQ_L);
+	}
+
+	/* 5.1.6 application tunable parameters */
+
+	he_writel(he_dev, 0x0, MCC);
+	he_writel(he_dev, 0x0, OEC);
+	he_writel(he_dev, 0x0, DCC);
+	he_writel(he_dev, 0x0, CEC);
+	
+	/* 5.1.7 cs block initialization */
+
+	he_init_cs_block(he_dev);
+
+	/* 5.1.8 cs block connection memory initialization */
+	
+	if (he_init_cs_block_rcm(he_dev) < 0)
+		return -ENOMEM;
+
+	/* 5.1.10 initialize host structures */
+
+	he_init_tpdrq(he_dev);
+
+#ifdef USE_TPD_POOL
+	he_dev->tpd_pool = pci_pool_create("tpd", he_dev->pci_dev,
+		sizeof(struct he_tpd), TPD_ALIGNMENT, 0);
+	if (he_dev->tpd_pool == NULL) {
+		hprintk("unable to create tpd pci_pool\n");
+		return -ENOMEM;         
+	}
+
+	INIT_LIST_HEAD(&he_dev->outstanding_tpds);
+#else
+	he_dev->tpd_base = (void *) pci_alloc_consistent(he_dev->pci_dev,
+			CONFIG_NUMTPDS * sizeof(struct he_tpd), &he_dev->tpd_base_phys);
+	if (!he_dev->tpd_base)
+		return -ENOMEM;
+
+	for (i = 0; i < CONFIG_NUMTPDS; ++i) {
+		he_dev->tpd_base[i].status = (i << TPD_ADDR_SHIFT);
+		he_dev->tpd_base[i].inuse = 0;
+	}
+		
+	he_dev->tpd_head = he_dev->tpd_base;
+	he_dev->tpd_end = &he_dev->tpd_base[CONFIG_NUMTPDS - 1];
+#endif
+
+	if (he_init_group(he_dev, 0) != 0)
+		return -ENOMEM;
+
+	for (group = 1; group < HE_NUM_GROUPS; ++group) {
+		he_writel(he_dev, 0x0, G0_RBPS_S + (group * 32));
+		he_writel(he_dev, 0x0, G0_RBPS_T + (group * 32));
+		he_writel(he_dev, 0x0, G0_RBPS_QI + (group * 32));
+		he_writel(he_dev, RBP_THRESH(0x1) | RBP_QSIZE(0x0),
+						G0_RBPS_BS + (group * 32));
+
+		he_writel(he_dev, 0x0, G0_RBPL_S + (group * 32));
+		he_writel(he_dev, 0x0, G0_RBPL_T + (group * 32));
+		he_writel(he_dev, RBP_THRESH(0x1) | RBP_QSIZE(0x0),
+						G0_RBPL_QI + (group * 32));
+		he_writel(he_dev, 0x0, G0_RBPL_BS + (group * 32));
+
+		he_writel(he_dev, 0x0, G0_RBRQ_ST + (group * 16));
+		he_writel(he_dev, 0x0, G0_RBRQ_H + (group * 16));
+		he_writel(he_dev, RBRQ_THRESH(0x1) | RBRQ_SIZE(0x0),
+						G0_RBRQ_Q + (group * 16));
+		he_writel(he_dev, 0x0, G0_RBRQ_I + (group * 16));
+
+		he_writel(he_dev, 0x0, G0_TBRQ_B_T + (group * 16));
+		he_writel(he_dev, 0x0, G0_TBRQ_H + (group * 16));
+		he_writel(he_dev, TBRQ_THRESH(0x1),
+						G0_TBRQ_THRESH + (group * 16));
+		he_writel(he_dev, 0x0, G0_TBRQ_S + (group * 16));
+	}
+
+	/* host status page */
+
+	he_dev->hsp = pci_alloc_consistent(he_dev->pci_dev,
+				sizeof(struct he_hsp), &he_dev->hsp_phys);
+	if (he_dev->hsp == NULL) {
+		hprintk("failed to allocate host status page\n");
+		return -ENOMEM;
+	}
+	memset(he_dev->hsp, 0, sizeof(struct he_hsp));
+	he_writel(he_dev, he_dev->hsp_phys, HSP_BA);
+
+	/* initialize framer */
+
+#ifdef CONFIG_ATM_HE_USE_SUNI
+	suni_init(he_dev->atm_dev);
+	if (he_dev->atm_dev->phy && he_dev->atm_dev->phy->start)
+		he_dev->atm_dev->phy->start(he_dev->atm_dev);
+#endif /* CONFIG_ATM_HE_USE_SUNI */
+
+	if (sdh) {
+		/* this really should be in suni.c but for now... */
+
+		int val;
+
+		val = he_phy_get(he_dev->atm_dev, SUNI_TPOP_APM);
+		val = (val & ~SUNI_TPOP_APM_S) | ( 0x2 << SUNI_TPOP_APM_S_SHIFT);
+		he_phy_put(he_dev->atm_dev, val, SUNI_TPOP_APM);
+	}
+
+	/* 5.1.12 enable transmit and receive */
+
+	reg = he_readl_mbox(he_dev, CS_ERCTL0);
+	reg |= TX_ENABLE|ER_ENABLE;
+	he_writel_mbox(he_dev, reg, CS_ERCTL0);
+
+	reg = he_readl(he_dev, RC_CONFIG);
+	reg |= RX_ENABLE;
+	he_writel(he_dev, reg, RC_CONFIG);
+
+	for (i = 0; i < HE_NUM_CS_STPER; ++i) {
+		he_dev->cs_stper[i].inuse = 0;
+		he_dev->cs_stper[i].pcr = -1;
+	}
+	he_dev->total_bw = 0;
+
+
+	/* atm linux initialization */
+
+	he_dev->atm_dev->ci_range.vpi_bits = he_dev->vpibits;
+	he_dev->atm_dev->ci_range.vci_bits = he_dev->vcibits;
+
+	he_dev->irq_peak = 0;
+	he_dev->rbrq_peak = 0;
+	he_dev->rbpl_peak = 0;
+	he_dev->tbrq_peak = 0;
+
+	HPRINTK("hell bent for leather!\n");
+
+	return 0;
+}
+
+static void
+he_stop(struct he_dev *he_dev)
+{
+	u16 command;
+	u32 gen_cntl_0, reg;
+	struct pci_dev *pci_dev;
+
+	pci_dev = he_dev->pci_dev;
+
+	/* disable interrupts */
+
+	if (he_dev->membase) {
+		pci_read_config_dword(pci_dev, GEN_CNTL_0, &gen_cntl_0);
+		gen_cntl_0 &= ~(INT_PROC_ENBL | INIT_ENB);
+		pci_write_config_dword(pci_dev, GEN_CNTL_0, gen_cntl_0);
+
+#ifdef USE_TASKLET
+		tasklet_disable(&he_dev->tasklet);
+#endif
+
+		/* disable recv and transmit */
+
+		reg = he_readl_mbox(he_dev, CS_ERCTL0);
+		reg &= ~(TX_ENABLE|ER_ENABLE);
+		he_writel_mbox(he_dev, reg, CS_ERCTL0);
+
+		reg = he_readl(he_dev, RC_CONFIG);
+		reg &= ~(RX_ENABLE);
+		he_writel(he_dev, reg, RC_CONFIG);
+	}
+
+#ifdef CONFIG_ATM_HE_USE_SUNI
+	if (he_dev->atm_dev->phy && he_dev->atm_dev->phy->stop)
+		he_dev->atm_dev->phy->stop(he_dev->atm_dev);
+#endif /* CONFIG_ATM_HE_USE_SUNI */
+
+	if (he_dev->irq)
+		free_irq(he_dev->irq, he_dev);
+
+	if (he_dev->irq_base)
+		pci_free_consistent(he_dev->pci_dev, (CONFIG_IRQ_SIZE+1)
+			* sizeof(struct he_irq), he_dev->irq_base, he_dev->irq_phys);
+
+	if (he_dev->hsp)
+		pci_free_consistent(he_dev->pci_dev, sizeof(struct he_hsp),
+						he_dev->hsp, he_dev->hsp_phys);
+
+	if (he_dev->rbpl_base) {
+#ifdef USE_RBPL_POOL
+		for (i = 0; i < CONFIG_RBPL_SIZE; ++i) {
+			void *cpuaddr = he_dev->rbpl_virt[i].virt;
+			dma_addr_t dma_handle = he_dev->rbpl_base[i].phys;
+
+			pci_pool_free(he_dev->rbpl_pool, cpuaddr, dma_handle);
