xref: /illumos-gate/usr/src/uts/common/io/blkdev/blkdev.c (revision e00bdde3c6d406f40f53f3025defadc22f7ec31a)
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  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2012 Garrett D'Amore <garrett@damore.org>.  All rights reserved.
24  * Copyright 2012 Alexey Zaytsev <alexey.zaytsev@gmail.com> All rights reserved.
25  * Copyright 2017 The MathWorks, Inc.  All rights reserved.
26  * Copyright 2020 Joyent, Inc.
27  * Copyright 2022 OmniOS Community Edition (OmniOSce) Association.
28  * Copyright 2022 Tintri by DDN, Inc. All rights reserved.
29  * Copyright 2023 Oxide Computer Company
30  */
31 
32 #include <sys/types.h>
33 #include <sys/ksynch.h>
34 #include <sys/kmem.h>
35 #include <sys/file.h>
36 #include <sys/errno.h>
37 #include <sys/open.h>
38 #include <sys/buf.h>
39 #include <sys/uio.h>
40 #include <sys/aio_req.h>
41 #include <sys/cred.h>
42 #include <sys/modctl.h>
43 #include <sys/cmlb.h>
44 #include <sys/conf.h>
45 #include <sys/devops.h>
46 #include <sys/list.h>
47 #include <sys/sysmacros.h>
48 #include <sys/dkio.h>
49 #include <sys/dkioc_free_util.h>
50 #include <sys/vtoc.h>
51 #include <sys/scsi/scsi.h>	/* for DTYPE_DIRECT */
52 #include <sys/kstat.h>
53 #include <sys/fs/dv_node.h>
54 #include <sys/ddi.h>
55 #include <sys/sunddi.h>
56 #include <sys/note.h>
57 #include <sys/blkdev.h>
58 #include <sys/scsi/impl/inquiry.h>
59 #include <sys/taskq.h>
60 #include <sys/taskq_impl.h>
61 #include <sys/disp.h>
62 #include <sys/sysevent/eventdefs.h>
63 #include <sys/sysevent/dev.h>
64 
65 /*
66  * blkdev is a driver which provides a lot of the common functionality
67  * a block device driver may need and helps by removing code which
68  * is frequently duplicated in block device drivers.
69  *
70  * Within this driver all the struct cb_ops functions required for a
71  * block device driver are written with appropriate call back functions
72  * to be provided by the parent driver.
73  *
74  * To use blkdev, a driver needs to:
75  *	1. Create a bd_ops_t structure which has the call back operations
76  *	   blkdev will use.
77  *	2. Create a handle by calling bd_alloc_handle(). One of the
78  *	   arguments to this function is the bd_ops_t.
79  *	3. Call bd_attach_handle(). This will instantiate a blkdev device
80  *	   as a child device node of the calling driver.
81  *
82  * A parent driver is not restricted to just allocating and attaching a
83  * single instance, it may attach as many as it wishes. For each handle
84  * attached, appropriate entries in /dev/[r]dsk are created.
85  *
86  * The bd_ops_t routines that a parent of blkdev need to provide are:
87  *
88  * o_drive_info: Provide information to blkdev such as how many I/O queues
89  *		 to create and the size of those queues. Also some device
90  *		 specifics such as EUI, vendor, product, model, serial
91  *		 number ....
92  *
93  * o_media_info: Provide information about the media. Eg size and block size.
94  *
95  * o_devid_init: Creates and initializes the device id. Typically calls
96  *		 ddi_devid_init().
97  *
98  * o_sync_cache: Issues a device appropriate command to flush any write
99  *		 caches.
100  *
101  * o_read:	 Read data as described by bd_xfer_t argument.
102  *
103  * o_write:	 Write data as described by bd_xfer_t argument.
104  *
105  * o_free_space: Free the space described by bd_xfer_t argument (optional).
106  *
107  * Queues
108  * ------
109  * Part of the drive_info data is a queue count. blkdev will create
110  * "queue count" number of waitq/runq pairs. Each waitq/runq pair
111  * operates independently. As an I/O is scheduled up to the parent
112  * driver via o_read or o_write its queue number is given. If the
113  * parent driver supports multiple hardware queues it can then select
114  * where to submit the I/O request.
115  *
116  * Currently blkdev uses a simplistic round-robin queue selection method.
117  * It has the advantage that it is lockless. In the future it will be
118  * worthwhile reviewing this strategy for something which prioritizes queues
119  * depending on how busy they are.
