in2000.c | searchcode

/drivers/scsi/in2000.c

Large files files are truncated, but you can click here to view the full file

/*
 *    in2000.c -  Linux device driver for the
 *                Always IN2000 ISA SCSI card.
 *
 * Copyright (c) 1996 John Shifflett, GeoLog Consulting
 *    john@geolog.com
 *    jshiffle@netcom.com
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2, 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.
 *
 * For the avoidance of doubt the "preferred form" of this code is one which
 * is in an open non patent encumbered format. Where cryptographic key signing
 * forms part of the process of creating an executable the information
 * including keys needed to generate an equivalently functional executable
 * are deemed to be part of the source code.
 *
 * Drew Eckhardt's excellent 'Generic NCR5380' sources provided
 * much of the inspiration and some of the code for this driver.
 * The Linux IN2000 driver distributed in the Linux kernels through
 * version 1.2.13 was an extremely valuable reference on the arcane
 * (and still mysterious) workings of the IN2000's fifo. It also
 * is where I lifted in2000_biosparam(), the gist of the card
 * detection scheme, and other bits of code. Many thanks to the
 * talented and courageous people who wrote, contributed to, and
 * maintained that driver (including Brad McLean, Shaun Savage,
 * Bill Earnest, Larry Doolittle, Roger Sunshine, John Luckey,
 * Matt Postiff, Peter Lu, zerucha@shell.portal.com, and Eric
 * Youngdale). I should also mention the driver written by
 * Hamish Macdonald for the (GASP!) Amiga A2091 card, included
 * in the Linux-m68k distribution; it gave me a good initial
 * understanding of the proper way to run a WD33c93 chip, and I
 * ended up stealing lots of code from it.
 *
 * _This_ driver is (I feel) an improvement over the old one in
 * several respects:
 *    -  All problems relating to the data size of a SCSI request are
 *          gone (as far as I know). The old driver couldn't handle
 *          swapping to partitions because that involved 4k blocks, nor
 *          could it deal with the st.c tape driver unmodified, because
 *          that usually involved 4k - 32k blocks. The old driver never
 *          quite got away from a morbid dependence on 2k block sizes -
 *          which of course is the size of the card's fifo.
 *
 *    -  Target Disconnection/Reconnection is now supported. Any
 *          system with more than one device active on the SCSI bus
 *          will benefit from this. The driver defaults to what I'm
 *          calling 'adaptive disconnect' - meaning that each command
 *          is evaluated individually as to whether or not it should
 *          be run with the option to disconnect/reselect (if the
 *          device chooses), or as a "SCSI-bus-hog".
 *
 *    -  Synchronous data transfers are now supported. Because there
 *          are a few devices (and many improperly terminated systems)
 *          that choke when doing sync, the default is sync DISABLED
 *          for all devices. This faster protocol can (and should!)
 *          be enabled on selected devices via the command-line.
 *
 *    -  Runtime operating parameters can now be specified through
 *       either the LILO or the 'insmod' command line. For LILO do:
 *          "in2000=blah,blah,blah"
 *       and with insmod go like:
 *          "insmod /usr/src/linux/modules/in2000.o setup_strings=blah,blah"
 *       The defaults should be good for most people. See the comment
 *       for 'setup_strings' below for more details.
 *
 *    -  The old driver relied exclusively on what the Western Digital
 *          docs call "Combination Level 2 Commands", which are a great
 *          idea in that the CPU is relieved of a lot of interrupt
 *          overhead. However, by accepting a certain (user-settable)
 *          amount of additional interrupts, this driver achieves
 *          better control over the SCSI bus, and data transfers are
 *          almost as fast while being much easier to define, track,
 *          and debug.
 *
 *    -  You can force detection of a card whose BIOS has been disabled.
 *
 *    -  Multiple IN2000 cards might almost be supported. I've tried to
 *       keep it in mind, but have no way to test...
 *
 *
 * TODO:
 *       tagged queuing. multiple cards.
 *
 *
 * NOTE:
 *       When using this or any other SCSI driver as a module, you'll
 *       find that with the stock kernel, at most _two_ SCSI hard
 *       drives will be linked into the device list (ie, usable).
 *       If your IN2000 card has more than 2 disks on its bus, you
 *       might want to change the define of 'SD_EXTRA_DEVS' in the
 *       'hosts.h' file from 2 to whatever is appropriate. It took
 *       me a while to track down this surprisingly obscure and
 *       undocumented little "feature".
 *
 *
 * People with bug reports, wish-lists, complaints, comments,
 * or improvements are asked to pah-leeez email me (John Shifflett)
 * at john@geolog.com or jshiffle@netcom.com! I'm anxious to get
 * this thing into as good a shape as possible, and I'm positive
 * there are lots of lurking bugs and "Stupid Places".
 *
 * Updated for Linux 2.5 by Alan Cox <alan@lxorguk.ukuu.org.uk>
 *	- Using new_eh handler
 *	- Hopefully got all the locking right again
 *	See "FIXME" notes for items that could do with more work
 */

#include <linux/module.h>
#include <linux/blkdev.h>
#include <linux/interrupt.h>
#include <linux/string.h>
#include <linux/delay.h>
#include <linux/proc_fs.h>
#include <linux/ioport.h>
#include <linux/stat.h>

#include <asm/io.h>
#include <asm/system.h>

#include "scsi.h"
#include <scsi/scsi_host.h>

#define IN2000_VERSION    "1.33-2.5"
#define IN2000_DATE       "2002/11/03"

#include "in2000.h"


/*
 * 'setup_strings' is a single string used to pass operating parameters and
 * settings from the kernel/module command-line to the driver. 'setup_args[]'
 * is an array of strings that define the compile-time default values for
 * these settings. If Linux boots with a LILO or insmod command-line, those
 * settings are combined with 'setup_args[]'. Note that LILO command-lines
 * are prefixed with "in2000=" while insmod uses a "setup_strings=" prefix.
 * The driver recognizes the following keywords (lower case required) and
 * arguments:
 *
 * -  ioport:addr    -Where addr is IO address of a (usually ROM-less) card.
 * -  noreset        -No optional args. Prevents SCSI bus reset at boot time.
 * -  nosync:x       -x is a bitmask where the 1st 7 bits correspond with
 *                    the 7 possible SCSI devices (bit 0 for device #0, etc).
 *                    Set a bit to PREVENT sync negotiation on that device.
 *                    The driver default is sync DISABLED on all devices.
 * -  period:ns      -ns is the minimum # of nanoseconds in a SCSI data transfer
 *                    period. Default is 500; acceptable values are 250 - 1000.
 * -  disconnect:x   -x = 0 to never allow disconnects, 2 to always allow them.
 *                    x = 1 does 'adaptive' disconnects, which is the default
 *                    and generally the best choice.
 * -  debug:x        -If 'DEBUGGING_ON' is defined, x is a bitmask that causes
 *                    various types of debug output to printed - see the DB_xxx
 *                    defines in in2000.h
 * -  proc:x         -If 'PROC_INTERFACE' is defined, x is a bitmask that
 *                    determines how the /proc interface works and what it
 *                    does - see the PR_xxx defines in in2000.h
 *
 * Syntax Notes:
 * -  Numeric arguments can be decimal or the '0x' form of hex notation. There
 *    _must_ be a colon between a keyword and its numeric argument, with no
 *    spaces.
 * -  Keywords are separated by commas, no spaces, in the standard kernel
 *    command-line manner.
 * -  A keyword in the 'nth' comma-separated command-line member will overwrite
 *    the 'nth' element of setup_args[]. A blank command-line member (in
 *    other words, a comma with no preceding keyword) will _not_ overwrite
 *    the corresponding setup_args[] element.
 *
 * A few LILO examples (for insmod, use 'setup_strings' instead of 'in2000'):
 * -  in2000=ioport:0x220,noreset
 * -  in2000=period:250,disconnect:2,nosync:0x03
 * -  in2000=debug:0x1e
 * -  in2000=proc:3
 */

