xref: /titanic_44/usr/src/uts/sun/io/dada/targets/dad.c (revision fcf3ce441efd61da9bb2884968af01cb7c1452cc)

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */


/*
 * Direct Attached  disk driver for SPARC machines.
 */

/*
 * Includes, Declarations and Local Data
 */
#include <sys/dada/dada.h>
#include <sys/dkbad.h>
#include <sys/dklabel.h>
#include <sys/dkio.h>
#include <sys/cdio.h>
#include <sys/vtoc.h>
#include <sys/dada/targets/daddef.h>
#include <sys/dada/targets/dadpriv.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/kstat.h>
#include <sys/vtrace.h>
#include <sys/aio_req.h>
#include <sys/note.h>
#include <sys/cmlb.h>

/*
 * Global Error Levels for Error Reporting
 */
int dcd_error_level	= DCD_ERR_RETRYABLE;
/*
 * Local Static Data
 */

static int dcd_io_time		= DCD_IO_TIME;
static int dcd_retry_count	= DCD_RETRY_COUNT;
#ifndef lint
static int dcd_report_pfa = 1;
#endif
static int dcd_rot_delay = 4;
static int dcd_poll_busycnt = DCD_POLL_TIMEOUT;

/*
 * Local Function Prototypes
 */

static int dcdopen(dev_t *dev_p, int flag, int otyp, cred_t *cred_p);
static int dcdclose(dev_t dev, int flag, int otyp, cred_t *cred_p);
static int dcdstrategy(struct buf *bp);
static int dcddump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
static int dcdioctl(dev_t, int, intptr_t, int, cred_t *, int *);
static int dcdread(dev_t dev, struct uio *uio, cred_t *cred_p);
static int dcdwrite(dev_t dev, struct uio *uio, cred_t *cred_p);
static int dcd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int,
    char *, caddr_t, int *);
static int dcdaread(dev_t dev, struct aio_req *aio, cred_t *cred_p);
static int dcdawrite(dev_t dev, struct aio_req *aio, cred_t *cred_p);


static void dcd_free_softstate(struct dcd_disk *un, dev_info_t *devi);
static int dcd_doattach(dev_info_t *devi, int (*f)());
static int dcd_validate_geometry(struct dcd_disk *un);
static ddi_devid_t dcd_get_devid(struct dcd_disk *un);
static ddi_devid_t  dcd_create_devid(struct dcd_disk *un);
static int dcd_make_devid_from_serial(struct dcd_disk *un);
static void dcd_validate_model_serial(char *str, int *retlen, int totallen);
static int dcd_read_deviceid(struct dcd_disk *un);
static int dcd_write_deviceid(struct dcd_disk *un);
static int dcd_poll(struct dcd_pkt *pkt);
static char *dcd_rname(int reason);
static void dcd_flush_cache(struct dcd_disk *un);

static int dcd_compute_dk_capacity(struct dcd_device *devp,
    diskaddr_t *capacity);
static int dcd_send_lb_rw_cmd(dev_info_t *devinfo, void *bufaddr,
    diskaddr_t start_block, size_t reqlength, uchar_t cmd);

static void dcdmin(struct buf *bp);

static int dcdioctl_cmd(dev_t, struct udcd_cmd *,
    enum uio_seg, enum uio_seg);

static void dcdstart(struct dcd_disk *un);
static void dcddone_and_mutex_exit(struct dcd_disk *un, struct buf *bp);
static void make_dcd_cmd(struct dcd_disk *un, struct buf *bp, int (*f)());
static void dcdudcdmin(struct buf *bp);

static int dcdrunout(caddr_t);
static int dcd_check_wp(dev_t dev);
static int dcd_unit_ready(dev_t dev);
static void dcd_handle_tran_busy(struct buf *bp, struct diskhd *dp,
    struct dcd_disk *un);
static void dcdintr(struct dcd_pkt *pkt);
static int dcd_handle_incomplete(struct dcd_disk *un, struct buf *bp);
static void dcd_offline(struct dcd_disk *un, int bechatty);
static int dcd_ready_and_valid(dev_t dev, struct dcd_disk *un);
static void dcd_reset_disk(struct dcd_disk *un, struct dcd_pkt *pkt);
static void dcd_translate(struct dadkio_status32 *statp, struct udcd_cmd *cmdp);
static int dcdflushdone(struct buf *bp);

/* Function prototypes for cmlb */

static int dcd_lb_rdwr(dev_info_t *devi, uchar_t cmd, void *bufaddr,
    diskaddr_t start_block, size_t reqlength, void *tg_cookie);

static int dcd_lb_getphygeom(dev_info_t *devi, cmlb_geom_t *phygeomp);
static int dcd_lb_getinfo(dev_info_t *devi, int cmd, void *arg,
    void *tg_cookie);


static cmlb_tg_ops_t dcd_lb_ops = {
	TG_DK_OPS_VERSION_1,
	dcd_lb_rdwr,
	dcd_lb_getinfo
};

/*
 * Error and Logging Functions
 */
#ifndef lint
static void clean_print(dev_info_t *dev, char *label, uint_t level,
    char *title, char *data, int len);
static void dcdrestart(void *arg);
#endif /* lint */

static int dcd_check_error(struct dcd_disk *un, struct buf *bp);

/*
 * Error statistics create/update functions
 */
static int dcd_create_errstats(struct dcd_disk *, int);



/*PRINTFLIKE4*/
extern void dcd_log(dev_info_t *, char *, uint_t, const char *, ...)
    __KPRINTFLIKE(4);
extern void makecommand(struct dcd_pkt *, int, uchar_t, uint32_t,
    uchar_t, uint32_t, uchar_t, uchar_t);


/*
 * Configuration Routines
 */
static int dcdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,
    void **result);
static int dcdprobe(dev_info_t *devi);
static int dcdattach(dev_info_t *devi, ddi_attach_cmd_t cmd);
static int dcddetach(dev_info_t *devi, ddi_detach_cmd_t cmd);
static int dcdreset(dev_info_t *dip, ddi_reset_cmd_t cmd);
static int dcd_dr_detach(dev_info_t *devi);
static int dcdpower(dev_info_t *devi, int component, int level);

static void *dcd_state;
static int dcd_max_instance;
static char *dcd_label = "dad";

static char *diskokay = "disk okay\n";

#if DEBUG || lint
#define	DCDDEBUG
#endif

int dcd_test_flag = 0;
/*
 * Debugging macros
 */
#ifdef	DCDDEBUG
static int dcddebug = 0;
#define	DEBUGGING	(dcddebug > 1)
#define	DAD_DEBUG	if (dcddebug == 1) dcd_log
#define	DAD_DEBUG2	if (dcddebug > 1) dcd_log
#else	/* DCDDEBUG */
#define	dcddebug		(0)
#define	DEBUGGING	(0)
#define	DAD_DEBUG	if (0) dcd_log
#define	DAD_DEBUG2	if (0) dcd_log
#endif

/*
 * we use pkt_private area for storing bp and retry_count
 * XXX: Really is this usefull.
 */
struct dcd_pkt_private {
	struct buf	*dcdpp_bp;
	short		 dcdpp_retry_count;
	short		 dcdpp_victim_retry_count;
};


_NOTE(SCHEME_PROTECTS_DATA("Unique per pkt", dcd_pkt_private buf))

#define	PP_LEN	(sizeof (struct dcd_pkt_private))

#define	PKT_SET_BP(pkt, bp)	\
	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_bp = bp
#define	PKT_GET_BP(pkt) \
	(((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_bp)


#define	PKT_SET_RETRY_CNT(pkt, n) \
	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_retry_count = n

#define	PKT_GET_RETRY_CNT(pkt) \
	(((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_retry_count)

#define	PKT_INCR_RETRY_CNT(pkt, n) \
	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_retry_count += n

#define	PKT_SET_VICTIM_RETRY_CNT(pkt, n) \
	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_victim_retry_count \
			= n

#define	PKT_GET_VICTIM_RETRY_CNT(pkt) \
	(((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_victim_retry_count)
#define	PKT_INCR_VICTIM_RETRY_CNT(pkt, n) \
	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_victim_retry_count \
			+= n

#define	DISK_NOT_READY_RETRY_COUNT	(dcd_retry_count / 2)


/*
 * Urk!
 */
#define	SET_BP_ERROR(bp, err)	\
	bioerror(bp, err);

#define	IOSP			KSTAT_IO_PTR(un->un_stats)
#define	IO_PARTITION_STATS	un->un_pstats[DCDPART(bp->b_edev)]
#define	IOSP_PARTITION		KSTAT_IO_PTR(IO_PARTITION_STATS)

#define	DCD_DO_KSTATS(un, kstat_function, bp) \
	ASSERT(mutex_owned(DCD_MUTEX)); \
	if (bp != un->un_sbufp) { \
		if (un->un_stats) { \
			kstat_function(IOSP); \
		} \
		if (IO_PARTITION_STATS) { \
			kstat_function(IOSP_PARTITION); \
		} \
	}

#define	DCD_DO_ERRSTATS(un, x) \
	if (un->un_errstats) { \
		struct dcd_errstats *dtp; \
		dtp = (struct dcd_errstats *)un->un_errstats->ks_data; \
		dtp->x.value.ui32++; \
	}

#define	GET_SOFT_STATE(dev)						\
	struct dcd_disk *un;					\
	int instance, part;					\
	minor_t minor = getminor(dev);				\
									\
	part = minor & DCDPART_MASK;					\
	instance = minor >> DCDUNIT_SHIFT;				\
	if ((un = ddi_get_soft_state(dcd_state, instance)) == NULL)	\
		return (ENXIO);

#define	LOGICAL_BLOCK_ALIGN(blkno, blknoshift) \
		(((blkno) & ((1 << (blknoshift)) - 1)) == 0)

/*
 * After the following number of sectors, the cylinder number spills over
 * 0xFFFF if sectors = 63 and heads = 16.
 */
#define	NUM_SECTORS_32G	0x3EFFC10

/*
 * Configuration Data
 */

/*
 * Device driver ops vector
 */

static struct cb_ops dcd_cb_ops = {
	dcdopen,		/* open */
	dcdclose,		/* close */
	dcdstrategy,		/* strategy */
	nodev,			/* print */
	dcddump,		/* dump */
	dcdread,		/* read */
	dcdwrite,		/* write */
	dcdioctl,		/* ioctl */
	nodev,			/* devmap */
	nodev,			/* mmap */
	nodev,			/* segmap */
	nochpoll,		/* poll */
	dcd_prop_op,		/* cb_prop_op */
	0,			/* streamtab  */
	D_64BIT | D_MP | D_NEW,	/* Driver compatibility flag */
	CB_REV,			/* cb_rev */
	dcdaread, 		/* async I/O read entry point */
	dcdawrite		/* async I/O write entry point */
};

static struct dev_ops dcd_ops = {
	DEVO_REV,		/* devo_rev, */
	0,			/* refcnt  */
	dcdinfo,		/* info */
	nulldev,		/* identify */
	dcdprobe,		/* probe */
	dcdattach,		/* attach */
	dcddetach,		/* detach */
	dcdreset,		/* reset */
	&dcd_cb_ops,		/* driver operations */
	(struct bus_ops *)0,	/* bus operations */
	dcdpower,		/* power */
	ddi_quiesce_not_supported,	/* devo_quiesce */
};


/*
 * This is the loadable module wrapper.
 */
#include <sys/modctl.h>

static struct modldrv modldrv = {
	&mod_driverops,		/* Type of module. This one is a driver */
	"DAD Disk Driver",	/* Name of the module. */
	&dcd_ops,	/* driver ops */
};



static struct modlinkage modlinkage = {
	MODREV_1, &modldrv, NULL
};

/*
 * the dcd_attach_mutex only protects dcd_max_instance in multi-threaded
 * attach situations
 */
static kmutex_t dcd_attach_mutex;

int
_init(void)
{
	int e;

	if ((e = ddi_soft_state_init(&dcd_state, sizeof (struct dcd_disk),
	    DCD_MAXUNIT)) != 0)
		return (e);

	mutex_init(&dcd_attach_mutex, NULL, MUTEX_DRIVER, NULL);
	e = mod_install(&modlinkage);
	if (e != 0) {
		mutex_destroy(&dcd_attach_mutex);
		ddi_soft_state_fini(&dcd_state);
		return (e);
	}

	return (e);
}

int
_fini(void)
{
	int e;

	if ((e = mod_remove(&modlinkage)) != 0)
		return (e);

	ddi_soft_state_fini(&dcd_state);
	mutex_destroy(&dcd_attach_mutex);

	return (e);
}

int
_info(struct modinfo *modinfop)
{

	return (mod_info(&modlinkage, modinfop));
}

static int
dcdprobe(dev_info_t *devi)
{
	struct dcd_device *devp;
	int rval = DDI_PROBE_PARTIAL;
	int instance;

	devp = ddi_get_driver_private(devi);
	instance = ddi_get_instance(devi);

	/*
	 * Keep a count of how many disks (ie. highest instance no) we have
	 * XXX currently not used but maybe useful later again
	 */
	mutex_enter(&dcd_attach_mutex);
	if (instance > dcd_max_instance)
		dcd_max_instance = instance;
	mutex_exit(&dcd_attach_mutex);

	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG, "dcdprobe:\n");

	if (ddi_get_soft_state(dcd_state, instance) != NULL)
		return (DDI_PROBE_PARTIAL);

	/*
	 * Turn around and call utility probe routine
	 * to see whether we actually have a disk at
	 */

	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG,
	    "dcdprobe: %x\n", dcd_probe(devp, NULL_FUNC));

	switch (dcd_probe(devp, NULL_FUNC)) {
	default:
	case DCDPROBE_NORESP:
	case DCDPROBE_NONCCS:
	case DCDPROBE_NOMEM:
	case DCDPROBE_FAILURE:
	case DCDPROBE_BUSY:
		break;

	case DCDPROBE_EXISTS:
		/*
		 * Check whether it is a ATA device and then
		 * return  SUCCESS.
		 */
		DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG,
		    "config %x\n", devp->dcd_ident->dcd_config);
		if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
			if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
				rval = DDI_PROBE_SUCCESS;
			} else
				rval = DDI_PROBE_FAILURE;
		} else {
			rval = DDI_PROBE_FAILURE;
		}
		break;
	}
	dcd_unprobe(devp);

	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG,
	    "dcdprobe returns %x\n", rval);

	return (rval);
}


/*ARGSUSED*/
static int
dcdattach(dev_info_t *devi, ddi_attach_cmd_t cmd)
{
	int instance, rval;
	struct dcd_device *devp;
	struct dcd_disk *un;
	struct diskhd *dp;
	char	*pm_comp[] =
	    { "NAME=ide-disk", "0=standby", "1=idle", "2=active" };

	/* CONSTCOND */
	ASSERT(NO_COMPETING_THREADS);


	devp = ddi_get_driver_private(devi);
	instance = ddi_get_instance(devi);
	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG, "Attach Started\n");

	switch (cmd) {
	case DDI_ATTACH:
		break;

	case DDI_RESUME:
		if (!(un = ddi_get_soft_state(dcd_state, instance)))
			return (DDI_FAILURE);
		mutex_enter(DCD_MUTEX);
		Restore_state(un);
		/*
		 * Restore the state which was saved to give the
		 * the right state in un_last_state
		 */
		un->un_last_state = un->un_save_state;
		un->un_throttle = 2;
		cv_broadcast(&un->un_suspend_cv);
		/*
		 * Raise the power level of the device to active.
		 */
		mutex_exit(DCD_MUTEX);
		(void) pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE);
		mutex_enter(DCD_MUTEX);

		/*
		 * start unit - if this is a low-activity device
		 * commands in queue will have to wait until new
		 * commands come in, which may take awhile.
		 * Also, we specifically don't check un_ncmds
		 * because we know that there really are no
		 * commands in progress after the unit was suspended
		 * and we could have reached the throttle level, been
		 * suspended, and have no new commands coming in for
		 * awhile.  Highly unlikely, but so is the low-
		 * activity disk scenario.
		 */
		dp = &un->un_utab;
		if (dp->b_actf && (dp->b_forw == NULL)) {
			dcdstart(un);
		}

		mutex_exit(DCD_MUTEX);
		return (DDI_SUCCESS);

	default:
		return (DDI_FAILURE);
	}

	if (dcd_doattach(devi, SLEEP_FUNC) == DDI_FAILURE) {
		return (DDI_FAILURE);
	}

	if (!(un = (struct dcd_disk *)
	    ddi_get_soft_state(dcd_state, instance))) {
		return (DDI_FAILURE);
	}
	devp->dcd_private = (ataopaque_t)un;

	/*
	 * Add a zero-length attribute to tell the world we support
	 * kernel ioctls (for layered drivers)
	 */
	(void) ddi_prop_create(DDI_DEV_T_NONE, devi, DDI_PROP_CANSLEEP,
	    DDI_KERNEL_IOCTL, NULL, 0);

	/*
	 * Since the dad device does not have the 'reg' property,
	 * cpr will not call its DDI_SUSPEND/DDI_RESUME entries.
	 * The following code is to tell cpr that this device
	 * does need to be suspended and resumed.
	 */
	(void) ddi_prop_update_string(DDI_DEV_T_NONE, devi,
	    "pm-hardware-state", (caddr_t)"needs-suspend-resume");

	/*
	 * Initialize power management bookkeeping;
	 * Create components - In IDE case there are 3 levels and one
	 * component. The levels being - active, idle, standby.
	 */

	rval = ddi_prop_update_string_array(DDI_DEV_T_NONE,
	    devi, "pm-components", pm_comp, 4);
	if (rval == DDI_PROP_SUCCESS) {
		/*
		 * Ignore the return value of pm_raise_power
		 * Even if we check the return values and
		 * remove the property created above, PM
		 * framework will not honour the change after
		 * first call to pm_raise_power. Hence, the
		 * removal of that property does not help if
		 * pm_raise_power fails.
		 */
		(void) pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE);
	}

	ddi_report_dev(devi);

	cmlb_alloc_handle(&un->un_dklbhandle);

	if (cmlb_attach(devi,
	    &dcd_lb_ops,
	    0,
	    0,
	    0,
	    DDI_NT_BLOCK_CHAN,
	    CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8,
	    un->un_dklbhandle,
	    0) != 0) {
		cmlb_free_handle(&un->un_dklbhandle);
		dcd_free_softstate(un, devi);
		return (DDI_FAILURE);
	}

	mutex_enter(DCD_MUTEX);
	(void) dcd_validate_geometry(un);