+		}
+#else
+		pci_free_consistent(he_dev->pci_dev, CONFIG_RBPL_SIZE
+			* CONFIG_RBPL_BUFSIZE, he_dev->rbpl_pages, he_dev->rbpl_pages_phys);
+#endif
+		pci_free_consistent(he_dev->pci_dev, CONFIG_RBPL_SIZE
+			* sizeof(struct he_rbp), he_dev->rbpl_base, he_dev->rbpl_phys);
+	}
+
+#ifdef USE_RBPL_POOL
+	if (he_dev->rbpl_pool)
+		pci_pool_destroy(he_dev->rbpl_pool);
+#endif
+
+#ifdef USE_RBPS
+	if (he_dev->rbps_base) {
+#ifdef USE_RBPS_POOL
+		for (i = 0; i < CONFIG_RBPS_SIZE; ++i) {
+			void *cpuaddr = he_dev->rbps_virt[i].virt;
+			dma_addr_t dma_handle = he_dev->rbps_base[i].phys;
+
+			pci_pool_free(he_dev->rbps_pool, cpuaddr, dma_handle);
+		}
+#else
+		pci_free_consistent(he_dev->pci_dev, CONFIG_RBPS_SIZE
+			* CONFIG_RBPS_BUFSIZE, he_dev->rbps_pages, he_dev->rbps_pages_phys);
+#endif
+		pci_free_consistent(he_dev->pci_dev, CONFIG_RBPS_SIZE
+			* sizeof(struct he_rbp), he_dev->rbps_base, he_dev->rbps_phys);
+	}
+
+#ifdef USE_RBPS_POOL
+	if (he_dev->rbps_pool)
+		pci_pool_destroy(he_dev->rbps_pool);
+#endif
+
+#endif /* USE_RBPS */
+
+	if (he_dev->rbrq_base)
+		pci_free_consistent(he_dev->pci_dev, CONFIG_RBRQ_SIZE * sizeof(struct he_rbrq),
+							he_dev->rbrq_base, he_dev->rbrq_phys);
+
+	if (he_dev->tbrq_base)
+		pci_free_consistent(he_dev->pci_dev, CONFIG_TBRQ_SIZE * sizeof(struct he_tbrq),
+							he_dev->tbrq_base, he_dev->tbrq_phys);
+
+	if (he_dev->tpdrq_base)
+		pci_free_consistent(he_dev->pci_dev, CONFIG_TBRQ_SIZE * sizeof(struct he_tbrq),
+							he_dev->tpdrq_base, he_dev->tpdrq_phys);
+
+#ifdef USE_TPD_POOL
+	if (he_dev->tpd_pool)
+		pci_pool_destroy(he_dev->tpd_pool);
+#else
+	if (he_dev->tpd_base)
+		pci_free_consistent(he_dev->pci_dev, CONFIG_NUMTPDS * sizeof(struct he_tpd),
+							he_dev->tpd_base, he_dev->tpd_base_phys);
+#endif
+
+	if (he_dev->pci_dev) {
+		pci_read_config_word(he_dev->pci_dev, PCI_COMMAND, &command);
+		command &= ~(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
+		pci_write_config_word(he_dev->pci_dev, PCI_COMMAND, command);
+	}
+	
+	if (he_dev->membase)
+		iounmap((void *) he_dev->membase);
+}
+
+static struct he_tpd *
+__alloc_tpd(struct he_dev *he_dev)
+{
+#ifdef USE_TPD_POOL
+	struct he_tpd *tpd;
+	dma_addr_t dma_handle; 
+
+	tpd = pci_pool_alloc(he_dev->tpd_pool, SLAB_ATOMIC|SLAB_DMA, &dma_handle);              
+	if (tpd == NULL)
+		return NULL;
+			
+	tpd->status = TPD_ADDR(dma_handle);
+	tpd->reserved = 0; 
+	tpd->iovec[0].addr = 0; tpd->iovec[0].len = 0;
+	tpd->iovec[1].addr = 0; tpd->iovec[1].len = 0;
+	tpd->iovec[2].addr = 0; tpd->iovec[2].len = 0;
+
+	return tpd;
+#else
+	int i;
+
+	for (i = 0; i < CONFIG_NUMTPDS; ++i) {
+		++he_dev->tpd_head;
+		if (he_dev->tpd_head > he_dev->tpd_end) {
+			he_dev->tpd_head = he_dev->tpd_base;
+		}
+
+		if (!he_dev->tpd_head->inuse) {
+			he_dev->tpd_head->inuse = 1;
+			he_dev->tpd_head->status &= TPD_MASK;
+			he_dev->tpd_head->iovec[0].addr = 0; he_dev->tpd_head->iovec[0].len = 0;
+			he_dev->tpd_head->iovec[1].addr = 0; he_dev->tpd_head->iovec[1].len = 0;
+			he_dev->tpd_head->iovec[2].addr = 0; he_dev->tpd_head->iovec[2].len = 0;
+			return he_dev->tpd_head;
+		}
+	}
+	hprintk("out of tpds -- increase CONFIG_NUMTPDS (%d)\n", CONFIG_NUMTPDS);
+	return NULL;
+#endif
+}
+
+#define AAL5_LEN(buf,len) 						\
+			((((unsigned char *)(buf))[(len)-6] << 8) |	\
+				(((unsigned char *)(buf))[(len)-5]))
+
+/* 2.10.1.2 receive
+ *
+ * aal5 packets can optionally return the tcp checksum in the lower
+ * 16 bits of the crc (RSR0_TCP_CKSUM)
+ */
+
+#define TCP_CKSUM(buf,len) 						\
+			((((unsigned char *)(buf))[(len)-2] << 8) |	\
+				(((unsigned char *)(buf))[(len-1)]))
+
+static int
+he_service_rbrq(struct he_dev *he_dev, int group)
+{
+	struct he_rbrq *rbrq_tail = (struct he_rbrq *)
+				((unsigned long)he_dev->rbrq_base |
+					he_dev->hsp->group[group].rbrq_tail);
+	struct he_rbp *rbp = NULL;
+	unsigned cid, lastcid = -1;
+	unsigned buf_len = 0;
+	struct sk_buff *skb;
+	struct atm_vcc *vcc = NULL;
+	struct he_vcc *he_vcc;
+	struct he_iovec *iov;
+	int pdus_assembled = 0;
+	int updated = 0;
+
+	while (he_dev->rbrq_head != rbrq_tail) {
+		++updated;
+
+		HPRINTK("%p rbrq%d 0x%x len=%d cid=0x%x %s%s%s%s%s%s\n",
+			he_dev->rbrq_head, group,
+			RBRQ_ADDR(he_dev->rbrq_head),
+			RBRQ_BUFLEN(he_dev->rbrq_head),
+			RBRQ_CID(he_dev->rbrq_head),
+			RBRQ_CRC_ERR(he_dev->rbrq_head) ? " CRC_ERR" : "",
+			RBRQ_LEN_ERR(he_dev->rbrq_head) ? " LEN_ERR" : "",
+			RBRQ_END_PDU(he_dev->rbrq_head) ? " END_PDU" : "",
+			RBRQ_AAL5_PROT(he_dev->rbrq_head) ? " AAL5_PROT" : "",
+			RBRQ_CON_CLOSED(he_dev->rbrq_head) ? " CON_CLOSED" : "",
+			RBRQ_HBUF_ERR(he_dev->rbrq_head) ? " HBUF_ERR" : "");
+
+#ifdef USE_RBPS
+		if (RBRQ_ADDR(he_dev->rbrq_head) & RBP_SMALLBUF)
+			rbp = &he_dev->rbps_base[RBP_INDEX(RBRQ_ADDR(he_dev->rbrq_head))];
+		else
+#endif
+			rbp = &he_dev->rbpl_base[RBP_INDEX(RBRQ_ADDR(he_dev->rbrq_head))];
+		
+		buf_len = RBRQ_BUFLEN(he_dev->rbrq_head) * 4;
+		cid = RBRQ_CID(he_dev->rbrq_head);
+
+		if (cid != lastcid)
+			vcc = he_find_vcc(he_dev, cid);
+		lastcid = cid;
+
+		if (vcc == NULL) {
+			hprintk("vcc == NULL  (cid 0x%x)\n", cid);
+			if (!RBRQ_HBUF_ERR(he_dev->rbrq_head))
+					rbp->status &= ~RBP_LOANED;
+					
+			goto next_rbrq_entry;
+		}
+
+		he_vcc = HE_VCC(vcc);
