Add Simon Shapiro's DPT driver

this shouldn't break anything existing.
Userland utilities to follow.

1 files changed, 3700 insertions, 0 deletions

diff --git a/sys/dev/dpt/dpt_scsi.c b/sys/dev/dpt/dpt_scsi.c
new file mode 100644
index 0000000..b1d8f35
--- /dev/null
+++ b/sys/dev/dpt/dpt_scsi.c
@@ -0,0 +1,3700 @@
+/**
+ *       Copyright (c) 1997 by Simon Shapiro
+ *       All Rights Reserved
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions, and the following disclaimer,
+ *    without modification, immediately at the beginning of the file.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the author may not be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
+ * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+/**
+ * dpt_scsi.c: SCSI dependant code for the DPT driver
+ *
+ * credits:	Assisted by Mike Neuffer in the early low level DPT code
+ *		Thanx to Mark Salyzyn of DPT for his assistance.
+ *		Special thanx to Justin Gibbs for invaluable help in
+ *		making this driver look and work like a FreeBSD component.
+ *		Last but not least, many thanx to UCB and the FreeBSD
+ *		team for creating and maintaining such a wonderful O/S.
+ *
+ * TODO: * Add EISA and ISA probe code.
+ *	     * Add driver-level RSID-0. This will allow interoperability with
+ *	       NiceTry, M$-Doze, Win-Dog, Slowlaris, etc. in recognizing RAID
+ *	       arrays that span controllers (Wow!).
+ */
+
+/**
+ * IMPORTANT:
+ *	There are two critical section "levels" used in this driver:
+ *	splcam() and splsoftcam().  Splcam() protects us from re-entrancy
+ *	from both our software and hardware interrupt handler.  Splsoftcam()
+ *	protects us only from our software interrupt handler.  The two
+ *	main data structures that need protection are the submitted and
+ *	completed queue.
+ *
+ *	There are three places where the submitted queue is accessed:
+ *
+ *       1.  dpt_run_queue        inserts into the queue
+ *       2.  dpt_intr             removes from the queue
+ *       3   dpt_handle_timeouts  potentially removes from the queue.
+ *
+ *	There are three places where the the completed queue is accessed:
+ *       1.  dpt_intr()            inserts into the queue
+ *       2.  dpt_sintr()           removes from the queue
+ *       3.  dpt_handle_timeouts   potentially inserts into the queue
+ */
+
+#ident "$Id: dpt_scsi.c,v 1.29 1998/01/21 04:32:08 ShimonR Exp $"
+#define _DPT_C_
+
+#include "opt_dpt.h"
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/time.h>
+#include <sys/malloc.h>
+#include <sys/buf.h>
+#include <sys/proc.h>
+#include <sys/kernel.h>
+#include <sys/acct.h>
+#include <sys/queue.h>
+
+#include <machine/endian.h>
+#include <machine/ipl.h>
+#include <scsi/scsi_all.h>
+#include <scsi/scsi_message.h>
+#include <scsi/scsiconf.h>
+#include <scsi/scsi_disk.h>
+
+#include <machine/clock.h>
+
+#include <vm/vm.h>
+#include <vm/vm_param.h>
+#include <vm/pmap.h>
+
+#include <sys/reboot.h>
+
+#include <sys/dpt.h>
+
+#ifdef INLINE
+#undef INLINE
+#endif
+
+#define INLINE	__inline
+#define INLINE_Q
+
+/* Function Prototypes */
+
+static INLINE u_int32_t dpt_inl(dpt_softc_t * dpt, u_int32_t offset);
+static INLINE u_int8_t dpt_inb(dpt_softc_t * dpt, u_int32_t offset);
+static INLINE void 
+dpt_outb(dpt_softc_t * dpt, u_int32_t offset,
+	 u_int8_t value);
+static INLINE void 
+dpt_outl(dpt_softc_t * dpt, u_int32_t offset,
+	 u_int32_t value);
+static INLINE_Q void dpt_Qpush_free(dpt_softc_t * dpt, dpt_ccb_t * ccb);
+static INLINE_Q dpt_ccb_t *dpt_Qpop_free(dpt_softc_t * dpt);
+static INLINE_Q void dpt_Qadd_waiting(dpt_softc_t * dpt, dpt_ccb_t * ccb);
+static INLINE_Q void dpt_Qpush_waiting(dpt_softc_t * dpt, dpt_ccb_t * ccb);
+static INLINE_Q void 
+dpt_Qremove_waiting(dpt_softc_t * dpt,
+		    dpt_ccb_t * ccb);
+static INLINE_Q void 
+dpt_Qadd_submitted(dpt_softc_t * dpt,
+		   dpt_ccb_t * ccb);
+static INLINE_Q void 
+dpt_Qremove_submitted(dpt_softc_t * dpt,
+		      dpt_ccb_t * ccb);
+static INLINE_Q void 
+dpt_Qadd_completed(dpt_softc_t * dpt,
+		   dpt_ccb_t * ccb);
+static INLINE_Q void 
+dpt_Qremove_completed(dpt_softc_t * dpt,
+		      dpt_ccb_t * ccb);
+static int 
+dpt_send_eata_command(dpt_softc_t * dpt,
+		      eata_ccb_t * cmd_block,
+		      u_int8_t command,
+		      int32_t retries,
+		      u_int8_t ifc, u_int8_t code,
+		      u_int8_t code2);
+static INLINE int 
+dpt_send_immediate(dpt_softc_t * dpt,
+		   eata_ccb_t * cmd_block,
+		   u_int8_t ifc, u_int8_t code,
+		   u_int8_t code2);
+static INLINE int dpt_just_reset(dpt_softc_t * dpt);
+static INLINE int dpt_raid_busy(dpt_softc_t * dpt);
+static INLINE void dpt_sched_queue(dpt_softc_t * dpt);
+
+#ifdef DPT_MEASURE_PERFORMANCE
+static void 
+dpt_IObySize(dpt_softc_t * dpt, dpt_ccb_t * ccb,
+	     int op, int index);
+#endif
+
+static void     dpt_swi_register(void *);
+
+#ifdef DPT_HANDLE_TIMEOUTS
+static void     dpt_handle_timeouts(dpt_softc_t * dpt);
+static void     dpt_timeout(void *dpt);
+#endif
+
+#ifdef DPT_LOST_IRQ
+static void     dpt_irq_timeout(void *dpt);
+#endif
+
+typedef struct scsi_inquiry_data s_inq_data_t;
+
+
+static int 
+dpt_scatter_gather(dpt_softc_t * dpt, dpt_ccb_t * ccb,
+		   u_int32_t data_length,
+		   caddr_t data);
+static int      dpt_alloc_freelist(dpt_softc_t * dpt);
+static void     dpt_run_queue(dpt_softc_t * dpt, int requests);
+static void     dpt_complete(dpt_softc_t * dpt);
+static int 
+dpt_process_completion(dpt_softc_t * dpt,
+		       dpt_ccb_t * ccb);
+static void 
+dpt_set_target(int redo, dpt_softc_t * dpt,
+	       u_int8_t bus, u_int8_t target, u_int8_t lun, int mode,
+	       u_int16_t length, u_int16_t offset, dpt_ccb_t * ccb);
+static void 
+dpt_target_ccb(dpt_softc_t * dpt, int bus, u_int8_t target, u_int8_t lun,
+	       dpt_ccb_t * ccb, int mode, u_int8_t command,
+	       u_int16_t length, u_int16_t offset);
+static void     dpt_target_done(dpt_softc_t * dpt, int bus, dpt_ccb_t * ccb);
+static void     dpt_user_cmd_done(dpt_softc_t * dpt, int bus, dpt_ccb_t * ccb);
+
+
+
+u_int8_t        dpt_blinking_led(dpt_softc_t * dpt);
+int 
+dpt_user_cmd(dpt_softc_t * dpt, eata_pt_t * user_cmd,
+	     caddr_t cmdarg, int minor_no);
+void            dpt_detect_cache(dpt_softc_t * dpt);
+void            dpt_shutdown(int howto, void *dpt);
+void            hex_dump(u_int8_t * data, int length, char *name, int no);
+char           *i2bin(unsigned int no, int length);
+dpt_conf_t     *
+dpt_get_conf(dpt_softc_t * dpt, u_int8_t page, u_int8_t target,
+	     u_int8_t size, int extent);
+dpt_inq_t      *dpt_get_board_data(dpt_softc_t * dpt, u_int32_t target_id);
+int             dpt_setup(dpt_softc_t * dpt, dpt_conf_t * conf);
+int             dpt_attach(dpt_softc_t * dpt);
+int32_t         dpt_scsi_cmd(struct scsi_xfer * xs);
+void            dptminphys(struct buf * bp);
+void            dpt_sintr(void);
+void            dpt_intr(void *arg);
+char           *scsi_cmd_name(u_int8_t cmd);
+dpt_rb_t 
+dpt_register_buffer(int unit,
+		    u_int8_t channel,
+		    u_int8_t target,
+		    u_int8_t lun,
+		    u_int8_t mode,
+		    u_int16_t length,
+		    u_int16_t offset,
+		    dpt_rec_buff callback,
+		    dpt_rb_op_t op);
+int 
+dpt_send_buffer(int unit,
+		u_int8_t channel,
+		u_int8_t target,
+		u_int8_t lun,
+		u_int8_t mode,
+		u_int16_t length,
+		u_int16_t offset,
+		void *data,
+		buff_wr_done callback);
+
+extern void     (*ihandlers[32]) __P((void));
+
+u_long          dpt_unit;	/* This one is kernel-related, do not touch! */
+
+/* The linked list of softc structures */
+TAILQ_HEAD(, dpt_softc) dpt_softc_list = TAILQ_HEAD_INITIALIZER(dpt_softc_list);
+
+/*
+ * These will have to be setup by parameters passed at boot/load time. For
+ * perfromance reasons, we make them constants for the time being.
+ */
+#define	dpt_min_segs	DPT_MAX_SEGS
+#define	dpt_max_segs	DPT_MAX_SEGS
+
+static struct scsi_adapter dpt_switch =
+{
+	dpt_scsi_cmd,
+	dptminphys,
+	NULL,
+	NULL,
+	NULL,
+	"dpt",
+	{0, 0}
+};
+
+static struct scsi_device dpt_dev =
+{
+	NULL,			/* Use default error handler */
+	NULL,			/* have a queue, served by this */
+	NULL,			/* have no async handler */
+	NULL,			/* Use default 'done' routine */
+	"dpt",
+	0,
+	{0, 0}
+};
+
+/* Software Interrupt Vector */
+
+static void
+dpt_swi_register(void *unused)
+{
+	ihandlers[SWI_CAMBIO] = dpt_sintr;
+}
+
+SYSINIT(dpt_camswi, SI_SUB_DRIVERS, SI_ORDER_FIRST, dpt_swi_register, NULL)
+/* These functions allows us to do memory mapped I/O, if hardware supported. */
+
+	static INLINE   u_int8_t
+	                dpt_inb(dpt_softc_t * dpt, u_int32_t offset)
+{
+	u_int8_t        result;
+
+	if (dpt->v_membase != NULL) {
+		result = dpt->v_membase[offset];
+	} else {
+		result = inb(dpt->io_base + offset);
+	}
+	return (result);
+}
+
+static INLINE   u_int32_t
+dpt_inl(dpt_softc_t * dpt, u_int32_t offset)
+{
+	u_int32_t       result;
+
+	if (dpt->v_membase != NULL) {
+		result = *(volatile u_int32_t *) (&dpt->v_membase[offset]);
+	} else {
+		result = inl(dpt->io_base + offset);
+	}
+	return (result);
+}
+
+static INLINE void
+dpt_outb(dpt_softc_t * dpt, u_int32_t offset, u_int8_t value)
+{
+	if (dpt->v_membase != NULL) {
+		dpt->v_membase[offset] = value;
+	} else {
+		outb(dpt->io_base + offset, value);
+	}
+}
+
+static INLINE void
+dpt_outl(dpt_softc_t * dpt, u_int32_t offset, u_int32_t value)
+{
+	if (dpt->v_membase != NULL) {
+		*(volatile u_int32_t *) (&dpt->v_membase[offset]) = value;
+	} else {
+		outl(dpt->io_base + offset, value);
+	}
+}
+
+static INLINE void
+dpt_sched_queue(dpt_softc_t * dpt)
+{
+	if (dpt->state & DPT_HA_QUIET) {
+		printf("dpt%d: Under Quiet Busses Condition.  "
+		       "No Commands are submitted\n", dpt->unit);
+		return;
+	}
+	setsoftcambio();
+}
+
+static INLINE int
+dpt_wait(dpt_softc_t * dpt, u_int8_t bits, u_int8_t state)
+{
+	int             i;
+	u_int8_t        c;
+
+	for (i = 0; i < 20000; i++) {	/* wait 20ms for not busy */
+		c = dpt_inb(dpt, HA_RSTATUS) & bits;
+		if (c == state)
+			return (0);
+		else
+			DELAY(50);
+	}
+	return (-1);
+}
+
+static INLINE int
+dpt_just_reset(dpt_softc_t * dpt)
+{
+	if ((dpt_inb(dpt, 2) == 'D')
+	    && (dpt_inb(dpt, 3) == 'P')
+	    && (dpt_inb(dpt, 4) == 'T')
+	    && (dpt_inb(dpt, 5) == 'H'))
+		return (1);
+	else
+		return (0);
+}
+
+static INLINE int
+dpt_raid_busy(dpt_softc_t * dpt)
+{
+	if ((dpt_inb(dpt, 0) == 'D')
+	    && (dpt_inb(dpt, 1) == 'P')
+	    && (dpt_inb(dpt, 2) == 'T'))
+		return (1);
+	else
+		return (0);
+}
+
+/**
+ * Build a Command Block for target mode READ/WRITE BUFFER,
+ * with the ``sync'' bit ON.
+ *
+ * Although the length and offset are 24 bit fields in the command, they cannot
+ * exceed 8192 bytes, so we take them as short integers andcheck their range.
+ *  If they are sensless, we round them to zero offset, maximum length and complain.
+ */
+
+static void
+dpt_target_ccb(dpt_softc_t * dpt, int bus, u_int8_t target, u_int8_t lun,
+	       dpt_ccb_t * ccb, int mode, u_int8_t command,
+	       u_int16_t length, u_int16_t offset)
+{
+	eata_ccb_t     *cp;
+	int             ospl;
+
+	if ((length + offset) > DPT_MAX_TARGET_MODE_BUFFER_SIZE) {
+		printf("dpt%d:  Length of %d, and offset of %d are wrong\n",
+		       dpt->unit, length, offset);
+		length = DPT_MAX_TARGET_MODE_BUFFER_SIZE;
+		offset = 0;
+	}
+	ccb->xs = NULL;
+	ccb->flags = 0;
+	ccb->state = DPT_CCB_STATE_NEW;
+	ccb->std_callback = (ccb_callback) dpt_target_done;
+	ccb->wrbuff_callback = NULL;
+
+	cp = &ccb->eata_ccb;
+	cp->CP_OpCode = EATA_CMD_DMA_SEND_CP;
+	cp->SCSI_Reset = 0;
+	cp->HBA_Init = 0;
+	cp->Auto_Req_Sen = 1;
+	cp->cp_id = target;
+	cp->DataIn = 1;
+	cp->DataOut = 0;
+	cp->Interpret = 0;
+	cp->reqlen = htonl(sizeof(struct scsi_sense_data));
+	cp->cp_statDMA = htonl(vtophys(&cp->cp_statDMA));
+	cp->cp_reqDMA = htonl(vtophys(&cp->cp_reqDMA));
+	cp->cp_viraddr = (u_int32_t) & ccb;
+
+	cp->cp_msg[0] = HA_IDENTIFY_MSG | HA_DISCO_RECO;
+
+	cp->cp_scsi_cmd = command;
+	cp->cp_cdb[1] = (u_int8_t) (mode & SCSI_TM_MODE_MASK);
+	cp->cp_lun = lun;	/* Order is important here! */
+	cp->cp_cdb[2] = 0x00;	/* Buffer Id, only 1 :-( */
+	cp->cp_cdb[3] = (length >> 16) & 0xFF;	/* Buffer offset MSB */
+	cp->cp_cdb[4] = (length >> 8) & 0xFF;
+	cp->cp_cdb[5] = length & 0xFF;
+	cp->cp_cdb[6] = (length >> 16) & 0xFF;	/* Length MSB */
+	cp->cp_cdb[7] = (length >> 8) & 0xFF;
+	cp->cp_cdb[8] = length & 0xFF;	/* Length LSB */
+	cp->cp_cdb[9] = 0;	/* No sync, no match bits */
+
+	/**
+	 * This could be optimized to live in dpt_register_buffer.