	/* Get devid; create a devid ONLY IF could not get ID */
	if (dcd_get_devid(un) == NULL) {
		/* Create the fab'd devid */
		(void) dcd_create_devid(un);
	}
	mutex_exit(DCD_MUTEX);

	return (DDI_SUCCESS);
}

static void
dcd_free_softstate(struct dcd_disk *un, dev_info_t *devi)
{
	struct dcd_device		*devp;
	int instance = ddi_get_instance(devi);

	devp = ddi_get_driver_private(devi);

	if (un) {
		sema_destroy(&un->un_semoclose);
		cv_destroy(&un->un_sbuf_cv);
		cv_destroy(&un->un_state_cv);
		cv_destroy(&un->un_disk_busy_cv);
		cv_destroy(&un->un_suspend_cv);

		/*
		 * Deallocate command packet resources.
		 */
		if (un->un_sbufp)
			freerbuf(un->un_sbufp);
		if (un->un_dp) {
			kmem_free((caddr_t)un->un_dp, sizeof (*un->un_dp));
		}
		/*
		 * Unregister the devid and free devid resources allocated
		 */
		ddi_devid_unregister(DCD_DEVINFO);
		if (un->un_devid) {
			ddi_devid_free(un->un_devid);
			un->un_devid = NULL;
		}

		/*
		 * Delete kstats. Kstats for non CD devices are deleted
		 * in dcdclose.
		 */
		if (un->un_stats) {
			kstat_delete(un->un_stats);
		}

	}

	/*
	 * Cleanup scsi_device resources.
	 */
	ddi_soft_state_free(dcd_state, instance);
	devp->dcd_private = (ataopaque_t)0;
	/* unprobe scsi device */
	dcd_unprobe(devp);

	/* Remove properties created during attach */
	ddi_prop_remove_all(devi);
}

static int
dcddetach(dev_info_t *devi, ddi_detach_cmd_t cmd)
{
	int instance;
	struct dcd_disk *un;
	clock_t	wait_cmds_complete;
	instance = ddi_get_instance(devi);

	if (!(un = ddi_get_soft_state(dcd_state, instance)))
		return (DDI_FAILURE);

	switch (cmd) {
	case DDI_DETACH:
		return (dcd_dr_detach(devi));

	case DDI_SUSPEND:
		mutex_enter(DCD_MUTEX);
		if (un->un_state == DCD_STATE_SUSPENDED) {
			mutex_exit(DCD_MUTEX);
			return (DDI_SUCCESS);
		}
		un->un_throttle = 0;
		/*
		 * Save the last state first
		 */
		un->un_save_state = un->un_last_state;

		New_state(un, DCD_STATE_SUSPENDED);

		/*
		 * wait till current operation completed. If we are
		 * in the resource wait state (with an intr outstanding)
		 * then we need to wait till the intr completes and
		 * starts the next cmd. We wait for
		 * DCD_WAIT_CMDS_COMPLETE seconds before failing the
		 * DDI_SUSPEND.
		 */
		wait_cmds_complete = ddi_get_lbolt();
		wait_cmds_complete +=
		    DCD_WAIT_CMDS_COMPLETE * drv_usectohz(1000000);

		while (un->un_ncmds) {
			if (cv_timedwait(&un->un_disk_busy_cv,
			    DCD_MUTEX, wait_cmds_complete) == -1) {
				/*
				 * commands Didn't finish in the
				 * specified time, fail the DDI_SUSPEND.
				 */
				DAD_DEBUG2(DCD_DEVINFO, dcd_label,
				    DCD_DEBUG, "dcddetach: SUSPEND "
				    "failed due to outstanding cmds\n");
				Restore_state(un);
				mutex_exit(DCD_MUTEX);
				return (DDI_FAILURE);
			}
		}
		mutex_exit(DCD_MUTEX);
		return (DDI_SUCCESS);
	}
	return (DDI_FAILURE);
}

/*
 * The reset entry point gets invoked at the system shutdown time or through
 * CPR code at system suspend.
 * Will be flushing the cache and expect this to be last I/O operation to the
 * disk before system reset/power off.
 */
/*ARGSUSED*/
static int
dcdreset(dev_info_t *dip, ddi_reset_cmd_t cmd)
{
	struct dcd_disk *un;
	int instance;

	instance = ddi_get_instance(dip);

	if (!(un = ddi_get_soft_state(dcd_state, instance)))
		return (DDI_FAILURE);

	dcd_flush_cache(un);

	return (DDI_SUCCESS);
}


static int
dcd_dr_detach(dev_info_t *devi)
{
	struct dcd_device	*devp;
	struct dcd_disk		*un;

	/*
	 * Get scsi_device structure for this instance.
	 */
	if ((devp = ddi_get_driver_private(devi)) == NULL)
		return (DDI_FAILURE);

	/*
	 * Get dcd_disk structure containing target 'private' information
	 */
	un = (struct dcd_disk *)devp->dcd_private;

	/*
	 * Verify there are NO outstanding commands issued to this device.
	 * ie, un_ncmds == 0.
	 * It's possible to have outstanding commands through the physio
	 * code path, even though everything's closed.
	 */
#ifndef lint
	_NOTE(COMPETING_THREADS_NOW);
#endif
	mutex_enter(DCD_MUTEX);
	if (un->un_ncmds) {
		mutex_exit(DCD_MUTEX);
		_NOTE(NO_COMPETING_THREADS_NOW);
		return (DDI_FAILURE);
	}

	mutex_exit(DCD_MUTEX);

	cmlb_detach(un->un_dklbhandle, 0);
	cmlb_free_handle(&un->un_dklbhandle);


	/*
	 * Lower the power state of the device
	 * i.e. the minimum power consumption state - sleep.
	 */
	(void) pm_lower_power(DCD_DEVINFO, 0, DCD_DEVICE_STANDBY);

	_NOTE(NO_COMPETING_THREADS_NOW);

	/*
	 * at this point there are no competing threads anymore
	 * release active MT locks and all device resources.
	 */
	dcd_free_softstate(un, devi);

	return (DDI_SUCCESS);
}

static int
dcdpower(dev_info_t *devi, int component, int level)
{
	struct dcd_pkt *pkt;
	struct dcd_disk *un;
	int	instance;
	uchar_t	cmd;


	instance = ddi_get_instance(devi);

	if (!(un = ddi_get_soft_state(dcd_state, instance)) ||
	    (DCD_DEVICE_STANDBY > level) || (level > DCD_DEVICE_ACTIVE) ||
	    component != 0) {
		return (DDI_FAILURE);
	}

	mutex_enter(DCD_MUTEX);
	/*
	 * if there are active commands for the device or device will be
	 * active soon. At the same time there is request to lower power
	 * return failure.
	 */
	if ((un->un_ncmds) && (level != DCD_DEVICE_ACTIVE)) {
		mutex_exit(DCD_MUTEX);
		return (DDI_FAILURE);
	}

	if ((un->un_state == DCD_STATE_OFFLINE) ||
	    (un->un_state == DCD_STATE_FATAL)) {
		mutex_exit(DCD_MUTEX);
		return (DDI_FAILURE);
	}

	if (level == DCD_DEVICE_ACTIVE) {
		/*
		 * No need to fire any command, just set the state structure
		 * to indicate previous state and set the level to active
		 */
		un->un_power_level = DCD_DEVICE_ACTIVE;
		if (un->un_state == DCD_STATE_PM_SUSPENDED)
			Restore_state(un);
		mutex_exit(DCD_MUTEX);
	} else {
		pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
		    NULL, (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
		    PKT_CONSISTENT, NULL_FUNC, NULL);

		if (pkt == (struct dcd_pkt *)NULL) {
			mutex_exit(DCD_MUTEX);
			return (DDI_FAILURE);
		}

		switch (level) {
		case DCD_DEVICE_IDLE:
			cmd = ATA_IDLE_IMMEDIATE;
			break;

		case DCD_DEVICE_STANDBY:
			cmd = ATA_STANDBY_IMMEDIATE;
			break;
		}

		makecommand(pkt, 0, cmd, 0, 0, 0, NO_DATA_XFER, 0);
		mutex_exit(DCD_MUTEX);
		/*
		 * Issue the appropriate command
		 */
		if ((dcd_poll(pkt)) || (SCBP_C(pkt) != STATUS_GOOD)) {
			dcd_destroy_pkt(pkt);
			return (DDI_FAILURE);
		}
		dcd_destroy_pkt(pkt);
		mutex_enter(DCD_MUTEX);
		if (un->un_state != DCD_STATE_PM_SUSPENDED)
			New_state(un, DCD_STATE_PM_SUSPENDED);
		un->un_power_level = level;
		mutex_exit(DCD_MUTEX);
	}

	return (DDI_SUCCESS);
}

static int
dcd_doattach(dev_info_t *devi, int (*canwait)())
{
	struct dcd_device *devp;
	struct dcd_disk *un = (struct dcd_disk *)0;
	int instance;
	int km_flags = (canwait != NULL_FUNC)? KM_SLEEP : KM_NOSLEEP;
	int rval;
	char *prop_template = "target%x-dcd-options";
	int options;
	char    prop_str[32];
	int target;
	diskaddr_t capacity;

	devp = ddi_get_driver_private(devi);

	/*
	 * Call the routine scsi_probe to do some of the dirty work.
	 * If the INQUIRY command succeeds, the field dcd_inq in the
	 * device structure will be filled in. The dcd_sense structure
	 * will also be allocated.
	 */

	switch (dcd_probe(devp, canwait)) {
	default:
		return (DDI_FAILURE);

	case DCDPROBE_EXISTS:
		if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
			if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
				rval = DDI_SUCCESS;
			} else {
				rval = DDI_FAILURE;
				goto error;
			}
		} else {
			rval = DDI_FAILURE;
			goto error;
		}
	}


	instance = ddi_get_instance(devp->dcd_dev);

	if (ddi_soft_state_zalloc(dcd_state, instance) != DDI_SUCCESS) {
		rval = DDI_FAILURE;
		goto error;
	}

	un = ddi_get_soft_state(dcd_state, instance);

	un->un_sbufp = getrbuf(km_flags);
	if (un->un_sbufp == (struct buf *)NULL) {
		rval = DDI_FAILURE;
		goto error;
	}


	un->un_dcd = devp;
	un->un_power_level = -1;
	un->un_tgattribute.media_is_writable = 1;

	sema_init(&un->un_semoclose, 1, NULL, SEMA_DRIVER, NULL);
	cv_init(&un->un_sbuf_cv, NULL, CV_DRIVER, NULL);
	cv_init(&un->un_state_cv, NULL, CV_DRIVER, NULL);
	/* Initialize power management conditional variable */
	cv_init(&un->un_disk_busy_cv, NULL, CV_DRIVER, NULL);
	cv_init(&un->un_suspend_cv, NULL, CV_DRIVER, NULL);

	if (un->un_dp == 0) {
		/*
		 * Assume CCS drive, assume parity, but call
		 * it a CDROM if it is a RODIRECT device.
		 */
		un->un_dp = (struct dcd_drivetype *)
		    kmem_zalloc(sizeof (struct dcd_drivetype), km_flags);
		if (!un->un_dp) {
			rval = DDI_FAILURE;
			goto error;
		}
		if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
			if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
				un->un_dp->ctype = CTYPE_DISK;
			}
		} else  {
			rval = DDI_FAILURE;
			goto error;
		}
		un->un_dp->name = "CCS";
		un->un_dp->options = 0;
	}

	/*
	 * Allow I/O requests at un_secsize offset in multiple of un_secsize.
	 */
	un->un_secsize = DEV_BSIZE;

	/*
	 * If the device is not a removable media device, make sure that
	 * that the device is ready, by issuing the another identify but
	 * not needed. Get the capacity from identify data and store here.
	 */
	if (dcd_compute_dk_capacity(devp, &capacity) == 0) {
		un->un_diskcapacity = capacity;
		un->un_lbasize = DEV_BSIZE;
	}

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "Geometry Data\n");
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "cyls %x, heads %x",
	    devp->dcd_ident->dcd_fixcyls,
	    devp->dcd_ident->dcd_heads);
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "sectors %x,",
	    devp->dcd_ident->dcd_sectors);
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "capacity %llx\n",
	    capacity);

	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "dcdprobe: drive selected\n");

	/*
	 * Check for the property target<n>-dcd-options to find the option
	 * set by the HBA driver for this target so that we can set the
	 * Unit structure variable so that we can send commands accordingly.
	 */
	target = devp->dcd_address->a_target;
	(void) sprintf(prop_str, prop_template, target);
	options = ddi_prop_get_int(DDI_DEV_T_ANY, devi, DDI_PROP_NOTPROM,
	    prop_str, -1);
	if (options < 0) {
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "No per target properties");
	} else {
		if ((options & DCD_DMA_MODE) == DCD_DMA_MODE) {
			un->un_dp->options |= DMA_SUPPORTTED;
			un->un_dp->dma_mode = (options >> 3) & 0x03;
			DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
			    "mode %x\n", un->un_dp->dma_mode);
		} else {
			un->un_dp->options &= ~DMA_SUPPORTTED;
			un->un_dp->pio_mode = options & 0x7;
			if (options & DCD_BLOCK_MODE)
				un->un_dp->options |= BLOCK_MODE;
			DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
			    "mode %x\n", un->un_dp->pio_mode);
		}
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "options %x,", un->un_dp->options);
	}

	un->un_throttle = 2;
	/*
	 * set default max_xfer_size - This should depend on whether the
	 * Block mode is supported by the device or not.
	 */
	un->un_max_xfer_size = MAX_ATA_XFER_SIZE;

	/*
	 * Set write cache enable softstate
	 *
	 * WCE is only supported in ATAPI-4 or higher; for
	 * lower rev devices, must assume write cache is
	 * enabled.
	 */
	mutex_enter(DCD_MUTEX);
	un->un_write_cache_enabled = (devp->dcd_ident->dcd_majvers == 0xffff) ||
	    ((devp->dcd_ident->dcd_majvers & IDENTIFY_80_ATAPI_4) == 0) ||
	    (devp->dcd_ident->dcd_features85 & IDENTIFY_85_WCE) != 0;
	mutex_exit(DCD_MUTEX);

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "dcd_doattach returns good\n");

	return (rval);

error:
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcd_doattach failed\n");
	dcd_free_softstate(un, devi);
	return (rval);
}

#ifdef NOTNEEDED
/*
 * This routine is used to set the block mode of operation by issuing the
 * Set Block mode ata command with the maximum block mode possible
 */
dcd_set_multiple(struct dcd_disk *un)
{
	int status;
	struct udcd_cmd ucmd;
	struct dcd_cmd cdb;
	dev_t	dev;


	/* Zero all the required structure */
	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));

	(void) bzero((caddr_t)&cdb, sizeof (struct dcd_cmd));

	cdb.cmd = ATA_SET_MULTIPLE;
	/*
	 * Here we should pass what needs to go into sector count REGISTER.
	 * Eventhough this field indicates the number of bytes to read we
	 * need to specify the block factor in terms of bytes so that it
	 * will be programmed by the HBA driver into the sector count register.
	 */
	cdb.size = un->un_lbasize * un->un_dp->block_factor;

	cdb.sector_num.lba_num = 0;
	cdb.address_mode = ADD_LBA_MODE;
	cdb.direction = NO_DATA_XFER;

	ucmd.udcd_flags = 0;
	ucmd.udcd_cmd = &cdb;
	ucmd.udcd_bufaddr = NULL;
	ucmd.udcd_buflen = 0;
	ucmd.udcd_flags |= UDCD_SILENT;

	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);


	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);

	return (status);
}
/*
 * The following routine is used only for setting the transfer mode
 * and it is not designed for transferring any other features subcommand.
 */
dcd_set_features(struct dcd_disk *un, uchar_t mode)
{
	int status;
	struct udcd_cmd ucmd;
	struct dcd_cmd cdb;
	dev_t	dev;


	/* Zero all the required structure */
	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));

	(void) bzero((caddr_t)&cdb, sizeof (struct dcd_cmd));

	cdb.cmd = ATA_SET_FEATURES;
	/*
	 * Here we need to pass what needs to go into the sector count register
	 * But in the case of SET FEATURES command the value taken in the
	 * sector count register depends what type of subcommand is
	 * passed in the features register. Since we have defined the size to
	 * be the size in bytes in this context it does not indicate bytes
	 * instead it indicates the mode to be programmed.
	 */
	cdb.size = un->un_lbasize * mode;

	cdb.sector_num.lba_num = 0;
	cdb.address_mode = ADD_LBA_MODE;
	cdb.direction = NO_DATA_XFER;
	cdb.features = ATA_FEATURE_SET_MODE;
	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "size %x, features %x, cmd %x\n",
	    cdb.size, cdb.features, cdb.cmd);

	ucmd.udcd_flags = 0;
	ucmd.udcd_cmd = &cdb;
	ucmd.udcd_bufaddr = NULL;
	ucmd.udcd_buflen = 0;
	ucmd.udcd_flags |= UDCD_SILENT;

	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);

	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);

	return (status);
}
#endif

/*
 * Validate the geometry for this disk, e.g.,
 * see whether it has a valid label.
 */
static int
dcd_validate_geometry(struct dcd_disk *un)
{
	int secsize = 0;
	struct  dcd_device *devp;
	int secdiv;
	int rval;

	ASSERT(mutex_owned(DCD_MUTEX));
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "dcd_validate_geometry: started \n");

	if (un->un_lbasize < 0) {
		return (DCD_BAD_LABEL);
	}

	if (un->un_state == DCD_STATE_PM_SUSPENDED) {
		mutex_exit(DCD_MUTEX);
		if (pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE) !=
		    DDI_SUCCESS) {
			mutex_enter(DCD_MUTEX);
			return (DCD_BAD_LABEL);
		}
		mutex_enter(DCD_MUTEX);
	}

	secsize = un->un_secsize;

	/*
	 * take a log base 2 of sector size (sorry)
	 */
	for (secdiv = 0; secsize = secsize >> 1; secdiv++)
		;
	un->un_secdiv = secdiv;