/* Normally, no defaults are specified... */
static char *setup_args[] = { "", "", "", "", "", "", "", "", "" };

/* filled in by 'insmod' */
static char *setup_strings;

module_param(setup_strings, charp, 0);

static inline uchar read_3393(struct IN2000_hostdata *hostdata, uchar reg_num)
{
	write1_io(reg_num, IO_WD_ADDR);
	return read1_io(IO_WD_DATA);
}


#define READ_AUX_STAT() read1_io(IO_WD_ASR)


static inline void write_3393(struct IN2000_hostdata *hostdata, uchar reg_num, uchar value)
{
	write1_io(reg_num, IO_WD_ADDR);
	write1_io(value, IO_WD_DATA);
}


static inline void write_3393_cmd(struct IN2000_hostdata *hostdata, uchar cmd)
{
/*   while (READ_AUX_STAT() & ASR_CIP)
      printk("|");*/
	write1_io(WD_COMMAND, IO_WD_ADDR);
	write1_io(cmd, IO_WD_DATA);
}


static uchar read_1_byte(struct IN2000_hostdata *hostdata)
{
	uchar asr, x = 0;

	write_3393(hostdata, WD_CONTROL, CTRL_IDI | CTRL_EDI | CTRL_POLLED);
	write_3393_cmd(hostdata, WD_CMD_TRANS_INFO | 0x80);
	do {
		asr = READ_AUX_STAT();
		if (asr & ASR_DBR)
			x = read_3393(hostdata, WD_DATA);
	} while (!(asr & ASR_INT));
	return x;
}


static void write_3393_count(struct IN2000_hostdata *hostdata, unsigned long value)
{
	write1_io(WD_TRANSFER_COUNT_MSB, IO_WD_ADDR);
	write1_io((value >> 16), IO_WD_DATA);
	write1_io((value >> 8), IO_WD_DATA);
	write1_io(value, IO_WD_DATA);
}


static unsigned long read_3393_count(struct IN2000_hostdata *hostdata)
{
	unsigned long value;

	write1_io(WD_TRANSFER_COUNT_MSB, IO_WD_ADDR);
	value = read1_io(IO_WD_DATA) << 16;
	value |= read1_io(IO_WD_DATA) << 8;
	value |= read1_io(IO_WD_DATA);
	return value;
}


/* The 33c93 needs to be told which direction a command transfers its
 * data; we use this function to figure it out. Returns true if there
 * will be a DATA_OUT phase with this command, false otherwise.
 * (Thanks to Joerg Dorchain for the research and suggestion.)
 */
static int is_dir_out(Scsi_Cmnd * cmd)
{
	switch (cmd->cmnd[0]) {
	case WRITE_6:
	case WRITE_10:
	case WRITE_12:
	case WRITE_LONG:
	case WRITE_SAME:
	case WRITE_BUFFER:
	case WRITE_VERIFY:
	case WRITE_VERIFY_12:
	case COMPARE:
	case COPY:
	case COPY_VERIFY:
	case SEARCH_EQUAL:
	case SEARCH_HIGH:
	case SEARCH_LOW:
	case SEARCH_EQUAL_12:
	case SEARCH_HIGH_12:
	case SEARCH_LOW_12:
	case FORMAT_UNIT:
	case REASSIGN_BLOCKS:
	case RESERVE:
	case MODE_SELECT:
	case MODE_SELECT_10:
	case LOG_SELECT:
	case SEND_DIAGNOSTIC:
	case CHANGE_DEFINITION:
	case UPDATE_BLOCK:
	case SET_WINDOW:
	case MEDIUM_SCAN:
	case SEND_VOLUME_TAG:
	case 0xea:
		return 1;
	default:
		return 0;
	}
}



static struct sx_period sx_table[] = {
	{1, 0x20},
	{252, 0x20},
	{376, 0x30},
	{500, 0x40},
	{624, 0x50},
	{752, 0x60},
	{876, 0x70},
	{1000, 0x00},
	{0, 0}
};

static int round_period(unsigned int period)
{
	int x;

	for (x = 1; sx_table[x].period_ns; x++) {
		if ((period <= sx_table[x - 0].period_ns) && (period > sx_table[x - 1].period_ns)) {
			return x;
		}
	}
	return 7;
}

static uchar calc_sync_xfer(unsigned int period, unsigned int offset)
{
	uchar result;

	period *= 4;		/* convert SDTR code to ns */
	result = sx_table[round_period(period)].reg_value;
	result |= (offset < OPTIMUM_SX_OFF) ? offset : OPTIMUM_SX_OFF;
	return result;
}



static void in2000_execute(struct Scsi_Host *instance);

static int in2000_queuecommand_lck(Scsi_Cmnd * cmd, void (*done) (Scsi_Cmnd *))
{
	struct Scsi_Host *instance;
	struct IN2000_hostdata *hostdata;
	Scsi_Cmnd *tmp;

	instance = cmd->device->host;
	hostdata = (struct IN2000_hostdata *) instance->hostdata;

	DB(DB_QUEUE_COMMAND, scmd_printk(KERN_DEBUG, cmd, "Q-%02x(", cmd->cmnd[0]))

/* Set up a few fields in the Scsi_Cmnd structure for our own use:
 *  - host_scribble is the pointer to the next cmd in the input queue
 *  - scsi_done points to the routine we call when a cmd is finished
 *  - result is what you'd expect
 */
	    cmd->host_scribble = NULL;
	cmd->scsi_done = done;
	cmd->result = 0;

/* We use the Scsi_Pointer structure that's included with each command
 * as a scratchpad (as it's intended to be used!). The handy thing about
 * the SCp.xxx fields is that they're always associated with a given
 * cmd, and are preserved across disconnect-reselect. This means we
 * can pretty much ignore SAVE_POINTERS and RESTORE_POINTERS messages
 * if we keep all the critical pointers and counters in SCp:
 *  - SCp.ptr is the pointer into the RAM buffer
 *  - SCp.this_residual is the size of that buffer
 *  - SCp.buffer points to the current scatter-gather buffer
 *  - SCp.buffers_residual tells us how many S.G. buffers there are
 *  - SCp.have_data_in helps keep track of >2048 byte transfers
 *  - SCp.sent_command is not used
 *  - SCp.phase records this command's SRCID_ER bit setting
 */

	if (scsi_bufflen(cmd)) {
		cmd->SCp.buffer = scsi_sglist(cmd);
		cmd->SCp.buffers_residual = scsi_sg_count(cmd) - 1;
		cmd->SCp.ptr = sg_virt(cmd->SCp.buffer);
		cmd->SCp.this_residual = cmd->SCp.buffer->length;
	} else {
		cmd->SCp.buffer = NULL;
		cmd->SCp.buffers_residual = 0;
		cmd->SCp.ptr = NULL;
		cmd->SCp.this_residual = 0;
	}
	cmd->SCp.have_data_in = 0;