+		if (he_vcc == NULL) {
+			hprintk("he_vcc == NULL  (cid 0x%x)\n", cid);
+			if (!RBRQ_HBUF_ERR(he_dev->rbrq_head))
+					rbp->status &= ~RBP_LOANED;
+			goto next_rbrq_entry;
+		}
+
+		if (RBRQ_HBUF_ERR(he_dev->rbrq_head)) {
+			hprintk("HBUF_ERR!  (cid 0x%x)\n", cid);
+				atomic_inc(&vcc->stats->rx_drop);
+			goto return_host_buffers;
+		}
+
+		he_vcc->iov_tail->iov_base = RBRQ_ADDR(he_dev->rbrq_head);
+		he_vcc->iov_tail->iov_len = buf_len;
+		he_vcc->pdu_len += buf_len;
+		++he_vcc->iov_tail;
+
+		if (RBRQ_CON_CLOSED(he_dev->rbrq_head)) {
+			lastcid = -1;
+			HPRINTK("wake_up rx_waitq  (cid 0x%x)\n", cid);
+			wake_up(&he_vcc->rx_waitq);
+			goto return_host_buffers;
+		}
+
+#ifdef notdef
+		if ((he_vcc->iov_tail - he_vcc->iov_head) > HE_MAXIOV) {
+			hprintk("iovec full!  cid 0x%x\n", cid);
+			goto return_host_buffers;
+		}
+#endif
+		if (!RBRQ_END_PDU(he_dev->rbrq_head))
+			goto next_rbrq_entry;
+
+		if (RBRQ_LEN_ERR(he_dev->rbrq_head)
+				|| RBRQ_CRC_ERR(he_dev->rbrq_head)) {
+			HPRINTK("%s%s (%d.%d)\n",
+				RBRQ_CRC_ERR(he_dev->rbrq_head)
+							? "CRC_ERR " : "",
+				RBRQ_LEN_ERR(he_dev->rbrq_head)
+							? "LEN_ERR" : "",
+							vcc->vpi, vcc->vci);
+			atomic_inc(&vcc->stats->rx_err);
+			goto return_host_buffers;
+		}
+
+		skb = atm_alloc_charge(vcc, he_vcc->pdu_len + rx_skb_reserve,
+							GFP_ATOMIC);
+		if (!skb) {
+			HPRINTK("charge failed (%d.%d)\n", vcc->vpi, vcc->vci);
+			goto return_host_buffers;
+		}
+
+		if (rx_skb_reserve > 0)
+			skb_reserve(skb, rx_skb_reserve);
+
+		do_gettimeofday(&skb->stamp);
+
+		for (iov = he_vcc->iov_head;
+				iov < he_vcc->iov_tail; ++iov) {
+#ifdef USE_RBPS
+			if (iov->iov_base & RBP_SMALLBUF)
+				memcpy(skb_put(skb, iov->iov_len),
+					he_dev->rbps_virt[RBP_INDEX(iov->iov_base)].virt, iov->iov_len);
+			else
+#endif
+				memcpy(skb_put(skb, iov->iov_len),
+					he_dev->rbpl_virt[RBP_INDEX(iov->iov_base)].virt, iov->iov_len);
+		}
+
+		switch (vcc->qos.aal) {
+			case ATM_AAL0:
+				/* 2.10.1.5 raw cell receive */
+				skb->len = ATM_AAL0_SDU;
+				skb->tail = skb->data + skb->len;
+				break;
+			case ATM_AAL5:
+				/* 2.10.1.2 aal5 receive */
+
+				skb->len = AAL5_LEN(skb->data, he_vcc->pdu_len);
+				skb->tail = skb->data + skb->len;
+#ifdef USE_CHECKSUM_HW
+				if (vcc->vpi == 0 && vcc->vci >= ATM_NOT_RSV_VCI) {
+					skb->ip_summed = CHECKSUM_HW;
+					skb->csum = TCP_CKSUM(skb->data,
+							he_vcc->pdu_len);
+				}
+#endif
+				break;
+		}
+
+#ifdef should_never_happen
+		if (skb->len > vcc->qos.rxtp.max_sdu)
+			hprintk("pdu_len (%d) > vcc->qos.rxtp.max_sdu (%d)!  cid 0x%x\n", skb->len, vcc->qos.rxtp.max_sdu, cid);
+#endif
+
+#ifdef notdef
+		ATM_SKB(skb)->vcc = vcc;
+#endif
+		vcc->push(vcc, skb);
+
+		atomic_inc(&vcc->stats->rx);
+
+return_host_buffers:
+		++pdus_assembled;
+
+		for (iov = he_vcc->iov_head;
+				iov < he_vcc->iov_tail; ++iov) {
+#ifdef USE_RBPS
+			if (iov->iov_base & RBP_SMALLBUF)
+				rbp = &he_dev->rbps_base[RBP_INDEX(iov->iov_base)];
+			else
+#endif
+				rbp = &he_dev->rbpl_base[RBP_INDEX(iov->iov_base)];
+
+			rbp->status &= ~RBP_LOANED;
+		}
+
+		he_vcc->iov_tail = he_vcc->iov_head;
+		he_vcc->pdu_len = 0;
+
+next_rbrq_entry:
+		he_dev->rbrq_head = (struct he_rbrq *)
+				((unsigned long) he_dev->rbrq_base |
+					RBRQ_MASK(++he_dev->rbrq_head));
+
+	}
+
+	if (updated) {
+		if (updated > he_dev->rbrq_peak)
+			he_dev->rbrq_peak = updated;
+
+		he_writel(he_dev, RBRQ_MASK(he_dev->rbrq_head),
+						G0_RBRQ_H + (group * 16));
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl(he_dev, G0_RBRQ_H + (group * 16));
+#endif
+	}
+
+	return pdus_assembled;
+}
+
+static void
+he_service_tbrq(struct he_dev *he_dev, int group)
+{
+	struct he_tbrq *tbrq_tail = (struct he_tbrq *)
+				((unsigned long)he_dev->tbrq_base |
+					he_dev->hsp->group[group].tbrq_tail);
+	struct he_tpd *tpd;
+	int slot, updated = 0;
+#ifdef USE_TPD_POOL
+	struct list_head *p;
+#endif
+
+	/* 2.1.6 transmit buffer return queue */
+
+	while (he_dev->tbrq_head != tbrq_tail) {
+		++updated;
+
+		HPRINTK("tbrq%d 0x%x%s%s\n",
+			group,
+			TBRQ_TPD(he_dev->tbrq_head), 
+			TBRQ_EOS(he_dev->tbrq_head) ? " EOS" : "",
+			TBRQ_MULTIPLE(he_dev->tbrq_head) ? " MULTIPLE" : "");
+#ifdef USE_TPD_POOL
+		tpd = NULL;
+		p = &he_dev->outstanding_tpds;
+		while ((p = p->next) != &he_dev->outstanding_tpds) {
+			struct he_tpd *__tpd = list_entry(p, struct he_tpd, entry);
+			if (TPD_ADDR(__tpd->status) == TBRQ_TPD(he_dev->tbrq_head)) {
+				tpd = __tpd;
+				list_del(&__tpd->entry);
+				break;
+			}
+		}
+
+		if (tpd == NULL) {
+			hprintk("unable to locate tpd for dma buffer %x\n",
+						TBRQ_TPD(he_dev->tbrq_head));
+			goto next_tbrq_entry;
+		}
+#else
+		tpd = &he_dev->tpd_base[ TPD_INDEX(TBRQ_TPD(he_dev->tbrq_head)) ];
+#endif
+
+		if (TBRQ_EOS(he_dev->tbrq_head)) {
+			HPRINTK("wake_up(tx_waitq) cid 0x%x\n",
+				he_mkcid(he_dev, tpd->vcc->vpi, tpd->vcc->vci));
+			if (tpd->vcc)
+				wake_up(&HE_VCC(tpd->vcc)->tx_waitq);
+
+			goto next_tbrq_entry;
+		}
+
+		for (slot = 0; slot < TPD_MAXIOV; ++slot) {
+			if (tpd->iovec[slot].addr)
+				pci_unmap_single(he_dev->pci_dev,
+					tpd->iovec[slot].addr,
+					tpd->iovec[slot].len & TPD_LEN_MASK,
+							PCI_DMA_TODEVICE);
+			if (tpd->iovec[slot].len & TPD_LST)
+				break;
+				
+		}
+