+	 *  We keep it here, just in case the kernel decides to reallocate pages
+	 */
+	if (dpt_scatter_gather(dpt, ccb, DPT_RW_BUFFER_SIZE,
+			       dpt->rw_buffer[bus][target][lun])) {
+		printf("dpt%d: Failed to setup Scatter/Gather for Target-Mode buffer\n",
+		       dpt->unit);
+	}
+}
+
+/* Setup a target mode READ command */
+
+#define cmd_ct dpt->performance.command_count[(int)ccb->eata_ccb.cp_scsi_cmd];
+
+static void
+dpt_set_target(int redo, dpt_softc_t * dpt,
+	       u_int8_t bus, u_int8_t target, u_int8_t lun, int mode,
+	       u_int16_t length, u_int16_t offset, dpt_ccb_t * ccb)
+{
+	int             ospl;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	struct timeval  now;
+#endif
+
+	if (dpt->target_mode_enabled) {
+		ospl = splcam();
+
+		if (!redo)
+			dpt_target_ccb(dpt, bus, target, lun, ccb, mode,
+				       SCSI_TM_READ_BUFFER, length, offset);
+
+		ccb->transaction_id = ++dpt->commands_processed;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+		++cmd_ct;
+		microtime(&now);
+		ccb->command_started = now;
+#endif
+		dpt_Qadd_waiting(dpt, ccb);
+		dpt_sched_queue(dpt);
+
+		splx(ospl);
+	} else {
+		printf("dpt%d:  Target Mode Request, but Target Mode is OFF\n",
+		       dpt->unit);
+	}
+}
+
+/**
+ * Schedule a buffer to be sent to another target.
+ * The work will be scheduled and the callback provided will be called when the work is
+ * actually done.
+ *
+ * Please NOTE:  ``Anyone'' can send a buffer, but only registered clients get notified
+                 of receipt of buffers.
+ */
+
+int
+dpt_send_buffer(int unit,
+		u_int8_t channel,
+		u_int8_t target,
+		u_int8_t lun,
+		u_int8_t mode,
+		u_int16_t length,
+		u_int16_t offset,
+		void *data,
+		buff_wr_done callback)
+{
+	dpt_softc_t    *dpt;
+	dpt_ccb_t      *ccb = NULL;
+	int             ospl;
+#ifdef DPT_MEASURE_PERFORMANCE
+	struct timeval  now;
+#endif
+
+	/* This is an external call.  Be a bit paranoid */
+	for (dpt = TAILQ_FIRST(&dpt_softc_list);
+	     dpt != NULL;
+	     dpt = TAILQ_NEXT(dpt, links)) {
+		if (dpt->unit == unit)
+			goto valid_unit;
+	}
+
+	return (INVALID_UNIT);
+
+valid_unit:
+
+	if (dpt->target_mode_enabled) {
+		if ((channel >= dpt->channels) || (target > dpt->max_id) ||
+		    (lun > dpt->max_lun)) {
+			return (INVALID_SENDER);
+		}
+		if ((dpt->rw_buffer[channel][target][lun] == NULL) ||
+		    (dpt->buffer_receiver[channel][target][lun] == NULL))
+			return (NOT_REGISTERED);
+
+		ospl = splsoftcam();
+		/* Process the free list */
+		if ((TAILQ_EMPTY(&dpt->free_ccbs)) && dpt_alloc_freelist(dpt)) {
+			printf("dpt%d ERROR: Cannot allocate any more free CCB's.\n"
+			       "             Please try later\n",
+			       dpt->unit);
+			splx(ospl);
+			return (NO_RESOURCES);
+		}
+		/* Now grab the newest CCB */
+		if ((ccb = dpt_Qpop_free(dpt)) == NULL) {
+			splx(ospl);
+			panic("dpt%d: Got a NULL CCB from pop_free()\n", dpt->unit);
+		}
+		splx(ospl);
+
+		bcopy(dpt->rw_buffer[channel][target][lun] + offset, data, length);
+		dpt_target_ccb(dpt, channel, target, lun, ccb, mode, SCSI_TM_WRITE_BUFFER,
+			       length, offset);
+		ccb->std_callback = (ccb_callback) callback;	/* A hack.  Potential
+								 * trouble */
+
+		ospl = splcam();
+		ccb->transaction_id = ++dpt->commands_processed;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+		++cmd_ct;
+		microtime(&now);
+		ccb->command_started = now;
+#endif
+		dpt_Qadd_waiting(dpt, ccb);
+		dpt_sched_queue(dpt);
+
+		splx(ospl);
+		return (0);
+	}
+	return (DRIVER_DOWN);
+}
+
+static void
+dpt_target_done(dpt_softc_t * dpt, int bus, dpt_ccb_t * ccb)
+{
+	int             ospl = splsoftcam();
+	eata_ccb_t     *cp;
+
+	cp = &ccb->eata_ccb;
+
+	/**
+	 * Remove the CCB from the waiting queue.
+	 *  We do NOT put it back on the free, etc., queues as it is a special
+	 * ccb, owned by the dpt_softc of this unit.
+	 */
+	ospl = splsoftcam();
+	dpt_Qremove_completed(dpt, ccb);
+	splx(ospl);
+
+#define br_channel           (ccb->eata_ccb.cp_channel)
+#define br_target            (ccb->eata_ccb.cp_id)
+#define br_lun               (ccb->eata_ccb.cp_LUN)
+#define br_index	     [br_channel][br_target][br_lun]
+#define read_buffer_callback (dpt->buffer_receiver br_index )
+#define	read_buffer	     (dpt->rw_buffer[br_channel][br_target][br_lun])
+#define cb(offset)           (ccb->eata_ccb.cp_cdb[offset])
+#define br_offset            ((cb(3) << 16) | (cb(4) << 8) | cb(5))
+#define br_length            ((cb(6) << 16) | (cb(7) << 8) | cb(8))
+
+	/* Different reasons for being here, you know... */
+	switch (ccb->eata_ccb.cp_scsi_cmd) {
+	case SCSI_TM_READ_BUFFER:
+		if (read_buffer_callback != NULL) {
+			/* This is a buffer generated by a kernel process */
+			read_buffer_callback(dpt->unit, br_channel,
+					     br_target, br_lun,
+					     read_buffer,
+					     br_offset, br_length);
+		} else {
+			/*
+			 * This is a buffer waited for by a user (sleeping)
+			 * command
+			 */
+			wakeup(ccb);
+		}
+
+		/* We ALWAYS re-issue the same command; args are don't-care  */
+		dpt_set_target(1, 0, 0, 0, 0, 0, 0, 0, 0);
+		break;
+
+	case SCSI_TM_WRITE_BUFFER:
+		(ccb->wrbuff_callback) (dpt->unit, br_channel, br_target,
+					br_offset, br_length,
+					br_lun, ccb->status_packet.hba_stat);
+		break;
+	default:
+		printf("dpt%d:  %s is an unsupported command for target mode\n",
+		       dpt->unit, scsi_cmd_name(ccb->eata_ccb.cp_scsi_cmd));
+	}
+	ospl = splsoftcam();
+	dpt->target_ccb[br_channel][br_target][br_lun] = NULL;
+	dpt_Qpush_free(dpt, ccb);
+	splx(ospl);
+
+}
+
+
+/**
+ * Use this function to register a client for a buffer read target operation.
+ * The function you register will be called every time a buffer is received
+ * by the target mode code.
+ */
+
+dpt_rb_t
+dpt_register_buffer(int unit,
+		    u_int8_t channel,
+		    u_int8_t target,
+		    u_int8_t lun,
+		    u_int8_t mode,
+		    u_int16_t length,
+		    u_int16_t offset,
+		    dpt_rec_buff callback,
+		    dpt_rb_op_t op)
+{
+	dpt_softc_t    *dpt;
+	dpt_ccb_t      *ccb = NULL;
+	int             ospl;
+
+	for (dpt = TAILQ_FIRST(&dpt_softc_list);
+	     dpt != NULL;
+	     dpt = TAILQ_NEXT(dpt, links)) {
+		if (dpt->unit == unit)
+			goto valid_unit;
+	}
+
+	return (INVALID_UNIT);
+
+valid_unit:
+
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		return (DRIVER_DOWN);
+
+	if ((channel > (dpt->channels - 1)) || (target > (dpt->max_id - 1)) ||
+	    (lun > (dpt->max_lun - 1)))
+		return (INVALID_SENDER);
+
+	if (dpt->buffer_receiver[channel][target][lun] == NULL) {
+		if (op == REGISTER_BUFFER) {
+			/* Assign the requested callback */
+			dpt->buffer_receiver[channel][target][lun] = callback;
+			/* Get a CCB */
+			ospl = splsoftcam();
+
+			/* Process the free list */
+			if ((TAILQ_EMPTY(&dpt->free_ccbs)) && dpt_alloc_freelist(dpt)) {
+				printf("dpt%d ERROR: Cannot allocate any more free CCB's.\n"
+				       "             Please try later\n",
+				       dpt->unit);
+				splx(ospl);
+				return (NO_RESOURCES);
+			}
+			/* Now grab the newest CCB */
+			if ((ccb = dpt_Qpop_free(dpt)) == NULL) {
+				splx(ospl);
+				panic("dpt%d: Got a NULL CCB from pop_free()\n", dpt->unit);
+			}
+			splx(ospl);
+
+			/* Clean up the leftover of the previous tenant */
+			ccb->status = DPT_CCB_STATE_NEW;
+			dpt->target_ccb[channel][target][lun] = ccb;
+
+			dpt->rw_buffer[channel][target][lun] = malloc(DPT_RW_BUFFER_SIZE,
+							M_DEVBUF, M_NOWAIT);
+			if (dpt->rw_buffer[channel][target][lun] == NULL) {
+				printf("dpt%d: Failed to allocate Target-Mode buffer\n",
+				       dpt->unit);
+				ospl = splsoftcam();
+				dpt_Qpush_free(dpt, ccb);
+				splx(ospl);
+				return (NO_RESOURCES);
+			}
+			dpt_set_target(0, dpt, channel, target, lun, mode, length, offset, ccb);
+			return (SUCCESSFULLY_REGISTERED);
+		} else
+			return (NOT_REGISTERED);
+	} else {
+		if (op == REGISTER_BUFFER) {
+			if (dpt->buffer_receiver[channel][target][lun] == callback)
+				return (ALREADY_REGISTERED);
+			else
+				return (REGISTERED_TO_ANOTHER);
+		} else {
+			if (dpt->buffer_receiver[channel][target][lun] == callback) {
+				dpt->buffer_receiver[channel][target][lun] = NULL;
+				ospl = splsoftcam();
+				dpt_Qpush_free(dpt, ccb);
+				splx(ospl);
+				free(dpt->rw_buffer[channel][target][lun], M_DEVBUF);
+				return (SUCCESSFULLY_REGISTERED);
+			} else
+				return (INVALID_CALLBACK);
+		}
+
+	}
+}
+
+/**
+ * This routine will try to send an EATA command to the DPT HBA.
+ * It will, by default, try AHZ times, waiting 10ms between tries.
+ * It returns 0 on success and 1 on failure.
+ * It assumes the caller protects it with splbio() or some such.
+ *
+ * IMPORTANT:  We do NOT protect the ports from multiple access in here.
+ *             You are expected to do it in the calling routine.
+ *             Here, we cannot have any clue as to the scope of your work.
+ */
+
+static int
+dpt_send_eata_command(dpt_softc_t * dpt, eata_ccb_t * cmd_block,
+		      u_int8_t command, int32_t retries,
+		      u_int8_t ifc, u_int8_t code, u_int8_t code2)
+{
+	int32_t         loop;
+	u_int8_t        result;
+	u_int32_t       test;
+	u_int32_t       swapped_cmdaddr;
+
+	if (!retries)
+		retries = 1000;
+
+	/*
+	 * I hate this polling nonsense. Wish there was a way to tell the DPT
+	 * to go get commands at its own pace,  or to interrupt when ready.
+	 * In the mean time we will measure how many itterations it really
+	 * takes.
+	 */
+	for (loop = 0; loop < retries; loop++) {
+		if ((dpt_inb(dpt, HA_RAUXSTAT) & HA_ABUSY) == 0)
+			break;
+		else
+			DELAY(50);
+	}
+
+	if (loop < retries) {
+#ifdef DPT_MEASURE_PERFORMANCE
+		if (loop > dpt->performance.max_eata_tries)
+			dpt->performance.max_eata_tries = loop;
+
+		if (loop < dpt->performance.min_eata_tries)
+			dpt->performance.min_eata_tries = loop;
+#endif
+	} else {
+#ifdef DPT_MEASURE_PERFORMANCE
+		++dpt->performance.command_too_busy;
+#endif
+		return (1);
+	}
+
+	if (cmd_block != NULL) {
+		swapped_cmdaddr = vtophys(cmd_block);
+
+#if (BYTE_ORDER == BIG_ENDIAN)
+		swapped_cmdaddr = ((swapped_cmdaddr >> 24) & 0xFF)
+			| ((swapped_cmdaddr >> 16) & 0xFF)
+			| ((swapped_cmdaddr >> 8) & 0xFF)
+			| (swapped_cmdaddr & 0xFF);
+#endif
+	} else {
+		swapped_cmdaddr = 0;
+	}
+	/* And now the address */
+	dpt_outl(dpt, HA_WDMAADDR, swapped_cmdaddr);
+
+	if (command == EATA_CMD_IMMEDIATE) {
+		if (cmd_block == NULL) {
+			dpt_outb(dpt, HA_WCODE2, code2);
+			dpt_outb(dpt, HA_WCODE, code);
+		}
+		dpt_outb(dpt, HA_WIFC, ifc);
+	}
+	dpt_outb(dpt, HA_WCOMMAND, command);
+
+	return (0);
+}
+
+/**
+ * Send a command for immediate execution by the DPT
+ * See above function for IMPORTANT notes.
+ */
+
+static INLINE int
+dpt_send_immediate(dpt_softc_t * dpt, eata_ccb_t * cmd_block,
+		   u_int8_t ifc, u_int8_t code, u_int8_t code2)
+{
+	return (dpt_send_eata_command(dpt, cmd_block, EATA_CMD_IMMEDIATE,
+				  /* retries */ 1000000, ifc, code, code2));
+}
+
+/* Return the state of the blinking DPT LED's */
+u_int8_t
+dpt_blinking_led(dpt_softc_t * dpt)
+{
+	int             ndx;
+	int             ospl;
+	u_int32_t       state;
+	u_int32_t       previous;
+	u_int8_t        result;
+
+	ospl = splcam();
+
+	result = 0;
+
+	for (ndx = 0, state = 0, previous = 0;
+	     (ndx < 10) && (state != previous);
+	     ndx++) {
+		previous = state;
+		state = dpt_inl(dpt, 1);
+	}
+
+	if ((state == previous) && (state == DPT_BLINK_INDICATOR))
+		result = dpt_inb(dpt, 5);
+
+	splx(ospl);
+	return (result);
+}
+
+/**
+ * Execute a command which did not come from the kernel's SCSI layer.