/* We don't set SCp.phase here - that's done in in2000_execute() */

/* WD docs state that at the conclusion of a "LEVEL2" command, the
 * status byte can be retrieved from the LUN register. Apparently,
 * this is the case only for *uninterrupted* LEVEL2 commands! If
 * there are any unexpected phases entered, even if they are 100%
 * legal (different devices may choose to do things differently),
 * the LEVEL2 command sequence is exited. This often occurs prior
 * to receiving the status byte, in which case the driver does a
 * status phase interrupt and gets the status byte on its own.
 * While such a command can then be "resumed" (ie restarted to
 * finish up as a LEVEL2 command), the LUN register will NOT be
 * a valid status byte at the command's conclusion, and we must
 * use the byte obtained during the earlier interrupt. Here, we
 * preset SCp.Status to an illegal value (0xff) so that when
 * this command finally completes, we can tell where the actual
 * status byte is stored.
 */

	cmd->SCp.Status = ILLEGAL_STATUS_BYTE;

/* We need to disable interrupts before messing with the input
 * queue and calling in2000_execute().
 */

	/*
	 * Add the cmd to the end of 'input_Q'. Note that REQUEST_SENSE
	 * commands are added to the head of the queue so that the desired
	 * sense data is not lost before REQUEST_SENSE executes.
	 */

	if (!(hostdata->input_Q) || (cmd->cmnd[0] == REQUEST_SENSE)) {
		cmd->host_scribble = (uchar *) hostdata->input_Q;
		hostdata->input_Q = cmd;
	} else {		/* find the end of the queue */
		for (tmp = (Scsi_Cmnd *) hostdata->input_Q; tmp->host_scribble; tmp = (Scsi_Cmnd *) tmp->host_scribble);
		tmp->host_scribble = (uchar *) cmd;
	}

/* We know that there's at least one command in 'input_Q' now.
 * Go see if any of them are runnable!
 */

	in2000_execute(cmd->device->host);

	DB(DB_QUEUE_COMMAND, printk(")Q "))
	    return 0;
}

static DEF_SCSI_QCMD(in2000_queuecommand)



/*
 * This routine attempts to start a scsi command. If the host_card is
 * already connected, we give up immediately. Otherwise, look through
 * the input_Q, using the first command we find that's intended
 * for a currently non-busy target/lun.
 * Note that this function is always called with interrupts already
 * disabled (either from in2000_queuecommand() or in2000_intr()).
 */
static void in2000_execute(struct Scsi_Host *instance)
{
	struct IN2000_hostdata *hostdata;
	Scsi_Cmnd *cmd, *prev;
	int i;
	unsigned short *sp;
	unsigned short f;
	unsigned short flushbuf[16];


	hostdata = (struct IN2000_hostdata *) instance->hostdata;

	DB(DB_EXECUTE, printk("EX("))

	    if (hostdata->selecting || hostdata->connected) {

		DB(DB_EXECUTE, printk(")EX-0 "))

		    return;
	}

	/*
	 * Search through the input_Q for a command destined
	 * for an idle target/lun.
	 */

	cmd = (Scsi_Cmnd *) hostdata->input_Q;
	prev = NULL;
	while (cmd) {
		if (!(hostdata->busy[cmd->device->id] & (1 << cmd->device->lun)))
			break;
		prev = cmd;
		cmd = (Scsi_Cmnd *) cmd->host_scribble;
	}

	/* quit if queue empty or all possible targets are busy */

	if (!cmd) {

		DB(DB_EXECUTE, printk(")EX-1 "))

		    return;
	}

	/*  remove command from queue */

	if (prev)
		prev->host_scribble = cmd->host_scribble;
	else
		hostdata->input_Q = (Scsi_Cmnd *) cmd->host_scribble;

#ifdef PROC_STATISTICS
	hostdata->cmd_cnt[cmd->device->id]++;
#endif

/*
 * Start the selection process
 */

	if (is_dir_out(cmd))
		write_3393(hostdata, WD_DESTINATION_ID, cmd->device->id);
	else
		write_3393(hostdata, WD_DESTINATION_ID, cmd->device->id | DSTID_DPD);

/* Now we need to figure out whether or not this command is a good
 * candidate for disconnect/reselect. We guess to the best of our
 * ability, based on a set of hierarchical rules. When several
 * devices are operating simultaneously, disconnects are usually
 * an advantage. In a single device system, or if only 1 device
 * is being accessed, transfers usually go faster if disconnects
 * are not allowed:
 *
 * + Commands should NEVER disconnect if hostdata->disconnect =
 *   DIS_NEVER (this holds for tape drives also), and ALWAYS
 *   disconnect if hostdata->disconnect = DIS_ALWAYS.
 * + Tape drive commands should always be allowed to disconnect.
 * + Disconnect should be allowed if disconnected_Q isn't empty.
 * + Commands should NOT disconnect if input_Q is empty.
 * + Disconnect should be allowed if there are commands in input_Q
 *   for a different target/lun. In this case, the other commands
 *   should be made disconnect-able, if not already.
 *
 * I know, I know - this code would flunk me out of any
 * "C Programming 101" class ever offered. But it's easy
 * to change around and experiment with for now.
 */

	cmd->SCp.phase = 0;	/* assume no disconnect */
	if (hostdata->disconnect == DIS_NEVER)
		goto no;
	if (hostdata->disconnect == DIS_ALWAYS)
		goto yes;
	if (cmd->device->type == 1)	/* tape drive? */
		goto yes;
	if (hostdata->disconnected_Q)	/* other commands disconnected? */
		goto yes;
	if (!(hostdata->input_Q))	/* input_Q empty? */
		goto no;
	for (prev = (Scsi_Cmnd *) hostdata->input_Q; prev; prev = (Scsi_Cmnd *) prev->host_scribble) {
		if ((prev->device->id != cmd->device->id) || (prev->device->lun != cmd->device->lun)) {
			for (prev = (Scsi_Cmnd *) hostdata->input_Q; prev; prev = (Scsi_Cmnd *) prev->host_scribble)
				prev->SCp.phase = 1;
			goto yes;
		}
	}
	goto no;

      yes:
	cmd->SCp.phase = 1;

#ifdef PROC_STATISTICS
	hostdata->disc_allowed_cnt[cmd->device->id]++;
#endif

      no:
	write_3393(hostdata, WD_SOURCE_ID, ((cmd->SCp.phase) ? SRCID_ER : 0));

	write_3393(hostdata, WD_TARGET_LUN, cmd->device->lun);
	write_3393(hostdata, WD_SYNCHRONOUS_TRANSFER, hostdata->sync_xfer[cmd->device->id]);
	hostdata->busy[cmd->device->id] |= (1 << cmd->device->lun);

	if ((hostdata->level2 <= L2_NONE) || (hostdata->sync_stat[cmd->device->id] == SS_UNSET)) {

		/*
		 * Do a 'Select-With-ATN' command. This will end with
		 * one of the following interrupts:
		 *    CSR_RESEL_AM:  failure - can try again later.
		 *    CSR_TIMEOUT:   failure - give up.
		 *    CSR_SELECT:    success - proceed.
		 */

		hostdata->selecting = cmd;