	/*
	 * Only DIRECT ACCESS devices will have Sun labels.
	 * CD's supposedly have a Sun label, too
	 */

	devp = un->un_dcd;

	if (((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) &&
	    (devp->dcd_ident->dcd_config & ATANON_REMOVABLE)) {
		mutex_exit(DCD_MUTEX);
		rval = cmlb_validate(un->un_dklbhandle, 0, 0);
		mutex_enter(DCD_MUTEX);
		if (rval == ENOMEM)
			return (DCD_NO_MEM_FOR_LABEL);
		else if (rval != 0)
			return (DCD_BAD_LABEL);
	} else {
		/* it should never get here. */
		return (DCD_BAD_LABEL);
	}

	/*
	 * take a log base 2 of logical block size
	 */
	secsize = un->un_lbasize;
	for (secdiv = 0; secsize = secsize >> 1; secdiv++)
		;
	un->un_lbadiv = secdiv;

	/*
	 * take a log base 2 of the multiple of DEV_BSIZE blocks that
	 * make up one logical block
	 */
	secsize = un->un_lbasize >> DEV_BSHIFT;
	for (secdiv = 0; secsize = secsize >> 1; secdiv++)
		;
	un->un_blknoshift = secdiv;
	return (0);
}

/*
 * Unix Entry Points
 */

/* ARGSUSED3 */
static int
dcdopen(dev_t *dev_p, int flag, int otyp, cred_t *cred_p)
{
	dev_t dev = *dev_p;
	int rval = EIO;
	int partmask;
	int nodelay = (flag & (FNDELAY | FNONBLOCK));
	int i;
	char kstatname[KSTAT_STRLEN];
	diskaddr_t lblocks;
	char *partname;

	GET_SOFT_STATE(dev);

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "Inside Open flag %x, otyp %x\n", flag, otyp);

	if (otyp >= OTYPCNT) {
		return (EINVAL);
	}

	partmask = 1 << part;

	/*
	 * We use a semaphore here in order to serialize
	 * open and close requests on the device.
	 */
	sema_p(&un->un_semoclose);

	mutex_enter(DCD_MUTEX);

	if ((un->un_state & DCD_STATE_FATAL) == DCD_STATE_FATAL) {
		rval = ENXIO;
		goto done;
	}

	while (un->un_state == DCD_STATE_SUSPENDED) {
		cv_wait(&un->un_suspend_cv, DCD_MUTEX);
	}

	if ((un->un_state == DCD_STATE_PM_SUSPENDED) && (!nodelay)) {
		mutex_exit(DCD_MUTEX);
		if (pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE)
		    != DDI_SUCCESS) {
			mutex_enter(DCD_MUTEX);
			rval = EIO;
			goto done;
		}
		mutex_enter(DCD_MUTEX);
	}

	/*
	 * set make_dcd_cmd() flags and stat_size here since these
	 * are unlikely to change
	 */
	un->un_cmd_flags = 0;

	un->un_cmd_stat_size = 2;

	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdopen un=0x%p\n",
	    (void *)un);
	/*
	 * check for previous exclusive open
	 */
	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "exclopen=%x, flag=%x, regopen=%x\n",
	    un->un_exclopen, flag, un->un_ocmap.regopen[otyp]);
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "Exclusive open flag %x, partmask %x\n",
	    un->un_exclopen, partmask);

	if (un->un_exclopen & (partmask)) {
failed_exclusive:
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "exclusive open fails\n");
		rval = EBUSY;
		goto done;
	}

	if (flag & FEXCL) {
		int i;
		if (un->un_ocmap.lyropen[part]) {
			goto failed_exclusive;
		}
		for (i = 0; i < (OTYPCNT - 1); i++) {
			if (un->un_ocmap.regopen[i] & (partmask)) {
				goto failed_exclusive;
			}
		}
	}
	if (flag & FWRITE) {
		mutex_exit(DCD_MUTEX);
		if (dcd_check_wp(dev)) {
			sema_v(&un->un_semoclose);
			return (EROFS);
		}
		mutex_enter(DCD_MUTEX);
	}

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "Check Write Protect handled\n");

	if (!nodelay) {
		mutex_exit(DCD_MUTEX);
		if ((rval = dcd_ready_and_valid(dev, un)) != 0) {
			rval = EIO;
		}
		(void) pm_idle_component(DCD_DEVINFO, 0);
		/*
		 * Fail if device is not ready or if the number of disk
		 * blocks is zero or negative for non CD devices.
		 */
		if (rval || cmlb_partinfo(un->un_dklbhandle,
		    part, &lblocks, NULL, &partname, NULL, 0) ||
		    lblocks <= 0) {
			rval = EIO;
			mutex_enter(DCD_MUTEX);
			goto done;
		}
		mutex_enter(DCD_MUTEX);
	}

	if (otyp == OTYP_LYR) {
		un->un_ocmap.lyropen[part]++;
	} else {
		un->un_ocmap.regopen[otyp] |= partmask;
	}

	/*
	 * set up open and exclusive open flags
	 */
	if (flag & FEXCL) {
		un->un_exclopen |= (partmask);
	}


	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "open of part %d type %d\n",
	    part, otyp);

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "Kstats getting updated\n");
	/*
	 * only create kstats for disks, CD kstats created in dcdattach
	 */
	_NOTE(NO_COMPETING_THREADS_NOW);
	mutex_exit(DCD_MUTEX);
	if (un->un_stats == (kstat_t *)0) {
		un->un_stats = kstat_create("dad", instance,
		    NULL, "disk", KSTAT_TYPE_IO, 1,
		    KSTAT_FLAG_PERSISTENT);
		if (un->un_stats) {
			un->un_stats->ks_lock = DCD_MUTEX;
			kstat_install(un->un_stats);
		}

		/*
		 * set up partition statistics for each partition
		 * with number of blocks > 0
		 */
		if (!nodelay) {
			for (i = 0; i < NDKMAP; i++) {
				if ((un->un_pstats[i] == (kstat_t *)0) &&
				    (cmlb_partinfo(un->un_dklbhandle,
				    i, &lblocks, NULL, &partname,
				    NULL, 0) == 0) && lblocks > 0) {
					(void) sprintf(kstatname, "dad%d,%s",
					    instance, partname);
					un->un_pstats[i] = kstat_create("dad",
					    instance,
					    kstatname,
					    "partition",
					    KSTAT_TYPE_IO,
					    1,
					    KSTAT_FLAG_PERSISTENT);
					if (un->un_pstats[i]) {
						un->un_pstats[i]->ks_lock =
						    DCD_MUTEX;
						kstat_install(un->un_pstats[i]);
					}
				}
			}
		}
		/*
		 * set up error kstats
		 */
		(void) dcd_create_errstats(un, instance);
	}
#ifndef lint
	_NOTE(COMPETING_THREADS_NOW);
#endif

	sema_v(&un->un_semoclose);
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "Open success\n");
	return (0);

done:
	mutex_exit(DCD_MUTEX);
	sema_v(&un->un_semoclose);
	return (rval);

}

/*
 * Test if disk is ready and has a valid geometry.
 */
static int
dcd_ready_and_valid(dev_t dev, struct dcd_disk *un)
{
	int rval = 1;
	int g_error = 0;

	mutex_enter(DCD_MUTEX);
	/*
	 * cmds outstanding
	 */
	if (un->un_ncmds == 0) {
		(void) dcd_unit_ready(dev);
	}

	/*
	 * If device is not yet ready here, inform it is offline
	 */
	if (un->un_state == DCD_STATE_NORMAL) {
		rval = dcd_unit_ready(dev);
		if (rval != 0 && rval != EACCES) {
			dcd_offline(un, 1);
			goto done;
		}
	}

	if (un->un_format_in_progress == 0) {
		g_error = dcd_validate_geometry(un);
	}

	/*
	 * check if geometry was valid. We don't check the validity of
	 * geometry for CDROMS.
	 */

	if (g_error == DCD_BAD_LABEL) {
		rval = 1;
		goto done;
	}


	/*
	 * the state has changed; inform the media watch routines
	 */
	un->un_mediastate = DKIO_INSERTED;
	cv_broadcast(&un->un_state_cv);
	rval = 0;

done:
	mutex_exit(DCD_MUTEX);
	return (rval);
}


/*ARGSUSED*/
static int
dcdclose(dev_t dev, int flag, int otyp, cred_t *cred_p)
{
	uchar_t *cp;
	int i;

	GET_SOFT_STATE(dev);


	if (otyp >= OTYPCNT)
		return (ENXIO);

	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "close of part %d type %d\n",
	    part, otyp);
	sema_p(&un->un_semoclose);

	mutex_enter(DCD_MUTEX);

	if (un->un_exclopen & (1<<part)) {
		un->un_exclopen &= ~(1<<part);
	}

	if (otyp == OTYP_LYR) {
		un->un_ocmap.lyropen[part] -= 1;
	} else {
		un->un_ocmap.regopen[otyp] &= ~(1<<part);
	}

	cp = &un->un_ocmap.chkd[0];
	while (cp < &un->un_ocmap.chkd[OCSIZE]) {
		if (*cp != (uchar_t)0) {
			break;
		}
		cp++;
	}

	if (cp == &un->un_ocmap.chkd[OCSIZE]) {
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "last close\n");
		if (un->un_state == DCD_STATE_OFFLINE) {
			dcd_offline(un, 1);
		}

		mutex_exit(DCD_MUTEX);
		(void) cmlb_close(un->un_dklbhandle, 0);

		_NOTE(NO_COMPETING_THREADS_NOW);
		if (un->un_stats) {
			kstat_delete(un->un_stats);
			un->un_stats = 0;
		}
		for (i = 0; i < NDKMAP; i++) {
			if (un->un_pstats[i]) {
				kstat_delete(un->un_pstats[i]);
				un->un_pstats[i] = (kstat_t *)0;
			}
		}

		if (un->un_errstats) {
			kstat_delete(un->un_errstats);
			un->un_errstats = (kstat_t *)0;
		}
		mutex_enter(DCD_MUTEX);

#ifndef lint
		_NOTE(COMPETING_THREADS_NOW);
#endif
	}

	mutex_exit(DCD_MUTEX);
	sema_v(&un->un_semoclose);
	return (0);
}

static void
dcd_offline(struct dcd_disk *un, int bechatty)
{
	if (bechatty)
		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN, "offline\n");

	mutex_exit(DCD_MUTEX);
	cmlb_invalidate(un->un_dklbhandle, 0);
	mutex_enter(DCD_MUTEX);
}

/*
 * Given the device number return the devinfo pointer
 * from the scsi_device structure.
 */
/*ARGSUSED*/
static int
dcdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
{
	dev_t dev;
	struct dcd_disk *un;
	int instance, error;


	switch (infocmd) {
	case DDI_INFO_DEVT2DEVINFO:
		dev = (dev_t)arg;
		instance = DCDUNIT(dev);
		if ((un = ddi_get_soft_state(dcd_state, instance)) == NULL)
			return (DDI_FAILURE);
		*result = (void *) DCD_DEVINFO;
		error = DDI_SUCCESS;
		break;
	case DDI_INFO_DEVT2INSTANCE:
		dev = (dev_t)arg;
		instance = DCDUNIT(dev);
		*result = (void *)(uintptr_t)instance;
		error = DDI_SUCCESS;
		break;
	default:
		error = DDI_FAILURE;
	}
	return (error);
}

/*
 * property operation routine.	return the number of blocks for the partition
 * in question or forward the request to the propery facilities.
 */
static int
dcd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
    char *name, caddr_t valuep, int *lengthp)
{
	struct dcd_disk	*un;

	if ((un = ddi_get_soft_state(dcd_state, ddi_get_instance(dip))) == NULL)
		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
		    name, valuep, lengthp));

	return (cmlb_prop_op(un->un_dklbhandle,
	    dev, dip, prop_op, mod_flags, name, valuep, lengthp,
	    DCDPART(dev), NULL));
}

/*
 * These routines perform raw i/o operations.
 */
/*ARGSUSED*/
void
dcduscsimin(struct buf *bp)
{

}


static void
dcdmin(struct buf *bp)
{
	struct dcd_disk *un;
	int instance;
	minor_t minor = getminor(bp->b_edev);
	instance = minor >> DCDUNIT_SHIFT;
	un = ddi_get_soft_state(dcd_state, instance);

	if (bp->b_bcount > un->un_max_xfer_size)
		bp->b_bcount = un->un_max_xfer_size;
}


/* ARGSUSED2 */
static int
dcdread(dev_t dev, struct uio *uio, cred_t *cred_p)
{
	int secmask;
	GET_SOFT_STATE(dev);
#ifdef lint
	part = part;
#endif /* lint */
	secmask = un->un_secsize - 1;

	if (uio->uio_loffset & ((offset_t)(secmask))) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "file offset not modulo %d\n",
		    un->un_secsize);
		return (EINVAL);
	} else if (uio->uio_iov->iov_len & (secmask)) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "transfer length not modulo %d\n", un->un_secsize);
		return (EINVAL);
	}
	return (physio(dcdstrategy, (struct buf *)0, dev, B_READ, dcdmin, uio));
}

/* ARGSUSED2 */
static int
dcdaread(dev_t dev, struct aio_req *aio, cred_t *cred_p)
{
	int secmask;
	struct uio *uio = aio->aio_uio;
	GET_SOFT_STATE(dev);
#ifdef lint
	part = part;
#endif /* lint */
	secmask = un->un_secsize - 1;

	if (uio->uio_loffset & ((offset_t)(secmask))) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "file offset not modulo %d\n",
		    un->un_secsize);
		return (EINVAL);
	} else if (uio->uio_iov->iov_len & (secmask)) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "transfer length not modulo %d\n", un->un_secsize);
		return (EINVAL);
	}
	return (aphysio(dcdstrategy, anocancel, dev, B_READ, dcdmin, aio));
}

/* ARGSUSED2 */
static int
dcdwrite(dev_t dev, struct uio *uio, cred_t *cred_p)
{
	int secmask;
	GET_SOFT_STATE(dev);
#ifdef lint
	part = part;
#endif /* lint */
	secmask = un->un_secsize - 1;

	if (uio->uio_loffset & ((offset_t)(secmask))) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "file offset not modulo %d\n",
		    un->un_secsize);
		return (EINVAL);
	} else if (uio->uio_iov->iov_len & (secmask)) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "transfer length not modulo %d\n", un->un_secsize);
		return (EINVAL);
	}
	return (physio(dcdstrategy, (struct buf *)0, dev, B_WRITE, dcdmin,
	    uio));
}

/* ARGSUSED2 */
static int
dcdawrite(dev_t dev, struct aio_req *aio, cred_t *cred_p)
{
	int secmask;
	struct uio *uio = aio->aio_uio;
	GET_SOFT_STATE(dev);
#ifdef lint
	part = part;
#endif /* lint */
	secmask = un->un_secsize - 1;

	if (uio->uio_loffset & ((offset_t)(secmask))) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "file offset not modulo %d\n",
		    un->un_secsize);
		return (EINVAL);
	} else if (uio->uio_iov->iov_len & (secmask)) {
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "transfer length not modulo %d\n", un->un_secsize);
		return (EINVAL);
	}
	return (aphysio(dcdstrategy, anocancel, dev, B_WRITE, dcdmin, aio));
}

/*
 * strategy routine
 */
static int
dcdstrategy(struct buf *bp)
{
	struct dcd_disk *un;
	struct diskhd *dp;
	int i;
	minor_t minor = getminor(bp->b_edev);
	diskaddr_t p_lblksrt;
	diskaddr_t lblocks;
	diskaddr_t bn;

	if ((un = ddi_get_soft_state(dcd_state,
	    minor >> DCDUNIT_SHIFT)) == NULL ||
	    un->un_state == DCD_STATE_DUMPING ||
	    ((un->un_state  & DCD_STATE_FATAL) == DCD_STATE_FATAL)) {
		SET_BP_ERROR(bp, ((un) ? ENXIO : EIO));
error:
		bp->b_resid = bp->b_bcount;
		biodone(bp);
		return (0);
	}

	/*
	 * If the request size (buf->b_bcount)is greater than the size
	 * (un->un_max_xfer_size) supported by the target driver fail
	 * the request with EINVAL error code.
	 *
	 * We are not supposed to receive requests exceeding
	 * un->un_max_xfer_size size because the caller is expected to
	 * check what is the maximum size that is supported by this
	 * driver either through ioctl or dcdmin routine(which is private
	 * to this driver).
	 * But we have seen cases (like meta driver(md))where dcdstrategy
	 * called with more than supported size and cause data corruption.
	 */

	if (bp->b_bcount > un->un_max_xfer_size) {
		SET_BP_ERROR(bp, EINVAL);
		goto error;
	}

	TRACE_2(TR_FAC_DADA, TR_DCDSTRATEGY_START,
	    "dcdstrategy_start: bp 0x%p un 0x%p", bp, un);

	/*
	 * Commands may sneak in while we released the mutex in
	 * DDI_SUSPEND, we should block new commands.
	 */
	mutex_enter(DCD_MUTEX);
	while (un->un_state == DCD_STATE_SUSPENDED) {
		cv_wait(&un->un_suspend_cv, DCD_MUTEX);
	}

	if (un->un_state == DCD_STATE_PM_SUSPENDED) {
		mutex_exit(DCD_MUTEX);
		(void) pm_idle_component(DCD_DEVINFO, 0);
		if (pm_raise_power(DCD_DEVINFO, 0,
		    DCD_DEVICE_ACTIVE) !=  DDI_SUCCESS) {
			SET_BP_ERROR(bp, EIO);
			goto error;
		}
		mutex_enter(DCD_MUTEX);
	}
	mutex_exit(DCD_MUTEX);

	/*
	 * Map-in the buffer in case starting address is not word aligned.
	 */

	if (((uintptr_t)bp->b_un.b_addr) & 0x1)
		bp_mapin(bp);

	bp->b_flags &= ~(B_DONE|B_ERROR);
	bp->b_resid = 0;
	bp->av_forw = 0;

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "bp->b_bcount %lx\n", bp->b_bcount);

	if (bp != un->un_sbufp) {
validated:	if (cmlb_partinfo(un->un_dklbhandle,
		    minor & DCDPART_MASK,
		    &lblocks,
		    &p_lblksrt,
		    NULL,
		    NULL,
		    0) == 0) {

			bn = dkblock(bp);

			DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
			    "dkblock(bp) is %llu\n", bn);

			i = 0;
			if (bn < 0) {
				i = -1;
			} else if (bn >= lblocks) {
				/*
				 * For proper comparison, file system block
				 * number has to be scaled to actual CD
				 * transfer size.
				 * Since all the CDROM operations
				 * that have Sun Labels are in the correct
				 * block size this will work for CD's.	This
				 * will have to change when we have different
				 * sector sizes.
				 *
				 * if bn == lblocks,
				 * Not an error, resid == count
				 */
				if (bn > lblocks) {
					i = -1;
				} else {
					i = 1;
				}
			} else if (bp->b_bcount & (un->un_secsize-1)) {
				/*
				 * This should really be:
				 *
				 * ... if (bp->b_bcount & (un->un_lbasize-1))
				 *
				 */
				i = -1;
			} else {
				if (!bp->b_bcount) {
					printf("Waring : Zero read or Write\n");
					goto error;
				}
				/*
				 * sort by absolute block number.
				 */
				bp->b_resid = bn;
				bp->b_resid += p_lblksrt;
				/*
				 * zero out av_back - this will be a signal
				 * to dcdstart to go and fetch the resources
				 */
				bp->av_back = NO_PKT_ALLOCATED;
			}

			/*
			 * Check to see whether or not we are done
			 * (with or without errors).
			 */

			if (i != 0) {
				if (i < 0) {
					bp->b_flags |= B_ERROR;
				}
				goto error;
			}
		} else {
			/*
			 * opened in NDELAY/NONBLOCK mode?
			 * Check if disk is ready and has a valid geometry
			 */
			if (dcd_ready_and_valid(bp->b_edev, un) == 0) {
				goto validated;
			} else {
				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
				    "i/o to invalid geometry\n");
				SET_BP_ERROR(bp, EIO);
				goto error;
			}
		}
	} else if (BP_HAS_NO_PKT(bp)) {
		struct udcd_cmd *tscmdp;
		struct dcd_cmd *tcmdp;
		/*
		 * This indicates that it is a special buffer
		 * This could be a udcd-cmd and hence call bp_mapin just
		 * in case that it could be a PIO command issued.
		 */
		tscmdp = (struct udcd_cmd *)bp->b_forw;
		tcmdp = tscmdp->udcd_cmd;
		if ((tcmdp->cmd != ATA_READ_DMA) && (tcmdp->cmd != 0xc9) &&
		    (tcmdp->cmd != ATA_WRITE_DMA) && (tcmdp->cmd != 0xcb) &&
		    (tcmdp->cmd != IDENTIFY_DMA) &&
		    (tcmdp->cmd != ATA_FLUSH_CACHE)) {
			bp_mapin(bp);
		}
	}

	/*
	 * We are doing it a bit non-standard. That is, the
	 * head of the b_actf chain is *not* the active command-
	 * it is just the head of the wait queue. The reason
	 * we do this is that the head of the b_actf chain is
	 * guaranteed to not be moved by disksort(), so that
	 * our restart command (pointed to by
	 * b_forw) and the head of the wait queue (b_actf) can
	 * have resources granted without it getting lost in
	 * the queue at some later point (where we would have
	 * to go and look for it).
	 */
	mutex_enter(DCD_MUTEX);

	DCD_DO_KSTATS(un, kstat_waitq_enter, bp);

	dp = &un->un_utab;

	if (dp->b_actf == NULL) {
		dp->b_actf = bp;
		dp->b_actl = bp;
	} else if ((un->un_state == DCD_STATE_SUSPENDED) &&
	    bp == un->un_sbufp) {
		bp->b_actf = dp->b_actf;
		dp->b_actf = bp;
	} else {
		TRACE_3(TR_FAC_DADA, TR_DCDSTRATEGY_DISKSORT_START,
		    "dcdstrategy_disksort_start: dp 0x%p bp 0x%p un 0x%p",
		    dp, bp, un);
		disksort(dp, bp);
		TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_DISKSORT_END,
		    "dcdstrategy_disksort_end");
	}

	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "ncmd %x , throttle %x, forw 0x%p\n",
	    un->un_ncmds, un->un_throttle, (void *)dp->b_forw);
	ASSERT(un->un_ncmds >= 0);
	ASSERT(un->un_throttle >= 0);
	if ((un->un_ncmds < un->un_throttle) && (dp->b_forw == NULL)) {
		dcdstart(un);
	} else if (BP_HAS_NO_PKT(dp->b_actf)) {
		struct buf *cmd_bp;

		cmd_bp = dp->b_actf;
		cmd_bp->av_back = ALLOCATING_PKT;
		mutex_exit(DCD_MUTEX);
		/*
		 * try and map this one
		 */
		TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_SMALL_WINDOW_START,
		    "dcdstrategy_small_window_call (begin)");

		make_dcd_cmd(un, cmd_bp, NULL_FUNC);

		TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_SMALL_WINDOW_END,
		    "dcdstrategy_small_window_call (end)");

		/*
		 * there is a small window where the active cmd
		 * completes before make_dcd_cmd returns.
		 * consequently, this cmd never gets started so
		 * we start it from here
		 */
		mutex_enter(DCD_MUTEX);
		if ((un->un_ncmds < un->un_throttle) &&
		    (dp->b_forw == NULL)) {
			dcdstart(un);
		}
	}
	mutex_exit(DCD_MUTEX);

done:
	TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_END, "dcdstrategy_end");
	return (0);
}


/*
 * Unit start and Completion
 * NOTE: we assume that the caller has at least checked for:
 *		(un->un_ncmds < un->un_throttle)
 *	if not, there is no real harm done, dcd_transport() will
 *	return BUSY
 */
static void
dcdstart(struct dcd_disk *un)
{
	int status, sort_key;
	struct buf *bp;
	struct diskhd *dp;
	uchar_t state = un->un_last_state;

	TRACE_1(TR_FAC_DADA, TR_DCDSTART_START, "dcdstart_start: un 0x%p", un);

retry:
	ASSERT(mutex_owned(DCD_MUTEX));

	dp = &un->un_utab;
	if (((bp = dp->b_actf) == NULL) || (bp->av_back == ALLOCATING_PKT) ||
	    (dp->b_forw != NULL)) {
		TRACE_0(TR_FAC_DADA, TR_DCDSTART_NO_WORK_END,
		    "dcdstart_end (no work)");
		return;
	}

	/*
	 * remove from active queue
	 */
	dp->b_actf = bp->b_actf;
	bp->b_actf = 0;

	/*
	 * increment ncmds before calling dcd_transport because dcdintr
	 * may be called before we return from dcd_transport!
	 */
	un->un_ncmds++;

	/*
	 * If measuring stats, mark exit from wait queue and
	 * entrance into run 'queue' if and only if we are
	 * going to actually start a command.
	 * Normally the bp already has a packet at this point
	 */
	DCD_DO_KSTATS(un, kstat_waitq_to_runq, bp);

	mutex_exit(DCD_MUTEX);

	if (BP_HAS_NO_PKT(bp)) {
		make_dcd_cmd(un, bp, dcdrunout);
		if (BP_HAS_NO_PKT(bp) && !(bp->b_flags & B_ERROR)) {
			mutex_enter(DCD_MUTEX);
			DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);

			bp->b_actf = dp->b_actf;
			dp->b_actf = bp;
			New_state(un, DCD_STATE_RWAIT);
			un->un_ncmds--;
			TRACE_0(TR_FAC_DADA, TR_DCDSTART_NO_RESOURCES_END,
			    "dcdstart_end (No Resources)");
			goto done;

		} else if (bp->b_flags & B_ERROR) {
			mutex_enter(DCD_MUTEX);
			DCD_DO_KSTATS(un, kstat_runq_exit, bp);

			un->un_ncmds--;
			bp->b_resid = bp->b_bcount;
			if (bp->b_error == 0) {
				SET_BP_ERROR(bp, EIO);
			}

			/*
			 * restore old state
			 */
			un->un_state = un->un_last_state;
			un->un_last_state = state;

			mutex_exit(DCD_MUTEX);

			biodone(bp);
			mutex_enter(DCD_MUTEX);
			if (un->un_state == DCD_STATE_SUSPENDED) {
				cv_broadcast(&un->un_disk_busy_cv);
			}

			if ((un->un_ncmds < un->un_throttle) &&
			    (dp->b_forw == NULL)) {
				goto retry;
			} else {
				goto done;
			}
		}
	}

	/*
	 * Restore resid from the packet, b_resid had been the
	 * disksort key.
	 */
	sort_key = bp->b_resid;
	bp->b_resid = BP_PKT(bp)->pkt_resid;
	BP_PKT(bp)->pkt_resid = 0;

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "bp->b_resid %lx, pkt_resid %lx\n",
	    bp->b_resid, BP_PKT(bp)->pkt_resid);

	/*
	 * We used to check whether or not to try and link commands here.
	 * Since we have found that there is no performance improvement
	 * for linked commands, this has not made much sense.
	 */
	if ((status = dcd_transport((struct dcd_pkt *)BP_PKT(bp)))
	    != TRAN_ACCEPT) {
		mutex_enter(DCD_MUTEX);
		un->un_ncmds--;
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "transport returned %x\n", status);
		if (status == TRAN_BUSY) {
			DCD_DO_ERRSTATS(un, dcd_transerrs);
			DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
			dcd_handle_tran_busy(bp, dp, un);
			if (un->un_ncmds > 0) {
				bp->b_resid = sort_key;
			}
		} else {
			DCD_DO_KSTATS(un, kstat_runq_exit, bp);
			mutex_exit(DCD_MUTEX);

			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "transport rejected (%d)\n",
			    status);
			SET_BP_ERROR(bp, EIO);
			bp->b_resid = bp->b_bcount;
			if (bp != un->un_sbufp) {
				dcd_destroy_pkt(BP_PKT(bp));
			}
			biodone(bp);

			mutex_enter(DCD_MUTEX);
			if (un->un_state == DCD_STATE_SUSPENDED) {
				cv_broadcast(&un->un_disk_busy_cv);
			}
			if ((un->un_ncmds < un->un_throttle) &&
			    (dp->b_forw == NULL)) {
					goto retry;
			}
		}
	} else {
		mutex_enter(DCD_MUTEX);

		if (dp->b_actf && BP_HAS_NO_PKT(dp->b_actf)) {
			struct buf *cmd_bp;

			cmd_bp = dp->b_actf;
			cmd_bp->av_back = ALLOCATING_PKT;
			mutex_exit(DCD_MUTEX);
			/*
			 * try and map this one
			 */
			TRACE_0(TR_FAC_DADA, TR_DCASTART_SMALL_WINDOW_START,
			    "dcdstart_small_window_start");

			make_dcd_cmd(un, cmd_bp, NULL_FUNC);

			TRACE_0(TR_FAC_DADA, TR_DCDSTART_SMALL_WINDOW_END,
			    "dcdstart_small_window_end");
			/*
			 * there is a small window where the active cmd
			 * completes before make_dcd_cmd returns.
			 * consequently, this cmd never gets started so
			 * we start it from here
			 */
			mutex_enter(DCD_MUTEX);
			if ((un->un_ncmds < un->un_throttle) &&
			    (dp->b_forw == NULL)) {
				goto retry;
			}
		}
	}

done:
	ASSERT(mutex_owned(DCD_MUTEX));
	TRACE_0(TR_FAC_DADA, TR_DCDSTART_END, "dcdstart_end");
}

/*
 * make_dcd_cmd: create a pkt
 */
static void
make_dcd_cmd(struct dcd_disk *un, struct buf *bp, int (*func)())
{
	auto int count, com, direction;
	struct dcd_pkt *pkt;
	int flags, tval;

	_NOTE(DATA_READABLE_WITHOUT_LOCK(dcd_disk::un_dp))
	TRACE_3(TR_FAC_DADA, TR_MAKE_DCD_CMD_START,
	    "make_dcd_cmd_start: un 0x%p bp 0x%p un 0x%p", un, bp, un);


	flags = un->un_cmd_flags;

	if (bp != un->un_sbufp) {
		int partition = DCDPART(bp->b_edev);
		diskaddr_t p_lblksrt;
		diskaddr_t lblocks;
		long secnt;
		uint32_t blkno;
		int dkl_nblk, delta;
		long resid;

		if (cmlb_partinfo(un->un_dklbhandle,
		    partition,
		    &lblocks,
		    &p_lblksrt,
		    NULL,
		    NULL,
		    0) != NULL) {
			lblocks = 0;
			p_lblksrt = 0;
		}

		dkl_nblk = (int)lblocks;

		/*
		 * Make sure we don't run off the end of a partition.
		 *
		 * Put this test here so that we can adjust b_count
		 * to accurately reflect the actual amount we are
		 * goint to transfer.
		 */

		/*
		 * First, compute partition-relative block number
		 */
		blkno = dkblock(bp);
		secnt = (bp->b_bcount + (un->un_secsize - 1)) >> un->un_secdiv;
		count = MIN(secnt, dkl_nblk - blkno);
		if (count != secnt) {
			/*
			 * We have an overrun
			 */
			resid = (secnt - count) << un->un_secdiv;
			DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
			    "overrun by %ld sectors\n",
			    secnt - count);
			bp->b_bcount -= resid;
		} else {
			resid = 0;
		}

		/*
		 * Adjust block number to absolute
		 */
		delta = (int)p_lblksrt;
		blkno += delta;

		mutex_enter(DCD_MUTEX);
		/*
		 * This is for devices having block size different from
		 * from DEV_BSIZE (e.g. 2K CDROMs).
		 */
		if (un->un_lbasize != un->un_secsize) {
			blkno >>= un->un_blknoshift;
			count >>= un->un_blknoshift;
		}
		mutex_exit(DCD_MUTEX);

		TRACE_0(TR_FAC_DADA, TR_MAKE_DCD_CMD_INIT_PKT_START,
		    "make_dcd_cmd_init_pkt_call (begin)");
		pkt = dcd_init_pkt(ROUTE, NULL, bp,
		    (uint32_t)sizeof (struct dcd_cmd),
		    un->un_cmd_stat_size, PP_LEN, PKT_CONSISTENT,
		    func, (caddr_t)un);
		TRACE_1(TR_FAC_DADA, TR_MAKE_DCD_CMD_INIT_PKT_END,
		    "make_dcd_cmd_init_pkt_call (end): pkt 0x%p", pkt);
		if (!pkt) {
			bp->b_bcount += resid;
			bp->av_back = NO_PKT_ALLOCATED;
			TRACE_0(TR_FAC_DADA,
			    TR_MAKE_DCD_CMD_NO_PKT_ALLOCATED1_END,
			    "make_dcd_cmd_end (NO_PKT_ALLOCATED1)");
			return;
		}
		if (bp->b_flags & B_READ) {
			if ((un->un_dp->options & DMA_SUPPORTTED) ==
			    DMA_SUPPORTTED) {
				com = ATA_READ_DMA;
			} else {
				if (un->un_dp->options & BLOCK_MODE)
					com = ATA_READ_MULTIPLE;
				else
					com = ATA_READ;
			}
			direction = DATA_READ;
		} else {
			if ((un->un_dp->options & DMA_SUPPORTTED) ==
			    DMA_SUPPORTTED) {
				com = ATA_WRITE_DMA;
			} else {
				if (un->un_dp->options & BLOCK_MODE)
					com = ATA_WRITE_MULTIPLE;
				else
					com = ATA_WRITE;
			}
			direction = DATA_WRITE;
		}

		/*
		 * Save the resid in the packet, temporarily until
		 * we transport the command.
		 */
		pkt->pkt_resid = resid;

		makecommand(pkt, flags, com, blkno, ADD_LBA_MODE,
		    bp->b_bcount, direction, 0);
		tval = dcd_io_time;
	} else {

		struct udcd_cmd *scmd = (struct udcd_cmd *)bp->b_forw;

		/*
		 * set options
		 */
		if ((scmd->udcd_flags & UDCD_SILENT) && !(DEBUGGING)) {
			flags |= FLAG_SILENT;
		}
		if (scmd->udcd_flags &  UDCD_DIAGNOSE)
			flags |= FLAG_DIAGNOSE;

		if (scmd->udcd_flags & UDCD_NOINTR)
			flags |= FLAG_NOINTR;

		pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
		    (bp->b_bcount)? bp: NULL,
		    (uint32_t)sizeof (struct dcd_cmd),
		    2, PP_LEN, PKT_CONSISTENT, func, (caddr_t)un);

		if (!pkt) {
			bp->av_back = NO_PKT_ALLOCATED;
			return;
		}

		makecommand(pkt, 0, scmd->udcd_cmd->cmd,
		    scmd->udcd_cmd->sector_num.lba_num,
		    scmd->udcd_cmd->address_mode,
		    scmd->udcd_cmd->size,
		    scmd->udcd_cmd->direction, scmd->udcd_cmd->features);

		pkt->pkt_flags = flags;
		if (scmd->udcd_timeout == 0)
			tval = dcd_io_time;
		else
			tval = scmd->udcd_timeout;
		/* UDAD interface should be decided. */
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "udcd interface\n");
	}

	pkt->pkt_comp = dcdintr;
	pkt->pkt_time = tval;
	PKT_SET_BP(pkt, bp);
	bp->av_back = (struct buf *)pkt;

	TRACE_0(TR_FAC_DADA, TR_MAKE_DCD_CMD_END, "make_dcd_cmd_end");
}

/*
 * Command completion processing
 */
static void
dcdintr(struct dcd_pkt *pkt)
{
	struct dcd_disk *un;
	struct buf *bp;
	int action;
	int status;

	bp = PKT_GET_BP(pkt);
	un = ddi_get_soft_state(dcd_state, DCDUNIT(bp->b_edev));

	TRACE_1(TR_FAC_DADA, TR_DCDINTR_START, "dcdintr_start: un 0x%p", un);
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdintr\n");

	mutex_enter(DCD_MUTEX);
	un->un_ncmds--;
	DCD_DO_KSTATS(un, kstat_runq_exit, bp);
	ASSERT(un->un_ncmds >= 0);

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "reason %x and Status %x\n", pkt->pkt_reason, SCBP_C(pkt));