+		if (tpd->skb) {	/* && !TBRQ_MULTIPLE(he_dev->tbrq_head) */
+			if (tpd->vcc && tpd->vcc->pop)
+				tpd->vcc->pop(tpd->vcc, tpd->skb);
+			else
+				dev_kfree_skb_any(tpd->skb);
+		}
+
+next_tbrq_entry:
+#ifdef USE_TPD_POOL
+		if (tpd)
+			pci_pool_free(he_dev->tpd_pool, tpd, TPD_ADDR(tpd->status));
+#else
+		tpd->inuse = 0;
+#endif
+		he_dev->tbrq_head = (struct he_tbrq *)
+				((unsigned long) he_dev->tbrq_base |
+					TBRQ_MASK(++he_dev->tbrq_head));
+	}
+
+	if (updated) {
+		if (updated > he_dev->tbrq_peak)
+			he_dev->tbrq_peak = updated;
+
+		he_writel(he_dev, TBRQ_MASK(he_dev->tbrq_head),
+						G0_TBRQ_H + (group * 16));
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl(he_dev, G0_TBRQ_H + (group * 16));
+#endif
+	}
+}
+
+
+static void
+he_service_rbpl(struct he_dev *he_dev, int group)
+{
+	struct he_rbp *newtail;
+	struct he_rbp *rbpl_head;
+	int moved = 0;
+
+	rbpl_head = (struct he_rbp *) ((unsigned long)he_dev->rbpl_base |
+					RBPL_MASK(he_readl(he_dev, G0_RBPL_S)));
+
+	for (;;) {
+		newtail = (struct he_rbp *) ((unsigned long)he_dev->rbpl_base |
+						RBPL_MASK(he_dev->rbpl_tail+1));
+
+		/* table 3.42 -- rbpl_tail should never be set to rbpl_head */
+		if ((newtail == rbpl_head) || (newtail->status & RBP_LOANED))
+			break;
+
+		newtail->status |= RBP_LOANED;
+		he_dev->rbpl_tail = newtail;
+		++moved;
+	} 
+
+	if (moved) {
+		he_writel(he_dev, RBPL_MASK(he_dev->rbpl_tail), G0_RBPL_T);
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl(he_dev, G0_RBPL_T);
+#endif
+	}
+}
+
+#ifdef USE_RBPS
+static void
+he_service_rbps(struct he_dev *he_dev, int group)
+{
+	struct he_rbp *newtail;
+	struct he_rbp *rbps_head;
+	int moved = 0;
+
+	rbps_head = (struct he_rbp *) ((unsigned long)he_dev->rbps_base |
+					RBPS_MASK(he_readl(he_dev, G0_RBPS_S)));
+
+	for (;;) {
+		newtail = (struct he_rbp *) ((unsigned long)he_dev->rbps_base |
+						RBPS_MASK(he_dev->rbps_tail+1));
+
+		/* table 3.42 -- rbps_tail should never be set to rbps_head */
+		if ((newtail == rbps_head) || (newtail->status & RBP_LOANED))
+			break;
+
+		newtail->status |= RBP_LOANED;
+		he_dev->rbps_tail = newtail;
+		++moved;
+	} 
+
+	if (moved) {
+		he_writel(he_dev, RBPS_MASK(he_dev->rbps_tail), G0_RBPS_T);
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl(he_dev, G0_RBPS_T);
+#endif
+	}
+}
+#endif /* USE_RBPS */
+
+static void
+he_tasklet(unsigned long data)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = (struct he_dev *) data;
+	int group, type;
+	int updated = 0;
+
+	HPRINTK("tasklet (0x%lx)\n", data);
+#ifdef USE_TASKLET
+	spin_lock_irqsave(&he_dev->global_lock, flags);
+#endif
+
+	while (he_dev->irq_head != he_dev->irq_tail) {
+		++updated;
+
+		type = ITYPE_TYPE(he_dev->irq_head->isw);
+		group = ITYPE_GROUP(he_dev->irq_head->isw);
+
+		switch (type) {
+			case ITYPE_RBRQ_THRESH:
+				HPRINTK("rbrq%d threshold\n", group);
+				/* fall through */
+			case ITYPE_RBRQ_TIMER:
+				if (he_service_rbrq(he_dev, group)) {
+					he_service_rbpl(he_dev, group);
+#ifdef USE_RBPS
+					he_service_rbps(he_dev, group);
+#endif /* USE_RBPS */
+				}
+				break;
+			case ITYPE_TBRQ_THRESH:
+				HPRINTK("tbrq%d threshold\n", group);
+				/* fall through */
+			case ITYPE_TPD_COMPLETE:
+				he_service_tbrq(he_dev, group);
+				break;
+			case ITYPE_RBPL_THRESH:
+				he_service_rbpl(he_dev, group);
+				break;
+			case ITYPE_RBPS_THRESH:
+#ifdef USE_RBPS
+				he_service_rbps(he_dev, group);
+#endif /* USE_RBPS */
+				break;
+			case ITYPE_PHY:
+				HPRINTK("phy interrupt\n");
+#ifdef CONFIG_ATM_HE_USE_SUNI
+				spin_unlock_irqrestore(&he_dev->global_lock, flags);
+				if (he_dev->atm_dev->phy && he_dev->atm_dev->phy->interrupt)
+					he_dev->atm_dev->phy->interrupt(he_dev->atm_dev);
+				spin_lock_irqsave(&he_dev->global_lock, flags);
+#endif
+				break;
+			case ITYPE_OTHER:
+				switch (type|group) {
+					case ITYPE_PARITY:
+						hprintk("parity error\n");
+						break;
+					case ITYPE_ABORT:
+						hprintk("abort 0x%x\n", he_readl(he_dev, ABORT_ADDR));
+						break;
+				}
+				break;
+			case ITYPE_TYPE(ITYPE_INVALID):
+				/* see 8.1.1 -- check all queues */
+
+				HPRINTK("isw not updated 0x%x\n", he_dev->irq_head->isw);
+
+				he_service_rbrq(he_dev, 0);
+				he_service_rbpl(he_dev, 0);
+#ifdef USE_RBPS
+				he_service_rbps(he_dev, 0);
+#endif /* USE_RBPS */
+				he_service_tbrq(he_dev, 0);
+				break;
+			default:
+				hprintk("bad isw 0x%x?\n", he_dev->irq_head->isw);
+		}
+
+		he_dev->irq_head->isw = ITYPE_INVALID;
+
+		he_dev->irq_head = (struct he_irq *) NEXT_ENTRY(he_dev->irq_base, he_dev->irq_head, IRQ_MASK);
+	}
+
+	if (updated) {
+		if (updated > he_dev->irq_peak)
+			he_dev->irq_peak = updated;
+
+		he_writel(he_dev,
+			IRQ_SIZE(CONFIG_IRQ_SIZE) |
+			IRQ_THRESH(CONFIG_IRQ_THRESH) |
+			IRQ_TAIL(he_dev->irq_tail), IRQ0_HEAD);
+		(void) he_readl(he_dev, INT_FIFO); /* 8.1.2 controller errata */
+	}
+#ifdef USE_TASKLET
+	spin_unlock_irqrestore(&he_dev->global_lock, flags);
+#endif
+}
+
+static irqreturn_t
+he_irq_handler(int irq, void *dev_id, struct pt_regs *regs)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = (struct he_dev * )dev_id;
+	int handled = 0;
+
+	if (he_dev == NULL)
+		return IRQ_NONE;
+
+	spin_lock_irqsave(&he_dev->global_lock, flags);
+