/* Every target has its own synchronous transfer setting, kept in
 * the sync_xfer array, and a corresponding status byte in sync_stat[].
 * Each target's sync_stat[] entry is initialized to SS_UNSET, and its
 * sync_xfer[] entry is initialized to the default/safe value. SS_UNSET
 * means that the parameters are undetermined as yet, and that we
 * need to send an SDTR message to this device after selection is
 * complete. We set SS_FIRST to tell the interrupt routine to do so,
 * unless we don't want to even _try_ synchronous transfers: In this
 * case we set SS_SET to make the defaults final.
 */
		if (hostdata->sync_stat[cmd->device->id] == SS_UNSET) {
			if (hostdata->sync_off & (1 << cmd->device->id))
				hostdata->sync_stat[cmd->device->id] = SS_SET;
			else
				hostdata->sync_stat[cmd->device->id] = SS_FIRST;
		}
		hostdata->state = S_SELECTING;
		write_3393_count(hostdata, 0);	/* this guarantees a DATA_PHASE interrupt */
		write_3393_cmd(hostdata, WD_CMD_SEL_ATN);
	}

	else {

		/*
		 * Do a 'Select-With-ATN-Xfer' command. This will end with
		 * one of the following interrupts:
		 *    CSR_RESEL_AM:  failure - can try again later.
		 *    CSR_TIMEOUT:   failure - give up.
		 *    anything else: success - proceed.
		 */

		hostdata->connected = cmd;
		write_3393(hostdata, WD_COMMAND_PHASE, 0);

		/* copy command_descriptor_block into WD chip
		 * (take advantage of auto-incrementing)
		 */

		write1_io(WD_CDB_1, IO_WD_ADDR);
		for (i = 0; i < cmd->cmd_len; i++)
			write1_io(cmd->cmnd[i], IO_WD_DATA);

		/* The wd33c93 only knows about Group 0, 1, and 5 commands when
		 * it's doing a 'select-and-transfer'. To be safe, we write the
		 * size of the CDB into the OWN_ID register for every case. This
		 * way there won't be problems with vendor-unique, audio, etc.
		 */

		write_3393(hostdata, WD_OWN_ID, cmd->cmd_len);

		/* When doing a non-disconnect command, we can save ourselves a DATA
		 * phase interrupt later by setting everything up now. With writes we
		 * need to pre-fill the fifo; if there's room for the 32 flush bytes,
		 * put them in there too - that'll avoid a fifo interrupt. Reads are
		 * somewhat simpler.
		 * KLUDGE NOTE: It seems that you can't completely fill the fifo here:
		 * This results in the IO_FIFO_COUNT register rolling over to zero,
		 * and apparently the gate array logic sees this as empty, not full,
		 * so the 3393 chip is never signalled to start reading from the
		 * fifo. Or maybe it's seen as a permanent fifo interrupt condition.
		 * Regardless, we fix this by temporarily pretending that the fifo
		 * is 16 bytes smaller. (I see now that the old driver has a comment
		 * about "don't fill completely" in an analogous place - must be the
		 * same deal.) This results in CDROM, swap partitions, and tape drives
		 * needing an extra interrupt per write command - I think we can live
		 * with that!
		 */

		if (!(cmd->SCp.phase)) {
			write_3393_count(hostdata, cmd->SCp.this_residual);
			write_3393(hostdata, WD_CONTROL, CTRL_IDI | CTRL_EDI | CTRL_BUS);
			write1_io(0, IO_FIFO_WRITE);	/* clear fifo counter, write mode */

			if (is_dir_out(cmd)) {
				hostdata->fifo = FI_FIFO_WRITING;
				if ((i = cmd->SCp.this_residual) > (IN2000_FIFO_SIZE - 16))
					i = IN2000_FIFO_SIZE - 16;
				cmd->SCp.have_data_in = i;	/* this much data in fifo */
				i >>= 1;	/* Gulp. Assuming modulo 2. */
				sp = (unsigned short *) cmd->SCp.ptr;
				f = hostdata->io_base + IO_FIFO;

#ifdef FAST_WRITE_IO

				FAST_WRITE2_IO();
#else
				while (i--)
					write2_io(*sp++, IO_FIFO);

#endif

				/* Is there room for the flush bytes? */

				if (cmd->SCp.have_data_in <= ((IN2000_FIFO_SIZE - 16) - 32)) {
					sp = flushbuf;
					i = 16;

#ifdef FAST_WRITE_IO

					FAST_WRITE2_IO();
#else
					while (i--)
						write2_io(0, IO_FIFO);

#endif

				}
			}

			else {
				write1_io(0, IO_FIFO_READ);	/* put fifo in read mode */
				hostdata->fifo = FI_FIFO_READING;
				cmd->SCp.have_data_in = 0;	/* nothing transferred yet */
			}

		} else {
			write_3393_count(hostdata, 0);	/* this guarantees a DATA_PHASE interrupt */
		}
		hostdata->state = S_RUNNING_LEVEL2;
		write_3393_cmd(hostdata, WD_CMD_SEL_ATN_XFER);
	}

	/*
	 * Since the SCSI bus can handle only 1 connection at a time,
	 * we get out of here now. If the selection fails, or when
	 * the command disconnects, we'll come back to this routine
	 * to search the input_Q again...
	 */

	DB(DB_EXECUTE, printk("%s)EX-2 ", (cmd->SCp.phase) ? "d:" : ""))

}



static void transfer_pio(uchar * buf, int cnt, int data_in_dir, struct IN2000_hostdata *hostdata)
{
	uchar asr;

	DB(DB_TRANSFER, printk("(%p,%d,%s)", buf, cnt, data_in_dir ? "in" : "out"))

	    write_3393(hostdata, WD_CONTROL, CTRL_IDI | CTRL_EDI | CTRL_POLLED);
	write_3393_count(hostdata, cnt);
	write_3393_cmd(hostdata, WD_CMD_TRANS_INFO);
	if (data_in_dir) {
		do {
			asr = READ_AUX_STAT();
			if (asr & ASR_DBR)
				*buf++ = read_3393(hostdata, WD_DATA);
		} while (!(asr & ASR_INT));
	} else {
		do {
			asr = READ_AUX_STAT();
			if (asr & ASR_DBR)
				write_3393(hostdata, WD_DATA, *buf++);
		} while (!(asr & ASR_INT));
	}

	/* Note: we are returning with the interrupt UN-cleared.
	 * Since (presumably) an entire I/O operation has
	 * completed, the bus phase is probably different, and
	 * the interrupt routine will discover this when it
	 * responds to the uncleared int.
	 */

}



static void transfer_bytes(Scsi_Cmnd * cmd, int data_in_dir)
{
	struct IN2000_hostdata *hostdata;
	unsigned short *sp;
	unsigned short f;
	int i;

	hostdata = (struct IN2000_hostdata *) cmd->device->host->hostdata;

/* Normally, you'd expect 'this_residual' to be non-zero here.
 * In a series of scatter-gather transfers, however, this
 * routine will usually be called with 'this_residual' equal
 * to 0 and 'buffers_residual' non-zero. This means that a
 * previous transfer completed, clearing 'this_residual', and
 * now we need to setup the next scatter-gather buffer as the
 * source or destination for THIS transfer.
 */
	if (!cmd->SCp.this_residual && cmd->SCp.buffers_residual) {
		++cmd->SCp.buffer;
		--cmd->SCp.buffers_residual;
		cmd->SCp.this_residual = cmd->SCp.buffer->length;
		cmd->SCp.ptr = sg_virt(cmd->SCp.buffer);
	}