+ * The only way to map user commands to bus and target is to comply with the
+ * standard DPT wire-down scheme:
+ */
+
+int
+dpt_user_cmd(dpt_softc_t * dpt, eata_pt_t * user_cmd,
+	     caddr_t cmdarg, int minor_no)
+{
+	int             channel, target, lun;
+	int             huh;
+	int             result;
+	int             ospl;
+	int             submitted;
+	dpt_ccb_t      *ccb;
+	void           *data;
+	struct timeval  now;
+
+	data = NULL;
+	channel = minor2hba(minor_no);
+	target = minor2target(minor_no);
+	lun = minor2lun(minor_no);
+
+	if ((channel > (dpt->channels - 1))
+	    || (target > dpt->max_id)
+	    || (lun > dpt->max_lun))
+		return (ENXIO);
+
+	if (target == dpt->sc_scsi_link[channel].adapter_targ) {
+		/* This one is for the controller itself */
+		if ((user_cmd->eataID[0] != 'E')
+		    || (user_cmd->eataID[1] != 'A')
+		    || (user_cmd->eataID[2] != 'T')
+		    || (user_cmd->eataID[3] != 'A')) {
+			return (ENXIO);
+		}
+	}
+	/* Get a DPT CCB, so we can prepare a command */
+	ospl = splsoftcam();
+
+	/* Process the free list */
+	if ((TAILQ_EMPTY(&dpt->free_ccbs)) && dpt_alloc_freelist(dpt)) {
+		printf("dpt%d ERROR: Cannot allocate any more free CCB's.\n"
+		       "             Please try later\n",
+		       dpt->unit);
+		splx(ospl);
+		return (EFAULT);
+	}
+	/* Now grab the newest CCB */
+	if ((ccb = dpt_Qpop_free(dpt)) == NULL) {
+		splx(ospl);
+		panic("dpt%d: Got a NULL CCB from pop_free()\n", dpt->unit);
+	} else {
+		splx(ospl);
+		/* Clean up the leftover of the previous tenant */
+		ccb->status = DPT_CCB_STATE_NEW;
+	}
+
+	bcopy((caddr_t) & user_cmd->command_packet, (caddr_t) & ccb->eata_ccb,
+	      sizeof(eata_ccb_t));
+
+	/* We do not want to do user specified scatter/gather.  Why?? */
+	if (ccb->eata_ccb.scatter == 1)
+		return (EINVAL);
+
+	ccb->eata_ccb.Auto_Req_Sen = 1;
+	ccb->eata_ccb.reqlen = htonl(sizeof(struct scsi_sense_data));
+	ccb->eata_ccb.cp_datalen = htonl(sizeof(ccb->eata_ccb.cp_datalen));
+	ccb->eata_ccb.cp_dataDMA = htonl(vtophys(ccb->eata_ccb.cp_dataDMA));
+	ccb->eata_ccb.cp_statDMA = htonl(vtophys(&ccb->eata_ccb.cp_statDMA));
+	ccb->eata_ccb.cp_reqDMA = htonl(vtophys(&ccb->eata_ccb.cp_reqDMA));
+	ccb->eata_ccb.cp_viraddr = (u_int32_t) & ccb;
+
+	if (ccb->eata_ccb.DataIn || ccb->eata_ccb.DataOut) {
+		/* Data I/O is involved in this command.  Alocate buffer */
+		if (ccb->eata_ccb.cp_datalen > PAGE_SIZE) {
+			data = contigmalloc(ccb->eata_ccb.cp_datalen,
+					    M_TEMP, M_WAITOK, 0, ~0,
+					    ccb->eata_ccb.cp_datalen,
+					    0x10000);
+		} else {
+			data = malloc(ccb->eata_ccb.cp_datalen, M_TEMP,
+				      M_WAITOK);
+		}
+
+		if (data == NULL) {
+			printf("dpt%d: Cannot allocate %d bytes "
+			       "for EATA command\n", dpt->unit,
+			       ccb->eata_ccb.cp_datalen);
+			return (EFAULT);
+		}
+#define usr_cmd_DMA (caddr_t)user_cmd->command_packet.cp_dataDMA
+		if (ccb->eata_ccb.DataIn == 1) {
+			if (copyin(usr_cmd_DMA,
+				   data, ccb->eata_ccb.cp_datalen) == -1)
+				return (EFAULT);
+		}
+	} else {
+		/* No data I/O involved here.  Make sure the DPT knows that */
+		ccb->eata_ccb.cp_datalen = 0;
+		data = NULL;
+	}
+
+	if (ccb->eata_ccb.FWNEST == 1)
+		ccb->eata_ccb.FWNEST = 0;
+
+	if (ccb->eata_ccb.cp_datalen != 0) {
+		if (dpt_scatter_gather(dpt, ccb, ccb->eata_ccb.cp_datalen,
+				       data) != 0) {
+			if (data != NULL)
+				free(data, M_TEMP);
+			return (EFAULT);
+		}
+	}
+	/**
+	 * We are required to quiet a SCSI bus.
+	 * since we do not queue comands on a bus basis,
+	 * we wait for ALL commands on a controller to complete.
+	 * In the mean time, sched_queue() will not schedule new commands.
+	 */
+	if ((ccb->eata_ccb.cp_cdb[0] == MULTIFUNCTION_CMD)
+	    && (ccb->eata_ccb.cp_cdb[2] == BUS_QUIET)) {
+		/* We wait for ALL traffic for this HBa to subside */
+		ospl = splsoftcam();
+		dpt->state |= DPT_HA_QUIET;
+		splx(ospl);
+
+		while ((submitted = dpt->submitted_ccbs_count) != 0) {
+			huh = tsleep((void *) dpt, PCATCH | PRIBIO, "dptqt",
+				     100 * hz);
+			switch (huh) {
+			case 0:
+				/* Wakeup call received */
+				break;
+			case EWOULDBLOCK:
+				/* Timer Expired */
+				break;
+			default:
+				/* anything else */
+				break;
+			}
+		}
+	}
+	/* Resume normal operation */
+	if ((ccb->eata_ccb.cp_cdb[0] == MULTIFUNCTION_CMD)
+	    && (ccb->eata_ccb.cp_cdb[2] == BUS_UNQUIET)) {
+		ospl = splsoftcam();
+		dpt->state &= ~DPT_HA_QUIET;
+		splx(ospl);
+	}
+	/**
+	 * Schedule the command and submit it.
+	 * We bypass dpt_sched_queue, as it will block on DPT_HA_QUIET
+	 */
+	ccb->xs = NULL;
+	ccb->flags = 0;
+	ccb->eata_ccb.Auto_Req_Sen = 1;	/* We always want this feature */
+
+	ccb->transaction_id = ++dpt->commands_processed;
+	ccb->std_callback = (ccb_callback) dpt_user_cmd_done;
+	ccb->result = (u_int32_t) & cmdarg;
+	ccb->data = data;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	++dpt->performance.command_count[(int) ccb->eata_ccb.cp_scsi_cmd];
+	microtime(&now);
+	ccb->command_started = now;
+#endif
+	ospl = splcam();
+	dpt_Qadd_waiting(dpt, ccb);
+	splx(ospl);
+
+	dpt_sched_queue(dpt);
+
+	/* Wait for the command to complete */
+	(void) tsleep((void *) ccb, PCATCH | PRIBIO, "dptucw", 100 * hz);
+
+	/* Free allocated memory */
+	if (data != NULL)
+		free(data, M_TEMP);
+
+	return (0);
+}
+
+static void
+dpt_user_cmd_done(dpt_softc_t * dpt, int bus, dpt_ccb_t * ccb)
+{
+	int             ospl = splsoftcam();
+	u_int32_t       result;
+	caddr_t         cmd_arg;
+
+	/**
+	 * If Auto Request Sense is on, copyout the sense struct
+	 */
+#define usr_pckt_DMA 	(caddr_t)ntohl(ccb->eata_ccb.cp_reqDMA)
+#define usr_pckt_len	ntohl(ccb->eata_ccb.cp_datalen)
+	if (ccb->eata_ccb.Auto_Req_Sen == 1) {
+		if (copyout((caddr_t) & ccb->sense_data, usr_pckt_DMA,
+			    sizeof(struct scsi_sense_data))) {
+			ccb->result = EFAULT;
+			dpt_Qpush_free(dpt, ccb);
+			splx(ospl);
+			wakeup(ccb);
+			return;
+		}
+	}
+	/* If DataIn is on, copyout the data */
+	if ((ccb->eata_ccb.DataIn == 1)
+	    && (ccb->status_packet.hba_stat == HA_NO_ERROR)) {
+		if (copyout(ccb->data, usr_pckt_DMA, usr_pckt_len)) {
+			dpt_Qpush_free(dpt, ccb);
+			ccb->result = EFAULT;
+
+			splx(ospl);
+			wakeup(ccb);
+			return;
+		}
+	}
+	/* Copyout the status */
+	result = ccb->status_packet.hba_stat;
+	cmd_arg = (caddr_t) ccb->result;
+
+	if (copyout((caddr_t) & result, cmd_arg, sizeof(result))) {
+		dpt_Qpush_free(dpt, ccb);
+		ccb->result = EFAULT;
+		splx(ospl);
+		wakeup(ccb);
+		return;
+	}
+	/* Put the CCB back in the freelist */
+	ccb->state |= DPT_CCB_STATE_COMPLETED;
+	dpt_Qpush_free(dpt, ccb);
+
+	/* Free allocated memory */
+	splx(ospl);
+	return;
+}
+
+/* Detect Cache parameters and size */
+
+void
+dpt_detect_cache(dpt_softc_t * dpt)
+{
+	int             size;
+	int             bytes;
+	int             result;
+	int             ospl;
+	int             ndx;
+	u_int8_t        status;
+	char            name[64];
+	char           *param;
+	char           *buff;
+	eata_ccb_t      cp;
+
+	dpt_sp_t        sp;
+	struct scsi_sense_data snp;
+
+	/**
+	 * We lock out the hardware early, so that we can either complete the
+	 * operation or bust out right away.
+	 */
+
+	sprintf(name, "FreeBSD DPT Driver, version %d.%d.%d",
+		DPT_RELEASE, DPT_VERSION, DPT_PATCH);
+
+	/**
+	 * Default setting, for best perfromance..
+	 * This is what virtually all cards default to..
+	 */
+	dpt->cache_type = DPT_CACHE_WRITEBACK;
+	dpt->cache_size = 0;
+
+	if ((buff = malloc(512, M_DEVBUF, M_NOWAIT)) == NULL) {
+		printf("dpt%d: Failed to allocate %d bytes for a work "
+		       "buffer\n",
+		       dpt->unit, 512);
+		return;
+	}
+	bzero(&cp, sizeof(eata_ccb_t));
+	bzero((int8_t *) & sp, sizeof(dpt_sp_t));
+	bzero((int8_t *) & snp, sizeof(struct scsi_sense_data));
+	bzero(buff, 512);
+
+	/* Setup the command structure */
+	cp.Interpret = 1;
+	cp.DataIn = 1;
+	cp.Auto_Req_Sen = 1;
+	cp.reqlen = (u_int8_t) sizeof(struct scsi_sense_data);
+
+	cp.cp_id = 0;		/* who cares?  The HBA will interpret.. */
+	cp.cp_LUN = 0;		/* In the EATA packet */
+	cp.cp_lun = 0;		/* In the SCSI command */
+	cp.cp_channel = 0;
+
+	cp.cp_scsi_cmd = EATA_CMD_DMA_SEND_CP;
+	cp.cp_len = 56;
+	cp.cp_dataDMA = htonl(vtophys(buff));
+	cp.cp_statDMA = htonl(vtophys(&sp));
+	cp.cp_reqDMA = htonl(vtophys(&snp));
+
+	cp.cp_identify = 1;
+	cp.cp_dispri = 1;
+
+	/**
+	 * Build the EATA Command Packet structure
+	 * for a Log Sense Command.
+	 */
+
+	cp.cp_cdb[0] = 0x4d;
+	cp.cp_cdb[1] = 0x0;
+	cp.cp_cdb[2] = 0x40 | 0x33;
+	cp.cp_cdb[7] = 1;
+
+	cp.cp_datalen = htonl(512);
+
+	ospl = splcam();
+	result = dpt_send_eata_command(dpt, &cp, EATA_CMD_DMA_SEND_CP,
+				       10000, 0, 0, 0);
+	if (result != 0) {
+		printf("dpt%d WARNING: detect_cache() failed (%d) to send "
+		       "EATA_CMD_DMA_SEND_CP\n", dpt->unit, result);
+		free(buff, M_TEMP);
+		splx(ospl);
+		return;
+	}
+	/* Wait for two seconds for a response.  This can be slow... */
+	for (ndx = 0;
+	     (ndx < 20000) &&
+	     !((status = dpt_inb(dpt, HA_RAUXSTAT)) & HA_AIRQ);
+	     ndx++) {
+		DELAY(50);
+	}
+
+	/* Grab the status and clear interrupts */
+	status = dpt_inb(dpt, HA_RSTATUS);
+	splx(ospl);
+
+	/**
+	 * Sanity check
+	 */
+	if (buff[0] != 0x33) {
+		return;
+	}
+	bytes = DPT_HCP_LENGTH(buff);
+	param = DPT_HCP_FIRST(buff);
+
+	if (DPT_HCP_CODE(param) != 1) {
+		/**
+		 * DPT Log Page layout error
+		 */
+		printf("dpt%d: NOTICE: Log Page (1) layout error\n",
+		       dpt->unit);
+		return;
+	}
+	if (!(param[4] & 0x4)) {
+		dpt->cache_type = DPT_NO_CACHE;
+		return;
+	}
+	while (DPT_HCP_CODE(param) != 6) {
+		param = DPT_HCP_NEXT(param);
+		if ((param < buff)
+		    || (param >= &buff[bytes])) {
+			return;
+		}
+	}
+
+	if (param[4] & 0x2) {
+		/**
+		 * Cache disabled
+		 */
+		dpt->cache_type = DPT_NO_CACHE;
+		return;
+	}
+	if (param[4] & 0x4) {
+		dpt->cache_type = DPT_CACHE_WRITETHROUGH;
+		return;
+	}
+	dpt->cache_size = param[5]
+		| (param[6] < 8)
+		| (param[7] << 16)
+		| (param[8] << 24);
+
+	return;
+}
+
+/**
+ * Initializes the softc structure and allocate all sorts of storage.
+ * Returns 0 on good luck, 1-n otherwise (error condition sensitive).
+ */
+
+int
+dpt_setup(dpt_softc_t * dpt, dpt_conf_t * conf)
+{
+	dpt_inq_t      *board_data;
+	u_long          rev;
+	int             ndx;
+	int             ospl;
+	dpt_ccb_t      *ccb;
+
+	board_data = dpt_get_board_data(dpt, conf->scsi_id0);
+	if (board_data == NULL) {
+		printf("dpt%d ERROR: Get_board_data() failure. "
+		       "Setup ignored!\n", dpt->unit);
+		return (1);
+	}
+	dpt->total_ccbs_count = 0;
+	dpt->free_ccbs_count = 0;
+	dpt->waiting_ccbs_count = 0;
+	dpt->submitted_ccbs_count = 0;
+	dpt->completed_ccbs_count = 0;
+
+	switch (ntohl(conf->splen)) {
+	case DPT_EATA_REVA:
+		dpt->EATA_revision = 'a';
+		break;
+	case DPT_EATA_REVB:
+		dpt->EATA_revision = 'b';
+		break;
+	case DPT_EATA_REVC:
+		dpt->EATA_revision = 'c';
+		break;
+	case DPT_EATA_REVZ:
+		dpt->EATA_revision = 'z';
+		break;
+	default:
+		dpt->EATA_revision = '?';
+	}
+
+	(void) memcpy(&dpt->board_data, board_data, sizeof(dpt_inq_t));
+
+	dpt->bustype = IS_PCI;	/* We only support and operate on PCI devices */
+	dpt->channels = conf->MAX_CHAN + 1;
+	dpt->max_id = conf->MAX_ID;
+	dpt->max_lun = conf->MAX_LUN;
+	dpt->state |= DPT_HA_OK;
+
+	if (conf->SECOND)
+		dpt->primary = FALSE;
+	else
+		dpt->primary = TRUE;
+
+	dpt->more_support = conf->MORE_support;
+
+	if (board_data == NULL) {
+		rev = ('?' << 24)
+			| ('-' << 16)
+			| ('?' << 8)
+			| '-';
+	} else {
+		/* Convert from network byte order to a "string" */
+		rev = (dpt->board_data.firmware[0] << 24)
+			| (dpt->board_data.firmware[1] << 16)
+			| (dpt->board_data.firmware[2] << 8)
+			| dpt->board_data.firmware[3];
+	}
+
+	if (rev >= (('0' << 24) + ('7' << 16) + ('G' << 8) + '0'))
+		dpt->immediate_support = 1;
+	else
+		dpt->immediate_support = 0;
+
+	dpt->broken_INQUIRY = FALSE;
+
+	for (ndx = 0; ndx < MAX_CHANNELS; ndx++)
+		dpt->resetlevel[ndx] = DPT_HA_OK;
+
+	dpt->cplen = ntohl(conf->cplen);
+	dpt->cppadlen = ntohs(conf->cppadlen);
+	dpt->queuesize = ntohs(conf->queuesiz);
+
+	dpt->hostid[0] = conf->scsi_id0;
+	dpt->hostid[1] = conf->scsi_id1;
+	dpt->hostid[2] = conf->scsi_id2;
+
+	if (conf->SG_64K) {
+		dpt->sgsize = SG_SIZE_BIG;
+	} else if ((ntohs(conf->SGsiz) < 1)
+		   || (ntohs(conf->SGsiz) > SG_SIZE)) {
+		/* Just a sanity check */
+		dpt->sgsize = SG_SIZE;
+	} else {
+		dpt->sgsize = ntohs(conf->SGsiz);
+	}
+
+	if (dpt->sgsize > dpt_max_segs)
+		dpt->sgsize = dpt_max_segs;
+
+	if (dpt_alloc_freelist(dpt) != 0) {
+		return (2);
+	}
+	/* Prepare for Target Mode */
+	ospl = splsoftcam();
+	dpt->target_mode_enabled = 1;
+	splx(ospl);
+
+	return (0);
+}
+
+/**
+ * The following function returns a pointer to a buffer which MUST be freed by
+ * The caller, a la free(result, M_DEVBUF)
+ *
+ * This function (and its like) assumes it is only running during system
+ * initialization!