+	he_dev->irq_tail = (struct he_irq *) (((unsigned long)he_dev->irq_base) |
+						(*he_dev->irq_tailoffset << 2));
+
+	if (he_dev->irq_tail == he_dev->irq_head) {
+		HPRINTK("tailoffset not updated?\n");
+		he_dev->irq_tail = (struct he_irq *) ((unsigned long)he_dev->irq_base |
+			((he_readl(he_dev, IRQ0_BASE) & IRQ_MASK) << 2));
+		(void) he_readl(he_dev, INT_FIFO);	/* 8.1.2 controller errata */
+	}
+
+#ifdef DEBUG
+	if (he_dev->irq_head == he_dev->irq_tail /* && !IRQ_PENDING */)
+		hprintk("spurious (or shared) interrupt?\n");
+#endif
+
+	if (he_dev->irq_head != he_dev->irq_tail) {
+		handled = 1;
+#ifdef USE_TASKLET
+		tasklet_schedule(&he_dev->tasklet);
+#else
+		he_tasklet((unsigned long) he_dev);
+#endif
+		he_writel(he_dev, INT_CLEAR_A, INT_FIFO);
+							/* clear interrupt */
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl(he_dev, INT_FIFO);
+#endif
+	}
+	spin_unlock_irqrestore(&he_dev->global_lock, flags);
+	return IRQ_RETVAL(handled);
+
+}
+
+static __inline__ void
+__enqueue_tpd(struct he_dev *he_dev, struct he_tpd *tpd, unsigned cid)
+{
+	struct he_tpdrq *new_tail;
+
+	HPRINTK("tpdrq %p cid 0x%x -> tpdrq_tail %p\n",
+					tpd, cid, he_dev->tpdrq_tail);
+
+	/* new_tail = he_dev->tpdrq_tail; */
+	new_tail = (struct he_tpdrq *) ((unsigned long) he_dev->tpdrq_base |
+					TPDRQ_MASK(he_dev->tpdrq_tail+1));
+
+	/*
+	 * check to see if we are about to set the tail == head
+	 * if true, update the head pointer from the adapter
+	 * to see if this is really the case (reading the queue
+	 * head for every enqueue would be unnecessarily slow)
+	 */
+
+	if (new_tail == he_dev->tpdrq_head) {
+		he_dev->tpdrq_head = (struct he_tpdrq *)
+			(((unsigned long)he_dev->tpdrq_base) |
+				TPDRQ_MASK(he_readl(he_dev, TPDRQ_B_H)));
+
+		if (new_tail == he_dev->tpdrq_head) {
+			hprintk("tpdrq full (cid 0x%x)\n", cid);
+			/*
+			 * FIXME
+			 * push tpd onto a transmit backlog queue
+			 * after service_tbrq, service the backlog
+			 * for now, we just drop the pdu
+			 */
+			if (tpd->skb) {
+				if (tpd->vcc->pop)
+					tpd->vcc->pop(tpd->vcc, tpd->skb);
+				else
+					dev_kfree_skb_any(tpd->skb);
+				atomic_inc(&tpd->vcc->stats->tx_err);
+			}
+#ifdef USE_TPD_POOL
+			pci_pool_free(he_dev->tpd_pool, tpd, TPD_ADDR(tpd->status));
+#else
+			tpd->inuse = 0;
+#endif
+			return;
+		}
+	}
+
+	/* 2.1.5 transmit packet descriptor ready queue */
+#ifdef USE_TPD_POOL
+	list_add_tail(&tpd->entry, &he_dev->outstanding_tpds);
+	he_dev->tpdrq_tail->tpd = TPD_ADDR(tpd->status);
+#else
+	he_dev->tpdrq_tail->tpd = he_dev->tpd_base_phys +
+				(TPD_INDEX(tpd->status) * sizeof(struct he_tpd));
+#endif
+	he_dev->tpdrq_tail->cid = cid;
+	wmb();
+
+	he_dev->tpdrq_tail = new_tail;
+
+	he_writel(he_dev, TPDRQ_MASK(he_dev->tpdrq_tail), TPDRQ_T);
+#ifdef CONFIG_IA64_SGI_SN2
+	(void) he_readl(he_dev, TPDRQ_T);
+#endif
+}
+
+static int
+he_open(struct atm_vcc *vcc, short vpi, int vci)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = HE_DEV(vcc->dev);
+	struct he_vcc *he_vcc;
+	int err = 0;
+	unsigned cid, rsr0, rsr1, rsr4, tsr0, tsr0_aal, tsr4, period, reg, clock;
+
+	
+	if ((err = atm_find_ci(vcc, &vpi, &vci))) {
+		HPRINTK("atm_find_ci err = %d\n", err);
+		return err;
+	}
+	if (vci == ATM_VCI_UNSPEC || vpi == ATM_VPI_UNSPEC)
+		return 0;
+	vcc->vpi = vpi;
+	vcc->vci = vci;
+
+	HPRINTK("open vcc %p %d.%d\n", vcc, vpi, vci);
+
+	set_bit(ATM_VF_ADDR, &vcc->flags);
+
+	cid = he_mkcid(he_dev, vpi, vci);
+
+	he_vcc = (struct he_vcc *) kmalloc(sizeof(struct he_vcc), GFP_ATOMIC);
+	if (he_vcc == NULL) {
+		hprintk("unable to allocate he_vcc during open\n");
+		return -ENOMEM;
+	}
+
+	he_vcc->iov_tail = he_vcc->iov_head;
+	he_vcc->pdu_len = 0;
+	he_vcc->rc_index = -1;
+
+	init_waitqueue_head(&he_vcc->rx_waitq);
+	init_waitqueue_head(&he_vcc->tx_waitq);
+
+	HE_VCC(vcc) = he_vcc;
+
+	if (vcc->qos.txtp.traffic_class != ATM_NONE) {
+		int pcr_goal;
+
+		pcr_goal = atm_pcr_goal(&vcc->qos.txtp);
+		if (pcr_goal == 0)
+			pcr_goal = he_dev->atm_dev->link_rate;
+		if (pcr_goal < 0)	/* means round down, technically */
+			pcr_goal = -pcr_goal;
+
+		HPRINTK("open tx cid 0x%x pcr_goal %d\n", cid, pcr_goal);
+
+		switch (vcc->qos.aal) {
+			case ATM_AAL5:
+				tsr0_aal = TSR0_AAL5;
+				tsr4 = TSR4_AAL5;
+				break;
+			case ATM_AAL0:
+				tsr0_aal = TSR0_AAL0_SDU;
+				tsr4 = TSR4_AAL0_SDU;
+				break;
+			default:
+				err = -EINVAL;
+				goto open_failed;
+		}
+
+		spin_lock_irqsave(&he_dev->global_lock, flags);
+		tsr0 = he_readl_tsr0(he_dev, cid);
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+		if (TSR0_CONN_STATE(tsr0) != 0) {
+			hprintk("cid 0x%x not idle (tsr0 = 0x%x)\n", cid, tsr0);
+			err = -EBUSY;
+			goto open_failed;
+		}
+
+		switch (vcc->qos.txtp.traffic_class) {
+			case ATM_UBR:
+				/* 2.3.3.1 open connection ubr */
+
+				tsr0 = TSR0_UBR | TSR0_GROUP(0) | tsr0_aal |
+					TSR0_USE_WMIN | TSR0_UPDATE_GER;
+				break;
+
+			case ATM_CBR:
+				/* 2.3.3.2 open connection cbr */
+
+				/* 8.2.3 cbr scheduler wrap problem -- limit to 90% total link rate */
+				if ((he_dev->total_bw + pcr_goal)
+					> (he_dev->atm_dev->link_rate * 9 / 10))
+				{
+					err = -EBUSY;
+					goto open_failed;
+				}
+
+				spin_lock_irqsave(&he_dev->global_lock, flags);			/* also protects he_dev->cs_stper[] */
+
+				/* find an unused cs_stper register */