/* Set up hardware registers */

	write_3393(hostdata, WD_SYNCHRONOUS_TRANSFER, hostdata->sync_xfer[cmd->device->id]);
	write_3393_count(hostdata, cmd->SCp.this_residual);
	write_3393(hostdata, WD_CONTROL, CTRL_IDI | CTRL_EDI | CTRL_BUS);
	write1_io(0, IO_FIFO_WRITE);	/* zero counter, assume write */

/* Reading is easy. Just issue the command and return - we'll
 * get an interrupt later when we have actual data to worry about.
 */

	if (data_in_dir) {
		write1_io(0, IO_FIFO_READ);
		if ((hostdata->level2 >= L2_DATA) || (hostdata->level2 == L2_BASIC && cmd->SCp.phase == 0)) {
			write_3393(hostdata, WD_COMMAND_PHASE, 0x45);
			write_3393_cmd(hostdata, WD_CMD_SEL_ATN_XFER);
			hostdata->state = S_RUNNING_LEVEL2;
		} else
			write_3393_cmd(hostdata, WD_CMD_TRANS_INFO);
		hostdata->fifo = FI_FIFO_READING;
		cmd->SCp.have_data_in = 0;
		return;
	}

/* Writing is more involved - we'll start the WD chip and write as
 * much data to the fifo as we can right now. Later interrupts will
 * write any bytes that don't make it at this stage.
 */

	if ((hostdata->level2 >= L2_DATA) || (hostdata->level2 == L2_BASIC && cmd->SCp.phase == 0)) {
		write_3393(hostdata, WD_COMMAND_PHASE, 0x45);
		write_3393_cmd(hostdata, WD_CMD_SEL_ATN_XFER);
		hostdata->state = S_RUNNING_LEVEL2;
	} else
		write_3393_cmd(hostdata, WD_CMD_TRANS_INFO);
	hostdata->fifo = FI_FIFO_WRITING;
	sp = (unsigned short *) cmd->SCp.ptr;

	if ((i = cmd->SCp.this_residual) > IN2000_FIFO_SIZE)
		i = IN2000_FIFO_SIZE;
	cmd->SCp.have_data_in = i;
	i >>= 1;		/* Gulp. We assume this_residual is modulo 2 */
	f = hostdata->io_base + IO_FIFO;

#ifdef FAST_WRITE_IO

	FAST_WRITE2_IO();
#else
	while (i--)
		write2_io(*sp++, IO_FIFO);

#endif

}


/* We need to use spin_lock_irqsave() & spin_unlock_irqrestore() in this
 * function in order to work in an SMP environment. (I'd be surprised
 * if the driver is ever used by anyone on a real multi-CPU motherboard,
 * but it _does_ need to be able to compile and run in an SMP kernel.)
 */

static irqreturn_t in2000_intr(int irqnum, void *dev_id)
{
	struct Scsi_Host *instance = dev_id;
	struct IN2000_hostdata *hostdata;
	Scsi_Cmnd *patch, *cmd;
	uchar asr, sr, phs, id, lun, *ucp, msg;
	int i, j;
	unsigned long length;
	unsigned short *sp;
	unsigned short f;
	unsigned long flags;

	hostdata = (struct IN2000_hostdata *) instance->hostdata;

/* Get the spin_lock and disable further ints, for SMP */

	spin_lock_irqsave(instance->host_lock, flags);

#ifdef PROC_STATISTICS
	hostdata->int_cnt++;
#endif

/* The IN2000 card has 2 interrupt sources OR'ed onto its IRQ line - the
 * WD3393 chip and the 2k fifo (which is actually a dual-port RAM combined
 * with a big logic array, so it's a little different than what you might
 * expect). As far as I know, there's no reason that BOTH can't be active
 * at the same time, but there's a problem: while we can read the 3393
 * to tell if _it_ wants an interrupt, I don't know of a way to ask the
 * fifo the same question. The best we can do is check the 3393 and if
 * it _isn't_ the source of the interrupt, then we can be pretty sure
 * that the fifo is the culprit.
 *  UPDATE: I have it on good authority (Bill Earnest) that bit 0 of the
 *          IO_FIFO_COUNT register mirrors the fifo interrupt state. I
 *          assume that bit clear means interrupt active. As it turns
 *          out, the driver really doesn't need to check for this after
 *          all, so my remarks above about a 'problem' can safely be
 *          ignored. The way the logic is set up, there's no advantage
 *          (that I can see) to worrying about it.
 *
 * It seems that the fifo interrupt signal is negated when we extract
 * bytes during read or write bytes during write.
 *  - fifo will interrupt when data is moving from it to the 3393, and
 *    there are 31 (or less?) bytes left to go. This is sort of short-
 *    sighted: what if you don't WANT to do more? In any case, our
 *    response is to push more into the fifo - either actual data or
 *    dummy bytes if need be. Note that we apparently have to write at
 *    least 32 additional bytes to the fifo after an interrupt in order
 *    to get it to release the ones it was holding on to - writing fewer
 *    than 32 will result in another fifo int.
 *  UPDATE: Again, info from Bill Earnest makes this more understandable:
 *          32 bytes = two counts of the fifo counter register. He tells
 *          me that the fifo interrupt is a non-latching signal derived
 *          from a straightforward boolean interpretation of the 7
 *          highest bits of the fifo counter and the fifo-read/fifo-write
 *          state. Who'd a thought?
 */

	write1_io(0, IO_LED_ON);
	asr = READ_AUX_STAT();
	if (!(asr & ASR_INT)) {	/* no WD33c93 interrupt? */

/* Ok. This is definitely a FIFO-only interrupt.
 *
 * If FI_FIFO_READING is set, there are up to 2048 bytes waiting to be read,
 * maybe more to come from the SCSI bus. Read as many as we can out of the
 * fifo and into memory at the location of SCp.ptr[SCp.have_data_in], and
 * update have_data_in afterwards.
 *
 * If we have FI_FIFO_WRITING, the FIFO has almost run out of bytes to move
 * into the WD3393 chip (I think the interrupt happens when there are 31
 * bytes left, but it may be fewer...). The 3393 is still waiting, so we
 * shove some more into the fifo, which gets things moving again. If the
 * original SCSI command specified more than 2048 bytes, there may still
 * be some of that data left: fine - use it (from SCp.ptr[SCp.have_data_in]).
 * Don't forget to update have_data_in. If we've already written out the
 * entire buffer, feed 32 dummy bytes to the fifo - they're needed to
 * push out the remaining real data.
 *    (Big thanks to Bill Earnest for getting me out of the mud in here.)
 */

		cmd = (Scsi_Cmnd *) hostdata->connected;	/* assume we're connected */
		CHECK_NULL(cmd, "fifo_int")

		    if (hostdata->fifo == FI_FIFO_READING) {

			DB(DB_FIFO, printk("{R:%02x} ", read1_io(IO_FIFO_COUNT)))

			    sp = (unsigned short *) (cmd->SCp.ptr + cmd->SCp.have_data_in);
			i = read1_io(IO_FIFO_COUNT) & 0xfe;
			i <<= 2;	/* # of words waiting in the fifo */
			f = hostdata->io_base + IO_FIFO;

#ifdef FAST_READ_IO

			FAST_READ2_IO();
#else
			while (i--)
				*sp++ = read2_io(IO_FIFO);