+ */
+dpt_inq_t      *
+dpt_get_board_data(dpt_softc_t * dpt, u_int32_t target_id)
+{
+	/* get_conf returns 512 bytes, most of which are zeros... */
+	return ((dpt_inq_t *) dpt_get_conf(dpt, 0, target_id,
+					   sizeof(dpt_inq_t), 0));
+}
+
+/**
+ * The following function returns a pointer to a buffer which MUST be freed by
+ * the caller, a la ``free(result, M_TEMP);''
+ */
+dpt_conf_t     *
+dpt_get_conf(dpt_softc_t * dpt, u_int8_t page, u_int8_t target,
+	     u_int8_t size, int extent)
+{
+	dpt_sp_t        sp;
+	eata_ccb_t      cp;
+
+	/* Get_conf returns 512 bytes, most of which are zeros... */
+	dpt_conf_t     *config;
+
+	u_short        *ip;
+	u_int8_t        status, sig1, sig2, sig3;
+
+	int             ndx;
+	int             ospl;
+	int             result;
+
+	struct scsi_sense_data snp;
+	if ((config = (dpt_conf_t *) malloc(512, M_TEMP, M_WAITOK)) == NULL)
+		return (NULL);
+
+	bzero(&cp, sizeof(eata_ccb_t));
+	bzero((int8_t *) & sp, sizeof(dpt_sp_t));
+	bzero(config, size);
+
+	cp.Interpret = 1;
+	cp.DataIn = 1;
+	cp.Auto_Req_Sen = 1;
+	cp.reqlen = sizeof(struct scsi_sense_data);
+
+	cp.cp_id = target;
+	cp.cp_LUN = 0;		/* In the EATA packet */
+	cp.cp_lun = 0;		/* In the SCSI command */
+
+	cp.cp_scsi_cmd = INQUIRY;
+	cp.cp_len = size;
+
+	cp.cp_extent = extent;
+
+	cp.cp_page = page;
+	cp.cp_channel = 0;	/* DNC, Interpret mode is set */
+	cp.cp_identify = 1;
+	cp.cp_datalen = htonl(size);
+	cp.cp_dataDMA = htonl(vtophys(config));
+	cp.cp_statDMA = htonl(vtophys(&sp));
+	cp.cp_reqDMA = htonl(vtophys(&snp));
+	cp.cp_viraddr = (u_int32_t) & cp;
+
+	ospl = splcam();
+
+#ifdef DPT_RESET_BOARD
+	printf("dpt%d: get_conf() resetting HBA at %x.\n",
+	       dpt->unit, BaseRegister(dpt));
+	dpt_outb(dpt, HA_WCOMMAND, EATA_CMD_RESET);
+	DELAY(750000);
+#endif
+
+	/**
+	 * This could be a simple for loop, but we suspected the compiler To
+	 * have optimized it a bit too much. Wait for the controller to
+	 * become ready
+	 */
+	while ((((status = dpt_inb(dpt, HA_RSTATUS)) != (HA_SREADY | HA_SSC))
+		&& (status != (HA_SREADY | HA_SSC | HA_SERROR))
+		&&		/* This results from the `wd' probe at our
+				 * addresses */
+		(status != (HA_SDRDY | HA_SERROR | HA_SDRQ)))
+	       || (dpt_wait(dpt, HA_SBUSY, 0))) {
+		/**
+		 * RAID Drives still Spinning up? (This should only occur if
+		 * the DPT controller is in a NON PC (PCI?) platform).
+		 */
+		if (dpt_raid_busy(dpt)) {
+			printf("dpt%d WARNING: Get_conf() RSUS failed for "
+			       "HBA at %x\n", dpt->unit, BaseRegister(dpt));
+			free(config, M_TEMP);
+			splx(ospl);
+			return (NULL);
+		}
+	}
+
+	DptStat_Reset_BUSY(&sp);
+
+	/**
+	 * XXXX We might want to do something more clever than aborting at
+	 * this point, like resetting (rebooting) the controller and trying
+	 * again.
+	 */
+	if ((result = dpt_send_eata_command(dpt, &cp, EATA_CMD_DMA_SEND_CP,
+					    10000, 0, 0, 0)) != 0) {
+		printf("dpt%d WARNING: Get_conf() failed (%d) to send "
+		       "EATA_CMD_DMA_READ_CONFIG\n",
+		       dpt->unit, result);
+		free(config, M_TEMP);
+		splx(ospl);
+		return (NULL);
+	}
+	/* Wait for two seconds for a response.  This can be slow  */
+	for (ndx = 0;
+	     (ndx < 20000)
+	     && !((status = dpt_inb(dpt, HA_RAUXSTAT)) & HA_AIRQ);
+	     ndx++) {
+		DELAY(50);
+	}
+
+	/* Grab the status and clear interrupts */
+	status = dpt_inb(dpt, HA_RSTATUS);
+
+	splx(ospl);
+
+	/**
+	 * Check the status carefully.  Return only if the
+	 * command was successful.
+	 */
+	if (((status & HA_SERROR) == 0)
+	    && (sp.hba_stat == 0)
+	    && (sp.scsi_stat == 0)
+	    && (sp.residue_len == 0)) {
+		return (config);
+	}
+	free(config, M_TEMP);
+	return (NULL);
+}
+
+/* This gets called once per SCSI bus defined in config! */
+
+int
+dpt_attach(dpt_softc_t * dpt)
+{
+	struct scsibus_data *scbus;
+
+	int             ndx;
+	int             idx;
+	int             channel;
+	int             target;
+	int             lun;
+
+	struct scsi_inquiry_data *inq;
+
+	for (ndx = 0; ndx < dpt->channels; ndx++) {
+		/**
+		 * We do not setup target nor lun on the assumption that
+		 * these are being set for individual devices that will be
+		 * attached to the bus later.
+		 */
+		dpt->sc_scsi_link[ndx].adapter_unit = dpt->unit;
+		dpt->sc_scsi_link[ndx].adapter_targ = dpt->hostid[ndx];
+		dpt->sc_scsi_link[ndx].fordriver = 0;
+		dpt->sc_scsi_link[ndx].adapter_softc = dpt;
+		dpt->sc_scsi_link[ndx].adapter = &dpt_switch;
+
+		/*
+		 * These appear to be the # of openings per that  DEVICE, not
+		 * the DPT!
+		 */
+		dpt->sc_scsi_link[ndx].opennings = dpt->queuesize;
+		dpt->sc_scsi_link[ndx].device = &dpt_dev;
+		dpt->sc_scsi_link[ndx].adapter_bus = ndx;
+
+		/**
+		 * Prepare the scsibus_data area for the upperlevel scsi
+		 * code.
+		 */
+		if ((scbus = scsi_alloc_bus()) == NULL)
+			return 0;
+
+		dpt->sc_scsi_link[ndx].scsibus = ndx;
+		scbus->maxtarg = dpt->max_id;
+		scbus->adapter_link = &dpt->sc_scsi_link[ndx];
+
+		/*
+		 * Invite the SCSI control layer to probe the busses.
+		 */
+
+		dpt->handle_interrupts = 1;	/* Now we are ready to work */
+		scsi_attachdevs(scbus);
+		scbus = (struct scsibus_data *) NULL;
+	}
+
+	return (1);
+}
+
+/**
+ * Allocate another chunk of CCB's. Return 0 on success, 1 otherwise.
+ * If the free list is empty, we allocate a block of entries and add them
+ * to the list.	We obtain, at most, DPT_FREE_LIST_INCREMENT CCB's at a time.
+ * If we cannot, we will try fewer entries until we succeed.
+ * For every CCB, we allocate a maximal Scatter/Gather list.
+ * This routine also initializes all the static data that pertains to this CCB.
+ */
+
+/**
+ * XXX JGibbs - How big are your SG lists?  Remeber that the kernel malloc
+ *		uses buckets and mallocs in powers of two.  So, if your
+ *		SG list is not a power of two (up to PAGESIZE), you might
+ *		waste a lot of memory.  This was the reason the ahc driver
+ *		allocats multiple SG lists at a time up to a PAGESIZE.
+ *		Just something to keep in mind.
+ * YYY Simon -  Up to 8192 entries, each entry is two ulongs, comes to 64K.
+ *              In reality they are much smaller, so you are right.
+ */
+static int
+dpt_alloc_freelist(dpt_softc_t * dpt)
+{
+	dpt_ccb_t      *nccbp;
+	dpt_sg_t       *sg;
+	u_int8_t       *buff;
+	int             ospl;
+	int             incr;
+	int             ndx;
+	int             ccb_count;
+
+	ccb_count = DPT_FREE_LIST_INCREMENT;
+
+#ifdef DPT_RESTRICTED_FREELIST
+	if (dpt->total_ccbs_count != 0) {
+		printf("dpt%d: Restricted FreeList, No more than %d entries "
+		       "allowed\n", dpt->unit, dpt->total_ccbs_count);
+		return (-1);
+	}
+#endif
+
+	/**
+	 * Allocate a group of dpt_ccb's. Work on the CCB's, one at a time
+	 */
+	ospl = splsoftcam();
+	for (ndx = 0; ndx < ccb_count; ndx++) {
+		size_t          alloc_size;
+		dpt_sg_t       *sgbuff;
+
+		alloc_size = sizeof(dpt_ccb_t);	/* About 200 bytes */
+
+		if (alloc_size > PAGE_SIZE) {
+			/*
+			 * Does not fit in a page. we try to fit in a
+			 * contigious block of memory. If not, we will, later
+			 * try to allocate smaller, and smaller chunks. There
+			 * is a tradeof between memory and performance here.
+			 * We know.this (crude) algorithm works well on
+			 * machines with plenty of memory. We have seen it
+			 * allocate in excess of 8MB.
+			 */
+			nccbp = (dpt_ccb_t *) contigmalloc(alloc_size,
+							 M_DEVBUF, M_NOWAIT,
+							   0, ~0,
+							   PAGE_SIZE,
+							   0x10000);
+		} else {
+			/* fits all in one page */
+			nccbp = (dpt_ccb_t *) malloc(alloc_size, M_DEVBUF,
+						     M_NOWAIT);
+		}
+
+		if (nccbp == (dpt_ccb_t *) NULL) {
+			printf("dpt%d ERROR: Alloc_free_list() failed to "
+			       "allocate %d\n",
+			       dpt->unit, ndx);
+			splx(ospl);
+			return (-1);
+		}
+		alloc_size = sizeof(dpt_sg_t) * dpt->sgsize;
+
+		if (alloc_size > PAGE_SIZE) {
+			/* Does not fit in a page */
+			sgbuff = (dpt_sg_t *) contigmalloc(alloc_size,
+							 M_DEVBUF, M_NOWAIT,
+							   0, ~0,
+							   PAGE_SIZE,
+							   0x10000);
+		} else {
+			/* fits all in one page */
+			sgbuff = (dpt_sg_t *) malloc(alloc_size, M_DEVBUF,
+						     M_NOWAIT);
+		}
+
+		/**
+		 * If we cannot allocate sg lists, we do not want the entire
+		 * list
+		 */
+		if (sgbuff == (dpt_sg_t *) NULL) {
+			free(nccbp, M_DEVBUF);
+			--ndx;
+			break;
+		}
+		/* Clean up the mailboxes */
+		bzero(sgbuff, alloc_size);
+		bzero(nccbp, sizeof(dpt_ccb_t));
+		/*
+		 * this line is nullified by the one below.
+		 * nccbp->eata_ccb.cp_dataDMA = (u_int32_t) sgbuff; Thanx,
+		 * Mike!
+		 */
+		nccbp->sg_list = sgbuff;
+
+		/**
+		 * Now that we have a new block of free CCB's, put them into
+		 * the free list. We always add to the head of the list and
+		 * always take form the head of the list (LIFO). Each ccb
+		 * has its own Scatter/Gather list. They are all of the same
+		 * size, Regardless of how much is used.
+		 *
+		 * While looping through all the new CCB's, we initialize them
+		 * properly. These items NEVER change; They are mostly
+		 * self-pointers, relative to the CCB itself.
+		 */
+		dpt_Qpush_free(dpt, nccbp);
+		++dpt->total_ccbs_count;
+
+		nccbp->eata_ccb.cp_dataDMA = htonl(vtophys(nccbp->sg_list));
+		nccbp->eata_ccb.cp_viraddr = (u_int32_t) nccbp;	/* Unique */
+		nccbp->eata_ccb.cp_statDMA = htonl(vtophys(&dpt->sp));
+
+		/**
+		 * See dpt_intr for why we make ALL CCB's ``have the same''
+		 * Status Packet
+		 */
+		nccbp->eata_ccb.cp_reqDMA = htonl(vtophys(&nccbp->sense_data));
+	}
+
+	splx(ospl);
+
+	return (0);
+}
+
+/**
+ * Prepare the data area for DMA.
+ */
+static int
+dpt_scatter_gather(dpt_softc_t * dpt, dpt_ccb_t * ccb, u_int32_t data_length,
+		   caddr_t data)
+{
+	int             seg;
+	int             thiskv;
+	int             bytes_this_seg;
+	int             bytes_this_page;
+	u_int32_t       datalen;
+	vm_offset_t     vaddr;
+	u_int32_t       paddr;
+	u_int32_t       nextpaddr;
+	dpt_sg_t       *sg;
+
+	/* we start with Scatter/Gather OFF */
+	ccb->eata_ccb.scatter = 0;
+
+	if (data_length) {
+		if (ccb->flags & SCSI_DATA_IN) {
+			ccb->eata_ccb.DataIn = 1;
+		}
+		if (ccb->flags & SCSI_DATA_OUT) {
+			ccb->eata_ccb.DataOut = 1;
+		}
+		seg = 0;
+		datalen = data_length;
+		vaddr = (vm_offset_t) data;
+		paddr = vtophys(vaddr);
+		ccb->eata_ccb.cp_dataDMA = htonl(vtophys(ccb->sg_list));
+		sg = ccb->sg_list;
+
+		while ((datalen > 0) && (seg < dpt->sgsize)) {
+			/* put in the base address and length */
+			sg->seg_addr = paddr;
+			sg->seg_len = 0;
+
+			/* do it at least once */
+			nextpaddr = paddr;
+
+			while ((datalen > 0) && (paddr == nextpaddr)) {
+				u_int32_t       size;
+
+				/**
+				 * This page is contiguous (physically) with
+				 * the the last, just extend the length
+				 */
+
+				/* how far to the end of the page */
+				nextpaddr = trunc_page(paddr) + PAGE_SIZE;
+
+				/* Compute the maximum size */
+
+				size = nextpaddr - paddr;
+				if (size > datalen)
+					size = datalen;
+
+				sg->seg_len += size;
+				vaddr += size;
+				datalen -= size;
+				if (datalen > 0)
+					paddr = vtophys(vaddr);
+			}
+
+			/* Next page isn't contiguous, finish the seg */
+			sg->seg_addr = htonl(sg->seg_addr);
+			sg->seg_len = htonl(sg->seg_len);
+			seg++;
+			sg++;
+		}
+
+		if (datalen) {
+			/* There's still data, must have run out of segs! */
+			printf("dpt%d: scsi_cmd() Too Many (%d) DMA segs "
+			       "(%d bytes left)\n",
+			       dpt->unit, dpt->sgsize, datalen);
+			return (1);
+		}
+		if (seg == 1) {
+			/**
+			 * After going through all this trouble, we
+			 * still have only one segment. As an
+			 * optimization measure, we will do the
+			 * I/O as a single, non-S/G operation.
+			 */
+			ccb->eata_ccb.cp_dataDMA = ccb->sg_list[0].seg_addr;
+			ccb->eata_ccb.cp_datalen = ccb->sg_list[0].seg_len;
+		} else {
+			/**
+			 * There is more than one segment. Use S/G.