120  *
121  * Each waitq/runq pair is protected by its mutex (q_iomutex). Incoming
122  * I/O requests are initially added to the waitq. They are taken off the
123  * waitq, added to the runq and submitted, providing the runq is less
124  * than the qsize as specified in the drive_info. As an I/O request
125  * completes, the parent driver is required to call bd_xfer_done(), which
126  * will remove the I/O request from the runq and pass I/O completion
127  * status up the stack.
128  *
129  * Locks
130  * -----
131  * There are 5 instance global locks d_ocmutex, d_ksmutex, d_errmutex,
132  * d_statemutex and d_dle_mutex. As well a q_iomutex per waitq/runq pair.
133  *
134  * Lock Hierarchy
135  * --------------
136  * The only two locks which may be held simultaneously are q_iomutex and
137  * d_ksmutex. In all cases q_iomutex must be acquired before d_ksmutex.
138  */
139 
140 #define	BD_MAXPART	64
141 #define	BDINST(dev)	(getminor(dev) / BD_MAXPART)
142 #define	BDPART(dev)	(getminor(dev) % BD_MAXPART)
143 
144 typedef struct bd bd_t;
145 typedef struct bd_xfer_impl bd_xfer_impl_t;
146 typedef struct bd_queue bd_queue_t;
147 
148 typedef enum {
149 	BD_DLE_PENDING	= 1 << 0,
150 	BD_DLE_RUNNING	= 1 << 1
151 } bd_dle_state_t;
152 
153 struct bd {
154 	void		*d_private;
155 	dev_info_t	*d_dip;
156 	kmutex_t	d_ocmutex;	/* open/close */
157 	kmutex_t	d_ksmutex;	/* kstat */
158 	kmutex_t	d_errmutex;
159 	kmutex_t	d_statemutex;
160 	kcondvar_t	d_statecv;
161 	enum dkio_state	d_state;
162 	cmlb_handle_t	d_cmlbh;
163 	unsigned	d_open_lyr[BD_MAXPART];	/* open count */
164 	uint64_t	d_open_excl;	/* bit mask indexed by partition */
165 	uint64_t	d_open_reg[OTYPCNT];		/* bit mask */
166 	uint64_t	d_io_counter;
167 
168 	uint32_t	d_qcount;
169 	uint32_t	d_qactive;
170 	uint32_t	d_maxxfer;
171 	uint32_t	d_blkshift;
172 	uint32_t	d_pblkshift;
173 	uint64_t	d_numblks;
174 	ddi_devid_t	d_devid;
175 
176 	uint64_t	d_max_free_seg;
177 	uint64_t	d_max_free_blks;
178 	uint64_t	d_max_free_seg_blks;
179 	uint64_t	d_free_align;
180 
181 	kmem_cache_t	*d_cache;
182 	bd_queue_t	*d_queues;
183 	kstat_t		*d_ksp;
184 	kstat_io_t	*d_kiop;
185 	kstat_t		*d_errstats;
186 	struct bd_errstats *d_kerr;
187 
188 	boolean_t	d_rdonly;
189 	boolean_t	d_ssd;
190 	boolean_t	d_removable;
191 	boolean_t	d_hotpluggable;
192 	boolean_t	d_use_dma;
193 
194 	ddi_dma_attr_t	d_dma;
195 	bd_ops_t	d_ops;
196 	bd_handle_t	d_handle;
197 
198 	kmutex_t	d_dle_mutex;
199 	taskq_ent_t	d_dle_ent;
200 	bd_dle_state_t	d_dle_state;
201 };
202 
203 struct bd_handle {
204 	bd_ops_t	h_ops;
205 	ddi_dma_attr_t	*h_dma;
206 	dev_info_t	*h_parent;
207 	dev_info_t	*h_child;
208 	void		*h_private;
209 	bd_t		*h_bd;
210 	char		*h_name;
211 	char		h_addr[50];	/* enough for w%0.32x,%X */
212 };
213 
214 struct bd_xfer_impl {
215 	bd_xfer_t	i_public;
216 	list_node_t	i_linkage;
217 	bd_t		*i_bd;
218 	buf_t		*i_bp;
219 	bd_queue_t	*i_bq;
220 	uint_t		i_num_win;
221 	uint_t		i_cur_win;
222 	off_t		i_offset;
223 	int		(*i_func)(void *, bd_xfer_t *);
224 	uint32_t	i_blkshift;
225 	size_t		i_len;
226 	size_t		i_resid;
227 };
228 
229 struct bd_queue {
230 	kmutex_t	q_iomutex;
231 	uint32_t	q_qsize;
232 	uint32_t	q_qactive;
233 	list_t		q_runq;
234 	list_t		q_waitq;
235 };
236 
237 #define	i_dmah		i_public.x_dmah
238 #define	i_dmac		i_public.x_dmac
239 #define	i_ndmac		i_public.x_ndmac
240 #define	i_kaddr		i_public.x_kaddr
241 #define	i_nblks		i_public.x_nblks
242 #define	i_blkno		i_public.x_blkno
243 #define	i_flags		i_public.x_flags
244 #define	i_qnum		i_public.x_qnum
245 #define	i_dfl		i_public.x_dfl
246 
247 #define	CAN_FREESPACE(bd) \
248 	(((bd)->d_ops.o_free_space == NULL) ? B_FALSE : B_TRUE)
249 
250 /*
251  * Private prototypes.