	/*
	 * do most common case first
	 */
	if ((pkt->pkt_reason == CMD_CMPLT) && (SCBP_C(pkt) == 0)) {
		int com = GETATACMD((struct dcd_cmd *)pkt->pkt_cdbp);

		if (un->un_state == DCD_STATE_OFFLINE) {
			un->un_state = un->un_last_state;
			dcd_log(DCD_DEVINFO, dcd_label, CE_NOTE,
			    (const char *) diskokay);
		}
		/*
		 * If the command is a read or a write, and we have
		 * a non-zero pkt_resid, that is an error. We should
		 * attempt to retry the operation if possible.
		 */
		action = COMMAND_DONE;
		if (pkt->pkt_resid && (com == ATA_READ || com == ATA_WRITE)) {
			DCD_DO_ERRSTATS(un, dcd_harderrs);
			if ((int)PKT_GET_RETRY_CNT(pkt) < dcd_retry_count) {
				PKT_INCR_RETRY_CNT(pkt, 1);
				action = QUE_COMMAND;
			} else {
				/*
				 * if we have exhausted retries
				 * a command with a residual is in error in
				 * this case.
				 */
				action = COMMAND_DONE_ERROR;
			}
			dcd_log(DCD_DEVINFO, dcd_label,
			    CE_WARN, "incomplete %s- %s\n",
			    (bp->b_flags & B_READ)? "read" : "write",
			    (action == QUE_COMMAND)? "retrying" :
			    "giving up");
		}

		/*
		 * pkt_resid will reflect, at this point, a residual
		 * of how many bytes left to be transferred there were
		 * from the actual scsi command. Add this to b_resid i.e
		 * the amount this driver could not see to transfer,
		 * to get the total number of bytes not transfered.
		 */
		if (action != QUE_COMMAND) {
			bp->b_resid += pkt->pkt_resid;
		}

	} else if (pkt->pkt_reason != CMD_CMPLT) {
		action = dcd_handle_incomplete(un, bp);
	}

	/*
	 * If we are in the middle of syncing or dumping, we have got
	 * here because dcd_transport has called us explictly after
	 * completing the command in a polled mode. We don't want to
	 * have a recursive call into dcd_transport again.
	 */
	if (ddi_in_panic() && (action == QUE_COMMAND)) {
		action = COMMAND_DONE_ERROR;
	}

	/*
	 * save pkt reason; consecutive failures are not reported unless
	 * fatal
	 * do not reset last_pkt_reason when the cmd was retried and
	 * succeeded because
	 * there maybe more commands comming back with last_pkt_reason
	 */
	if ((un->un_last_pkt_reason != pkt->pkt_reason) &&
	    ((pkt->pkt_reason != CMD_CMPLT) ||
	    (PKT_GET_RETRY_CNT(pkt) == 0))) {
		un->un_last_pkt_reason = pkt->pkt_reason;
	}

	switch (action) {
	case COMMAND_DONE_ERROR:
error:
		if (bp->b_resid == 0) {
			bp->b_resid = bp->b_bcount;
		}
		if (bp->b_error == 0) {
			struct	dcd_cmd *cdbp = (struct dcd_cmd *)pkt->pkt_cdbp;
			if (cdbp->cmd == ATA_FLUSH_CACHE &&
			    (pkt->pkt_scbp[0] & STATUS_ATA_ERR) &&
			    (pkt->pkt_scbp[1] & ERR_ABORT)) {
				SET_BP_ERROR(bp, ENOTSUP);
				un->un_flush_not_supported = 1;
			} else {
				SET_BP_ERROR(bp, EIO);
			}
		}
		bp->b_flags |= B_ERROR;
		/*FALLTHROUGH*/
	case COMMAND_DONE:
		dcddone_and_mutex_exit(un, bp);

		TRACE_0(TR_FAC_DADA, TR_DCDINTR_COMMAND_DONE_END,
		    "dcdintr_end (COMMAND_DONE)");
		return;

	case QUE_COMMAND:
		if (un->un_ncmds >= un->un_throttle) {
			struct diskhd *dp = &un->un_utab;

			bp->b_actf = dp->b_actf;
			dp->b_actf = bp;

			DCD_DO_KSTATS(un, kstat_waitq_enter, bp);

			mutex_exit(DCD_MUTEX);
			goto exit;
		}

		un->un_ncmds++;
		/* reset the pkt reason again */
		pkt->pkt_reason = 0;
		DCD_DO_KSTATS(un, kstat_runq_enter, bp);
		mutex_exit(DCD_MUTEX);
		if ((status = dcd_transport(BP_PKT(bp))) != TRAN_ACCEPT) {
			struct diskhd *dp = &un->un_utab;

			mutex_enter(DCD_MUTEX);
			un->un_ncmds--;
			if (status == TRAN_BUSY) {
				DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
				dcd_handle_tran_busy(bp, dp, un);
				mutex_exit(DCD_MUTEX);
				goto exit;
			}
			DCD_DO_ERRSTATS(un, dcd_transerrs);
			DCD_DO_KSTATS(un, kstat_runq_exit, bp);

			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "requeue of command fails (%x)\n", status);
			SET_BP_ERROR(bp, EIO);
			bp->b_resid = bp->b_bcount;

			dcddone_and_mutex_exit(un, bp);
			goto exit;
		}
		break;

	case JUST_RETURN:
	default:
		DCD_DO_KSTATS(un, kstat_waitq_enter, bp);
		mutex_exit(DCD_MUTEX);
		break;
	}

exit:
	TRACE_0(TR_FAC_DADA, TR_DCDINTR_END, "dcdintr_end");
}


/*
 * Done with a command.
 */
static void
dcddone_and_mutex_exit(struct dcd_disk *un, register struct buf *bp)
{
	struct diskhd *dp;

	TRACE_1(TR_FAC_DADA, TR_DCDONE_START, "dcddone_start: un 0x%p", un);

	_NOTE(LOCK_RELEASED_AS_SIDE_EFFECT(&un->un_dcd->dcd_mutex));

	dp = &un->un_utab;
	if (bp == dp->b_forw) {
		dp->b_forw = NULL;
	}

	if (un->un_stats) {
		ulong_t n_done = bp->b_bcount - bp->b_resid;
		if (bp->b_flags & B_READ) {
			IOSP->reads++;
			IOSP->nread += n_done;
		} else {
			IOSP->writes++;
			IOSP->nwritten += n_done;
		}
	}
	if (IO_PARTITION_STATS) {
		ulong_t n_done = bp->b_bcount - bp->b_resid;
		if (bp->b_flags & B_READ) {
			IOSP_PARTITION->reads++;
			IOSP_PARTITION->nread += n_done;
		} else {
			IOSP_PARTITION->writes++;
			IOSP_PARTITION->nwritten += n_done;
		}
	}

	/*
	 * Start the next one before releasing resources on this one
	 */
	if (un->un_state == DCD_STATE_SUSPENDED) {
		cv_broadcast(&un->un_disk_busy_cv);
	} else if (dp->b_actf && (un->un_ncmds < un->un_throttle) &&
	    (dp->b_forw == NULL && un->un_state != DCD_STATE_SUSPENDED)) {
		dcdstart(un);
	}

	mutex_exit(DCD_MUTEX);

	if (bp != un->un_sbufp) {
		dcd_destroy_pkt(BP_PKT(bp));
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "regular done: resid %ld\n", bp->b_resid);
	} else {
		ASSERT(un->un_sbuf_busy);
	}
	TRACE_0(TR_FAC_DADA, TR_DCDDONE_BIODONE_CALL, "dcddone_biodone_call");

	biodone(bp);

	(void) pm_idle_component(DCD_DEVINFO, 0);

	TRACE_0(TR_FAC_DADA, TR_DCDDONE_END, "dcddone end");
}


/*
 * reset the disk unless the transport layer has already
 * cleared the problem
 */
#define	C1	(STAT_ATA_BUS_RESET|STAT_ATA_DEV_RESET|STAT_ATA_ABORTED)
static void
dcd_reset_disk(struct dcd_disk *un, struct dcd_pkt *pkt)
{

	if ((pkt->pkt_statistics & C1) == 0) {
		mutex_exit(DCD_MUTEX);
		if (!dcd_reset(ROUTE, RESET_ALL)) {
			DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
			    "Reset failed");
		}
		mutex_enter(DCD_MUTEX);
	}
}

static int
dcd_handle_incomplete(struct dcd_disk *un, struct buf *bp)
{
	static char *fail = "ATA transport failed: reason '%s': %s\n";
	static char *notresp = "disk not responding to selection\n";
	int rval = COMMAND_DONE_ERROR;
	int action = COMMAND_SOFT_ERROR;
	struct dcd_pkt *pkt = BP_PKT(bp);
	int be_chatty = (un->un_state != DCD_STATE_SUSPENDED) &&
	    (bp != un->un_sbufp || !(pkt->pkt_flags & FLAG_SILENT));

	ASSERT(mutex_owned(DCD_MUTEX));

	switch (pkt->pkt_reason) {

	case CMD_TIMEOUT:
		/*
		 * This Indicates the already the HBA would  have reset
		 * so Just indicate to retry the command
		 */
		break;

	case CMD_INCOMPLETE:
		action = dcd_check_error(un, bp);
		DCD_DO_ERRSTATS(un, dcd_transerrs);
		if (action == COMMAND_HARD_ERROR) {
			(void) dcd_reset_disk(un, pkt);
		}
		break;

	case CMD_FATAL:
		/*
		 * Something drastic has gone wrong
		 */
		break;
	case CMD_DMA_DERR:
	case CMD_DATA_OVR:
		/* FALLTHROUGH */

	default:
		/*
		 * the target may still be running the	command,
		 * so we should try and reset that target.
		 */
		DCD_DO_ERRSTATS(un, dcd_transerrs);
		if ((pkt->pkt_reason != CMD_RESET) &&
		    (pkt->pkt_reason != CMD_ABORTED)) {
			(void) dcd_reset_disk(un, pkt);
		}
		break;
	}

	/*
	 * If pkt_reason is CMD_RESET/ABORTED, chances are that this pkt got
	 * reset/aborted because another disk on this bus caused it.
	 * The disk that caused it, should get CMD_TIMEOUT with pkt_statistics
	 * of STAT_TIMEOUT/STAT_DEV_RESET
	 */
	if ((pkt->pkt_reason == CMD_RESET) ||(pkt->pkt_reason == CMD_ABORTED)) {
		/* To be written : XXX */
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "Command aborted\n");
	}

	if (bp == un->un_sbufp && (pkt->pkt_flags & FLAG_DIAGNOSE)) {
		rval = COMMAND_DONE_ERROR;
	} else {
		if ((rval == COMMAND_DONE_ERROR) &&
		    (action == COMMAND_SOFT_ERROR) &&
		    ((int)PKT_GET_RETRY_CNT(pkt) < dcd_retry_count)) {
			PKT_INCR_RETRY_CNT(pkt, 1);
			rval = QUE_COMMAND;
		}
	}

	if (pkt->pkt_reason == CMD_INCOMPLETE && rval == COMMAND_DONE_ERROR) {
		/*
		 * Looks like someone turned off this shoebox.
		 */
		if (un->un_state != DCD_STATE_OFFLINE) {
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    (const char *) notresp);
			New_state(un, DCD_STATE_OFFLINE);
		}
	} else if (pkt->pkt_reason == CMD_FATAL) {
		/*
		 * Suppressing the following message for the time being
		 * dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
		 * (const char *) notresp);
		 */
		PKT_INCR_RETRY_CNT(pkt, 6);
		rval = COMMAND_DONE_ERROR;
		New_state(un, DCD_STATE_FATAL);
	} else if (be_chatty) {
		int in_panic = ddi_in_panic();
		if (!in_panic || (rval == COMMAND_DONE_ERROR)) {
			if (((pkt->pkt_reason != un->un_last_pkt_reason) &&
			    (pkt->pkt_reason != CMD_RESET)) ||
			    (rval == COMMAND_DONE_ERROR) ||
			    (dcd_error_level == DCD_ERR_ALL)) {
				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
				    fail, dcd_rname(pkt->pkt_reason),
				    (rval == COMMAND_DONE_ERROR) ?
				    "giving up": "retrying command");
				DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
				    "retrycount=%x\n",
				    PKT_GET_RETRY_CNT(pkt));
			}
		}
	}
error:
	return (rval);
}

static int
dcd_check_error(struct dcd_disk *un, struct buf *bp)
{
	struct diskhd *dp = &un->un_utab;
	struct dcd_pkt *pkt = BP_PKT(bp);
	int rval = 0;
	unsigned char status;
	unsigned char error;

	TRACE_0(TR_FAC_DADA, TR_DCD_CHECK_ERROR_START, "dcd_check_error_start");
	ASSERT(mutex_owned(DCD_MUTEX));

	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "Pkt: 0x%p dp: 0x%p\n", (void *)pkt, (void *)dp);

	/*
	 * Here we need to check status first and then if error is indicated
	 * Then the error register.
	 */

	status = (pkt->pkt_scbp)[0];
	if ((status & STATUS_ATA_DWF) == STATUS_ATA_DWF) {
		/*
		 * There has been a Device Fault  - reason for such error
		 * is vendor specific
		 * Action to be taken is - Indicate error and reset device.
		 */

		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN, "Device Fault\n");
		rval = COMMAND_HARD_ERROR;
	} else if ((status & STATUS_ATA_CORR) == STATUS_ATA_CORR) {

		/*
		 * The sector read or written is marginal and hence ECC
		 * Correction has been applied. Indicate to repair
		 * Here we need to probably re-assign based on the badblock
		 * mapping.
		 */

		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
		    "Soft Error on block %x\n",
		    ((struct dcd_cmd *)pkt->pkt_cdbp)->sector_num.lba_num);
		rval = COMMAND_SOFT_ERROR;
	} else if ((status & STATUS_ATA_ERR) == STATUS_ATA_ERR) {
		error = pkt->pkt_scbp[1];

		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
		    "Command:0x%x,Error:0x%x,Status:0x%x\n",
		    GETATACMD((struct dcd_cmd *)pkt->pkt_cdbp),
		    error, status);
		if ((error &  ERR_AMNF) == ERR_AMNF) {
			/* Address make not found */
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "Address Mark Not Found");
		} else if ((error & ERR_TKONF) == ERR_TKONF) {
			/* Track 0 Not found */
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "Track 0 Not found \n");
		} else if ((error & ERR_IDNF) == ERR_IDNF) {
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    " ID not found \n");
		} else if ((error &  ERR_UNC) == ERR_UNC) {
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "Uncorrectable data Error: Block %x\n",
			    ((struct dcd_cmd *)pkt->pkt_cdbp)->
			    sector_num.lba_num);
		} else if ((error & ERR_BBK) == ERR_BBK) {
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "Bad block detected: Block %x\n",
			    ((struct dcd_cmd *)pkt->pkt_cdbp)->
			    sector_num.lba_num);
		} else if ((error & ERR_ABORT) == ERR_ABORT) {
			/* Aborted Command */
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    " Aborted Command \n");
		}
		/*
		 * Return the soft error so that the command
		 * will be retried.
		 */
		rval = COMMAND_SOFT_ERROR;
	}

	TRACE_0(TR_FAC_DADA, TR_DCD_CHECK_ERROR_END, "dcd_check_error_end");
	return (rval);
}


/*
 *	System Crash Dump routine
 */

#define	NDUMP_RETRIES	5

static int
dcddump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
{
	struct dcd_pkt *pkt;
	int i;
	struct buf local, *bp;
	int err;
	unsigned char com;
	diskaddr_t p_lblksrt;
	diskaddr_t lblocks;

	GET_SOFT_STATE(dev);
#ifdef lint
	part = part;
#endif /* lint */

	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*un))

	if ((un->un_state & DCD_STATE_FATAL) == DCD_STATE_FATAL)
		return (ENXIO);

	if (cmlb_partinfo(un->un_dklbhandle, DCDPART(dev),
	    &lblocks, &p_lblksrt, NULL, NULL, 0))
		return (ENXIO);

	if (blkno+nblk > lblocks) {
		return (EINVAL);
	}


	if ((un->un_state == DCD_STATE_SUSPENDED) ||
	    (un->un_state == DCD_STATE_PM_SUSPENDED)) {
		if (pm_raise_power(DCD_DEVINFO, 0,
		    DCD_DEVICE_ACTIVE) != DDI_SUCCESS) {
			return (EIO);
		}
	}

	/*
	 * When cpr calls dcddump, we know that dad is in a
	 * a good state, so no bus reset is required
	 */
	un->un_throttle = 0;

	if ((un->un_state != DCD_STATE_SUSPENDED) &&
	    (un->un_state != DCD_STATE_DUMPING)) {

		New_state(un, DCD_STATE_DUMPING);

		/*
		 * Reset the bus. I'd like to not have to do this,
		 * but this is the safest thing to do...
		 */

		if (dcd_reset(ROUTE, RESET_ALL) == 0) {
			return (EIO);
		}

	}

	blkno += p_lblksrt;

	/*
	 * It should be safe to call the allocator here without
	 * worrying about being locked for DVMA mapping because
	 * the address we're passed is already a DVMA mapping
	 *
	 * We are also not going to worry about semaphore ownership
	 * in the dump buffer. Dumping is single threaded at present.
	 */

	bp = &local;
	bzero((caddr_t)bp, sizeof (*bp));
	bp->b_flags = B_BUSY;
	bp->b_un.b_addr = addr;
	bp->b_bcount = nblk << DEV_BSHIFT;
	bp->b_resid = 0;

	for (i = 0; i < NDUMP_RETRIES; i++) {
		bp->b_flags &= ~B_ERROR;
		if ((pkt = dcd_init_pkt(ROUTE, NULL, bp,
		    (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
		    PKT_CONSISTENT, NULL_FUNC, NULL)) != NULL) {
			break;
		}
		if (i == 0) {
			if (bp->b_flags & B_ERROR) {
				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
				    "no resources for dumping; "
				    "error code: 0x%x, retrying",
				    geterror(bp));
			} else {
				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
				    "no resources for dumping; retrying");
			}
		} else if (i != (NDUMP_RETRIES - 1)) {
			if (bp->b_flags & B_ERROR) {
				dcd_log(DCD_DEVINFO, dcd_label, CE_CONT, "no "
				    "resources for dumping; error code: 0x%x, "
				    "retrying\n", geterror(bp));
			}
		} else {
			if (bp->b_flags & B_ERROR) {
				dcd_log(DCD_DEVINFO, dcd_label, CE_CONT,
				    "no resources for dumping; "
				    "error code: 0x%x, retries failed, "
				    "giving up.\n", geterror(bp));
			} else {
				dcd_log(DCD_DEVINFO, dcd_label, CE_CONT,
				    "no resources for dumping; "
				    "retries failed, giving up.\n");
			}
			return (EIO);
		}
		delay(10);
	}
	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
		com = ATA_WRITE_DMA;
	} else {
		if (un->un_dp->options & BLOCK_MODE)
			com = ATA_WRITE_MULTIPLE;
		else
			com = ATA_WRITE;
	}

	makecommand(pkt, 0, com, blkno, ADD_LBA_MODE,
	    (int)nblk*un->un_secsize, DATA_WRITE, 0);

	for (err = EIO, i = 0; i < NDUMP_RETRIES && err == EIO; i++) {

		if (dcd_poll(pkt) == 0) {
			switch (SCBP_C(pkt)) {
			case STATUS_GOOD:
				if (pkt->pkt_resid == 0) {
					err = 0;
				}
				break;
			case STATUS_ATA_BUSY:
				(void) dcd_reset(ROUTE, RESET_TARGET);
				break;
			default:
				mutex_enter(DCD_MUTEX);
				(void) dcd_reset_disk(un, pkt);
				mutex_exit(DCD_MUTEX);
				break;
			}
		} else if (i > NDUMP_RETRIES/2) {
			(void) dcd_reset(ROUTE, RESET_ALL);
		}