#endif

			i = sp - (unsigned short *) (cmd->SCp.ptr + cmd->SCp.have_data_in);
			i <<= 1;
			cmd->SCp.have_data_in += i;
		}

		else if (hostdata->fifo == FI_FIFO_WRITING) {

			DB(DB_FIFO, printk("{W:%02x} ", read1_io(IO_FIFO_COUNT)))

/* If all bytes have been written to the fifo, flush out the stragglers.
 * Note that while writing 16 dummy words seems arbitrary, we don't
 * have another choice that I can see. What we really want is to read
 * the 3393 transfer count register (that would tell us how many bytes
 * needed flushing), but the TRANSFER_INFO command hasn't completed
 * yet (not enough bytes!) and that register won't be accessible. So,
 * we use 16 words - a number obtained through trial and error.
 *  UPDATE: Bill says this is exactly what Always does, so there.
 *          More thanks due him for help in this section.
 */
			    if (cmd->SCp.this_residual == cmd->SCp.have_data_in) {
				i = 16;
				while (i--)	/* write 32 dummy bytes */
					write2_io(0, IO_FIFO);
			}

/* If there are still bytes left in the SCSI buffer, write as many as we
 * can out to the fifo.
 */

			else {
				sp = (unsigned short *) (cmd->SCp.ptr + cmd->SCp.have_data_in);
				i = cmd->SCp.this_residual - cmd->SCp.have_data_in;	/* bytes yet to go */
				j = read1_io(IO_FIFO_COUNT) & 0xfe;
				j <<= 2;	/* how many words the fifo has room for */
				if ((j << 1) > i)
					j = (i >> 1);
				while (j--)
					write2_io(*sp++, IO_FIFO);

				i = sp - (unsigned short *) (cmd->SCp.ptr + cmd->SCp.have_data_in);
				i <<= 1;
				cmd->SCp.have_data_in += i;
			}
		}

		else {
			printk("*** Spurious FIFO interrupt ***");
		}

		write1_io(0, IO_LED_OFF);

/* release the SMP spin_lock and restore irq state */
		spin_unlock_irqrestore(instance->host_lock, flags);
		return IRQ_HANDLED;
	}

/* This interrupt was triggered by the WD33c93 chip. The fifo interrupt
 * may also be asserted, but we don't bother to check it: we get more
 * detailed info from FIFO_READING and FIFO_WRITING (see below).
 */

	cmd = (Scsi_Cmnd *) hostdata->connected;	/* assume we're connected */
	sr = read_3393(hostdata, WD_SCSI_STATUS);	/* clear the interrupt */
	phs = read_3393(hostdata, WD_COMMAND_PHASE);

	if (!cmd && (sr != CSR_RESEL_AM && sr != CSR_TIMEOUT && sr != CSR_SELECT)) {
		printk("\nNR:wd-intr-1\n");
		write1_io(0, IO_LED_OFF);

/* release the SMP spin_lock and restore irq state */
		spin_unlock_irqrestore(instance->host_lock, flags);
		return IRQ_HANDLED;
	}

	DB(DB_INTR, printk("{%02x:%02x-", asr, sr))

/* After starting a FIFO-based transfer, the next _WD3393_ interrupt is
 * guaranteed to be in response to the completion of the transfer.
 * If we were reading, there's probably data in the fifo that needs
 * to be copied into RAM - do that here. Also, we have to update
 * 'this_residual' and 'ptr' based on the contents of the
 * TRANSFER_COUNT register, in case the device decided to do an
 * intermediate disconnect (a device may do this if it has to
 * do a seek,  or just to be nice and let other devices have
 * some bus time during long transfers).
 * After doing whatever is necessary with the fifo, we go on and
 * service the WD3393 interrupt normally.
 */
	    if (hostdata->fifo == FI_FIFO_READING) {

/* buffer index = start-of-buffer + #-of-bytes-already-read */

		sp = (unsigned short *) (cmd->SCp.ptr + cmd->SCp.have_data_in);

/* bytes remaining in fifo = (total-wanted - #-not-got) - #-already-read */

		i = (cmd->SCp.this_residual - read_3393_count(hostdata)) - cmd->SCp.have_data_in;
		i >>= 1;	/* Gulp. We assume this will always be modulo 2 */
		f = hostdata->io_base + IO_FIFO;

#ifdef FAST_READ_IO

		FAST_READ2_IO();
#else
		while (i--)
			*sp++ = read2_io(IO_FIFO);

#endif

		hostdata->fifo = FI_FIFO_UNUSED;
		length = cmd->SCp.this_residual;
		cmd->SCp.this_residual = read_3393_count(hostdata);
		cmd->SCp.ptr += (length - cmd->SCp.this_residual);

		DB(DB_TRANSFER, printk("(%p,%d)", cmd->SCp.ptr, cmd->SCp.this_residual))

	}

	else if (hostdata->fifo == FI_FIFO_WRITING) {
		hostdata->fifo = FI_FIFO_UNUSED;
		length = cmd->SCp.this_residual;
		cmd->SCp.this_residual = read_3393_count(hostdata);
		cmd->SCp.ptr += (length - cmd->SCp.this_residual);

		DB(DB_TRANSFER, printk("(%p,%d)", cmd->SCp.ptr, cmd->SCp.this_residual))

	}

/* Respond to the specific WD3393 interrupt - there are quite a few! */

	switch (sr) {

	case CSR_TIMEOUT:
		DB(DB_INTR, printk("TIMEOUT"))

		    if (hostdata->state == S_RUNNING_LEVEL2)
			hostdata->connected = NULL;
		else {
			cmd = (Scsi_Cmnd *) hostdata->selecting;	/* get a valid cmd */
			CHECK_NULL(cmd, "csr_timeout")
			    hostdata->selecting = NULL;
		}

		cmd->result = DID_NO_CONNECT << 16;
		hostdata->busy[cmd->device->id] &= ~(1 << cmd->device->lun);
		hostdata->state = S_UNCONNECTED;
		cmd->scsi_done(cmd);

/* We are not connected to a target - check to see if there
 * are commands waiting to be executed.
 */

		in2000_execute(instance);
		break;


/* Note: this interrupt should not occur in a LEVEL2 command */

	case CSR_SELECT:
		DB(DB_INTR, printk("SELECT"))
		    hostdata->connected = cmd = (Scsi_Cmnd *) hostdata->selecting;
		CHECK_NULL(cmd, "csr_select")
		    hostdata->selecting = NULL;

		/* construct an IDENTIFY message with correct disconnect bit */

		hostdata->outgoing_msg[0] = (0x80 | 0x00 | cmd->device->lun);
		if (cmd->SCp.phase)
			hostdata->outgoing_msg[0] |= 0x40;

		if (hostdata->sync_stat[cmd->device->id] == SS_FIRST) {
#ifdef SYNC_DEBUG
			printk(" sending SDTR ");
#endif

			hostdata->sync_stat[cmd->device->id] = SS_WAITING;