252  */
253 
254 static void bd_prop_update_inqstring(dev_info_t *, char *, char *, size_t);
255 static void bd_create_inquiry_props(dev_info_t *, bd_drive_t *);
256 static void bd_create_errstats(bd_t *, int, bd_drive_t *);
257 static void bd_destroy_errstats(bd_t *);
258 static void bd_errstats_setstr(kstat_named_t *, char *, size_t, char *);
259 static void bd_init_errstats(bd_t *, bd_drive_t *);
260 static void bd_fini_errstats(bd_t *);
261 
262 static int bd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
263 static int bd_attach(dev_info_t *, ddi_attach_cmd_t);
264 static int bd_detach(dev_info_t *, ddi_detach_cmd_t);
265 
266 static int bd_open(dev_t *, int, int, cred_t *);
267 static int bd_close(dev_t, int, int, cred_t *);
268 static int bd_strategy(struct buf *);
269 static int bd_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
270 static int bd_dump(dev_t, caddr_t, daddr_t, int);
271 static int bd_read(dev_t, struct uio *, cred_t *);
272 static int bd_write(dev_t, struct uio *, cred_t *);
273 static int bd_aread(dev_t, struct aio_req *, cred_t *);
274 static int bd_awrite(dev_t, struct aio_req *, cred_t *);
275 static int bd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *,
276     caddr_t, int *);
277 
278 static int bd_tg_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t,
279     void *);
280 static int bd_tg_getinfo(dev_info_t *, int, void *, void *);
281 static int bd_xfer_ctor(void *, void *, int);
282 static void bd_xfer_dtor(void *, void *);
283 static void bd_sched(bd_t *, bd_queue_t *);
284 static void bd_submit(bd_t *, bd_xfer_impl_t *);
285 static void bd_runq_exit(bd_xfer_impl_t *, int);
286 static void bd_update_state(bd_t *);
287 static int bd_check_state(bd_t *, enum dkio_state *);
288 static int bd_flush_write_cache(bd_t *, struct dk_callback *);
289 static int bd_check_uio(dev_t, struct uio *);
290 static int bd_free_space(dev_t, bd_t *, dkioc_free_list_t *);
291 
292 struct cmlb_tg_ops bd_tg_ops = {
293 	TG_DK_OPS_VERSION_1,
294 	bd_tg_rdwr,
295 	bd_tg_getinfo,
296 };
297 
298 static struct cb_ops bd_cb_ops = {
299 	bd_open,		/* open */
300 	bd_close,		/* close */
301 	bd_strategy,		/* strategy */
302 	nodev,			/* print */
303 	bd_dump,		/* dump */
304 	bd_read,		/* read */
305 	bd_write,		/* write */
306 	bd_ioctl,		/* ioctl */
307 	nodev,			/* devmap */
308 	nodev,			/* mmap */
309 	nodev,			/* segmap */
310 	nochpoll,		/* poll */
311 	bd_prop_op,		/* cb_prop_op */
312 	0,			/* streamtab  */
313 	D_64BIT | D_MP,		/* Driver comaptibility flag */
314 	CB_REV,			/* cb_rev */
315 	bd_aread,		/* async read */
316 	bd_awrite		/* async write */
317 };
318 
319 struct dev_ops bd_dev_ops = {
320 	DEVO_REV,		/* devo_rev, */
321 	0,			/* refcnt  */
322 	bd_getinfo,		/* getinfo */
323 	nulldev,		/* identify */
324 	nulldev,		/* probe */
325 	bd_attach,		/* attach */
326 	bd_detach,		/* detach */
327 	nodev,			/* reset */
328 	&bd_cb_ops,		/* driver operations */
329 	NULL,			/* bus operations */
330 	NULL,			/* power */
331 	ddi_quiesce_not_needed,	/* quiesce */
332 };
333 
334 static struct modldrv modldrv = {