+				for (reg = 0; reg < HE_NUM_CS_STPER; ++reg)
+					if (he_dev->cs_stper[reg].inuse == 0 || 
+					    he_dev->cs_stper[reg].pcr == pcr_goal)
+							break;
+
+				if (reg == HE_NUM_CS_STPER) {
+					err = -EBUSY;
+					spin_unlock_irqrestore(&he_dev->global_lock, flags);
+					goto open_failed;
+				}
+
+				he_dev->total_bw += pcr_goal;
+
+				he_vcc->rc_index = reg;
+				++he_dev->cs_stper[reg].inuse;
+				he_dev->cs_stper[reg].pcr = pcr_goal;
+
+				clock = he_is622(he_dev) ? 66667000 : 50000000;
+				period = clock / pcr_goal;
+				
+				HPRINTK("rc_index = %d period = %d\n",
+								reg, period);
+
+				he_writel_mbox(he_dev, rate_to_atmf(period/2),
+							CS_STPER0 + reg);
+				spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+				tsr0 = TSR0_CBR | TSR0_GROUP(0) | tsr0_aal |
+							TSR0_RC_INDEX(reg);
+
+				break;
+			default:
+				err = -EINVAL;
+				goto open_failed;
+		}
+
+		spin_lock_irqsave(&he_dev->global_lock, flags);
+
+		he_writel_tsr0(he_dev, tsr0, cid);
+		he_writel_tsr4(he_dev, tsr4 | 1, cid);
+		he_writel_tsr1(he_dev, TSR1_MCR(rate_to_atmf(0)) |
+					TSR1_PCR(rate_to_atmf(pcr_goal)), cid);
+		he_writel_tsr2(he_dev, TSR2_ACR(rate_to_atmf(pcr_goal)), cid);
+		he_writel_tsr9(he_dev, TSR9_OPEN_CONN, cid);
+
+		he_writel_tsr3(he_dev, 0x0, cid);
+		he_writel_tsr5(he_dev, 0x0, cid);
+		he_writel_tsr6(he_dev, 0x0, cid);
+		he_writel_tsr7(he_dev, 0x0, cid);
+		he_writel_tsr8(he_dev, 0x0, cid);
+		he_writel_tsr10(he_dev, 0x0, cid);
+		he_writel_tsr11(he_dev, 0x0, cid);
+		he_writel_tsr12(he_dev, 0x0, cid);
+		he_writel_tsr13(he_dev, 0x0, cid);
+		he_writel_tsr14(he_dev, 0x0, cid);
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl_tsr0(he_dev, cid);
+#endif
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+	}
+
+	if (vcc->qos.rxtp.traffic_class != ATM_NONE) {
+		unsigned aal;
+
+		HPRINTK("open rx cid 0x%x (rx_waitq %p)\n", cid,
+		 				&HE_VCC(vcc)->rx_waitq);
+
+		switch (vcc->qos.aal) {
+			case ATM_AAL5:
+				aal = RSR0_AAL5;
+				break;
+			case ATM_AAL0:
+				aal = RSR0_RAWCELL;
+				break;
+			default:
+				err = -EINVAL;
+				goto open_failed;
+		}
+
+		spin_lock_irqsave(&he_dev->global_lock, flags);
+
+		rsr0 = he_readl_rsr0(he_dev, cid);
+		if (rsr0 & RSR0_OPEN_CONN) {
+			spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+			hprintk("cid 0x%x not idle (rsr0 = 0x%x)\n", cid, rsr0);
+			err = -EBUSY;
+			goto open_failed;
+		}
+
+#ifdef USE_RBPS
+		rsr1 = RSR1_GROUP(0);
+		rsr4 = RSR4_GROUP(0);
+#else /* !USE_RBPS */
+		rsr1 = RSR1_GROUP(0)|RSR1_RBPL_ONLY;
+		rsr4 = RSR4_GROUP(0)|RSR4_RBPL_ONLY;
+#endif /* USE_RBPS */
+		rsr0 = vcc->qos.rxtp.traffic_class == ATM_UBR ? 
+				(RSR0_EPD_ENABLE|RSR0_PPD_ENABLE) : 0;
+
+#ifdef USE_CHECKSUM_HW
+		if (vpi == 0 && vci >= ATM_NOT_RSV_VCI)
+			rsr0 |= RSR0_TCP_CKSUM;
+#endif
+
+		he_writel_rsr4(he_dev, rsr4, cid);
+		he_writel_rsr1(he_dev, rsr1, cid);
+		/* 5.1.11 last parameter initialized should be
+			  the open/closed indication in rsr0 */
+		he_writel_rsr0(he_dev,
+			rsr0 | RSR0_START_PDU | RSR0_OPEN_CONN | aal, cid);
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl_rsr0(he_dev, cid);
+#endif
+
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+	}
+
+open_failed:
+
+	if (err) {
+		if (he_vcc)
+			kfree(he_vcc);
+		clear_bit(ATM_VF_ADDR, &vcc->flags);
+	}
+	else
+		set_bit(ATM_VF_READY, &vcc->flags);
+
+	return err;
+}
+
+static void
+he_close(struct atm_vcc *vcc)
+{
+	unsigned long flags;
+	DECLARE_WAITQUEUE(wait, current);
+	struct he_dev *he_dev = HE_DEV(vcc->dev);
+	struct he_tpd *tpd;
+	unsigned cid;
+	struct he_vcc *he_vcc = HE_VCC(vcc);
+#define MAX_RETRY 30
+	int retry = 0, sleep = 1, tx_inuse;
+
+	HPRINTK("close vcc %p %d.%d\n", vcc, vcc->vpi, vcc->vci);
+
+	clear_bit(ATM_VF_READY, &vcc->flags);
+	cid = he_mkcid(he_dev, vcc->vpi, vcc->vci);
+
+	if (vcc->qos.rxtp.traffic_class != ATM_NONE) {
+		int timeout;
+
+		HPRINTK("close rx cid 0x%x\n", cid);
+
+		/* 2.7.2.2 close receive operation */
+
+		/* wait for previous close (if any) to finish */
+
+		spin_lock_irqsave(&he_dev->global_lock, flags);
+		while (he_readl(he_dev, RCC_STAT) & RCC_BUSY) {
+			HPRINTK("close cid 0x%x RCC_BUSY\n", cid);
+			udelay(250);
+		}
+
+		add_wait_queue(&he_vcc->rx_waitq, &wait);
+		set_current_state(TASK_UNINTERRUPTIBLE);
+
+		he_writel_rsr0(he_dev, RSR0_CLOSE_CONN, cid);
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl_rsr0(he_dev, cid);
+#endif
+		he_writel_mbox(he_dev, cid, RXCON_CLOSE);
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+		timeout = schedule_timeout(30*HZ);
+
+		remove_wait_queue(&he_vcc->rx_waitq, &wait);
+		set_current_state(TASK_RUNNING);
+
+		if (timeout == 0)
+			hprintk("close rx timeout cid 0x%x\n", cid);
+
+		HPRINTK("close rx cid 0x%x complete\n", cid);
+
+	}
+
+	if (vcc->qos.txtp.traffic_class != ATM_NONE) {
+		volatile unsigned tsr4, tsr0;
+		int timeout;
+
+		HPRINTK("close tx cid 0x%x\n", cid);
+		
+		/* 2.1.2
+		 *
+		 * ... the host must first stop queueing packets to the TPDRQ
+		 * on the connection to be closed, then wait for all outstanding
+		 * packets to be transmitted and their buffers returned to the
+		 * TBRQ. When the last packet on the connection arrives in the
+		 * TBRQ, the host issues the close command to the adapter.
