xref: /titanic_50/usr/src/uts/sun/io/dada/targets/dad.c (revision 9a5d73e03cd3312ddb571a748c40a63c58bd66e5)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 
28 /*
29  * Direct Attached  disk driver for SPARC machines.
30  */
31 
32 /*
33  * Includes, Declarations and Local Data
34  */
35 #include <sys/dada/dada.h>
36 #include <sys/dkbad.h>
37 #include <sys/dklabel.h>
38 #include <sys/dkio.h>
39 #include <sys/cdio.h>
40 #include <sys/vtoc.h>
41 #include <sys/dada/targets/daddef.h>
42 #include <sys/dada/targets/dadpriv.h>
43 #include <sys/file.h>
44 #include <sys/stat.h>
45 #include <sys/kstat.h>
46 #include <sys/vtrace.h>
47 #include <sys/aio_req.h>
48 #include <sys/note.h>
49 #include <sys/cmlb.h>
50 
51 /*
52  * Global Error Levels for Error Reporting
53  */
54 int dcd_error_level	= DCD_ERR_RETRYABLE;
55 /*
56  * Local Static Data
57  */
58 
59 static int dcd_io_time		= DCD_IO_TIME;
60 static int dcd_retry_count	= DCD_RETRY_COUNT;
61 #ifndef lint
62 static int dcd_report_pfa = 1;
63 #endif
64 static int dcd_rot_delay = 4;
65 static int dcd_poll_busycnt = DCD_POLL_TIMEOUT;
66 
67 /*
68  * Local Function Prototypes
69  */
70 
71 static int dcdopen(dev_t *dev_p, int flag, int otyp, cred_t *cred_p);
72 static int dcdclose(dev_t dev, int flag, int otyp, cred_t *cred_p);
73 static int dcdstrategy(struct buf *bp);
74 static int dcddump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
75 static int dcdioctl(dev_t, int, intptr_t, int, cred_t *, int *);
76 static int dcdread(dev_t dev, struct uio *uio, cred_t *cred_p);
77 static int dcdwrite(dev_t dev, struct uio *uio, cred_t *cred_p);
78 static int dcd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int,
79     char *, caddr_t, int *);
80 static int dcdaread(dev_t dev, struct aio_req *aio, cred_t *cred_p);
81 static int dcdawrite(dev_t dev, struct aio_req *aio, cred_t *cred_p);
82 
83 
84 static void dcd_free_softstate(struct dcd_disk *un, dev_info_t *devi);
85 static int dcd_doattach(dev_info_t *devi, int (*f)());
86 static int dcd_validate_geometry(struct dcd_disk *un);
87 static ddi_devid_t dcd_get_devid(struct dcd_disk *un);
88 static ddi_devid_t  dcd_create_devid(struct dcd_disk *un);
89 static int dcd_make_devid_from_serial(struct dcd_disk *un);
90 static void dcd_validate_model_serial(char *str, int *retlen, int totallen);
91 static int dcd_read_deviceid(struct dcd_disk *un);
92 static int dcd_write_deviceid(struct dcd_disk *un);
93 static int dcd_poll(struct dcd_pkt *pkt);
94 static char *dcd_rname(int reason);
95 static void dcd_flush_cache(struct dcd_disk *un);
96 
97 static int dcd_compute_dk_capacity(struct dcd_device *devp,
98     diskaddr_t *capacity);
99 static int dcd_send_lb_rw_cmd(dev_info_t *devinfo, void *bufaddr,
100     diskaddr_t start_block, size_t reqlength, uchar_t cmd);
101 
102 static void dcdmin(struct buf *bp);
103 
104 static int dcdioctl_cmd(dev_t, struct udcd_cmd *,
105     enum uio_seg, enum uio_seg);
106 
107 static void dcdstart(struct dcd_disk *un);
108 static void dcddone_and_mutex_exit(struct dcd_disk *un, struct buf *bp);
109 static void make_dcd_cmd(struct dcd_disk *un, struct buf *bp, int (*f)());
110 static void dcdudcdmin(struct buf *bp);
111 
112 static int dcdrunout(caddr_t);
113 static int dcd_check_wp(dev_t dev);
114 static int dcd_unit_ready(dev_t dev);
115 static void dcd_handle_tran_busy(struct buf *bp, struct diskhd *dp,
116     struct dcd_disk *un);
117 static void dcdintr(struct dcd_pkt *pkt);
118 static int dcd_handle_incomplete(struct dcd_disk *un, struct buf *bp);
119 static void dcd_offline(struct dcd_disk *un, int bechatty);
120 static int dcd_ready_and_valid(dev_t dev, struct dcd_disk *un);
121 static void dcd_reset_disk(struct dcd_disk *un, struct dcd_pkt *pkt);
122 static void dcd_translate(struct dadkio_status32 *statp, struct udcd_cmd *cmdp);
123 static int dcdflushdone(struct buf *bp);
124 
125 /* Function prototypes for cmlb */
126 
127 static int dcd_lb_rdwr(dev_info_t *devi, uchar_t cmd, void *bufaddr,
128     diskaddr_t start_block, size_t reqlength, void *tg_cookie);
129 
130 static int dcd_lb_getphygeom(dev_info_t *devi, cmlb_geom_t *phygeomp);
131 static int dcd_lb_getinfo(dev_info_t *devi, int cmd, void *arg,
132     void *tg_cookie);
133 
134 
135 static cmlb_tg_ops_t dcd_lb_ops = {
136 	TG_DK_OPS_VERSION_1,
137 	dcd_lb_rdwr,
138 	dcd_lb_getinfo
139 };
140 
141 /*
142  * Error and Logging Functions
143  */
144 #ifndef lint
145 static void clean_print(dev_info_t *dev, char *label, uint_t level,
146     char *title, char *data, int len);
147 static void dcdrestart(void *arg);
148 #endif /* lint */
149 
150 static int dcd_check_error(struct dcd_disk *un, struct buf *bp);
151 
152 /*
153  * Error statistics create/update functions
154  */
155 static int dcd_create_errstats(struct dcd_disk *, int);
156 
157 
158 
159 /*PRINTFLIKE4*/
160 extern void dcd_log(dev_info_t *, char *, uint_t, const char *, ...)
161     __KPRINTFLIKE(4);
162 extern void makecommand(struct dcd_pkt *, int, uchar_t, uint32_t,
163     uchar_t, uint32_t, uchar_t, uchar_t);
164 
165 
166 /*
167  * Configuration Routines
168  */
169 static int dcdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,
170     void **result);
171 static int dcdprobe(dev_info_t *devi);
172 static int dcdattach(dev_info_t *devi, ddi_attach_cmd_t cmd);
173 static int dcddetach(dev_info_t *devi, ddi_detach_cmd_t cmd);
174 static int dcdreset(dev_info_t *dip, ddi_reset_cmd_t cmd);
175 static int dcd_dr_detach(dev_info_t *devi);
176 static int dcdpower(dev_info_t *devi, int component, int level);
177 
178 static void *dcd_state;
179 static int dcd_max_instance;
180 static char *dcd_label = "dad";
181 
182 static char *diskokay = "disk okay\n";
183 
184 #if DEBUG || lint
185 #define	DCDDEBUG
186 #endif
187 
188 int dcd_test_flag = 0;
189 /*
190  * Debugging macros
191  */
192 #ifdef	DCDDEBUG
193 static int dcddebug = 0;
194 #define	DEBUGGING	(dcddebug > 1)
195 #define	DAD_DEBUG	if (dcddebug == 1) dcd_log
196 #define	DAD_DEBUG2	if (dcddebug > 1) dcd_log
197 #else	/* DCDDEBUG */
198 #define	dcddebug		(0)
199 #define	DEBUGGING	(0)
200 #define	DAD_DEBUG	if (0) dcd_log
201 #define	DAD_DEBUG2	if (0) dcd_log
202 #endif
203 
204 /*
205  * we use pkt_private area for storing bp and retry_count
206  * XXX: Really is this usefull.
207  */
208 struct dcd_pkt_private {
209 	struct buf	*dcdpp_bp;
210 	short		 dcdpp_retry_count;
211 	short		 dcdpp_victim_retry_count;
212 };
213 
214 
215 _NOTE(SCHEME_PROTECTS_DATA("Unique per pkt", dcd_pkt_private buf))
216 
217 #define	PP_LEN	(sizeof (struct dcd_pkt_private))
218 
219 #define	PKT_SET_BP(pkt, bp)	\
220 	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_bp = bp
221 #define	PKT_GET_BP(pkt) \
222 	(((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_bp)
223 
224 
225 #define	PKT_SET_RETRY_CNT(pkt, n) \
226 	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_retry_count = n
227 
228 #define	PKT_GET_RETRY_CNT(pkt) \
229 	(((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_retry_count)
230 
231 #define	PKT_INCR_RETRY_CNT(pkt, n) \
232 	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_retry_count += n
233 
234 #define	PKT_SET_VICTIM_RETRY_CNT(pkt, n) \
235 	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_victim_retry_count \
236 			= n
237 
238 #define	PKT_GET_VICTIM_RETRY_CNT(pkt) \
239 	(((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_victim_retry_count)
240 #define	PKT_INCR_VICTIM_RETRY_CNT(pkt, n) \
241 	((struct dcd_pkt_private *)pkt->pkt_private)->dcdpp_victim_retry_count \
242 			+= n
243 
244 #define	DISK_NOT_READY_RETRY_COUNT	(dcd_retry_count / 2)
245 
246 
247 /*
248  * Urk!
249  */
250 #define	SET_BP_ERROR(bp, err)	\
251 	bioerror(bp, err);
252 
253 #define	IOSP			KSTAT_IO_PTR(un->un_stats)
254 #define	IO_PARTITION_STATS	un->un_pstats[DCDPART(bp->b_edev)]
255 #define	IOSP_PARTITION		KSTAT_IO_PTR(IO_PARTITION_STATS)
256 
257 #define	DCD_DO_KSTATS(un, kstat_function, bp) \
258 	ASSERT(mutex_owned(DCD_MUTEX)); \
259 	if (bp != un->un_sbufp) { \
260 		if (un->un_stats) { \
261 			kstat_function(IOSP); \
262 		} \
263 		if (IO_PARTITION_STATS) { \
264 			kstat_function(IOSP_PARTITION); \
265 		} \
266 	}
267 
268 #define	DCD_DO_ERRSTATS(un, x) \
269 	if (un->un_errstats) { \
270 		struct dcd_errstats *dtp; \
271 		dtp = (struct dcd_errstats *)un->un_errstats->ks_data; \
272 		dtp->x.value.ui32++; \
273 	}
274 
275 #define	GET_SOFT_STATE(dev)						\
276 	struct dcd_disk *un;					\
277 	int instance, part;					\
278 	minor_t minor = getminor(dev);				\
279 									\
280 	part = minor & DCDPART_MASK;					\
281 	instance = minor >> DCDUNIT_SHIFT;				\
282 	if ((un = ddi_get_soft_state(dcd_state, instance)) == NULL)	\
283 		return (ENXIO);
284 
285 #define	LOGICAL_BLOCK_ALIGN(blkno, blknoshift) \
286 		(((blkno) & ((1 << (blknoshift)) - 1)) == 0)
287 
288 /*
289  * After the following number of sectors, the cylinder number spills over
290  * 0xFFFF if sectors = 63 and heads = 16.
291  */
292 #define	NUM_SECTORS_32G	0x3EFFC10
293 
294 /*
295  * Configuration Data
296  */
297 
298 /*
299  * Device driver ops vector
300  */
301 
302 static struct cb_ops dcd_cb_ops = {
303 	dcdopen,		/* open */
304 	dcdclose,		/* close */
305 	dcdstrategy,		/* strategy */
306 	nodev,			/* print */
307 	dcddump,		/* dump */
308 	dcdread,		/* read */
309 	dcdwrite,		/* write */
310 	dcdioctl,		/* ioctl */
311 	nodev,			/* devmap */
312 	nodev,			/* mmap */
313 	nodev,			/* segmap */
314 	nochpoll,		/* poll */
315 	dcd_prop_op,		/* cb_prop_op */
316 	0,			/* streamtab  */
317 	D_64BIT | D_MP | D_NEW,	/* Driver compatibility flag */
318 	CB_REV,			/* cb_rev */
319 	dcdaread, 		/* async I/O read entry point */
320 	dcdawrite		/* async I/O write entry point */
321 };
322 
323 static struct dev_ops dcd_ops = {
324 	DEVO_REV,		/* devo_rev, */
325 	0,			/* refcnt  */
326 	dcdinfo,		/* info */
327 	nulldev,		/* identify */
328 	dcdprobe,		/* probe */
329 	dcdattach,		/* attach */
330 	dcddetach,		/* detach */
331 	dcdreset,		/* reset */
332 	&dcd_cb_ops,		/* driver operations */
333 	(struct bus_ops *)0,	/* bus operations */
334 	dcdpower,		/* power */
335 	ddi_quiesce_not_supported,	/* devo_quiesce */
336 };
337 
338 
339 /*
340  * This is the loadable module wrapper.
341  */
342 #include <sys/modctl.h>
343 
344 static struct modldrv modldrv = {
345 	&mod_driverops,		/* Type of module. This one is a driver */
346 	"DAD Disk Driver",	/* Name of the module. */
347 	&dcd_ops,	/* driver ops */
348 };
349 
350 
351 
352 static struct modlinkage modlinkage = {
353 	MODREV_1, &modldrv, NULL
354 };
355 
356 /*
357  * the dcd_attach_mutex only protects dcd_max_instance in multi-threaded
358  * attach situations
359  */
360 static kmutex_t dcd_attach_mutex;
361 
362 int
363 _init(void)
364 {
365 	int e;
366 
367 	if ((e = ddi_soft_state_init(&dcd_state, sizeof (struct dcd_disk),
368 	    DCD_MAXUNIT)) != 0)
369 		return (e);
370 
371 	mutex_init(&dcd_attach_mutex, NULL, MUTEX_DRIVER, NULL);
372 	e = mod_install(&modlinkage);
373 	if (e != 0) {
374 		mutex_destroy(&dcd_attach_mutex);
375 		ddi_soft_state_fini(&dcd_state);
376 		return (e);
377 	}
378 
379 	return (e);
380 }
381 
382 int
383 _fini(void)
384 {
385 	int e;
386 
387 	if ((e = mod_remove(&modlinkage)) != 0)
388 		return (e);
389 
390 	ddi_soft_state_fini(&dcd_state);
391 	mutex_destroy(&dcd_attach_mutex);
392 
393 	return (e);
394 }
395 
396 int
397 _info(struct modinfo *modinfop)
398 {
399 
400 	return (mod_info(&modlinkage, modinfop));
401 }
402 
403 static int
404 dcdprobe(dev_info_t *devi)
405 {
406 	struct dcd_device *devp;
407 	int rval = DDI_PROBE_PARTIAL;
408 	int instance;
409 
410 	devp = ddi_get_driver_private(devi);
411 	instance = ddi_get_instance(devi);
412 
413 	/*
414 	 * Keep a count of how many disks (ie. highest instance no) we have
415 	 * XXX currently not used but maybe useful later again
416 	 */
417 	mutex_enter(&dcd_attach_mutex);
418 	if (instance > dcd_max_instance)
419 		dcd_max_instance = instance;
420 	mutex_exit(&dcd_attach_mutex);
421 
422 	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG, "dcdprobe:\n");
423 
424 	if (ddi_get_soft_state(dcd_state, instance) != NULL)
425 		return (DDI_PROBE_PARTIAL);
426 
427 	/*
428 	 * Turn around and call utility probe routine
429 	 * to see whether we actually have a disk at
430 	 */
431 
432 	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG,
433 	    "dcdprobe: %x\n", dcd_probe(devp, NULL_FUNC));
434 
435 	switch (dcd_probe(devp, NULL_FUNC)) {
436 	default:
437 	case DCDPROBE_NORESP:
438 	case DCDPROBE_NONCCS:
439 	case DCDPROBE_NOMEM:
440 	case DCDPROBE_FAILURE:
441 	case DCDPROBE_BUSY:
442 		break;
443 
444 	case DCDPROBE_EXISTS:
445 		/*
446 		 * Check whether it is a ATA device and then
447 		 * return  SUCCESS.
448 		 */
449 		DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG,
450 		    "config %x\n", devp->dcd_ident->dcd_config);
451 		if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
452 			if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
453 				rval = DDI_PROBE_SUCCESS;
454 			} else
455 				rval = DDI_PROBE_FAILURE;
456 		} else {
457 			rval = DDI_PROBE_FAILURE;
458 		}
459 		break;
460 	}
461 	dcd_unprobe(devp);
462 
463 	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG,
464 	    "dcdprobe returns %x\n", rval);
465 
466 	return (rval);
467 }
468 
469 
470 /*ARGSUSED*/
471 static int
472 dcdattach(dev_info_t *devi, ddi_attach_cmd_t cmd)
473 {
474 	int instance, rval;
475 	struct dcd_device *devp;
476 	struct dcd_disk *un;
477 	struct diskhd *dp;
478 	char	*pm_comp[] =
479 	    { "NAME=ide-disk", "0=standby", "1=idle", "2=active" };
480 
481 	/* CONSTCOND */
482 	ASSERT(NO_COMPETING_THREADS);
483 
484 
485 	devp = ddi_get_driver_private(devi);
486 	instance = ddi_get_instance(devi);
487 	DAD_DEBUG2(devp->dcd_dev, dcd_label, DCD_DEBUG, "Attach Started\n");
488 
489 	switch (cmd) {
490 	case DDI_ATTACH:
491 		break;
492 
493 	case DDI_RESUME:
494 		if (!(un = ddi_get_soft_state(dcd_state, instance)))
495 			return (DDI_FAILURE);
496 		mutex_enter(DCD_MUTEX);
497 		Restore_state(un);
498 		/*
499 		 * Restore the state which was saved to give the
500 		 * the right state in un_last_state
501 		 */
502 		un->un_last_state = un->un_save_state;
503 		un->un_throttle = 2;
504 		cv_broadcast(&un->un_suspend_cv);
505 		/*
506 		 * Raise the power level of the device to active.
507 		 */
508 		mutex_exit(DCD_MUTEX);
509 		(void) pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE);
510 		mutex_enter(DCD_MUTEX);
511 
512 		/*
513 		 * start unit - if this is a low-activity device
514 		 * commands in queue will have to wait until new
515 		 * commands come in, which may take awhile.
516 		 * Also, we specifically don't check un_ncmds
517 		 * because we know that there really are no
518 		 * commands in progress after the unit was suspended
519 		 * and we could have reached the throttle level, been
520 		 * suspended, and have no new commands coming in for
521 		 * awhile.  Highly unlikely, but so is the low-
522 		 * activity disk scenario.
523 		 */
524 		dp = &un->un_utab;
525 		if (dp->b_actf && (dp->b_forw == NULL)) {
526 			dcdstart(un);
527 		}
528 
529 		mutex_exit(DCD_MUTEX);
530 		return (DDI_SUCCESS);
531 
532 	default:
533 		return (DDI_FAILURE);
534 	}
535 
536 	if (dcd_doattach(devi, SLEEP_FUNC) == DDI_FAILURE) {
537 		return (DDI_FAILURE);
538 	}
539 
540 	if (!(un = (struct dcd_disk *)
541 	    ddi_get_soft_state(dcd_state, instance))) {
542 		return (DDI_FAILURE);
543 	}
544 	devp->dcd_private = (ataopaque_t)un;
545 
546 	/*
547 	 * Add a zero-length attribute to tell the world we support
548 	 * kernel ioctls (for layered drivers)
549 	 */
550 	(void) ddi_prop_create(DDI_DEV_T_NONE, devi, DDI_PROP_CANSLEEP,
551 	    DDI_KERNEL_IOCTL, NULL, 0);
552 
553 	/*
554 	 * Since the dad device does not have the 'reg' property,
555 	 * cpr will not call its DDI_SUSPEND/DDI_RESUME entries.
556 	 * The following code is to tell cpr that this device
557 	 * does need to be suspended and resumed.
558 	 */
559 	(void) ddi_prop_update_string(DDI_DEV_T_NONE, devi,
560 	    "pm-hardware-state", (caddr_t)"needs-suspend-resume");
561 
562 	/*
563 	 * Initialize power management bookkeeping;
564 	 * Create components - In IDE case there are 3 levels and one
565 	 * component. The levels being - active, idle, standby.
566 	 */
567 
568 	rval = ddi_prop_update_string_array(DDI_DEV_T_NONE,
569 	    devi, "pm-components", pm_comp, 4);
570 	if (rval == DDI_PROP_SUCCESS) {
571 		/*
572 		 * Ignore the return value of pm_raise_power
573 		 * Even if we check the return values and
574 		 * remove the property created above, PM
575 		 * framework will not honour the change after
576 		 * first call to pm_raise_power. Hence, the
577 		 * removal of that property does not help if
578 		 * pm_raise_power fails.
579 		 */
580 		(void) pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE);
581 	}
582 
583 	ddi_report_dev(devi);
584 
585 	cmlb_alloc_handle(&un->un_dklbhandle);
586 
587 	if (cmlb_attach(devi,
588 	    &dcd_lb_ops,
589 	    0,
590 	    B_FALSE,
591 	    B_FALSE,
592 	    DDI_NT_BLOCK_CHAN,
593 	    CMLB_FAKE_GEOM_LABEL_IOCTLS_VTOC8,
594 	    un->un_dklbhandle,
595 	    0) != 0) {
596 		cmlb_free_handle(&un->un_dklbhandle);
597 		dcd_free_softstate(un, devi);
598 		return (DDI_FAILURE);
599 	}
600 
601 	mutex_enter(DCD_MUTEX);
602 	(void) dcd_validate_geometry(un);
603 
604 	/* Get devid; create a devid ONLY IF could not get ID */
605 	if (dcd_get_devid(un) == NULL) {
606 		/* Create the fab'd devid */
607 		(void) dcd_create_devid(un);
608 	}
609 	mutex_exit(DCD_MUTEX);
610 
611 	return (DDI_SUCCESS);
612 }
613 
614 static void
615 dcd_free_softstate(struct dcd_disk *un, dev_info_t *devi)
616 {
617 	struct dcd_device		*devp;
618 	int instance = ddi_get_instance(devi);
619 
620 	devp = ddi_get_driver_private(devi);
621 
622 	if (un) {
623 		sema_destroy(&un->un_semoclose);
624 		cv_destroy(&un->un_sbuf_cv);
625 		cv_destroy(&un->un_state_cv);
626 		cv_destroy(&un->un_disk_busy_cv);
627 		cv_destroy(&un->un_suspend_cv);
628 
629 		/*
630 		 * Deallocate command packet resources.