335 	&mod_driverops,
336 	"Generic Block Device",
337 	&bd_dev_ops,
338 };
339 
340 static struct modlinkage modlinkage = {
341 	MODREV_1, { &modldrv, NULL }
342 };
343 
344 static void *bd_state;
345 static krwlock_t bd_lock;
346 static taskq_t *bd_taskq;
347 
348 int
349 _init(void)
350 {
351 	char taskq_name[TASKQ_NAMELEN];
352 	const char *name;
353 	int rv;
354 
355 	rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2);
356 	if (rv != DDI_SUCCESS)
357 		return (rv);
358 
359 	name = mod_modname(&modlinkage);
360 	(void) snprintf(taskq_name, sizeof (taskq_name), "%s_taskq", name);
361 	bd_taskq = taskq_create(taskq_name, 1, minclsyspri, 0, 0, 0);
362 	if (bd_taskq == NULL) {
363 		cmn_err(CE_WARN, "%s: unable to create %s", name, taskq_name);
364 		ddi_soft_state_fini(&bd_state);
365 		return (DDI_FAILURE);
366 	}
367 
368 	rw_init(&bd_lock, NULL, RW_DRIVER, NULL);
369 
370 	rv = mod_install(&modlinkage);
371 	if (rv != DDI_SUCCESS) {
372 		rw_destroy(&bd_lock);
373 		taskq_destroy(bd_taskq);
374 		ddi_soft_state_fini(&bd_state);
375 	}
376 	return (rv);
377 }
378 
379 int
380 _fini(void)
381 {
382 	int	rv;
383 
384 	rv = mod_remove(&modlinkage);
385 	if (rv == DDI_SUCCESS) {
386 		rw_destroy(&bd_lock);
387 		taskq_destroy(bd_taskq);
388 		ddi_soft_state_fini(&bd_state);
389 	}
390 	return (rv);
391 }
392 
393 int
394 _info(struct modinfo *modinfop)
395 {
396 	return (mod_info(&modlinkage, modinfop));
397 }
398 
399 static int
400 bd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resultp)
401 {
402 	bd_t	*bd;
403 	minor_t	inst;
404 
405 	_NOTE(ARGUNUSED(dip));
406 
407 	inst = BDINST((dev_t)arg);
408 
409 	switch (cmd) {
410 	case DDI_INFO_DEVT2DEVINFO:
411 		bd = ddi_get_soft_state(bd_state, inst);
412 		if (bd == NULL) {
413 			return (DDI_FAILURE);
414 		}
415 		*resultp = (void *)bd->d_dip;
416 		break;
417 
418 	case DDI_INFO_DEVT2INSTANCE:
419 		*resultp = (void *)(intptr_t)inst;
420 		break;
421 
422 	default:
423 		return (DDI_FAILURE);
424 	}
425 	return (DDI_SUCCESS);
426 }
427 
428 static void
429 bd_prop_update_inqstring(dev_info_t *dip, char *name, char *data, size_t len)
430 {
431 	int	ilen;
432 	char	*data_string;
433 
434 	ilen = scsi_ascii_inquiry_len(data, len);
435 	ASSERT3U(ilen, <=, len);
436 	if (ilen <= 0)
437 		return;
438 	/* ensure null termination */
439 	data_string = kmem_zalloc(ilen + 1, KM_SLEEP);
440 	bcopy(data, data_string, ilen);
441 	(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip, name, data_string);
442 	kmem_free(data_string, ilen + 1);
443 }
444 
445 static void
446 bd_create_inquiry_props(dev_info_t *dip, bd_drive_t *drive)
447 {
448 	if (drive->d_vendor_len > 0)
449 		bd_prop_update_inqstring(dip, INQUIRY_VENDOR_ID,
450 		    drive->d_vendor, drive->d_vendor_len);
451 
452 	if (drive->d_product_len > 0)
453 		bd_prop_update_inqstring(dip, INQUIRY_PRODUCT_ID,
454 		    drive->d_product, drive->d_product_len);
455 
456 	if (drive->d_serial_len > 0)
457 		bd_prop_update_inqstring(dip, INQUIRY_SERIAL_NO,
458 		    drive->d_serial, drive->d_serial_len);
459 