			/* tack on a 2nd message to ask about synchronous transfers */

			hostdata->outgoing_msg[1] = EXTENDED_MESSAGE;
			hostdata->outgoing_msg[2] = 3;
			hostdata->outgoing_msg[3] = EXTENDED_SDTR;
			hostdata->outgoing_msg[4] = OPTIMUM_SX_PER / 4;
			hostdata->outgoing_msg[5] = OPTIMUM_SX_OFF;
			hostdata->outgoing_len = 6;
		} else
			hostdata->outgoing_len = 1;

		hostdata->state = S_CONNECTED;
		break;


	case CSR_XFER_DONE | PHS_DATA_IN:
	case CSR_UNEXP | PHS_DATA_IN:
	case CSR_SRV_REQ | PHS_DATA_IN:
		DB(DB_INTR, printk("IN-%d.%d", cmd->SCp.this_residual, cmd->SCp.buffers_residual))
		    transfer_bytes(cmd, DATA_IN_DIR);
		if (hostdata->state != S_RUNNING_LEVEL2)
			hostdata->state = S_CONNECTED;
		break;


	case CSR_XFER_DONE | PHS_DATA_OUT:
	case CSR_UNEXP | PHS_DATA_OUT:
	case CSR_SRV_REQ | PHS_DATA_OUT:
		DB(DB_INTR, printk("OUT-%d.%d", cmd->SCp.this_residual, cmd->SCp.buffers_residual))
		    transfer_bytes(cmd, DATA_OUT_DIR);
		if (hostdata->state != S_RUNNING_LEVEL2)
			hostdata->state = S_CONNECTED;
		break;


/* Note: this interrupt should not occur in a LEVEL2 command */

	case CSR_XFER_DONE | PHS_COMMAND:
	case CSR_UNEXP | PHS_COMMAND:
	case CSR_SRV_REQ | PHS_COMMAND:
		DB(DB_INTR, printk("CMND-%02x", cmd->cmnd[0]))
		    transfer_pio(cmd->cmnd, cmd->cmd_len, DATA_OUT_DIR, hostdata);
		hostdata->state = S_CONNECTED;
		break;


	case CSR_XFER_DONE | PHS_STATUS:
	case CSR_UNEXP | PHS_STATUS:
	case CSR_SRV_REQ | PHS_STATUS:
		DB(DB_INTR, printk("STATUS="))

		    cmd->SCp.Status = read_1_byte(hostdata);
		DB(DB_INTR, printk("%02x", cmd->SCp.Status))
		    if (hostdata->level2 >= L2_BASIC) {
			sr = read_3393(hostdata, WD_SCSI_STATUS);	/* clear interrupt */
			hostdata->state = S_RUNNING_LEVEL2;
			write_3393(hostdata, WD_COMMAND_PHASE, 0x50);
			write_3393_cmd(hostdata, WD_CMD_SEL_ATN_XFER);
		} else {
			hostdata->state = S_CONNECTED;
		}
		break;


	case CSR_XFER_DONE | PHS_MESS_IN:
	case CSR_UNEXP | PHS_MESS_IN:
	case CSR_SRV_REQ | PHS_MESS_IN:
		DB(DB_INTR, printk("MSG_IN="))

		    msg = read_1_byte(hostdata);
		sr = read_3393(hostdata, WD_SCSI_STATUS);	/* clear interrupt */

		hostdata->incoming_msg[hostdata->incoming_ptr] = msg;
		if (hostdata->incoming_msg[0] == EXTENDED_MESSAGE)
			msg = EXTENDED_MESSAGE;
		else
			hostdata->incoming_ptr = 0;

		cmd->SCp.Message = msg;
		switch (msg) {

		case COMMAND_COMPLETE:
			DB(DB_INTR, printk("CCMP"))
			    write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
			hostdata->state = S_PRE_CMP_DISC;
			break;

		case SAVE_POINTERS:
			DB(DB_INTR, printk("SDP"))
			    write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
			hostdata->state = S_CONNECTED;
			break;

		case RESTORE_POINTERS:
			DB(DB_INTR, printk("RDP"))
			    if (hostdata->level2 >= L2_BASIC) {
				write_3393(hostdata, WD_COMMAND_PHASE, 0x45);
				write_3393_cmd(hostdata, WD_CMD_SEL_ATN_XFER);
				hostdata->state = S_RUNNING_LEVEL2;
			} else {
				write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
				hostdata->state = S_CONNECTED;
			}
			break;

		case DISCONNECT:
			DB(DB_INTR, printk("DIS"))
			    cmd->device->disconnect = 1;
			write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
			hostdata->state = S_PRE_TMP_DISC;
			break;

		case MESSAGE_REJECT:
			DB(DB_INTR, printk("REJ"))
#ifdef SYNC_DEBUG
			    printk("-REJ-");
#endif
			if (hostdata->sync_stat[cmd->device->id] == SS_WAITING)
				hostdata->sync_stat[cmd->device->id] = SS_SET;
			write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
			hostdata->state = S_CONNECTED;
			break;

		case EXTENDED_MESSAGE:
			DB(DB_INTR, printk("EXT"))

			    ucp = hostdata->incoming_msg;

#ifdef SYNC_DEBUG
			printk("%02x", ucp[hostdata->incoming_ptr]);
#endif
			/* Is this the last byte of the extended message? */

			if ((hostdata->incoming_ptr >= 2) && (hostdata->incoming_ptr == (ucp[1] + 1))) {

				switch (ucp[2]) {	/* what's the EXTENDED code? */
				case EXTENDED_SDTR:
					id = calc_sync_xfer(ucp[3], ucp[4]);
					if (hostdata->sync_stat[cmd->device->id] != SS_WAITING) {

/* A device has sent an unsolicited SDTR message; rather than go
 * through the effort of decoding it and then figuring out what
 * our reply should be, we're just gonna say that we have a
 * synchronous fifo depth of 0. This will result in asynchronous
 * transfers - not ideal but so much easier.
 * Actually, this is OK because it assures us that if we don't
 * specifically ask for sync transfers, we won't do any.
 */

						write_3393_cmd(hostdata, WD_CMD_ASSERT_ATN);	/* want MESS_OUT */
						hostdata->outgoing_msg[0] = EXTENDED_MESSAGE;
						hostdata->outgoing_msg[1] = 3;
						hostdata->outgoing_msg[2] = EXTENDED_SDTR;
						hostdata->outgoing_msg[3] = hostdata->default_sx_per / 4;
						hostdata->outgoing_msg[4] = 0;
						hostdata->outgoing_len = 5;
						hostdata->sync_xfer[cmd->device->id] = calc_sync_xfer(hostdata->default_sx_per / 4, 0);
					} else {
						hostdata->sync_xfer[cmd->device->id] = id;
					}
#ifdef SYNC_DEBUG
					printk("sync_xfer=%02x", hostdata->sync_xfer[cmd->device->id]);
#endif
					hostdata->sync_stat[cmd->device->id] = SS_SET;
					write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
					hostdata->state = S_CONNECTED;
					break;
				case EXTENDED_WDTR:
					write_3393_cmd(hostdata, WD_CMD_ASSERT_ATN);	/* want MESS_OUT */
					printk("sending WDTR ");
					hostdata->outgoing_msg[0] = EXTENDED_MESSAGE;
					hostdata->outgoing_msg[1] = 2;
					hostdata->outgoing_msg[2] = EXTENDED_WDTR;
					hostdata->outgoing_msg[3] = 0;	/* 8 bit transfer width */
					hostdata->outgoing_len = 4;
					write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
					hostdata->state = S_CONNECTED;
					break;
				default:
					write_3393_cmd(hostdata, WD_CMD_ASSERT_ATN);	/* want MESS_OUT */
					printk("Rejecting Unknown Extended Message(%02x). ", ucp[2]);
					hostdata->outgoing_msg[0] = MESSAGE_REJECT;
					hostdata->outgoing_len = 1;
					write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
					hostdata->state = S_CONNECTED;
					break;
				}
				hostdata->incoming_ptr = 0;
			}

