drivers/ieee1394/ohci1394.c

/*
 * ohci1394.c - driver for OHCI 1394 boards
 * Copyright (C)1999,2000 Sebastien Rougeaux <sebastien.rougeaux@anu.edu.au>
 *                        Gord Peters <GordPeters@smarttech.com>
 *              2001      Ben Collins <bcollins@debian.org>
 *
 * 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.
 */

/*
 * Things known to be working:
 * . Async Request Transmit
 * . Async Response Receive
 * . Async Request Receive
 * . Async Response Transmit
 * . Iso Receive
 * . DMA mmap for iso receive
 * . Config ROM generation
 *
 * Things implemented, but still in test phase:
 * . Iso Transmit
 * . Async Stream Packets Transmit (Receive done via Iso interface)
 *
 * Things not implemented:
 * . DMA error recovery
 *
 * Known bugs:
 * . devctl BUS_RESET arg confusion (reset type or root holdoff?)
 *   added LONG_RESET_ROOT and SHORT_RESET_ROOT for root holdoff --kk
 */

/*
 * Acknowledgments:
 *
 * Adam J Richter <adam@yggdrasil.com>
 *  . Use of pci_class to find device
 *
 * Emilie Chung	<emilie.chung@axis.com>
 *  . Tip on Async Request Filter
 *
 * Pascal Drolet <pascal.drolet@informission.ca>
 *  . Various tips for optimization and functionnalities
 *
 * Robert Ficklin <rficklin@westengineering.com>
 *  . Loop in irq_handler
 *
 * James Goodwin <jamesg@Filanet.com>
 *  . Various tips on initialization, self-id reception, etc.
 *
 * Albrecht Dress <ad@mpifr-bonn.mpg.de>
 *  . Apple PowerBook detection
 *
 * Daniel Kobras <daniel.kobras@student.uni-tuebingen.de>
 *  . Reset the board properly before leaving + misc cleanups
 *
 * Leon van Stuivenberg <leonvs@iae.nl>
 *  . Bug fixes
 *
 * Ben Collins <bcollins@debian.org>
 *  . Working big-endian support
 *  . Updated to 2.4.x module scheme (PCI aswell)
 *  . Config ROM generation
 *
 * Manfred Weihs <weihs@ict.tuwien.ac.at>
 *  . Reworked code for initiating bus resets
 *    (long, short, with or without hold-off)
 *
 * Nandu Santhi <contactnandu@users.sourceforge.net>
 *  . Added support for nVidia nForce2 onboard Firewire chipset
 *
 */

#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/slab.h>
#include <linux/interrupt.h>
#include <linux/wait.h>
#include <linux/errno.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/pci.h>
#include <linux/fs.h>
#include <linux/poll.h>
#include <asm/byteorder.h>
#include <asm/atomic.h>
#include <asm/uaccess.h>
#include <linux/delay.h>
#include <linux/spinlock.h>

#include <asm/pgtable.h>
#include <asm/page.h>
#include <asm/irq.h>
#include <linux/sched.h>
#include <linux/types.h>
#include <linux/vmalloc.h>
#include <linux/init.h>

#ifdef CONFIG_PPC_PMAC
#include <asm/machdep.h>
#include <asm/pmac_feature.h>
#include <asm/prom.h>
#include <asm/pci-bridge.h>
#endif

#include "csr1212.h"
#include "ieee1394.h"
#include "ieee1394_types.h"
#include "hosts.h"
#include "dma.h"
#include "iso.h"
#include "ieee1394_core.h"
#include "highlevel.h"
#include "ohci1394.h"

#ifdef CONFIG_IEEE1394_VERBOSEDEBUG
#define OHCI1394_DEBUG
#endif

#ifdef DBGMSG
#undef DBGMSG
#endif

#ifdef OHCI1394_DEBUG
#define DBGMSG(fmt, args...) \
printk(KERN_INFO "%s: fw-host%d: " fmt "\n" , OHCI1394_DRIVER_NAME, ohci->host->id , ## args)
#else
#define DBGMSG(fmt, args...) do {} while (0)
#endif

#ifdef CONFIG_IEEE1394_OHCI_DMA_DEBUG
#define OHCI_DMA_ALLOC(fmt, args...) \
	HPSB_ERR("%s(%s)alloc(%d): "fmt, OHCI1394_DRIVER_NAME, __FUNCTION__, \
		++global_outstanding_dmas, ## args)
#define OHCI_DMA_FREE(fmt, args...) \
	HPSB_ERR("%s(%s)free(%d): "fmt, OHCI1394_DRIVER_NAME, __FUNCTION__, \
		--global_outstanding_dmas, ## args)
static int global_outstanding_dmas = 0;
#else
#define OHCI_DMA_ALLOC(fmt, args...) do {} while (0)
#define OHCI_DMA_FREE(fmt, args...) do {} while (0)
#endif

/* print general (card independent) information */
#define PRINT_G(level, fmt, args...) \
printk(level "%s: " fmt "\n" , OHCI1394_DRIVER_NAME , ## args)

/* print card specific information */
#define PRINT(level, fmt, args...) \
printk(level "%s: fw-host%d: " fmt "\n" , OHCI1394_DRIVER_NAME, ohci->host->id , ## args)

/* Module Parameters */
static int phys_dma = 1;
module_param(phys_dma, int, 0444);
MODULE_PARM_DESC(phys_dma, "Enable physical dma (default = 1).");

static void dma_trm_tasklet(unsigned long data);
static void dma_trm_reset(struct dma_trm_ctx *d);

static int alloc_dma_rcv_ctx(struct ti_ohci *ohci, struct dma_rcv_ctx *d,
			     enum context_type type, int ctx, int num_desc,
			     int buf_size, int split_buf_size, int context_base);
static void stop_dma_rcv_ctx(struct dma_rcv_ctx *d);
static void free_dma_rcv_ctx(struct dma_rcv_ctx *d);

static int alloc_dma_trm_ctx(struct ti_ohci *ohci, struct dma_trm_ctx *d,
			     enum context_type type, int ctx, int num_desc,
			     int context_base);

static void ohci1394_pci_remove(struct pci_dev *pdev);

#ifndef __LITTLE_ENDIAN
const static size_t hdr_sizes[] = {
	3,	/* TCODE_WRITEQ */
	4,	/* TCODE_WRITEB */
	3,	/* TCODE_WRITE_RESPONSE */
	0,	/* reserved */
	3,	/* TCODE_READQ */
	4,	/* TCODE_READB */
	3,	/* TCODE_READQ_RESPONSE */
	4,	/* TCODE_READB_RESPONSE */
	1,	/* TCODE_CYCLE_START */
	4,	/* TCODE_LOCK_REQUEST */
	2,	/* TCODE_ISO_DATA */
	4,	/* TCODE_LOCK_RESPONSE */
		/* rest is reserved or link-internal */
};

static inline void header_le32_to_cpu(quadlet_t *data, unsigned char tcode)
{
	size_t size;

	if (unlikely(tcode >= ARRAY_SIZE(hdr_sizes)))
		return;

	size = hdr_sizes[tcode];
	while (size--)
		data[size] = le32_to_cpu(data[size]);
}
#else
#define header_le32_to_cpu(w,x) do {} while (0)
#endif /* !LITTLE_ENDIAN */

/***********************************
 * IEEE-1394 functionality section *
 ***********************************/

static u8 get_phy_reg(struct ti_ohci *ohci, u8 addr)
{
	int i;
	unsigned long flags;
	quadlet_t r;

	spin_lock_irqsave (&ohci->phy_reg_lock, flags);

	reg_write(ohci, OHCI1394_PhyControl, (addr << 8) | 0x00008000);

	for (i = 0; i < OHCI_LOOP_COUNT; i++) {
		if (reg_read(ohci, OHCI1394_PhyControl) & 0x80000000)
			break;

		mdelay(1);
	}

	r = reg_read(ohci, OHCI1394_PhyControl);

	if (i >= OHCI_LOOP_COUNT)
		PRINT (KERN_ERR, "Get PHY Reg timeout [0x%08x/0x%08x/%d]",
		       r, r & 0x80000000, i);

	spin_unlock_irqrestore (&ohci->phy_reg_lock, flags);

	return (r & 0x00ff0000) >> 16;
}

static void set_phy_reg(struct ti_ohci *ohci, u8 addr, u8 data)
{
	int i;
	unsigned long flags;
	u32 r = 0;

	spin_lock_irqsave (&ohci->phy_reg_lock, flags);

	reg_write(ohci, OHCI1394_PhyControl, (addr << 8) | data | 0x00004000);

	for (i = 0; i < OHCI_LOOP_COUNT; i++) {
		r = reg_read(ohci, OHCI1394_PhyControl);
		if (!(r & 0x00004000))
			break;

		mdelay(1);
	}

	if (i == OHCI_LOOP_COUNT)
		PRINT (KERN_ERR, "Set PHY Reg timeout [0x%08x/0x%08x/%d]",
		       r, r & 0x00004000, i);

	spin_unlock_irqrestore (&ohci->phy_reg_lock, flags);

	return;
}

/* Or's our value into the current value */
static void set_phy_reg_mask(struct ti_ohci *ohci, u8 addr, u8 data)
{
	u8 old;

	old = get_phy_reg (ohci, addr);
	old |= data;
	set_phy_reg (ohci, addr, old);

	return;
}

static void handle_selfid(struct ti_ohci *ohci, struct hpsb_host *host,
				int phyid, int isroot)
{
	quadlet_t *q = ohci->selfid_buf_cpu;
	quadlet_t self_id_count=reg_read(ohci, OHCI1394_SelfIDCount);
	size_t size;
	quadlet_t q0, q1;

	/* Check status of self-id reception */

	if (ohci->selfid_swap)
		q0 = le32_to_cpu(q[0]);
	else
		q0 = q[0];

	if ((self_id_count & 0x80000000) ||
	    ((self_id_count & 0x00FF0000) != (q0 & 0x00FF0000))) {
		PRINT(KERN_ERR,
		      "Error in reception of SelfID packets [0x%08x/0x%08x] (count: %d)",
		      self_id_count, q0, ohci->self_id_errors);

		/* Tip by James Goodwin <jamesg@Filanet.com>:
		 * We had an error, generate another bus reset in response.  */
		if (ohci->self_id_errors<OHCI1394_MAX_SELF_ID_ERRORS) {
			set_phy_reg_mask (ohci, 1, 0x40);
			ohci->self_id_errors++;
		} else {
			PRINT(KERN_ERR,
			      "Too many errors on SelfID error reception, giving up!");
		}
		return;
	}

	/* SelfID Ok, reset error counter. */
	ohci->self_id_errors = 0;

	size = ((self_id_count & 0x00001FFC) >> 2) - 1;
	q++;

	while (size > 0) {
		if (ohci->selfid_swap) {
			q0 = le32_to_cpu(q[0]);
			q1 = le32_to_cpu(q[1]);
		} else {
			q0 = q[0];
			q1 = q[1];
		}

		if (q0 == ~q1) {
			DBGMSG ("SelfID packet 0x%x received", q0);
			hpsb_selfid_received(host, cpu_to_be32(q0));
			if (((q0 & 0x3f000000) >> 24) == phyid)
				DBGMSG ("SelfID for this node is 0x%08x", q0);
		} else {
			PRINT(KERN_ERR,
			      "SelfID is inconsistent [0x%08x/0x%08x]", q0, q1);
		}
		q += 2;
		size -= 2;
	}

	DBGMSG("SelfID complete");

	return;
}

static void ohci_soft_reset(struct ti_ohci *ohci) {
	int i;

	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);

	for (i = 0; i < OHCI_LOOP_COUNT; i++) {
		if (!(reg_read(ohci, OHCI1394_HCControlSet) & OHCI1394_HCControl_softReset))
			break;
		mdelay(1);
	}
	DBGMSG ("Soft reset finished");
}


/* Generate the dma receive prgs and start the context */
static void initialize_dma_rcv_ctx(struct dma_rcv_ctx *d, int generate_irq)
{
	struct ti_ohci *ohci = (struct ti_ohci*)(d->ohci);
	int i;

	ohci1394_stop_context(ohci, d->ctrlClear, NULL);

	for (i=0; i<d->num_desc; i++) {
		u32 c;

		c = DMA_CTL_INPUT_MORE | DMA_CTL_UPDATE | DMA_CTL_BRANCH;
		if (generate_irq)
			c |= DMA_CTL_IRQ;

		d->prg_cpu[i]->control = cpu_to_le32(c | d->buf_size);

		/* End of descriptor list? */
		if (i + 1 < d->num_desc) {
			d->prg_cpu[i]->branchAddress =
				cpu_to_le32((d->prg_bus[i+1] & 0xfffffff0) | 0x1);
		} else {
			d->prg_cpu[i]->branchAddress =
				cpu_to_le32((d->prg_bus[0] & 0xfffffff0));
		}

		d->prg_cpu[i]->address = cpu_to_le32(d->buf_bus[i]);
		d->prg_cpu[i]->status = cpu_to_le32(d->buf_size);
	}

        d->buf_ind = 0;
        d->buf_offset = 0;

	if (d->type == DMA_CTX_ISO) {
		/* Clear contextControl */
		reg_write(ohci, d->ctrlClear, 0xffffffff);

		/* Set bufferFill, isochHeader, multichannel for IR context */
		reg_write(ohci, d->ctrlSet, 0xd0000000);

		/* Set the context match register to match on all tags */
		reg_write(ohci, d->ctxtMatch, 0xf0000000);

		/* Clear the multi channel mask high and low registers */
		reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear, 0xffffffff);
		reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear, 0xffffffff);

		/* Set up isoRecvIntMask to generate interrupts */
		reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << d->ctx);
	}

	/* Tell the controller where the first AR program is */
	reg_write(ohci, d->cmdPtr, d->prg_bus[0] | 0x1);

	/* Run context */
	reg_write(ohci, d->ctrlSet, 0x00008000);

	DBGMSG("Receive DMA ctx=%d initialized", d->ctx);
}

/* Initialize the dma transmit context */
static void initialize_dma_trm_ctx(struct dma_trm_ctx *d)
{
	struct ti_ohci *ohci = (struct ti_ohci*)(d->ohci);

	/* Stop the context */
	ohci1394_stop_context(ohci, d->ctrlClear, NULL);

        d->prg_ind = 0;
	d->sent_ind = 0;
	d->free_prgs = d->num_desc;
        d->branchAddrPtr = NULL;
	INIT_LIST_HEAD(&d->fifo_list);
	INIT_LIST_HEAD(&d->pending_list);

	if (d->type == DMA_CTX_ISO) {
		/* enable interrupts */
		reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << d->ctx);
	}

	DBGMSG("Transmit DMA ctx=%d initialized", d->ctx);
}

/* Count the number of available iso contexts */
static int get_nb_iso_ctx(struct ti_ohci *ohci, int reg)
{
	int i,ctx=0;
	u32 tmp;

	reg_write(ohci, reg, 0xffffffff);
	tmp = reg_read(ohci, reg);

	DBGMSG("Iso contexts reg: %08x implemented: %08x", reg, tmp);

	/* Count the number of contexts */
	for (i=0; i<32; i++) {
	    	if (tmp & 1) ctx++;
		tmp >>= 1;
	}
	return ctx;
}

/* Global initialization */
static void ohci_initialize(struct ti_ohci *ohci)
{
	quadlet_t buf;
	int num_ports, i;

	spin_lock_init(&ohci->phy_reg_lock);

	/* Put some defaults to these undefined bus options */
	buf = reg_read(ohci, OHCI1394_BusOptions);
	buf |=  0x60000000; /* Enable CMC and ISC */
	if (hpsb_disable_irm)
		buf &= ~0x80000000;
	else
		buf |=  0x80000000; /* Enable IRMC */
	buf &= ~0x00ff0000; /* XXX: Set cyc_clk_acc to zero for now */
	buf &= ~0x18000000; /* Disable PMC and BMC */
	reg_write(ohci, OHCI1394_BusOptions, buf);

	/* Set the bus number */
	reg_write(ohci, OHCI1394_NodeID, 0x0000ffc0);

	/* Enable posted writes */
	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_postedWriteEnable);

	/* Clear link control register */
	reg_write(ohci, OHCI1394_LinkControlClear, 0xffffffff);

	/* Enable cycle timer and cycle master and set the IRM
	 * contender bit in our self ID packets if appropriate. */
	reg_write(ohci, OHCI1394_LinkControlSet,
		  OHCI1394_LinkControl_CycleTimerEnable |
		  OHCI1394_LinkControl_CycleMaster);
	i = get_phy_reg(ohci, 4) | PHY_04_LCTRL;
	if (hpsb_disable_irm)
		i &= ~PHY_04_CONTENDER;
	else
		i |= PHY_04_CONTENDER;
	set_phy_reg(ohci, 4, i);

	/* Set up self-id dma buffer */
	reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->selfid_buf_bus);

	/* enable self-id and phys */
	reg_write(ohci, OHCI1394_LinkControlSet, OHCI1394_LinkControl_RcvSelfID |
		  OHCI1394_LinkControl_RcvPhyPkt);

	/* Set the Config ROM mapping register */
	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->csr_config_rom_bus);

	/* Now get our max packet size */
	ohci->max_packet_size =
		1<<(((reg_read(ohci, OHCI1394_BusOptions)>>12)&0xf)+1);
		
	/* Don't accept phy packets into AR request context */
	reg_write(ohci, OHCI1394_LinkControlClear, 0x00000400);

	/* Clear the interrupt mask */
	reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 0xffffffff);