460 	if (drive->d_revision_len > 0)
461 		bd_prop_update_inqstring(dip, INQUIRY_REVISION_ID,
462 		    drive->d_revision, drive->d_revision_len);
463 }
464 
465 static void
466 bd_create_errstats(bd_t *bd, int inst, bd_drive_t *drive)
467 {
468 	char	ks_module[KSTAT_STRLEN];
469 	char	ks_name[KSTAT_STRLEN];
470 	int	ndata = sizeof (struct bd_errstats) / sizeof (kstat_named_t);
471 
472 	if (bd->d_errstats != NULL)
473 		return;
474 
475 	(void) snprintf(ks_module, sizeof (ks_module), "%serr",
476 	    ddi_driver_name(bd->d_dip));
477 	(void) snprintf(ks_name, sizeof (ks_name), "%s%d,err",
478 	    ddi_driver_name(bd->d_dip), inst);
479 
480 	bd->d_errstats = kstat_create(ks_module, inst, ks_name, "device_error",
481 	    KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
482 
483 	mutex_init(&bd->d_errmutex, NULL, MUTEX_DRIVER, NULL);
484 	if (bd->d_errstats == NULL) {
485 		/*
486 		 * Even if we cannot create the kstat, we create a
487 		 * scratch kstat.  The reason for this is to ensure
488 		 * that we can update the kstat all of the time,
489 		 * without adding an extra branch instruction.
490 		 */
491 		bd->d_kerr = kmem_zalloc(sizeof (struct bd_errstats),
492 		    KM_SLEEP);
493 	} else {
494 		bd->d_errstats->ks_lock = &bd->d_errmutex;
495 		bd->d_kerr = (struct bd_errstats *)bd->d_errstats->ks_data;
496 	}
497 
498 	kstat_named_init(&bd->d_kerr->bd_softerrs,	"Soft Errors",
499 	    KSTAT_DATA_UINT32);
500 	kstat_named_init(&bd->d_kerr->bd_harderrs,	"Hard Errors",
501 	    KSTAT_DATA_UINT32);
502 	kstat_named_init(&bd->d_kerr->bd_transerrs,	"Transport Errors",
503 	    KSTAT_DATA_UINT32);
504 
505 	if (drive->d_model_len > 0) {
506 		kstat_named_init(&bd->d_kerr->bd_model,	"Model",
507 		    KSTAT_DATA_STRING);
508 	} else {
509 		kstat_named_init(&bd->d_kerr->bd_vid,	"Vendor",
510 		    KSTAT_DATA_STRING);
511 		kstat_named_init(&bd->d_kerr->bd_pid,	"Product",
512 		    KSTAT_DATA_STRING);
513 	}
514 
515 	kstat_named_init(&bd->d_kerr->bd_revision,	"Revision",
516 	    KSTAT_DATA_STRING);
517 	kstat_named_init(&bd->d_kerr->bd_serial,	"Serial No",
518 	    KSTAT_DATA_STRING);
519 	kstat_named_init(&bd->d_kerr->bd_capacity,	"Size",
520 	    KSTAT_DATA_ULONGLONG);
521 	kstat_named_init(&bd->d_kerr->bd_rq_media_err,	"Media Error",
522 	    KSTAT_DATA_UINT32);
523 	kstat_named_init(&bd->d_kerr->bd_rq_ntrdy_err,	"Device Not Ready",
524 	    KSTAT_DATA_UINT32);
525 	kstat_named_init(&bd->d_kerr->bd_rq_nodev_err,	"No Device",
526 	    KSTAT_DATA_UINT32);
527 	kstat_named_init(&bd->d_kerr->bd_rq_recov_err,	"Recoverable",
528 	    KSTAT_DATA_UINT32);
529 	kstat_named_init(&bd->d_kerr->bd_rq_illrq_err,	"Illegal Request",
530 	    KSTAT_DATA_UINT32);
531 	kstat_named_init(&bd->d_kerr->bd_rq_pfa_err,
532 	    "Predictive Failure Analysis", KSTAT_DATA_UINT32);
533 
534 	bd->d_errstats->ks_private = bd;
535 
536 	kstat_install(bd->d_errstats);
537 	bd_init_errstats(bd, drive);
538 }
539 
540 static void
541 bd_destroy_errstats(bd_t *bd)
542 {
543 	if (bd->d_errstats != NULL) {
544 		bd_fini_errstats(bd);
545 		kstat_delete(bd->d_errstats);