+			 */
+			ccb->eata_ccb.scatter = 1;
+			ccb->eata_ccb.cp_datalen =
+				htonl(seg * sizeof(dpt_sg_t));
+		}
+	} else {		/* datalen == 0 */
+		/* No data xfer */
+		ccb->eata_ccb.cp_datalen = 0;
+		ccb->eata_ccb.cp_dataDMA = 0;
+	}
+
+	return (0);
+}
+
+/**
+ * This function obtains a CCB for a command and attempts to queue it to the
+ * Controller.
+ *
+ * CCB Obtaining: Is done by getting the first entry in the free list for the
+ * HBA. If we fail to get an scb, we send a TRY_LATER to the caller.
+ *
+ * XXX - JGibbs: XS_DRIVER_STUFFUP is equivalent to failing the I/O in the
+ *               current SCSI layer.
+ *
+ * Command Queuing: Is done by putting the command at the end of the waiting
+ * queue. This assures fair chance for all commands to be processed.
+ * If the queue was empty (has only this, current command in it, we try to
+ * submit it to the HBA. Otherwise we return SUCCESSFULLY_QUEUED.
+ */
+
+int32_t
+dpt_scsi_cmd(struct scsi_xfer * xs)
+{
+	dpt_softc_t    *dpt;
+	int             incr;
+	int             ndx;
+	int             ospl;
+	int             huh;
+
+	u_int32_t       flags;
+	dpt_ccb_t      *ccb;
+	u_int8_t        status;
+	u_int32_t       aux_status = 0;	/* Initialized to shut GCC up */
+	int             result;
+
+	int             channel, target, lun;
+
+	struct scsi_inquiry_data *inq;
+
+	dpt = (dpt_softc_t *) xs->sc_link->adapter_softc;
+
+	flags = xs->flags;
+	channel = xs->sc_link->adapter_bus;
+	target = xs->sc_link->target;
+	lun = xs->sc_link->lun;
+
+#ifdef DPT_HANDLE_TIMEOUTS
+	ospl = splsoftcam();
+	if ((dpt->state & DPT_HA_TIMEOUTS_SET) == 0) {
+		dpt->state |= DPT_HA_TIMEOUTS_SET;
+		timeout(dpt_timeout, dpt, hz * 10);
+	}
+	splx(ospl);
+#endif
+
+#ifdef DPT_LOST_IRQ
+	ospl = splcam();
+	if ((dpt->state & DPT_LOST_IRQ_SET) == 0) {
+		printf("dpt%d: Initializing Lost IRQ Timer\n", dpt->unit);
+		dpt->state |= DPT_LOST_IRQ_SET;
+		timeout(dpt_irq_timeout, dpt, hz);
+	}
+	splx(ospl);
+#endif
+
+	/**
+	 * Examine the command flags and handle properly. XXXX We are not
+	 * handling external resets right now.	Needs to be added. We do not
+	 * care about the SCSI_NOSLEEP flag as we do not sleep here. We have
+	 * to observe the SCSI_NOMASK flag, though.
+	 */
+	if (xs->flags & SCSI_RESET) {
+		printf("dpt%d: Unsupported option...\n"
+		       "      I refuse to Reset b%dt%du%d...!\n",
+		       __FILE__, __LINE__, channel, target, lun);
+		xs->error = XS_DRIVER_STUFFUP;
+		return (COMPLETE);
+	}
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE) {
+		printf("dpt%d ERROR: Command \"%s\" recieved for b%dt%du%d\n"
+		    "	    but controller is shutdown; Aborting...\n",
+		       dpt->unit,
+		       scsi_cmd_name(xs->cmd->opcode),
+		       channel, target, lun);
+		xs->error = XS_DRIVER_STUFFUP;
+		return (COMPLETE);
+	}
+	if (flags & ITSDONE) {
+		printf("dpt%d WARNING: scsi_cmd(%s) already done on "
+		       "b%dt%du%d?!\n",
+		       dpt->unit, scsi_cmd_name(xs->cmd->opcode),
+		       channel, target, lun);
+		xs->flags &= ~ITSDONE;
+	}
+	if (!(flags & INUSE)) {
+		printf("dpt%d WARNING: Unit not in use in scsi_cmd(%s) "
+		       "on b%dt%du%d?!\n",
+		       dpt->unit, scsi_cmd_name(xs->cmd->opcode), channel,
+		       target, lun);
+		xs->flags |= INUSE;
+	}
+	/**
+	 * We do not want to be disrupted when doing this, or another caller
+	 * may do the same thing.
+	 */
+	ospl = splsoftcam();
+
+	/* Process the free list */
+	if ((TAILQ_EMPTY(&dpt->free_ccbs)) && dpt_alloc_freelist(dpt)) {
+		printf("dpt%d ERROR: Cannot allocate any more free CCB's.\n"
+		       "            Will try later\n",
+		       dpt->unit);
+		xs->error = XS_DRIVER_STUFFUP;
+		splx(ospl);
+		return (COMPLETE);
+	}
+	/* Now grab the newest CCB */
+	if ((ccb = dpt_Qpop_free(dpt)) == NULL) {
+		/*
+		 * No need to panic here.  We can continue with only as many
+		 * CCBs as we have.
+		 */
+		printf("dpt%d ERROR: Got a NULL CCB from pop_free()\n",
+		       dpt->unit);
+		xs->error = XS_DRIVER_STUFFUP;
+		splx(ospl);
+		return (COMPLETE);
+	}
+#ifdef DPT_HANDLE_TIMEOUTS
+	ccb->status &= ~(DPT_CCB_STATE_ABORTED | DPT_CCB_STATE_MARKED_LOST);
+#endif
+
+	splx(ospl);
+	bcopy(xs->cmd, ccb->eata_ccb.cp_cdb, xs->cmdlen);
+
+	/* Put all the CCB population stuff below */
+	ccb->xs = xs;
+	ccb->flags = flags;
+	/* We NEVER reset the bus from a command   */
+	ccb->eata_ccb.SCSI_Reset = 0;
+	/* We NEVER re-boot the HBA from a * command */
+	ccb->eata_ccb.HBA_Init = 0;
+	ccb->eata_ccb.Auto_Req_Sen = 1;	/* We always want this feature */
+	ccb->eata_ccb.reqlen = htonl(sizeof(struct scsi_sense_data));
+	ccb->std_callback = NULL;
+	ccb->wrbuff_callback = NULL;
+
+	if (xs->sc_link->target == xs->sc_link->adapter_targ) {
+		ccb->eata_ccb.Interpret = 1;
+	} else {
+		ccb->eata_ccb.Interpret = 0;
+	}
+
+	ccb->eata_ccb.scatter = 0;	/* S/G is OFF now */
+	ccb->eata_ccb.DataIn = 0;
+	ccb->eata_ccb.DataOut = 0;
+
+	/* At this time we do not deal with the RAID internals */
+	ccb->eata_ccb.FWNEST = 0;
+	ccb->eata_ccb.Phsunit = 0;
+	/* We do not do SMARTROM kind of things */
+	ccb->eata_ccb.I_AT = 0;
+	/* We do not inhibit the cache at this time */
+	ccb->eata_ccb.Disable_Cache = 0;
+	ccb->eata_ccb.cp_channel = channel;
+	ccb->eata_ccb.cp_id = target;
+	ccb->eata_ccb.cp_LUN = lun;	/**
+					 * In the EATA packet. We do not
+					 * change the SCSI command yet
+					 */
+	/* We are currently dealing with target LUN's, not ROUTINEs */
+	ccb->eata_ccb.cp_luntar = 0;
+
+	/**
+	 * XXXX - We grant the target disconnect prvileges, except in polled
+	 * mode (????).
+	 */
+	if ((ccb->flags & SCSI_NOMASK) || !dpt->handle_interrupts) {
+		ccb->eata_ccb.cp_dispri = 0;
+	} else {
+		ccb->eata_ccb.cp_dispri = 1;
+	}
+
+	/* we always ask for Identify */
+	ccb->eata_ccb.cp_identify = 1;
+
+	/**
+	 * These three are used for command queues and tags. How do we use
+	 * them?
+	 *
+	 * XXX - JGibbs: Most likely like so: ccb->eata_ccb.cp_msg[0] =
+	 * MSG_SIMPLEQ_TAG; ccb->eata_ccb.cp_msg[1] = tagid;
+	 * ccb->eata_ccb.cp_msg[2] = 0;
+	 *
+	 * YYY - Shimon: Thanx!	We still do not do that as the current
+	 * firmware does it automatically, including on RAID arrays.
+	 */
+
+	ccb->eata_ccb.cp_msg[0] = 0;
+	ccb->eata_ccb.cp_msg[1] = 0;
+	ccb->eata_ccb.cp_msg[2] = 0;
+
+	/* End of CCB population */
+
+	if (dpt_scatter_gather(dpt, ccb, xs->datalen, xs->data) != 0) {
+		xs->error = XS_DRIVER_STUFFUP;
+		ospl = splsoftcam();
+		dpt_Qpush_free(dpt, ccb);
+		splx(ospl);
+		return (COMPLETE);
+	}
+	xs->resid = 0;
+	xs->status = 0;
+
+	/**
+	 * This is the polled mode section. If we are here to honor
+	 * SCSI_NOMASK, during scsi_attachdevs(), please notice that
+	 * interrupts are ENABLED in the system (2.2.1) and that the DPT
+	 * WILL generate them, unless we turn them off!
+	 */
+
+	/**
+	 * XXX - JGibbs: Polled mode was a botch at best. It's nice to
+	 *               know that it goes completely away with the CAM code.
+	 * YYY - Simon:  Take it out once the rest is stable. Be careful about
+	 *               how you wait for commands to complete when you switch
+	 *               to interrupt mode in the scanning code (initiated by
+	 *               scsi_attachdevs).
+	 *               Disabling it in 2.2 causes a hung system.
+	 */
+
+	if ((ccb->flags & SCSI_NOMASK) || !dpt->handle_interrupts) {
+		/**
+		 * This is an ``immediate'' command.	Poll it! We poll by
+		 * partially bypassing the queues. We first submit the
+		 * command by asking dpt_run_queue() to queue it. Then we
+		 * poll its status packet, until it completes. Then we give
+		 * it to dpt_process_completion() to analyze and then we
+		 * return.
+		 */
+
+		/*
+		 * Increase the number of commands queueable for a device. We
+		 * force each device to the maximum allowed for its HBA. This
+		 * appears wrong but all it will do is cause excessive
+		 * commands to sit in our queue. On the other hand, we can
+		 * burst as many commands as the DPT can take for a single
+		 * device. We do it here, so only while in polled mode (early
+		 * boot) do we waste time on it.	We have no clean way
+		 * to overrule sdattach() zeal in depressing the opennings
+		 * back to one if it is more than 1.
+		 */
+		if (xs->sc_link->opennings < dpt->queuesize) {
+			xs->sc_link->opennings = dpt->queuesize;
+		}
+		/**
+		 * This test only protects us from submitting polled
+		 * commands during Non-polled times.  We assumed polled
+		 * commands go in serially, one at a time. BTW, we have NOT
+		 * checked, nor verified the scope of the disaster that WILL
+		 * follow going into polled mode after being in interrupt
+		 * mode for any length of time.
+		 */
+		if (dpt->submitted_ccbs_count < dpt->queuesize) {
+			/**
+			 * Submit the request to the DPT. Unfortunately, ALL
+			 * this must be done as an atomic operation :-(
+			 */
+			ccb->eata_ccb.cp_viraddr = (u_int32_t) & ccb;
+#define dpt_SP		htonl(vtophys(&ccb->status_packet))
+#define dpt_sense	htonl(vtophys(&ccb->sense_data))
+			ccb->eata_ccb.cp_statDMA = dpt_SP;
+			ccb->eata_ccb.cp_reqDMA = dpt_sense;
+
+			/* Try to queue a command */
+			ospl = splcam();
+			result = dpt_send_eata_command(dpt, &ccb->eata_ccb,
+						       EATA_CMD_DMA_SEND_CP,
+						       0, 0, 0, 0);
+
+			if (result != 0) {
+				dpt_Qpush_free(dpt, ccb);
+				xs->error = XS_DRIVER_STUFFUP;
+				splx(ospl);
+				return (COMPLETE);
+			}
+		} else {
+			xs->error = XS_DRIVER_STUFFUP;
+			dpt_Qpush_free(dpt, ccb);
+			splx(ospl);
+			return (COMPLETE);
+		}
+
+		for (ndx = 0;
+		     (ndx < xs->timeout)
+		     && !((aux_status = dpt_inb(dpt, HA_RAUXSTAT))
+			  & HA_AIRQ);
+		     ndx++) {
+			DELAY(50);
+		}
+
+		/**
+		 * Get the status and clear the interrupt flag on the
+		 * controller
+		 */
+		status = dpt_inb(dpt, HA_RSTATUS);
+		splx(ospl);
+
+		ccb->status_reg = status;
+		ccb->aux_status_reg = aux_status;
+		/* This will setup the xs flags */
+		dpt_process_completion(dpt, ccb);
+
+		if (status & HA_SERROR) {
+			ospl = splsoftcam();
+			dpt_Qpush_free(dpt, ccb);
+			splx(ospl);
+			return (COMPLETE);
+		}
+		ospl = splsoftcam();
+		dpt_Qpush_free(dpt, ccb);
+		splx(ospl);
+		return (COMPLETE);
+	} else {
+		struct timeval  junk;
+
+		/**
+		 * Not a polled command.
+		 * The command can be queued normally.
+		 * We start a critical section PRIOR to submitting to the DPT,
+		 * and end it AFTER it moves to the submitted queue.
+		 * If not, we cal (and will!) be hit with a completion
+		 * interrupt while the command is in suspense between states.
+		 */
+
+		ospl = splsoftcam();
+		ccb->transaction_id = ++dpt->commands_processed;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+#define cmd_ndx (int)ccb->eata_ccb.cp_scsi_cmd
+		++dpt->performance.command_count[cmd_ndx];
+		microtime(&junk);
+		ccb->command_started = junk;
+#endif
+		dpt_Qadd_waiting(dpt, ccb);
+		splx(ospl);
+
+		dpt_sched_queue(dpt);
+	}
+
+	return (SUCCESSFULLY_QUEUED);
+}
+
+/**
+ * This function returns the transfer size in bytes,
+ * as a function of the maximum number of Scatter/Gather
+ * segments.  It should do so for a given HBA, but right now it returns
+ * dpt_min_segs, which is the SMALLEST number, from the ``weakest'' HBA found.
+ */
+
+void
+dptminphys(struct buf * bp)
+{
+	/**
+	 * This IS a performance sensitive routine.
+	 * It gets called at least once per I/O. Sometimes more
+	 */
+
+	if (dpt_min_segs == 0) {
+		panic("DPT:  Minphys without attach!\n");
+	}
+	if (bp->b_bcount > ((dpt_min_segs - 1) * PAGE_SIZE)) {
+#ifdef DPT_DEBUG_MINPHYS
+		printf("DPT: Block size of %x is larger than %x.  Truncating\n",
+		       bp->b_bcount, ((dpt_min_segs - 1) * PAGE_SIZE));
+#endif
+		bp->b_bcount = ((dpt_min_segs - 1) * PAGE_SIZE);
+	}
+}
+
+/*
+ * This function goes to the waiting queue, peels off a request, gives it to
+ * the DPT HBA and returns. It takes care of some housekeeping details first.
+ * The requests argument tells us how many requests to try and send to the
+ * DPT. A requests = 0 will attempt to send as many as the controller can
+ * take.
+ */
+
+static void
+dpt_run_queue(dpt_softc_t * dpt, int requests)
+{
+	int             req;
+	int             ospl;
+	int             ndx;
+	int             result;
+
+	u_int8_t        status, aux_status;
+
+	eata_ccb_t     *ccb;
+	dpt_ccb_t      *dccb;
+
+	if (TAILQ_EMPTY(&dpt->waiting_ccbs)) {
+		return;		/* Nothing to do if the list is empty */
+	}
+	if (!requests)
+		requests = dpt->queuesize;
+
+	/* Main work loop */
+	for (req = 0; (req < requests) && dpt->waiting_ccbs_count
+	     && (dpt->submitted_ccbs_count < dpt->queuesize); req++) {
+		/**
+		 * Move the request from the waiting list to the submitted
+		 * list, and submit to the DPT.
+		 * We enter a critical section BEFORE even looking at the
+		 * queue, and exit it AFTER the ccb has moved to a
+		 * destination queue.