	/* Clear the interrupt mask */
	reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 0xffffffff);

	/* Initialize AR dma */
	initialize_dma_rcv_ctx(&ohci->ar_req_context, 0);
	initialize_dma_rcv_ctx(&ohci->ar_resp_context, 0);

	/* Initialize AT dma */
	initialize_dma_trm_ctx(&ohci->at_req_context);
	initialize_dma_trm_ctx(&ohci->at_resp_context);
	
	/* Initialize IR Legacy DMA channel mask */
	ohci->ir_legacy_channels = 0;

	/* Accept AR requests from all nodes */
	reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);

	/* Set the address range of the physical response unit.
	 * Most controllers do not implement it as a writable register though.
	 * They will keep a hardwired offset of 0x00010000 and show 0x0 as
	 * register content.
	 * To actually enable physical responses is the job of our interrupt
	 * handler which programs the physical request filter. */
	reg_write(ohci, OHCI1394_PhyUpperBound,
		  OHCI1394_PHYS_UPPER_BOUND_PROGRAMMED >> 16);

	DBGMSG("physUpperBoundOffset=%08x",
	       reg_read(ohci, OHCI1394_PhyUpperBound));

	/* Specify AT retries */
	reg_write(ohci, OHCI1394_ATRetries,
		  OHCI1394_MAX_AT_REQ_RETRIES |
		  (OHCI1394_MAX_AT_RESP_RETRIES<<4) |
		  (OHCI1394_MAX_PHYS_RESP_RETRIES<<8));

	/* We don't want hardware swapping */
	reg_write(ohci, OHCI1394_HCControlClear, OHCI1394_HCControl_noByteSwap);

	/* Enable interrupts */
	reg_write(ohci, OHCI1394_IntMaskSet,
		  OHCI1394_unrecoverableError |
		  OHCI1394_masterIntEnable |
		  OHCI1394_busReset |
		  OHCI1394_selfIDComplete |
		  OHCI1394_RSPkt |
		  OHCI1394_RQPkt |
		  OHCI1394_respTxComplete |
		  OHCI1394_reqTxComplete |
		  OHCI1394_isochRx |
		  OHCI1394_isochTx |
		  OHCI1394_postedWriteErr |
		  OHCI1394_cycleTooLong |
		  OHCI1394_cycleInconsistent);

	/* Enable link */
	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_linkEnable);

	buf = reg_read(ohci, OHCI1394_Version);
	PRINT(KERN_INFO, "OHCI-1394 %d.%d (PCI): IRQ=[%d]  "
	      "MMIO=[%llx-%llx]  Max Packet=[%d]  IR/IT contexts=[%d/%d]",
	      ((((buf) >> 16) & 0xf) + (((buf) >> 20) & 0xf) * 10),
	      ((((buf) >> 4) & 0xf) + ((buf) & 0xf) * 10), ohci->dev->irq,
	      (unsigned long long)pci_resource_start(ohci->dev, 0),
	      (unsigned long long)pci_resource_start(ohci->dev, 0) + OHCI1394_REGISTER_SIZE - 1,
	      ohci->max_packet_size,
	      ohci->nb_iso_rcv_ctx, ohci->nb_iso_xmit_ctx);

	/* Check all of our ports to make sure that if anything is
	 * connected, we enable that port. */
	num_ports = get_phy_reg(ohci, 2) & 0xf;
	for (i = 0; i < num_ports; i++) {
		unsigned int status;

		set_phy_reg(ohci, 7, i);
		status = get_phy_reg(ohci, 8);

		if (status & 0x20)
			set_phy_reg(ohci, 8, status & ~1);
	}

        /* Serial EEPROM Sanity check. */
        if ((ohci->max_packet_size < 512) ||
	    (ohci->max_packet_size > 4096)) {
		/* Serial EEPROM contents are suspect, set a sane max packet
		 * size and print the raw contents for bug reports if verbose
		 * debug is enabled. */
#ifdef CONFIG_IEEE1394_VERBOSEDEBUG
		int i;
#endif

		PRINT(KERN_DEBUG, "Serial EEPROM has suspicious values, "
                      "attempting to setting max_packet_size to 512 bytes");
		reg_write(ohci, OHCI1394_BusOptions,
			  (reg_read(ohci, OHCI1394_BusOptions) & 0xf007) | 0x8002);
		ohci->max_packet_size = 512;
#ifdef CONFIG_IEEE1394_VERBOSEDEBUG
		PRINT(KERN_DEBUG, "    EEPROM Present: %d",
		      (reg_read(ohci, OHCI1394_Version) >> 24) & 0x1);
		reg_write(ohci, OHCI1394_GUID_ROM, 0x80000000);

		for (i = 0;
		     ((i < 1000) &&
		      (reg_read(ohci, OHCI1394_GUID_ROM) & 0x80000000)); i++)
			udelay(10);

		for (i = 0; i < 0x20; i++) {
			reg_write(ohci, OHCI1394_GUID_ROM, 0x02000000);
			PRINT(KERN_DEBUG, "    EEPROM %02x: %02x", i,
			      (reg_read(ohci, OHCI1394_GUID_ROM) >> 16) & 0xff);
		}
#endif
	}
}

/*
 * Insert a packet in the DMA fifo and generate the DMA prg
 * FIXME: rewrite the program in order to accept packets crossing
 *        page boundaries.
 *        check also that a single dma descriptor doesn't cross a
 *        page boundary.
 */
static void insert_packet(struct ti_ohci *ohci,
			  struct dma_trm_ctx *d, struct hpsb_packet *packet)
{
	u32 cycleTimer;
	int idx = d->prg_ind;

	DBGMSG("Inserting packet for node " NODE_BUS_FMT
	       ", tlabel=%d, tcode=0x%x, speed=%d",
	       NODE_BUS_ARGS(ohci->host, packet->node_id), packet->tlabel,
	       packet->tcode, packet->speed_code);

	d->prg_cpu[idx]->begin.address = 0;
	d->prg_cpu[idx]->begin.branchAddress = 0;

	if (d->type == DMA_CTX_ASYNC_RESP) {
		/*
		 * For response packets, we need to put a timeout value in
		 * the 16 lower bits of the status... let's try 1 sec timeout
		 */
		cycleTimer = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
		d->prg_cpu[idx]->begin.status = cpu_to_le32(
			(((((cycleTimer>>25)&0x7)+1)&0x7)<<13) |
			((cycleTimer&0x01fff000)>>12));

		DBGMSG("cycleTimer: %08x timeStamp: %08x",
		       cycleTimer, d->prg_cpu[idx]->begin.status);
	} else 
		d->prg_cpu[idx]->begin.status = 0;

        if ( (packet->type == hpsb_async) || (packet->type == hpsb_raw) ) {

                if (packet->type == hpsb_raw) {
			d->prg_cpu[idx]->data[0] = cpu_to_le32(OHCI1394_TCODE_PHY<<4);
                        d->prg_cpu[idx]->data[1] = cpu_to_le32(packet->header[0]);
                        d->prg_cpu[idx]->data[2] = cpu_to_le32(packet->header[1]);
                } else {
                        d->prg_cpu[idx]->data[0] = packet->speed_code<<16 |
                                (packet->header[0] & 0xFFFF);

			if (packet->tcode == TCODE_ISO_DATA) {
				/* Sending an async stream packet */
				d->prg_cpu[idx]->data[1] = packet->header[0] & 0xFFFF0000;
			} else {
				/* Sending a normal async request or response */
				d->prg_cpu[idx]->data[1] =
					(packet->header[1] & 0xFFFF) |
					(packet->header[0] & 0xFFFF0000);
				d->prg_cpu[idx]->data[2] = packet->header[2];
				d->prg_cpu[idx]->data[3] = packet->header[3];
			}
			header_le32_to_cpu(d->prg_cpu[idx]->data, packet->tcode);
                }

                if (packet->data_size) { /* block transmit */
			if (packet->tcode == TCODE_STREAM_DATA){
				d->prg_cpu[idx]->begin.control =
					cpu_to_le32(DMA_CTL_OUTPUT_MORE |
						    DMA_CTL_IMMEDIATE | 0x8);
			} else {
				d->prg_cpu[idx]->begin.control =
					cpu_to_le32(DMA_CTL_OUTPUT_MORE |
						    DMA_CTL_IMMEDIATE | 0x10);
			}
                        d->prg_cpu[idx]->end.control =
                                cpu_to_le32(DMA_CTL_OUTPUT_LAST |
					    DMA_CTL_IRQ |
					    DMA_CTL_BRANCH |
					    packet->data_size);
                        /*
                         * Check that the packet data buffer
                         * does not cross a page boundary.
			 *
			 * XXX Fix this some day. eth1394 seems to trigger
			 * it, but ignoring it doesn't seem to cause a
			 * problem.
                         */
#if 0
                        if (cross_bound((unsigned long)packet->data,
                                        packet->data_size)>0) {
                                /* FIXME: do something about it */
                                PRINT(KERN_ERR,
                                      "%s: packet data addr: %p size %Zd bytes "
                                      "cross page boundary", __FUNCTION__,
                                      packet->data, packet->data_size);
                        }
#endif
                        d->prg_cpu[idx]->end.address = cpu_to_le32(
                                pci_map_single(ohci->dev, packet->data,
                                               packet->data_size,
                                               PCI_DMA_TODEVICE));
			OHCI_DMA_ALLOC("single, block transmit packet");

                        d->prg_cpu[idx]->end.branchAddress = 0;
                        d->prg_cpu[idx]->end.status = 0;
                        if (d->branchAddrPtr)
                                *(d->branchAddrPtr) =
					cpu_to_le32(d->prg_bus[idx] | 0x3);
                        d->branchAddrPtr =
                                &(d->prg_cpu[idx]->end.branchAddress);
                } else { /* quadlet transmit */
                        if (packet->type == hpsb_raw)
                                d->prg_cpu[idx]->begin.control =
					cpu_to_le32(DMA_CTL_OUTPUT_LAST |
						    DMA_CTL_IMMEDIATE |
						    DMA_CTL_IRQ |
						    DMA_CTL_BRANCH |
						    (packet->header_size + 4));
                        else
                                d->prg_cpu[idx]->begin.control =
					cpu_to_le32(DMA_CTL_OUTPUT_LAST |
						    DMA_CTL_IMMEDIATE |
						    DMA_CTL_IRQ |
						    DMA_CTL_BRANCH |
						    packet->header_size);

                        if (d->branchAddrPtr)
                                *(d->branchAddrPtr) =
					cpu_to_le32(d->prg_bus[idx] | 0x2);
                        d->branchAddrPtr =
                                &(d->prg_cpu[idx]->begin.branchAddress);
                }

        } else { /* iso packet */
                d->prg_cpu[idx]->data[0] = packet->speed_code<<16 |
                        (packet->header[0] & 0xFFFF);
                d->prg_cpu[idx]->data[1] = packet->header[0] & 0xFFFF0000;
		header_le32_to_cpu(d->prg_cpu[idx]->data, packet->tcode);

                d->prg_cpu[idx]->begin.control =
			cpu_to_le32(DMA_CTL_OUTPUT_MORE |
				    DMA_CTL_IMMEDIATE | 0x8);
                d->prg_cpu[idx]->end.control =
			cpu_to_le32(DMA_CTL_OUTPUT_LAST |
				    DMA_CTL_UPDATE |
				    DMA_CTL_IRQ |
				    DMA_CTL_BRANCH |
				    packet->data_size);
                d->prg_cpu[idx]->end.address = cpu_to_le32(
				pci_map_single(ohci->dev, packet->data,
				packet->data_size, PCI_DMA_TODEVICE));
		OHCI_DMA_ALLOC("single, iso transmit packet");

                d->prg_cpu[idx]->end.branchAddress = 0;
                d->prg_cpu[idx]->end.status = 0;
                DBGMSG("Iso xmit context info: header[%08x %08x]\n"
                       "                       begin=%08x %08x %08x %08x\n"
                       "                             %08x %08x %08x %08x\n"
                       "                       end  =%08x %08x %08x %08x",
                       d->prg_cpu[idx]->data[0], d->prg_cpu[idx]->data[1],
                       d->prg_cpu[idx]->begin.control,
                       d->prg_cpu[idx]->begin.address,
                       d->prg_cpu[idx]->begin.branchAddress,
                       d->prg_cpu[idx]->begin.status,
                       d->prg_cpu[idx]->data[0],
                       d->prg_cpu[idx]->data[1],
                       d->prg_cpu[idx]->data[2],
                       d->prg_cpu[idx]->data[3],
                       d->prg_cpu[idx]->end.control,
                       d->prg_cpu[idx]->end.address,
                       d->prg_cpu[idx]->end.branchAddress,
                       d->prg_cpu[idx]->end.status);
                if (d->branchAddrPtr)
  		        *(d->branchAddrPtr) = cpu_to_le32(d->prg_bus[idx] | 0x3);
                d->branchAddrPtr = &(d->prg_cpu[idx]->end.branchAddress);
        }
	d->free_prgs--;

	/* queue the packet in the appropriate context queue */
	list_add_tail(&packet->driver_list, &d->fifo_list);
	d->prg_ind = (d->prg_ind + 1) % d->num_desc;
}

/*
 * This function fills the FIFO with the (eventual) pending packets
 * and runs or wakes up the DMA prg if necessary.
 *
 * The function MUST be called with the d->lock held.
 */
static void dma_trm_flush(struct ti_ohci *ohci, struct dma_trm_ctx *d)
{
	struct hpsb_packet *packet, *ptmp;
	int idx = d->prg_ind;
	int z = 0;

	/* insert the packets into the dma fifo */
	list_for_each_entry_safe(packet, ptmp, &d->pending_list, driver_list) {
		if (!d->free_prgs)
			break;

		/* For the first packet only */
		if (!z)
			z = (packet->data_size) ? 3 : 2;

		/* Insert the packet */
		list_del_init(&packet->driver_list);
		insert_packet(ohci, d, packet);
	}

	/* Nothing must have been done, either no free_prgs or no packets */
	if (z == 0)
		return;

	/* Is the context running ? (should be unless it is
	   the first packet to be sent in this context) */
	if (!(reg_read(ohci, d->ctrlSet) & 0x8000)) {
		u32 nodeId = reg_read(ohci, OHCI1394_NodeID);

		DBGMSG("Starting transmit DMA ctx=%d",d->ctx);
		reg_write(ohci, d->cmdPtr, d->prg_bus[idx] | z);

		/* Check that the node id is valid, and not 63 */
		if (!(nodeId & 0x80000000) || (nodeId & 0x3f) == 63)
			PRINT(KERN_ERR, "Running dma failed because Node ID is not valid");
		else
			reg_write(ohci, d->ctrlSet, 0x8000);
	} else {
		/* Wake up the dma context if necessary */
		if (!(reg_read(ohci, d->ctrlSet) & 0x400))
			DBGMSG("Waking transmit DMA ctx=%d",d->ctx);

		/* do this always, to avoid race condition */
		reg_write(ohci, d->ctrlSet, 0x1000);
	}

	return;
}

/* Transmission of an async or iso packet */
static int ohci_transmit(struct hpsb_host *host, struct hpsb_packet *packet)
{
	struct ti_ohci *ohci = host->hostdata;
	struct dma_trm_ctx *d;
	unsigned long flags;

	if (packet->data_size > ohci->max_packet_size) {
		PRINT(KERN_ERR,
		      "Transmit packet size %Zd is too big",
		      packet->data_size);
		return -EOVERFLOW;
	}