631 		 */
632 		if (un->un_sbufp)
633 			freerbuf(un->un_sbufp);
634 		if (un->un_dp) {
635 			kmem_free((caddr_t)un->un_dp, sizeof (*un->un_dp));
636 		}
637 		/*
638 		 * Unregister the devid and free devid resources allocated
639 		 */
640 		ddi_devid_unregister(DCD_DEVINFO);
641 		if (un->un_devid) {
642 			ddi_devid_free(un->un_devid);
643 			un->un_devid = NULL;
644 		}
645 
646 		/*
647 		 * Delete kstats. Kstats for non CD devices are deleted
648 		 * in dcdclose.
649 		 */
650 		if (un->un_stats) {
651 			kstat_delete(un->un_stats);
652 		}
653 
654 	}
655 
656 	/*
657 	 * Cleanup scsi_device resources.
658 	 */
659 	ddi_soft_state_free(dcd_state, instance);
660 	devp->dcd_private = (ataopaque_t)0;
661 	/* unprobe scsi device */
662 	dcd_unprobe(devp);
663 
664 	/* Remove properties created during attach */
665 	ddi_prop_remove_all(devi);
666 }
667 
668 static int
669 dcddetach(dev_info_t *devi, ddi_detach_cmd_t cmd)
670 {
671 	int instance;
672 	struct dcd_disk *un;
673 	clock_t	wait_cmds_complete;
674 	instance = ddi_get_instance(devi);
675 
676 	if (!(un = ddi_get_soft_state(dcd_state, instance)))
677 		return (DDI_FAILURE);
678 
679 	switch (cmd) {
680 	case DDI_DETACH:
681 		return (dcd_dr_detach(devi));
682 
683 	case DDI_SUSPEND:
684 		mutex_enter(DCD_MUTEX);
685 		if (un->un_state == DCD_STATE_SUSPENDED) {
686 			mutex_exit(DCD_MUTEX);
687 			return (DDI_SUCCESS);
688 		}
689 		un->un_throttle = 0;
690 		/*
691 		 * Save the last state first
692 		 */
693 		un->un_save_state = un->un_last_state;
694 
695 		New_state(un, DCD_STATE_SUSPENDED);
696 
697 		/*
698 		 * wait till current operation completed. If we are
699 		 * in the resource wait state (with an intr outstanding)
700 		 * then we need to wait till the intr completes and
701 		 * starts the next cmd. We wait for
702 		 * DCD_WAIT_CMDS_COMPLETE seconds before failing the
703 		 * DDI_SUSPEND.
704 		 */
705 		wait_cmds_complete = ddi_get_lbolt();
706 		wait_cmds_complete +=
707 		    DCD_WAIT_CMDS_COMPLETE * drv_usectohz(1000000);
708 
709 		while (un->un_ncmds) {
710 			if (cv_timedwait(&un->un_disk_busy_cv,
711 			    DCD_MUTEX, wait_cmds_complete) == -1) {
712 				/*
713 				 * commands Didn't finish in the
714 				 * specified time, fail the DDI_SUSPEND.
715 				 */
716 				DAD_DEBUG2(DCD_DEVINFO, dcd_label,
717 				    DCD_DEBUG, "dcddetach: SUSPEND "
718 				    "failed due to outstanding cmds\n");
719 				Restore_state(un);
720 				mutex_exit(DCD_MUTEX);
721 				return (DDI_FAILURE);
722 			}
723 		}
724 		mutex_exit(DCD_MUTEX);
725 		return (DDI_SUCCESS);
726 	}
727 	return (DDI_FAILURE);
728 }
729 
730 /*
731  * The reset entry point gets invoked at the system shutdown time or through
732  * CPR code at system suspend.
733  * Will be flushing the cache and expect this to be last I/O operation to the
734  * disk before system reset/power off.
735  */
736 /*ARGSUSED*/
737 static int
738 dcdreset(dev_info_t *dip, ddi_reset_cmd_t cmd)
739 {
740 	struct dcd_disk *un;
741 	int instance;
742 
743 	instance = ddi_get_instance(dip);
744 
745 	if (!(un = ddi_get_soft_state(dcd_state, instance)))
746 		return (DDI_FAILURE);
747 
748 	dcd_flush_cache(un);
749 
750 	return (DDI_SUCCESS);
751 }
752 
753 
754 static int
755 dcd_dr_detach(dev_info_t *devi)
756 {
757 	struct dcd_device	*devp;
758 	struct dcd_disk		*un;
759 
760 	/*
761 	 * Get scsi_device structure for this instance.
762 	 */
763 	if ((devp = ddi_get_driver_private(devi)) == NULL)
764 		return (DDI_FAILURE);
765 
766 	/*
767 	 * Get dcd_disk structure containing target 'private' information
768 	 */
769 	un = (struct dcd_disk *)devp->dcd_private;
770 
771 	/*
772 	 * Verify there are NO outstanding commands issued to this device.
773 	 * ie, un_ncmds == 0.
774 	 * It's possible to have outstanding commands through the physio
775 	 * code path, even though everything's closed.
776 	 */
777 #ifndef lint
778 	_NOTE(COMPETING_THREADS_NOW);
779 #endif
780 	mutex_enter(DCD_MUTEX);
781 	if (un->un_ncmds) {
782 		mutex_exit(DCD_MUTEX);
783 		_NOTE(NO_COMPETING_THREADS_NOW);
784 		return (DDI_FAILURE);
785 	}
786 
787 	mutex_exit(DCD_MUTEX);
788 
789 	cmlb_detach(un->un_dklbhandle, 0);
790 	cmlb_free_handle(&un->un_dklbhandle);
791 
792 
793 	/*
794 	 * Lower the power state of the device
795 	 * i.e. the minimum power consumption state - sleep.
796 	 */
797 	(void) pm_lower_power(DCD_DEVINFO, 0, DCD_DEVICE_STANDBY);
798 
799 	_NOTE(NO_COMPETING_THREADS_NOW);
800 
801 	/*
802 	 * at this point there are no competing threads anymore
803 	 * release active MT locks and all device resources.
804 	 */
805 	dcd_free_softstate(un, devi);
806 
807 	return (DDI_SUCCESS);
808 }
809 
810 static int
811 dcdpower(dev_info_t *devi, int component, int level)
812 {
813 	struct dcd_pkt *pkt;
814 	struct dcd_disk *un;
815 	int	instance;
816 	uchar_t	cmd;
817 
818 
819 	instance = ddi_get_instance(devi);
820 
821 	if (!(un = ddi_get_soft_state(dcd_state, instance)) ||
822 	    (DCD_DEVICE_STANDBY > level) || (level > DCD_DEVICE_ACTIVE) ||
823 	    component != 0) {
824 		return (DDI_FAILURE);
825 	}
826 
827 	mutex_enter(DCD_MUTEX);
828 	/*
829 	 * if there are active commands for the device or device will be
830 	 * active soon. At the same time there is request to lower power
831 	 * return failure.
832 	 */
833 	if ((un->un_ncmds) && (level != DCD_DEVICE_ACTIVE)) {
834 		mutex_exit(DCD_MUTEX);
835 		return (DDI_FAILURE);
836 	}
837 
838 	if ((un->un_state == DCD_STATE_OFFLINE) ||
839 	    (un->un_state == DCD_STATE_FATAL)) {
840 		mutex_exit(DCD_MUTEX);
841 		return (DDI_FAILURE);
842 	}
843 
844 	if (level == DCD_DEVICE_ACTIVE) {
845 		/*
846 		 * No need to fire any command, just set the state structure
847 		 * to indicate previous state and set the level to active
848 		 */
849 		un->un_power_level = DCD_DEVICE_ACTIVE;
850 		if (un->un_state == DCD_STATE_PM_SUSPENDED)
851 			Restore_state(un);
852 		mutex_exit(DCD_MUTEX);
853 	} else {
854 		pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
855 		    NULL, (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
856 		    PKT_CONSISTENT, NULL_FUNC, NULL);
857 
858 		if (pkt == (struct dcd_pkt *)NULL) {
859 			mutex_exit(DCD_MUTEX);
860 			return (DDI_FAILURE);
861 		}
862 
863 		switch (level) {
864 		case DCD_DEVICE_IDLE:
865 			cmd = ATA_IDLE_IMMEDIATE;
866 			break;
867 
868 		case DCD_DEVICE_STANDBY:
869 			cmd = ATA_STANDBY_IMMEDIATE;
870 			break;
871 		}
872 
873 		makecommand(pkt, 0, cmd, 0, 0, 0, NO_DATA_XFER, 0);
874 		mutex_exit(DCD_MUTEX);
875 		/*
876 		 * Issue the appropriate command
877 		 */
878 		if ((dcd_poll(pkt)) || (SCBP_C(pkt) != STATUS_GOOD)) {
879 			dcd_destroy_pkt(pkt);
880 			return (DDI_FAILURE);
881 		}
882 		dcd_destroy_pkt(pkt);
883 		mutex_enter(DCD_MUTEX);
884 		if (un->un_state != DCD_STATE_PM_SUSPENDED)
885 			New_state(un, DCD_STATE_PM_SUSPENDED);
886 		un->un_power_level = level;
887 		mutex_exit(DCD_MUTEX);
888 	}
889 
890 	return (DDI_SUCCESS);
891 }
892 
893 static int
894 dcd_doattach(dev_info_t *devi, int (*canwait)())
895 {
896 	struct dcd_device *devp;
897 	struct dcd_disk *un = (struct dcd_disk *)0;
898 	int instance;
899 	int km_flags = (canwait != NULL_FUNC)? KM_SLEEP : KM_NOSLEEP;
900 	int rval;
901 	char *prop_template = "target%x-dcd-options";
902 	int options;
903 	char    prop_str[32];
904 	int target;
905 	diskaddr_t capacity;
906 
907 	devp = ddi_get_driver_private(devi);
908 
909 	/*
910 	 * Call the routine scsi_probe to do some of the dirty work.
911 	 * If the INQUIRY command succeeds, the field dcd_inq in the
912 	 * device structure will be filled in. The dcd_sense structure
913 	 * will also be allocated.
914 	 */
915 
916 	switch (dcd_probe(devp, canwait)) {
917 	default:
918 		return (DDI_FAILURE);
919 
920 	case DCDPROBE_EXISTS:
921 		if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
922 			if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
923 				rval = DDI_SUCCESS;
924 			} else {
925 				rval = DDI_FAILURE;
926 				goto error;
927 			}
928 		} else {
929 			rval = DDI_FAILURE;
930 			goto error;
931 		}
932 	}
933 
934 
935 	instance = ddi_get_instance(devp->dcd_dev);
936 
937 	if (ddi_soft_state_zalloc(dcd_state, instance) != DDI_SUCCESS) {
938 		rval = DDI_FAILURE;
939 		goto error;
940 	}
941 
942 	un = ddi_get_soft_state(dcd_state, instance);
943 
944 	un->un_sbufp = getrbuf(km_flags);
945 	if (un->un_sbufp == (struct buf *)NULL) {
946 		rval = DDI_FAILURE;
947 		goto error;
948 	}
949 
950 
951 	un->un_dcd = devp;
952 	un->un_power_level = -1;
953 	un->un_tgattribute.media_is_writable = 1;
954 
955 	sema_init(&un->un_semoclose, 1, NULL, SEMA_DRIVER, NULL);
956 	cv_init(&un->un_sbuf_cv, NULL, CV_DRIVER, NULL);
957 	cv_init(&un->un_state_cv, NULL, CV_DRIVER, NULL);
958 	/* Initialize power management conditional variable */
959 	cv_init(&un->un_disk_busy_cv, NULL, CV_DRIVER, NULL);
960 	cv_init(&un->un_suspend_cv, NULL, CV_DRIVER, NULL);
961 
962 	if (un->un_dp == 0) {
963 		/*
964 		 * Assume CCS drive, assume parity, but call
965 		 * it a CDROM if it is a RODIRECT device.
966 		 */
967 		un->un_dp = (struct dcd_drivetype *)
968 		    kmem_zalloc(sizeof (struct dcd_drivetype), km_flags);
969 		if (!un->un_dp) {
970 			rval = DDI_FAILURE;
971 			goto error;
972 		}
973 		if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
974 			if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
975 				un->un_dp->ctype = CTYPE_DISK;
976 			}
977 		} else  {
978 			rval = DDI_FAILURE;
979 			goto error;
980 		}
981 		un->un_dp->name = "CCS";
982 		un->un_dp->options = 0;
983 	}
984 
985 	/*
986 	 * Allow I/O requests at un_secsize offset in multiple of un_secsize.
987 	 */
988 	un->un_secsize = DEV_BSIZE;
989 
990 	/*
991 	 * If the device is not a removable media device, make sure that
992 	 * that the device is ready, by issuing the another identify but
993 	 * not needed. Get the capacity from identify data and store here.
994 	 */
995 	if (dcd_compute_dk_capacity(devp, &capacity) == 0) {
996 		un->un_diskcapacity = capacity;
997 		un->un_lbasize = DEV_BSIZE;
998 	}
999 
1000 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "Geometry Data\n");
1001 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "cyls %x, heads %x",
1002 	    devp->dcd_ident->dcd_fixcyls,
1003 	    devp->dcd_ident->dcd_heads);
1004 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "sectors %x,",
1005 	    devp->dcd_ident->dcd_sectors);
1006 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "capacity %llx\n",
1007 	    capacity);
1008 
1009 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1010 	    "dcdprobe: drive selected\n");
1011 
1012 	/*
1013 	 * Check for the property target<n>-dcd-options to find the option
1014 	 * set by the HBA driver for this target so that we can set the
1015 	 * Unit structure variable so that we can send commands accordingly.
1016 	 */
1017 	target = devp->dcd_address->da_target;
1018 	(void) sprintf(prop_str, prop_template, target);
1019 	options = ddi_prop_get_int(DDI_DEV_T_ANY, devi, DDI_PROP_NOTPROM,
1020 	    prop_str, -1);
1021 	if (options < 0) {
1022 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1023 		    "No per target properties");
1024 	} else {
1025 		if ((options & DCD_DMA_MODE) == DCD_DMA_MODE) {
1026 			un->un_dp->options |= DMA_SUPPORTTED;
1027 			un->un_dp->dma_mode = (options >> 3) & 0x03;
1028 			DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1029 			    "mode %x\n", un->un_dp->dma_mode);
1030 		} else {
1031 			un->un_dp->options &= ~DMA_SUPPORTTED;
1032 			un->un_dp->pio_mode = options & 0x7;
1033 			if (options & DCD_BLOCK_MODE)
1034 				un->un_dp->options |= BLOCK_MODE;
1035 			DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1036 			    "mode %x\n", un->un_dp->pio_mode);
1037 		}
1038 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1039 		    "options %x,", un->un_dp->options);
1040 	}
1041 
1042 	un->un_throttle = 2;
1043 	/*
1044 	 * set default max_xfer_size - This should depend on whether the
1045 	 * Block mode is supported by the device or not.
1046 	 */
1047 	un->un_max_xfer_size = MAX_ATA_XFER_SIZE;
1048 
1049 	/*
1050 	 * Set write cache enable softstate
1051 	 *
1052 	 * WCE is only supported in ATAPI-4 or higher; for
1053 	 * lower rev devices, must assume write cache is
1054 	 * enabled.
1055 	 */
1056 	mutex_enter(DCD_MUTEX);
1057 	un->un_write_cache_enabled = (devp->dcd_ident->dcd_majvers == 0xffff) ||
1058 	    ((devp->dcd_ident->dcd_majvers & IDENTIFY_80_ATAPI_4) == 0) ||
1059 	    (devp->dcd_ident->dcd_features85 & IDENTIFY_85_WCE) != 0;
1060 	mutex_exit(DCD_MUTEX);
1061 
1062 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1063 	    "dcd_doattach returns good\n");
1064 
1065 	return (rval);
1066 
1067 error:
1068 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcd_doattach failed\n");
1069 	dcd_free_softstate(un, devi);
1070 	return (rval);
1071 }
1072 
1073 #ifdef NOTNEEDED
1074 /*
1075  * This routine is used to set the block mode of operation by issuing the
1076  * Set Block mode ata command with the maximum block mode possible
1077  */
1078 dcd_set_multiple(struct dcd_disk *un)
1079 {
1080 	int status;
1081 	struct udcd_cmd ucmd;
1082 	struct dcd_cmd cdb;
1083 	dev_t	dev;
1084 
1085 
1086 	/* Zero all the required structure */
1087 	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));
1088 
1089 	(void) bzero((caddr_t)&cdb, sizeof (struct dcd_cmd));
1090 
1091 	cdb.cmd = ATA_SET_MULTIPLE;
1092 	/*
1093 	 * Here we should pass what needs to go into sector count REGISTER.
1094 	 * Eventhough this field indicates the number of bytes to read we
1095 	 * need to specify the block factor in terms of bytes so that it
1096 	 * will be programmed by the HBA driver into the sector count register.
1097 	 */
1098 	cdb.size = un->un_lbasize * un->un_dp->block_factor;
1099 
1100 	cdb.sector_num.lba_num = 0;
1101 	cdb.address_mode = ADD_LBA_MODE;
1102 	cdb.direction = NO_DATA_XFER;
1103 
1104 	ucmd.udcd_flags = 0;
1105 	ucmd.udcd_cmd = &cdb;
1106 	ucmd.udcd_bufaddr = NULL;
1107 	ucmd.udcd_buflen = 0;
1108 	ucmd.udcd_flags |= UDCD_SILENT;
1109 
1110 	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
1111 	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);
1112 
1113 
1114 	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);
1115 
1116 	return (status);
1117 }
1118 /*
1119  * The following routine is used only for setting the transfer mode
1120  * and it is not designed for transferring any other features subcommand.
1121  */
1122 dcd_set_features(struct dcd_disk *un, uchar_t mode)
1123 {
1124 	int status;
1125 	struct udcd_cmd ucmd;
1126 	struct dcd_cmd cdb;
1127 	dev_t	dev;
1128 
1129 
1130 	/* Zero all the required structure */
1131 	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));
1132 
1133 	(void) bzero((caddr_t)&cdb, sizeof (struct dcd_cmd));
1134 
1135 	cdb.cmd = ATA_SET_FEATURES;
1136 	/*
1137 	 * Here we need to pass what needs to go into the sector count register
1138 	 * But in the case of SET FEATURES command the value taken in the
1139 	 * sector count register depends what type of subcommand is
1140 	 * passed in the features register. Since we have defined the size to
1141 	 * be the size in bytes in this context it does not indicate bytes
1142 	 * instead it indicates the mode to be programmed.
1143 	 */
1144 	cdb.size = un->un_lbasize * mode;
1145 
1146 	cdb.sector_num.lba_num = 0;
1147 	cdb.address_mode = ADD_LBA_MODE;
1148 	cdb.direction = NO_DATA_XFER;
1149 	cdb.features = ATA_FEATURE_SET_MODE;
1150 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1151 	    "size %x, features %x, cmd %x\n",
1152 	    cdb.size, cdb.features, cdb.cmd);
1153 
1154 	ucmd.udcd_flags = 0;
1155 	ucmd.udcd_cmd = &cdb;
1156 	ucmd.udcd_bufaddr = NULL;
1157 	ucmd.udcd_buflen = 0;
1158 	ucmd.udcd_flags |= UDCD_SILENT;
1159 
1160 	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
1161 	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);
1162 
1163 	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);
1164 
1165 	return (status);
1166 }
1167 #endif
1168 
1169 /*
1170  * Validate the geometry for this disk, e.g.,
1171  * see whether it has a valid label.
1172  */
1173 static int
1174 dcd_validate_geometry(struct dcd_disk *un)
1175 {
1176 	int secsize = 0;
1177 	struct  dcd_device *devp;
1178 	int secdiv;
1179 	int rval;
1180 
1181 	ASSERT(mutex_owned(DCD_MUTEX));
1182 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1183 	    "dcd_validate_geometry: started \n");
1184 
1185 	if (un->un_lbasize < 0) {
1186 		return (DCD_BAD_LABEL);
1187 	}
1188 
1189 	if (un->un_state == DCD_STATE_PM_SUSPENDED) {
1190 		mutex_exit(DCD_MUTEX);
1191 		if (pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE) !=
1192 		    DDI_SUCCESS) {
1193 			mutex_enter(DCD_MUTEX);
1194 			return (DCD_BAD_LABEL);
1195 		}
1196 		mutex_enter(DCD_MUTEX);
1197 	}
1198 
1199 	secsize = un->un_secsize;
1200 
1201 	/*
1202 	 * take a log base 2 of sector size (sorry)
1203 	 */
1204 	for (secdiv = 0; secsize = secsize >> 1; secdiv++)
1205 		;
1206 	un->un_secdiv = secdiv;
1207 
1208 	/*
1209 	 * Only DIRECT ACCESS devices will have Sun labels.
1210 	 * CD's supposedly have a Sun label, too
1211 	 */
1212 
1213 	devp = un->un_dcd;
1214 
1215 	if (((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) &&
1216 	    (devp->dcd_ident->dcd_config & ATANON_REMOVABLE)) {
1217 		mutex_exit(DCD_MUTEX);
1218 		rval = cmlb_validate(un->un_dklbhandle, 0, 0);
1219 		mutex_enter(DCD_MUTEX);
1220 		if (rval == ENOMEM)
1221 			return (DCD_NO_MEM_FOR_LABEL);
1222 		else if (rval != 0)
1223 			return (DCD_BAD_LABEL);
1224 	} else {
1225 		/* it should never get here. */
1226 		return (DCD_BAD_LABEL);
1227 	}
1228 
1229 	/*
1230 	 * take a log base 2 of logical block size
1231 	 */
1232 	secsize = un->un_lbasize;
1233 	for (secdiv = 0; secsize = secsize >> 1; secdiv++)
1234 		;
1235 	un->un_lbadiv = secdiv;
1236 
1237 	/*
1238 	 * take a log base 2 of the multiple of DEV_BSIZE blocks that
1239 	 * make up one logical block
1240 	 */
1241 	secsize = un->un_lbasize >> DEV_BSHIFT;
1242 	for (secdiv = 0; secsize = secsize >> 1; secdiv++)
1243 		;
1244 	un->un_blknoshift = secdiv;
1245 	return (0);
1246 }
1247 
1248 /*
1249  * Unix Entry Points
1250  */
1251 
1252 /* ARGSUSED3 */
1253 static int
1254 dcdopen(dev_t *dev_p, int flag, int otyp, cred_t *cred_p)
1255 {
1256 	dev_t dev = *dev_p;
1257 	int rval = EIO;
1258 	int partmask;
1259 	int nodelay = (flag & (FNDELAY | FNONBLOCK));
1260 	int i;
1261 	char kstatname[KSTAT_STRLEN];
1262 	diskaddr_t lblocks;
1263 	char *partname;
1264 
1265 	GET_SOFT_STATE(dev);
1266 
1267 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1268 	    "Inside Open flag %x, otyp %x\n", flag, otyp);
1269 
1270 	if (otyp >= OTYPCNT) {
1271 		return (EINVAL);
1272 	}
1273 
1274 	partmask = 1 << part;
1275 
1276 	/*
1277 	 * We use a semaphore here in order to serialize
1278 	 * open and close requests on the device.