546 		bd->d_errstats = NULL;
547 	} else {
548 		kmem_free(bd->d_kerr, sizeof (struct bd_errstats));
549 		bd->d_kerr = NULL;
550 		mutex_destroy(&bd->d_errmutex);
551 	}
552 }
553 
554 static void
555 bd_errstats_setstr(kstat_named_t *k, char *str, size_t len, char *alt)
556 {
557 	char	*tmp;
558 	size_t	km_len;
559 
560 	if (KSTAT_NAMED_STR_PTR(k) == NULL) {
561 		if (len > 0)
562 			km_len = strnlen(str, len);
563 		else if (alt != NULL)
564 			km_len = strlen(alt);
565 		else
566 			return;
567 
568 		tmp = kmem_alloc(km_len + 1, KM_SLEEP);
569 		bcopy(len > 0 ? str : alt, tmp, km_len);
570 		tmp[km_len] = '\0';
571 
572 		kstat_named_setstr(k, tmp);
573 	}
574 }
575 
576 static void
577 bd_errstats_clrstr(kstat_named_t *k)
578 {
579 	if (KSTAT_NAMED_STR_PTR(k) == NULL)
580 		return;
581 
582 	kmem_free(KSTAT_NAMED_STR_PTR(k), KSTAT_NAMED_STR_BUFLEN(k));
583 	kstat_named_setstr(k, NULL);
584 }
585 
586 static void
587 bd_init_errstats(bd_t *bd, bd_drive_t *drive)
588 {
589 	struct bd_errstats	*est = bd->d_kerr;
590 
591 	mutex_enter(&bd->d_errmutex);
592 
593 	if (drive->d_model_len > 0 &&
594 	    KSTAT_NAMED_STR_PTR(&est->bd_model) == NULL) {
595 		bd_errstats_setstr(&est->bd_model, drive->d_model,
596 		    drive->d_model_len, NULL);
597 	} else {
598 		bd_errstats_setstr(&est->bd_vid, drive->d_vendor,
599 		    drive->d_vendor_len, "Unknown ");
600 		bd_errstats_setstr(&est->bd_pid, drive->d_product,
601 		    drive->d_product_len, "Unknown         ");
602 	}
603 
604 	bd_errstats_setstr(&est->bd_revision, drive->d_revision,
605 	    drive->d_revision_len, "0001");
606 	bd_errstats_setstr(&est->bd_serial, drive->d_serial,
607 	    drive->d_serial_len, "0               ");
608 
609 	mutex_exit(&bd->d_errmutex);
610 }
611 
612 static void
613 bd_fini_errstats(bd_t *bd)
614 {
615 	struct bd_errstats	*est = bd->d_kerr;
616 
617 	mutex_enter(&bd->d_errmutex);
618 
619 	bd_errstats_clrstr(&est->bd_model);
620 	bd_errstats_clrstr(&est->bd_vid);
621 	bd_errstats_clrstr(&est->bd_pid);
622 	bd_errstats_clrstr(&est->bd_revision);
623 	bd_errstats_clrstr(&est->bd_serial);
624 
625 	mutex_exit(&bd->d_errmutex);
626 }
627 
628 static void
629 bd_queues_free(bd_t *bd)
630 {
631 	uint32_t i;
632 
633 	for (i = 0; i < bd->d_qcount; i++) {
634 		bd_queue_t *bq = &bd->d_queues[i];
635 
636 		mutex_destroy(&bq->q_iomutex);
637 		list_destroy(&bq->q_waitq);
638 		list_destroy(&bq->q_runq);
639 	}
640 
641 	kmem_free(bd->d_queues, sizeof (*bd->d_queues) * bd->d_qcount);
642 }
643 
644 static int
645 bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
646 {
647 	int		inst;
648 	bd_handle_t	hdl;
649 	bd_t		*bd;
650 	bd_drive_t	drive;
651 	uint32_t	i;
652 	int		rv;
653 	char		name[16];
654 	char		kcache[32];
655 	char		*node_type;
656 
657 	switch (cmd) {
658 	case DDI_ATTACH:
659 		break;
660 	case DDI_RESUME:
661 		/* We don't do anything native for suspend/resume */
662 		return (DDI_SUCCESS);
663 	default:
664 		return (DDI_FAILURE);
665 	}
666 
667 	inst = ddi_get_instance(dip);
668 	hdl = ddi_get_parent_data(dip);
669 