	/* Decide whether we have an iso, a request, or a response packet */
	if (packet->type == hpsb_raw)
		d = &ohci->at_req_context;
	else if ((packet->tcode == TCODE_ISO_DATA) && (packet->type == hpsb_iso)) {
		/* The legacy IT DMA context is initialized on first
		 * use.  However, the alloc cannot be run from
		 * interrupt context, so we bail out if that is the
		 * case. I don't see anyone sending ISO packets from
		 * interrupt context anyway... */

		if (ohci->it_legacy_context.ohci == NULL) {
			if (in_interrupt()) {
				PRINT(KERN_ERR,
				      "legacy IT context cannot be initialized during interrupt");
				return -EINVAL;
			}

			if (alloc_dma_trm_ctx(ohci, &ohci->it_legacy_context,
					      DMA_CTX_ISO, 0, IT_NUM_DESC,
					      OHCI1394_IsoXmitContextBase) < 0) {
				PRINT(KERN_ERR,
				      "error initializing legacy IT context");
				return -ENOMEM;
			}

			initialize_dma_trm_ctx(&ohci->it_legacy_context);
		}

		d = &ohci->it_legacy_context;
	} else if ((packet->tcode & 0x02) && (packet->tcode != TCODE_ISO_DATA))
		d = &ohci->at_resp_context;
	else
		d = &ohci->at_req_context;

	spin_lock_irqsave(&d->lock,flags);

	list_add_tail(&packet->driver_list, &d->pending_list);

	dma_trm_flush(ohci, d);

	spin_unlock_irqrestore(&d->lock,flags);

	return 0;
}

static int ohci_devctl(struct hpsb_host *host, enum devctl_cmd cmd, int arg)
{
	struct ti_ohci *ohci = host->hostdata;
	int retval = 0;
	unsigned long flags;
	int phy_reg;

	switch (cmd) {
	case RESET_BUS:
		switch (arg) {
		case SHORT_RESET:
			phy_reg = get_phy_reg(ohci, 5);
			phy_reg |= 0x40;
			set_phy_reg(ohci, 5, phy_reg); /* set ISBR */
			break;
		case LONG_RESET:
			phy_reg = get_phy_reg(ohci, 1);
			phy_reg |= 0x40;
			set_phy_reg(ohci, 1, phy_reg); /* set IBR */
			break;
		case SHORT_RESET_NO_FORCE_ROOT:
			phy_reg = get_phy_reg(ohci, 1);
			if (phy_reg & 0x80) {
				phy_reg &= ~0x80;
				set_phy_reg(ohci, 1, phy_reg); /* clear RHB */
			}

			phy_reg = get_phy_reg(ohci, 5);
			phy_reg |= 0x40;
			set_phy_reg(ohci, 5, phy_reg); /* set ISBR */
			break;
		case LONG_RESET_NO_FORCE_ROOT:
			phy_reg = get_phy_reg(ohci, 1);
			phy_reg &= ~0x80;
			phy_reg |= 0x40;
			set_phy_reg(ohci, 1, phy_reg); /* clear RHB, set IBR */
			break;
		case SHORT_RESET_FORCE_ROOT:
			phy_reg = get_phy_reg(ohci, 1);
			if (!(phy_reg & 0x80)) {
				phy_reg |= 0x80;
				set_phy_reg(ohci, 1, phy_reg); /* set RHB */
			}

			phy_reg = get_phy_reg(ohci, 5);
			phy_reg |= 0x40;
			set_phy_reg(ohci, 5, phy_reg); /* set ISBR */
			break;
		case LONG_RESET_FORCE_ROOT:
			phy_reg = get_phy_reg(ohci, 1);
			phy_reg |= 0xc0;
			set_phy_reg(ohci, 1, phy_reg); /* set RHB and IBR */
			break;
		default:
			retval = -1;
		}
		break;

	case GET_CYCLE_COUNTER:
		retval = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
		break;

	case SET_CYCLE_COUNTER:
		reg_write(ohci, OHCI1394_IsochronousCycleTimer, arg);
		break;

	case SET_BUS_ID:
		PRINT(KERN_ERR, "devctl command SET_BUS_ID err");
		break;

	case ACT_CYCLE_MASTER:
		if (arg) {
			/* check if we are root and other nodes are present */
			u32 nodeId = reg_read(ohci, OHCI1394_NodeID);
			if ((nodeId & (1<<30)) && (nodeId & 0x3f)) {
				/*
				 * enable cycleTimer, cycleMaster
				 */
				DBGMSG("Cycle master enabled");
				reg_write(ohci, OHCI1394_LinkControlSet,
					  OHCI1394_LinkControl_CycleTimerEnable |
					  OHCI1394_LinkControl_CycleMaster);
			}
		} else {
			/* disable cycleTimer, cycleMaster, cycleSource */
			reg_write(ohci, OHCI1394_LinkControlClear,
				  OHCI1394_LinkControl_CycleTimerEnable |
				  OHCI1394_LinkControl_CycleMaster |
				  OHCI1394_LinkControl_CycleSource);
		}
		break;

	case CANCEL_REQUESTS:
		DBGMSG("Cancel request received");
		dma_trm_reset(&ohci->at_req_context);
		dma_trm_reset(&ohci->at_resp_context);
		break;

	case ISO_LISTEN_CHANNEL:
        {
		u64 mask;
		struct dma_rcv_ctx *d = &ohci->ir_legacy_context;
		int ir_legacy_active;

		if (arg<0 || arg>63) {
			PRINT(KERN_ERR,
			      "%s: IS0 listen channel %d is out of range",
			      __FUNCTION__, arg);
			return -EFAULT;
		}

		mask = (u64)0x1<<arg;

                spin_lock_irqsave(&ohci->IR_channel_lock, flags);

		if (ohci->ISO_channel_usage & mask) {
			PRINT(KERN_ERR,
			      "%s: IS0 listen channel %d is already used",
			      __FUNCTION__, arg);
			spin_unlock_irqrestore(&ohci->IR_channel_lock, flags);
			return -EFAULT;
		}

		ir_legacy_active = ohci->ir_legacy_channels;

		ohci->ISO_channel_usage |= mask;
		ohci->ir_legacy_channels |= mask;

                spin_unlock_irqrestore(&ohci->IR_channel_lock, flags);

		if (!ir_legacy_active) {
			if (ohci1394_register_iso_tasklet(ohci,
					  &ohci->ir_legacy_tasklet) < 0) {
				PRINT(KERN_ERR, "No IR DMA context available");
				return -EBUSY;
			}

			/* the IR context can be assigned to any DMA context
			 * by ohci1394_register_iso_tasklet */
			d->ctx = ohci->ir_legacy_tasklet.context;
			d->ctrlSet = OHCI1394_IsoRcvContextControlSet +
				32*d->ctx;
			d->ctrlClear = OHCI1394_IsoRcvContextControlClear +
				32*d->ctx;
			d->cmdPtr = OHCI1394_IsoRcvCommandPtr + 32*d->ctx;
			d->ctxtMatch = OHCI1394_IsoRcvContextMatch + 32*d->ctx;

			initialize_dma_rcv_ctx(&ohci->ir_legacy_context, 1);

			if (printk_ratelimit())
				DBGMSG("IR legacy activated");
		}

                spin_lock_irqsave(&ohci->IR_channel_lock, flags);

		if (arg>31)
			reg_write(ohci, OHCI1394_IRMultiChanMaskHiSet,
				  1<<(arg-32));
		else
			reg_write(ohci, OHCI1394_IRMultiChanMaskLoSet,
				  1<<arg);

                spin_unlock_irqrestore(&ohci->IR_channel_lock, flags);
                DBGMSG("Listening enabled on channel %d", arg);
                break;
        }
	case ISO_UNLISTEN_CHANNEL:
        {
		u64 mask;

		if (arg<0 || arg>63) {
			PRINT(KERN_ERR,
			      "%s: IS0 unlisten channel %d is out of range",
			      __FUNCTION__, arg);
			return -EFAULT;
		}

		mask = (u64)0x1<<arg;

                spin_lock_irqsave(&ohci->IR_channel_lock, flags);

		if (!(ohci->ISO_channel_usage & mask)) {
			PRINT(KERN_ERR,
			      "%s: IS0 unlisten channel %d is not used",
			      __FUNCTION__, arg);
			spin_unlock_irqrestore(&ohci->IR_channel_lock, flags);
			return -EFAULT;
		}

		ohci->ISO_channel_usage &= ~mask;
		ohci->ir_legacy_channels &= ~mask;

		if (arg>31)
			reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear,
				  1<<(arg-32));
		else
			reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear,
				  1<<arg);

                spin_unlock_irqrestore(&ohci->IR_channel_lock, flags);
                DBGMSG("Listening disabled on channel %d", arg);

		if (ohci->ir_legacy_channels == 0) {
			stop_dma_rcv_ctx(&ohci->ir_legacy_context);
			DBGMSG("ISO legacy receive context stopped");
		}

                break;
        }
	default:
		PRINT_G(KERN_ERR, "ohci_devctl cmd %d not implemented yet",
			cmd);
		break;
	}
	return retval;
}

/***********************************
 * rawiso ISO reception            *
 ***********************************/

/*
  We use either buffer-fill or packet-per-buffer DMA mode. The DMA
  buffer is split into "blocks" (regions described by one DMA
  descriptor). Each block must be one page or less in size, and
  must not cross a page boundary.

  There is one little wrinkle with buffer-fill mode: a packet that
  starts in the final block may wrap around into the first block. But
  the user API expects all packets to be contiguous. Our solution is
  to keep the very last page of the DMA buffer in reserve - if a
  packet spans the gap, we copy its tail into this page.
*/

struct ohci_iso_recv {
	struct ti_ohci *ohci;

	struct ohci1394_iso_tasklet task;
	int task_active;

	enum { BUFFER_FILL_MODE = 0,
	       PACKET_PER_BUFFER_MODE = 1 } dma_mode;

	/* memory and PCI mapping for the DMA descriptors */
	struct dma_prog_region prog;
	struct dma_cmd *block; /* = (struct dma_cmd*) prog.virt */

	/* how many DMA blocks fit in the buffer */
	unsigned int nblocks;

	/* stride of DMA blocks */
	unsigned int buf_stride;

	/* number of blocks to batch between interrupts */
	int block_irq_interval;

	/* block that DMA will finish next */
	int block_dma;

	/* (buffer-fill only) block that the reader will release next */
	int block_reader;

	/* (buffer-fill only) bytes of buffer the reader has released,
	   less than one block */
	int released_bytes;

	/* (buffer-fill only) buffer offset at which the next packet will appear */
	int dma_offset;

	/* OHCI DMA context control registers */
	u32 ContextControlSet;
	u32 ContextControlClear;
	u32 CommandPtr;
	u32 ContextMatch;
};

static void ohci_iso_recv_task(unsigned long data);
static void ohci_iso_recv_stop(struct hpsb_iso *iso);
static void ohci_iso_recv_shutdown(struct hpsb_iso *iso);
static int  ohci_iso_recv_start(struct hpsb_iso *iso, int cycle, int tag_mask, int sync);
static void ohci_iso_recv_program(struct hpsb_iso *iso);

static int ohci_iso_recv_init(struct hpsb_iso *iso)
{
	struct ti_ohci *ohci = iso->host->hostdata;
	struct ohci_iso_recv *recv;
	int ctx;
	int ret = -ENOMEM;

	recv = kmalloc(sizeof(*recv), GFP_KERNEL);
	if (!recv)
		return -ENOMEM;

	iso->hostdata = recv;
	recv->ohci = ohci;
	recv->task_active = 0;
	dma_prog_region_init(&recv->prog);
	recv->block = NULL;

	/* use buffer-fill mode, unless irq_interval is 1
	   (note: multichannel requires buffer-fill) */

	if (((iso->irq_interval == 1 && iso->dma_mode == HPSB_ISO_DMA_OLD_ABI) ||
	     iso->dma_mode == HPSB_ISO_DMA_PACKET_PER_BUFFER) && iso->channel != -1) {
		recv->dma_mode = PACKET_PER_BUFFER_MODE;
	} else {
		recv->dma_mode = BUFFER_FILL_MODE;
	}

	/* set nblocks, buf_stride, block_irq_interval */

	if (recv->dma_mode == BUFFER_FILL_MODE) {
		recv->buf_stride = PAGE_SIZE;

		/* one block per page of data in the DMA buffer, minus the final guard page */
		recv->nblocks = iso->buf_size/PAGE_SIZE - 1;
		if (recv->nblocks < 3) {
			DBGMSG("ohci_iso_recv_init: DMA buffer too small");
			goto err;
		}

		/* iso->irq_interval is in packets - translate that to blocks */
		if (iso->irq_interval == 1)
			recv->block_irq_interval = 1;
		else
			recv->block_irq_interval = iso->irq_interval *
							((recv->nblocks+1)/iso->buf_packets);
		if (recv->block_irq_interval*4 > recv->nblocks)
			recv->block_irq_interval = recv->nblocks/4;
		if (recv->block_irq_interval < 1)
			recv->block_irq_interval = 1;

	} else {
		int max_packet_size;

		recv->nblocks = iso->buf_packets;
		recv->block_irq_interval = iso->irq_interval;
		if (recv->block_irq_interval * 4 > iso->buf_packets)
			recv->block_irq_interval = iso->buf_packets / 4;
		if (recv->block_irq_interval < 1)
		recv->block_irq_interval = 1;

		/* choose a buffer stride */
		/* must be a power of 2, and <= PAGE_SIZE */

		max_packet_size = iso->buf_size / iso->buf_packets;

		for (recv->buf_stride = 8; recv->buf_stride < max_packet_size;
		    recv->buf_stride *= 2);

		if (recv->buf_stride*iso->buf_packets > iso->buf_size ||
		   recv->buf_stride > PAGE_SIZE) {
			/* this shouldn't happen, but anyway... */
			DBGMSG("ohci_iso_recv_init: problem choosing a buffer stride");
			goto err;
		}
	}

	recv->block_reader = 0;
	recv->released_bytes = 0;
	recv->block_dma = 0;
	recv->dma_offset = 0;

	/* size of DMA program = one descriptor per block */
	if (dma_prog_region_alloc(&recv->prog,
				 sizeof(struct dma_cmd) * recv->nblocks,
				 recv->ohci->dev))
		goto err;

	recv->block = (struct dma_cmd*) recv->prog.kvirt;

	ohci1394_init_iso_tasklet(&recv->task,
				  iso->channel == -1 ? OHCI_ISO_MULTICHANNEL_RECEIVE :
				                       OHCI_ISO_RECEIVE,
				  ohci_iso_recv_task, (unsigned long) iso);

	if (ohci1394_register_iso_tasklet(recv->ohci, &recv->task) < 0) {
		ret = -EBUSY;
		goto err;
	}

	recv->task_active = 1;

	/* recv context registers are spaced 32 bytes apart */
	ctx = recv->task.context;
	recv->ContextControlSet = OHCI1394_IsoRcvContextControlSet + 32 * ctx;
	recv->ContextControlClear = OHCI1394_IsoRcvContextControlClear + 32 * ctx;
	recv->CommandPtr = OHCI1394_IsoRcvCommandPtr + 32 * ctx;
	recv->ContextMatch = OHCI1394_IsoRcvContextMatch + 32 * ctx;

	if (iso->channel == -1) {
		/* clear multi-channel selection mask */
		reg_write(recv->ohci, OHCI1394_IRMultiChanMaskHiClear, 0xFFFFFFFF);
		reg_write(recv->ohci, OHCI1394_IRMultiChanMaskLoClear, 0xFFFFFFFF);
	}

	/* write the DMA program */
	ohci_iso_recv_program(iso);

	DBGMSG("ohci_iso_recv_init: %s mode, DMA buffer is %lu pages"
	       " (%u bytes), using %u blocks, buf_stride %u, block_irq_interval %d",
	       recv->dma_mode == BUFFER_FILL_MODE ?
	       "buffer-fill" : "packet-per-buffer",
	       iso->buf_size/PAGE_SIZE, iso->buf_size,
	       recv->nblocks, recv->buf_stride, recv->block_irq_interval);

	return 0;

err:
	ohci_iso_recv_shutdown(iso);
	return ret;
}

static void ohci_iso_recv_stop(struct hpsb_iso *iso)
{
	struct ohci_iso_recv *recv = iso->hostdata;

	/* disable interrupts */
	reg_write(recv->ohci, OHCI1394_IsoRecvIntMaskClear, 1 << recv->task.context);

	/* halt DMA */
	ohci1394_stop_context(recv->ohci, recv->ContextControlClear, NULL);
}

static void ohci_iso_recv_shutdown(struct hpsb_iso *iso)
{
	struct ohci_iso_recv *recv = iso->hostdata;

	if (recv->task_active) {
		ohci_iso_recv_stop(iso);
		ohci1394_unregister_iso_tasklet(recv->ohci, &recv->task);
		recv->task_active = 0;
	}

	dma_prog_region_free(&recv->prog);
	kfree(recv);
	iso->hostdata = NULL;
}

/* set up a "gapped" ring buffer DMA program */
static void ohci_iso_recv_program(struct hpsb_iso *iso)
{
	struct ohci_iso_recv *recv = iso->hostdata;
	int blk;

	/* address of 'branch' field in previous DMA descriptor */
	u32 *prev_branch = NULL;

	for (blk = 0; blk < recv->nblocks; blk++) {
		u32 control;

		/* the DMA descriptor */
		struct dma_cmd *cmd = &recv->block[blk];

		/* offset of the DMA descriptor relative to the DMA prog buffer */
		unsigned long prog_offset = blk * sizeof(struct dma_cmd);

		/* offset of this packet's data within the DMA buffer */
		unsigned long buf_offset = blk * recv->buf_stride;

		if (recv->dma_mode == BUFFER_FILL_MODE) {
			control = 2 << 28; /* INPUT_MORE */
		} else {
			control = 3 << 28; /* INPUT_LAST */
		}

		control |= 8 << 24; /* s = 1, update xferStatus and resCount */

		/* interrupt on last block, and at intervals */
		if (blk == recv->nblocks-1 || (blk % recv->block_irq_interval) == 0) {
			control |= 3 << 20; /* want interrupt */
		}

		control |= 3 << 18; /* enable branch to address */
		control |= recv->buf_stride;

		cmd->control = cpu_to_le32(control);
		cmd->address = cpu_to_le32(dma_region_offset_to_bus(&iso->data_buf, buf_offset));
		cmd->branchAddress = 0; /* filled in on next loop */
		cmd->status = cpu_to_le32(recv->buf_stride);

		/* link the previous descriptor to this one */
		if (prev_branch) {
			*prev_branch = cpu_to_le32(dma_prog_region_offset_to_bus(&recv->prog, prog_offset) | 1);
		}

		prev_branch = &cmd->branchAddress;
	}

	/* the final descriptor's branch address and Z should be left at 0 */
}

/* listen or unlisten to a specific channel (multi-channel mode only) */
static void ohci_iso_recv_change_channel(struct hpsb_iso *iso, unsigned char channel, int listen)
{
	struct ohci_iso_recv *recv = iso->hostdata;
	int reg, i;

	if (channel < 32) {
		reg = listen ? OHCI1394_IRMultiChanMaskLoSet : OHCI1394_IRMultiChanMaskLoClear;
		i = channel;
	} else {
		reg = listen ? OHCI1394_IRMultiChanMaskHiSet : OHCI1394_IRMultiChanMaskHiClear;
		i = channel - 32;
	}

	reg_write(recv->ohci, reg, (1 << i));