1279 	 */
1280 	sema_p(&un->un_semoclose);
1281 
1282 	mutex_enter(DCD_MUTEX);
1283 
1284 	if ((un->un_state & DCD_STATE_FATAL) == DCD_STATE_FATAL) {
1285 		rval = ENXIO;
1286 		goto done;
1287 	}
1288 
1289 	while (un->un_state == DCD_STATE_SUSPENDED) {
1290 		cv_wait(&un->un_suspend_cv, DCD_MUTEX);
1291 	}
1292 
1293 	if ((un->un_state == DCD_STATE_PM_SUSPENDED) && (!nodelay)) {
1294 		mutex_exit(DCD_MUTEX);
1295 		if (pm_raise_power(DCD_DEVINFO, 0, DCD_DEVICE_ACTIVE)
1296 		    != DDI_SUCCESS) {
1297 			mutex_enter(DCD_MUTEX);
1298 			rval = EIO;
1299 			goto done;
1300 		}
1301 		mutex_enter(DCD_MUTEX);
1302 	}
1303 
1304 	/*
1305 	 * set make_dcd_cmd() flags and stat_size here since these
1306 	 * are unlikely to change
1307 	 */
1308 	un->un_cmd_flags = 0;
1309 
1310 	un->un_cmd_stat_size = 2;
1311 
1312 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdopen un=0x%p\n",
1313 	    (void *)un);
1314 	/*
1315 	 * check for previous exclusive open
1316 	 */
1317 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1318 	    "exclopen=%x, flag=%x, regopen=%x\n",
1319 	    un->un_exclopen, flag, un->un_ocmap.regopen[otyp]);
1320 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1321 	    "Exclusive open flag %x, partmask %x\n",
1322 	    un->un_exclopen, partmask);
1323 
1324 	if (un->un_exclopen & (partmask)) {
1325 failed_exclusive:
1326 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1327 		    "exclusive open fails\n");
1328 		rval = EBUSY;
1329 		goto done;
1330 	}
1331 
1332 	if (flag & FEXCL) {
1333 		int i;
1334 		if (un->un_ocmap.lyropen[part]) {
1335 			goto failed_exclusive;
1336 		}
1337 		for (i = 0; i < (OTYPCNT - 1); i++) {
1338 			if (un->un_ocmap.regopen[i] & (partmask)) {
1339 				goto failed_exclusive;
1340 			}
1341 		}
1342 	}
1343 	if (flag & FWRITE) {
1344 		mutex_exit(DCD_MUTEX);
1345 		if (dcd_check_wp(dev)) {
1346 			sema_v(&un->un_semoclose);
1347 			return (EROFS);
1348 		}
1349 		mutex_enter(DCD_MUTEX);
1350 	}
1351 
1352 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1353 	    "Check Write Protect handled\n");
1354 
1355 	if (!nodelay) {
1356 		mutex_exit(DCD_MUTEX);
1357 		if ((rval = dcd_ready_and_valid(dev, un)) != 0) {
1358 			rval = EIO;
1359 		}
1360 		(void) pm_idle_component(DCD_DEVINFO, 0);
1361 		/*
1362 		 * Fail if device is not ready or if the number of disk
1363 		 * blocks is zero or negative for non CD devices.
1364 		 */
1365 		if (rval || cmlb_partinfo(un->un_dklbhandle,
1366 		    part, &lblocks, NULL, &partname, NULL, 0) ||
1367 		    lblocks <= 0) {
1368 			rval = EIO;
1369 			mutex_enter(DCD_MUTEX);
1370 			goto done;
1371 		}
1372 		mutex_enter(DCD_MUTEX);
1373 	}
1374 
1375 	if (otyp == OTYP_LYR) {
1376 		un->un_ocmap.lyropen[part]++;
1377 	} else {
1378 		un->un_ocmap.regopen[otyp] |= partmask;
1379 	}
1380 
1381 	/*
1382 	 * set up open and exclusive open flags
1383 	 */
1384 	if (flag & FEXCL) {
1385 		un->un_exclopen |= (partmask);
1386 	}
1387 
1388 
1389 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1390 	    "open of part %d type %d\n",
1391 	    part, otyp);
1392 
1393 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1394 	    "Kstats getting updated\n");
1395 	/*
1396 	 * only create kstats for disks, CD kstats created in dcdattach
1397 	 */
1398 	_NOTE(NO_COMPETING_THREADS_NOW);
1399 	mutex_exit(DCD_MUTEX);
1400 	if (un->un_stats == (kstat_t *)0) {
1401 		un->un_stats = kstat_create("dad", instance,
1402 		    NULL, "disk", KSTAT_TYPE_IO, 1,
1403 		    KSTAT_FLAG_PERSISTENT);
1404 		if (un->un_stats) {
1405 			un->un_stats->ks_lock = DCD_MUTEX;
1406 			kstat_install(un->un_stats);
1407 		}
1408 
1409 		/*
1410 		 * set up partition statistics for each partition
1411 		 * with number of blocks > 0
1412 		 */
1413 		if (!nodelay) {
1414 			for (i = 0; i < NDKMAP; i++) {
1415 				if ((un->un_pstats[i] == (kstat_t *)0) &&
1416 				    (cmlb_partinfo(un->un_dklbhandle,
1417 				    i, &lblocks, NULL, &partname,
1418 				    NULL, 0) == 0) && lblocks > 0) {
1419 					(void) sprintf(kstatname, "dad%d,%s",
1420 					    instance, partname);
1421 					un->un_pstats[i] = kstat_create("dad",
1422 					    instance,
1423 					    kstatname,
1424 					    "partition",
1425 					    KSTAT_TYPE_IO,
1426 					    1,
1427 					    KSTAT_FLAG_PERSISTENT);
1428 					if (un->un_pstats[i]) {
1429 						un->un_pstats[i]->ks_lock =
1430 						    DCD_MUTEX;
1431 						kstat_install(un->un_pstats[i]);
1432 					}
1433 				}
1434 			}
1435 		}
1436 		/*
1437 		 * set up error kstats
1438 		 */
1439 		(void) dcd_create_errstats(un, instance);
1440 	}
1441 #ifndef lint
1442 	_NOTE(COMPETING_THREADS_NOW);
1443 #endif
1444 
1445 	sema_v(&un->un_semoclose);
1446 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "Open success\n");
1447 	return (0);
1448 
1449 done:
1450 	mutex_exit(DCD_MUTEX);
1451 	sema_v(&un->un_semoclose);
1452 	return (rval);
1453 
1454 }
1455 
1456 /*
1457  * Test if disk is ready and has a valid geometry.
1458  */
1459 static int
1460 dcd_ready_and_valid(dev_t dev, struct dcd_disk *un)
1461 {
1462 	int rval = 1;
1463 	int g_error = 0;
1464 
1465 	mutex_enter(DCD_MUTEX);
1466 	/*
1467 	 * cmds outstanding
1468 	 */
1469 	if (un->un_ncmds == 0) {
1470 		(void) dcd_unit_ready(dev);
1471 	}
1472 
1473 	/*
1474 	 * If device is not yet ready here, inform it is offline
1475 	 */
1476 	if (un->un_state == DCD_STATE_NORMAL) {
1477 		rval = dcd_unit_ready(dev);
1478 		if (rval != 0 && rval != EACCES) {
1479 			dcd_offline(un, 1);
1480 			goto done;
1481 		}
1482 	}
1483 
1484 	if (un->un_format_in_progress == 0) {
1485 		g_error = dcd_validate_geometry(un);
1486 	}
1487 
1488 	/*
1489 	 * check if geometry was valid. We don't check the validity of
1490 	 * geometry for CDROMS.
1491 	 */
1492 
1493 	if (g_error == DCD_BAD_LABEL) {
1494 		rval = 1;
1495 		goto done;
1496 	}
1497 
1498 
1499 	/*
1500 	 * the state has changed; inform the media watch routines
1501 	 */
1502 	un->un_mediastate = DKIO_INSERTED;
1503 	cv_broadcast(&un->un_state_cv);
1504 	rval = 0;
1505 
1506 done:
1507 	mutex_exit(DCD_MUTEX);
1508 	return (rval);
1509 }
1510 
1511 
1512 /*ARGSUSED*/
1513 static int
1514 dcdclose(dev_t dev, int flag, int otyp, cred_t *cred_p)
1515 {
1516 	uchar_t *cp;
1517 	int i;
1518 
1519 	GET_SOFT_STATE(dev);
1520 
1521 
1522 	if (otyp >= OTYPCNT)
1523 		return (ENXIO);
1524 
1525 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1526 	    "close of part %d type %d\n",
1527 	    part, otyp);
1528 	sema_p(&un->un_semoclose);
1529 
1530 	mutex_enter(DCD_MUTEX);
1531 
1532 	if (un->un_exclopen & (1<<part)) {
1533 		un->un_exclopen &= ~(1<<part);
1534 	}
1535 
1536 	if (otyp == OTYP_LYR) {
1537 		un->un_ocmap.lyropen[part] -= 1;
1538 	} else {
1539 		un->un_ocmap.regopen[otyp] &= ~(1<<part);
1540 	}
1541 
1542 	cp = &un->un_ocmap.chkd[0];
1543 	while (cp < &un->un_ocmap.chkd[OCSIZE]) {
1544 		if (*cp != (uchar_t)0) {
1545 			break;
1546 		}
1547 		cp++;
1548 	}
1549 
1550 	if (cp == &un->un_ocmap.chkd[OCSIZE]) {
1551 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "last close\n");
1552 		if (un->un_state == DCD_STATE_OFFLINE) {
1553 			dcd_offline(un, 1);
1554 		}
1555 
1556 		mutex_exit(DCD_MUTEX);
1557 		(void) cmlb_close(un->un_dklbhandle, 0);
1558 
1559 		_NOTE(NO_COMPETING_THREADS_NOW);
1560 		if (un->un_stats) {
1561 			kstat_delete(un->un_stats);
1562 			un->un_stats = 0;
1563 		}
1564 		for (i = 0; i < NDKMAP; i++) {
1565 			if (un->un_pstats[i]) {
1566 				kstat_delete(un->un_pstats[i]);
1567 				un->un_pstats[i] = (kstat_t *)0;
1568 			}
1569 		}
1570 
1571 		if (un->un_errstats) {
1572 			kstat_delete(un->un_errstats);
1573 			un->un_errstats = (kstat_t *)0;
1574 		}
1575 		mutex_enter(DCD_MUTEX);
1576 
1577 #ifndef lint
1578 		_NOTE(COMPETING_THREADS_NOW);
1579 #endif
1580 	}
1581 
1582 	mutex_exit(DCD_MUTEX);
1583 	sema_v(&un->un_semoclose);
1584 	return (0);
1585 }
1586 
1587 static void
1588 dcd_offline(struct dcd_disk *un, int bechatty)
1589 {
1590 	if (bechatty)
1591 		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN, "offline\n");
1592 
1593 	mutex_exit(DCD_MUTEX);
1594 	cmlb_invalidate(un->un_dklbhandle, 0);
1595 	mutex_enter(DCD_MUTEX);
1596 }
1597 
1598 /*
1599  * Given the device number return the devinfo pointer
1600  * from the scsi_device structure.
1601  */
1602 /*ARGSUSED*/
1603 static int
1604 dcdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
1605 {
1606 	dev_t dev;
1607 	struct dcd_disk *un;
1608 	int instance, error;
1609 
1610 
1611 	switch (infocmd) {
1612 	case DDI_INFO_DEVT2DEVINFO:
1613 		dev = (dev_t)arg;
1614 		instance = DCDUNIT(dev);
1615 		if ((un = ddi_get_soft_state(dcd_state, instance)) == NULL)
1616 			return (DDI_FAILURE);
1617 		*result = (void *) DCD_DEVINFO;
1618 		error = DDI_SUCCESS;
1619 		break;
1620 	case DDI_INFO_DEVT2INSTANCE:
1621 		dev = (dev_t)arg;
1622 		instance = DCDUNIT(dev);
1623 		*result = (void *)(uintptr_t)instance;
1624 		error = DDI_SUCCESS;
1625 		break;
1626 	default:
1627 		error = DDI_FAILURE;
1628 	}
1629 	return (error);
1630 }
1631 
1632 /*
1633  * property operation routine.	return the number of blocks for the partition
1634  * in question or forward the request to the propery facilities.
1635  */
1636 static int
1637 dcd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1638     char *name, caddr_t valuep, int *lengthp)
1639 {
1640 	struct dcd_disk	*un;
1641 
1642 	if ((un = ddi_get_soft_state(dcd_state, ddi_get_instance(dip))) == NULL)
1643 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1644 		    name, valuep, lengthp));
1645 
1646 	return (cmlb_prop_op(un->un_dklbhandle,
1647 	    dev, dip, prop_op, mod_flags, name, valuep, lengthp,
1648 	    DCDPART(dev), NULL));
1649 }
1650 
1651 /*
1652  * These routines perform raw i/o operations.
1653  */
1654 /*ARGSUSED*/
1655 void
1656 dcduscsimin(struct buf *bp)
1657 {
1658 
1659 }
1660 
1661 
1662 static void
1663 dcdmin(struct buf *bp)
1664 {
1665 	struct dcd_disk *un;
1666 	int instance;
1667 	minor_t minor = getminor(bp->b_edev);
1668 	instance = minor >> DCDUNIT_SHIFT;
1669 	un = ddi_get_soft_state(dcd_state, instance);
1670 
1671 	if (bp->b_bcount > un->un_max_xfer_size)
1672 		bp->b_bcount = un->un_max_xfer_size;
1673 }
1674 
1675 
1676 /* ARGSUSED2 */
1677 static int
1678 dcdread(dev_t dev, struct uio *uio, cred_t *cred_p)
1679 {
1680 	int secmask;
1681 	GET_SOFT_STATE(dev);
1682 #ifdef lint
1683 	part = part;
1684 #endif /* lint */
1685 	secmask = un->un_secsize - 1;
1686 
1687 	if (uio->uio_loffset & ((offset_t)(secmask))) {
1688 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1689 		    "file offset not modulo %d\n",
1690 		    un->un_secsize);
1691 		return (EINVAL);
1692 	} else if (uio->uio_iov->iov_len & (secmask)) {
1693 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1694 		    "transfer length not modulo %d\n", un->un_secsize);
1695 		return (EINVAL);
1696 	}
1697 	return (physio(dcdstrategy, (struct buf *)0, dev, B_READ, dcdmin, uio));
1698 }
1699 
1700 /* ARGSUSED2 */
1701 static int
1702 dcdaread(dev_t dev, struct aio_req *aio, cred_t *cred_p)
1703 {
1704 	int secmask;
1705 	struct uio *uio = aio->aio_uio;
1706 	GET_SOFT_STATE(dev);
1707 #ifdef lint
1708 	part = part;
1709 #endif /* lint */
1710 	secmask = un->un_secsize - 1;
1711 
1712 	if (uio->uio_loffset & ((offset_t)(secmask))) {
1713 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1714 		    "file offset not modulo %d\n",
1715 		    un->un_secsize);
1716 		return (EINVAL);
1717 	} else if (uio->uio_iov->iov_len & (secmask)) {
1718 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1719 		    "transfer length not modulo %d\n", un->un_secsize);
1720 		return (EINVAL);
1721 	}
1722 	return (aphysio(dcdstrategy, anocancel, dev, B_READ, dcdmin, aio));
1723 }
1724 
1725 /* ARGSUSED2 */
1726 static int
1727 dcdwrite(dev_t dev, struct uio *uio, cred_t *cred_p)
1728 {
1729 	int secmask;
1730 	GET_SOFT_STATE(dev);
1731 #ifdef lint
1732 	part = part;
1733 #endif /* lint */
1734 	secmask = un->un_secsize - 1;
1735 
1736 	if (uio->uio_loffset & ((offset_t)(secmask))) {
1737 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1738 		    "file offset not modulo %d\n",
1739 		    un->un_secsize);
1740 		return (EINVAL);
1741 	} else if (uio->uio_iov->iov_len & (secmask)) {
1742 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1743 		    "transfer length not modulo %d\n", un->un_secsize);
1744 		return (EINVAL);
1745 	}
1746 	return (physio(dcdstrategy, (struct buf *)0, dev, B_WRITE, dcdmin,
1747 	    uio));
1748 }
1749 
1750 /* ARGSUSED2 */
1751 static int
1752 dcdawrite(dev_t dev, struct aio_req *aio, cred_t *cred_p)
1753 {
1754 	int secmask;
1755 	struct uio *uio = aio->aio_uio;
1756 	GET_SOFT_STATE(dev);
1757 #ifdef lint
1758 	part = part;
1759 #endif /* lint */
1760 	secmask = un->un_secsize - 1;
1761 
1762 	if (uio->uio_loffset & ((offset_t)(secmask))) {
1763 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1764 		    "file offset not modulo %d\n",
1765 		    un->un_secsize);
1766 		return (EINVAL);
1767 	} else if (uio->uio_iov->iov_len & (secmask)) {
1768 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1769 		    "transfer length not modulo %d\n", un->un_secsize);
1770 		return (EINVAL);
1771 	}
1772 	return (aphysio(dcdstrategy, anocancel, dev, B_WRITE, dcdmin, aio));
1773 }
1774 
1775 /*
1776  * strategy routine
1777  */
1778 static int
1779 dcdstrategy(struct buf *bp)
1780 {
1781 	struct dcd_disk *un;
1782 	struct diskhd *dp;
1783 	int i;
1784 	minor_t minor = getminor(bp->b_edev);
1785 	diskaddr_t p_lblksrt;
1786 	diskaddr_t lblocks;
1787 	diskaddr_t bn;
1788 
1789 	if ((un = ddi_get_soft_state(dcd_state,
1790 	    minor >> DCDUNIT_SHIFT)) == NULL ||
1791 	    un->un_state == DCD_STATE_DUMPING ||
1792 	    ((un->un_state  & DCD_STATE_FATAL) == DCD_STATE_FATAL)) {
1793 		SET_BP_ERROR(bp, ((un) ? ENXIO : EIO));
1794 error:
1795 		bp->b_resid = bp->b_bcount;
1796 		biodone(bp);
1797 		return (0);
1798 	}
1799 
1800 	/*
1801 	 * If the request size (buf->b_bcount)is greater than the size
1802 	 * (un->un_max_xfer_size) supported by the target driver fail
1803 	 * the request with EINVAL error code.
1804 	 *
1805 	 * We are not supposed to receive requests exceeding
1806 	 * un->un_max_xfer_size size because the caller is expected to
1807 	 * check what is the maximum size that is supported by this
1808 	 * driver either through ioctl or dcdmin routine(which is private
1809 	 * to this driver).
1810 	 * But we have seen cases (like meta driver(md))where dcdstrategy
1811 	 * called with more than supported size and cause data corruption.
1812 	 */
1813 
1814 	if (bp->b_bcount > un->un_max_xfer_size) {
1815 		SET_BP_ERROR(bp, EINVAL);
1816 		goto error;
1817 	}
1818 
1819 	TRACE_2(TR_FAC_DADA, TR_DCDSTRATEGY_START,
1820 	    "dcdstrategy_start: bp 0x%p un 0x%p", bp, un);
1821 
1822 	/*
1823 	 * Commands may sneak in while we released the mutex in
1824 	 * DDI_SUSPEND, we should block new commands.
1825 	 */
1826 	mutex_enter(DCD_MUTEX);
1827 	while (un->un_state == DCD_STATE_SUSPENDED) {
1828 		cv_wait(&un->un_suspend_cv, DCD_MUTEX);
1829 	}
1830 
1831 	if (un->un_state == DCD_STATE_PM_SUSPENDED) {
1832 		mutex_exit(DCD_MUTEX);
1833 		(void) pm_idle_component(DCD_DEVINFO, 0);
1834 		if (pm_raise_power(DCD_DEVINFO, 0,
1835 		    DCD_DEVICE_ACTIVE) !=  DDI_SUCCESS) {
1836 			SET_BP_ERROR(bp, EIO);
1837 			goto error;
1838 		}
1839 		mutex_enter(DCD_MUTEX);
1840 	}
1841 	mutex_exit(DCD_MUTEX);
1842 
1843 	/*
1844 	 * Map-in the buffer in case starting address is not word aligned.
1845 	 */
1846 
1847 	if (((uintptr_t)bp->b_un.b_addr) & 0x1)
1848 		bp_mapin(bp);
1849 
1850 	bp->b_flags &= ~(B_DONE|B_ERROR);
1851 	bp->b_resid = 0;
1852 	bp->av_forw = 0;
1853 
1854 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1855 	    "bp->b_bcount %lx\n", bp->b_bcount);
1856 
1857 	if (bp != un->un_sbufp) {
1858 validated:	if (cmlb_partinfo(un->un_dklbhandle,
1859 		    minor & DCDPART_MASK,
1860 		    &lblocks,
1861 		    &p_lblksrt,
1862 		    NULL,
1863 		    NULL,
1864 		    0) == 0) {
1865 
1866 			bn = dkblock(bp);
1867 
1868 			DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1869 			    "dkblock(bp) is %llu\n", bn);
1870 
1871 			i = 0;
1872 			if (bn < 0) {
1873 				i = -1;
1874 			} else if (bn >= lblocks) {
1875 				/*
1876 				 * For proper comparison, file system block
1877 				 * number has to be scaled to actual CD
1878 				 * transfer size.
1879 				 * Since all the CDROM operations
1880 				 * that have Sun Labels are in the correct
1881 				 * block size this will work for CD's.	This
1882 				 * will have to change when we have different
1883 				 * sector sizes.
1884 				 *
1885 				 * if bn == lblocks,
1886 				 * Not an error, resid == count
1887 				 */
1888 				if (bn > lblocks) {
1889 					i = -1;
1890 				} else {
1891 					i = 1;
1892 				}
1893 			} else if (bp->b_bcount & (un->un_secsize-1)) {
1894 				/*
1895 				 * This should really be:
1896 				 *
1897 				 * ... if (bp->b_bcount & (un->un_lbasize-1))
1898 				 *
1899 				 */
1900 				i = -1;
1901 			} else {
1902 				if (!bp->b_bcount) {
1903 					printf("Waring : Zero read or Write\n");
1904 					goto error;
1905 				}
1906 				/*
1907 				 * sort by absolute block number.