+		 * This is normally the submitted queue but can be the waiting
+		 * queue again, if pushing the command into the DPT failed.
+		 */
+
+		ospl = splsoftcam();
+		dccb = TAILQ_FIRST(&dpt->waiting_ccbs);
+
+		if (dccb == NULL) {
+			/* We have yet to see one report of this condition */
+			panic("dpt%d ERROR: Race condition in run_queue "
+			      "(w%ds%d)\n",
+			      dpt->unit, dpt->waiting_ccbs_count,
+			      dpt->submitted_ccbs_count);
+			splx(ospl);
+			return;
+		}
+		dpt_Qremove_waiting(dpt, dccb);
+		splx(ospl);
+
+		/**
+		 * Assign exact values here. We manipulate these values
+		 * indirectly elsewhere, so BE CAREFUL!
+		 */
+		dccb->eata_ccb.cp_viraddr = (u_int32_t) dccb;
+		dccb->eata_ccb.cp_statDMA = htonl(vtophys(&dpt->sp));
+		dccb->eata_ccb.cp_reqDMA = htonl(vtophys(&dccb->sense_data));
+
+		if (dccb->xs != NULL)
+			bzero(&dccb->xs->sense, sizeof(struct scsi_sense_data));
+
+		/* Try to queue a command */
+		ospl = splcam();
+
+		if ((result = dpt_send_eata_command(dpt, &dccb->eata_ccb,
+						    EATA_CMD_DMA_SEND_CP, 0,
+						    0, 0, 0)) != 0) {
+			dpt_Qpush_waiting(dpt, dccb);
+			splx(ospl);
+			return;
+		}
+		dpt_Qadd_submitted(dpt, dccb);
+		splx(ospl);
+	}
+}
+
+/**
+ * This is the interrupt handler for the DPT driver.
+ * This routine runs at splcam (or whatever was configured for this device).
+ */
+
+void
+dpt_intr(void *arg)
+{
+	dpt_softc_t    *dpt;
+	dpt_softc_t    *ldpt;
+
+	u_int8_t        status, aux_status;
+
+	dpt_ccb_t      *dccb;
+	dpt_ccb_t      *tccb;
+	eata_ccb_t     *ccb;
+
+	dpt = (dpt_softc_t *) arg;
+
+#ifdef DPT_INTR_DELAY
+	DELAY(DPT_INTR_DELAY);
+#endif
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	{
+		struct timeval  junk;
+
+		microtime(&junk);
+		dpt->performance.intr_started = junk;
+	}
+#endif
+
+	/* First order of business is to check if this interrupt is for us */
+	aux_status = dpt_inb(dpt, HA_RAUXSTAT);
+	if (!(aux_status & HA_AIRQ)) {
+#ifdef DPT_LOST_IRQ
+		if (dpt->state & DPT_LOST_IRQ_ACTIVE) {
+			dpt->state &= ~DPT_LOST_IRQ_ACTIVE;
+			return;
+		}
+#endif
+#ifdef DPT_MEASURE_PERFORMANCE
+		++dpt->performance.spurious_interrupts;
+#endif
+		return;
+	}
+	if (!dpt->handle_interrupts) {
+#ifdef DPT_MEASURE_PERFORMANCE
+		++dpt->performance.aborted_interrupts;
+#endif
+		status = dpt_inb(dpt, HA_RSTATUS);	/* This CLEARS
+							 * interrupts */
+		return;
+	}
+	/**
+	 * What we want to do now, is to capture the status, all of it, move
+	 * it where it belongs, wake up whoever sleeps waiting to process
+	 * this result, and get out of here.
+	 */
+
+	dccb = dpt->sp.ccb;	/**
+				 * There is a very SERIOUS and dangerous
+				 * assumption here. We assume that EVERY
+				 * interrupt is in response to some request we
+				 * put to the DPT. IOW, we assume that the
+				 * Virtual Address of CP always has a valid
+				 * pointer that we put in! How will the DPT
+				 * behave if it is in Target mode? How does it
+				 * (and our driver) know it switches from
+				 * Initiator to target? What will the SP be
+				 * when a target mode interrupt is received?
+				 */
+
+#ifdef DPT_VERIFY_HINTR
+	dpt->sp.ccb = (dpt_ccb_t *) 0x55555555;
+#else
+	dpt->sp.ccb = (dpt_ccb_t *) NULL;
+#endif
+
+#ifdef DPT_HANDLE_TIMEOUTS
+	if (dccb->state & DPT_CCB_STATE_MARKED_LOST) {
+		struct timeval  now;
+		u_int32_t       age;
+		struct scsi_xfer *xs = dccb->xs;
+
+		microtime(&now);
+		age = dpt_time_delta(dccb->command_started, now);
+
+		printf("dpt%d: Salvaging Tx %d from the jaws of destruction "
+		       "(%d/%d)\n",
+		       dpt->unit, dccb->transaction_id, xs->timeout, age);
+		dccb->state |= DPT_CCB_STATE_MARKED_SALVAGED;
+		dccb->state &= ~DPT_CCB_STATE_MARKED_LOST;
+	}
+#endif
+
+	/* Ignore status packets with EOC not set */
+	if (dpt->sp.EOC == 0) {
+		printf("dpt%d ERROR: Request %d recieved with clear EOC.\n"
+		       "     Marking as LOST.\n",
+		       dpt->unit, dccb->transaction_id);
+#ifdef DPT_VERIFY_HINTR
+		dpt->sp.ccb = (dpt_sp_t *) 0x55555555;
+#else
+		dpt->sp.ccb = (dpt_sp_t *) NULL;
+#endif
+
+#ifdef DPT_MEASURE_PERFORMANCE
+		++dpt->performance.aborted_interrupts;
+#endif
+
+#ifdef DPT_HANDLE_TIMEOUTS
+		dccb->state |= DPT_CCB_STATE_MARKED_LOST;
+#endif
+		/* This CLEARS the interrupt! */
+		status = dpt_inb(dpt, HA_RSTATUS);
+		return;
+	}
+	dpt->sp.EOC = 0;
+
+#ifdef DPT_VERIFY_HINTR
+	/*
+	 * Make SURE the next caller is legitimate. If they are not, we will
+	 * find 0x55555555 here. We see 0x000000 or 0xffffffff when the PCi
+	 * bus has DMA troubles (as when behing a PCI-PCI * bridge .
+	 */
+	if ((dccb == NULL)
+	    || (dccb == (dpt_ccb_t *) ~ 0)
+	    || (dccb == (dpt_ccb_t *) 0x55555555)) {
+		printf("dpt%d: BAD (%x) CCB in SP (AUX status = %s).\n",
+		       dpt->unit, dccb, i2bin((unsigned long) aux_status,
+					      sizeof(aux_status) * 8));
+#ifdef DPT_MEASURE_PERFORMANCE
+		++dpt->performance.aborted_interrupts;
+#endif
+		/* This CLEARS the interrupt! */
+		status = dpt_inb(dpt, HA_RSTATUS);
+		return;
+	}
+	for (tccb = TAILQ_FIRST(&dpt->submitted_ccbs);
+	     (tccb != NULL) && (tccb != dccb);
+	     tccb = TAILQ_NEXT(tccb, links));
+	if (tccb == NULL) {
+		printf("dpt%d: %x is not in the SUBMITTED queue\n",
+		       dpt->unit, dccb);
+
+		for (tccb = TAILQ_FIRST(&dpt->completed_ccbs);
+		     (tccb != NULL) && (tccb != dccb);
+		     tccb = TAILQ_NEXT(tccb, links));
+		if (tccb != NULL)
+			printf("dpt%d: %x is in the COMPLETED queue\n",
+			       dpt->unit, dccb);
+
+		for (tccb = TAILQ_FIRST(&dpt->waiting_ccbs);
+		     (tccb != NULL) && (tccb != dccb);
+		     tccb = TAILQ_NEXT(tccb, links));
+		if (tccb != NULL)
+			printf("dpt%d: %x is in the WAITING queue\n",
+			       dpt->unit, dccb);
+
+		for (tccb = TAILQ_FIRST(&dpt->free_ccbs);
+		     (tccb != NULL) && (tccb != dccb);
+		     tccb = TAILQ_NEXT(tccb, links));
+		if (tccb != NULL)
+			printf("dpt%d: %x is in the FREE queue\n",
+			       dpt->unit, dccb);
+
+#ifdef DPT_MEASURE_PERFORMANCE
+		++dpt->performance.aborted_interrupts;
+#endif
+		/* This CLEARS the interrupt! */
+		status = dpt_inb(dpt, HA_RSTATUS);
+		return;
+	}
+#endif				/* DPT_VERIFY_HINTR */
+
+	/**
+	 * Copy the status packet from the general area to the dpt_ccb.
+	 * According to Mark Salyzyn, we only need few pieces of it.
+	 * Originally we had:
+	 * bcopy((void *) &dpt->sp, (void *) &dccb->status_packet,
+	 *        sizeof(dpt_sp_t));
+	 */
+	dccb->status_packet.hba_stat = dpt->sp.hba_stat;
+	dccb->status_packet.scsi_stat = dpt->sp.scsi_stat;
+	dccb->status_packet.residue_len = dpt->sp.residue_len;
+
+	/* Make sure the EOC bit is OFF! */
+	dpt->sp.EOC = 0;
+
+	/* Clear interrupts, check for error */
+	if ((status = dpt_inb(dpt, HA_RSTATUS)) & HA_SERROR) {
+		/**
+		 * Error Condition. Check for magic cookie. Exit this test
+		 * on earliest sign of non-reset condition
+		 */
+
+		/* Check that this is not a board reset interrupt */
+		if (dpt_just_reset(dpt)) {
+			printf("dpt%d: HBA rebooted.\n"
+			       "      All transactions should be "
+			       "resubmitted\n",
+			       dpt->unit);
+
+			printf("dpt%d: >>---->>  This is incomplete, fix me"
+			       "....  <<----<<",
+			       dpt->unit);
+			printf("      Incomplete Code; Re-queue the lost "
+			       "commands\n",
+			       dpt->unit);
+			Debugger("DPT Rebooted");
+
+#ifdef DPT_MEASURE_PERFORMANCE
+			++dpt->performance.aborted_interrupts;
+#endif
+			return;
+		}
+	}
+	dccb->status_reg = status;
+	dccb->aux_status_reg = aux_status;
+
+	/* Mark BOTH queues as busy */
+	dpt->queue_status |= (DPT_SUBMITTED_QUEUE_ACTIVE
+			      | DPT_COMPLETED_QUEUE_ACTIVE);
+	dpt_Qremove_submitted(dpt, dccb);
+	dpt_Qadd_completed(dpt, dccb);
+	dpt->queue_status &= ~(DPT_SUBMITTED_QUEUE_ACTIVE
+			       | DPT_COMPLETED_QUEUE_ACTIVE);
+	dpt_sched_queue(dpt);
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	{
+		u_int32_t       result;
+		struct timeval  junk;
+
+		microtime(&junk);
+
+		result = dpt_time_delta(dpt->performance.intr_started, junk);
+
+		if (result != ~0) {
+			if (dpt->performance.max_intr_time < result)
+				dpt->performance.max_intr_time = result;
+
+			if (result < dpt->performance.min_intr_time) {
+				dpt->performance.min_intr_time = result;
+			}
+		}
+	}
+#endif
+}
+
+/*
+ * This function is the DPT_ISR Software Interrupt Service Routine. When the
+ * DPT completes a SCSI command, it puts the results in a Status Packet, sets
+ * up two 1-byte registers and generates an interrupt.  We catch this
+ * interrupt in dpt_intr and copy the whole status to the proper CCB. Once
+ * this is done, we generate a software interrupt that calls this routine.
+ * The routine then scans ALL the complete queues of all the DPT HBA's and
+ * processes ALL the commands that are in the queue.
+ * 
+ * XXXX REMEMBER:	We always scan ALL the queues of all the HBA's. Always
+ * starting with the first controller registered (dpt0).	This creates
+ * an ``unfair'' opportunity for the first controllers in being served.
+ * Careful instrumentation may prove a need to change this policy.
+ * 
+ * This command rns at splSOFTcam.  Remember that.
+ */
+
+void
+dpt_sintr(void)
+{
+	dpt_softc_t    *dpt;
+	int             ospl;
+
+	/* Find which DPT needs help */
+	for (dpt = TAILQ_FIRST(&dpt_softc_list);
+	     dpt != NULL;
+	     dpt = TAILQ_NEXT(dpt, links)) {
+		/*
+		 * Drain the completed queue, to make room for new, " waiting
+		 * requests. We change to splcam to block interrupts from
+		 * mucking with " the completed queue
+		 */
+		ospl = splcam();
+		if (dpt->queue_status & DPT_SINTR_ACTIVE) {
+			splx(ospl);
+			continue;
+		}
+		dpt->queue_status |= DPT_SINTR_ACTIVE;
+
+		if (!TAILQ_EMPTY(&dpt->completed_ccbs)) {
+			splx(ospl);
+			dpt_complete(dpt);
+			ospl = splcam();
+		}
+		/* Submit as many waiting requests as the DPT can take */
+		if (!TAILQ_EMPTY(&dpt->waiting_ccbs)) {
+			dpt_run_queue(dpt, 0);
+		}
+		dpt->queue_status &= ~DPT_SINTR_ACTIVE;
+		splx(ospl);
+	}
+}
+
+/**
+ * Scan the complete queue for a given controller and process ALL the completed
+ * commands in the queue.
+ */
+
+static void
+dpt_complete(dpt_softc_t * dpt)
+{
+	dpt_ccb_t      *ccb;
+	int             ospl;
+
+	ospl = splcam();
+
+	if (dpt->queue_status & DPT_COMPLETED_QUEUE_ACTIVE) {
+		splx(ospl);
+		return;
+	}
+	dpt->queue_status |= DPT_COMPLETED_QUEUE_ACTIVE;
+
+	while ((ccb = TAILQ_FIRST(&dpt->completed_ccbs)) != NULL) {
+		struct scsi_xfer *xs;
+
+		dpt_Qremove_completed(dpt, ccb);
+		splx(ospl);
+
+		/* Process this completed request */
+		if (dpt_process_completion(dpt, ccb) == 0) {
+			xs = ccb->xs;
+
+			if (ccb->std_callback != NULL) {
+				(ccb->std_callback) (dpt, ccb->eata_ccb.cp_channel,
+						     ccb);
+			} else {
+				ospl = splcam();
+				dpt_Qpush_free(dpt, ccb);
+				splx(ospl);
+
+#ifdef DPT_MEASURE_PERFORMANCE
+				{
+					u_int32_t       result;
+					struct timeval  junk;
+
+					microtime(&junk);
+					ccb->command_ended = junk;
+#define time_delta dpt_time_delta(ccb->command_started,	ccb->command_ended)
+					result = time_delta;
+#define maxctime dpt->performance.max_command_time[ccb->eata_ccb.cp_scsi_cmd]
+#define minctime dpt->performance.min_command_time[ccb->eata_ccb.cp_scsi_cmd]
+
+					if (result != ~0) {
+						if (maxctime < result) {
+							maxctime = result;
+						}
+						if ((minctime == 0)
+						    || (minctime > result))
+							minctime = result;
+					}
+				}
+#endif
+
+				scsi_done(xs);
+			}
+			ospl = splcam();
+		}
+	}
+	splx(ospl);
+
+	/**
+	 * As per Justin's suggestion, we now will call the run_queue for
+	 * this HBA. This is done in case there are left-over requests that
+	 * were not submitted yet.
+	 */
+	dpt_run_queue(dpt, 0);
+	ospl = splsoftcam();
+	dpt->queue_status &= ~DPT_COMPLETED_QUEUE_ACTIVE;
+	splx(ospl);
+}
+
+#ifdef DPT_MEASURE_PERFORMANCE
+/**
+ * Given a dpt_ccb and a scsi_xfr structures,
+ * this functions translates the result of a SCSI operation.
+ * It returns values in the structures pointed by the arguments.
+ * This function does NOT attempt to protect itself from bad influence!