	}
	dcd_destroy_pkt(pkt);
	return (err);
}

/*
 * This routine implements the ioctl calls.  It is called
 * from the device switch at normal priority.
 */
/* ARGSUSED3 */
static int
dcdioctl(dev_t dev, int cmd, intptr_t arg, int flag,
	cred_t *cred_p, int *rval_p)
{
	auto int32_t data[512 / (sizeof (int32_t))];
	struct dk_cinfo *info;
	struct dk_minfo media_info;
	struct udcd_cmd *scmd;
	int i, err;
	enum uio_seg uioseg = 0;
	enum dkio_state state = 0;
#ifdef _MULTI_DATAMODEL
	struct dadkio_rwcmd rwcmd;
#endif
	struct dadkio_rwcmd32 rwcmd32;
	struct dcd_cmd dcdcmd;

	GET_SOFT_STATE(dev);
#ifdef lint
	part = part;
	state = state;
	uioseg = uioseg;
#endif  /* lint */

	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "dcd_ioctl : cmd %x, arg %lx\n", cmd, arg);

	bzero((caddr_t)data, sizeof (data));

	switch (cmd) {

#ifdef DCDDEBUG
/*
 * Following ioctl are for testing RESET/ABORTS
 */
#define	DKIOCRESET	(DKIOC|14)
#define	DKIOCABORT	(DKIOC|15)

	case DKIOCRESET:
		if (ddi_copyin((caddr_t)arg, (caddr_t)data, 4, flag))
			return (EFAULT);
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "DKIOCRESET: data = 0x%x\n", data[0]);
		if (dcd_reset(ROUTE, data[0])) {
			return (0);
		} else {
			return (EIO);
		}
	case DKIOCABORT:
		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "DKIOCABORT:\n");
		if (dcd_abort(ROUTE, (struct dcd_pkt *)0)) {
			return (0);
		} else {
			return (EIO);
		}
#endif

	case DKIOCINFO:
		/*
		 * Controller Information
		 */
		info = (struct dk_cinfo *)data;

		mutex_enter(DCD_MUTEX);
		switch (un->un_dp->ctype) {
		default:
			info->dki_ctype = DKC_DIRECT;
			break;
		}
		mutex_exit(DCD_MUTEX);
		info->dki_cnum = ddi_get_instance(ddi_get_parent(DCD_DEVINFO));
		(void) strcpy(info->dki_cname,
		    ddi_get_name(ddi_get_parent(DCD_DEVINFO)));
		/*
		 * Unit Information
		 */
		info->dki_unit = ddi_get_instance(DCD_DEVINFO);
		info->dki_slave = (Tgt(DCD_DCD_DEVP)<<3);
		(void) strcpy(info->dki_dname, ddi_driver_name(DCD_DEVINFO));
		info->dki_flags = DKI_FMTVOL;
		info->dki_partition = DCDPART(dev);

		/*
		 * Max Transfer size of this device in blocks
		 */
		info->dki_maxtransfer = un->un_max_xfer_size / DEV_BSIZE;

		/*
		 * We can't get from here to there yet
		 */
		info->dki_addr = 0;
		info->dki_space = 0;
		info->dki_prio = 0;
		info->dki_vec = 0;

		i = sizeof (struct dk_cinfo);
		if (ddi_copyout((caddr_t)data, (caddr_t)arg, i, flag))
			return (EFAULT);
		else
			return (0);

	case DKIOCGMEDIAINFO:
		/*
		 * As dad target driver is used for IDE disks only
		 * Can keep the return value hardcoded to FIXED_DISK
		 */
		media_info.dki_media_type = DK_FIXED_DISK;

		mutex_enter(DCD_MUTEX);
		media_info.dki_lbsize = un->un_lbasize;
		media_info.dki_capacity = un->un_diskcapacity;
		mutex_exit(DCD_MUTEX);

		if (ddi_copyout(&media_info, (caddr_t)arg,
		    sizeof (struct dk_minfo), flag))
			return (EFAULT);
		else
			return (0);

	case DKIOCGGEOM:
	case DKIOCGVTOC:
	case DKIOCGETEFI:

		mutex_enter(DCD_MUTEX);
		if (un->un_ncmds == 0) {
			if ((err = dcd_unit_ready(dev)) != 0) {
				mutex_exit(DCD_MUTEX);
				return (err);
			}
		}

		mutex_exit(DCD_MUTEX);
		err = cmlb_ioctl(un->un_dklbhandle, dev, cmd,
		    arg, flag, cred_p, rval_p, 0);
		return (err);

	case DKIOCGAPART:
	case DKIOCSAPART:
	case DKIOCSGEOM:
	case DKIOCSVTOC:
	case DKIOCSETEFI:
	case DKIOCPARTITION:
	case DKIOCPARTINFO:
	case DKIOCGMBOOT:
	case DKIOCSMBOOT:

		err = cmlb_ioctl(un->un_dklbhandle, dev, cmd,
		    arg, flag, cred_p, rval_p, 0);
		return (err);

	case DIOCTL_RWCMD:
		if (drv_priv(cred_p) != 0) {
			return (EPERM);
		}

#ifdef _MULTI_DATAMODEL
		switch (ddi_model_convert_from(flag & FMODELS)) {
		case DDI_MODEL_NONE:
			if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
			    sizeof (struct dadkio_rwcmd), flag)) {
				return (EFAULT);
			}
			rwcmd32.cmd = rwcmd.cmd;
			rwcmd32.flags = rwcmd.flags;
			rwcmd32.blkaddr = rwcmd.blkaddr;
			rwcmd32.buflen = rwcmd.buflen;
			rwcmd32.bufaddr = (caddr32_t)(uintptr_t)rwcmd.bufaddr;
			break;
		case DDI_MODEL_ILP32:
			if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
			    sizeof (struct dadkio_rwcmd32), flag)) {
				return (EFAULT);
			}
			break;
		}
#else
		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
		    sizeof (struct dadkio_rwcmd32), flag)) {
			return (EFAULT);
		}
#endif
		mutex_enter(DCD_MUTEX);

		uioseg  = UIO_SYSSPACE;
		scmd = (struct udcd_cmd *)data;
		scmd->udcd_cmd = &dcdcmd;
		/*
		 * Convert the dadkio_rwcmd structure to udcd_cmd so that
		 * it can take the normal path to get the io done
		 */
		if (rwcmd32.cmd == DADKIO_RWCMD_READ) {
			if ((un->un_dp->options & DMA_SUPPORTTED) ==
			    DMA_SUPPORTTED)
				scmd->udcd_cmd->cmd = ATA_READ_DMA;
			else
				scmd->udcd_cmd->cmd = ATA_READ;
			scmd->udcd_cmd->address_mode = ADD_LBA_MODE;
			scmd->udcd_cmd->direction = DATA_READ;
			scmd->udcd_flags |= UDCD_READ|UDCD_SILENT;
		} else if (rwcmd32.cmd == DADKIO_RWCMD_WRITE) {
			if ((un->un_dp->options & DMA_SUPPORTTED) ==
			    DMA_SUPPORTTED)
				scmd->udcd_cmd->cmd = ATA_WRITE_DMA;
			else
				scmd->udcd_cmd->cmd = ATA_WRITE;
			scmd->udcd_cmd->direction = DATA_WRITE;
			scmd->udcd_flags |= UDCD_WRITE|UDCD_SILENT;
		} else {
			mutex_exit(DCD_MUTEX);
			return (EINVAL);
		}

		scmd->udcd_cmd->address_mode = ADD_LBA_MODE;
		scmd->udcd_cmd->features = 0;
		scmd->udcd_cmd->size = rwcmd32.buflen;
		scmd->udcd_cmd->sector_num.lba_num = rwcmd32.blkaddr;
		scmd->udcd_bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
		scmd->udcd_buflen = rwcmd32.buflen;
		scmd->udcd_timeout = (ushort_t)dcd_io_time;
		scmd->udcd_resid = 0ULL;
		scmd->udcd_status = 0;
		scmd->udcd_error_reg = 0;
		scmd->udcd_status_reg = 0;

		mutex_exit(DCD_MUTEX);

		i = dcdioctl_cmd(dev, scmd, UIO_SYSSPACE, UIO_USERSPACE);
		mutex_enter(DCD_MUTEX);
		/*
		 * After return convert the status from scmd to
		 * dadkio_status
		 */
		(void) dcd_translate(&(rwcmd32.status), scmd);
		rwcmd32.status.resid = scmd->udcd_resid;
		mutex_exit(DCD_MUTEX);

#ifdef _MULTI_DATAMODEL
		switch (ddi_model_convert_from(flag & FMODELS)) {
		case DDI_MODEL_NONE: {
			int counter;
			rwcmd.status.status = rwcmd32.status.status;
			rwcmd.status.resid  = rwcmd32.status.resid;
			rwcmd.status.failed_blk_is_valid =
			    rwcmd32.status.failed_blk_is_valid;
			rwcmd.status.failed_blk = rwcmd32.status.failed_blk;
			rwcmd.status.fru_code_is_valid =
			    rwcmd32.status.fru_code_is_valid;
			rwcmd.status.fru_code = rwcmd32.status.fru_code;
			for (counter = 0;
			    counter < DADKIO_ERROR_INFO_LEN; counter++)
				rwcmd.status.add_error_info[counter] =
				    rwcmd32.status.add_error_info[counter];
			}
			/* Copy out the result back to the user program */
			if (ddi_copyout((caddr_t)&rwcmd, (caddr_t)arg,
			    sizeof (struct dadkio_rwcmd), flag)) {
				if (i != 0) {
					i = EFAULT;
				}
			}
			break;
		case DDI_MODEL_ILP32:
			/* Copy out the result back to the user program */
			if (ddi_copyout((caddr_t)&rwcmd32, (caddr_t)arg,
			    sizeof (struct dadkio_rwcmd32), flag)) {
				if (i != 0) {
					i = EFAULT;
				}
			}
			break;
		}
#else
		/* Copy out the result back to the user program  */
		if (ddi_copyout((caddr_t)&rwcmd32, (caddr_t)arg,
		    sizeof (struct dadkio_rwcmd32), flag)) {
			if (i != 0)
				i = EFAULT;
		}
#endif
		return (i);

	case UDCDCMD:	{
#ifdef	_MULTI_DATAMODEL
		/*
		 * For use when a 32 bit app makes a call into a
		 * 64 bit ioctl
		 */
		struct udcd_cmd32	udcd_cmd_32_for_64;
		struct udcd_cmd32	*ucmd32 = &udcd_cmd_32_for_64;
		model_t			model;
#endif /* _MULTI_DATAMODEL */

		if (drv_priv(cred_p) != 0) {
			return (EPERM);
		}

		scmd = (struct udcd_cmd *)data;

#ifdef _MULTI_DATAMODEL
		switch (model = ddi_model_convert_from(flag & FMODELS)) {
		case DDI_MODEL_ILP32:
			if (ddi_copyin((caddr_t)arg, ucmd32,
			    sizeof (struct udcd_cmd32), flag)) {
				return (EFAULT);
			}
			/*
			 * Convert the ILP32 uscsi data from the
			 * application to LP64 for internal use.
			 */
			udcd_cmd32toudcd_cmd(ucmd32, scmd);
			break;
		case DDI_MODEL_NONE:
			if (ddi_copyin((caddr_t)arg, scmd, sizeof (*scmd),
			    flag)) {
				return (EFAULT);
			}
			break;
		}
#else /* ! _MULTI_DATAMODEL */
		if (ddi_copyin((caddr_t)arg, (caddr_t)scmd,
		    sizeof (*scmd), flag)) {
			return (EFAULT);
		}
#endif /* ! _MULTI_DATAMODEL */

		scmd->udcd_flags &= ~UDCD_NOINTR;
		uioseg = (flag & FKIOCTL)? UIO_SYSSPACE: UIO_USERSPACE;

		i = dcdioctl_cmd(dev, scmd, uioseg, uioseg);
#ifdef _MULTI_DATAMODEL
		switch (model) {
		case DDI_MODEL_ILP32:
			/*
			 * Convert back to ILP32 before copyout to the
			 * application
			 */
			udcd_cmdtoudcd_cmd32(scmd, ucmd32);
			if (ddi_copyout(ucmd32, (caddr_t)arg,
			    sizeof (*ucmd32), flag)) {
				if (i != 0)
					i = EFAULT;
			}
			break;
		case DDI_MODEL_NONE:
			if (ddi_copyout(scmd, (caddr_t)arg, sizeof (*scmd),
			    flag)) {
				if (i != 0)
					i = EFAULT;
			}
			break;
		}
#else /* ! _MULTI_DATAMODE */
		if (ddi_copyout((caddr_t)scmd, (caddr_t)arg,
		    sizeof (*scmd), flag)) {
			if (i != 0)
				i = EFAULT;
		}
#endif
		return (i);
	}
	case DKIOCFLUSHWRITECACHE:	{
		struct dk_callback *dkc = (struct dk_callback *)arg;
		struct dcd_pkt *pkt;
		struct buf *bp;
		int is_sync = 1;

		mutex_enter(DCD_MUTEX);
		if (un->un_flush_not_supported ||
		    ! un->un_write_cache_enabled) {
			i = un->un_flush_not_supported ? ENOTSUP : 0;
			mutex_exit(DCD_MUTEX);
			/*
			 * If a callback was requested: a callback will
			 * always be done if the caller saw the
			 * DKIOCFLUSHWRITECACHE ioctl return 0, and
			 * never done if the caller saw the ioctl return
			 * an error.
			 */
			if ((flag & FKIOCTL) && dkc != NULL &&
			    dkc->dkc_callback != NULL) {
				(*dkc->dkc_callback)(dkc->dkc_cookie, i);
				/*
				 * Did callback and reported error.
				 * Since we did a callback, ioctl
				 * should return 0.
				 */
				i = 0;
			}
			return (i);
		}

		/*
		 * Get the special buffer
		 */
		while (un->un_sbuf_busy) {
			cv_wait(&un->un_sbuf_cv, DCD_MUTEX);
		}
		un->un_sbuf_busy = 1;
		bp  = un->un_sbufp;
		mutex_exit(DCD_MUTEX);

		pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
		    NULL, (uint32_t)sizeof (struct dcd_cmd),
		    2, PP_LEN, PKT_CONSISTENT, SLEEP_FUNC, (caddr_t)un);
		ASSERT(pkt != NULL);

		makecommand(pkt, un->un_cmd_flags | FLAG_SILENT,
		    ATA_FLUSH_CACHE, 0, ADD_LBA_MODE, 0, NO_DATA_XFER, 0);

		pkt->pkt_comp = dcdintr;
		pkt->pkt_time = DCD_FLUSH_TIME;
		PKT_SET_BP(pkt, bp);

		bp->av_back = (struct buf *)pkt;
		bp->b_forw = NULL;
		bp->b_flags = B_BUSY;
		bp->b_error = 0;
		bp->b_edev = dev;
		bp->b_dev = cmpdev(dev);
		bp->b_bcount = 0;
		bp->b_blkno = 0;
		bp->b_un.b_addr = 0;
		bp->b_iodone = NULL;
		bp->b_list = NULL;

		if ((flag & FKIOCTL) && dkc != NULL &&
		    dkc->dkc_callback != NULL) {
			struct dk_callback *dkc2 = (struct dk_callback *)
			    kmem_zalloc(sizeof (*dkc2), KM_SLEEP);
			bcopy(dkc, dkc2, sizeof (*dkc2));

			bp->b_list = (struct buf *)dkc2;
			bp->b_iodone = dcdflushdone;
			is_sync = 0;
		}

		(void) dcdstrategy(bp);

		i = 0;
		if (is_sync) {
			i = biowait(bp);
			(void) dcdflushdone(bp);
		}

		return (i);
	}
	default:
		break;
	}
	return (ENOTTY);
}


static int
dcdflushdone(struct buf *bp)
{
	struct dcd_disk *un = ddi_get_soft_state(dcd_state,
	    DCDUNIT(bp->b_edev));
	struct dcd_pkt *pkt = BP_PKT(bp);
	struct dk_callback *dkc = (struct dk_callback *)bp->b_list;

	ASSERT(un != NULL);
	ASSERT(bp == un->un_sbufp);
	ASSERT(pkt != NULL);

	dcd_destroy_pkt(pkt);
	bp->av_back = NO_PKT_ALLOCATED;

	if (dkc != NULL) {
		ASSERT(bp->b_iodone != NULL);
		(*dkc->dkc_callback)(dkc->dkc_cookie, geterror(bp));
		kmem_free(dkc, sizeof (*dkc));
		bp->b_iodone = NULL;
		bp->b_list = NULL;
	}

	/*
	 * Tell anybody who cares that the buffer is now free
	 */
	mutex_enter(DCD_MUTEX);
	un->un_sbuf_busy = 0;
	cv_signal(&un->un_sbuf_cv);
	mutex_exit(DCD_MUTEX);
	return (0);
}