1908 				 */
1909 				bp->b_resid = bn;
1910 				bp->b_resid += p_lblksrt;
1911 				/*
1912 				 * zero out av_back - this will be a signal
1913 				 * to dcdstart to go and fetch the resources
1914 				 */
1915 				bp->av_back = NO_PKT_ALLOCATED;
1916 			}
1917 
1918 			/*
1919 			 * Check to see whether or not we are done
1920 			 * (with or without errors).
1921 			 */
1922 
1923 			if (i != 0) {
1924 				if (i < 0) {
1925 					bp->b_flags |= B_ERROR;
1926 				}
1927 				goto error;
1928 			}
1929 		} else {
1930 			/*
1931 			 * opened in NDELAY/NONBLOCK mode?
1932 			 * Check if disk is ready and has a valid geometry
1933 			 */
1934 			if (dcd_ready_and_valid(bp->b_edev, un) == 0) {
1935 				goto validated;
1936 			} else {
1937 				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
1938 				    "i/o to invalid geometry\n");
1939 				SET_BP_ERROR(bp, EIO);
1940 				goto error;
1941 			}
1942 		}
1943 	} else if (BP_HAS_NO_PKT(bp)) {
1944 		struct udcd_cmd *tscmdp;
1945 		struct dcd_cmd *tcmdp;
1946 		/*
1947 		 * This indicates that it is a special buffer
1948 		 * This could be a udcd-cmd and hence call bp_mapin just
1949 		 * in case that it could be a PIO command issued.
1950 		 */
1951 		tscmdp = (struct udcd_cmd *)bp->b_forw;
1952 		tcmdp = tscmdp->udcd_cmd;
1953 		if ((tcmdp->cmd != ATA_READ_DMA) && (tcmdp->cmd != 0xc9) &&
1954 		    (tcmdp->cmd != ATA_WRITE_DMA) && (tcmdp->cmd != 0xcb) &&
1955 		    (tcmdp->cmd != IDENTIFY_DMA) &&
1956 		    (tcmdp->cmd != ATA_FLUSH_CACHE)) {
1957 			bp_mapin(bp);
1958 		}
1959 	}
1960 
1961 	/*
1962 	 * We are doing it a bit non-standard. That is, the
1963 	 * head of the b_actf chain is *not* the active command-
1964 	 * it is just the head of the wait queue. The reason
1965 	 * we do this is that the head of the b_actf chain is
1966 	 * guaranteed to not be moved by disksort(), so that
1967 	 * our restart command (pointed to by
1968 	 * b_forw) and the head of the wait queue (b_actf) can
1969 	 * have resources granted without it getting lost in
1970 	 * the queue at some later point (where we would have
1971 	 * to go and look for it).
1972 	 */
1973 	mutex_enter(DCD_MUTEX);
1974 
1975 	DCD_DO_KSTATS(un, kstat_waitq_enter, bp);
1976 
1977 	dp = &un->un_utab;
1978 
1979 	if (dp->b_actf == NULL) {
1980 		dp->b_actf = bp;
1981 		dp->b_actl = bp;
1982 	} else if ((un->un_state == DCD_STATE_SUSPENDED) &&
1983 	    bp == un->un_sbufp) {
1984 		bp->b_actf = dp->b_actf;
1985 		dp->b_actf = bp;
1986 	} else {
1987 		TRACE_3(TR_FAC_DADA, TR_DCDSTRATEGY_DISKSORT_START,
1988 		    "dcdstrategy_disksort_start: dp 0x%p bp 0x%p un 0x%p",
1989 		    dp, bp, un);
1990 		disksort(dp, bp);
1991 		TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_DISKSORT_END,
1992 		    "dcdstrategy_disksort_end");
1993 	}
1994 
1995 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
1996 	    "ncmd %x , throttle %x, forw 0x%p\n",
1997 	    un->un_ncmds, un->un_throttle, (void *)dp->b_forw);
1998 	ASSERT(un->un_ncmds >= 0);
1999 	ASSERT(un->un_throttle >= 0);
2000 	if ((un->un_ncmds < un->un_throttle) && (dp->b_forw == NULL)) {
2001 		dcdstart(un);
2002 	} else if (BP_HAS_NO_PKT(dp->b_actf)) {
2003 		struct buf *cmd_bp;
2004 
2005 		cmd_bp = dp->b_actf;
2006 		cmd_bp->av_back = ALLOCATING_PKT;
2007 		mutex_exit(DCD_MUTEX);
2008 		/*
2009 		 * try and map this one
2010 		 */
2011 		TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_SMALL_WINDOW_START,
2012 		    "dcdstrategy_small_window_call (begin)");
2013 
2014 		make_dcd_cmd(un, cmd_bp, NULL_FUNC);
2015 
2016 		TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_SMALL_WINDOW_END,
2017 		    "dcdstrategy_small_window_call (end)");
2018 
2019 		/*
2020 		 * there is a small window where the active cmd
2021 		 * completes before make_dcd_cmd returns.
2022 		 * consequently, this cmd never gets started so
2023 		 * we start it from here
2024 		 */
2025 		mutex_enter(DCD_MUTEX);
2026 		if ((un->un_ncmds < un->un_throttle) &&
2027 		    (dp->b_forw == NULL)) {
2028 			dcdstart(un);
2029 		}
2030 	}
2031 	mutex_exit(DCD_MUTEX);
2032 
2033 done:
2034 	TRACE_0(TR_FAC_DADA, TR_DCDSTRATEGY_END, "dcdstrategy_end");
2035 	return (0);
2036 }
2037 
2038 
2039 /*
2040  * Unit start and Completion
2041  * NOTE: we assume that the caller has at least checked for:
2042  *		(un->un_ncmds < un->un_throttle)
2043  *	if not, there is no real harm done, dcd_transport() will
2044  *	return BUSY
2045  */
2046 static void
2047 dcdstart(struct dcd_disk *un)
2048 {
2049 	int status, sort_key;
2050 	struct buf *bp;
2051 	struct diskhd *dp;
2052 	uchar_t state = un->un_last_state;
2053 
2054 	TRACE_1(TR_FAC_DADA, TR_DCDSTART_START, "dcdstart_start: un 0x%p", un);
2055 
2056 retry:
2057 	ASSERT(mutex_owned(DCD_MUTEX));
2058 
2059 	dp = &un->un_utab;
2060 	if (((bp = dp->b_actf) == NULL) || (bp->av_back == ALLOCATING_PKT) ||
2061 	    (dp->b_forw != NULL)) {
2062 		TRACE_0(TR_FAC_DADA, TR_DCDSTART_NO_WORK_END,
2063 		    "dcdstart_end (no work)");
2064 		return;
2065 	}
2066 
2067 	/*
2068 	 * remove from active queue
2069 	 */
2070 	dp->b_actf = bp->b_actf;
2071 	bp->b_actf = 0;
2072 
2073 	/*
2074 	 * increment ncmds before calling dcd_transport because dcdintr
2075 	 * may be called before we return from dcd_transport!
2076 	 */
2077 	un->un_ncmds++;
2078 
2079 	/*
2080 	 * If measuring stats, mark exit from wait queue and
2081 	 * entrance into run 'queue' if and only if we are
2082 	 * going to actually start a command.
2083 	 * Normally the bp already has a packet at this point
2084 	 */
2085 	DCD_DO_KSTATS(un, kstat_waitq_to_runq, bp);
2086 
2087 	mutex_exit(DCD_MUTEX);
2088 
2089 	if (BP_HAS_NO_PKT(bp)) {
2090 		make_dcd_cmd(un, bp, dcdrunout);
2091 		if (BP_HAS_NO_PKT(bp) && !(bp->b_flags & B_ERROR)) {
2092 			mutex_enter(DCD_MUTEX);
2093 			DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
2094 
2095 			bp->b_actf = dp->b_actf;
2096 			dp->b_actf = bp;
2097 			New_state(un, DCD_STATE_RWAIT);
2098 			un->un_ncmds--;
2099 			TRACE_0(TR_FAC_DADA, TR_DCDSTART_NO_RESOURCES_END,
2100 			    "dcdstart_end (No Resources)");
2101 			goto done;
2102 
2103 		} else if (bp->b_flags & B_ERROR) {
2104 			mutex_enter(DCD_MUTEX);
2105 			DCD_DO_KSTATS(un, kstat_runq_exit, bp);
2106 
2107 			un->un_ncmds--;
2108 			bp->b_resid = bp->b_bcount;
2109 			if (bp->b_error == 0) {
2110 				SET_BP_ERROR(bp, EIO);
2111 			}
2112 
2113 			/*
2114 			 * restore old state
2115 			 */
2116 			un->un_state = un->un_last_state;
2117 			un->un_last_state = state;
2118 
2119 			mutex_exit(DCD_MUTEX);
2120 
2121 			biodone(bp);
2122 			mutex_enter(DCD_MUTEX);
2123 			if (un->un_state == DCD_STATE_SUSPENDED) {
2124 				cv_broadcast(&un->un_disk_busy_cv);
2125 			}
2126 
2127 			if ((un->un_ncmds < un->un_throttle) &&
2128 			    (dp->b_forw == NULL)) {
2129 				goto retry;
2130 			} else {
2131 				goto done;
2132 			}
2133 		}
2134 	}
2135 
2136 	/*
2137 	 * Restore resid from the packet, b_resid had been the
2138 	 * disksort key.
2139 	 */
2140 	sort_key = bp->b_resid;
2141 	bp->b_resid = BP_PKT(bp)->pkt_resid;
2142 	BP_PKT(bp)->pkt_resid = 0;
2143 
2144 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2145 	    "bp->b_resid %lx, pkt_resid %lx\n",
2146 	    bp->b_resid, BP_PKT(bp)->pkt_resid);
2147 
2148 	/*
2149 	 * We used to check whether or not to try and link commands here.
2150 	 * Since we have found that there is no performance improvement
2151 	 * for linked commands, this has not made much sense.
2152 	 */
2153 	if ((status = dcd_transport((struct dcd_pkt *)BP_PKT(bp)))
2154 	    != TRAN_ACCEPT) {
2155 		mutex_enter(DCD_MUTEX);
2156 		un->un_ncmds--;
2157 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2158 		    "transport returned %x\n", status);
2159 		if (status == TRAN_BUSY) {
2160 			DCD_DO_ERRSTATS(un, dcd_transerrs);
2161 			DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
2162 			dcd_handle_tran_busy(bp, dp, un);
2163 			if (un->un_ncmds > 0) {
2164 				bp->b_resid = sort_key;
2165 			}
2166 		} else {
2167 			DCD_DO_KSTATS(un, kstat_runq_exit, bp);
2168 			mutex_exit(DCD_MUTEX);
2169 
2170 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2171 			    "transport rejected (%d)\n",
2172 			    status);
2173 			SET_BP_ERROR(bp, EIO);
2174 			bp->b_resid = bp->b_bcount;
2175 			if (bp != un->un_sbufp) {
2176 				dcd_destroy_pkt(BP_PKT(bp));
2177 			}
2178 			biodone(bp);
2179 
2180 			mutex_enter(DCD_MUTEX);
2181 			if (un->un_state == DCD_STATE_SUSPENDED) {
2182 				cv_broadcast(&un->un_disk_busy_cv);
2183 			}
2184 			if ((un->un_ncmds < un->un_throttle) &&
2185 			    (dp->b_forw == NULL)) {
2186 					goto retry;
2187 			}
2188 		}
2189 	} else {
2190 		mutex_enter(DCD_MUTEX);
2191 
2192 		if (dp->b_actf && BP_HAS_NO_PKT(dp->b_actf)) {
2193 			struct buf *cmd_bp;
2194 
2195 			cmd_bp = dp->b_actf;
2196 			cmd_bp->av_back = ALLOCATING_PKT;
2197 			mutex_exit(DCD_MUTEX);
2198 			/*
2199 			 * try and map this one
2200 			 */
2201 			TRACE_0(TR_FAC_DADA, TR_DCASTART_SMALL_WINDOW_START,
2202 			    "dcdstart_small_window_start");
2203 
2204 			make_dcd_cmd(un, cmd_bp, NULL_FUNC);
2205 
2206 			TRACE_0(TR_FAC_DADA, TR_DCDSTART_SMALL_WINDOW_END,
2207 			    "dcdstart_small_window_end");
2208 			/*
2209 			 * there is a small window where the active cmd
2210 			 * completes before make_dcd_cmd returns.
2211 			 * consequently, this cmd never gets started so
2212 			 * we start it from here
2213 			 */
2214 			mutex_enter(DCD_MUTEX);
2215 			if ((un->un_ncmds < un->un_throttle) &&
2216 			    (dp->b_forw == NULL)) {
2217 				goto retry;
2218 			}
2219 		}
2220 	}
2221 
2222 done:
2223 	ASSERT(mutex_owned(DCD_MUTEX));
2224 	TRACE_0(TR_FAC_DADA, TR_DCDSTART_END, "dcdstart_end");
2225 }
2226 
2227 /*
2228  * make_dcd_cmd: create a pkt
2229  */
2230 static void
2231 make_dcd_cmd(struct dcd_disk *un, struct buf *bp, int (*func)())
2232 {
2233 	auto int count, com, direction;
2234 	struct dcd_pkt *pkt;
2235 	int flags, tval;
2236 
2237 	_NOTE(DATA_READABLE_WITHOUT_LOCK(dcd_disk::un_dp))
2238 	TRACE_3(TR_FAC_DADA, TR_MAKE_DCD_CMD_START,
2239 	    "make_dcd_cmd_start: un 0x%p bp 0x%p un 0x%p", un, bp, un);
2240 
2241 
2242 	flags = un->un_cmd_flags;
2243 
2244 	if (bp != un->un_sbufp) {
2245 		int partition = DCDPART(bp->b_edev);
2246 		diskaddr_t p_lblksrt;
2247 		diskaddr_t lblocks;
2248 		long secnt;
2249 		uint32_t blkno;
2250 		int dkl_nblk, delta;
2251 		long resid;
2252 
2253 		if (cmlb_partinfo(un->un_dklbhandle,
2254 		    partition,
2255 		    &lblocks,
2256 		    &p_lblksrt,
2257 		    NULL,
2258 		    NULL,
2259 		    0) != NULL) {
2260 			lblocks = 0;
2261 			p_lblksrt = 0;
2262 		}
2263 
2264 		dkl_nblk = (int)lblocks;
2265 
2266 		/*
2267 		 * Make sure we don't run off the end of a partition.
2268 		 *
2269 		 * Put this test here so that we can adjust b_count
2270 		 * to accurately reflect the actual amount we are
2271 		 * goint to transfer.
2272 		 */
2273 
2274 		/*
2275 		 * First, compute partition-relative block number
2276 		 */
2277 		blkno = dkblock(bp);
2278 		secnt = (bp->b_bcount + (un->un_secsize - 1)) >> un->un_secdiv;
2279 		count = MIN(secnt, dkl_nblk - blkno);
2280 		if (count != secnt) {
2281 			/*
2282 			 * We have an overrun
2283 			 */
2284 			resid = (secnt - count) << un->un_secdiv;
2285 			DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2286 			    "overrun by %ld sectors\n",
2287 			    secnt - count);
2288 			bp->b_bcount -= resid;
2289 		} else {
2290 			resid = 0;
2291 		}
2292 
2293 		/*
2294 		 * Adjust block number to absolute
2295 		 */
2296 		delta = (int)p_lblksrt;
2297 		blkno += delta;
2298 
2299 		mutex_enter(DCD_MUTEX);
2300 		/*
2301 		 * This is for devices having block size different from
2302 		 * from DEV_BSIZE (e.g. 2K CDROMs).
2303 		 */
2304 		if (un->un_lbasize != un->un_secsize) {
2305 			blkno >>= un->un_blknoshift;
2306 			count >>= un->un_blknoshift;
2307 		}
2308 		mutex_exit(DCD_MUTEX);
2309 
2310 		TRACE_0(TR_FAC_DADA, TR_MAKE_DCD_CMD_INIT_PKT_START,
2311 		    "make_dcd_cmd_init_pkt_call (begin)");
2312 		pkt = dcd_init_pkt(ROUTE, NULL, bp,
2313 		    (uint32_t)sizeof (struct dcd_cmd),
2314 		    un->un_cmd_stat_size, PP_LEN, PKT_CONSISTENT,
2315 		    func, (caddr_t)un);
2316 		TRACE_1(TR_FAC_DADA, TR_MAKE_DCD_CMD_INIT_PKT_END,
2317 		    "make_dcd_cmd_init_pkt_call (end): pkt 0x%p", pkt);
2318 		if (!pkt) {
2319 			bp->b_bcount += resid;
2320 			bp->av_back = NO_PKT_ALLOCATED;
2321 			TRACE_0(TR_FAC_DADA,
2322 			    TR_MAKE_DCD_CMD_NO_PKT_ALLOCATED1_END,
2323 			    "make_dcd_cmd_end (NO_PKT_ALLOCATED1)");
2324 			return;
2325 		}
2326 		if (bp->b_flags & B_READ) {
2327 			if ((un->un_dp->options & DMA_SUPPORTTED) ==
2328 			    DMA_SUPPORTTED) {
2329 				com = ATA_READ_DMA;
2330 			} else {
2331 				if (un->un_dp->options & BLOCK_MODE)
2332 					com = ATA_READ_MULTIPLE;
2333 				else
2334 					com = ATA_READ;
2335 			}
2336 			direction = DATA_READ;
2337 		} else {
2338 			if ((un->un_dp->options & DMA_SUPPORTTED) ==
2339 			    DMA_SUPPORTTED) {
2340 				com = ATA_WRITE_DMA;
2341 			} else {
2342 				if (un->un_dp->options & BLOCK_MODE)
2343 					com = ATA_WRITE_MULTIPLE;
2344 				else
2345 					com = ATA_WRITE;
2346 			}
2347 			direction = DATA_WRITE;
2348 		}
2349 
2350 		/*
2351 		 * Save the resid in the packet, temporarily until
2352 		 * we transport the command.
2353 		 */
2354 		pkt->pkt_resid = resid;
2355 
2356 		makecommand(pkt, flags, com, blkno, ADD_LBA_MODE,
2357 		    bp->b_bcount, direction, 0);
2358 		tval = dcd_io_time;
2359 	} else {
2360 
2361 		struct udcd_cmd *scmd = (struct udcd_cmd *)bp->b_forw;
2362 
2363 		/*
2364 		 * set options
2365 		 */
2366 		if ((scmd->udcd_flags & UDCD_SILENT) && !(DEBUGGING)) {
2367 			flags |= FLAG_SILENT;
2368 		}
2369 		if (scmd->udcd_flags &  UDCD_DIAGNOSE)
2370 			flags |= FLAG_DIAGNOSE;
2371 
2372 		if (scmd->udcd_flags & UDCD_NOINTR)
2373 			flags |= FLAG_NOINTR;
2374 
2375 		pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
2376 		    (bp->b_bcount)? bp: NULL,
2377 		    (uint32_t)sizeof (struct dcd_cmd),
2378 		    2, PP_LEN, PKT_CONSISTENT, func, (caddr_t)un);
2379 
2380 		if (!pkt) {
2381 			bp->av_back = NO_PKT_ALLOCATED;
2382 			return;
2383 		}
2384 
2385 		makecommand(pkt, 0, scmd->udcd_cmd->cmd,
2386 		    scmd->udcd_cmd->sector_num.lba_num,
2387 		    scmd->udcd_cmd->address_mode,
2388 		    scmd->udcd_cmd->size,
2389 		    scmd->udcd_cmd->direction, scmd->udcd_cmd->features);
2390 
2391 		pkt->pkt_flags = flags;
2392 		if (scmd->udcd_timeout == 0)
2393 			tval = dcd_io_time;
2394 		else
2395 			tval = scmd->udcd_timeout;
2396 		/* UDAD interface should be decided. */
2397 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2398 		    "udcd interface\n");
2399 	}
2400 
2401 	pkt->pkt_comp = dcdintr;
2402 	pkt->pkt_time = tval;
2403 	PKT_SET_BP(pkt, bp);
2404 	bp->av_back = (struct buf *)pkt;
2405 
2406 	TRACE_0(TR_FAC_DADA, TR_MAKE_DCD_CMD_END, "make_dcd_cmd_end");
2407 }
2408 
2409 /*
2410  * Command completion processing
2411  */
2412 static void
2413 dcdintr(struct dcd_pkt *pkt)
2414 {
2415 	struct dcd_disk *un;
2416 	struct buf *bp;
2417 	int action;
2418 	int status;
2419 
2420 	bp = PKT_GET_BP(pkt);
2421 	un = ddi_get_soft_state(dcd_state, DCDUNIT(bp->b_edev));
2422 
2423 	TRACE_1(TR_FAC_DADA, TR_DCDINTR_START, "dcdintr_start: un 0x%p", un);
2424 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdintr\n");
2425 
2426 	mutex_enter(DCD_MUTEX);
2427 	un->un_ncmds--;
2428 	DCD_DO_KSTATS(un, kstat_runq_exit, bp);
2429 	ASSERT(un->un_ncmds >= 0);
2430 
2431 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2432 	    "reason %x and Status %x\n", pkt->pkt_reason, SCBP_C(pkt));
2433 
2434 	/*
2435 	 * do most common case first
2436 	 */
2437 	if ((pkt->pkt_reason == CMD_CMPLT) && (SCBP_C(pkt) == 0)) {
2438 		int com = GETATACMD((struct dcd_cmd *)pkt->pkt_cdbp);
2439 
2440 		if (un->un_state == DCD_STATE_OFFLINE) {
2441 			un->un_state = un->un_last_state;
2442 			dcd_log(DCD_DEVINFO, dcd_label, CE_NOTE,
2443 			    (const char *) diskokay);
2444 		}
2445 		/*
2446 		 * If the command is a read or a write, and we have
2447 		 * a non-zero pkt_resid, that is an error. We should
2448 		 * attempt to retry the operation if possible.
2449 		 */
2450 		action = COMMAND_DONE;
2451 		if (pkt->pkt_resid && (com == ATA_READ || com == ATA_WRITE)) {
2452 			DCD_DO_ERRSTATS(un, dcd_harderrs);
2453 			if ((int)PKT_GET_RETRY_CNT(pkt) < dcd_retry_count) {
2454 				PKT_INCR_RETRY_CNT(pkt, 1);
2455 				action = QUE_COMMAND;
2456 			} else {
2457 				/*
2458 				 * if we have exhausted retries
2459 				 * a command with a residual is in error in
2460 				 * this case.