	/* issue a dummy read to force all PCI writes to be posted immediately */
	mb();
	reg_read(recv->ohci, OHCI1394_IsochronousCycleTimer);
}

static void ohci_iso_recv_set_channel_mask(struct hpsb_iso *iso, u64 mask)
{
	struct ohci_iso_recv *recv = iso->hostdata;
	int i;

	for (i = 0; i < 64; i++) {
		if (mask & (1ULL << i)) {
			if (i < 32)
				reg_write(recv->ohci, OHCI1394_IRMultiChanMaskLoSet, (1 << i));
			else
				reg_write(recv->ohci, OHCI1394_IRMultiChanMaskHiSet, (1 << (i-32)));
		} else {
			if (i < 32)
				reg_write(recv->ohci, OHCI1394_IRMultiChanMaskLoClear, (1 << i));
			else
				reg_write(recv->ohci, OHCI1394_IRMultiChanMaskHiClear, (1 << (i-32)));
		}
	}

	/* issue a dummy read to force all PCI writes to be posted immediately */
	mb();
	reg_read(recv->ohci, OHCI1394_IsochronousCycleTimer);
}

static int ohci_iso_recv_start(struct hpsb_iso *iso, int cycle, int tag_mask, int sync)
{
	struct ohci_iso_recv *recv = iso->hostdata;
	struct ti_ohci *ohci = recv->ohci;
	u32 command, contextMatch;

	reg_write(recv->ohci, recv->ContextControlClear, 0xFFFFFFFF);
	wmb();

	/* always keep ISO headers */
	command = (1 << 30);

	if (recv->dma_mode == BUFFER_FILL_MODE)
		command |= (1 << 31);

	reg_write(recv->ohci, recv->ContextControlSet, command);

	/* match on specified tags */
	contextMatch = tag_mask << 28;

	if (iso->channel == -1) {
		/* enable multichannel reception */
		reg_write(recv->ohci, recv->ContextControlSet, (1 << 28));
	} else {
		/* listen on channel */
		contextMatch |= iso->channel;
	}

	if (cycle != -1) {
		u32 seconds;

		/* enable cycleMatch */
		reg_write(recv->ohci, recv->ContextControlSet, (1 << 29));

		/* set starting cycle */
		cycle &= 0x1FFF;

		/* 'cycle' is only mod 8000, but we also need two 'seconds' bits -
		   just snarf them from the current time */
		seconds = reg_read(recv->ohci, OHCI1394_IsochronousCycleTimer) >> 25;

		/* advance one second to give some extra time for DMA to start */
		seconds += 1;

		cycle |= (seconds & 3) << 13;

		contextMatch |= cycle << 12;
	}

	if (sync != -1) {
		/* set sync flag on first DMA descriptor */
		struct dma_cmd *cmd = &recv->block[recv->block_dma];
		cmd->control |= cpu_to_le32(DMA_CTL_WAIT);

		/* match sync field */
		contextMatch |= (sync&0xf)<<8;
	}

	reg_write(recv->ohci, recv->ContextMatch, contextMatch);

	/* address of first descriptor block */
	command = dma_prog_region_offset_to_bus(&recv->prog,
						recv->block_dma * sizeof(struct dma_cmd));
	command |= 1; /* Z=1 */

	reg_write(recv->ohci, recv->CommandPtr, command);

	/* enable interrupts */
	reg_write(recv->ohci, OHCI1394_IsoRecvIntMaskSet, 1 << recv->task.context);

	wmb();

	/* run */
	reg_write(recv->ohci, recv->ContextControlSet, 0x8000);

	/* issue a dummy read of the cycle timer register to force
	   all PCI writes to be posted immediately */
	mb();
	reg_read(recv->ohci, OHCI1394_IsochronousCycleTimer);

	/* check RUN */
	if (!(reg_read(recv->ohci, recv->ContextControlSet) & 0x8000)) {
		PRINT(KERN_ERR,
		      "Error starting IR DMA (ContextControl 0x%08x)\n",
		      reg_read(recv->ohci, recv->ContextControlSet));
		return -1;
	}

	return 0;
}

static void ohci_iso_recv_release_block(struct ohci_iso_recv *recv, int block)
{
	/* re-use the DMA descriptor for the block */
	/* by linking the previous descriptor to it */

	int next_i = block;
	int prev_i = (next_i == 0) ? (recv->nblocks - 1) : (next_i - 1);

	struct dma_cmd *next = &recv->block[next_i];
	struct dma_cmd *prev = &recv->block[prev_i];
	
	/* ignore out-of-range requests */
	if ((block < 0) || (block > recv->nblocks))
		return;

	/* 'next' becomes the new end of the DMA chain,
	   so disable branch and enable interrupt */
	next->branchAddress = 0;
	next->control |= cpu_to_le32(3 << 20);
	next->status = cpu_to_le32(recv->buf_stride);

	/* link prev to next */
	prev->branchAddress = cpu_to_le32(dma_prog_region_offset_to_bus(&recv->prog,
									sizeof(struct dma_cmd) * next_i)
					  | 1); /* Z=1 */

	/* disable interrupt on previous DMA descriptor, except at intervals */
	if ((prev_i % recv->block_irq_interval) == 0) {
		prev->control |= cpu_to_le32(3 << 20); /* enable interrupt */
	} else {
		prev->control &= cpu_to_le32(~(3<<20)); /* disable interrupt */
	}
	wmb();

	/* wake up DMA in case it fell asleep */
	reg_write(recv->ohci, recv->ContextControlSet, (1 << 12));
}

static void ohci_iso_recv_bufferfill_release(struct ohci_iso_recv *recv,
					     struct hpsb_iso_packet_info *info)
{
	/* release the memory where the packet was */
	recv->released_bytes += info->total_len;

	/* have we released enough memory for one block? */
	while (recv->released_bytes > recv->buf_stride) {
		ohci_iso_recv_release_block(recv, recv->block_reader);
		recv->block_reader = (recv->block_reader + 1) % recv->nblocks;
		recv->released_bytes -= recv->buf_stride;
	}
}

static inline void ohci_iso_recv_release(struct hpsb_iso *iso, struct hpsb_iso_packet_info *info)
{
	struct ohci_iso_recv *recv = iso->hostdata;
	if (recv->dma_mode == BUFFER_FILL_MODE) {
		ohci_iso_recv_bufferfill_release(recv, info);
	} else {
		ohci_iso_recv_release_block(recv, info - iso->infos);
	}
}

/* parse all packets from blocks that have been fully received */
static void ohci_iso_recv_bufferfill_parse(struct hpsb_iso *iso, struct ohci_iso_recv *recv)
{
	int wake = 0;
	int runaway = 0;
	struct ti_ohci *ohci = recv->ohci;

	while (1) {
		/* we expect the next parsable packet to begin at recv->dma_offset */
		/* note: packet layout is as shown in section 10.6.1.1 of the OHCI spec */

		unsigned int offset;
		unsigned short len, cycle, total_len;
		unsigned char channel, tag, sy;

		unsigned char *p = iso->data_buf.kvirt;

		unsigned int this_block = recv->dma_offset/recv->buf_stride;

		/* don't loop indefinitely */
		if (runaway++ > 100000) {
			atomic_inc(&iso->overflows);
			PRINT(KERN_ERR,
			      "IR DMA error - Runaway during buffer parsing!\n");
			break;
		}

		/* stop parsing once we arrive at block_dma (i.e. don't get ahead of DMA) */
		if (this_block == recv->block_dma)
			break;

		wake = 1;

		/* parse data length, tag, channel, and sy */

		/* note: we keep our own local copies of 'len' and 'offset'
		   so the user can't mess with them by poking in the mmap area */

		len = p[recv->dma_offset+2] | (p[recv->dma_offset+3] << 8);

		if (len > 4096) {
			PRINT(KERN_ERR,
			      "IR DMA error - bogus 'len' value %u\n", len);
		}

		channel = p[recv->dma_offset+1] & 0x3F;
		tag = p[recv->dma_offset+1] >> 6;
		sy = p[recv->dma_offset+0] & 0xF;

		/* advance to data payload */
		recv->dma_offset += 4;

		/* check for wrap-around */
		if (recv->dma_offset >= recv->buf_stride*recv->nblocks) {
			recv->dma_offset -= recv->buf_stride*recv->nblocks;
		}

		/* dma_offset now points to the first byte of the data payload */
		offset = recv->dma_offset;

		/* advance to xferStatus/timeStamp */
		recv->dma_offset += len;

		total_len = len + 8; /* 8 bytes header+trailer in OHCI packet */
		/* payload is padded to 4 bytes */
		if (len % 4) {
			recv->dma_offset += 4 - (len%4);
			total_len += 4 - (len%4);
		}

		/* check for wrap-around */
		if (recv->dma_offset >= recv->buf_stride*recv->nblocks) {
			/* uh oh, the packet data wraps from the last
                           to the first DMA block - make the packet
                           contiguous by copying its "tail" into the
                           guard page */

			int guard_off = recv->buf_stride*recv->nblocks;
			int tail_len = len - (guard_off - offset);

			if (tail_len > 0  && tail_len < recv->buf_stride) {
				memcpy(iso->data_buf.kvirt + guard_off,
				       iso->data_buf.kvirt,
				       tail_len);
			}

			recv->dma_offset -= recv->buf_stride*recv->nblocks;
		}

		/* parse timestamp */
		cycle = p[recv->dma_offset+0] | (p[recv->dma_offset+1]<<8);
		cycle &= 0x1FFF;

		/* advance to next packet */
		recv->dma_offset += 4;

		/* check for wrap-around */
		if (recv->dma_offset >= recv->buf_stride*recv->nblocks) {
			recv->dma_offset -= recv->buf_stride*recv->nblocks;
		}

		hpsb_iso_packet_received(iso, offset, len, total_len, cycle, channel, tag, sy);
	}

	if (wake)
		hpsb_iso_wake(iso);
}

static void ohci_iso_recv_bufferfill_task(struct hpsb_iso *iso, struct ohci_iso_recv *recv)
{
	int loop;
	struct ti_ohci *ohci = recv->ohci;

	/* loop over all blocks */
	for (loop = 0; loop < recv->nblocks; loop++) {

		/* check block_dma to see if it's done */
		struct dma_cmd *im = &recv->block[recv->block_dma];

		/* check the DMA descriptor for new writes to xferStatus */
		u16 xferstatus = le32_to_cpu(im->status) >> 16;

		/* rescount is the number of bytes *remaining to be written* in the block */
		u16 rescount = le32_to_cpu(im->status) & 0xFFFF;

		unsigned char event = xferstatus & 0x1F;

		if (!event) {
			/* nothing has happened to this block yet */
			break;
		}

		if (event != 0x11) {
			atomic_inc(&iso->overflows);
			PRINT(KERN_ERR,
			      "IR DMA error - OHCI error code 0x%02x\n", event);
		}

		if (rescount != 0) {
			/* the card is still writing to this block;
			   we can't touch it until it's done */
			break;
		}

		/* OK, the block is finished... */

		/* sync our view of the block */
		dma_region_sync_for_cpu(&iso->data_buf, recv->block_dma*recv->buf_stride, recv->buf_stride);

		/* reset the DMA descriptor */
		im->status = recv->buf_stride;

		/* advance block_dma */
		recv->block_dma = (recv->block_dma + 1) % recv->nblocks;

		if ((recv->block_dma+1) % recv->nblocks == recv->block_reader) {
			atomic_inc(&iso->overflows);
			DBGMSG("ISO reception overflow - "
			       "ran out of DMA blocks");
		}
	}

	/* parse any packets that have arrived */
	ohci_iso_recv_bufferfill_parse(iso, recv);
}

static void ohci_iso_recv_packetperbuf_task(struct hpsb_iso *iso, struct ohci_iso_recv *recv)
{
	int count;
	int wake = 0;
	struct ti_ohci *ohci = recv->ohci;

	/* loop over the entire buffer */
	for (count = 0; count < recv->nblocks; count++) {
		u32 packet_len = 0;

		/* pointer to the DMA descriptor */
		struct dma_cmd *il = ((struct dma_cmd*) recv->prog.kvirt) + iso->pkt_dma;

		/* check the DMA descriptor for new writes to xferStatus */
		u16 xferstatus = le32_to_cpu(il->status) >> 16;
		u16 rescount = le32_to_cpu(il->status) & 0xFFFF;

		unsigned char event = xferstatus & 0x1F;

		if (!event) {
			/* this packet hasn't come in yet; we are done for now */
			goto out;
		}

		if (event == 0x11) {
			/* packet received successfully! */

			/* rescount is the number of bytes *remaining* in the packet buffer,
			   after the packet was written */
			packet_len = recv->buf_stride - rescount;

		} else if (event == 0x02) {
			PRINT(KERN_ERR, "IR DMA error - packet too long for buffer\n");
		} else if (event) {
			PRINT(KERN_ERR, "IR DMA error - OHCI error code 0x%02x\n", event);
		}

		/* sync our view of the buffer */
		dma_region_sync_for_cpu(&iso->data_buf, iso->pkt_dma * recv->buf_stride, recv->buf_stride);

		/* record the per-packet info */
		{
			/* iso header is 8 bytes ahead of the data payload */
			unsigned char *hdr;

			unsigned int offset;
			unsigned short cycle;
			unsigned char channel, tag, sy;

			offset = iso->pkt_dma * recv->buf_stride;
			hdr = iso->data_buf.kvirt + offset;

			/* skip iso header */
			offset += 8;
			packet_len -= 8;

			cycle = (hdr[0] | (hdr[1] << 8)) & 0x1FFF;
			channel = hdr[5] & 0x3F;
			tag = hdr[5] >> 6;
			sy = hdr[4] & 0xF;

			hpsb_iso_packet_received(iso, offset, packet_len,
					recv->buf_stride, cycle, channel, tag, sy);
		}

		/* reset the DMA descriptor */
		il->status = recv->buf_stride;

		wake = 1;
		recv->block_dma = iso->pkt_dma;
	}

out:
	if (wake)
		hpsb_iso_wake(iso);
}

static void ohci_iso_recv_task(unsigned long data)
{
	struct hpsb_iso *iso = (struct hpsb_iso*) data;
	struct ohci_iso_recv *recv = iso->hostdata;

	if (recv->dma_mode == BUFFER_FILL_MODE)
		ohci_iso_recv_bufferfill_task(iso, recv);
	else
		ohci_iso_recv_packetperbuf_task(iso, recv);
}

/***********************************
 * rawiso ISO transmission         *
 ***********************************/

struct ohci_iso_xmit {
	struct ti_ohci *ohci;
	struct dma_prog_region prog;
	struct ohci1394_iso_tasklet task;
	int task_active;

	u32 ContextControlSet;
	u32 ContextControlClear;
	u32 CommandPtr;
};

/* transmission DMA program:
   one OUTPUT_MORE_IMMEDIATE for the IT header
   one OUTPUT_LAST for the buffer data */

struct iso_xmit_cmd {
	struct dma_cmd output_more_immediate;
	u8 iso_hdr[8];
	u32 unused[2];
	struct dma_cmd output_last;
};

static int ohci_iso_xmit_init(struct hpsb_iso *iso);
static int ohci_iso_xmit_start(struct hpsb_iso *iso, int cycle);
static void ohci_iso_xmit_shutdown(struct hpsb_iso *iso);
static void ohci_iso_xmit_task(unsigned long data);

static int ohci_iso_xmit_init(struct hpsb_iso *iso)
{
	struct ohci_iso_xmit *xmit;
	unsigned int prog_size;
	int ctx;
	int ret = -ENOMEM;

	xmit = kmalloc(sizeof(*xmit), GFP_KERNEL);
	if (!xmit)
		return -ENOMEM;

	iso->hostdata = xmit;
	xmit->ohci = iso->host->hostdata;
	xmit->task_active = 0;

	dma_prog_region_init(&xmit->prog);

	prog_size = sizeof(struct iso_xmit_cmd) * iso->buf_packets;

	if (dma_prog_region_alloc(&xmit->prog, prog_size, xmit->ohci->dev))
		goto err;

	ohci1394_init_iso_tasklet(&xmit->task, OHCI_ISO_TRANSMIT,
				  ohci_iso_xmit_task, (unsigned long) iso);

	if (ohci1394_register_iso_tasklet(xmit->ohci, &xmit->task) < 0) {
		ret = -EBUSY;
		goto err;
	}

	xmit->task_active = 1;

	/* xmit context registers are spaced 16 bytes apart */
	ctx = xmit->task.context;
	xmit->ContextControlSet = OHCI1394_IsoXmitContextControlSet + 16 * ctx;
	xmit->ContextControlClear = OHCI1394_IsoXmitContextControlClear + 16 * ctx;
	xmit->CommandPtr = OHCI1394_IsoXmitCommandPtr + 16 * ctx;

	return 0;

err:
	ohci_iso_xmit_shutdown(iso);
	return ret;
}

static void ohci_iso_xmit_stop(struct hpsb_iso *iso)
{
	struct ohci_iso_xmit *xmit = iso->hostdata;
	struct ti_ohci *ohci = xmit->ohci;

	/* disable interrupts */
	reg_write(xmit->ohci, OHCI1394_IsoXmitIntMaskClear, 1 << xmit->task.context);