/*
 * dcdrunout:
 *	the callback function for resource allocation
 *
 * XXX it would be preferable that dcdrunout() scans the whole
 *	list for possible candidates for dcdstart(); this avoids
 *	that a bp at the head of the list whose request cannot be
 *	satisfied is retried again and again
 */
/*ARGSUSED*/
static int
dcdrunout(caddr_t arg)
{
	int serviced;
	struct dcd_disk *un;
	struct diskhd *dp;

	TRACE_1(TR_FAC_DADA, TR_DCDRUNOUT_START, "dcdrunout_start: arg 0x%p",
	    arg);
	serviced = 1;

	un = (struct dcd_disk *)arg;
	dp = &un->un_utab;

	/*
	 * We now support passing a structure to the callback
	 * routine.
	 */
	ASSERT(un != NULL);
	mutex_enter(DCD_MUTEX);
	if ((un->un_ncmds < un->un_throttle) && (dp->b_forw == NULL)) {
		dcdstart(un);
	}
	if (un->un_state == DCD_STATE_RWAIT) {
		serviced = 0;
	}
	mutex_exit(DCD_MUTEX);
	TRACE_1(TR_FAC_DADA, TR_DCDRUNOUT_END,
	    "dcdrunout_end: serviced %d", serviced);
	return (serviced);
}


/*
 * This routine called to see whether unit is (still) there. Must not
 * be called when un->un_sbufp is in use, and must not be called with
 * an unattached disk. Soft state of disk is restored to what it was
 * upon entry- up to caller to set the correct state.
 *
 * We enter with the disk mutex held.
 */

/* ARGSUSED0 */
static int
dcd_unit_ready(dev_t dev)
{
#ifndef lint
	auto struct udcd_cmd dcmd, *com = &dcmd;
	auto struct dcd_cmd cmdblk;
#endif
	int error;
#ifndef lint
	GET_SOFT_STATE(dev);
#endif

	/*
	 * Now that we protect the special buffer with
	 * a mutex, we could probably do a mutex_tryenter
	 * on it here and return failure if it were held...
	 */

	error = 0;
	return (error);
}

/* ARGSUSED0 */
int
dcdioctl_cmd(dev_t devp, struct udcd_cmd *in, enum uio_seg cdbspace,
    enum uio_seg dataspace)
{

	struct buf *bp;
	struct	udcd_cmd *scmd;
	struct dcd_pkt *pkt;
	int	err, rw;
	caddr_t	cdb;
	int	flags = 0;

	GET_SOFT_STATE(devp);

#ifdef lint
	part = part;
#endif

	/*
	 * Is this a request to reset the bus?
	 * if so, we need to do reseting.
	 */

	if (in->udcd_flags & UDCD_RESET) {
		int flag = RESET_TARGET;
		err = dcd_reset(ROUTE, flag) ? 0: EIO;
		return (err);
	}

	scmd = in;


	/* Do some sanity checks */
	if (scmd->udcd_buflen <= 0) {
		if (scmd->udcd_flags & (UDCD_READ | UDCD_WRITE)) {
			return (EINVAL);
		} else {
			scmd->udcd_buflen = 0;
		}
	}

	/* Make a copy of the dcd_cmd passed  */
	cdb = kmem_zalloc(sizeof (struct dcd_cmd), KM_SLEEP);
	if (cdbspace == UIO_SYSSPACE) {
		flags |= FKIOCTL;
	}

	if (ddi_copyin((void *)scmd->udcd_cmd, cdb, sizeof (struct dcd_cmd),
	    flags)) {
		kmem_free(cdb, sizeof (struct dcd_cmd));
		return (EFAULT);
	}
	scmd = (struct udcd_cmd *)kmem_alloc(sizeof (*scmd), KM_SLEEP);
	bcopy((caddr_t)in, (caddr_t)scmd, sizeof (*scmd));
	scmd->udcd_cmd = (struct dcd_cmd *)cdb;
	rw = (scmd->udcd_flags & UDCD_READ) ? B_READ: B_WRITE;


	/*
	 * Get the special buffer
	 */

	mutex_enter(DCD_MUTEX);
	while (un->un_sbuf_busy) {
		if (cv_wait_sig(&un->un_sbuf_cv, DCD_MUTEX) == 0) {
			kmem_free(scmd->udcd_cmd, sizeof (struct dcd_cmd));
			kmem_free((caddr_t)scmd, sizeof (*scmd));
			mutex_exit(DCD_MUTEX);
			return (EINTR);
		}
	}

	un->un_sbuf_busy = 1;
	bp  = un->un_sbufp;
	mutex_exit(DCD_MUTEX);


	/*
	 * If we are going to do actual I/O, let physio do all the
	 * things
	 */
	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
	    "dcdioctl_cmd : buflen %x\n", scmd->udcd_buflen);

	if (scmd->udcd_buflen) {
		auto struct iovec aiov;
		auto struct uio auio;
		struct uio *uio = &auio;

		bzero((caddr_t)&auio, sizeof (struct uio));
		bzero((caddr_t)&aiov, sizeof (struct iovec));

		aiov.iov_base = scmd->udcd_bufaddr;
		aiov.iov_len = scmd->udcd_buflen;

		uio->uio_iov = &aiov;
		uio->uio_iovcnt = 1;
		uio->uio_resid = scmd->udcd_buflen;
		uio->uio_segflg = dataspace;

		/*
		 * Let physio do the rest...
		 */
		bp->av_back = NO_PKT_ALLOCATED;
		bp->b_forw = (struct buf *)scmd;
		err = physio(dcdstrategy, bp, devp, rw, dcdudcdmin, uio);
	} else {
		/*
		 * We have to mimic what physio would do here.
		 */
		bp->av_back = NO_PKT_ALLOCATED;
		bp->b_forw = (struct buf *)scmd;
		bp->b_flags = B_BUSY | rw;
		bp->b_edev = devp;
		bp->b_dev = cmpdev(devp);
		bp->b_bcount = bp->b_blkno = 0;
		(void) dcdstrategy(bp);
		err = biowait(bp);
	}

done:
	if ((pkt = BP_PKT(bp)) != NULL) {
		bp->av_back = NO_PKT_ALLOCATED;
		/* we need to update the completion status of udcd command */
		in->udcd_resid = bp->b_resid;
		in->udcd_status_reg = SCBP_C(pkt);
		/* XXX: we need to give error_reg also */
		dcd_destroy_pkt(pkt);
	}
	/*
	 * Tell anybody who cares that the buffer is now free
	 */
	mutex_enter(DCD_MUTEX);
	un->un_sbuf_busy = 0;
	cv_signal(&un->un_sbuf_cv);
	mutex_exit(DCD_MUTEX);

	kmem_free(scmd->udcd_cmd, sizeof (struct dcd_cmd));
	kmem_free((caddr_t)scmd, sizeof (*scmd));
	return (err);
}

static void
dcdudcdmin(struct buf *bp)
{

#ifdef lint
	bp = bp;
#endif

}

/*
 * restart a cmd from timeout() context
 *
 * the cmd is expected to be in un_utab.b_forw. If this pointer is non-zero
 * a restart timeout request has been issued and no new timeouts should
 * be requested. b_forw is reset when the cmd eventually completes in
 * dcddone_and_mutex_exit()
 */
void
dcdrestart(void *arg)
{
	struct dcd_disk *un = (struct dcd_disk *)arg;
	struct buf *bp;
	int status;

	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdrestart\n");

	mutex_enter(DCD_MUTEX);
	bp = un->un_utab.b_forw;
	if (bp) {
		un->un_ncmds++;
		DCD_DO_KSTATS(un, kstat_waitq_to_runq, bp);
	}


	if (bp) {
		struct dcd_pkt *pkt = BP_PKT(bp);

		mutex_exit(DCD_MUTEX);

		pkt->pkt_flags = 0;

		if ((status = dcd_transport(pkt)) != TRAN_ACCEPT) {
			mutex_enter(DCD_MUTEX);
			DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
			un->un_ncmds--;
			if (status == TRAN_BUSY) {
				/* XXX : To be checked */
				/*
				 * if (un->un_throttle > 1) {
				 *	ASSERT(un->un_ncmds >= 0);
				 *	un->un_throttle = un->un_ncmds;
				 * }
				 */
				un->un_reissued_timeid =
				    timeout(dcdrestart, (caddr_t)un,
				    DCD_BSY_TIMEOUT/500);
				mutex_exit(DCD_MUTEX);
				return;
			}
			DCD_DO_ERRSTATS(un, dcd_transerrs);
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "dcdrestart transport failed (%x)\n", status);
			bp->b_resid = bp->b_bcount;
			SET_BP_ERROR(bp, EIO);

			DCD_DO_KSTATS(un, kstat_waitq_exit, bp);
			un->un_reissued_timeid = 0L;
			dcddone_and_mutex_exit(un, bp);
			return;
		}
		mutex_enter(DCD_MUTEX);
	}
	un->un_reissued_timeid = 0L;
	mutex_exit(DCD_MUTEX);
	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdrestart done\n");
}

/*
 * This routine gets called to reset the throttle to its saved
 * value wheneven we lower the throttle.
 */
void
dcd_reset_throttle(caddr_t arg)
{
	struct dcd_disk *un = (struct dcd_disk *)arg;
	struct diskhd *dp;

	mutex_enter(DCD_MUTEX);
	dp = &un->un_utab;

	/*
	 * start any commands that didn't start while throttling.
	 */
	if (dp->b_actf && (un->un_ncmds < un->un_throttle) &&
	    (dp->b_forw == NULL)) {
		dcdstart(un);
	}
	mutex_exit(DCD_MUTEX);
}


/*
 * This routine handles the case when a TRAN_BUSY is
 * returned by HBA.
 *
 * If there are some commands already in the transport, the
 * bp can be put back on queue and it will
 * be retried when the queue is emptied after command
 * completes. But if there is no command in the tranport
 * and it still return busy, we have to retry the command
 * after some time like 10ms.
 */
/* ARGSUSED0 */
static void
dcd_handle_tran_busy(struct buf *bp, struct diskhd *dp, struct dcd_disk *un)
{
	ASSERT(mutex_owned(DCD_MUTEX));


	if (dp->b_forw == NULL || dp->b_forw == bp) {
		dp->b_forw = bp;
	} else if (dp->b_forw != bp) {
		bp->b_actf = dp->b_actf;
		dp->b_actf = bp;

	}
	if (!un->un_reissued_timeid) {
		un->un_reissued_timeid =
		    timeout(dcdrestart, (caddr_t)un, DCD_BSY_TIMEOUT/500);
	}
}

static int
dcd_write_deviceid(struct dcd_disk *un)
{

	int 	status;
	diskaddr_t blk;
	struct udcd_cmd ucmd;
	struct dcd_cmd cdb;
	struct dk_devid	*dkdevid;
	uint_t *ip, chksum;
	int	i;
	dev_t	dev;

	mutex_exit(DCD_MUTEX);
	if (cmlb_get_devid_block(un->un_dklbhandle, &blk, 0)) {
		mutex_enter(DCD_MUTEX);
		return (EINVAL);
	}
	mutex_enter(DCD_MUTEX);

	/* Allocate the buffer */
	dkdevid = kmem_zalloc(un->un_secsize, KM_SLEEP);

	/* Fill in the revision */
	dkdevid->dkd_rev_hi = DK_DEVID_REV_MSB;
	dkdevid->dkd_rev_lo = DK_DEVID_REV_LSB;

	/* Copy in the device id */
	bcopy(un->un_devid, &dkdevid->dkd_devid,
	    ddi_devid_sizeof(un->un_devid));

	/* Calculate the chksum */
	chksum = 0;
	ip = (uint_t *)dkdevid;
	for (i = 0; i < ((un->un_secsize - sizeof (int))/sizeof (int)); i++)
		chksum ^= ip[i];

	/* Fill in the checksum */
	DKD_FORMCHKSUM(chksum, dkdevid);

	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));
	(void) bzero((caddr_t)&cdb, sizeof (struct dcd_cmd));

	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
		cdb.cmd = ATA_WRITE_DMA;
	} else {
		if (un->un_dp->options & BLOCK_MODE)
			cdb.cmd = ATA_WRITE_MULTIPLE;
		else
			cdb.cmd = ATA_WRITE;
	}
	cdb.size = un->un_secsize;
	cdb.sector_num.lba_num = blk;
	cdb.address_mode = ADD_LBA_MODE;
	cdb.direction = DATA_WRITE;

	ucmd.udcd_flags = UDCD_WRITE;
	ucmd.udcd_cmd =  &cdb;
	ucmd.udcd_bufaddr = (caddr_t)dkdevid;
	ucmd.udcd_buflen = un->un_secsize;
	ucmd.udcd_flags |= UDCD_SILENT;
	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);
	mutex_exit(DCD_MUTEX);
	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);
	mutex_enter(DCD_MUTEX);

	kmem_free(dkdevid, un->un_secsize);
	return (status);
}

static int
dcd_read_deviceid(struct dcd_disk *un)
{
	int status;
	diskaddr_t blk;
	struct udcd_cmd ucmd;
	struct dcd_cmd cdb;
	struct dk_devid *dkdevid;
	uint_t *ip;
	int chksum;
	int i, sz;
	dev_t dev;

	mutex_exit(DCD_MUTEX);
	if (cmlb_get_devid_block(un->un_dklbhandle, &blk, 0)) {
		mutex_enter(DCD_MUTEX);
		return (EINVAL);
	}
	mutex_enter(DCD_MUTEX);

	dkdevid = kmem_alloc(un->un_secsize, KM_SLEEP);

	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));
	(void) bzero((caddr_t)&cdb, sizeof (cdb));

	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
		cdb.cmd = ATA_READ_DMA;
	} else {
		if (un->un_dp->options & BLOCK_MODE)
			cdb.cmd = ATA_READ_MULTIPLE;
		else
			cdb.cmd = ATA_READ;
	}
	cdb.size = un->un_secsize;
	cdb.sector_num.lba_num = blk;
	cdb.address_mode = ADD_LBA_MODE;
	cdb.direction = DATA_READ;

	ucmd.udcd_flags = UDCD_READ;
	ucmd.udcd_cmd =  &cdb;
	ucmd.udcd_bufaddr = (caddr_t)dkdevid;
	ucmd.udcd_buflen = un->un_secsize;
	ucmd.udcd_flags |= UDCD_SILENT;
	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);
	mutex_exit(DCD_MUTEX);
	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);
	mutex_enter(DCD_MUTEX);

	if (status != 0) {
		kmem_free((caddr_t)dkdevid, un->un_secsize);
		return (status);
	}

	/* Validate the revision */

	if ((dkdevid->dkd_rev_hi != DK_DEVID_REV_MSB) ||
	    (dkdevid->dkd_rev_lo != DK_DEVID_REV_LSB)) {
		kmem_free((caddr_t)dkdevid, un->un_secsize);
		return (EINVAL);
	}

	/* Calculate the checksum */
	chksum = 0;
	ip = (uint_t *)dkdevid;
	for (i = 0; i < ((un->un_secsize - sizeof (int))/sizeof (int)); i++)
		chksum ^= ip[i];

	/* Compare the checksums */

	if (DKD_GETCHKSUM(dkdevid) != chksum) {
		kmem_free((caddr_t)dkdevid, un->un_secsize);
		return (EINVAL);
	}

	/* VAlidate the device id */
	if (ddi_devid_valid((ddi_devid_t)&dkdevid->dkd_devid) != DDI_SUCCESS) {
		kmem_free((caddr_t)dkdevid, un->un_secsize);
		return (EINVAL);
	}

	/* return a copy of the device id */
	sz = ddi_devid_sizeof((ddi_devid_t)&dkdevid->dkd_devid);
	un->un_devid = (ddi_devid_t)kmem_alloc(sz, KM_SLEEP);
	bcopy(&dkdevid->dkd_devid, un->un_devid, sz);
	kmem_free((caddr_t)dkdevid, un->un_secsize);

	return (0);
}

/*
 * Return the device id for the device.
 * 1. If the device ID exists then just return it - nothing to do in that case.
 * 2. Build one from the drives model number and serial number.
 * 3. If there is a problem in building it from serial/model #, then try
 * to read it from the acyl region of the disk.
 * Note: If this function is unable to return a valid ID then the calling
 * point will invoke the routine to create a fabricated ID ans stor it on the
 * acyl region of the disk.
 */
static ddi_devid_t
dcd_get_devid(struct dcd_disk *un)
{
	int		rc;

	/* If already registered, return that value */
	if (un->un_devid != NULL)
		return (un->un_devid);

	/* Build a devid from model and serial number, if present */
	rc = dcd_make_devid_from_serial(un);

	if (rc != DDI_SUCCESS) {
		/* Read the devid from the disk. */
		if (dcd_read_deviceid(un))
			return (NULL);
	}

	(void) ddi_devid_register(DCD_DEVINFO, un->un_devid);
	return (un->un_devid);
}


static ddi_devid_t
dcd_create_devid(struct dcd_disk *un)
{
	if (ddi_devid_init(DCD_DEVINFO, DEVID_FAB, 0, NULL, (ddi_devid_t *)
	    &un->un_devid) == DDI_FAILURE)
		return (NULL);

	if (dcd_write_deviceid(un)) {
		ddi_devid_free(un->un_devid);
		un->un_devid = NULL;
		return (NULL);
	}

	(void) ddi_devid_register(DCD_DEVINFO, un->un_devid);
	return (un->un_devid);
}

/*
 * Build a devid from the model and serial number, if present
 * Return DDI_SUCCESS or DDI_FAILURE.
 */
static int
dcd_make_devid_from_serial(struct dcd_disk *un)
{
	int	rc = DDI_SUCCESS;
	char	*hwid;
	char	*model;
	int	model_len;
	char	*serno;
	int	serno_len;
	int	total_len;

	/* initialize the model and serial number information */
	model = un->un_dcd->dcd_ident->dcd_model;
	model_len = DCD_MODEL_NUMBER_LENGTH;
	serno = un->un_dcd->dcd_ident->dcd_drvser;
	serno_len = DCD_SERIAL_NUMBER_LENGTH;

	/* Verify the model and serial number */
	dcd_validate_model_serial(model, &model_len, model_len);
	if (model_len == 0) {
		rc = DDI_FAILURE;
		goto out;
	}
	dcd_validate_model_serial(serno, &serno_len, serno_len);
	if (serno_len == 0) {
		rc = DDI_FAILURE;
		goto out;
	}