2461 				 */
2462 				action = COMMAND_DONE_ERROR;
2463 			}
2464 			dcd_log(DCD_DEVINFO, dcd_label,
2465 			    CE_WARN, "incomplete %s- %s\n",
2466 			    (bp->b_flags & B_READ)? "read" : "write",
2467 			    (action == QUE_COMMAND)? "retrying" :
2468 			    "giving up");
2469 		}
2470 
2471 		/*
2472 		 * pkt_resid will reflect, at this point, a residual
2473 		 * of how many bytes left to be transferred there were
2474 		 * from the actual scsi command. Add this to b_resid i.e
2475 		 * the amount this driver could not see to transfer,
2476 		 * to get the total number of bytes not transfered.
2477 		 */
2478 		if (action != QUE_COMMAND) {
2479 			bp->b_resid += pkt->pkt_resid;
2480 		}
2481 
2482 	} else if (pkt->pkt_reason != CMD_CMPLT) {
2483 		action = dcd_handle_incomplete(un, bp);
2484 	}
2485 
2486 	/*
2487 	 * If we are in the middle of syncing or dumping, we have got
2488 	 * here because dcd_transport has called us explictly after
2489 	 * completing the command in a polled mode. We don't want to
2490 	 * have a recursive call into dcd_transport again.
2491 	 */
2492 	if (ddi_in_panic() && (action == QUE_COMMAND)) {
2493 		action = COMMAND_DONE_ERROR;
2494 	}
2495 
2496 	/*
2497 	 * save pkt reason; consecutive failures are not reported unless
2498 	 * fatal
2499 	 * do not reset last_pkt_reason when the cmd was retried and
2500 	 * succeeded because
2501 	 * there maybe more commands comming back with last_pkt_reason
2502 	 */
2503 	if ((un->un_last_pkt_reason != pkt->pkt_reason) &&
2504 	    ((pkt->pkt_reason != CMD_CMPLT) ||
2505 	    (PKT_GET_RETRY_CNT(pkt) == 0))) {
2506 		un->un_last_pkt_reason = pkt->pkt_reason;
2507 	}
2508 
2509 	switch (action) {
2510 	case COMMAND_DONE_ERROR:
2511 error:
2512 		if (bp->b_resid == 0) {
2513 			bp->b_resid = bp->b_bcount;
2514 		}
2515 		if (bp->b_error == 0) {
2516 			struct	dcd_cmd *cdbp = (struct dcd_cmd *)pkt->pkt_cdbp;
2517 			if (cdbp->cmd == ATA_FLUSH_CACHE &&
2518 			    (pkt->pkt_scbp[0] & STATUS_ATA_ERR) &&
2519 			    (pkt->pkt_scbp[1] & ERR_ABORT)) {
2520 				SET_BP_ERROR(bp, ENOTSUP);
2521 				un->un_flush_not_supported = 1;
2522 			} else {
2523 				SET_BP_ERROR(bp, EIO);
2524 			}
2525 		}
2526 		bp->b_flags |= B_ERROR;
2527 		/*FALLTHROUGH*/
2528 	case COMMAND_DONE:
2529 		dcddone_and_mutex_exit(un, bp);
2530 
2531 		TRACE_0(TR_FAC_DADA, TR_DCDINTR_COMMAND_DONE_END,
2532 		    "dcdintr_end (COMMAND_DONE)");
2533 		return;
2534 
2535 	case QUE_COMMAND:
2536 		if (un->un_ncmds >= un->un_throttle) {
2537 			struct diskhd *dp = &un->un_utab;
2538 
2539 			bp->b_actf = dp->b_actf;
2540 			dp->b_actf = bp;
2541 
2542 			DCD_DO_KSTATS(un, kstat_waitq_enter, bp);
2543 
2544 			mutex_exit(DCD_MUTEX);
2545 			goto exit;
2546 		}
2547 
2548 		un->un_ncmds++;
2549 		/* reset the pkt reason again */
2550 		pkt->pkt_reason = 0;
2551 		DCD_DO_KSTATS(un, kstat_runq_enter, bp);
2552 		mutex_exit(DCD_MUTEX);
2553 		if ((status = dcd_transport(BP_PKT(bp))) != TRAN_ACCEPT) {
2554 			struct diskhd *dp = &un->un_utab;
2555 
2556 			mutex_enter(DCD_MUTEX);
2557 			un->un_ncmds--;
2558 			if (status == TRAN_BUSY) {
2559 				DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
2560 				dcd_handle_tran_busy(bp, dp, un);
2561 				mutex_exit(DCD_MUTEX);
2562 				goto exit;
2563 			}
2564 			DCD_DO_ERRSTATS(un, dcd_transerrs);
2565 			DCD_DO_KSTATS(un, kstat_runq_exit, bp);
2566 
2567 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2568 			    "requeue of command fails (%x)\n", status);
2569 			SET_BP_ERROR(bp, EIO);
2570 			bp->b_resid = bp->b_bcount;
2571 
2572 			dcddone_and_mutex_exit(un, bp);
2573 			goto exit;
2574 		}
2575 		break;
2576 
2577 	case JUST_RETURN:
2578 	default:
2579 		DCD_DO_KSTATS(un, kstat_waitq_enter, bp);
2580 		mutex_exit(DCD_MUTEX);
2581 		break;
2582 	}
2583 
2584 exit:
2585 	TRACE_0(TR_FAC_DADA, TR_DCDINTR_END, "dcdintr_end");
2586 }
2587 
2588 
2589 /*
2590  * Done with a command.
2591  */
2592 static void
2593 dcddone_and_mutex_exit(struct dcd_disk *un, register struct buf *bp)
2594 {
2595 	struct diskhd *dp;
2596 
2597 	TRACE_1(TR_FAC_DADA, TR_DCDONE_START, "dcddone_start: un 0x%p", un);
2598 
2599 	_NOTE(LOCK_RELEASED_AS_SIDE_EFFECT(&un->un_dcd->dcd_mutex));
2600 
2601 	dp = &un->un_utab;
2602 	if (bp == dp->b_forw) {
2603 		dp->b_forw = NULL;
2604 	}
2605 
2606 	if (un->un_stats) {
2607 		ulong_t n_done = bp->b_bcount - bp->b_resid;
2608 		if (bp->b_flags & B_READ) {
2609 			IOSP->reads++;
2610 			IOSP->nread += n_done;
2611 		} else {
2612 			IOSP->writes++;
2613 			IOSP->nwritten += n_done;
2614 		}
2615 	}
2616 	if (IO_PARTITION_STATS) {
2617 		ulong_t n_done = bp->b_bcount - bp->b_resid;
2618 		if (bp->b_flags & B_READ) {
2619 			IOSP_PARTITION->reads++;
2620 			IOSP_PARTITION->nread += n_done;
2621 		} else {
2622 			IOSP_PARTITION->writes++;
2623 			IOSP_PARTITION->nwritten += n_done;
2624 		}
2625 	}
2626 
2627 	/*
2628 	 * Start the next one before releasing resources on this one
2629 	 */
2630 	if (un->un_state == DCD_STATE_SUSPENDED) {
2631 		cv_broadcast(&un->un_disk_busy_cv);
2632 	} else if (dp->b_actf && (un->un_ncmds < un->un_throttle) &&
2633 	    (dp->b_forw == NULL && un->un_state != DCD_STATE_SUSPENDED)) {
2634 		dcdstart(un);
2635 	}
2636 
2637 	mutex_exit(DCD_MUTEX);
2638 
2639 	if (bp != un->un_sbufp) {
2640 		dcd_destroy_pkt(BP_PKT(bp));
2641 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2642 		    "regular done: resid %ld\n", bp->b_resid);
2643 	} else {
2644 		ASSERT(un->un_sbuf_busy);
2645 	}
2646 	TRACE_0(TR_FAC_DADA, TR_DCDDONE_BIODONE_CALL, "dcddone_biodone_call");
2647 
2648 	biodone(bp);
2649 
2650 	(void) pm_idle_component(DCD_DEVINFO, 0);
2651 
2652 	TRACE_0(TR_FAC_DADA, TR_DCDDONE_END, "dcddone end");
2653 }
2654 
2655 
2656 /*
2657  * reset the disk unless the transport layer has already
2658  * cleared the problem
2659  */
2660 #define	C1	(STAT_ATA_BUS_RESET|STAT_ATA_DEV_RESET|STAT_ATA_ABORTED)
2661 static void
2662 dcd_reset_disk(struct dcd_disk *un, struct dcd_pkt *pkt)
2663 {
2664 
2665 	if ((pkt->pkt_statistics & C1) == 0) {
2666 		mutex_exit(DCD_MUTEX);
2667 		if (!dcd_reset(ROUTE, RESET_ALL)) {
2668 			DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2669 			    "Reset failed");
2670 		}
2671 		mutex_enter(DCD_MUTEX);
2672 	}
2673 }
2674 
2675 static int
2676 dcd_handle_incomplete(struct dcd_disk *un, struct buf *bp)
2677 {
2678 	static char *fail = "ATA transport failed: reason '%s': %s\n";
2679 	static char *notresp = "disk not responding to selection\n";
2680 	int rval = COMMAND_DONE_ERROR;
2681 	int action = COMMAND_SOFT_ERROR;
2682 	struct dcd_pkt *pkt = BP_PKT(bp);
2683 	int be_chatty = (un->un_state != DCD_STATE_SUSPENDED) &&
2684 	    (bp != un->un_sbufp || !(pkt->pkt_flags & FLAG_SILENT));
2685 
2686 	ASSERT(mutex_owned(DCD_MUTEX));
2687 
2688 	switch (pkt->pkt_reason) {
2689 
2690 	case CMD_TIMEOUT:
2691 		/*
2692 		 * This Indicates the already the HBA would  have reset
2693 		 * so Just indicate to retry the command
2694 		 */
2695 		break;
2696 
2697 	case CMD_INCOMPLETE:
2698 		action = dcd_check_error(un, bp);
2699 		DCD_DO_ERRSTATS(un, dcd_transerrs);
2700 		if (action == COMMAND_HARD_ERROR) {
2701 			(void) dcd_reset_disk(un, pkt);
2702 		}
2703 		break;
2704 
2705 	case CMD_FATAL:
2706 		/*
2707 		 * Something drastic has gone wrong
2708 		 */
2709 		break;
2710 	case CMD_DMA_DERR:
2711 	case CMD_DATA_OVR:
2712 		/* FALLTHROUGH */
2713 
2714 	default:
2715 		/*
2716 		 * the target may still be running the	command,
2717 		 * so we should try and reset that target.
2718 		 */
2719 		DCD_DO_ERRSTATS(un, dcd_transerrs);
2720 		if ((pkt->pkt_reason != CMD_RESET) &&
2721 		    (pkt->pkt_reason != CMD_ABORTED)) {
2722 			(void) dcd_reset_disk(un, pkt);
2723 		}
2724 		break;
2725 	}
2726 
2727 	/*
2728 	 * If pkt_reason is CMD_RESET/ABORTED, chances are that this pkt got
2729 	 * reset/aborted because another disk on this bus caused it.
2730 	 * The disk that caused it, should get CMD_TIMEOUT with pkt_statistics
2731 	 * of STAT_TIMEOUT/STAT_DEV_RESET
2732 	 */
2733 	if ((pkt->pkt_reason == CMD_RESET) ||(pkt->pkt_reason == CMD_ABORTED)) {
2734 		/* To be written : XXX */
2735 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2736 		    "Command aborted\n");
2737 	}
2738 
2739 	if (bp == un->un_sbufp && (pkt->pkt_flags & FLAG_DIAGNOSE)) {
2740 		rval = COMMAND_DONE_ERROR;
2741 	} else {
2742 		if ((rval == COMMAND_DONE_ERROR) &&
2743 		    (action == COMMAND_SOFT_ERROR) &&
2744 		    ((int)PKT_GET_RETRY_CNT(pkt) < dcd_retry_count)) {
2745 			PKT_INCR_RETRY_CNT(pkt, 1);
2746 			rval = QUE_COMMAND;
2747 		}
2748 	}
2749 
2750 	if (pkt->pkt_reason == CMD_INCOMPLETE && rval == COMMAND_DONE_ERROR) {
2751 		/*
2752 		 * Looks like someone turned off this shoebox.
2753 		 */
2754 		if (un->un_state != DCD_STATE_OFFLINE) {
2755 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2756 			    (const char *) notresp);
2757 			New_state(un, DCD_STATE_OFFLINE);
2758 		}
2759 	} else if (pkt->pkt_reason == CMD_FATAL) {
2760 		/*
2761 		 * Suppressing the following message for the time being
2762 		 * dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2763 		 * (const char *) notresp);
2764 		 */
2765 		PKT_INCR_RETRY_CNT(pkt, 6);
2766 		rval = COMMAND_DONE_ERROR;
2767 		New_state(un, DCD_STATE_FATAL);
2768 	} else if (be_chatty) {
2769 		int in_panic = ddi_in_panic();
2770 		if (!in_panic || (rval == COMMAND_DONE_ERROR)) {
2771 			if (((pkt->pkt_reason != un->un_last_pkt_reason) &&
2772 			    (pkt->pkt_reason != CMD_RESET)) ||
2773 			    (rval == COMMAND_DONE_ERROR) ||
2774 			    (dcd_error_level == DCD_ERR_ALL)) {
2775 				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2776 				    fail, dcd_rname(pkt->pkt_reason),
2777 				    (rval == COMMAND_DONE_ERROR) ?
2778 				    "giving up": "retrying command");
2779 				DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2780 				    "retrycount=%x\n",
2781 				    PKT_GET_RETRY_CNT(pkt));
2782 			}
2783 		}
2784 	}
2785 error:
2786 	return (rval);
2787 }
2788 
2789 static int
2790 dcd_check_error(struct dcd_disk *un, struct buf *bp)
2791 {
2792 	struct diskhd *dp = &un->un_utab;
2793 	struct dcd_pkt *pkt = BP_PKT(bp);
2794 	int rval = 0;
2795 	unsigned char status;
2796 	unsigned char error;
2797 
2798 	TRACE_0(TR_FAC_DADA, TR_DCD_CHECK_ERROR_START, "dcd_check_error_start");
2799 	ASSERT(mutex_owned(DCD_MUTEX));
2800 
2801 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
2802 	    "Pkt: 0x%p dp: 0x%p\n", (void *)pkt, (void *)dp);
2803 
2804 	/*
2805 	 * Here we need to check status first and then if error is indicated
2806 	 * Then the error register.
2807 	 */
2808 
2809 	status = (pkt->pkt_scbp)[0];
2810 	if ((status & STATUS_ATA_DWF) == STATUS_ATA_DWF) {
2811 		/*
2812 		 * There has been a Device Fault  - reason for such error
2813 		 * is vendor specific
2814 		 * Action to be taken is - Indicate error and reset device.
2815 		 */
2816 
2817 		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN, "Device Fault\n");
2818 		rval = COMMAND_HARD_ERROR;
2819 	} else if ((status & STATUS_ATA_CORR) == STATUS_ATA_CORR) {
2820 
2821 		/*
2822 		 * The sector read or written is marginal and hence ECC
2823 		 * Correction has been applied. Indicate to repair
2824 		 * Here we need to probably re-assign based on the badblock
2825 		 * mapping.
2826 		 */
2827 
2828 		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2829 		    "Soft Error on block %x\n",
2830 		    ((struct dcd_cmd *)pkt->pkt_cdbp)->sector_num.lba_num);
2831 		rval = COMMAND_SOFT_ERROR;
2832 	} else if ((status & STATUS_ATA_ERR) == STATUS_ATA_ERR) {
2833 		error = pkt->pkt_scbp[1];
2834 
2835 		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2836 		    "Command:0x%x,Error:0x%x,Status:0x%x\n",
2837 		    GETATACMD((struct dcd_cmd *)pkt->pkt_cdbp),
2838 		    error, status);
2839 		if ((error &  ERR_AMNF) == ERR_AMNF) {
2840 			/* Address make not found */
2841 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2842 			    "Address Mark Not Found");
2843 		} else if ((error & ERR_TKONF) == ERR_TKONF) {
2844 			/* Track 0 Not found */
2845 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2846 			    "Track 0 Not found \n");
2847 		} else if ((error & ERR_IDNF) == ERR_IDNF) {
2848 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2849 			    " ID not found \n");
2850 		} else if ((error &  ERR_UNC) == ERR_UNC) {
2851 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2852 			    "Uncorrectable data Error: Block %x\n",
2853 			    ((struct dcd_cmd *)pkt->pkt_cdbp)->
2854 			    sector_num.lba_num);
2855 		} else if ((error & ERR_BBK) == ERR_BBK) {
2856 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2857 			    "Bad block detected: Block %x\n",
2858 			    ((struct dcd_cmd *)pkt->pkt_cdbp)->
2859 			    sector_num.lba_num);
2860 		} else if ((error & ERR_ABORT) == ERR_ABORT) {
2861 			/* Aborted Command */
2862 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2863 			    " Aborted Command \n");
2864 		}
2865 		/*
2866 		 * Return the soft error so that the command
2867 		 * will be retried.
2868 		 */
2869 		rval = COMMAND_SOFT_ERROR;
2870 	}
2871 
2872 	TRACE_0(TR_FAC_DADA, TR_DCD_CHECK_ERROR_END, "dcd_check_error_end");
2873 	return (rval);
2874 }
2875 
2876 
2877 /*
2878  *	System Crash Dump routine
2879  */
2880 
2881 #define	NDUMP_RETRIES	5
2882 
2883 static int
2884 dcddump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
2885 {
2886 	struct dcd_pkt *pkt;
2887 	int i;
2888 	struct buf local, *bp;
2889 	int err;
2890 	unsigned char com;
2891 	diskaddr_t p_lblksrt;
2892 	diskaddr_t lblocks;
2893 
2894 	GET_SOFT_STATE(dev);
2895 #ifdef lint
2896 	part = part;
2897 #endif /* lint */
2898 
2899 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*un))
2900 
2901 	if ((un->un_state & DCD_STATE_FATAL) == DCD_STATE_FATAL)
2902 		return (ENXIO);
2903 
2904 	if (cmlb_partinfo(un->un_dklbhandle, DCDPART(dev),
2905 	    &lblocks, &p_lblksrt, NULL, NULL, 0))
2906 		return (ENXIO);
2907 
2908 	if (blkno+nblk > lblocks) {
2909 		return (EINVAL);
2910 	}
2911 
2912 
2913 	if ((un->un_state == DCD_STATE_SUSPENDED) ||
2914 	    (un->un_state == DCD_STATE_PM_SUSPENDED)) {
2915 		if (pm_raise_power(DCD_DEVINFO, 0,
2916 		    DCD_DEVICE_ACTIVE) != DDI_SUCCESS) {
2917 			return (EIO);
2918 		}
2919 	}
2920 
2921 	/*
2922 	 * When cpr calls dcddump, we know that dad is in a
2923 	 * a good state, so no bus reset is required
2924 	 */
2925 	un->un_throttle = 0;
2926 
2927 	if ((un->un_state != DCD_STATE_SUSPENDED) &&
2928 	    (un->un_state != DCD_STATE_DUMPING)) {
2929 
2930 		New_state(un, DCD_STATE_DUMPING);
2931 
2932 		/*
2933 		 * Reset the bus. I'd like to not have to do this,
2934 		 * but this is the safest thing to do...
2935 		 */
2936 
2937 		if (dcd_reset(ROUTE, RESET_ALL) == 0) {
2938 			return (EIO);
2939 		}
2940 
2941 	}
2942 
2943 	blkno += p_lblksrt;
2944 
2945 	/*
2946 	 * It should be safe to call the allocator here without
2947 	 * worrying about being locked for DVMA mapping because
2948 	 * the address we're passed is already a DVMA mapping
2949 	 *
2950 	 * We are also not going to worry about semaphore ownership
2951 	 * in the dump buffer. Dumping is single threaded at present.
2952 	 */
2953 
2954 	bp = &local;
2955 	bzero((caddr_t)bp, sizeof (*bp));
2956 	bp->b_flags = B_BUSY;
2957 	bp->b_un.b_addr = addr;
2958 	bp->b_bcount = nblk << DEV_BSHIFT;
2959 	bp->b_resid = 0;
2960 
2961 	for (i = 0; i < NDUMP_RETRIES; i++) {
2962 		bp->b_flags &= ~B_ERROR;
2963 		if ((pkt = dcd_init_pkt(ROUTE, NULL, bp,
2964 		    (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
2965 		    PKT_CONSISTENT, NULL_FUNC, NULL)) != NULL) {
2966 			break;
2967 		}
2968 		if (i == 0) {
2969 			if (bp->b_flags & B_ERROR) {
2970 				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2971 				    "no resources for dumping; "
2972 				    "error code: 0x%x, retrying",
2973 				    geterror(bp));
2974 			} else {
2975 				dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
2976 				    "no resources for dumping; retrying");
2977 			}
2978 		} else if (i != (NDUMP_RETRIES - 1)) {
2979 			if (bp->b_flags & B_ERROR) {
2980 				dcd_log(DCD_DEVINFO, dcd_label, CE_CONT, "no "
2981 				    "resources for dumping; error code: 0x%x, "
2982 				    "retrying\n", geterror(bp));
2983 			}
2984 		} else {
2985 			if (bp->b_flags & B_ERROR) {
2986 				dcd_log(DCD_DEVINFO, dcd_label, CE_CONT,
2987 				    "no resources for dumping; "
2988 				    "error code: 0x%x, retries failed, "
2989 				    "giving up.\n", geterror(bp));
2990 			} else {
2991 				dcd_log(DCD_DEVINFO, dcd_label, CE_CONT,
2992 				    "no resources for dumping; "
2993 				    "retries failed, giving up.\n");
2994 			}
2995 			return (EIO);
2996 		}
2997 		delay(10);
2998 	}
2999 	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
3000 		com = ATA_WRITE_DMA;
3001 	} else {
3002 		if (un->un_dp->options & BLOCK_MODE)
3003 			com = ATA_WRITE_MULTIPLE;
3004 		else
3005 			com = ATA_WRITE;
3006 	}
3007 
3008 	makecommand(pkt, 0, com, blkno, ADD_LBA_MODE,
3009 	    (int)nblk*un->un_secsize, DATA_WRITE, 0);
3010 
3011 	for (err = EIO, i = 0; i < NDUMP_RETRIES && err == EIO; i++) {
3012 
3013 		if (dcd_poll(pkt) == 0) {
3014 			switch (SCBP_C(pkt)) {
3015 			case STATUS_GOOD:
3016 				if (pkt->pkt_resid == 0) {
3017 					err = 0;
3018 				}
3019 				break;
3020 			case STATUS_ATA_BUSY:
3021 				(void) dcd_reset(ROUTE, RESET_TARGET);
3022 				break;
3023 			default:
3024 				mutex_enter(DCD_MUTEX);
3025 				(void) dcd_reset_disk(un, pkt);
3026 				mutex_exit(DCD_MUTEX);
3027 				break;
3028 			}
3029 		} else if (i > NDUMP_RETRIES/2) {
3030 			(void) dcd_reset(ROUTE, RESET_ALL);
3031 		}
3032 
3033 	}
3034 	dcd_destroy_pkt(pkt);
3035 	return (err);
3036 }
3037 
3038 /*
3039  * This routine implements the ioctl calls.  It is called
3040  * from the device switch at normal priority.