+ */
+
+#define WRITE_OP 1
+#define READ_OP 2
+#define min_submitR	dpt->performance.read_by_size_min_time[index]
+#define max_submitR	dpt->performance.read_by_size_max_time[index]
+#define min_submitW	dpt->performance.write_by_size_min_time[index]
+#define max_submitW	dpt->performance.write_by_size_max_time[index]
+
+static void
+dpt_IObySize(dpt_softc_t * dpt, dpt_ccb_t * ccb, int op, int index)
+{
+	if (op == READ_OP) {
+		++dpt->performance.read_by_size_count[index];
+		if (ccb->submitted_time < min_submitR)
+			min_submitR = ccb->submitted_time;
+
+		if (ccb->submitted_time > max_submitR)
+			max_submitR = ccb->submitted_time;
+	} else {		/* WRITE operation */
+		++dpt->performance.write_by_size_count[index];
+		if (ccb->submitted_time < min_submitW)
+			min_submitW = ccb->submitted_time;
+
+		if (ccb->submitted_time > max_submitW)
+			max_submitW = ccb->submitted_time;
+	}
+}
+#endif
+
+static int
+dpt_process_completion(dpt_softc_t * dpt,
+		       dpt_ccb_t * ccb)
+{
+	int             ospl;
+	struct scsi_xfer *xs;
+
+	if (ccb == NULL) {
+		panic("dpt%d: Improper argumet to process_completion (%p%p)\n",
+		      dpt->unit, ccb);
+	} else {
+		xs = ccb->xs;
+	}
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	{
+		u_int32_t       size;
+		struct scsi_rw_big *cmd;
+		int             op_type;
+
+		cmd = (struct scsi_rw_big *) & ccb->eata_ccb.cp_scsi_cmd;
+
+		switch (cmd->op_code) {
+		case 0xa8:	/* 12-byte READ	 */
+		case 0x08:	/* 6-byte READ	 */
+		case 0x28:	/* 10-byte READ	 */
+			op_type = READ_OP;
+			break;
+		case 0x0a:	/* 6-byte WRITE */
+		case 0xaa:	/* 12-byte WRITE */
+		case 0x2a:	/* 10-byte WRITE */
+			op_type = WRITE_OP;
+			break;
+		default:
+			op_type = 0;
+			break;
+		}
+
+		if (op_type != 0) {
+
+			size = (((u_int32_t) cmd->length2 << 8)
+				| ((u_int32_t) cmd->length1)) << 9;
+
+			switch (size) {
+			case 512:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_512);
+				break;
+			case 1024:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_1K);
+				break;
+			case 2048:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_2K);
+				break;
+			case 4096:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_4K);
+				break;
+			case 8192:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_8K);
+				break;
+			case 16384:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_16K);
+				break;
+			case 32768:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_32K);
+				break;
+			case 65536:
+				dpt_IObySize(dpt, ccb, op_type, SIZE_64K);
+				break;
+			default:
+				if (size > (1 << 16))
+					dpt_IObySize(dpt, ccb, op_type,
+						     SIZE_BIGGER);
+
+				else
+					dpt_IObySize(dpt, ccb, op_type,
+						     SIZE_OTHER);
+				break;
+			}
+		}
+	}
+#endif				/* DPT_MEASURE_PERFORMANCE */
+
+
+	switch ((int) ccb->status_packet.hba_stat) {
+	case HA_NO_ERROR:
+		if (xs != NULL) {
+			xs->error = XS_NOERROR;
+			xs->flags |= SCSI_ITSDONE;
+		}
+		break;
+	case HA_ERR_SEL_TO:
+	case HA_ERR_CMD_TO:
+		if (xs != NULL) {
+			xs->error |= XS_SELTIMEOUT;
+			xs->flags |= SCSI_ITSDONE;
+		}
+		break;
+	case HA_SCSIBUS_RESET:
+	case HA_CP_ABORTED:
+	case HA_CP_RESET:
+	case HA_PCI_PARITY:
+	case HA_PCI_MABORT:
+	case HA_PCI_TABORT:
+	case HA_PCI_STABORT:
+	case HA_BUS_PARITY:
+	case HA_UNX_MSGRJCT:
+		if (ccb->retries++ > DPT_RETRIES) {
+			if (xs != NULL) {
+				xs->error |= XS_SENSE;
+				xs->flags |= SCSI_ITSDONE;
+			}
+		} else {
+			ospl = splsoftcam();
+			dpt_Qpush_waiting(dpt, ccb);
+			splx(ospl);
+			dpt_sched_queue(dpt);
+		}
+		break;
+	case HA_HBA_POWER_UP:
+	case HA_UNX_BUSPHASE:
+	case HA_UNX_BUS_FREE:
+	case HA_SCSI_HUNG:
+	case HA_RESET_STUCK:
+		if (ccb->retries++ > DPT_RETRIES) {
+			if (xs != NULL) {
+				xs->error |= XS_SENSE;
+				xs->flags |= SCSI_ITSDONE;
+			}
+		} else {
+			ospl = splsoftcam();
+			dpt_Qpush_waiting(dpt, ccb);
+			splx(ospl);
+			dpt_sched_queue(dpt);
+			return (1);
+		}
+		break;
+	case HA_RSENSE_FAIL:
+		if (ccb->status_packet.EOC) {
+			if (xs != NULL) {
+				xs->error |= XS_SENSE;
+				xs->flags |= SCSI_ITSDONE;
+			}
+		} else {
+			if (ccb->retries++ > DPT_RETRIES) {
+				if (xs != NULL) {
+					xs->error |= XS_SENSE;
+					xs->flags |= SCSI_ITSDONE;
+				}
+			} else {
+				ospl = splsoftcam();
+				dpt_Qpush_waiting(dpt, ccb);
+				splx(ospl);
+				dpt_sched_queue(dpt);
+				return (1);
+			}
+		}
+		break;
+	case HA_PARITY_ERR:
+	case HA_CP_ABORT_NA:
+	case HA_CP_RESET_NA:
+	case HA_ECC_ERR:
+		if (xs != NULL) {
+			xs->error |= XS_SENSE;
+			xs->flags |= SCSI_ITSDONE;
+		}
+		break;
+	default:
+		printf("dpt%d: Undocumented Error %x",
+		       dpt->unit, ccb->status_packet.hba_stat);
+		if (xs != NULL) {
+			xs->error |= XS_SENSE;
+			xs->flags |= SCSI_ITSDONE;
+		}
+		Debugger("Please mail this message to shimon@i-connect.net");
+		break;
+	}
+
+	if (xs != NULL) {
+		if ((xs->error & XS_SENSE))
+			bcopy(&ccb->sense_data, &xs->sense,
+			      sizeof(struct scsi_sense_data));
+
+		if (ccb->status_packet.residue_len != 0) {
+			xs->flags |= SCSI_RESID_VALID;
+			xs->resid = ccb->status_packet.residue_len;
+		}
+	}
+	return (0);
+}
+
+#ifdef DPT_LOST_IRQ
+/**
+ * This functions handles the calling of the interrupt routine on a periodic
+ * basis.
+ * It is a completely ugly hack which purpose is to handle the problem of
+ * missing interrupts on certain platforms..
+ */
+
+static void
+dpt_irq_timeout(void *arg)
+{
+	dpt_softc_t    *dpt = (dpt_softc_t *) arg;
+	int             ospl;
+
+
+	if (!(dpt->state & DPT_LOST_IRQ_ACTIVE)) {
+		ospl = splcam();
+		dpt->state |= DPT_LOST_IRQ_ACTIVE;
+		dpt_intr(dpt);
+		splx(ospl);
+		if (dpt->state & DPT_LOST_IRQ_ACTIVE) {
+			printf("dpt %d: %d lost Interrupts Recovered\n",
+			       dpt->unit, ++dpt->lost_interrupts);
+		}
+		dpt->state &= ~DPT_LOST_IRQ_ACTIVE;
+	}
+	timeout(dpt_irq_timeout, (caddr_t) dpt, hz * 1);
+}
+
+#endif				/* DPT_LOST_IRQ */
+
+#ifdef DPT_HANDLE_TIMEOUTS
+/**
+ * This function walks down the SUBMITTED queue.
+ * Every request that is too old gets aborted and marked.
+ * Since the DPT will complete (interrupt) immediately (what does that mean?),
+ * We just walk the list, aborting old commands and marking them as such.
+ * The dpt_complete function will get rid of the that were interrupted in the
+ * normal manner.
+ *
+ * This function needs to run at splcam(), as it interacts with the submitted
+ * queue, as well as the completed and free queues.  Just like dpt_intr() does.
+ * To run it at any ISPL other than that of dpt_intr(), will mean that dpt_intr
+ * willbe able to pre-empt it, grab a transaction in progress (towards
+ * destruction) and operate on it.  The state of this transaction will be not
+ * very clear.
+ * The only other option, is to lock it only as long as necessary but have
+ * dpt_intr() spin-wait on it. In a UP environment this makes no sense and in
+ * a SMP environment, the advantage is dubvious for a function that runs once
+ * every ten seconds for few microseconds and, on systems with healthy
+ * hardware, does not do anything anyway.
+ */
+
+static void
+dpt_handle_timeouts(dpt_softc_t * dpt)
+{
+	dpt_ccb_t      *ccb;
+	int             ospl;
+
+	ospl = splcam();
+
+	if (dpt->state & DPT_HA_TIMEOUTS_ACTIVE) {
+		printf("dpt%d WARNING: Timeout Handling Collision\n",
+		       dpt->unit);
+		splx(ospl);
+		return;
+	}
+	dpt->state |= DPT_HA_TIMEOUTS_ACTIVE;
+
+	/* Loop through the entire submitted queue, looking for lost souls */
+	for (ccb = TAILQ_FIRST(&dpt->submitted_ccbs);
+	     ccb != NULL;
+	     ccb = TAILQ_NEXT(ccb, links)) {
+		struct scsi_xfer *xs;
+		struct timeval  now;
+		u_int32_t       age, max_age;
+
+		xs = ccb->xs;
+
+		microtime(&now);
+		age = dpt_time_delta(ccb->command_started, now);
+
+#define TenSec	10000000
+
+		if (xs == NULL) {	/* Local, non-kernel call */
+			max_age = TenSec;
+		} else {
+			max_age = (((xs->timeout * (dpt->submitted_ccbs_count
+						    + DPT_TIMEOUT_FACTOR))
+				    > TenSec)
+				 ? (xs->timeout * (dpt->submitted_ccbs_count
+						   + DPT_TIMEOUT_FACTOR))
+				   : TenSec);
+		}
+
+		/*
+		 * If a transaction is marked lost and is TWICE as old as we
+		 * care, then, and only then do we destroy it!
+		 */
+		if (ccb->state & DPT_CCB_STATE_MARKED_LOST) {
+			/* Remember who is next */
+			if (age > (max_age * 2)) {
+				dpt_Qremove_submitted(dpt, ccb);
+				ccb->state &= ~DPT_CCB_STATE_MARKED_LOST;
+				ccb->state |= DPT_CCB_STATE_ABORTED;
+#define cmd_name scsi_cmd_name(ccb->eata_ccb.cp_scsi_cmd)
+				if (ccb->retries++ > DPT_RETRIES) {
+					printf("dpt%d ERROR: Destroying stale "
+					       "%d (%s)\n"
+					       "		on "
+					       "c%db%dt%du%d (%d/%d)\n",
+					     dpt->unit, ccb->transaction_id,
+					       cmd_name,
+					       dpt->unit,
+					       ccb->eata_ccb.cp_channel,
+					       ccb->eata_ccb.cp_id,
+					       ccb->eata_ccb.cp_LUN, age,
+					       ccb->retries);
+#define send_ccb &ccb->eata_ccb
+#define ESA	 EATA_SPECIFIC_ABORT
+					(void) dpt_send_immediate(dpt,
+								  send_ccb,
+								  ESA,
+								  0, 0);
+					dpt_Qpush_free(dpt, ccb);
+
+					/* The SCSI layer should re-try */
+					xs->error |= XS_TIMEOUT;
+					xs->flags |= SCSI_ITSDONE;
+					scsi_done(xs);
+				} else {
+					printf("dpt%d ERROR: Stale %d (%s) on "
+					       "c%db%dt%du%d (%d)\n"
+					     "		gets another "
+					       "chance(%d/%d)\n",
+					     dpt->unit, ccb->transaction_id,
+					       cmd_name,
+					       dpt->unit,
+					       ccb->eata_ccb.cp_channel,
+					       ccb->eata_ccb.cp_id,
+					       ccb->eata_ccb.cp_LUN,
+					    age, ccb->retries, DPT_RETRIES);
+
+					dpt_Qpush_waiting(dpt, ccb);
+					dpt_sched_queue(dpt);
+				}
+			}
+		} else {
+			/*
+			 * This is a transaction that is not to be destroyed
+			 * (yet) But it is too old for our liking. We wait as
+			 * long as the upper layer thinks. Not really, we
+			 * multiply that by the number of commands in the
+			 * submitted queue + 1.
+			 */
+			if (!(ccb->state & DPT_CCB_STATE_MARKED_LOST) &&
+			    (age != ~0) && (age > max_age)) {
+				printf("dpt%d ERROR: Marking %d (%s) on "
+				       "c%db%dt%du%d \n"
+				       "            as late after %dusec\n",
+				       dpt->unit, ccb->transaction_id,
+				       cmd_name,
+				       dpt->unit, ccb->eata_ccb.cp_channel,
+				       ccb->eata_ccb.cp_id,
+				       ccb->eata_ccb.cp_LUN, age);
+				ccb->state |= DPT_CCB_STATE_MARKED_LOST;
+			}
+		}
+	}
+
+	dpt->state &= ~DPT_HA_TIMEOUTS_ACTIVE;
+	splx(ospl);
+}
+
+static void
+dpt_timeout(void *arg)
+{
+	dpt_softc_t    *dpt = (dpt_softc_t *) arg;
+
+	if (!(dpt->state & DPT_HA_TIMEOUTS_ACTIVE))
+		dpt_handle_timeouts(dpt);
+
+	timeout(dpt_timeout, (caddr_t) dpt, hz * 10);
+}
+
+#endif				/* DPT_HANDLE_TIMEOUTS */
+
+/*
+ * Remove a ccb from the completed queue
+ */
+static INLINE_Q void
+dpt_Qremove_completed(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+#ifdef DPT_MEASURE_PERFORMANCE
+	u_int32_t       complete_time;
+	struct timeval  now;
+
+	microtime(&now);
+	complete_time = dpt_time_delta(ccb->command_ended, now);
+
+	if (complete_time != ~0) {
+		if (dpt->performance.max_complete_time < complete_time)
+			dpt->performance.max_complete_time = complete_time;
+		if (complete_time < dpt->performance.min_complete_time)
+			dpt->performance.min_complete_time = complete_time;
+	}
+#endif
+
+	TAILQ_REMOVE(&dpt->completed_ccbs, ccb, links);
+	--dpt->completed_ccbs_count;	/* One less completed ccb in the
+					 * queue */
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		wakeup(&dpt);
+}
+
+/**
+ * Pop the most recently used ccb off the (HEAD of the) FREE ccb queue
+ */
+static INLINE_Q dpt_ccb_t *
+dpt_Qpop_free(dpt_softc_t * dpt)
+{
+	dpt_ccb_t      *ccb;
+
+	if ((ccb = TAILQ_FIRST(&dpt->free_ccbs)) == NULL) {
+		if (dpt_alloc_freelist(dpt))
+			return (ccb);
+		else
+			return (dpt_Qpop_free(dpt));
+	} else {
+		TAILQ_REMOVE(&dpt->free_ccbs, ccb, links);
+		--dpt->free_ccbs_count;
+	}
+
+	return (ccb);
+}
+
+/**
+ * Put a (now freed) ccb back into the HEAD of the FREE ccb queue
+ */
+static INLINE_Q void
+dpt_Qpush_free(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+#ifdef DPT_FREELIST_IS_STACK
+	TAILQ_INSERT_HEAD(&dpt->free_ccbs, ccb, links)
+#else
+	TAILQ_INSERT_TAIL(&dpt->free_ccbs, ccb, links);
+#endif
+
+	++dpt->free_ccbs_count;
+}
+
+/**
+ *	Add a request to the TAIL of the WAITING ccb queue
+ */
+static INLINE_Q void
+dpt_Qadd_waiting(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+	struct timeval  junk;
+
+	TAILQ_INSERT_TAIL(&dpt->waiting_ccbs, ccb, links);
+	++dpt->waiting_ccbs_count;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	microtime(&junk);
+	ccb->command_ended = junk;
+	if (dpt->waiting_ccbs_count > dpt->performance.max_waiting_count)
+		dpt->performance.max_waiting_count = dpt->waiting_ccbs_count;
+#endif
+
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		wakeup(&dpt);
+}
+
+/**
+ *	Add a request to the HEAD of the WAITING ccb queue
+ */
+static INLINE_Q void
+dpt_Qpush_waiting(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+	struct timeval  junk;
+
+	TAILQ_INSERT_HEAD(&dpt->waiting_ccbs, ccb, links);
+	++dpt->waiting_ccbs_count;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	microtime(&junk);
+	ccb->command_ended = junk;
+
+	if (dpt->performance.max_waiting_count < dpt->waiting_ccbs_count)
+		dpt->performance.max_waiting_count = dpt->waiting_ccbs_count;
+
+#endif
+
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		wakeup(&dpt);
+}
+
+/**
+ * Remove a ccb from the waiting queue
+ */
+static INLINE_Q void
+dpt_Qremove_waiting(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+#ifdef DPT_MEASURE_PERFORMANCE
+	struct timeval  now;
+	u_int32_t       waiting_time;
+
+	microtime(&now);
+	waiting_time = dpt_time_delta(ccb->command_ended, now);
+
+	if (waiting_time != ~0) {
+		if (dpt->performance.max_waiting_time < waiting_time)
+			dpt->performance.max_waiting_time = waiting_time;
+		if (waiting_time < dpt->performance.min_waiting_time)
+			dpt->performance.min_waiting_time = waiting_time;
+	}
+#endif
+
+	TAILQ_REMOVE(&dpt->waiting_ccbs, ccb, links);
+	--dpt->waiting_ccbs_count;	/* One less waiting ccb in the queue	 */
+
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		wakeup(&dpt);
+}
+
+/**
+ * Add a request to the TAIL of the SUBMITTED ccb queue
+ */
+static INLINE_Q void
+dpt_Qadd_submitted(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+	struct timeval  junk;
+
+	TAILQ_INSERT_TAIL(&dpt->submitted_ccbs, ccb, links);
+	++dpt->submitted_ccbs_count;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	microtime(&junk);
+	ccb->command_ended = junk;
+	if (dpt->performance.max_submit_count < dpt->submitted_ccbs_count)
+		dpt->performance.max_submit_count = dpt->submitted_ccbs_count;
+#endif
+
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		wakeup(&dpt);
+}
+
+/**
+ * Add a request to the TAIL of the Completed ccb queue
+ */
+static INLINE_Q void
+dpt_Qadd_completed(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+	struct timeval  junk;
+
+	TAILQ_INSERT_TAIL(&dpt->completed_ccbs, ccb, links);
+	++dpt->completed_ccbs_count;
+
+#ifdef DPT_MEASURE_PERFORMANCE
+	microtime(&junk);
+	ccb->command_ended = junk;
+	if (dpt->performance.max_complete_count < dpt->completed_ccbs_count)
+		dpt->performance.max_complete_count =
+			dpt->completed_ccbs_count;
+#endif
+
+	if (dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+		wakeup(&dpt);
+}
+
+/**
+ * Remove a ccb from the submitted queue
+ */
+static INLINE_Q void
+dpt_Qremove_submitted(dpt_softc_t * dpt, dpt_ccb_t * ccb)
+{
+#ifdef DPT_MEASURE_PERFORMANCE
+	struct timeval  now;
+	u_int32_t       submit_time;
+
+	microtime(&now);
+	submit_time = dpt_time_delta(ccb->command_ended, now);
+
+	if (submit_time != ~0) {
+		ccb->submitted_time = submit_time;
+		if (dpt->performance.max_submit_time < submit_time)
+			dpt->performance.max_submit_time = submit_time;
+		if (submit_time < dpt->performance.min_submit_time)
+			dpt->performance.min_submit_time = submit_time;
+	} else {
+		ccb->submitted_time = 0;
+	}
+
+#endif
+
+	TAILQ_REMOVE(&dpt->submitted_ccbs, ccb, links);
+	--dpt->submitted_ccbs_count;	/* One less submitted ccb in the
+					 * queue */
+
+	if ((dpt->state & DPT_HA_SHUTDOWN_ACTIVE)
+	    || (dpt->state & DPT_HA_QUIET))
+		wakeup(&dpt);
+}
+
+/**
+ * Handle Shutdowns.