			/* We need to read more MESS_IN bytes for the extended message */

			else {
				hostdata->incoming_ptr++;
				write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
				hostdata->state = S_CONNECTED;
			}
			break;

		default:
			printk("Rejecting Unknown Message(%02x) ", msg);
			write_3393_cmd(hostdata, WD_CMD_ASSERT_ATN);	/* want MESS_OUT */
			hostdata->outgoing_msg[0] = MESSAGE_REJECT;
			hostdata->outgoing_len = 1;
			write_3393_cmd(hostdata, WD_CMD_NEGATE_ACK);
			hostdata->state = S_CONNECTED;
		}
		break;


/* Note: this interrupt will occur only after a LEVEL2 command */

	case CSR_SEL_XFER_DONE:

/* Make sure that reselection is enabled at this point - it may
 * have been turned off for the command that just completed.
 */

		write_3393(hostdata, WD_SOURCE_ID, SRCID_ER);
		if (phs == 0x60) {
			DB(DB_INTR, printk("SX-DONE"))
			    cmd->SCp.Message = COMMAND_COMPLETE;
			lun = read_3393(hostdata, WD_TARGET_LUN);
			DB(DB_INTR, printk(":%d.%d", cmd->SCp.Status, lun))
			    hostdata->connected = NULL;
			hostdata->busy[cmd->device->id] &= ~(1 << cmd->device->lun);
			hostdata->state = S_UNCONNECTED;
			if (cmd->SCp.Status == ILLEGAL_STATUS_BYTE)
				cmd->SCp.Status = lun;
			if (cmd->cmnd[0] == REQUEST_SENSE && cmd->SCp.Status != GOOD)
				cmd->result = (cmd->result & 0x00ffff) | (DID_ERROR << 16);
			else
				cmd->result = cmd->SCp.Status | (cmd->SCp.Message << 8);
			cmd->scsi_done(cmd);

/* We are no longer connected to a target - check to see if
 * there are commands waiting to be executed.
 */

			in2000_execute(instance);
		} else {
			printk("%02x:%02x:%02x: Unknown SEL_XFER_DONE phase!!---", asr, sr, phs);
		}
		break;


/* Note: this interrupt will occur only after a LEVEL2 command */

	case CSR_SDP:
		DB(DB_INTR, printk("SDP"))
		    hostdata->state = S_RUNNING_LEVEL2;
		write_3393(hostdata, WD_COMMAND_PHASE, 0x41);
		write_3393_cmd(hostdata, WD_CMD_SEL_ATN_XFER);
		break;


	case CSR_XFER_DONE | PHS_MESS_OUT:
	case CSR_UNEXP | PHS_MESS_OUT:
	case CSR_SRV_REQ | PHS_MESS_OUT:
		DB(DB_INTR, printk("MSG_OUT="))

/* To get here, we've probably requested MESSAGE_OUT and have
 * already put the correct bytes in outgoing_msg[] and filled
 * in outgoing_len. We simply send them out to the SCSI bus.
 * Sometimes we get MESSAGE_OUT phase when we're not expecting
 * it - like when our SDTR message is rejected by a target. Some
 * targets send the REJECT before receiving all of the extended
 * message, and then seem to go back to MESSAGE_OUT for a byte
 * or two. Not sure why, or if I'm doing something wrong to
 * cause this to happen. Regardless, it seems that sending
 * NOP messages in these situations results in no harm and
 * makes everyone happy.
 */
		    if (hostdata->outgoing_len == 0) {
			hostdata->outgoing_len = 1;
			hostdata->outgoing_msg[0] = NOP;
		}
		transfer_pio(hostdata->outgoing_msg, hostdata->outgoing_len, DATA_OUT_DIR, hostdata);
		DB(DB_INTR, printk("%02x", hostdata->outgoing_msg[0]))
		    hostdata->outgoing_len = 0;
		hostdata->state = S_CONNECTED;
		break;


	case CSR_UNEXP_DISC:

/* I think I've seen this after a request-sense that was in response
 * to an error condition, but not sure. We certainly need to do
 * something when we get this interrupt - the question is 'what?'.
 * Let's think positively, and assume some command has finished
 * in a legal manner (like a command that provokes a request-sense),
 * so we treat it as a normal command-complete-disconnect.
 */


/* Make sure that reselection is enabled at this point - it may
 * have been turned off for the command that just completed.
 */

		write_3393(hostdata, WD_SOURCE_ID, SRCID_ER);
		if (cmd == NULL) {
			printk(" - Already disconnected! ");
			hostdata->state = S_UNCONNECTED;

/* release the SMP spin_lock and restore irq state */
			spin_unlock_irqrestore(instance->host_lock, flags);
			return IRQ_HANDLED;
		}
		DB(DB_INTR, printk("UNEXP_DISC"))
		    hostdata->connected = NULL;
		hostdata->busy[cmd->device->id] &= ~(1 << cmd->device->lun);
		hostdata->state = S_UNCONNECTED;
		if (cmd->cmnd[0] == REQUEST_SENSE && cmd->SCp.Status != GOOD)
			cmd->result = (cmd->result & 0x00ffff) | (DID_ERROR << 16);
		else
			cmd->result = cmd->SCp.Status | (cmd->SCp.Message << 8);
		cmd->scsi_done(cmd);

/* We are no longer connected to a target - check to see if
 * there are commands waiting to be executed.
 */

		in2000_execute(instance);
		break;


	case CSR_DISC:

/* Make sure that reselection is enabled at this point - it may
 * have been turned off for the command that just completed.
 */

		write_3393(hostdata, WD_SOURCE_ID, SRCID_ER);
		DB(DB_INTR, printk("DISC"))
		    if (cmd == NULL) {
			printk(" - Already disconnected! ");
			hostdata->state = S_UNCONNECTED;
		}
		switch (hostdata->state) {
		case S_PRE_CMP_DISC:
			hostdata->connected = NULL;
			hostdata->busy[cmd->device->id] &= ~(1 << cmd->device->lun);
			hostdata->state = S_UNCONNECTED;
			DB(DB_INTR, printk(":%d", cmd->SCp.Status))
			    if (cmd->cmnd[0] == REQUEST_SENSE && cmd->SCp.Status != GOOD)
				cmd->result = (cmd->result & 0x00ffff) | (DID_ERROR << 16);
			else
				cmd->result = cmd->SCp.Status | (cmd->SCp.Message << 8);
			cmd->scsi_done(cmd);
			break;
		case S_PRE_TMP_DISC:
		case S_RUNNING_LEVEL2:
			cmd->host_scribble = (uchar *) hostdata->disconnected_Q;
			hostdata->disconnected_Q = cmd;
			hostdata->connected = NULL;
			hostdata->state = S_UNCONNECTED;

#ifdef PROC_STATISTICS
			hostdata->disc_done_cnt[cmd->device->id]++;
#endif

			break;
		default:
			printk("*** Unexpected DISCONNECT interrupt! ***");
			hostdata->state = S_UNCONNECTED;
		}

/* We are no longer connected to a target - check to see if
 * there are commands waiting to be executed.
 */

		in2000_execute(instance);
		break;


	case CSR_RESEL_AM:
		DB(DB_INTR, printk("RESEL"))

		    /* First we have to make sure this reselection didn't */
		    /* happen during Arbitration/Selection of some other device. */
		    /* If yes, put losing command back on top of input_Q. */
		    if (hostdat…
Large files files are truncated, but you can click here to view the full file