3041  */
3042 /* ARGSUSED3 */
3043 static int
3044 dcdioctl(dev_t dev, int cmd, intptr_t arg, int flag,
3045 	cred_t *cred_p, int *rval_p)
3046 {
3047 	auto int32_t data[512 / (sizeof (int32_t))];
3048 	struct dk_cinfo *info;
3049 	struct dk_minfo media_info;
3050 	struct udcd_cmd *scmd;
3051 	int i, err;
3052 	enum uio_seg uioseg = 0;
3053 	enum dkio_state state = 0;
3054 #ifdef _MULTI_DATAMODEL
3055 	struct dadkio_rwcmd rwcmd;
3056 #endif
3057 	struct dadkio_rwcmd32 rwcmd32;
3058 	struct dcd_cmd dcdcmd;
3059 
3060 	GET_SOFT_STATE(dev);
3061 #ifdef lint
3062 	part = part;
3063 	state = state;
3064 	uioseg = uioseg;
3065 #endif  /* lint */
3066 
3067 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
3068 	    "dcd_ioctl : cmd %x, arg %lx\n", cmd, arg);
3069 
3070 	bzero((caddr_t)data, sizeof (data));
3071 
3072 	switch (cmd) {
3073 
3074 #ifdef DCDDEBUG
3075 /*
3076  * Following ioctl are for testing RESET/ABORTS
3077  */
3078 #define	DKIOCRESET	(DKIOC|14)
3079 #define	DKIOCABORT	(DKIOC|15)
3080 
3081 	case DKIOCRESET:
3082 		if (ddi_copyin((caddr_t)arg, (caddr_t)data, 4, flag))
3083 			return (EFAULT);
3084 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
3085 		    "DKIOCRESET: data = 0x%x\n", data[0]);
3086 		if (dcd_reset(ROUTE, data[0])) {
3087 			return (0);
3088 		} else {
3089 			return (EIO);
3090 		}
3091 	case DKIOCABORT:
3092 		DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG,
3093 		    "DKIOCABORT:\n");
3094 		if (dcd_abort(ROUTE, (struct dcd_pkt *)0)) {
3095 			return (0);
3096 		} else {
3097 			return (EIO);
3098 		}
3099 #endif
3100 
3101 	case DKIOCINFO:
3102 		/*
3103 		 * Controller Information
3104 		 */
3105 		info = (struct dk_cinfo *)data;
3106 
3107 		mutex_enter(DCD_MUTEX);
3108 		switch (un->un_dp->ctype) {
3109 		default:
3110 			info->dki_ctype = DKC_DIRECT;
3111 			break;
3112 		}
3113 		mutex_exit(DCD_MUTEX);
3114 		info->dki_cnum = ddi_get_instance(ddi_get_parent(DCD_DEVINFO));
3115 		(void) strcpy(info->dki_cname,
3116 		    ddi_get_name(ddi_get_parent(DCD_DEVINFO)));
3117 		/*
3118 		 * Unit Information
3119 		 */
3120 		info->dki_unit = ddi_get_instance(DCD_DEVINFO);
3121 		info->dki_slave = (Tgt(DCD_DCD_DEVP)<<3);
3122 		(void) strcpy(info->dki_dname, ddi_driver_name(DCD_DEVINFO));
3123 		info->dki_flags = DKI_FMTVOL;
3124 		info->dki_partition = DCDPART(dev);
3125 
3126 		/*
3127 		 * Max Transfer size of this device in blocks
3128 		 */
3129 		info->dki_maxtransfer = un->un_max_xfer_size / DEV_BSIZE;
3130 
3131 		/*
3132 		 * We can't get from here to there yet
3133 		 */
3134 		info->dki_addr = 0;
3135 		info->dki_space = 0;
3136 		info->dki_prio = 0;
3137 		info->dki_vec = 0;
3138 
3139 		i = sizeof (struct dk_cinfo);
3140 		if (ddi_copyout((caddr_t)data, (caddr_t)arg, i, flag))
3141 			return (EFAULT);
3142 		else
3143 			return (0);
3144 
3145 	case DKIOCGMEDIAINFO:
3146 		/*
3147 		 * As dad target driver is used for IDE disks only
3148 		 * Can keep the return value hardcoded to FIXED_DISK
3149 		 */
3150 		media_info.dki_media_type = DK_FIXED_DISK;
3151 
3152 		mutex_enter(DCD_MUTEX);
3153 		media_info.dki_lbsize = un->un_lbasize;
3154 		media_info.dki_capacity = un->un_diskcapacity;
3155 		mutex_exit(DCD_MUTEX);
3156 
3157 		if (ddi_copyout(&media_info, (caddr_t)arg,
3158 		    sizeof (struct dk_minfo), flag))
3159 			return (EFAULT);
3160 		else
3161 			return (0);
3162 
3163 	case DKIOCGGEOM:
3164 	case DKIOCGVTOC:
3165 	case DKIOCGETEFI:
3166 
3167 		mutex_enter(DCD_MUTEX);
3168 		if (un->un_ncmds == 0) {
3169 			if ((err = dcd_unit_ready(dev)) != 0) {
3170 				mutex_exit(DCD_MUTEX);
3171 				return (err);
3172 			}
3173 		}
3174 
3175 		mutex_exit(DCD_MUTEX);
3176 		err = cmlb_ioctl(un->un_dklbhandle, dev, cmd,
3177 		    arg, flag, cred_p, rval_p, 0);
3178 		return (err);
3179 
3180 	case DKIOCGAPART:
3181 	case DKIOCSAPART:
3182 	case DKIOCSGEOM:
3183 	case DKIOCSVTOC:
3184 	case DKIOCSETEFI:
3185 	case DKIOCPARTITION:
3186 	case DKIOCPARTINFO:
3187 	case DKIOCGMBOOT:
3188 	case DKIOCSMBOOT:
3189 
3190 		err = cmlb_ioctl(un->un_dklbhandle, dev, cmd,
3191 		    arg, flag, cred_p, rval_p, 0);
3192 		return (err);
3193 
3194 	case DIOCTL_RWCMD:
3195 		if (drv_priv(cred_p) != 0) {
3196 			return (EPERM);
3197 		}
3198 
3199 #ifdef _MULTI_DATAMODEL
3200 		switch (ddi_model_convert_from(flag & FMODELS)) {
3201 		case DDI_MODEL_NONE:
3202 			if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd,
3203 			    sizeof (struct dadkio_rwcmd), flag)) {
3204 				return (EFAULT);
3205 			}
3206 			rwcmd32.cmd = rwcmd.cmd;
3207 			rwcmd32.flags = rwcmd.flags;
3208 			rwcmd32.blkaddr = rwcmd.blkaddr;
3209 			rwcmd32.buflen = rwcmd.buflen;
3210 			rwcmd32.bufaddr = (caddr32_t)(uintptr_t)rwcmd.bufaddr;
3211 			break;
3212 		case DDI_MODEL_ILP32:
3213 			if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
3214 			    sizeof (struct dadkio_rwcmd32), flag)) {
3215 				return (EFAULT);
3216 			}
3217 			break;
3218 		}
3219 #else
3220 		if (ddi_copyin((caddr_t)arg, (caddr_t)&rwcmd32,
3221 		    sizeof (struct dadkio_rwcmd32), flag)) {
3222 			return (EFAULT);
3223 		}
3224 #endif
3225 		mutex_enter(DCD_MUTEX);
3226 
3227 		uioseg  = UIO_SYSSPACE;
3228 		scmd = (struct udcd_cmd *)data;
3229 		scmd->udcd_cmd = &dcdcmd;
3230 		/*
3231 		 * Convert the dadkio_rwcmd structure to udcd_cmd so that
3232 		 * it can take the normal path to get the io done
3233 		 */
3234 		if (rwcmd32.cmd == DADKIO_RWCMD_READ) {
3235 			if ((un->un_dp->options & DMA_SUPPORTTED) ==
3236 			    DMA_SUPPORTTED)
3237 				scmd->udcd_cmd->cmd = ATA_READ_DMA;
3238 			else
3239 				scmd->udcd_cmd->cmd = ATA_READ;
3240 			scmd->udcd_cmd->address_mode = ADD_LBA_MODE;
3241 			scmd->udcd_cmd->direction = DATA_READ;
3242 			scmd->udcd_flags |= UDCD_READ|UDCD_SILENT;
3243 		} else if (rwcmd32.cmd == DADKIO_RWCMD_WRITE) {
3244 			if ((un->un_dp->options & DMA_SUPPORTTED) ==
3245 			    DMA_SUPPORTTED)
3246 				scmd->udcd_cmd->cmd = ATA_WRITE_DMA;
3247 			else
3248 				scmd->udcd_cmd->cmd = ATA_WRITE;
3249 			scmd->udcd_cmd->direction = DATA_WRITE;
3250 			scmd->udcd_flags |= UDCD_WRITE|UDCD_SILENT;
3251 		} else {
3252 			mutex_exit(DCD_MUTEX);
3253 			return (EINVAL);
3254 		}
3255 
3256 		scmd->udcd_cmd->address_mode = ADD_LBA_MODE;
3257 		scmd->udcd_cmd->features = 0;
3258 		scmd->udcd_cmd->size = rwcmd32.buflen;
3259 		scmd->udcd_cmd->sector_num.lba_num = rwcmd32.blkaddr;
3260 		scmd->udcd_bufaddr = (caddr_t)(uintptr_t)rwcmd32.bufaddr;
3261 		scmd->udcd_buflen = rwcmd32.buflen;
3262 		scmd->udcd_timeout = (ushort_t)dcd_io_time;
3263 		scmd->udcd_resid = 0ULL;
3264 		scmd->udcd_status = 0;
3265 		scmd->udcd_error_reg = 0;
3266 		scmd->udcd_status_reg = 0;
3267 
3268 		mutex_exit(DCD_MUTEX);
3269 
3270 		i = dcdioctl_cmd(dev, scmd, UIO_SYSSPACE, UIO_USERSPACE);
3271 		mutex_enter(DCD_MUTEX);
3272 		/*
3273 		 * After return convert the status from scmd to
3274 		 * dadkio_status
3275 		 */
3276 		(void) dcd_translate(&(rwcmd32.status), scmd);
3277 		rwcmd32.status.resid = scmd->udcd_resid;
3278 		mutex_exit(DCD_MUTEX);
3279 
3280 #ifdef _MULTI_DATAMODEL
3281 		switch (ddi_model_convert_from(flag & FMODELS)) {
3282 		case DDI_MODEL_NONE: {
3283 			int counter;
3284 			rwcmd.status.status = rwcmd32.status.status;
3285 			rwcmd.status.resid  = rwcmd32.status.resid;
3286 			rwcmd.status.failed_blk_is_valid =
3287 			    rwcmd32.status.failed_blk_is_valid;
3288 			rwcmd.status.failed_blk = rwcmd32.status.failed_blk;
3289 			rwcmd.status.fru_code_is_valid =
3290 			    rwcmd32.status.fru_code_is_valid;
3291 			rwcmd.status.fru_code = rwcmd32.status.fru_code;
3292 			for (counter = 0;
3293 			    counter < DADKIO_ERROR_INFO_LEN; counter++)
3294 				rwcmd.status.add_error_info[counter] =
3295 				    rwcmd32.status.add_error_info[counter];
3296 			}
3297 			/* Copy out the result back to the user program */
3298 			if (ddi_copyout((caddr_t)&rwcmd, (caddr_t)arg,
3299 			    sizeof (struct dadkio_rwcmd), flag)) {
3300 				if (i != 0) {
3301 					i = EFAULT;
3302 				}
3303 			}
3304 			break;
3305 		case DDI_MODEL_ILP32:
3306 			/* Copy out the result back to the user program */
3307 			if (ddi_copyout((caddr_t)&rwcmd32, (caddr_t)arg,
3308 			    sizeof (struct dadkio_rwcmd32), flag)) {
3309 				if (i != 0) {
3310 					i = EFAULT;
3311 				}
3312 			}
3313 			break;
3314 		}
3315 #else
3316 		/* Copy out the result back to the user program  */
3317 		if (ddi_copyout((caddr_t)&rwcmd32, (caddr_t)arg,
3318 		    sizeof (struct dadkio_rwcmd32), flag)) {
3319 			if (i != 0)
3320 				i = EFAULT;
3321 		}
3322 #endif
3323 		return (i);
3324 
3325 	case UDCDCMD:	{
3326 #ifdef	_MULTI_DATAMODEL
3327 		/*
3328 		 * For use when a 32 bit app makes a call into a
3329 		 * 64 bit ioctl
3330 		 */
3331 		struct udcd_cmd32	udcd_cmd_32_for_64;
3332 		struct udcd_cmd32	*ucmd32 = &udcd_cmd_32_for_64;
3333 		model_t			model;
3334 #endif /* _MULTI_DATAMODEL */
3335 
3336 		if (drv_priv(cred_p) != 0) {
3337 			return (EPERM);
3338 		}
3339 
3340 		scmd = (struct udcd_cmd *)data;
3341 
3342 #ifdef _MULTI_DATAMODEL
3343 		switch (model = ddi_model_convert_from(flag & FMODELS)) {
3344 		case DDI_MODEL_ILP32:
3345 			if (ddi_copyin((caddr_t)arg, ucmd32,
3346 			    sizeof (struct udcd_cmd32), flag)) {
3347 				return (EFAULT);
3348 			}
3349 			/*
3350 			 * Convert the ILP32 uscsi data from the
3351 			 * application to LP64 for internal use.
3352 			 */
3353 			udcd_cmd32toudcd_cmd(ucmd32, scmd);
3354 			break;
3355 		case DDI_MODEL_NONE:
3356 			if (ddi_copyin((caddr_t)arg, scmd, sizeof (*scmd),
3357 			    flag)) {
3358 				return (EFAULT);
3359 			}
3360 			break;
3361 		}
3362 #else /* ! _MULTI_DATAMODEL */
3363 		if (ddi_copyin((caddr_t)arg, (caddr_t)scmd,
3364 		    sizeof (*scmd), flag)) {
3365 			return (EFAULT);
3366 		}
3367 #endif /* ! _MULTI_DATAMODEL */
3368 
3369 		scmd->udcd_flags &= ~UDCD_NOINTR;
3370 		uioseg = (flag & FKIOCTL)? UIO_SYSSPACE: UIO_USERSPACE;
3371 
3372 		i = dcdioctl_cmd(dev, scmd, uioseg, uioseg);
3373 #ifdef _MULTI_DATAMODEL
3374 		switch (model) {
3375 		case DDI_MODEL_ILP32:
3376 			/*
3377 			 * Convert back to ILP32 before copyout to the
3378 			 * application
3379 			 */
3380 			udcd_cmdtoudcd_cmd32(scmd, ucmd32);
3381 			if (ddi_copyout(ucmd32, (caddr_t)arg,
3382 			    sizeof (*ucmd32), flag)) {
3383 				if (i != 0)
3384 					i = EFAULT;
3385 			}
3386 			break;
3387 		case DDI_MODEL_NONE:
3388 			if (ddi_copyout(scmd, (caddr_t)arg, sizeof (*scmd),
3389 			    flag)) {
3390 				if (i != 0)
3391 					i = EFAULT;
3392 			}
3393 			break;
3394 		}
3395 #else /* ! _MULTI_DATAMODE */
3396 		if (ddi_copyout((caddr_t)scmd, (caddr_t)arg,
3397 		    sizeof (*scmd), flag)) {
3398 			if (i != 0)
3399 				i = EFAULT;
3400 		}
3401 #endif
3402 		return (i);
3403 	}
3404 	case DKIOCFLUSHWRITECACHE:	{
3405 		struct dk_callback *dkc = (struct dk_callback *)arg;
3406 		struct dcd_pkt *pkt;
3407 		struct buf *bp;
3408 		int is_sync = 1;
3409 
3410 		mutex_enter(DCD_MUTEX);
3411 		if (un->un_flush_not_supported ||
3412 		    ! un->un_write_cache_enabled) {
3413 			i = un->un_flush_not_supported ? ENOTSUP : 0;
3414 			mutex_exit(DCD_MUTEX);
3415 			/*
3416 			 * If a callback was requested: a callback will
3417 			 * always be done if the caller saw the
3418 			 * DKIOCFLUSHWRITECACHE ioctl return 0, and
3419 			 * never done if the caller saw the ioctl return
3420 			 * an error.
3421 			 */
3422 			if ((flag & FKIOCTL) && dkc != NULL &&
3423 			    dkc->dkc_callback != NULL) {
3424 				(*dkc->dkc_callback)(dkc->dkc_cookie, i);
3425 				/*
3426 				 * Did callback and reported error.
3427 				 * Since we did a callback, ioctl
3428 				 * should return 0.
3429 				 */
3430 				i = 0;
3431 			}
3432 			return (i);
3433 		}
3434 
3435 		/*
3436 		 * Get the special buffer
3437 		 */
3438 		while (un->un_sbuf_busy) {
3439 			cv_wait(&un->un_sbuf_cv, DCD_MUTEX);
3440 		}
3441 		un->un_sbuf_busy = 1;
3442 		bp  = un->un_sbufp;
3443 		mutex_exit(DCD_MUTEX);
3444 
3445 		pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
3446 		    NULL, (uint32_t)sizeof (struct dcd_cmd),
3447 		    2, PP_LEN, PKT_CONSISTENT, SLEEP_FUNC, (caddr_t)un);
3448 		ASSERT(pkt != NULL);
3449 
3450 		makecommand(pkt, un->un_cmd_flags | FLAG_SILENT,
3451 		    ATA_FLUSH_CACHE, 0, ADD_LBA_MODE, 0, NO_DATA_XFER, 0);
3452 
3453 		pkt->pkt_comp = dcdintr;
3454 		pkt->pkt_time = DCD_FLUSH_TIME;
3455 		PKT_SET_BP(pkt, bp);
3456 
3457 		bp->av_back = (struct buf *)pkt;
3458 		bp->b_forw = NULL;
3459 		bp->b_flags = B_BUSY;
3460 		bp->b_error = 0;
3461 		bp->b_edev = dev;
3462 		bp->b_dev = cmpdev(dev);
3463 		bp->b_bcount = 0;
3464 		bp->b_blkno = 0;
3465 		bp->b_un.b_addr = 0;
3466 		bp->b_iodone = NULL;
3467 		bp->b_list = NULL;
3468 
3469 		if ((flag & FKIOCTL) && dkc != NULL &&
3470 		    dkc->dkc_callback != NULL) {
3471 			struct dk_callback *dkc2 = (struct dk_callback *)
3472 			    kmem_zalloc(sizeof (*dkc2), KM_SLEEP);
3473 			bcopy(dkc, dkc2, sizeof (*dkc2));
3474 
3475 			bp->b_list = (struct buf *)dkc2;
3476 			bp->b_iodone = dcdflushdone;
3477 			is_sync = 0;
3478 		}
3479 
3480 		(void) dcdstrategy(bp);
3481 
3482 		i = 0;
3483 		if (is_sync) {
3484 			i = biowait(bp);
3485 			(void) dcdflushdone(bp);
3486 		}
3487 
3488 		return (i);
3489 	}
3490 	default:
3491 		break;
3492 	}
3493 	return (ENOTTY);
3494 }
3495 
3496 
3497 static int
3498 dcdflushdone(struct buf *bp)
3499 {
3500 	struct dcd_disk *un = ddi_get_soft_state(dcd_state,
3501 	    DCDUNIT(bp->b_edev));
3502 	struct dcd_pkt *pkt = BP_PKT(bp);
3503 	struct dk_callback *dkc = (struct dk_callback *)bp->b_list;
3504 
3505 	ASSERT(un != NULL);
3506 	ASSERT(bp == un->un_sbufp);
3507 	ASSERT(pkt != NULL);
3508 
3509 	dcd_destroy_pkt(pkt);
3510 	bp->av_back = NO_PKT_ALLOCATED;
3511 
3512 	if (dkc != NULL) {
3513 		ASSERT(bp->b_iodone != NULL);
3514 		(*dkc->dkc_callback)(dkc->dkc_cookie, geterror(bp));
3515 		kmem_free(dkc, sizeof (*dkc));
3516 		bp->b_iodone = NULL;
3517 		bp->b_list = NULL;
3518 	}
3519 
3520 	/*
3521 	 * Tell anybody who cares that the buffer is now free
3522 	 */
3523 	mutex_enter(DCD_MUTEX);
3524 	un->un_sbuf_busy = 0;
3525 	cv_signal(&un->un_sbuf_cv);
3526 	mutex_exit(DCD_MUTEX);
3527 	return (0);
3528 }
3529 
3530 /*
3531  * dcdrunout:
3532  *	the callback function for resource allocation
3533  *
3534  * XXX it would be preferable that dcdrunout() scans the whole
3535  *	list for possible candidates for dcdstart(); this avoids
3536  *	that a bp at the head of the list whose request cannot be
3537  *	satisfied is retried again and again
3538  */
3539 /*ARGSUSED*/
3540 static int
3541 dcdrunout(caddr_t arg)
3542 {
3543 	int serviced;
3544 	struct dcd_disk *un;
3545 	struct diskhd *dp;
3546 
3547 	TRACE_1(TR_FAC_DADA, TR_DCDRUNOUT_START, "dcdrunout_start: arg 0x%p",
3548 	    arg);
3549 	serviced = 1;
3550 
3551 	un = (struct dcd_disk *)arg;
3552 	dp = &un->un_utab;
3553 
3554 	/*
3555 	 * We now support passing a structure to the callback
3556 	 * routine.
3557 	 */
3558 	ASSERT(un != NULL);
3559 	mutex_enter(DCD_MUTEX);
3560 	if ((un->un_ncmds < un->un_throttle) && (dp->b_forw == NULL)) {
3561 		dcdstart(un);
3562 	}
3563 	if (un->un_state == DCD_STATE_RWAIT) {
3564 		serviced = 0;
3565 	}
3566 	mutex_exit(DCD_MUTEX);
3567 	TRACE_1(TR_FAC_DADA, TR_DCDRUNOUT_END,
3568 	    "dcdrunout_end: serviced %d", serviced);
3569 	return (serviced);
3570 }
3571 
3572 
3573 /*
3574  * This routine called to see whether unit is (still) there. Must not
3575  * be called when un->un_sbufp is in use, and must not be called with
3576  * an unattached disk. Soft state of disk is restored to what it was
3577  * upon entry- up to caller to set the correct state.