+ * Gets registered by the dpt_pci.c registar and called AFTER the system did
+ * all its sync work.
+ */
+
+void
+dpt_shutdown(int howto, void *arg_dpt)
+{
+	dpt_softc_t    *ldpt;
+	u_int8_t        channel;
+	u_int32_t       target;
+	u_int32_t       lun;
+	int             waiting;
+	int             submitted;
+	int             completed;
+	int             huh;
+	int             wait_is_over;
+	int             ospl;
+	dpt_softc_t    *dpt;
+
+	dpt = (dpt_softc_t *) arg_dpt;
+
+	printf("dpt%d: Shutting down (mode %d) HBA.	Please wait...",
+	       dpt->unit, howto);
+	wait_is_over = 0;
+
+	ospl = splcam();
+	dpt->state |= DPT_HA_SHUTDOWN_ACTIVE;
+	splx(ospl);
+
+	while ((((waiting = dpt->waiting_ccbs_count) != 0)
+		|| ((submitted = dpt->submitted_ccbs_count) != 0)
+		|| ((completed = dpt->completed_ccbs_count) != 0))
+	       && (wait_is_over == 0)) {
+#ifdef DPT_DEBUG_SHUTDOWN
+		printf("dpt%d: Waiting for queues w%ds%dc%d to deplete\n",
+		       dpt->unit, dpt->waiting_ccbs_count,
+		       dpt->submitted_ccbs_count,
+		       dpt->completed_ccbs_count);
+#endif
+		huh = tsleep((void *) dpt, PCATCH | PRIBIO, "dptoff", 100 * hz);
+		switch (huh) {
+		case 0:
+			/* Wakeup call received */
+			goto checkit;
+			break;
+		case EWOULDBLOCK:
+			/* Timer Expired */
+			printf("dpt%d: Shutdown timer expired with queues at "
+			       "w%ds%dc%d\n",
+			       dpt->unit, dpt->waiting_ccbs_count,
+			       dpt->submitted_ccbs_count,
+			       dpt->completed_ccbs_count);
+			++wait_is_over;
+			break;
+		default:
+			/* anything else */
+			printf("dpt%d: Shutdown UNKNOWN with qeueues at "
+			       "w%ds%dc%d\n",
+			       dpt->unit, dpt->waiting_ccbs_count,
+			       dpt->submitted_ccbs_count,
+			       dpt->completed_ccbs_count);
+			++wait_is_over;
+			break;
+		}
+checkit:
+
+	}
+
+	/**
+	 * What we do for a shutdown, is give the DPT early power loss
+	 * warning
+       . */
+	(void) dpt_send_immediate(dpt, NULL, EATA_POWER_OFF_WARN, 0, 0);
+	printf("dpt%d: Controller was warned of shutdown and is now "
+	       "disabled\n",
+	       dpt->unit);
+
+	return;
+}
+
+/* A primitive subset of isgraph.	Used by hex_dump below */
+#define IsGraph(val)	((((val) >= ' ') && ((val) <= '~')))
+
+/**
+ * This function dumps bytes to the screen in hex format.
+ */
+void
+hex_dump(u_int8_t * data, int length, char *name, int no)
+{
+	int             line, column, ndx;
+
+	printf("Kernel Hex Dump for %s-%d at %p (%d bytes)\n",
+	       name, no, data, length);
+
+	/* Zero out all the counters and repeat for as many bytes as we have */
+	for (ndx = 0, column = 0, line = 0; ndx < length; ndx++) {
+		/* Print relative offset at the beginning of every line */
+		if (column == 0)
+			printf("%04x ", ndx);
+
+		/* Print the byte as two hex digits, followed by a space */
+		printf("%02x ", data[ndx]);
+
+		/* Split the row of 16 bytes in half */
+		if (++column == 8) {
+			printf(" ");
+		}
+		/* St the end of each row of 16 bytes, put a space ... */
+		if (column == 16) {
+			printf("	");
+
+			/* ... and then print the ASCII-visible on a line. */
+			for (column = 0; column < 16; column++) {
+				int             ascii_pos = ndx - 15 + column;
+
+				/**
+				 * Non-printable and non-ASCII are just a
+				 * dot. ;-(
+				 */
+				if (IsGraph(data[ascii_pos]))
+					printf("%c", data[ascii_pos]);
+				else
+					printf(".");
+			}
+
+			/* Each line ends with a new line */
+			printf("\n");
+			column = 0;
+
+			/**
+			 * Every 256 bytes (16 lines of 16 bytes each) have
+			 * an empty line, separating them from the next
+			 * ``page''. Yes, I programmed on a Z-80, where a
+			 * page was 256 bytes :-)
+			 */
+			if (++line > 15) {
+				printf("\n");
+				line = 0;
+			}
+		}
+	}
+
+	/**
+	 * We are basically done. We do want, however, to handle the ASCII
+	 * translation of fractional lines.
+	 */
+	if ((ndx == length) && (column != 0)) {
+		int             modulus = 16 - column, spaces = modulus * 3,
+		                skip;
+
+		/**
+		 * Skip to the right, as many spaces as there are bytes
+		 * ``missing'' ...
+		 */
+		for (skip = 0; skip < spaces; skip++)
+			printf(" ");
+
+		/* ... And the gap separating the hex dump from the ASCII */
+		printf("  ");
+
+		/**
+		 * Do not forget the extra space that splits the hex dump
+		 * vertically
+		 */
+		if (column < 8)
+			printf(" ");
+
+		for (column = 0; column < (16 - modulus); column++) {
+			int             ascii_pos = ndx - (16 - modulus) + column;
+
+			if (IsGraph(data[ascii_pos]))
+				printf("%c", data[ascii_pos]);
+			else
+				printf(".");
+		}
+		printf("\n");
+	}
+}
+
+/**
+ * and this one presents an integer as ones and zeros
+ */
+static char     i2bin_bitmap[48];	/* Used for binary dump of registers */
+
+char           *
+i2bin(unsigned int no, int length)
+{
+	int             ndx, rind;
+
+	for (ndx = 0, rind = 0; ndx < 32; ndx++, rind++) {
+		i2bin_bitmap[rind] = (((no << ndx) & 0x80000000) ? '1' : '0');
+
+		if (((ndx % 4) == 3))
+			i2bin_bitmap[++rind] = ' ';
+	}
+
+	if ((ndx % 4) == 3)
+		i2bin_bitmap[rind - 1] = '\0';
+	else
+		i2bin_bitmap[rind] = '\0';
+
+	switch (length) {
+	case 8:
+		return (i2bin_bitmap + 30);
+		break;
+	case 16:
+		return (i2bin_bitmap + 20);
+		break;
+	case 24:
+		return (i2bin_bitmap + 10);
+		break;
+	case 32:
+		return (i2bin_bitmap);
+	default:
+		return ("i2bin: Invalid length Specs");
+		break;
+	}
+}
+
+/**
+ * This function translates a SCSI command numeric code to a human readable
+ * string.
+ * The string contains the class of devices, scope, description, (length),
+ * and [SCSI III documentation section].
+ */
+
+char           *
+scsi_cmd_name(u_int8_t cmd)
+{
+	switch (cmd) {
+		case 0x40:
+		return ("Change Definition [7.1]");
+		break;
+	case 0x39:
+		return ("Compare [7,2]");
+		break;
+	case 0x18:
+		return ("Copy [7.3]");
+		break;
+	case 0x3a:
+		return ("Copy and Verify [7.4]");
+		break;
+	case 0x04:
+		return ("Format Unit [6.1.1]");
+		break;
+	case 0x12:
+		return ("Inquiry [7.5]");
+		break;
+	case 0x36:
+		return ("lock/Unlock Cache [6.1.2]");
+		break;
+	case 0x4c:
+		return ("Log Select [7.6]");
+		break;
+	case 0x4d:
+		return ("Log Sense [7.7]");
+		break;
+	case 0x15:
+		return ("Mode select (6) [7.8]");
+		break;
+	case 0x55:
+		return ("Mode Select (10) [7.9]");
+		break;
+	case 0x1a:
+		return ("Mode Sense (6) [7.10]");
+		break;
+	case 0x5a:
+		return ("Mode Sense (10) [7.11]");
+		break;
+	case 0xa7:
+		return ("Move Medium Attached [SMC]");
+		break;
+	case 0x5e:
+		return ("Persistent Reserve In [7.12]");
+		break;
+	case 0x5f:
+		return ("Persistent Reserve Out [7.13]");
+		break;
+	case 0x1e:
+		return ("Prevent/Allow Medium Removal [7.14]");
+		break;
+	case 0x08:
+		return ("Read, Receive (6) [6.1.5]");
+		break;
+	case 0x28:
+		return ("Read (10) [6.1.5]");
+		break;
+	case 0xa8:
+		return ("Read (12) [6.1.5]");
+		break;
+	case 0x3c:
+		return ("Read Buffer [7.15]");
+		break;
+	case 0x25:
+		return ("Read Capacity [6.1.6]");
+		break;
+	case 0x37:
+		return ("Read Defect Data (10) [6.1.7]");
+		break;
+	case 0xb7:
+		return ("Read Defect Data (12) [6.2.5]");
+		break;
+	case 0xb4:
+		return ("Read Element Status Attached [SMC]");
+		break;
+	case 0x3e:
+		return ("Read Long [6.1.8]");
+		break;
+	case 0x07:
+		return ("Reassign Blocks [6.1.9]");
+		break;
+	case 0x81:
+		return ("Rebuild [6.1.10]");
+		break;
+	case 0x1c:
+		return ("Receive Diagnostics Result [7.16]");
+		break;
+	case 0x82:
+		return ("Regenerate [6.1.11]");
+		break;
+	case 0x17:
+		return ("Release(6) [7.17]");
+		break;
+	case 0x57:
+		return ("Release(10) [7.18]");
+		break;
+	case 0xa0:
+		return ("Report LUNs [7.19]");
+		break;
+	case 0x03:
+		return ("Request Sense [7.20]");
+		break;
+	case 0x16:
+		return ("Resereve (6) [7.21]");
+		break;
+	case 0x56:
+		return ("Reserve(10) [7.22]");
+		break;
+	case 0x2b:
+		return ("Reserve(10) [6.1.12]");
+		break;
+	case 0x1d:
+		return ("Send Disagnostics [7.23]");
+		break;
+	case 0x33:
+		return ("Set Limit (10) [6.1.13]");
+		break;
+	case 0xb3:
+		return ("Set Limit (12) [6.2.8]");
+		break;
+	case 0x1b:
+		return ("Start/Stop Unit [6.1.14]");
+		break;
+	case 0x35:
+		return ("Synchronize Cache [6.1.15]");
+		break;
+	case 0x00:
+		return ("Test Unit Ready [7.24]");
+		break;
+	case 0x3d:
+		return ("Update Block (6.2.9");
+		break;
+	case 0x2f:
+		return ("Verify (10) [6.1.16, 6.2.10]");
+		break;
+	case 0xaf:
+		return ("Verify (12) [6.2.11]");
+		break;
+	case 0x0a:
+		return ("Write, Send (6) [6.1.17, 9.2]");
+		break;
+	case 0x2a:
+		return ("Write (10) [6.1.18]");
+		break;
+	case 0xaa:
+		return ("Write (12) [6.2.13]");
+		break;
+	case 0x2e:
+		return ("Write and Verify (10) [6.1.19, 6.2.14]");
+		break;
+	case 0xae:
+		return ("Write and Verify (12) [6.1.19, 6.2.15]");
+		break;
+	case 0x03b:
+		return ("Write Buffer [7.25]");
+		break;
+	case 0x03f:
+		return ("Write Long [6.1.20]");
+		break;
+	case 0x041:
+		return ("Write Same [6.1.21]");
+		break;
+	case 0x052:
+		return ("XD Read [6.1.22]");
+		break;
+	case 0x050:
+		return ("XD Write [6.1.22]");
+		break;
+	case 0x080:
+		return ("XD Write Extended [6.1.22]");
+		break;
+	case 0x051:
+		return ("XO Write [6.1.22]");
+		break;
+	default:
+		return ("Unknown SCSI Command");
+	}
+}
+
+/* End of the DPT driver */
+
+/**
+ * Hello emacs, these are the
+ * Local Variables:
+ *  c-indent-level:               8
+ *  c-continued-statement-offset: 8
+ *  c-continued-brace-offset:     0
+ *  c-brace-offset:              -8
+ *  c-brace-imaginary-offset:     0
+ *  c-argdecl-indent:             8
+ *  c-label-offset:              -8
+ *  c++-hanging-braces:           1
+ *  c++-access-specifier-offset: -8
+ *  c++-empty-arglist-indent:     8
+ *  c++-friend-offset:            0
+ * End:
+ */