	/* halt DMA */
	if (ohci1394_stop_context(xmit->ohci, xmit->ContextControlClear, NULL)) {
		/* XXX the DMA context will lock up if you try to send too much data! */
		PRINT(KERN_ERR,
		      "you probably exceeded the OHCI card's bandwidth limit - "
		      "reload the module and reduce xmit bandwidth");
	}
}

static void ohci_iso_xmit_shutdown(struct hpsb_iso *iso)
{
	struct ohci_iso_xmit *xmit = iso->hostdata;

	if (xmit->task_active) {
		ohci_iso_xmit_stop(iso);
		ohci1394_unregister_iso_tasklet(xmit->ohci, &xmit->task);
		xmit->task_active = 0;
	}

	dma_prog_region_free(&xmit->prog);
	kfree(xmit);
	iso->hostdata = NULL;
}

static void ohci_iso_xmit_task(unsigned long data)
{
	struct hpsb_iso *iso = (struct hpsb_iso*) data;
	struct ohci_iso_xmit *xmit = iso->hostdata;
	struct ti_ohci *ohci = xmit->ohci;
	int wake = 0;
	int count;

	/* check the whole buffer if necessary, starting at pkt_dma */
	for (count = 0; count < iso->buf_packets; count++) {
		int cycle;

		/* DMA descriptor */
		struct iso_xmit_cmd *cmd = dma_region_i(&xmit->prog, struct iso_xmit_cmd, iso->pkt_dma);

		/* check for new writes to xferStatus */
		u16 xferstatus = le32_to_cpu(cmd->output_last.status) >> 16;
		u8  event = xferstatus & 0x1F;

		if (!event) {
			/* packet hasn't been sent yet; we are done for now */
			break;
		}

		if (event != 0x11)
			PRINT(KERN_ERR,
			      "IT DMA error - OHCI error code 0x%02x\n", event);

		/* at least one packet went out, so wake up the writer */
		wake = 1;

		/* parse cycle */
		cycle = le32_to_cpu(cmd->output_last.status) & 0x1FFF;

		/* tell the subsystem the packet has gone out */
		hpsb_iso_packet_sent(iso, cycle, event != 0x11);

		/* reset the DMA descriptor for next time */
		cmd->output_last.status = 0;
	}

	if (wake)
		hpsb_iso_wake(iso);
}

static int ohci_iso_xmit_queue(struct hpsb_iso *iso, struct hpsb_iso_packet_info *info)
{
	struct ohci_iso_xmit *xmit = iso->hostdata;
	struct ti_ohci *ohci = xmit->ohci;

	int next_i, prev_i;
	struct iso_xmit_cmd *next, *prev;

	unsigned int offset;
	unsigned short len;
	unsigned char tag, sy;

	/* check that the packet doesn't cross a page boundary
	   (we could allow this if we added OUTPUT_MORE descriptor support) */
	if (cross_bound(info->offset, info->len)) {
		PRINT(KERN_ERR,
		      "rawiso xmit: packet %u crosses a page boundary",
		      iso->first_packet);
		return -EINVAL;
	}

	offset = info->offset;
	len = info->len;
	tag = info->tag;
	sy = info->sy;

	/* sync up the card's view of the buffer */
	dma_region_sync_for_device(&iso->data_buf, offset, len);

	/* append first_packet to the DMA chain */
	/* by linking the previous descriptor to it */
	/* (next will become the new end of the DMA chain) */

	next_i = iso->first_packet;
	prev_i = (next_i == 0) ? (iso->buf_packets - 1) : (next_i - 1);

	next = dma_region_i(&xmit->prog, struct iso_xmit_cmd, next_i);
	prev = dma_region_i(&xmit->prog, struct iso_xmit_cmd, prev_i);

	/* set up the OUTPUT_MORE_IMMEDIATE descriptor */
	memset(next, 0, sizeof(struct iso_xmit_cmd));
	next->output_more_immediate.control = cpu_to_le32(0x02000008);

	/* ISO packet header is embedded in the OUTPUT_MORE_IMMEDIATE */

	/* tcode = 0xA, and sy */
	next->iso_hdr[0] = 0xA0 | (sy & 0xF);

	/* tag and channel number */
	next->iso_hdr[1] = (tag << 6) | (iso->channel & 0x3F);

	/* transmission speed */
	next->iso_hdr[2] = iso->speed & 0x7;

	/* payload size */
	next->iso_hdr[6] = len & 0xFF;
	next->iso_hdr[7] = len >> 8;

	/* set up the OUTPUT_LAST */
	next->output_last.control = cpu_to_le32(1 << 28);
	next->output_last.control |= cpu_to_le32(1 << 27); /* update timeStamp */
	next->output_last.control |= cpu_to_le32(3 << 20); /* want interrupt */
	next->output_last.control |= cpu_to_le32(3 << 18); /* enable branch */
	next->output_last.control |= cpu_to_le32(len);

	/* payload bus address */
	next->output_last.address = cpu_to_le32(dma_region_offset_to_bus(&iso->data_buf, offset));

	/* leave branchAddress at zero for now */

	/* re-write the previous DMA descriptor to chain to this one */

	/* set prev branch address to point to next (Z=3) */
	prev->output_last.branchAddress = cpu_to_le32(
		dma_prog_region_offset_to_bus(&xmit->prog, sizeof(struct iso_xmit_cmd) * next_i) | 3);

	/* disable interrupt, unless required by the IRQ interval */
	if (prev_i % iso->irq_interval) {
		prev->output_last.control &= cpu_to_le32(~(3 << 20)); /* no interrupt */
	} else {
		prev->output_last.control |= cpu_to_le32(3 << 20); /* enable interrupt */
	}

	wmb();

	/* wake DMA in case it is sleeping */
	reg_write(xmit->ohci, xmit->ContextControlSet, 1 << 12);

	/* issue a dummy read of the cycle timer to force all PCI
	   writes to be posted immediately */
	mb();
	reg_read(xmit->ohci, OHCI1394_IsochronousCycleTimer);

	return 0;
}

static int ohci_iso_xmit_start(struct hpsb_iso *iso, int cycle)
{
	struct ohci_iso_xmit *xmit = iso->hostdata;
	struct ti_ohci *ohci = xmit->ohci;

	/* clear out the control register */
	reg_write(xmit->ohci, xmit->ContextControlClear, 0xFFFFFFFF);
	wmb();

	/* address and length of first descriptor block (Z=3) */
	reg_write(xmit->ohci, xmit->CommandPtr,
		  dma_prog_region_offset_to_bus(&xmit->prog, iso->pkt_dma * sizeof(struct iso_xmit_cmd)) | 3);

	/* cycle match */
	if (cycle != -1) {
		u32 start = cycle & 0x1FFF;

		/* 'cycle' is only mod 8000, but we also need two 'seconds' bits -
		   just snarf them from the current time */
		u32 seconds = reg_read(xmit->ohci, OHCI1394_IsochronousCycleTimer) >> 25;

		/* advance one second to give some extra time for DMA to start */
		seconds += 1;

		start |= (seconds & 3) << 13;

		reg_write(xmit->ohci, xmit->ContextControlSet, 0x80000000 | (start << 16));
	}

	/* enable interrupts */
	reg_write(xmit->ohci, OHCI1394_IsoXmitIntMaskSet, 1 << xmit->task.context);

	/* run */
	reg_write(xmit->ohci, xmit->ContextControlSet, 0x8000);
	mb();

	/* wait 100 usec to give the card time to go active */
	udelay(100);

	/* check the RUN bit */
	if (!(reg_read(xmit->ohci, xmit->ContextControlSet) & 0x8000)) {
		PRINT(KERN_ERR, "Error starting IT DMA (ContextControl 0x%08x)\n",
		      reg_read(xmit->ohci, xmit->ContextControlSet));
		return -1;
	}

	return 0;
}

static int ohci_isoctl(struct hpsb_iso *iso, enum isoctl_cmd cmd, unsigned long arg)
{

	switch(cmd) {
	case XMIT_INIT:
		return ohci_iso_xmit_init(iso);
	case XMIT_START:
		return ohci_iso_xmit_start(iso, arg);
	case XMIT_STOP:
		ohci_iso_xmit_stop(iso);
		return 0;
	case XMIT_QUEUE:
		return ohci_iso_xmit_queue(iso, (struct hpsb_iso_packet_info*) arg);
	case XMIT_SHUTDOWN:
		ohci_iso_xmit_shutdown(iso);
		return 0;

	case RECV_INIT:
		return ohci_iso_recv_init(iso);
	case RECV_START: {
		int *args = (int*) arg;
		return ohci_iso_recv_start(iso, args[0], args[1], args[2]);
	}
	case RECV_STOP:
		ohci_iso_recv_stop(iso);
		return 0;
	case RECV_RELEASE:
		ohci_iso_recv_release(iso, (struct hpsb_iso_packet_info*) arg);
		return 0;
	case RECV_FLUSH:
		ohci_iso_recv_task((unsigned long) iso);
		return 0;
	case RECV_SHUTDOWN:
		ohci_iso_recv_shutdown(iso);
		return 0;
	case RECV_LISTEN_CHANNEL:
		ohci_iso_recv_change_channel(iso, arg, 1);
		return 0;
	case RECV_UNLISTEN_CHANNEL:
		ohci_iso_recv_change_channel(iso, arg, 0);
		return 0;
	case RECV_SET_CHANNEL_MASK:
		ohci_iso_recv_set_channel_mask(iso, *((u64*) arg));
		return 0;

	default:
		PRINT_G(KERN_ERR, "ohci_isoctl cmd %d not implemented yet",
			cmd);
		break;
	}
	return -EINVAL;
}

/***************************************
 * IEEE-1394 functionality section END *
 ***************************************/


/********************************************************
 * Global stuff (interrupt handler, init/shutdown code) *
 ********************************************************/

static void dma_trm_reset(struct dma_trm_ctx *d)
{
	unsigned long flags;
	LIST_HEAD(packet_list);
	struct ti_ohci *ohci = d->ohci;
	struct hpsb_packet *packet, *ptmp;

	ohci1394_stop_context(ohci, d->ctrlClear, NULL);

	/* Lock the context, reset it and release it. Move the packets
	 * that were pending in the context to packet_list and free
	 * them after releasing the lock. */

	spin_lock_irqsave(&d->lock, flags);

	list_splice(&d->fifo_list, &packet_list);
	list_splice(&d->pending_list, &packet_list);
	INIT_LIST_HEAD(&d->fifo_list);
	INIT_LIST_HEAD(&d->pending_list);

	d->branchAddrPtr = NULL;
	d->sent_ind = d->prg_ind;
	d->free_prgs = d->num_desc;

	spin_unlock_irqrestore(&d->lock, flags);

	if (list_empty(&packet_list))
		return;

	PRINT(KERN_INFO, "AT dma reset ctx=%d, aborting transmission", d->ctx);

	/* Now process subsystem callbacks for the packets from this
	 * context. */
	list_for_each_entry_safe(packet, ptmp, &packet_list, driver_list) {
		list_del_init(&packet->driver_list);
		hpsb_packet_sent(ohci->host, packet, ACKX_ABORTED);
	}
}

static void ohci_schedule_iso_tasklets(struct ti_ohci *ohci,
				       quadlet_t rx_event,
				       quadlet_t tx_event)
{
	struct ohci1394_iso_tasklet *t;
	unsigned long mask;
	unsigned long flags;

	spin_lock_irqsave(&ohci->iso_tasklet_list_lock, flags);

	list_for_each_entry(t, &ohci->iso_tasklet_list, link) {
		mask = 1 << t->context;

		if (t->type == OHCI_ISO_TRANSMIT && tx_event & mask)
			tasklet_schedule(&t->tasklet);
		else if (rx_event & mask)
			tasklet_schedule(&t->tasklet);
	}

	spin_unlock_irqrestore(&ohci->iso_tasklet_list_lock, flags);
}

static irqreturn_t ohci_irq_handler(int irq, void *dev_id)
{
	quadlet_t event, node_id;
	struct ti_ohci *ohci = (struct ti_ohci *)dev_id;
	struct hpsb_host *host = ohci->host;
	int phyid = -1, isroot = 0;
	unsigned long flags;

	/* Read and clear the interrupt event register.  Don't clear
	 * the busReset event, though. This is done when we get the
	 * selfIDComplete interrupt. */
	spin_lock_irqsave(&ohci->event_lock, flags);
	event = reg_read(ohci, OHCI1394_IntEventClear);
	reg_write(ohci, OHCI1394_IntEventClear, event & ~OHCI1394_busReset);
	spin_unlock_irqrestore(&ohci->event_lock, flags);

	if (!event)
		return IRQ_NONE;

	/* If event is ~(u32)0 cardbus card was ejected.  In this case
	 * we just return, and clean up in the ohci1394_pci_remove
	 * function. */
	if (event == ~(u32) 0) {
		DBGMSG("Device removed.");
		return IRQ_NONE;
	}

	DBGMSG("IntEvent: %08x", event);

	if (event & OHCI1394_unrecoverableError) {
		int ctx;
		PRINT(KERN_ERR, "Unrecoverable error!");

		if (reg_read(ohci, OHCI1394_AsReqTrContextControlSet) & 0x800)
			PRINT(KERN_ERR, "Async Req Tx Context died: "
				"ctrl[%08x] cmdptr[%08x]",
				reg_read(ohci, OHCI1394_AsReqTrContextControlSet),
				reg_read(ohci, OHCI1394_AsReqTrCommandPtr));

		if (reg_read(ohci, OHCI1394_AsRspTrContextControlSet) & 0x800)
			PRINT(KERN_ERR, "Async Rsp Tx Context died: "
				"ctrl[%08x] cmdptr[%08x]",
				reg_read(ohci, OHCI1394_AsRspTrContextControlSet),
				reg_read(ohci, OHCI1394_AsRspTrCommandPtr));

		if (reg_read(ohci, OHCI1394_AsReqRcvContextControlSet) & 0x800)
			PRINT(KERN_ERR, "Async Req Rcv Context died: "
				"ctrl[%08x] cmdptr[%08x]",
				reg_read(ohci, OHCI1394_AsReqRcvContextControlSet),
				reg_read(ohci, OHCI1394_AsReqRcvCommandPtr));

		if (reg_read(ohci, OHCI1394_AsRspRcvContextControlSet) & 0x800)
			PRINT(KERN_ERR, "Async Rsp Rcv Context died: "
				"ctrl[%08x] cmdptr[%08x]",
				reg_read(ohci, OHCI1394_AsRspRcvContextControlSet),
				reg_read(ohci, OHCI1394_AsRspRcvCommandPtr));

		for (ctx = 0; ctx < ohci->nb_iso_xmit_ctx; ctx++) {
			if (reg_read(ohci, OHCI1394_IsoXmitContextControlSet + (16 * ctx)) & 0x800)
				PRINT(KERN_ERR, "Iso Xmit %d Context died: "
					"ctrl[%08x] cmdptr[%08x]", ctx,
					reg_read(ohci, OHCI1394_IsoXmitContextControlSet + (16 * ctx)),
					reg_read(ohci, OHCI1394_IsoXmitCommandPtr + (16 * ctx)));
		}

		for (ctx = 0; ctx < ohci->nb_iso_rcv_ctx; ctx++) {
			if (reg_read(ohci, OHCI1394_IsoRcvContextControlSet + (32 * ctx)) & 0x800)
				PRINT(KERN_ERR, "Iso Recv %d Context died: "
					"ctrl[%08x] cmdptr[%08x] match[%08x]", ctx,
					reg_read(ohci, OHCI1394_IsoRcvContextControlSet + (32 * ctx)),
					reg_read(ohci, OHCI1394_IsoRcvCommandPtr + (32 * ctx)),
					reg_read(ohci, OHCI1394_IsoRcvContextMatch + (32 * ctx)));
		}

		event &= ~OHCI1394_unrecoverableError;
	}
	if (event & OHCI1394_postedWriteErr) {
		PRINT(KERN_ERR, "physical posted write error");
		/* no recovery strategy yet, had to involve protocol drivers */
	}
	if (event & OHCI1394_cycleTooLong) {
		if(printk_ratelimit())
			PRINT(KERN_WARNING, "isochronous cycle too long");
		else
			DBGMSG("OHCI1394_cycleTooLong");
		reg_write(ohci, OHCI1394_LinkControlSet,
			  OHCI1394_LinkControl_CycleMaster);
		event &= ~OHCI1394_cycleTooLong;
	}
	if (event & OHCI1394_cycleInconsistent) {
		/* We subscribe to the cycleInconsistent event only to
		 * clear the corresponding event bit... otherwise,
		 * isochronous cycleMatch DMA won't work. */
		DBGMSG("OHCI1394_cycleInconsistent");
		event &= ~OHCI1394_cycleInconsistent;
	}
	if (event & OHCI1394_busReset) {
		/* The busReset event bit can't be cleared during the
		 * selfID phase, so we disable busReset interrupts, to
		 * avoid burying the cpu in interrupt requests. */
		spin_lock_irqsave(&ohci->event_lock, flags);
		reg_write(ohci, OHCI1394_IntMaskClear, OHCI1394_busReset);

		if (ohci->check_busreset) {
			int loop_count = 0;

			udelay(10);

			while (reg_read(ohci, OHCI1394_IntEventSet) & OHCI1394_busReset) {
				reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);

				spin_unlock_irqrestore(&ohci->event_lock, flags);
				udelay(10);
				spin_lock_irqsave(&ohci->event_lock, flags);

				/* The loop counter check is to prevent the driver
				 * from remaining in this state forever. For the
				 * initial bus reset, the loop continues for ever
				 * and the system hangs, until some device is plugged-in
				 * or out manually into a port! The forced reset seems
				 * to solve this problem. This mainly effects nForce2. */
				if (loop_count > 10000) {
					ohci_devctl(host, RESET_BUS, LONG_RESET);
					DBGMSG("Detected bus-reset loop. Forced a bus reset!");
					loop_count = 0;
				}

				loop_count++;
			}
		}
		spin_unlock_irqrestore(&ohci->event_lock, flags);
		if (!host->in_bus_reset) {
			DBGMSG("irq_handler: Bus reset requested");