3578  *
3579  * We enter with the disk mutex held.
3580  */
3581 
3582 /* ARGSUSED0 */
3583 static int
3584 dcd_unit_ready(dev_t dev)
3585 {
3586 #ifndef lint
3587 	auto struct udcd_cmd dcmd, *com = &dcmd;
3588 	auto struct dcd_cmd cmdblk;
3589 #endif
3590 	int error;
3591 #ifndef lint
3592 	GET_SOFT_STATE(dev);
3593 #endif
3594 
3595 	/*
3596 	 * Now that we protect the special buffer with
3597 	 * a mutex, we could probably do a mutex_tryenter
3598 	 * on it here and return failure if it were held...
3599 	 */
3600 
3601 	error = 0;
3602 	return (error);
3603 }
3604 
3605 /* ARGSUSED0 */
3606 int
3607 dcdioctl_cmd(dev_t devp, struct udcd_cmd *in, enum uio_seg cdbspace,
3608     enum uio_seg dataspace)
3609 {
3610 
3611 	struct buf *bp;
3612 	struct	udcd_cmd *scmd;
3613 	struct dcd_pkt *pkt;
3614 	int	err, rw;
3615 	caddr_t	cdb;
3616 	int	flags = 0;
3617 
3618 	GET_SOFT_STATE(devp);
3619 
3620 #ifdef lint
3621 	part = part;
3622 #endif
3623 
3624 	/*
3625 	 * Is this a request to reset the bus?
3626 	 * if so, we need to do reseting.
3627 	 */
3628 
3629 	if (in->udcd_flags & UDCD_RESET) {
3630 		int flag = RESET_TARGET;
3631 		err = dcd_reset(ROUTE, flag) ? 0: EIO;
3632 		return (err);
3633 	}
3634 
3635 	scmd = in;
3636 
3637 
3638 	/* Do some sanity checks */
3639 	if (scmd->udcd_buflen <= 0) {
3640 		if (scmd->udcd_flags & (UDCD_READ | UDCD_WRITE)) {
3641 			return (EINVAL);
3642 		} else {
3643 			scmd->udcd_buflen = 0;
3644 		}
3645 	}
3646 
3647 	/* Make a copy of the dcd_cmd passed  */
3648 	cdb = kmem_zalloc(sizeof (struct dcd_cmd), KM_SLEEP);
3649 	if (cdbspace == UIO_SYSSPACE) {
3650 		flags |= FKIOCTL;
3651 	}
3652 
3653 	if (ddi_copyin((void *)scmd->udcd_cmd, cdb, sizeof (struct dcd_cmd),
3654 	    flags)) {
3655 		kmem_free(cdb, sizeof (struct dcd_cmd));
3656 		return (EFAULT);
3657 	}
3658 	scmd = (struct udcd_cmd *)kmem_alloc(sizeof (*scmd), KM_SLEEP);
3659 	bcopy((caddr_t)in, (caddr_t)scmd, sizeof (*scmd));
3660 	scmd->udcd_cmd = (struct dcd_cmd *)cdb;
3661 	rw = (scmd->udcd_flags & UDCD_READ) ? B_READ: B_WRITE;
3662 
3663 
3664 	/*
3665 	 * Get the special buffer
3666 	 */
3667 
3668 	mutex_enter(DCD_MUTEX);
3669 	while (un->un_sbuf_busy) {
3670 		if (cv_wait_sig(&un->un_sbuf_cv, DCD_MUTEX) == 0) {
3671 			kmem_free(scmd->udcd_cmd, sizeof (struct dcd_cmd));
3672 			kmem_free((caddr_t)scmd, sizeof (*scmd));
3673 			mutex_exit(DCD_MUTEX);
3674 			return (EINTR);
3675 		}
3676 	}
3677 
3678 	un->un_sbuf_busy = 1;
3679 	bp  = un->un_sbufp;
3680 	mutex_exit(DCD_MUTEX);
3681 
3682 
3683 	/*
3684 	 * If we are going to do actual I/O, let physio do all the
3685 	 * things
3686 	 */
3687 	DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
3688 	    "dcdioctl_cmd : buflen %x\n", scmd->udcd_buflen);
3689 
3690 	if (scmd->udcd_buflen) {
3691 		auto struct iovec aiov;
3692 		auto struct uio auio;
3693 		struct uio *uio = &auio;
3694 
3695 		bzero((caddr_t)&auio, sizeof (struct uio));
3696 		bzero((caddr_t)&aiov, sizeof (struct iovec));
3697 
3698 		aiov.iov_base = scmd->udcd_bufaddr;
3699 		aiov.iov_len = scmd->udcd_buflen;
3700 
3701 		uio->uio_iov = &aiov;
3702 		uio->uio_iovcnt = 1;
3703 		uio->uio_resid = scmd->udcd_buflen;
3704 		uio->uio_segflg = dataspace;
3705 
3706 		/*
3707 		 * Let physio do the rest...
3708 		 */
3709 		bp->av_back = NO_PKT_ALLOCATED;
3710 		bp->b_forw = (struct buf *)scmd;
3711 		err = physio(dcdstrategy, bp, devp, rw, dcdudcdmin, uio);
3712 	} else {
3713 		/*
3714 		 * We have to mimic what physio would do here.
3715 		 */
3716 		bp->av_back = NO_PKT_ALLOCATED;
3717 		bp->b_forw = (struct buf *)scmd;
3718 		bp->b_flags = B_BUSY | rw;
3719 		bp->b_edev = devp;
3720 		bp->b_dev = cmpdev(devp);
3721 		bp->b_bcount = bp->b_blkno = 0;
3722 		(void) dcdstrategy(bp);
3723 		err = biowait(bp);
3724 	}
3725 
3726 done:
3727 	if ((pkt = BP_PKT(bp)) != NULL) {
3728 		bp->av_back = NO_PKT_ALLOCATED;
3729 		/* we need to update the completion status of udcd command */
3730 		in->udcd_resid = bp->b_resid;
3731 		in->udcd_status_reg = SCBP_C(pkt);
3732 		/* XXX: we need to give error_reg also */
3733 		dcd_destroy_pkt(pkt);
3734 	}
3735 	/*
3736 	 * Tell anybody who cares that the buffer is now free
3737 	 */
3738 	mutex_enter(DCD_MUTEX);
3739 	un->un_sbuf_busy = 0;
3740 	cv_signal(&un->un_sbuf_cv);
3741 	mutex_exit(DCD_MUTEX);
3742 
3743 	kmem_free(scmd->udcd_cmd, sizeof (struct dcd_cmd));
3744 	kmem_free((caddr_t)scmd, sizeof (*scmd));
3745 	return (err);
3746 }
3747 
3748 static void
3749 dcdudcdmin(struct buf *bp)
3750 {
3751 
3752 #ifdef lint
3753 	bp = bp;
3754 #endif
3755 
3756 }
3757 
3758 /*
3759  * restart a cmd from timeout() context
3760  *
3761  * the cmd is expected to be in un_utab.b_forw. If this pointer is non-zero
3762  * a restart timeout request has been issued and no new timeouts should
3763  * be requested. b_forw is reset when the cmd eventually completes in
3764  * dcddone_and_mutex_exit()
3765  */
3766 void
3767 dcdrestart(void *arg)
3768 {
3769 	struct dcd_disk *un = (struct dcd_disk *)arg;
3770 	struct buf *bp;
3771 	int status;
3772 
3773 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdrestart\n");
3774 
3775 	mutex_enter(DCD_MUTEX);
3776 	bp = un->un_utab.b_forw;
3777 	if (bp) {
3778 		un->un_ncmds++;
3779 		DCD_DO_KSTATS(un, kstat_waitq_to_runq, bp);
3780 	}
3781 
3782 
3783 	if (bp) {
3784 		struct dcd_pkt *pkt = BP_PKT(bp);
3785 
3786 		mutex_exit(DCD_MUTEX);
3787 
3788 		pkt->pkt_flags = 0;
3789 
3790 		if ((status = dcd_transport(pkt)) != TRAN_ACCEPT) {
3791 			mutex_enter(DCD_MUTEX);
3792 			DCD_DO_KSTATS(un, kstat_runq_back_to_waitq, bp);
3793 			un->un_ncmds--;
3794 			if (status == TRAN_BUSY) {
3795 				/* XXX : To be checked */
3796 				/*
3797 				 * if (un->un_throttle > 1) {
3798 				 *	ASSERT(un->un_ncmds >= 0);
3799 				 *	un->un_throttle = un->un_ncmds;
3800 				 * }
3801 				 */
3802 				un->un_reissued_timeid =
3803 				    timeout(dcdrestart, (caddr_t)un,
3804 				    DCD_BSY_TIMEOUT/500);
3805 				mutex_exit(DCD_MUTEX);
3806 				return;
3807 			}
3808 			DCD_DO_ERRSTATS(un, dcd_transerrs);
3809 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
3810 			    "dcdrestart transport failed (%x)\n", status);
3811 			bp->b_resid = bp->b_bcount;
3812 			SET_BP_ERROR(bp, EIO);
3813 
3814 			DCD_DO_KSTATS(un, kstat_waitq_exit, bp);
3815 			un->un_reissued_timeid = 0L;
3816 			dcddone_and_mutex_exit(un, bp);
3817 			return;
3818 		}
3819 		mutex_enter(DCD_MUTEX);
3820 	}
3821 	un->un_reissued_timeid = 0L;
3822 	mutex_exit(DCD_MUTEX);
3823 	DAD_DEBUG(DCD_DEVINFO, dcd_label, DCD_DEBUG, "dcdrestart done\n");
3824 }
3825 
3826 /*
3827  * This routine gets called to reset the throttle to its saved
3828  * value wheneven we lower the throttle.
3829  */
3830 void
3831 dcd_reset_throttle(caddr_t arg)
3832 {
3833 	struct dcd_disk *un = (struct dcd_disk *)arg;
3834 	struct diskhd *dp;
3835 
3836 	mutex_enter(DCD_MUTEX);
3837 	dp = &un->un_utab;
3838 
3839 	/*
3840 	 * start any commands that didn't start while throttling.
3841 	 */
3842 	if (dp->b_actf && (un->un_ncmds < un->un_throttle) &&
3843 	    (dp->b_forw == NULL)) {
3844 		dcdstart(un);
3845 	}
3846 	mutex_exit(DCD_MUTEX);
3847 }
3848 
3849 
3850 /*
3851  * This routine handles the case when a TRAN_BUSY is
3852  * returned by HBA.
3853  *
3854  * If there are some commands already in the transport, the
3855  * bp can be put back on queue and it will
3856  * be retried when the queue is emptied after command
3857  * completes. But if there is no command in the tranport
3858  * and it still return busy, we have to retry the command
3859  * after some time like 10ms.
3860  */
3861 /* ARGSUSED0 */
3862 static void
3863 dcd_handle_tran_busy(struct buf *bp, struct diskhd *dp, struct dcd_disk *un)
3864 {
3865 	ASSERT(mutex_owned(DCD_MUTEX));
3866 
3867 
3868 	if (dp->b_forw == NULL || dp->b_forw == bp) {
3869 		dp->b_forw = bp;
3870 	} else if (dp->b_forw != bp) {
3871 		bp->b_actf = dp->b_actf;
3872 		dp->b_actf = bp;
3873 
3874 	}
3875 	if (!un->un_reissued_timeid) {
3876 		un->un_reissued_timeid =
3877 		    timeout(dcdrestart, (caddr_t)un, DCD_BSY_TIMEOUT/500);
3878 	}
3879 }
3880 
3881 static int
3882 dcd_write_deviceid(struct dcd_disk *un)
3883 {
3884 
3885 	int 	status;
3886 	diskaddr_t blk;
3887 	struct udcd_cmd ucmd;
3888 	struct dcd_cmd cdb;
3889 	struct dk_devid	*dkdevid;
3890 	uint_t *ip, chksum;
3891 	int	i;
3892 	dev_t	dev;
3893 
3894 	mutex_exit(DCD_MUTEX);
3895 	if (cmlb_get_devid_block(un->un_dklbhandle, &blk, 0)) {
3896 		mutex_enter(DCD_MUTEX);
3897 		return (EINVAL);
3898 	}
3899 	mutex_enter(DCD_MUTEX);
3900 
3901 	/* Allocate the buffer */
3902 	dkdevid = kmem_zalloc(un->un_secsize, KM_SLEEP);
3903 
3904 	/* Fill in the revision */
3905 	dkdevid->dkd_rev_hi = DK_DEVID_REV_MSB;
3906 	dkdevid->dkd_rev_lo = DK_DEVID_REV_LSB;
3907 
3908 	/* Copy in the device id */
3909 	bcopy(un->un_devid, &dkdevid->dkd_devid,
3910 	    ddi_devid_sizeof(un->un_devid));
3911 
3912 	/* Calculate the chksum */
3913 	chksum = 0;
3914 	ip = (uint_t *)dkdevid;
3915 	for (i = 0; i < ((un->un_secsize - sizeof (int))/sizeof (int)); i++)
3916 		chksum ^= ip[i];
3917 
3918 	/* Fill in the checksum */
3919 	DKD_FORMCHKSUM(chksum, dkdevid);
3920 
3921 	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));
3922 	(void) bzero((caddr_t)&cdb, sizeof (struct dcd_cmd));
3923 
3924 	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
3925 		cdb.cmd = ATA_WRITE_DMA;
3926 	} else {
3927 		if (un->un_dp->options & BLOCK_MODE)
3928 			cdb.cmd = ATA_WRITE_MULTIPLE;
3929 		else
3930 			cdb.cmd = ATA_WRITE;
3931 	}
3932 	cdb.size = un->un_secsize;
3933 	cdb.sector_num.lba_num = blk;
3934 	cdb.address_mode = ADD_LBA_MODE;
3935 	cdb.direction = DATA_WRITE;
3936 
3937 	ucmd.udcd_flags = UDCD_WRITE;
3938 	ucmd.udcd_cmd =  &cdb;
3939 	ucmd.udcd_bufaddr = (caddr_t)dkdevid;
3940 	ucmd.udcd_buflen = un->un_secsize;
3941 	ucmd.udcd_flags |= UDCD_SILENT;
3942 	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
3943 	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);
3944 	mutex_exit(DCD_MUTEX);
3945 	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);
3946 	mutex_enter(DCD_MUTEX);
3947 
3948 	kmem_free(dkdevid, un->un_secsize);
3949 	return (status);
3950 }
3951 
3952 static int
3953 dcd_read_deviceid(struct dcd_disk *un)
3954 {
3955 	int status;
3956 	diskaddr_t blk;
3957 	struct udcd_cmd ucmd;
3958 	struct dcd_cmd cdb;
3959 	struct dk_devid *dkdevid;
3960 	uint_t *ip;
3961 	int chksum;
3962 	int i, sz;
3963 	dev_t dev;
3964 
3965 	mutex_exit(DCD_MUTEX);
3966 	if (cmlb_get_devid_block(un->un_dklbhandle, &blk, 0)) {
3967 		mutex_enter(DCD_MUTEX);
3968 		return (EINVAL);
3969 	}
3970 	mutex_enter(DCD_MUTEX);
3971 
3972 	dkdevid = kmem_alloc(un->un_secsize, KM_SLEEP);
3973 
3974 	(void) bzero((caddr_t)&ucmd, sizeof (ucmd));
3975 	(void) bzero((caddr_t)&cdb, sizeof (cdb));
3976 
3977 	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
3978 		cdb.cmd = ATA_READ_DMA;
3979 	} else {
3980 		if (un->un_dp->options & BLOCK_MODE)
3981 			cdb.cmd = ATA_READ_MULTIPLE;
3982 		else
3983 			cdb.cmd = ATA_READ;
3984 	}
3985 	cdb.size = un->un_secsize;
3986 	cdb.sector_num.lba_num = blk;
3987 	cdb.address_mode = ADD_LBA_MODE;
3988 	cdb.direction = DATA_READ;
3989 
3990 	ucmd.udcd_flags = UDCD_READ;
3991 	ucmd.udcd_cmd =  &cdb;
3992 	ucmd.udcd_bufaddr = (caddr_t)dkdevid;
3993 	ucmd.udcd_buflen = un->un_secsize;
3994 	ucmd.udcd_flags |= UDCD_SILENT;
3995 	dev = makedevice(ddi_driver_major(DCD_DEVINFO),
3996 	    ddi_get_instance(DCD_DEVINFO) << DCDUNIT_SHIFT);
3997 	mutex_exit(DCD_MUTEX);
3998 	status = dcdioctl_cmd(dev, &ucmd, UIO_SYSSPACE, UIO_SYSSPACE);
3999 	mutex_enter(DCD_MUTEX);
4000 
4001 	if (status != 0) {
4002 		kmem_free((caddr_t)dkdevid, un->un_secsize);
4003 		return (status);
4004 	}
4005 
4006 	/* Validate the revision */
4007 
4008 	if ((dkdevid->dkd_rev_hi != DK_DEVID_REV_MSB) ||
4009 	    (dkdevid->dkd_rev_lo != DK_DEVID_REV_LSB)) {
4010 		kmem_free((caddr_t)dkdevid, un->un_secsize);
4011 		return (EINVAL);
4012 	}
4013 
4014 	/* Calculate the checksum */
4015 	chksum = 0;
4016 	ip = (uint_t *)dkdevid;
4017 	for (i = 0; i < ((un->un_secsize - sizeof (int))/sizeof (int)); i++)
4018 		chksum ^= ip[i];
4019 
4020 	/* Compare the checksums */
4021 
4022 	if (DKD_GETCHKSUM(dkdevid) != chksum) {
4023 		kmem_free((caddr_t)dkdevid, un->un_secsize);
4024 		return (EINVAL);
4025 	}
4026 
4027 	/* VAlidate the device id */
4028 	if (ddi_devid_valid((ddi_devid_t)&dkdevid->dkd_devid) != DDI_SUCCESS) {
4029 		kmem_free((caddr_t)dkdevid, un->un_secsize);
4030 		return (EINVAL);
4031 	}
4032 
4033 	/* return a copy of the device id */
4034 	sz = ddi_devid_sizeof((ddi_devid_t)&dkdevid->dkd_devid);
4035 	un->un_devid = (ddi_devid_t)kmem_alloc(sz, KM_SLEEP);
4036 	bcopy(&dkdevid->dkd_devid, un->un_devid, sz);
4037 	kmem_free((caddr_t)dkdevid, un->un_secsize);
4038 
4039 	return (0);
4040 }
4041 
4042 /*
4043  * Return the device id for the device.
4044  * 1. If the device ID exists then just return it - nothing to do in that case.
4045  * 2. Build one from the drives model number and serial number.
4046  * 3. If there is a problem in building it from serial/model #, then try
4047  * to read it from the acyl region of the disk.
4048  * Note: If this function is unable to return a valid ID then the calling
4049  * point will invoke the routine to create a fabricated ID ans stor it on the
4050  * acyl region of the disk.
4051  */
4052 static ddi_devid_t
4053 dcd_get_devid(struct dcd_disk *un)
4054 {
4055 	int		rc;
4056 
4057 	/* If already registered, return that value */
4058 	if (un->un_devid != NULL)
4059 		return (un->un_devid);
4060 
4061 	/* Build a devid from model and serial number, if present */
4062 	rc = dcd_make_devid_from_serial(un);
4063 
4064 	if (rc != DDI_SUCCESS) {
4065 		/* Read the devid from the disk. */
4066 		if (dcd_read_deviceid(un))
4067 			return (NULL);
4068 	}
4069 
4070 	(void) ddi_devid_register(DCD_DEVINFO, un->un_devid);
4071 	return (un->un_devid);
4072 }
4073 
4074 
4075 static ddi_devid_t
4076 dcd_create_devid(struct dcd_disk *un)
4077 {
4078 	if (ddi_devid_init(DCD_DEVINFO, DEVID_FAB, 0, NULL, (ddi_devid_t *)
4079 	    &un->un_devid) == DDI_FAILURE)
4080 		return (NULL);
4081 
4082 	if (dcd_write_deviceid(un)) {
4083 		ddi_devid_free(un->un_devid);
4084 		un->un_devid = NULL;
4085 		return (NULL);
4086 	}
4087 
4088 	(void) ddi_devid_register(DCD_DEVINFO, un->un_devid);
4089 	return (un->un_devid);
4090 }
4091 
4092 /*
4093  * Build a devid from the model and serial number, if present
4094  * Return DDI_SUCCESS or DDI_FAILURE.
4095  */
4096 static int
4097 dcd_make_devid_from_serial(struct dcd_disk *un)
4098 {
4099 	int	rc = DDI_SUCCESS;
4100 	char	*hwid;
4101 	char	*model;
4102 	int	model_len;
4103 	char	*serno;
4104 	int	serno_len;
4105 	int	total_len;
4106 
4107 	/* initialize the model and serial number information */
4108 	model = un->un_dcd->dcd_ident->dcd_model;
4109 	model_len = DCD_MODEL_NUMBER_LENGTH;
4110 	serno = un->un_dcd->dcd_ident->dcd_drvser;
4111 	serno_len = DCD_SERIAL_NUMBER_LENGTH;
4112 
4113 	/* Verify the model and serial number */
4114 	dcd_validate_model_serial(model, &model_len, model_len);
4115 	if (model_len == 0) {
4116 		rc = DDI_FAILURE;
4117 		goto out;
4118 	}
4119 	dcd_validate_model_serial(serno, &serno_len, serno_len);
4120 	if (serno_len == 0) {
4121 		rc = DDI_FAILURE;
4122 		goto out;
4123 	}
4124 
4125 	/*
4126 	 * The device ID will be concatenation of the model number,
4127 	 * the '=' separator, the serial number. Allocate
4128 	 * the string and concatenate the components.
4129 	 */
4130 	total_len = model_len + 1 + serno_len;
4131 	hwid = kmem_alloc(total_len, KM_SLEEP);
4132 	bcopy((caddr_t)model, (caddr_t)hwid, model_len);
4133 	bcopy((caddr_t)"=", (caddr_t)&hwid[model_len], 1);
4134 	bcopy((caddr_t)serno, (caddr_t)&hwid[model_len + 1], serno_len);
4135 
4136 	/* Initialize the device ID, trailing NULL not included */
4137 	rc = ddi_devid_init(DCD_DEVINFO, DEVID_ATA_SERIAL, total_len,
4138 	    hwid, (ddi_devid_t *)&un->un_devid);
4139 
4140 	/* Free the allocated string */
4141 	kmem_free(hwid, total_len);
4142 
4143 out:	return (rc);
4144 }
4145 
4146 /*
4147  * Test for a valid model or serial number. Assume that a valid representation
4148  * contains at least one character that is neither a space, 0 digit, or NULL.