	/*
	 * The device ID will be concatenation of the model number,
	 * the '=' separator, the serial number. Allocate
	 * the string and concatenate the components.
	 */
	total_len = model_len + 1 + serno_len;
	hwid = kmem_alloc(total_len, KM_SLEEP);
	bcopy((caddr_t)model, (caddr_t)hwid, model_len);
	bcopy((caddr_t)"=", (caddr_t)&hwid[model_len], 1);
	bcopy((caddr_t)serno, (caddr_t)&hwid[model_len + 1], serno_len);

	/* Initialize the device ID, trailing NULL not included */
	rc = ddi_devid_init(DCD_DEVINFO, DEVID_ATA_SERIAL, total_len,
	    hwid, (ddi_devid_t *)&un->un_devid);

	/* Free the allocated string */
	kmem_free(hwid, total_len);

out:	return (rc);
}

/*
 * Test for a valid model or serial number. Assume that a valid representation
 * contains at least one character that is neither a space, 0 digit, or NULL.
 * Trim trailing blanks and NULLS from returned length.
 */
static void
dcd_validate_model_serial(char *str, int *retlen, int totallen)
{
	char		ch;
	boolean_t	ret = B_FALSE;
	int		i;
	int		tb;

	for (i = 0, tb = 0; i < totallen; i++) {
		ch = *str++;
		if ((ch != ' ') && (ch != '\0') && (ch != '0'))
			ret = B_TRUE;
		if ((ch == ' ') || (ch == '\0'))
			tb++;
		else
			tb = 0;
	}

	if (ret == B_TRUE) {
		/* Atleast one non 0 or blank character. */
		*retlen = totallen - tb;
	} else {
		*retlen = 0;
	}
}

#ifndef lint
void
clean_print(dev_info_t *dev, char *label, uint_t level,
	char *title, char *data, int len)
{
	int	i;
	char	buf[256];

	(void) sprintf(buf, "%s:", title);
	for (i = 0; i < len; i++) {
		(void) sprintf(&buf[strlen(buf)], "0x%x ", (data[i] & 0xff));
	}
	(void) sprintf(&buf[strlen(buf)], "\n");

	dcd_log(dev, label, level, "%s", buf);
}
#endif /* Not lint */

#ifndef lint
/*
 * Print a piece of inquiry data- cleaned up for non-printable characters
 * and stopping at the first space character after the beginning of the
 * passed string;
 */

void
inq_fill(char *p, int l, char *s)
{
	unsigned i = 0;
	char c;

	while (i++ < l) {
		if ((c = *p++) < ' ' || c >= 0177) {
			c = '*';
		} else if (i != 1 && c == ' ') {
			break;
		}
		*s++ = c;
	}
	*s++ = 0;
}
#endif /* Not lint */

char *
dcd_sname(uchar_t status)
{
	switch (status & STATUS_ATA_MASK) {
	case STATUS_GOOD:
		return ("good status");

	case STATUS_ATA_BUSY:
		return ("busy");

	default:
		return ("<unknown status>");
	}
}

/* ARGSUSED0 */
char *
dcd_rname(int reason)
{
	static char *rnames[] = {
		"cmplt",
		"incomplete",
		"dma_derr",
		"tran_err",
		"reset",
		"aborted",
		"timeout",
		"data_ovr",
	};
	if (reason > CMD_DATA_OVR) {
		return ("<unknown reason>");
	} else {
		return (rnames[reason]);
	}
}



/* ARGSUSED0 */
int
dcd_check_wp(dev_t dev)
{

	return (0);
}

/*
 * Create device error kstats
 */
static int
dcd_create_errstats(struct dcd_disk *un, int instance)
{

	char kstatname[KSTAT_STRLEN];

	if (un->un_errstats == (kstat_t *)0) {
		(void) sprintf(kstatname, "dad%d,error", instance);
		un->un_errstats = kstat_create("daderror", instance, kstatname,
		    "device_error", KSTAT_TYPE_NAMED,
		    sizeof (struct dcd_errstats)/ sizeof (kstat_named_t),
		    KSTAT_FLAG_PERSISTENT);

		if (un->un_errstats) {
			struct dcd_errstats *dtp;

			dtp = (struct dcd_errstats *)un->un_errstats->ks_data;
			kstat_named_init(&dtp->dcd_softerrs, "Soft Errors",
			    KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_harderrs, "Hard Errors",
			    KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_transerrs,
			    "Transport Errors", KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_model, "Model",
			    KSTAT_DATA_CHAR);
			kstat_named_init(&dtp->dcd_revision, "Revision",
			    KSTAT_DATA_CHAR);
			kstat_named_init(&dtp->dcd_serial, "Serial No",
			    KSTAT_DATA_CHAR);
			kstat_named_init(&dtp->dcd_capacity, "Size",
			    KSTAT_DATA_ULONGLONG);
			kstat_named_init(&dtp->dcd_rq_media_err, "Media Error",
			    KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_rq_ntrdy_err,
			    "Device Not Ready", KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_rq_nodev_err, " No Device",
			    KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_rq_recov_err, "Recoverable",
			    KSTAT_DATA_UINT32);
			kstat_named_init(&dtp->dcd_rq_illrq_err,
			    "Illegal Request", KSTAT_DATA_UINT32);

			un->un_errstats->ks_private = un;
			un->un_errstats->ks_update = nulldev;
			kstat_install(un->un_errstats);

			(void) strncpy(&dtp->dcd_model.value.c[0],
			    un->un_dcd->dcd_ident->dcd_model, 16);
			(void) strncpy(&dtp->dcd_serial.value.c[0],
			    un->un_dcd->dcd_ident->dcd_drvser, 16);
			(void) strncpy(&dtp->dcd_revision.value.c[0],
			    un->un_dcd->dcd_ident->dcd_fw, 8);
			dtp->dcd_capacity.value.ui64 =
			    (uint64_t)((uint64_t)un->un_diskcapacity *
			    (uint64_t)un->un_lbasize);
		}
	}
	return (0);
}


/*
 * This has been moved from DADA layer as this does not do anything other than
 * retrying the command when it is busy or it does not complete
 */
int
dcd_poll(struct dcd_pkt *pkt)
{
	int	busy_count, rval = -1, savef;
	clock_t	savet;
	void	(*savec)();


	/*
	 * Save old flags
	 */
	savef = pkt->pkt_flags;
	savec = pkt->pkt_comp;
	savet = pkt->pkt_time;

	pkt->pkt_flags |= FLAG_NOINTR;


	/*
	 * Set the Pkt_comp to NULL
	 */

	pkt->pkt_comp = 0;

	/*
	 * Set the Pkt time for the polled command
	 */
	if (pkt->pkt_time == 0) {
		pkt->pkt_time = DCD_POLL_TIMEOUT;
	}


	/* Now transport the command */
	for (busy_count = 0; busy_count < dcd_poll_busycnt; busy_count++) {
		if ((rval = dcd_transport(pkt)) == TRAN_ACCEPT) {
			if (pkt->pkt_reason == CMD_INCOMPLETE &&
			    pkt->pkt_state == 0) {
				delay(100);
			} else if (pkt->pkt_reason  == CMD_CMPLT) {
				rval = 0;
				break;
			}
		}
		if (rval == TRAN_BUSY)  {
			delay(100);
			continue;
		}
	}

	pkt->pkt_flags = savef;
	pkt->pkt_comp = savec;
	pkt->pkt_time = savet;
	return (rval);
}


void
dcd_translate(struct dadkio_status32 *statp, struct udcd_cmd *cmdp)
{
	if (cmdp->udcd_status_reg & STATUS_ATA_BUSY)
		statp->status = DADKIO_STAT_NOT_READY;
	else if (cmdp->udcd_status_reg & STATUS_ATA_DWF)
		statp->status = DADKIO_STAT_HARDWARE_ERROR;
	else if (cmdp->udcd_status_reg & STATUS_ATA_CORR)
		statp->status = DADKIO_STAT_SOFT_ERROR;
	else if (cmdp->udcd_status_reg & STATUS_ATA_ERR) {
		/*
		 * The error register is valid only when BSY and DRQ not set
		 * Assumed that HBA has checked this before it gives the data
		 */
		if (cmdp->udcd_error_reg & ERR_AMNF)
			statp->status = DADKIO_STAT_NOT_FORMATTED;
		else if (cmdp->udcd_error_reg & ERR_TKONF)
			statp->status = DADKIO_STAT_NOT_FORMATTED;
		else if (cmdp->udcd_error_reg & ERR_ABORT)
			statp->status = DADKIO_STAT_ILLEGAL_REQUEST;
		else if (cmdp->udcd_error_reg & ERR_IDNF)
			statp->status = DADKIO_STAT_NOT_FORMATTED;
		else if (cmdp->udcd_error_reg & ERR_UNC)
			statp->status = DADKIO_STAT_BUS_ERROR;
		else if (cmdp->udcd_error_reg & ERR_BBK)
			statp->status = DADKIO_STAT_MEDIUM_ERROR;
	} else
		statp->status = DADKIO_STAT_NO_ERROR;
}

static void
dcd_flush_cache(struct dcd_disk *un)
{
	struct dcd_pkt *pkt;
	int retry_count;


	if ((pkt = dcd_init_pkt(ROUTE, NULL, NULL,
	    (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
	    PKT_CONSISTENT, NULL_FUNC, NULL)) == NULL) {
		return;
	}

	makecommand(pkt, 0, ATA_FLUSH_CACHE, 0, ADD_LBA_MODE, 0,
	    NO_DATA_XFER, 0);

	/*
	 * Send the command. There are chances it might fail on some
	 * disks since it is not a mandatory command as per ata-4. Try
	 * 3 times if it fails. The retry count has been randomly selected.
	 * There is a need for retry since as per the spec FLUSH CACHE can fail
	 * as a result of unrecoverable error encountered during execution
	 * of writing data and subsequent command should continue flushing
	 * cache.
	 */
	for (retry_count = 0; retry_count < 3; retry_count++) {
		/*
		 * Set the packet fields.
		 */
		pkt->pkt_comp = 0;
		pkt->pkt_time = DCD_POLL_TIMEOUT;
		pkt->pkt_flags |= FLAG_FORCENOINTR;
		pkt->pkt_flags |= FLAG_NOINTR;
		if (dcd_transport(pkt) == TRAN_ACCEPT) {
			if (pkt->pkt_reason  == CMD_CMPLT) {
				break;
			}
		}
		/*
		 * Note the wait time value of 100ms is same as in the
		 * dcd_poll routine.
		 */
		drv_usecwait(1000000);
	}
	(void) dcd_destroy_pkt(pkt);
}

static int
dcd_send_lb_rw_cmd(dev_info_t *devi, void *bufaddr,
    diskaddr_t start_block, size_t reqlength, uchar_t cmd)
{
	struct dcd_pkt *pkt;
	struct buf *bp;
	diskaddr_t real_addr = start_block;
	size_t buffer_size = reqlength;
	uchar_t command, tmp;
	int i, rval = 0;
	struct dcd_disk *un;

	un = ddi_get_soft_state(dcd_state, ddi_get_instance(devi));
	if (un == NULL)
		return (ENXIO);

	bp = dcd_alloc_consistent_buf(ROUTE, (struct buf *)NULL,
	    buffer_size, B_READ, NULL_FUNC, NULL);
	if (!bp) {
		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
		    "no bp for disk label\n");
		return (ENOMEM);
	}

	pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
	    bp, (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
	    PKT_CONSISTENT, NULL_FUNC, NULL);

	if (!pkt) {
		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
		    "no memory for disk label\n");
		dcd_free_consistent_buf(bp);
		return (ENOMEM);
	}

	if (cmd == TG_READ) {
		bzero(bp->b_un.b_addr, buffer_size);
		tmp = DATA_READ;
	} else {
		bcopy((caddr_t)bufaddr, bp->b_un.b_addr, buffer_size);
		tmp = DATA_WRITE;
	}

	mutex_enter(DCD_MUTEX);
	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
		if (cmd == TG_READ) {
			command = ATA_READ_DMA;
		} else {
			command = ATA_WRITE_DMA;
		}
	} else {
		if (cmd == TG_READ) {
			if (un->un_dp->options & BLOCK_MODE)
				command = ATA_READ_MULTIPLE;
			else
				command = ATA_READ;
		} else {
			if (un->un_dp->options & BLOCK_MODE)
				command = ATA_READ_MULTIPLE;
			else
				command = ATA_WRITE;
		}
	}
	mutex_exit(DCD_MUTEX);
	(void) makecommand(pkt, 0, command, real_addr, ADD_LBA_MODE,
	    buffer_size, tmp, 0);

	for (i = 0; i < 3; i++) {
		if (dcd_poll(pkt) || SCBP_C(pkt) != STATUS_GOOD ||
		    (pkt->pkt_state & STATE_XFERRED_DATA) == 0 ||
		    (pkt->pkt_resid != 0)) {
			DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
			    "Status %x, state %x, resid %lx\n",
			    SCBP_C(pkt), pkt->pkt_state, pkt->pkt_resid);
			rval = EIO;
		} else {
			break;
		}
	}

	if (rval != 0) {
		dcd_destroy_pkt(pkt);
		dcd_free_consistent_buf(bp);
		return (EIO);
	}

	if (cmd == TG_READ) {
		bcopy(bp->b_un.b_addr, bufaddr, reqlength);
		rval = 0;
	}

	dcd_destroy_pkt(pkt);
	dcd_free_consistent_buf(bp);
	return (rval);
}

static int dcd_compute_dk_capacity(struct dcd_device *devp,
    diskaddr_t *capacity)
{
	diskaddr_t cap;
	diskaddr_t no_of_lbasec;

	cap = devp->dcd_ident->dcd_fixcyls *
	    devp->dcd_ident->dcd_heads *
	    devp->dcd_ident->dcd_sectors;
	no_of_lbasec = devp->dcd_ident->dcd_addrsec[1];
	no_of_lbasec = no_of_lbasec << 16;
	no_of_lbasec = no_of_lbasec | devp->dcd_ident->dcd_addrsec[0];

	if (no_of_lbasec > cap) {
		cap = no_of_lbasec;
	}

	if (cap != ((uint32_t)-1))
		*capacity = cap;
	else
		return (EINVAL);
	return (0);
}

/*ARGSUSED5*/
static int
dcd_lb_rdwr(dev_info_t *devi, uchar_t cmd, void *bufaddr,
    diskaddr_t start_block, size_t reqlength, void *tg_cookie)
{
	if (cmd != TG_READ && cmd != TG_WRITE)
		return (EINVAL);

	return (dcd_send_lb_rw_cmd(devi, bufaddr, start_block,
	    reqlength, cmd));
}

static int
dcd_lb_getphygeom(dev_info_t *devi, cmlb_geom_t *phygeomp)
{
	struct dcd_device *devp;
	uint32_t no_of_lbasec, capacity, calculated_cylinders;

	devp = ddi_get_driver_private(devi);

	if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
		if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
			phygeomp->g_ncyl = devp->dcd_ident->dcd_fixcyls - 2;
			phygeomp->g_acyl = 2;
			phygeomp->g_nhead = devp->dcd_ident->dcd_heads;
			phygeomp->g_nsect = devp->dcd_ident->dcd_sectors;

			no_of_lbasec = devp->dcd_ident->dcd_addrsec[1];
			no_of_lbasec = no_of_lbasec << 16;
			no_of_lbasec = no_of_lbasec |
			    devp->dcd_ident->dcd_addrsec[0];
			capacity = devp->dcd_ident->dcd_fixcyls *
			    devp->dcd_ident->dcd_heads *
			    devp->dcd_ident->dcd_sectors;
			if (no_of_lbasec > capacity) {
				capacity = no_of_lbasec;
				if (capacity > NUM_SECTORS_32G) {
					/*
					 * if the capacity is greater than 32G,
					 * then 255 is the sectors per track.
					 * This should be good until 128G disk
					 * capacity, which is the current ATA-4
					 * limitation.
					 */
					phygeomp->g_nsect = 255;
				}

				/*
				 * If the disk capacity is >= 128GB then no. of
				 * addressable sectors will be set to 0xfffffff
				 * in the IDENTIFY info. In that case set the
				 *  no. of pcyl to the Max. 16bit value.
				 */

				calculated_cylinders = (capacity) /
				    (phygeomp->g_nhead * phygeomp->g_nsect);
				if (calculated_cylinders >= USHRT_MAX) {
					phygeomp->g_ncyl = USHRT_MAX - 2;
				} else {
					phygeomp->g_ncyl =
					    calculated_cylinders - 2;
				}
			}

			phygeomp->g_capacity = capacity;
			phygeomp->g_intrlv = 0;
			phygeomp->g_rpm = 5400;
			phygeomp->g_secsize = devp->dcd_ident->dcd_secsiz;

			return (0);
		} else
			return (ENOTSUP);
	} else {
		return (EINVAL);
	}
}


/*ARGSUSED3*/
static int
dcd_lb_getinfo(dev_info_t *devi, int cmd,  void *arg, void *tg_cookie)
{
	struct dcd_disk *un;

	un = ddi_get_soft_state(dcd_state, ddi_get_instance(devi));

	if (un == NULL)
		return (ENXIO);

	switch (cmd) {
	case TG_GETPHYGEOM:
		return (dcd_lb_getphygeom(devi, (cmlb_geom_t *)arg));

	case TG_GETVIRTGEOM:
		return (-1);

	case TG_GETCAPACITY:
	case TG_GETBLOCKSIZE:
		mutex_enter(DCD_MUTEX);
		if (un->un_diskcapacity <= 0) {
			mutex_exit(DCD_MUTEX);
			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
			    "invalid disk capacity\n");
			return (EIO);
		}
		if (cmd == TG_GETCAPACITY)
			*(diskaddr_t *)arg = un->un_diskcapacity;
		else
			*(uint32_t *)arg = DEV_BSIZE;

		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "capacity %x\n",
		    un->un_diskcapacity);
		mutex_exit(DCD_MUTEX);
		return (0);

	case TG_GETATTR:
		mutex_enter(DCD_MUTEX);
		*(tg_attribute_t *)arg = un->un_tgattribute;
		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
		    "media_is_writable %x\n",
		    un->un_tgattribute.media_is_writable);
		mutex_exit(DCD_MUTEX);
		return (0);
	default:
		return (ENOTTY);
	}
}