			/* Subsystem call */
			hpsb_bus_reset(ohci->host);
		}
		event &= ~OHCI1394_busReset;
	}
	if (event & OHCI1394_reqTxComplete) {
		struct dma_trm_ctx *d = &ohci->at_req_context;
		DBGMSG("Got reqTxComplete interrupt "
		       "status=0x%08X", reg_read(ohci, d->ctrlSet));
		if (reg_read(ohci, d->ctrlSet) & 0x800)
			ohci1394_stop_context(ohci, d->ctrlClear,
					      "reqTxComplete");
		else
			dma_trm_tasklet((unsigned long)d);
			//tasklet_schedule(&d->task);
		event &= ~OHCI1394_reqTxComplete;
	}
	if (event & OHCI1394_respTxComplete) {
		struct dma_trm_ctx *d = &ohci->at_resp_context;
		DBGMSG("Got respTxComplete interrupt "
		       "status=0x%08X", reg_read(ohci, d->ctrlSet));
		if (reg_read(ohci, d->ctrlSet) & 0x800)
			ohci1394_stop_context(ohci, d->ctrlClear,
					      "respTxComplete");
		else
			tasklet_schedule(&d->task);
		event &= ~OHCI1394_respTxComplete;
	}
	if (event & OHCI1394_RQPkt) {
		struct dma_rcv_ctx *d = &ohci->ar_req_context;
		DBGMSG("Got RQPkt interrupt status=0x%08X",
		       reg_read(ohci, d->ctrlSet));
		if (reg_read(ohci, d->ctrlSet) & 0x800)
			ohci1394_stop_context(ohci, d->ctrlClear, "RQPkt");
		else
			tasklet_schedule(&d->task);
		event &= ~OHCI1394_RQPkt;
	}
	if (event & OHCI1394_RSPkt) {
		struct dma_rcv_ctx *d = &ohci->ar_resp_context;
		DBGMSG("Got RSPkt interrupt status=0x%08X",
		       reg_read(ohci, d->ctrlSet));
		if (reg_read(ohci, d->ctrlSet) & 0x800)
			ohci1394_stop_context(ohci, d->ctrlClear, "RSPkt");
		else
			tasklet_schedule(&d->task);
		event &= ~OHCI1394_RSPkt;
	}
	if (event & OHCI1394_isochRx) {
		quadlet_t rx_event;

		rx_event = reg_read(ohci, OHCI1394_IsoRecvIntEventSet);
		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, rx_event);
		ohci_schedule_iso_tasklets(ohci, rx_event, 0);
		event &= ~OHCI1394_isochRx;
	}
	if (event & OHCI1394_isochTx) {
		quadlet_t tx_event;

		tx_event = reg_read(ohci, OHCI1394_IsoXmitIntEventSet);
		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, tx_event);
		ohci_schedule_iso_tasklets(ohci, 0, tx_event);
		event &= ~OHCI1394_isochTx;
	}
	if (event & OHCI1394_selfIDComplete) {
		if (host->in_bus_reset) {
			node_id = reg_read(ohci, OHCI1394_NodeID);

			if (!(node_id & 0x80000000)) {
				PRINT(KERN_ERR,
				      "SelfID received, but NodeID invalid "
				      "(probably new bus reset occurred): %08X",
				      node_id);
				goto selfid_not_valid;
			}

			phyid =  node_id & 0x0000003f;
			isroot = (node_id & 0x40000000) != 0;

			DBGMSG("SelfID interrupt received "
			      "(phyid %d, %s)", phyid,
			      (isroot ? "root" : "not root"));

			handle_selfid(ohci, host, phyid, isroot);

			/* Clear the bus reset event and re-enable the
			 * busReset interrupt.  */
			spin_lock_irqsave(&ohci->event_lock, flags);
			reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
			reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_busReset);
			spin_unlock_irqrestore(&ohci->event_lock, flags);

			/* Turn on phys dma reception.
			 *
			 * TODO: Enable some sort of filtering management.
			 */
			if (phys_dma) {
				reg_write(ohci, OHCI1394_PhyReqFilterHiSet,
					  0xffffffff);
				reg_write(ohci, OHCI1394_PhyReqFilterLoSet,
					  0xffffffff);
			}

			DBGMSG("PhyReqFilter=%08x%08x",
			       reg_read(ohci, OHCI1394_PhyReqFilterHiSet),
			       reg_read(ohci, OHCI1394_PhyReqFilterLoSet));

			hpsb_selfid_complete(host, phyid, isroot);
		} else
			PRINT(KERN_ERR,
			      "SelfID received outside of bus reset sequence");

selfid_not_valid:
		event &= ~OHCI1394_selfIDComplete;
	}

	/* Make sure we handle everything, just in case we accidentally
	 * enabled an interrupt that we didn't write a handler for.  */
	if (event)
		PRINT(KERN_ERR, "Unhandled interrupt(s) 0x%08x",
		      event);

	return IRQ_HANDLED;
}

/* Put the buffer back into the dma context */
static void insert_dma_buffer(struct dma_rcv_ctx *d, int idx)
{
	struct ti_ohci *ohci = (struct ti_ohci*)(d->ohci);
	DBGMSG("Inserting dma buf ctx=%d idx=%d", d->ctx, idx);

	d->prg_cpu[idx]->status = cpu_to_le32(d->buf_size);
	d->prg_cpu[idx]->branchAddress &= le32_to_cpu(0xfffffff0);
	idx = (idx + d->num_desc - 1 ) % d->num_desc;
	d->prg_cpu[idx]->branchAddress |= le32_to_cpu(0x00000001);

	/* To avoid a race, ensure 1394 interface hardware sees the inserted
	 * context program descriptors before it sees the wakeup bit set. */
	wmb();
	
	/* wake up the dma context if necessary */
	if (!(reg_read(ohci, d->ctrlSet) & 0x400)) {
		PRINT(KERN_INFO,
		      "Waking dma ctx=%d ... processing is probably too slow",
		      d->ctx);
	}

	/* do this always, to avoid race condition */
	reg_write(ohci, d->ctrlSet, 0x1000);
}

#define cond_le32_to_cpu(data, noswap) \
	(noswap ? data : le32_to_cpu(data))

static const int TCODE_SIZE[16] = {20, 0, 16, -1, 16, 20, 20, 0,
			    -1, 0, -1, 0, -1, -1, 16, -1};

/*
 * Determine the length of a packet in the buffer
 * Optimization suggested by Pascal Drolet <pascal.drolet@informission.ca>
 */
static inline int packet_length(struct dma_rcv_ctx *d, int idx,
				quadlet_t *buf_ptr, int offset,
				unsigned char tcode, int noswap)
{
	int length = -1;

	if (d->type == DMA_CTX_ASYNC_REQ || d->type == DMA_CTX_ASYNC_RESP) {
		length = TCODE_SIZE[tcode];
		if (length == 0) {
			if (offset + 12 >= d->buf_size) {
				length = (cond_le32_to_cpu(d->buf_cpu[(idx + 1) % d->num_desc]
						[3 - ((d->buf_size - offset) >> 2)], noswap) >> 16);
			} else {
				length = (cond_le32_to_cpu(buf_ptr[3], noswap) >> 16);
			}
			length += 20;
		}
	} else if (d->type == DMA_CTX_ISO) {
		/* Assumption: buffer fill mode with header/trailer */
		length = (cond_le32_to_cpu(buf_ptr[0], noswap) >> 16) + 8;
	}

	if (length > 0 && length % 4)
		length += 4 - (length % 4);

	return length;
}

/* Tasklet that processes dma receive buffers */
static void dma_rcv_tasklet (unsigned long data)
{
	struct dma_rcv_ctx *d = (struct dma_rcv_ctx*)data;
	struct ti_ohci *ohci = (struct ti_ohci*)(d->ohci);
	unsigned int split_left, idx, offset, rescount;
	unsigned char tcode;
	int length, bytes_left, ack;
	unsigned long flags;
	quadlet_t *buf_ptr;
	char *split_ptr;
	char msg[256];

	spin_lock_irqsave(&d->lock, flags);

	idx = d->buf_ind;
	offset = d->buf_offset;
	buf_ptr = d->buf_cpu[idx] + offset/4;

	rescount = le32_to_cpu(d->prg_cpu[idx]->status) & 0xffff;
	bytes_left = d->buf_size - rescount - offset;

	while (bytes_left > 0) {
		tcode = (cond_le32_to_cpu(buf_ptr[0], ohci->no_swap_incoming) >> 4) & 0xf;

		/* packet_length() will return < 4 for an error */
		length = packet_length(d, idx, buf_ptr, offset, tcode, ohci->no_swap_incoming);

		if (length < 4) { /* something is wrong */
			sprintf(msg,"Unexpected tcode 0x%x(0x%08x) in AR ctx=%d, length=%d",
				tcode, cond_le32_to_cpu(buf_ptr[0], ohci->no_swap_incoming),
				d->ctx, length);
			ohci1394_stop_context(ohci, d->ctrlClear, msg);
			spin_unlock_irqrestore(&d->lock, flags);
			return;
		}

		/* The first case is where we have a packet that crosses
		 * over more than one descriptor. The next case is where
		 * it's all in the first descriptor.  */
		if ((offset + length) > d->buf_size) {
			DBGMSG("Split packet rcv'd");
			if (length > d->split_buf_size) {
				ohci1394_stop_context(ohci, d->ctrlClear,
					     "Split packet size exceeded");
				d->buf_ind = idx;
				d->buf_offset = offset;
				spin_unlock_irqrestore(&d->lock, flags);
				return;
			}

			if (le32_to_cpu(d->prg_cpu[(idx+1)%d->num_desc]->status)
			    == d->buf_size) {
				/* Other part of packet not written yet.
				 * this should never happen I think
				 * anyway we'll get it on the next call.  */
				PRINT(KERN_INFO,
				      "Got only half a packet!");
				d->buf_ind = idx;
				d->buf_offset = offset;
				spin_unlock_irqrestore(&d->lock, flags);
				return;
			}

			split_left = length;
			split_ptr = (char *)d->spb;
			memcpy(split_ptr,buf_ptr,d->buf_size-offset);
			split_left -= d->buf_size-offset;
			split_ptr += d->buf_size-offset;
			insert_dma_buffer(d, idx);
			idx = (idx+1) % d->num_desc;
			buf_ptr = d->buf_cpu[idx];
			offset=0;

			while (split_left >= d->buf_size) {
				memcpy(split_ptr,buf_ptr,d->buf_size);
				split_ptr += d->buf_size;
				split_left -= d->buf_size;
				insert_dma_buffer(d, idx);
				idx = (idx+1) % d->num_desc;
				buf_ptr = d->buf_cpu[idx];
			}

			if (split_left > 0) {
				memcpy(split_ptr, buf_ptr, split_left);
				offset = split_left;
				buf_ptr += offset/4;
			}
		} else {
			DBGMSG("Single packet rcv'd");
			memcpy(d->spb, buf_ptr, length);
			offset += length;
			buf_ptr += length/4;
			if (offset==d->buf_size) {
				insert_dma_buffer(d, idx);
				idx = (idx+1) % d->num_desc;
				buf_ptr = d->buf_cpu[idx];
				offset=0;
			}
		}

		/* We get one phy packet to the async descriptor for each
		 * bus reset. We always ignore it.  */
		if (tcode != OHCI1394_TCODE_PHY) {
			if (!ohci->no_swap_incoming)
				header_le32_to_cpu(d->spb, tcode);
			DBGMSG("Packet received from node"
				" %d ack=0x%02X spd=%d tcode=0x%X"
				" length=%d ctx=%d tlabel=%d",
				(d->spb[1]>>16)&0x3f,
				(cond_le32_to_cpu(d->spb[length/4-1], ohci->no_swap_incoming)>>16)&0x1f,
				(cond_le32_to_cpu(d->spb[length/4-1], ohci->no_swap_incoming)>>21)&0x3,
				tcode, length, d->ctx,
				(d->spb[0]>>10)&0x3f);

			ack = (((cond_le32_to_cpu(d->spb[length/4-1], ohci->no_swap_incoming)>>16)&0x1f)
				== 0x11) ? 1 : 0;

			hpsb_packet_received(ohci->host, d->spb,
					     length-4, ack);
		}
#ifdef OHCI1394_DEBUG
		else
			PRINT (KERN_DEBUG, "Got phy packet ctx=%d ... discarded",
			       d->ctx);
#endif

	       	rescount = le32_to_cpu(d->prg_cpu[idx]->status) & 0xffff;

		bytes_left = d->buf_size - rescount - offset;

	}

	d->buf_ind = idx;
	d->buf_offset = offset;

	spin_unlock_irqrestore(&d->lock, flags);
}

/* Bottom half that processes sent packets */
static void dma_trm_tasklet (unsigned long data)
{
	struct dma_trm_ctx *d = (struct dma_trm_ctx*)data;
	struct ti_ohci *ohci = (struct ti_ohci*)(d->ohci);
	struct hpsb_packet *packet, *ptmp;
	unsigned long flags;
	u32 status, ack;
        size_t datasize;

	spin_lock_irqsave(&d->lock, flags);

	list_for_each_entry_safe(packet, ptmp, &d->fifo_list, driver_list) {
                datasize = packet->data_size;
		if (datasize && packet->type != hpsb_raw)
			status = le32_to_cpu(
				d->prg_cpu[d->sent_ind]->end.status) >> 16;
		else
			status = le32_to_cpu(
				d->prg_cpu[d->sent_ind]->begin.status) >> 16;

		if (status == 0)
			/* this packet hasn't been sent yet*/
			break;

#ifdef OHCI1394_DEBUG
		if (datasize)
			if (((le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])>>4)&0xf) == 0xa)
				DBGMSG("Stream packet sent to channel %d tcode=0x%X "
				       "ack=0x%X spd=%d dataLength=%d ctx=%d",
				       (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])>>8)&0x3f,
				       (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])>>4)&0xf,
				       status&0x1f, (status>>5)&0x3,
				       le32_to_cpu(d->prg_cpu[d->sent_ind]->data[1])>>16,
				       d->ctx);
			else
				DBGMSG("Packet sent to node %d tcode=0x%X tLabel="
				       "%d ack=0x%X spd=%d dataLength=%d ctx=%d",
				       (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[1])>>16)&0x3f,
				       (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])>>4)&0xf,
				       (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])>>10)&0x3f,
				       status&0x1f, (status>>5)&0x3,
				       le32_to_cpu(d->prg_cpu[d->sent_ind]->data[3])>>16,
				       d->ctx);
		else
			DBGMSG("Packet sent to node %d tcode=0x%X tLabel="
			       "%d ack=0x%X spd=%d data=0x%08X ctx=%d",
                                (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[1])
                                        >>16)&0x3f,
                                (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])
                                        >>4)&0xf,
                                (le32_to_cpu(d->prg_cpu[d->sent_ind]->data[0])
                                        >>10)&0x3f,
                                status&0x1f, (status>>5)&0x3,
                                le32_to_cpu(d->prg_cpu[d->sent_ind]->data[3]),
                                d->ctx);
#endif

		if (status & 0x10) {
			ack = status & 0xf;
		} else {
			switch (status & 0x1f) {
			case EVT_NO_STATUS: /* that should never happen */
			case EVT_RESERVED_A: /* that should never happen */
			case EVT_LONG_PACKET: /* that should never happen */
				PRINT(KERN_WARNING, "Received OHCI evt_* error 0x%x", status & 0x1f);
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_MISSING_ACK:
				ack = ACKX_TIMEOUT;
				break;
			case EVT_UNDERRUN:
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_OVERRUN: /* that should never happen */
				PRINT(KERN_WARNING, "Received OHCI evt_* error 0x%x", status & 0x1f);
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_DESCRIPTOR_READ:
			case EVT_DATA_READ:
			case EVT_DATA_WRITE:
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_BUS_RESET: /* that should never happen */
				PRINT(KERN_WARNING, "Received OHCI evt_* error 0x%x", status & 0x1f);
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_TIMEOUT:
				ack = ACKX_TIMEOUT;
				break;
			case EVT_TCODE_ERR:
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_RESERVED_B: /* that should never happen */
			case EVT_RESERVED_C: /* that should never happen */
				PRINT(KERN_WARNING, "Received OHCI evt_* error 0x%x", status & 0x1f);
				ack = ACKX_SEND_ERROR;
				break;
			case EVT_UNKNOWN:
			case EVT_FLUSHED:
				ack = ACKX_SEND_ERROR;
				break;
			default:
				PRINT(KERN_ERR, "Unhandled OHCI evt_* error 0x%x", status & 0x1f);
				ack = ACKX_SEND_ERROR;
				BUG();
			}
		}

		list_del_init(&packet->driver_list);
		hpsb_packet_sent(ohci->host, packet, ack);

		if (datasize) {
			pci_unmap_single(ohci->dev,
					 cpu_to_le32(d->prg_cpu[d->sent_ind]->end.address),
					 datasize, PCI_DMA_TODEVICE);
			OHCI_DMA_FREE("single Xmit data packet");
		}

		d->sent_ind = (d->sent_ind+1)%d->num_desc;
		d->free_prgs++;
	}

	dma_trm_flush(ohci, d);

	spin_unlock_irqrestore(&d->lock, flags);
}

static void stop_dma_rcv_ctx(struct dma_rcv_ctx *d)
{
	if (d->ctrlClear) {
		ohci1394_stop_context(d->ohci, d->ctrlClear, NULL);

		if (d->type == DMA_CTX_ISO) {
			/* disable interrupts */
			reg_write(d->ohci, OHCI1394_IsoRecvIntMaskClear, 1 << d->ctx);
			ohci1394_unregister_iso_tasklet(d->ohci, &d->ohci->ir_legacy_tasklet);
		} else {
			tasklet_kill(&d->task);
		}
	}
}


static void free_dma_rcv_ctx(struct dma_rcv_ctx *d)
{
	int i;
	struct ti_ohci *ohci = d->ohci;

	if (ohci == NULL)
		return;