4149  * Trim trailing blanks and NULLS from returned length.
4150  */
4151 static void
4152 dcd_validate_model_serial(char *str, int *retlen, int totallen)
4153 {
4154 	char		ch;
4155 	boolean_t	ret = B_FALSE;
4156 	int		i;
4157 	int		tb;
4158 
4159 	for (i = 0, tb = 0; i < totallen; i++) {
4160 		ch = *str++;
4161 		if ((ch != ' ') && (ch != '\0') && (ch != '0'))
4162 			ret = B_TRUE;
4163 		if ((ch == ' ') || (ch == '\0'))
4164 			tb++;
4165 		else
4166 			tb = 0;
4167 	}
4168 
4169 	if (ret == B_TRUE) {
4170 		/* Atleast one non 0 or blank character. */
4171 		*retlen = totallen - tb;
4172 	} else {
4173 		*retlen = 0;
4174 	}
4175 }
4176 
4177 #ifndef lint
4178 void
4179 clean_print(dev_info_t *dev, char *label, uint_t level,
4180 	char *title, char *data, int len)
4181 {
4182 	int	i;
4183 	char	buf[256];
4184 
4185 	(void) sprintf(buf, "%s:", title);
4186 	for (i = 0; i < len; i++) {
4187 		(void) sprintf(&buf[strlen(buf)], "0x%x ", (data[i] & 0xff));
4188 	}
4189 	(void) sprintf(&buf[strlen(buf)], "\n");
4190 
4191 	dcd_log(dev, label, level, "%s", buf);
4192 }
4193 #endif /* Not lint */
4194 
4195 #ifndef lint
4196 /*
4197  * Print a piece of inquiry data- cleaned up for non-printable characters
4198  * and stopping at the first space character after the beginning of the
4199  * passed string;
4200  */
4201 
4202 void
4203 inq_fill(char *p, int l, char *s)
4204 {
4205 	unsigned i = 0;
4206 	char c;
4207 
4208 	while (i++ < l) {
4209 		if ((c = *p++) < ' ' || c >= 0177) {
4210 			c = '*';
4211 		} else if (i != 1 && c == ' ') {
4212 			break;
4213 		}
4214 		*s++ = c;
4215 	}
4216 	*s++ = 0;
4217 }
4218 #endif /* Not lint */
4219 
4220 char *
4221 dcd_sname(uchar_t status)
4222 {
4223 	switch (status & STATUS_ATA_MASK) {
4224 	case STATUS_GOOD:
4225 		return ("good status");
4226 
4227 	case STATUS_ATA_BUSY:
4228 		return ("busy");
4229 
4230 	default:
4231 		return ("<unknown status>");
4232 	}
4233 }
4234 
4235 /* ARGSUSED0 */
4236 char *
4237 dcd_rname(int reason)
4238 {
4239 	static char *rnames[] = {
4240 		"cmplt",
4241 		"incomplete",
4242 		"dma_derr",
4243 		"tran_err",
4244 		"reset",
4245 		"aborted",
4246 		"timeout",
4247 		"data_ovr",
4248 	};
4249 	if (reason > CMD_DATA_OVR) {
4250 		return ("<unknown reason>");
4251 	} else {
4252 		return (rnames[reason]);
4253 	}
4254 }
4255 
4256 
4257 
4258 /* ARGSUSED0 */
4259 int
4260 dcd_check_wp(dev_t dev)
4261 {
4262 
4263 	return (0);
4264 }
4265 
4266 /*
4267  * Create device error kstats
4268  */
4269 static int
4270 dcd_create_errstats(struct dcd_disk *un, int instance)
4271 {
4272 
4273 	char kstatname[KSTAT_STRLEN];
4274 
4275 	if (un->un_errstats == (kstat_t *)0) {
4276 		(void) sprintf(kstatname, "dad%d,error", instance);
4277 		un->un_errstats = kstat_create("daderror", instance, kstatname,
4278 		    "device_error", KSTAT_TYPE_NAMED,
4279 		    sizeof (struct dcd_errstats)/ sizeof (kstat_named_t),
4280 		    KSTAT_FLAG_PERSISTENT);
4281 
4282 		if (un->un_errstats) {
4283 			struct dcd_errstats *dtp;
4284 
4285 			dtp = (struct dcd_errstats *)un->un_errstats->ks_data;
4286 			kstat_named_init(&dtp->dcd_softerrs, "Soft Errors",
4287 			    KSTAT_DATA_UINT32);
4288 			kstat_named_init(&dtp->dcd_harderrs, "Hard Errors",
4289 			    KSTAT_DATA_UINT32);
4290 			kstat_named_init(&dtp->dcd_transerrs,
4291 			    "Transport Errors", KSTAT_DATA_UINT32);
4292 			kstat_named_init(&dtp->dcd_model, "Model",
4293 			    KSTAT_DATA_CHAR);
4294 			kstat_named_init(&dtp->dcd_revision, "Revision",
4295 			    KSTAT_DATA_CHAR);
4296 			kstat_named_init(&dtp->dcd_serial, "Serial No",
4297 			    KSTAT_DATA_CHAR);
4298 			kstat_named_init(&dtp->dcd_capacity, "Size",
4299 			    KSTAT_DATA_ULONGLONG);
4300 			kstat_named_init(&dtp->dcd_rq_media_err, "Media Error",
4301 			    KSTAT_DATA_UINT32);
4302 			kstat_named_init(&dtp->dcd_rq_ntrdy_err,
4303 			    "Device Not Ready", KSTAT_DATA_UINT32);
4304 			kstat_named_init(&dtp->dcd_rq_nodev_err, " No Device",
4305 			    KSTAT_DATA_UINT32);
4306 			kstat_named_init(&dtp->dcd_rq_recov_err, "Recoverable",
4307 			    KSTAT_DATA_UINT32);
4308 			kstat_named_init(&dtp->dcd_rq_illrq_err,
4309 			    "Illegal Request", KSTAT_DATA_UINT32);
4310 
4311 			un->un_errstats->ks_private = un;
4312 			un->un_errstats->ks_update = nulldev;
4313 			kstat_install(un->un_errstats);
4314 
4315 			(void) strncpy(&dtp->dcd_model.value.c[0],
4316 			    un->un_dcd->dcd_ident->dcd_model, 16);
4317 			(void) strncpy(&dtp->dcd_serial.value.c[0],
4318 			    un->un_dcd->dcd_ident->dcd_drvser, 16);
4319 			(void) strncpy(&dtp->dcd_revision.value.c[0],
4320 			    un->un_dcd->dcd_ident->dcd_fw, 8);
4321 			dtp->dcd_capacity.value.ui64 =
4322 			    (uint64_t)((uint64_t)un->un_diskcapacity *
4323 			    (uint64_t)un->un_lbasize);
4324 		}
4325 	}
4326 	return (0);
4327 }
4328 
4329 
4330 /*
4331  * This has been moved from DADA layer as this does not do anything other than
4332  * retrying the command when it is busy or it does not complete
4333  */
4334 int
4335 dcd_poll(struct dcd_pkt *pkt)
4336 {
4337 	int	busy_count, rval = -1, savef;
4338 	clock_t	savet;
4339 	void	(*savec)();
4340 
4341 
4342 	/*
4343 	 * Save old flags
4344 	 */
4345 	savef = pkt->pkt_flags;
4346 	savec = pkt->pkt_comp;
4347 	savet = pkt->pkt_time;
4348 
4349 	pkt->pkt_flags |= FLAG_NOINTR;
4350 
4351 
4352 	/*
4353 	 * Set the Pkt_comp to NULL
4354 	 */
4355 
4356 	pkt->pkt_comp = 0;
4357 
4358 	/*
4359 	 * Set the Pkt time for the polled command
4360 	 */
4361 	if (pkt->pkt_time == 0) {
4362 		pkt->pkt_time = DCD_POLL_TIMEOUT;
4363 	}
4364 
4365 
4366 	/* Now transport the command */
4367 	for (busy_count = 0; busy_count < dcd_poll_busycnt; busy_count++) {
4368 		if ((rval = dcd_transport(pkt)) == TRAN_ACCEPT) {
4369 			if (pkt->pkt_reason == CMD_INCOMPLETE &&
4370 			    pkt->pkt_state == 0) {
4371 				delay(100);
4372 			} else if (pkt->pkt_reason  == CMD_CMPLT) {
4373 				rval = 0;
4374 				break;
4375 			}
4376 		}
4377 		if (rval == TRAN_BUSY)  {
4378 			delay(100);
4379 			continue;
4380 		}
4381 	}
4382 
4383 	pkt->pkt_flags = savef;
4384 	pkt->pkt_comp = savec;
4385 	pkt->pkt_time = savet;
4386 	return (rval);
4387 }
4388 
4389 
4390 void
4391 dcd_translate(struct dadkio_status32 *statp, struct udcd_cmd *cmdp)
4392 {
4393 	if (cmdp->udcd_status_reg & STATUS_ATA_BUSY)
4394 		statp->status = DADKIO_STAT_NOT_READY;
4395 	else if (cmdp->udcd_status_reg & STATUS_ATA_DWF)
4396 		statp->status = DADKIO_STAT_HARDWARE_ERROR;
4397 	else if (cmdp->udcd_status_reg & STATUS_ATA_CORR)
4398 		statp->status = DADKIO_STAT_SOFT_ERROR;
4399 	else if (cmdp->udcd_status_reg & STATUS_ATA_ERR) {
4400 		/*
4401 		 * The error register is valid only when BSY and DRQ not set
4402 		 * Assumed that HBA has checked this before it gives the data
4403 		 */
4404 		if (cmdp->udcd_error_reg & ERR_AMNF)
4405 			statp->status = DADKIO_STAT_NOT_FORMATTED;
4406 		else if (cmdp->udcd_error_reg & ERR_TKONF)
4407 			statp->status = DADKIO_STAT_NOT_FORMATTED;
4408 		else if (cmdp->udcd_error_reg & ERR_ABORT)
4409 			statp->status = DADKIO_STAT_ILLEGAL_REQUEST;
4410 		else if (cmdp->udcd_error_reg & ERR_IDNF)
4411 			statp->status = DADKIO_STAT_NOT_FORMATTED;
4412 		else if (cmdp->udcd_error_reg & ERR_UNC)
4413 			statp->status = DADKIO_STAT_BUS_ERROR;
4414 		else if (cmdp->udcd_error_reg & ERR_BBK)
4415 			statp->status = DADKIO_STAT_MEDIUM_ERROR;
4416 	} else
4417 		statp->status = DADKIO_STAT_NO_ERROR;
4418 }
4419 
4420 static void
4421 dcd_flush_cache(struct dcd_disk *un)
4422 {
4423 	struct dcd_pkt *pkt;
4424 	int retry_count;
4425 
4426 
4427 	if ((pkt = dcd_init_pkt(ROUTE, NULL, NULL,
4428 	    (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
4429 	    PKT_CONSISTENT, NULL_FUNC, NULL)) == NULL) {
4430 		return;
4431 	}
4432 
4433 	makecommand(pkt, 0, ATA_FLUSH_CACHE, 0, ADD_LBA_MODE, 0,
4434 	    NO_DATA_XFER, 0);
4435 
4436 	/*
4437 	 * Send the command. There are chances it might fail on some
4438 	 * disks since it is not a mandatory command as per ata-4. Try
4439 	 * 3 times if it fails. The retry count has been randomly selected.
4440 	 * There is a need for retry since as per the spec FLUSH CACHE can fail
4441 	 * as a result of unrecoverable error encountered during execution
4442 	 * of writing data and subsequent command should continue flushing
4443 	 * cache.
4444 	 */
4445 	for (retry_count = 0; retry_count < 3; retry_count++) {
4446 		/*
4447 		 * Set the packet fields.
4448 		 */
4449 		pkt->pkt_comp = 0;
4450 		pkt->pkt_time = DCD_POLL_TIMEOUT;
4451 		pkt->pkt_flags |= FLAG_FORCENOINTR;
4452 		pkt->pkt_flags |= FLAG_NOINTR;
4453 		if (dcd_transport(pkt) == TRAN_ACCEPT) {
4454 			if (pkt->pkt_reason  == CMD_CMPLT) {
4455 				break;
4456 			}
4457 		}
4458 		/*
4459 		 * Note the wait time value of 100ms is same as in the
4460 		 * dcd_poll routine.
4461 		 */
4462 		drv_usecwait(1000000);
4463 	}
4464 	(void) dcd_destroy_pkt(pkt);
4465 }
4466 
4467 static int
4468 dcd_send_lb_rw_cmd(dev_info_t *devi, void *bufaddr,
4469     diskaddr_t start_block, size_t reqlength, uchar_t cmd)
4470 {
4471 	struct dcd_pkt *pkt;
4472 	struct buf *bp;
4473 	diskaddr_t real_addr = start_block;
4474 	size_t buffer_size = reqlength;
4475 	uchar_t command, tmp;
4476 	int i, rval = 0;
4477 	struct dcd_disk *un;
4478 
4479 	un = ddi_get_soft_state(dcd_state, ddi_get_instance(devi));
4480 	if (un == NULL)
4481 		return (ENXIO);
4482 
4483 	bp = dcd_alloc_consistent_buf(ROUTE, (struct buf *)NULL,
4484 	    buffer_size, B_READ, NULL_FUNC, NULL);
4485 	if (!bp) {
4486 		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
4487 		    "no bp for disk label\n");
4488 		return (ENOMEM);
4489 	}
4490 
4491 	pkt = dcd_init_pkt(ROUTE, (struct dcd_pkt *)NULL,
4492 	    bp, (uint32_t)sizeof (struct dcd_cmd), 2, PP_LEN,
4493 	    PKT_CONSISTENT, NULL_FUNC, NULL);
4494 
4495 	if (!pkt) {
4496 		dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
4497 		    "no memory for disk label\n");
4498 		dcd_free_consistent_buf(bp);
4499 		return (ENOMEM);
4500 	}
4501 
4502 	if (cmd == TG_READ) {
4503 		bzero(bp->b_un.b_addr, buffer_size);
4504 		tmp = DATA_READ;
4505 	} else {
4506 		bcopy((caddr_t)bufaddr, bp->b_un.b_addr, buffer_size);
4507 		tmp = DATA_WRITE;
4508 	}
4509 
4510 	mutex_enter(DCD_MUTEX);
4511 	if ((un->un_dp->options & DMA_SUPPORTTED) == DMA_SUPPORTTED) {
4512 		if (cmd == TG_READ) {
4513 			command = ATA_READ_DMA;
4514 		} else {
4515 			command = ATA_WRITE_DMA;
4516 		}
4517 	} else {
4518 		if (cmd == TG_READ) {
4519 			if (un->un_dp->options & BLOCK_MODE)
4520 				command = ATA_READ_MULTIPLE;
4521 			else
4522 				command = ATA_READ;
4523 		} else {
4524 			if (un->un_dp->options & BLOCK_MODE)
4525 				command = ATA_READ_MULTIPLE;
4526 			else
4527 				command = ATA_WRITE;
4528 		}
4529 	}
4530 	mutex_exit(DCD_MUTEX);
4531 	(void) makecommand(pkt, 0, command, real_addr, ADD_LBA_MODE,
4532 	    buffer_size, tmp, 0);
4533 
4534 	for (i = 0; i < 3; i++) {
4535 		if (dcd_poll(pkt) || SCBP_C(pkt) != STATUS_GOOD ||
4536 		    (pkt->pkt_state & STATE_XFERRED_DATA) == 0 ||
4537 		    (pkt->pkt_resid != 0)) {
4538 			DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
4539 			    "Status %x, state %x, resid %lx\n",
4540 			    SCBP_C(pkt), pkt->pkt_state, pkt->pkt_resid);
4541 			rval = EIO;
4542 		} else {
4543 			break;
4544 		}
4545 	}
4546 
4547 	if (rval != 0) {
4548 		dcd_destroy_pkt(pkt);
4549 		dcd_free_consistent_buf(bp);
4550 		return (EIO);
4551 	}
4552 
4553 	if (cmd == TG_READ) {
4554 		bcopy(bp->b_un.b_addr, bufaddr, reqlength);
4555 		rval = 0;
4556 	}
4557 
4558 	dcd_destroy_pkt(pkt);
4559 	dcd_free_consistent_buf(bp);
4560 	return (rval);
4561 }
4562 
4563 static int dcd_compute_dk_capacity(struct dcd_device *devp,
4564     diskaddr_t *capacity)
4565 {
4566 	diskaddr_t cap;
4567 	diskaddr_t no_of_lbasec;
4568 
4569 	cap = devp->dcd_ident->dcd_fixcyls *
4570 	    devp->dcd_ident->dcd_heads *
4571 	    devp->dcd_ident->dcd_sectors;
4572 	no_of_lbasec = devp->dcd_ident->dcd_addrsec[1];
4573 	no_of_lbasec = no_of_lbasec << 16;
4574 	no_of_lbasec = no_of_lbasec | devp->dcd_ident->dcd_addrsec[0];
4575 
4576 	if (no_of_lbasec > cap) {
4577 		cap = no_of_lbasec;
4578 	}
4579 
4580 	if (cap != ((uint32_t)-1))
4581 		*capacity = cap;
4582 	else
4583 		return (EINVAL);
4584 	return (0);
4585 }
4586 
4587 /*ARGSUSED5*/
4588 static int
4589 dcd_lb_rdwr(dev_info_t *devi, uchar_t cmd, void *bufaddr,
4590     diskaddr_t start_block, size_t reqlength, void *tg_cookie)
4591 {
4592 	if (cmd != TG_READ && cmd != TG_WRITE)
4593 		return (EINVAL);
4594 
4595 	return (dcd_send_lb_rw_cmd(devi, bufaddr, start_block,
4596 	    reqlength, cmd));
4597 }
4598 
4599 static int
4600 dcd_lb_getphygeom(dev_info_t *devi, cmlb_geom_t *phygeomp)
4601 {
4602 	struct dcd_device *devp;
4603 	uint32_t no_of_lbasec, capacity, calculated_cylinders;
4604 
4605 	devp = ddi_get_driver_private(devi);
4606 
4607 	if ((devp->dcd_ident->dcd_config & ATAPI_DEVICE) == 0) {
4608 		if (devp->dcd_ident->dcd_config & ATANON_REMOVABLE) {
4609 			phygeomp->g_ncyl = devp->dcd_ident->dcd_fixcyls - 2;
4610 			phygeomp->g_acyl = 2;
4611 			phygeomp->g_nhead = devp->dcd_ident->dcd_heads;
4612 			phygeomp->g_nsect = devp->dcd_ident->dcd_sectors;
4613 
4614 			no_of_lbasec = devp->dcd_ident->dcd_addrsec[1];
4615 			no_of_lbasec = no_of_lbasec << 16;
4616 			no_of_lbasec = no_of_lbasec |
4617 			    devp->dcd_ident->dcd_addrsec[0];
4618 			capacity = devp->dcd_ident->dcd_fixcyls *
4619 			    devp->dcd_ident->dcd_heads *
4620 			    devp->dcd_ident->dcd_sectors;
4621 			if (no_of_lbasec > capacity) {
4622 				capacity = no_of_lbasec;
4623 				if (capacity > NUM_SECTORS_32G) {
4624 					/*
4625 					 * if the capacity is greater than 32G,
4626 					 * then 255 is the sectors per track.
4627 					 * This should be good until 128G disk
4628 					 * capacity, which is the current ATA-4
4629 					 * limitation.
4630 					 */
4631 					phygeomp->g_nsect = 255;
4632 				}
4633 
4634 				/*
4635 				 * If the disk capacity is >= 128GB then no. of
4636 				 * addressable sectors will be set to 0xfffffff
4637 				 * in the IDENTIFY info. In that case set the
4638 				 *  no. of pcyl to the Max. 16bit value.
4639 				 */
4640 
4641 				calculated_cylinders = (capacity) /
4642 				    (phygeomp->g_nhead * phygeomp->g_nsect);
4643 				if (calculated_cylinders >= USHRT_MAX) {
4644 					phygeomp->g_ncyl = USHRT_MAX - 2;
4645 				} else {
4646 					phygeomp->g_ncyl =
4647 					    calculated_cylinders - 2;
4648 				}
4649 			}
4650 
4651 			phygeomp->g_capacity = capacity;
4652 			phygeomp->g_intrlv = 0;
4653 			phygeomp->g_rpm = 5400;
4654 			phygeomp->g_secsize = devp->dcd_ident->dcd_secsiz;
4655 
4656 			return (0);
4657 		} else
4658 			return (ENOTSUP);
4659 	} else {
4660 		return (EINVAL);
4661 	}
4662 }
4663 
4664 
4665 /*ARGSUSED3*/
4666 static int
4667 dcd_lb_getinfo(dev_info_t *devi, int cmd,  void *arg, void *tg_cookie)
4668 {
4669 	struct dcd_disk *un;
4670 
4671 	un = ddi_get_soft_state(dcd_state, ddi_get_instance(devi));
4672 
4673 	if (un == NULL)
4674 		return (ENXIO);
4675 
4676 	switch (cmd) {
4677 	case TG_GETPHYGEOM:
4678 		return (dcd_lb_getphygeom(devi, (cmlb_geom_t *)arg));
4679 
4680 	case TG_GETVIRTGEOM:
4681 		return (-1);
4682 
4683 	case TG_GETCAPACITY:
4684 	case TG_GETBLOCKSIZE:
4685 		mutex_enter(DCD_MUTEX);
4686 		if (un->un_diskcapacity <= 0) {
4687 			mutex_exit(DCD_MUTEX);
4688 			dcd_log(DCD_DEVINFO, dcd_label, CE_WARN,
4689 			    "invalid disk capacity\n");
4690 			return (EIO);
4691 		}
4692 		if (cmd == TG_GETCAPACITY)
4693 			*(diskaddr_t *)arg = un->un_diskcapacity;
4694 		else
4695 			*(uint32_t *)arg = DEV_BSIZE;
4696 
4697 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG, "capacity %x\n",
4698 		    un->un_diskcapacity);
4699 		mutex_exit(DCD_MUTEX);
4700 		return (0);
4701 
4702 	case TG_GETATTR:
4703 		mutex_enter(DCD_MUTEX);
4704 		*(tg_attribute_t *)arg = un->un_tgattribute;
4705 		DAD_DEBUG2(DCD_DEVINFO, dcd_label, DCD_DEBUG,
4706 		    "media_is_writable %x\n",
4707 		    un->un_tgattribute.media_is_writable);
4708 		mutex_exit(DCD_MUTEX);
4709 		return (0);
4710 	default:
4711 		return (ENOTTY);
4712 	}
4713 }
4714