+		 */
+
+		while (((tx_inuse = atomic_read(&vcc->sk->wmem_alloc)) > 0) &&
+		       (retry < MAX_RETRY)) {
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			(void) schedule_timeout(sleep);
+			set_current_state(TASK_RUNNING);
+			if (sleep < HZ)
+				sleep = sleep * 2;
+
+			++retry;
+		}
+
+		if (tx_inuse)
+			hprintk("close tx cid 0x%x tx_inuse = %d\n", cid, tx_inuse);
+
+		/* 2.3.1.1 generic close operations with flush */
+
+		spin_lock_irqsave(&he_dev->global_lock, flags);
+		he_writel_tsr4_upper(he_dev, TSR4_FLUSH_CONN, cid);
+					/* also clears TSR4_SESSION_ENDED */
+#ifdef CONFIG_IA64_SGI_SN2
+		(void) he_readl_tsr4(he_dev, cid);
+#endif
+
+		switch (vcc->qos.txtp.traffic_class) {
+			case ATM_UBR:
+				he_writel_tsr1(he_dev, 
+					TSR1_MCR(rate_to_atmf(200000))
+					| TSR1_PCR(0), cid);
+				break;
+			case ATM_CBR:
+				he_writel_tsr14_upper(he_dev, TSR14_DELETE, cid);
+				break;
+		}
+
+		tpd = __alloc_tpd(he_dev);
+		if (tpd == NULL) {
+			hprintk("close tx he_alloc_tpd failed cid 0x%x\n", cid);
+			goto close_tx_incomplete;
+		}
+		tpd->status |= TPD_EOS | TPD_INT;
+		tpd->skb = NULL;
+		tpd->vcc = vcc;
+		wmb();
+
+		add_wait_queue(&he_vcc->tx_waitq, &wait);
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		__enqueue_tpd(he_dev, tpd, cid);
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+		timeout = schedule_timeout(30*HZ);
+
+		remove_wait_queue(&he_vcc->tx_waitq, &wait);
+		set_current_state(TASK_RUNNING);
+
+		spin_lock_irqsave(&he_dev->global_lock, flags);
+
+		if (timeout == 0) {
+			hprintk("close tx timeout cid 0x%x\n", cid);
+			goto close_tx_incomplete;
+		}
+
+		while (!((tsr4 = he_readl_tsr4(he_dev, cid)) & TSR4_SESSION_ENDED)) {
+			HPRINTK("close tx cid 0x%x !TSR4_SESSION_ENDED (tsr4 = 0x%x)\n", cid, tsr4);
+			udelay(250);
+		}
+
+		while (TSR0_CONN_STATE(tsr0 = he_readl_tsr0(he_dev, cid)) != 0) {
+			HPRINTK("close tx cid 0x%x TSR0_CONN_STATE != 0 (tsr0 = 0x%x)\n", cid, tsr0);
+			udelay(250);
+		}
+
+close_tx_incomplete:
+
+		if (vcc->qos.txtp.traffic_class == ATM_CBR) {
+			int reg = he_vcc->rc_index;
+
+			HPRINTK("cs_stper reg = %d\n", reg);
+
+			if (he_dev->cs_stper[reg].inuse == 0)
+				hprintk("cs_stper[%d].inuse = 0!\n", reg);
+			else
+				--he_dev->cs_stper[reg].inuse;
+
+			he_dev->total_bw -= he_dev->cs_stper[reg].pcr;
+		}
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+		HPRINTK("close tx cid 0x%x complete\n", cid);
+	}
+
+	kfree(he_vcc);
+
+	clear_bit(ATM_VF_ADDR, &vcc->flags);
+}
+
+static int
+he_sg_send(struct atm_vcc *vcc, unsigned long start, unsigned long size)
+{
+#ifdef USE_SCATTERGATHER
+	return 1;
+#else
+	return 0;
+#endif
+}
+
+static int
+he_send(struct atm_vcc *vcc, struct sk_buff *skb)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = HE_DEV(vcc->dev);
+	unsigned cid = he_mkcid(he_dev, vcc->vpi, vcc->vci);
+	struct he_tpd *tpd;
+#ifdef USE_SCATTERGATHER
+	int i, slot = 0;
+#endif
+
+#define HE_TPD_BUFSIZE 0xffff
+
+	HPRINTK("send %d.%d\n", vcc->vpi, vcc->vci);
+
+	if ((skb->len > HE_TPD_BUFSIZE) ||
+	    ((vcc->qos.aal == ATM_AAL0) && (skb->len != ATM_AAL0_SDU))) {
+		hprintk("buffer too large (or small) -- %d bytes\n", skb->len );
+		if (vcc->pop)
+			vcc->pop(vcc, skb);
+		else
+			dev_kfree_skb_any(skb);
+		atomic_inc(&vcc->stats->tx_err);
+		return -EINVAL;
+	}
+
+#ifndef USE_SCATTERGATHER
+	if (skb_shinfo(skb)->nr_frags) {
+		hprintk("no scatter/gather support\n");
+		if (vcc->pop)
+			vcc->pop(vcc, skb);
+		else
+			dev_kfree_skb_any(skb);
+		atomic_inc(&vcc->stats->tx_err);
+		return -EINVAL;
+	}
+#endif
+	spin_lock_irqsave(&he_dev->global_lock, flags);
+
+	tpd = __alloc_tpd(he_dev);
+	if (tpd == NULL) {
+		if (vcc->pop)
+			vcc->pop(vcc, skb);
+		else
+			dev_kfree_skb_any(skb);
+		atomic_inc(&vcc->stats->tx_err);
+		spin_unlock_irqrestore(&he_dev->global_lock, flags);
+		return -ENOMEM;
+	}
+
+	if (vcc->qos.aal == ATM_AAL5)
+		tpd->status |= TPD_CELLTYPE(TPD_USERCELL);
+	else {
+		char *pti_clp = (void *) (skb->data + 3);
+		int clp, pti;
+
+		pti = (*pti_clp & ATM_HDR_PTI_MASK) >> ATM_HDR_PTI_SHIFT; 
+		clp = (*pti_clp & ATM_HDR_CLP);
+		tpd->status |= TPD_CELLTYPE(pti);
+		if (clp)
+			tpd->status |= TPD_CLP;
+
+		skb_pull(skb, ATM_AAL0_SDU - ATM_CELL_PAYLOAD);
+	}
+
+#ifdef USE_SCATTERGATHER
+	tpd->iovec[slot].addr = pci_map_single(he_dev->pci_dev, skb->data,
+				skb->len - skb->data_len, PCI_DMA_TODEVICE);
+	tpd->iovec[slot].len = skb->len - skb->data_len;
+	++slot;
+
+	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+		if (slot == TPD_MAXIOV) {	/* queue tpd; start new tpd */
+			tpd->vcc = vcc;
+			tpd->skb = NULL;	/* not the last fragment
+						   so dont ->push() yet */
+			wmb();
+
+			__enqueue_tpd(he_dev, tpd, cid);
+			tpd = __alloc_tpd(he_dev);
+			if (tpd == NULL) {
+				if (vcc->pop)
+					vcc->pop(vcc, skb);
+				else
+					dev_kfree_skb_any(skb);
+				atomic_inc(&vcc->stats->tx_err);
+				spin_unlock_irqrestore(&he_dev->global_lock, flags);
+				return -ENOMEM;
+			}
+			tpd->status |= TPD_USERCELL;
+			slot = 0;
+		}
+
+		tpd->iovec[slot].addr = pci_map_single(he_dev->pci_dev,
+			(void *) page_address(frag->page) + frag->page_offset,
+				frag->size, PCI_DMA_TODEVICE);
+		tpd->iovec[slot].len = frag->size;
+		++slot;
+
+	}
+
+	tpd->iovec[slot - 1].len |= TPD_LST;
+#else
+	tpd->address0 = pci_map_single(he_dev->pci_dev, skb->data, skb->len, PCI_DMA_TODEVICE);
+	tpd->length0 = skb->len | TPD_LST;
+#endif
+	tpd->status |= TPD_INT;
+
+	tpd->vcc = vcc;
+	tpd->skb = skb;
+	wmb();
+	ATM_SKB(skb)->vcc = vcc;
+
+	__enqueue_tpd(he_dev, tpd, cid);
+	spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+	atomic_inc(&vcc->stats->tx);
+
+	return 0;
+}
+
+static int
+he_ioctl(struct atm_dev *atm_dev, unsigned int cmd, void *arg)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = HE_DEV(atm_dev);
+	struct he_ioctl_reg reg;
+	int err = 0;
+
+	switch (cmd) {
+		case HE_GET_REG:
+			if (!capable(CAP_NET_ADMIN))
+				return -EPERM;
+
+			copy_from_user(&reg, (struct he_ioctl_reg *) arg,
+						sizeof(struct he_ioctl_reg));
+			spin_lock_irqsave(&he_dev->global_lock, flags);
+			switch (reg.type) {
+				case HE_REGTYPE_PCI:
+					reg.val = he_readl(he_dev, reg.addr);
+					break;
+				case HE_REGTYPE_RCM:
+					reg.val =
+						he_readl_rcm(he_dev, reg.addr);
+					break;
+				case HE_REGTYPE_TCM:
+					reg.val =
+						he_readl_tcm(he_dev, reg.addr);
+					break;
+				case HE_REGTYPE_MBOX:
+					reg.val =
+						he_readl_mbox(he_dev, reg.addr);
+					break;
+				default:
+					err = -EINVAL;
+					break;
+			}
+			spin_unlock_irqrestore(&he_dev->global_lock, flags);
+			if (err == 0)
+				copy_to_user((struct he_ioctl_reg *) arg, &reg,
+							sizeof(struct he_ioctl_reg));
+			break;
+		default:
+#ifdef CONFIG_ATM_HE_USE_SUNI
+			if (atm_dev->phy && atm_dev->phy->ioctl)
+				err = atm_dev->phy->ioctl(atm_dev, cmd, arg);
+#else /* CONFIG_ATM_HE_USE_SUNI */
+			err = -EINVAL;
+#endif /* CONFIG_ATM_HE_USE_SUNI */
+			break;
+	}
+
+	return err;
+}
+
+static void
+he_phy_put(struct atm_dev *atm_dev, unsigned char val, unsigned long addr)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = HE_DEV(atm_dev);
+
+	HPRINTK("phy_put(val 0x%x, addr 0x%lx)\n", val, addr);
+
+	spin_lock_irqsave(&he_dev->global_lock, flags);
+	he_writel(he_dev, val, FRAMER + (addr*4));
+#ifdef CONFIG_IA64_SGI_SN2
+	(void) he_readl(he_dev, FRAMER + (addr*4));
+#endif
+	spin_unlock_irqrestore(&he_dev->global_lock, flags);
+}
+ 
+	
+static unsigned char
+he_phy_get(struct atm_dev *atm_dev, unsigned long addr)
+{ 
+	unsigned long flags;
+	struct he_dev *he_dev = HE_DEV(atm_dev);
+	unsigned reg;
+
+	spin_lock_irqsave(&he_dev->global_lock, flags);
+	reg = he_readl(he_dev, FRAMER + (addr*4));
+	spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+	HPRINTK("phy_get(addr 0x%lx) =0x%x\n", addr, reg);
+	return reg;
+}
+
+static int
+he_proc_read(struct atm_dev *dev, loff_t *pos, char *page)
+{
+	unsigned long flags;
+	struct he_dev *he_dev = HE_DEV(dev);
+	int left, i;
+#ifdef notdef
+	struct he_rbrq *rbrq_tail;
+	struct he_tpdrq *tpdrq_head;
+	int rbpl_head, rbpl_tail;
+#endif
+	static long mcc = 0, oec = 0, dcc = 0, cec = 0;
+
+
+	left = *pos;
+	if (!left--)
+		return sprintf(page, "%s\n", version);
+
+	if (!left--)
+		return sprintf(page, "%s%s\n\n",
+			he_dev->prod_id, he_dev->media & 0x40 ? "SM" : "MM");
+
+	if (!left--)
+		return sprintf(page, "Mismatched Cells  VPI/VCI Not Open  Dropped Cells  RCM Dropped Cells\n");
+
+	spin_lock_irqsave(&he_dev->global_lock, flags);
+	mcc += he_readl(he_dev, MCC);
+	oec += he_readl(he_dev, OEC);
+	dcc += he_readl(he_dev, DCC);
+	cec += he_readl(he_dev, CEC);
+	spin_unlock_irqrestore(&he_dev->global_lock, flags);
+
+	if (!left--)
+		return sprintf(page, "%16ld  %16ld  %13ld  %17ld\n\n", 
+							mcc, oec, dcc, cec);
+
+	if (!left--)
+		return sprintf(page, "irq_size = %d  inuse = ?  peak = %d\n",
+				CONFIG_IRQ_SIZE, he_dev->irq_peak);
+
+	if (!left--)
+		return sprintf(page, "tpdrq_size = %d  inuse = ?\n",
+						CONFIG_TPDRQ_SIZE);
+
+	if (!left--)
+		return sprintf(page, "rbrq_size = %d  inuse = ?  peak = %d\n",
+				CONFIG_RBRQ_SIZE, he_dev->rbrq_peak);
+
+	if (!left--)
+		return sprintf(page, "tbrq_size = %d  peak = %d\n",
+					CONFIG_TBRQ_SIZE, he_dev->tbrq_peak);
+
+
+#ifdef notdef
+	rbpl_head = RBPL_MASK(he_readl(he_dev, G0_RBPL_S));
+	rbpl_tail = RBPL_MASK(he_readl(he_dev, G0_RBPL_T));
+
+	inuse = rbpl_head - rbpl_tail;
+	if (inuse < 0)
+		inuse += CONFIG_RBPL_SIZE * sizeof(struct he_rbp);
+	inuse /= sizeof(struct he_rbp);
+
+	if (!left--)
+		return sprintf(page, "rbpl_size = %d  inuse = %d\n\n",
+						CONFIG_RBPL_SIZE, inuse);
+#endif
+
+	if (!left--)
+		return sprintf(page, "rate controller periods (cbr)\n                 pcr  #vc\n");
+
+	for (i = 0; i < HE_NUM_CS_STPER; ++i)
+		if (!left--)
+			return sprintf(page, "cs_stper%-2d  %8ld  %3d\n", i,
+						he_dev->cs_stper[i].pcr,
+						he_dev->cs_stper[i].inuse);
+
+	if (!left--)
+		return sprintf(page, "total bw (cbr): %d  (limit %d)\n",
+			he_dev->total_bw, he_dev->atm_dev->link_rate * 10 / 9);
+
+	return 0;
+}
+
+/* eeprom routines  -- see 4.7 */
+
+u8
+read_prom_byte(struct he_dev *he_dev, int addr)
+{
+	u32 val = 0, tmp_read = 0;
+	int i, j = 0;
+	u8 byte_read = 0;
+
+	val = readl(he_dev->membase + HOST_CNTL);
+	val &= 0xFFFFE0FF;
+       
+	/* Turn on write enable */
+	val |= 0x800;
+	he_writel(he_dev, val, HOST_CNTL);
+       
+	/* Send READ instruction */
+	for (i = 0; i < sizeof(readtab)/sizeof(readtab[0]); i++) {
+		he_writel(he_dev, val | readtab[i], HOST_CNTL);
+		udelay(EEPROM_DELAY);
+	}
+       
+	/* Next, we need to send the byte address to read from */
+	for (i = 7; i >= 0; i--) {
+		he_writel(he_dev, val | clocktab[j++] | (((addr >> i) & 1) << 9), HOST_CNTL);
+		udelay(EEPROM_DELAY);
+		he_writel(he_dev, val | clocktab[j++] | (((addr >> i) & 1) << 9), HOST_CNTL);
+		udelay(EEPROM_DELAY);
+	}
+       
+	j = 0;
+
+	val &= 0xFFFFF7FF;      /* Turn off write enable */
+	he_writel(he_dev, val, HOST_CNTL);
+       
+	/* Now, we can read data from the EEPROM by clocking it in */
+	for (i = 7; i >= 0; i--) {
+		he_writel(he_dev, val | clocktab[j++], HOST_CNTL);
+		udelay(EEPROM_DELAY);
+		tmp_read = he_readl(he_dev, HOST_CNTL);
+		byte_read |= (unsigned char)
+			   ((tmp_read & ID_DOUT) >> ID_DOFFSET << i);
+		he_writel(he_dev, val | clocktab[j++], HOST_CNTL);
+		udelay(EEPROM_DELAY);
+	}
+       
+	he_writel(he_dev, val | ID_CS, HOST_CNTL);
+	udelay(EEPROM_DELAY);
+
+	return byte_read;
+}
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("chas williams <chas@cmf.nrl.navy.mil>");
+MODULE_DESCRIPTION("ForeRunnerHE ATM Adapter driver");
+MODULE_PARM(disable64, "h");
+MODULE_PARM_DESC(disable64, "disable 64-bit pci bus transfers");
+MODULE_PARM(nvpibits, "i");
+MODULE_PARM_DESC(nvpibits, "numbers of bits for vpi (default 0)");
+MODULE_PARM(nvcibits, "i");
+MODULE_PARM_DESC(nvcibits, "numbers of bits for vci (default 12)");
+MODULE_PARM(rx_skb_reserve, "i");
+MODULE_PARM_DESC(rx_skb_reserve, "padding for receive skb (default 16)");
+MODULE_PARM(irq_coalesce, "i");
+MODULE_PARM_DESC(irq_coalesce, "use interrupt coalescing (default 1)");
+MODULE_PARM(sdh, "i");
+MODULE_PARM_DESC(sdh, "use SDH framing (default 0)");
+
+static struct pci_device_id he_pci_tbl[] __devinitdata = {
+	{ PCI_VENDOR_ID_FORE, PCI_DEVICE_ID_FORE_HE, PCI_ANY_ID, PCI_ANY_ID,
+	  0, 0, 0 },
+	{ 0, }
+};
+
+static struct pci_driver he_driver = {
+	.name =		"he",
+	.probe =	he_init_one,
+	.remove =	__devexit_p(he_remove_one),
+	.id_table =	he_pci_tbl,
+};
+
+static int __init he_init(void)
+{
+	return pci_module_init(&he_driver);
+}
+
+static void __exit he_cleanup(void)
+{
+	pci_unregister_driver(&he_driver);
+}
+
+module_init(he_init);
+module_exit(he_cleanup);

FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)