	DBGMSG("Freeing dma_rcv_ctx %d", d->ctx);

	if (d->buf_cpu) {
		for (i=0; i<d->num_desc; i++)
			if (d->buf_cpu[i] && d->buf_bus[i]) {
				pci_free_consistent(
					ohci->dev, d->buf_size,
					d->buf_cpu[i], d->buf_bus[i]);
				OHCI_DMA_FREE("consistent dma_rcv buf[%d]", i);
			}
		kfree(d->buf_cpu);
		kfree(d->buf_bus);
	}
	if (d->prg_cpu) {
		for (i=0; i<d->num_desc; i++)
			if (d->prg_cpu[i] && d->prg_bus[i]) {
				pci_pool_free(d->prg_pool, d->prg_cpu[i], d->prg_bus[i]);
				OHCI_DMA_FREE("consistent dma_rcv prg[%d]", i);
			}
		pci_pool_destroy(d->prg_pool);
		OHCI_DMA_FREE("dma_rcv prg pool");
		kfree(d->prg_cpu);
		kfree(d->prg_bus);
	}
	kfree(d->spb);

	/* Mark this context as freed. */
	d->ohci = NULL;
}

static int
alloc_dma_rcv_ctx(struct ti_ohci *ohci, struct dma_rcv_ctx *d,
		  enum context_type type, int ctx, int num_desc,
		  int buf_size, int split_buf_size, int context_base)
{
	int i, len;
	static int num_allocs;
	static char pool_name[20];

	d->ohci = ohci;
	d->type = type;
	d->ctx = ctx;

	d->num_desc = num_desc;
	d->buf_size = buf_size;
	d->split_buf_size = split_buf_size;

	d->ctrlSet = 0;
	d->ctrlClear = 0;
	d->cmdPtr = 0;

	d->buf_cpu = kzalloc(d->num_desc * sizeof(*d->buf_cpu), GFP_ATOMIC);
	d->buf_bus = kzalloc(d->num_desc * sizeof(*d->buf_bus), GFP_ATOMIC);

	if (d->buf_cpu == NULL || d->buf_bus == NULL) {
		PRINT(KERN_ERR, "Failed to allocate dma buffer");
		free_dma_rcv_ctx(d);
		return -ENOMEM;
	}

	d->prg_cpu = kzalloc(d->num_desc * sizeof(*d->prg_cpu), GFP_ATOMIC);
	d->prg_bus = kzalloc(d->num_desc * sizeof(*d->prg_bus), GFP_ATOMIC);

	if (d->prg_cpu == NULL || d->prg_bus == NULL) {
		PRINT(KERN_ERR, "Failed to allocate dma prg");
		free_dma_rcv_ctx(d);
		return -ENOMEM;
	}

	d->spb = kmalloc(d->split_buf_size, GFP_ATOMIC);

	if (d->spb == NULL) {
		PRINT(KERN_ERR, "Failed to allocate split buffer");
		free_dma_rcv_ctx(d);
		return -ENOMEM;
	}
	
	len = sprintf(pool_name, "ohci1394_rcv_prg");
	sprintf(pool_name+len, "%d", num_allocs);
	d->prg_pool = pci_pool_create(pool_name, ohci->dev,
				sizeof(struct dma_cmd), 4, 0);
	if(d->prg_pool == NULL)
	{
		PRINT(KERN_ERR, "pci_pool_create failed for %s", pool_name);
		free_dma_rcv_ctx(d);
		return -ENOMEM;
	}
	num_allocs++;

	OHCI_DMA_ALLOC("dma_rcv prg pool");

	for (i=0; i<d->num_desc; i++) {
		d->buf_cpu[i] = pci_alloc_consistent(ohci->dev,
						     d->buf_size,
						     d->buf_bus+i);
		OHCI_DMA_ALLOC("consistent dma_rcv buf[%d]", i);

		if (d->buf_cpu[i] != NULL) {
			memset(d->buf_cpu[i], 0, d->buf_size);
		} else {
			PRINT(KERN_ERR,
			      "Failed to allocate dma buffer");
			free_dma_rcv_ctx(d);
			return -ENOMEM;
		}

		d->prg_cpu[i] = pci_pool_alloc(d->prg_pool, GFP_KERNEL, d->prg_bus+i);
		OHCI_DMA_ALLOC("pool dma_rcv prg[%d]", i);

                if (d->prg_cpu[i] != NULL) {
                        memset(d->prg_cpu[i], 0, sizeof(struct dma_cmd));
		} else {
			PRINT(KERN_ERR,
			      "Failed to allocate dma prg");
			free_dma_rcv_ctx(d);
			return -ENOMEM;
		}
	}

        spin_lock_init(&d->lock);

	if (type == DMA_CTX_ISO) {
		ohci1394_init_iso_tasklet(&ohci->ir_legacy_tasklet,
					  OHCI_ISO_MULTICHANNEL_RECEIVE,
					  dma_rcv_tasklet, (unsigned long) d);
	} else {
		d->ctrlSet = context_base + OHCI1394_ContextControlSet;
		d->ctrlClear = context_base + OHCI1394_ContextControlClear;
		d->cmdPtr = context_base + OHCI1394_ContextCommandPtr;

		tasklet_init (&d->task, dma_rcv_tasklet, (unsigned long) d);
	}

	return 0;
}

static void free_dma_trm_ctx(struct dma_trm_ctx *d)
{
	int i;
	struct ti_ohci *ohci = d->ohci;

	if (ohci == NULL)
		return;

	DBGMSG("Freeing dma_trm_ctx %d", d->ctx);

	if (d->prg_cpu) {
		for (i=0; i<d->num_desc; i++)
			if (d->prg_cpu[i] && d->prg_bus[i]) {
				pci_pool_free(d->prg_pool, d->prg_cpu[i], d->prg_bus[i]);
				OHCI_DMA_FREE("pool dma_trm prg[%d]", i);
			}
		pci_pool_destroy(d->prg_pool);
		OHCI_DMA_FREE("dma_trm prg pool");
		kfree(d->prg_cpu);
		kfree(d->prg_bus);
	}

	/* Mark this context as freed. */
	d->ohci = NULL;
}

static int
alloc_dma_trm_ctx(struct ti_ohci *ohci, struct dma_trm_ctx *d,
		  enum context_type type, int ctx, int num_desc,
		  int context_base)
{
	int i, len;
	static char pool_name[20];
	static int num_allocs=0;

	d->ohci = ohci;
	d->type = type;
	d->ctx = ctx;
	d->num_desc = num_desc;
	d->ctrlSet = 0;
	d->ctrlClear = 0;
	d->cmdPtr = 0;

	d->prg_cpu = kzalloc(d->num_desc * sizeof(*d->prg_cpu), GFP_KERNEL);
	d->prg_bus = kzalloc(d->num_desc * sizeof(*d->prg_bus), GFP_KERNEL);

	if (d->prg_cpu == NULL || d->prg_bus == NULL) {
		PRINT(KERN_ERR, "Failed to allocate at dma prg");
		free_dma_trm_ctx(d);
		return -ENOMEM;
	}

	len = sprintf(pool_name, "ohci1394_trm_prg");
	sprintf(pool_name+len, "%d", num_allocs);
	d->prg_pool = pci_pool_create(pool_name, ohci->dev,
				sizeof(struct at_dma_prg), 4, 0);
	if (d->prg_pool == NULL) {
		PRINT(KERN_ERR, "pci_pool_create failed for %s", pool_name);
		free_dma_trm_ctx(d);
		return -ENOMEM;
	}
	num_allocs++;

	OHCI_DMA_ALLOC("dma_rcv prg pool");

	for (i = 0; i < d->num_desc; i++) {
		d->prg_cpu[i] = pci_pool_alloc(d->prg_pool, GFP_KERNEL, d->prg_bus+i);
		OHCI_DMA_ALLOC("pool dma_trm prg[%d]", i);

                if (d->prg_cpu[i] != NULL) {
                        memset(d->prg_cpu[i], 0, sizeof(struct at_dma_prg));
		} else {
			PRINT(KERN_ERR,
			      "Failed to allocate at dma prg");
			free_dma_trm_ctx(d);
			return -ENOMEM;
		}
	}

        spin_lock_init(&d->lock);

	/* initialize tasklet */
	if (type == DMA_CTX_ISO) {
		ohci1394_init_iso_tasklet(&ohci->it_legacy_tasklet, OHCI_ISO_TRANSMIT,
					  dma_trm_tasklet, (unsigned long) d);
		if (ohci1394_register_iso_tasklet(ohci,
						  &ohci->it_legacy_tasklet) < 0) {
			PRINT(KERN_ERR, "No IT DMA context available");
			free_dma_trm_ctx(d);
			return -EBUSY;
		}

		/* IT can be assigned to any context by register_iso_tasklet */
		d->ctx = ohci->it_legacy_tasklet.context;
		d->ctrlSet = OHCI1394_IsoXmitContextControlSet + 16 * d->ctx;
		d->ctrlClear = OHCI1394_IsoXmitContextControlClear + 16 * d->ctx;
		d->cmdPtr = OHCI1394_IsoXmitCommandPtr + 16 * d->ctx;
	} else {
		d->ctrlSet = context_base + OHCI1394_ContextControlSet;
		d->ctrlClear = context_base + OHCI1394_ContextControlClear;
		d->cmdPtr = context_base + OHCI1394_ContextCommandPtr;
		tasklet_init (&d->task, dma_trm_tasklet, (unsigned long)d);
	}

	return 0;
}

static void ohci_set_hw_config_rom(struct hpsb_host *host, quadlet_t *config_rom)
{
	struct ti_ohci *ohci = host->hostdata;

	reg_write(ohci, OHCI1394_ConfigROMhdr, be32_to_cpu(config_rom[0]));
	reg_write(ohci, OHCI1394_BusOptions, be32_to_cpu(config_rom[2]));

	memcpy(ohci->csr_config_rom_cpu, config_rom, OHCI_CONFIG_ROM_LEN);
}


static quadlet_t ohci_hw_csr_reg(struct hpsb_host *host, int reg,
                                 quadlet_t data, quadlet_t compare)
{
	struct ti_ohci *ohci = host->hostdata;
	int i;

	reg_write(ohci, OHCI1394_CSRData, data);
	reg_write(ohci, OHCI1394_CSRCompareData, compare);
	reg_write(ohci, OHCI1394_CSRControl, reg & 0x3);

	for (i = 0; i < OHCI_LOOP_COUNT; i++) {
		if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000)
			break;

		mdelay(1);
	}

	return reg_read(ohci, OHCI1394_CSRData);
}

static struct hpsb_host_driver ohci1394_driver = {
	.owner =		THIS_MODULE,
	.name =			OHCI1394_DRIVER_NAME,
	.set_hw_config_rom =	ohci_set_hw_config_rom,
	.transmit_packet =	ohci_transmit,
	.devctl =		ohci_devctl,
	.isoctl =               ohci_isoctl,
	.hw_csr_reg =		ohci_hw_csr_reg,
};

/***********************************
 * PCI Driver Interface functions  *
 ***********************************/

#define FAIL(err, fmt, args...)			\
do {						\
	PRINT_G(KERN_ERR, fmt , ## args);	\
        ohci1394_pci_remove(dev);               \
	return err;				\
} while (0)

static int __devinit ohci1394_pci_probe(struct pci_dev *dev,
					const struct pci_device_id *ent)
{
	struct hpsb_host *host;
	struct ti_ohci *ohci;	/* shortcut to currently handled device */
	resource_size_t ohci_base;

#ifdef CONFIG_PPC_PMAC
	/* Necessary on some machines if ohci1394 was loaded/ unloaded before */
	if (machine_is(powermac)) {
		struct device_node *ofn = pci_device_to_OF_node(dev);

		if (ofn) {
			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
		}
	}
#endif /* CONFIG_PPC_PMAC */

        if (pci_enable_device(dev))
		FAIL(-ENXIO, "Failed to enable OHCI hardware");
        pci_set_master(dev);

	host = hpsb_alloc_host(&ohci1394_driver, sizeof(struct ti_ohci), &dev->dev);
	if (!host) FAIL(-ENOMEM, "Failed to allocate host structure");

	ohci = host->hostdata;
	ohci->dev = dev;
	ohci->host = host;
	ohci->init_state = OHCI_INIT_ALLOC_HOST;
	host->pdev = dev;
	pci_set_drvdata(dev, ohci);

	/* We don't want hardware swapping */
	pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);

	/* Some oddball Apple controllers do not order the selfid
	 * properly, so we make up for it here.  */
#ifndef __LITTLE_ENDIAN
	/* XXX: Need a better way to check this. I'm wondering if we can
	 * read the values of the OHCI1394_PCI_HCI_Control and the
	 * noByteSwapData registers to see if they were not cleared to
	 * zero. Should this work? Obviously it's not defined what these
	 * registers will read when they aren't supported. Bleh! */
	if (dev->vendor == PCI_VENDOR_ID_APPLE &&
	    dev->device == PCI_DEVICE_ID_APPLE_UNI_N_FW) {
		ohci->no_swap_incoming = 1;
		ohci->selfid_swap = 0;
	} else
		ohci->selfid_swap = 1;
#endif


#ifndef PCI_DEVICE_ID_NVIDIA_NFORCE2_FW
#define PCI_DEVICE_ID_NVIDIA_NFORCE2_FW 0x006e
#endif

	/* These chipsets require a bit of extra care when checking after
	 * a busreset.  */
	if ((dev->vendor == PCI_VENDOR_ID_APPLE &&
	     dev->device == PCI_DEVICE_ID_APPLE_UNI_N_FW) ||
	    (dev->vendor ==  PCI_VENDOR_ID_NVIDIA &&
	     dev->device == PCI_DEVICE_ID_NVIDIA_NFORCE2_FW))
		ohci->check_busreset = 1;

	/* We hardwire the MMIO length, since some CardBus adaptors
	 * fail to report the right length.  Anyway, the ohci spec
	 * clearly says it's 2kb, so this shouldn't be a problem. */
	ohci_base = pci_resource_start(dev, 0);
	if (pci_resource_len(dev, 0) < OHCI1394_REGISTER_SIZE)
		PRINT(KERN_WARNING, "PCI resource length of 0x%llx too small!",
		      (unsigned long long)pci_resource_len(dev, 0));

	/* Seems PCMCIA handles this internally. Not sure why. Seems
	 * pretty bogus to force a driver to special case this.  */
#ifndef PCMCIA
	if (!request_mem_region (ohci_base, OHCI1394_REGISTER_SIZE, OHCI1394_DRIVER_NAME))
		FAIL(-ENOMEM, "MMIO resource (0x%llx - 0x%llx) unavailable",
			(unsigned long long)ohci_base,
			(unsigned long long)ohci_base + OHCI1394_REGISTER_SIZE);
#endif
	ohci->init_state = OHCI_INIT_HAVE_MEM_REGION;

	ohci->registers = ioremap(ohci_base, OHCI1394_REGISTER_SIZE);
	if (ohci->registers == NULL)
		FAIL(-ENXIO, "Failed to remap registers - card not accessible");
	ohci->init_state = OHCI_INIT_HAVE_IOMAPPING;
	DBGMSG("Remapped memory spaces reg 0x%p", ohci->registers);

	/* csr_config rom allocation */
	ohci->csr_config_rom_cpu =
		pci_alloc_consistent(ohci->dev, OHCI_CONFIG_ROM_LEN,
				     &ohci->csr_config_rom_bus);
	OHCI_DMA_ALLOC("consistent csr_config_rom");
	if (ohci->csr_config_rom_cpu == NULL)
		FAIL(-ENOMEM, "Failed to allocate buffer config rom");
	ohci->init_state = OHCI_INIT_HAVE_CONFIG_ROM_BUFFER;

	/* self-id dma buffer allocation */
	ohci->selfid_buf_cpu =
		pci_alloc_consistent(ohci->dev, OHCI1394_SI_DMA_BUF_SIZE,
                      &ohci->selfid_buf_bus);
	OHCI_DMA_ALLOC("consistent selfid_buf");

	if (ohci->selfid_buf_cpu == NULL)
		FAIL(-ENOMEM, "Failed to allocate DMA buffer for self-id packets");
	ohci->init_state = OHCI_INIT_HAVE_SELFID_BUFFER;

	if ((unsigned long)ohci->selfid_buf_cpu & 0x1fff)
		PRINT(KERN_INFO, "SelfID buffer %p is not aligned on "
		      "8Kb boundary... may cause problems on some CXD3222 chip",
		      ohci->selfid_buf_cpu);

	/* No self-id errors at startup */
	ohci->self_id_errors = 0;

	ohci->init_state = OHCI_INIT_HAVE_TXRX_BUFFERS__MAYBE;
	/* AR DMA request context allocation */
	if (alloc_dma_rcv_ctx(ohci, &ohci->ar_req_context,
			      DMA_CTX_ASYNC_REQ, 0, AR_REQ_NUM_DESC,
			      AR_REQ_BUF_SIZE, AR_REQ_SPLIT_BUF_SIZE,
			      OHCI1394_AsReqRcvContextBase) < 0)
		FAIL(-ENOMEM, "Failed to allocate AR Req context");

	/* AR DMA response context allocation */
	if (alloc_dma_rcv_ctx(ohci, &ohci->ar_resp_context,
			      DMA_CTX_ASYNC_RESP, 0, AR_RESP_NUM_DESC,
			      AR_RESP_BUF_SIZE, AR_RESP_SPLIT_BUF_SIZE,
			      OHCI1394_AsRspRcvContextBase) < 0)
		FAIL(-ENOMEM, "Failed to allocate AR Resp context");

	/* AT DMA request context */
	if (alloc_dma_trm_ctx(ohci, &ohci->at_req_context,
			      DMA_CTX_ASYNC_REQ, 0, AT_REQ_NUM_DESC,
			      OHCI1394_AsReqTrContextBase) < 0)
		FAIL(-ENOMEM, "Failed to allocate AT Req context");

	/* AT DMA response context */
	if (alloc_dma_trm_ctx(ohci, &ohci->at_resp_context,
			      DMA_CTX_ASYNC_RESP, 1, AT_RESP_NUM_DESC,
			      OHCI1394_AsRspTrContextBase) < 0)
		FAIL(-ENOMEM, "Failed to allocate AT Resp context");

	/* Start off with a soft reset, to clear everything to a sane
	 * state. */
	ohci_soft_reset(ohci);

	/* Now enable LPS, which we need in order to start accessing
	 * most of the registers.  In fact, on some cards (ALI M5251),
	 * accessing registers in the SClk domain without LPS enabled
	 * will lock up the machine.  Wait 50msec to make sure we have
	 * full link enabled.  */
	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_LPS);

	/* Disable and clear interrupts */
	reg_write(ohci, OHCI1394_IntEventClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IntMaskClear, 0xffffffff);

	mdelay(50);

	/* Determine the number of available IR and IT contexts. */
	ohci->nb_iso_rcv_ctx =
		get_nb_iso_ctx(ohci, OHCI1394_IsoRecvIntMaskSet);
	ohci->nb_iso_xmit_ctx =
		get_nb_iso_ctx(ohci, OHCI1394_IsoXmitIntMaskSet);

	/* Set the usage bits for non-existent contexts so they can't
	 * be allocated */
	ohci->ir_ctx_usage = ~0 << ohci->nb_iso_rcv_ctx;
	ohci->it_ctx_usage = ~0 << ohci->nb_iso_xmit_ctx;

	INIT_LIST_HEAD(&ohci->iso_tasklet_list);
	spin_lock_init(&ohci->iso_tasklet_list_lock);
	ohci->ISO_channel_usage = 0;
        spin_lock_init(&ohci->IR_channel_lock);

	/* Allocate the IR DMA context right here so we don't have
	 * to do it in interrupt path - note that this doesn't
	 * waste much memory and avoids the jugglery required to
	 * allocate it in IRQ path. */
	if (alloc_dma_rcv_ctx(ohci, &ohci->ir_legacy_context,
			      DMA_CTX_ISO, 0, IR_NUM_DESC,
			      IR_BUF_SIZE, IR_SPLIT_BUF_SIZE,
			      OHCI1394_IsoRcvContextBase) < 0) {
		FAIL(-ENOMEM, "Cannot allocate IR Legacy DMA context");
	}

	/* We hopefully don't have to pre-allocate IT DMA like we did
	 * for IR DMA above. Allocate it on-demand and mark inactive. */
	ohci->it_legacy_context.ohci = NULL;
	spin_lock_init(&ohci->event_lock);

	/*
	 * interrupts are disabled, all right, but... due to IRQF_SHARED we
	 * might get called anyway.  We'll see no event, of course, but
	 * we need to get to that "no event", so enough should be initialized
	 * by that point.
	 */
	if (request_irq(dev->irq, ohci_irq_handler, IRQF_SHARED,
			 OHCI1394_DRIVER_NAME, ohci))
		FAIL(-ENOMEM, "Failed to allocate shared interrupt %d", dev->irq);

	ohci->init_state = OHCI_INIT_HAVE_IRQ;
	ohci_initialize(ohci);

	/* Set certain csr values */
	host->csr.guid_hi = reg_read(ohci, OHCI1394_GUIDHi);
	host->csr.guid_lo = reg_read(ohci, OHCI1394_GUIDLo);
	host->csr.cyc_clk_acc = 100;  /* how do we determine clk accuracy? */
	host->csr.max_rec = (reg_read(ohci, OHCI1394_BusOptions) >> 12) & 0xf;
	host->csr.lnk_spd = reg_read(ohci, OHCI1394_BusOptions) & 0x7;

	if (phys_dma) {
		host->low_addr_space =
			(u64) reg_read(ohci, OHCI1394_PhyUpperBound) << 16;
		if (!host->low_addr_space)
			host->low_addr_space = OHCI1394_PHYS_UPPER_BOUND_FIXED;
	}
	host->middle_addr_space = OHCI1394_MIDDLE_ADDRESS_SPACE;

	/* Tell the highlevel this host is ready */
	if (hpsb_add_host(host))
		FAIL(-ENOMEM, "Failed to register host with highlevel");

	ohci->init_state = OHCI_INIT_DONE;

	return 0;
#undef FAIL
}

static void ohci1394_pci_remove(struct pci_dev *pdev)
{
	struct ti_ohci *ohci;
	struct device *dev;

	ohci = pci_get_drvdata(pdev);
	if (!ohci)
		return;

	dev = get_device(&ohci->host->device);

	switch (ohci->init_state) {
	case OHCI_INIT_DONE:
		hpsb_remove_host(ohci->host);

		/* Clear out BUS Options */
		reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
		reg_write(ohci, OHCI1394_BusOptions,
			  (reg_read(ohci, OHCI1394_BusOptions) & 0x0000f007) |
			  0x00ff0000);
		memset(ohci->csr_config_rom_cpu, 0, OHCI_CONFIG_ROM_LEN);

	case OHCI_INIT_HAVE_IRQ:
		/* Clear interrupt registers */
		reg_write(ohci, OHCI1394_IntMaskClear, 0xffffffff);
		reg_write(ohci, OHCI1394_IntEventClear, 0xffffffff);
		reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 0xffffffff);
		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 0xffffffff);
		reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 0xffffffff);
		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 0xffffffff);

		/* Disable IRM Contender */
		set_phy_reg(ohci, 4, ~0xc0 & get_phy_reg(ohci, 4));

		/* Clear link control register */
		reg_write(ohci, OHCI1394_LinkControlClear, 0xffffffff);

		/* Let all other nodes know to ignore us */
		ohci_devctl(ohci->host, RESET_BUS, LONG_RESET_NO_FORCE_ROOT);

		/* Soft reset before we start - this disables
		 * interrupts and clears linkEnable and LPS. */
		ohci_soft_reset(ohci);
		free_irq(ohci->dev->irq, ohci);

	case OHCI_INIT_HAVE_TXRX_BUFFERS__MAYBE:
		/* The ohci_soft_reset() stops all DMA contexts, so we
		 * dont need to do this.  */
		free_dma_rcv_ctx(&ohci->ar_req_context);
		free_dma_rcv_ctx(&ohci->ar_resp_context);
		free_dma_trm_ctx(&ohci->at_req_context);
		free_dma_trm_ctx(&ohci->at_resp_context);
		free_dma_rcv_ctx(&ohci->ir_legacy_context);
		free_dma_trm_ctx(&ohci->it_legacy_context);

	case OHCI_INIT_HAVE_SELFID_BUFFER:
		pci_free_consistent(ohci->dev, OHCI1394_SI_DMA_BUF_SIZE,
				    ohci->selfid_buf_cpu,
				    ohci->selfid_buf_bus);
		OHCI_DMA_FREE("consistent selfid_buf");

	case OHCI_INIT_HAVE_CONFIG_ROM_BUFFER:
		pci_free_consistent(ohci->dev, OHCI_CONFIG_ROM_LEN,
				    ohci->csr_config_rom_cpu,
				    ohci->csr_config_rom_bus);
		OHCI_DMA_FREE("consistent csr_config_rom");

	case OHCI_INIT_HAVE_IOMAPPING:
		iounmap(ohci->registers);

	case OHCI_INIT_HAVE_MEM_REGION:
#ifndef PCMCIA
		release_mem_region(pci_resource_start(ohci->dev, 0),
				   OHCI1394_REGISTER_SIZE);
#endif

#ifdef CONFIG_PPC_PMAC
	/* On UniNorth, power down the cable and turn off the chip clock
	 * to save power on laptops */
	if (machine_is(powermac)) {
		struct device_node* ofn = pci_device_to_OF_node(ohci->dev);

		if (ofn) {
			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
		}
	}
#endif /* CONFIG_PPC_PMAC */

	case OHCI_INIT_ALLOC_HOST:
		pci_set_drvdata(ohci->dev, NULL);
	}

	if (dev)
		put_device(dev);
}

#ifdef CONFIG_PM
static int ohci1394_pci_suspend(struct pci_dev *pdev, pm_message_t state)
{
	int err;
	struct ti_ohci *ohci = pci_get_drvdata(pdev);

	printk(KERN_INFO "%s does not fully support suspend and resume yet\n",
	       OHCI1394_DRIVER_NAME);

	if (!ohci) {
		printk(KERN_ERR "%s: tried to suspend nonexisting host\n",
		       OHCI1394_DRIVER_NAME);
		return -ENXIO;
	}
	DBGMSG("suspend called");

	/* Clear the async DMA contexts and stop using the controller */
	hpsb_bus_reset(ohci->host);

	/* See ohci1394_pci_remove() for comments on this sequence */
	reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
	reg_write(ohci, OHCI1394_BusOptions,
		  (reg_read(ohci, OHCI1394_BusOptions) & 0x0000f007) |
		  0x00ff0000);
	reg_write(ohci, OHCI1394_IntMaskClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IntEventClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 0xffffffff);
	set_phy_reg(ohci, 4, ~0xc0 & get_phy_reg(ohci, 4));
	reg_write(ohci, OHCI1394_LinkControlClear, 0xffffffff);
	ohci_devctl(ohci->host, RESET_BUS, LONG_RESET_NO_FORCE_ROOT);
	ohci_soft_reset(ohci);

	err = pci_save_state(pdev);
	if (err) {
		PRINT(KERN_ERR, "pci_save_state failed with %d", err);
		return err;
	}
	err = pci_set_power_state(pdev, pci_choose_state(pdev, state));
	if (err)
		DBGMSG("pci_set_power_state failed with %d", err);

/* PowerMac suspend code comes last */
#ifdef CONFIG_PPC_PMAC
	if (machine_is(powermac)) {
		struct device_node *ofn = pci_device_to_OF_node(pdev);

		if (ofn)
			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
	}
#endif /* CONFIG_PPC_PMAC */

	return 0;
}

static int ohci1394_pci_resume(struct pci_dev *pdev)
{
	int err;
	struct ti_ohci *ohci = pci_get_drvdata(pdev);

	if (!ohci) {
		printk(KERN_ERR "%s: tried to resume nonexisting host\n",
		       OHCI1394_DRIVER_NAME);
		return -ENXIO;
	}
	DBGMSG("resume called");

/* PowerMac resume code comes first */
#ifdef CONFIG_PPC_PMAC
	if (machine_is(powermac)) {
		struct device_node *ofn = pci_device_to_OF_node(pdev);

		if (ofn)
			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
	}
#endif /* CONFIG_PPC_PMAC */

	pci_set_power_state(pdev, PCI_D0);
	pci_restore_state(pdev);
	err = pci_enable_device(pdev);
	if (err) {
		PRINT(KERN_ERR, "pci_enable_device failed with %d", err);
		return err;
	}

	/* See ohci1394_pci_probe() for comments on this sequence */
	ohci_soft_reset(ohci);
	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_LPS);
	reg_write(ohci, OHCI1394_IntEventClear, 0xffffffff);
	reg_write(ohci, OHCI1394_IntMaskClear, 0xffffffff);
	mdelay(50);
	ohci_initialize(ohci);

	return 0;
}
#endif /* CONFIG_PM */

#define PCI_CLASS_FIREWIRE_OHCI     ((PCI_CLASS_SERIAL_FIREWIRE << 8) | 0x10)

static struct pci_device_id ohci1394_pci_tbl[] = {
	{
		.class = 	PCI_CLASS_FIREWIRE_OHCI,
		.class_mask = 	PCI_ANY_ID,
		.vendor =	PCI_ANY_ID,
		.device =	PCI_ANY_ID,
		.subvendor =	PCI_ANY_ID,
		.subdevice =	PCI_ANY_ID,
	},
	{ 0, },
};

MODULE_DEVICE_TABLE(pci, ohci1394_pci_tbl);

static struct pci_driver ohci1394_pci_driver = {
	.name =		OHCI1394_DRIVER_NAME,
	.id_table =	ohci1394_pci_tbl,
	.probe =	ohci1394_pci_probe,
	.remove =	ohci1394_pci_remove,
#ifdef CONFIG_PM
	.resume =	ohci1394_pci_resume,
	.suspend =	ohci1394_pci_suspend,
#endif
};

/***********************************
 * OHCI1394 Video Interface        *
 ***********************************/

/* essentially the only purpose of this code is to allow another
   module to hook into ohci's interrupt handler */

int ohci1394_stop_context(struct ti_ohci *ohci, int reg, char *msg)
{
	int i=0;

	/* stop the channel program if it's still running */
	reg_write(ohci, reg, 0x8000);

	/* Wait until it effectively stops */
	while (reg_read(ohci, reg) & 0x400) {
		i++;
		if (i>5000) {
			PRINT(KERN_ERR,
			      "Runaway loop while stopping context: %s...", msg ? msg : "");
			return 1;
		}

		mb();
		udelay(10);
	}
	if (msg) PRINT(KERN_ERR, "%s: dma prg stopped", msg);
	return 0;
}

void ohci1394_init_iso_tasklet(struct ohci1394_iso_tasklet *tasklet, int type,
			       void (*func)(unsigned long), unsigned long data)
{
	tasklet_init(&tasklet->tasklet, func, data);
	tasklet->type = type;
	/* We init the tasklet->link field, so we can list_del() it
	 * without worrying whether it was added to the list or not. */
	INIT_LIST_HEAD(&tasklet->link);
}

int ohci1394_register_iso_tasklet(struct ti_ohci *ohci,
				  struct ohci1394_iso_tasklet *tasklet)
{
	unsigned long flags, *usage;
	int n, i, r = -EBUSY;

	if (tasklet->type == OHCI_ISO_TRANSMIT) {
		n = ohci->nb_iso_xmit_ctx;
		usage = &ohci->it_ctx_usage;
	}
	else {
		n = ohci->nb_iso_rcv_ctx;
		usage = &ohci->ir_ctx_usage;

		/* only one receive context can be multichannel (OHCI sec 10.4.1) */
		if (tasklet->type == OHCI_ISO_MULTICHANNEL_RECEIVE) {
			if (test_and_set_bit(0, &ohci->ir_multichannel_used)) {
				return r;
			}
		}
	}

	spin_lock_irqsave(&ohci->iso_tasklet_list_lock, flags);

	for (i = 0; i < n; i++)
		if (!test_and_set_bit(i, usage)) {
			tasklet->context = i;
			list_add_tail(&tasklet->link, &ohci->iso_tasklet_list);
			r = 0;
			break;
		}

	spin_unlock_irqrestore(&ohci->iso_tasklet_list_lock, flags);

	return r;
}

void ohci1394_unregister_iso_tasklet(struct ti_ohci *ohci,
				     struct ohci1394_iso_tasklet *tasklet)
{
	unsigned long flags;

	tasklet_kill(&tasklet->tasklet);

	spin_lock_irqsave(&ohci->iso_tasklet_list_lock, flags);

	if (tasklet->type == OHCI_ISO_TRANSMIT)
		clear_bit(tasklet->context, &ohci->it_ctx_usage);
	else {
		clear_bit(tasklet->context, &ohci->ir_ctx_usage);

		if (tasklet->type == OHCI_ISO_MULTICHANNEL_RECEIVE) {
			clear_bit(0, &ohci->ir_multichannel_used);
		}
	}

	list_del(&tasklet->link);

	spin_unlock_irqrestore(&ohci->iso_tasklet_list_lock, flags);
}

EXPORT_SYMBOL(ohci1394_stop_context);
EXPORT_SYMBOL(ohci1394_init_iso_tasklet);
EXPORT_SYMBOL(ohci1394_register_iso_tasklet);
EXPORT_SYMBOL(ohci1394_unregister_iso_tasklet);

/***********************************
 * General module initialization   *
 ***********************************/

MODULE_AUTHOR("Sebastien Rougeaux <sebastien.rougeaux@anu.edu.au>");
MODULE_DESCRIPTION("Driver for PCI OHCI IEEE-1394 controllers");
MODULE_LICENSE("GPL");

static void __exit ohci1394_cleanup (void)
{
	pci_unregister_driver(&ohci1394_pci_driver);
}

static int __init ohci1394_init(void)
{
	return pci_register_driver(&ohci1394_pci_driver);
}

/* Register before most other device drivers.
 * Useful for remote debugging via physical DMA, e.g. using firescope. */
fs_initcall(ohci1394_init);
module_exit(ohci1394_cleanup);