670 	(void) snprintf(name, sizeof (name), "%s%d",
671 	    ddi_driver_name(dip), ddi_get_instance(dip));
672 	(void) snprintf(kcache, sizeof (kcache), "%s_xfer", name);
673 
674 	if (hdl == NULL) {
675 		cmn_err(CE_WARN, "%s: missing parent data!", name);
676 		return (DDI_FAILURE);
677 	}
678 
679 	if (ddi_soft_state_zalloc(bd_state, inst) != DDI_SUCCESS) {
680 		cmn_err(CE_WARN, "%s: unable to zalloc soft state!", name);
681 		return (DDI_FAILURE);
682 	}
683 	bd = ddi_get_soft_state(bd_state, inst);
684 
685 	if (hdl->h_dma) {
686 		bd->d_dma = *(hdl->h_dma);
687 		bd->d_dma.dma_attr_granular =
688 		    max(DEV_BSIZE, bd->d_dma.dma_attr_granular);
689 		bd->d_use_dma = B_TRUE;
690 
691 		if (bd->d_maxxfer &&
692 		    (bd->d_maxxfer != bd->d_dma.dma_attr_maxxfer)) {
693 			cmn_err(CE_WARN,
694 			    "%s: inconsistent maximum transfer size!",
695 			    name);
696 			/* We force it */
697 			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
698 		} else {
699 			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
700 		}
701 	} else {
702 		bd->d_use_dma = B_FALSE;
703 		if (bd->d_maxxfer == 0) {
704 			bd->d_maxxfer = 1024 * 1024;
705 		}
706 	}
707 	bd->d_ops = hdl->h_ops;
708 	bd->d_private = hdl->h_private;
709 	bd->d_blkshift = DEV_BSHIFT;	/* 512 bytes, to start */
710 
711 	if (bd->d_maxxfer % DEV_BSIZE) {
712 		cmn_err(CE_WARN, "%s: maximum transfer misaligned!", name);
713 		bd->d_maxxfer &= ~(DEV_BSIZE - 1);
714 	}
715 	if (bd->d_maxxfer < DEV_BSIZE) {
716 		cmn_err(CE_WARN, "%s: maximum transfer size too small!", name);
717 		ddi_soft_state_free(bd_state, inst);
718 		return (DDI_FAILURE);
719 	}
720 
721 	bd->d_dip = dip;
722 	bd->d_handle = hdl;
723 	ddi_set_driver_private(dip, bd);
724 
725 	mutex_init(&bd->d_ksmutex, NULL, MUTEX_DRIVER, NULL);
726 	mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL);
727 	mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL);
728 	cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL);
729 	mutex_init(&bd->d_dle_mutex, NULL, MUTEX_DRIVER, NULL);
730 	bd->d_dle_state = 0;
731 
732 	bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8,
733 	    bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0);
734 
735 	bd->d_ksp = kstat_create(ddi_driver_name(dip), inst, NULL, "disk",
736 	    KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
737 	if (bd->d_ksp != NULL) {
738 		bd->d_ksp->ks_lock = &bd->d_ksmutex;
739 		kstat_install(bd->d_ksp);
740 		bd->d_kiop = bd->d_ksp->ks_data;
741 	} else {
742 		/*
743 		 * Even if we cannot create the kstat, we create a
744 		 * scratch kstat.  The reason for this is to ensure
745 		 * that we can update the kstat all of the time,
746 		 * without adding an extra branch instruction.
747 		 */
748 		bd->d_kiop = kmem_zalloc(sizeof (kstat_io_t), KM_SLEEP);
749 	}
750 
751 	cmlb_alloc_handle(&bd->d_cmlbh);
752 
753 	bd->d_state = DKIO_NONE;
754 
755 	bzero(&drive, sizeof (drive));
756 	/*
757 	 * Default to one queue, and no restrictions on free space requests
758 	 * (if driver provides method) parent driver can override.
759 	 */
760 	drive.d_qcount = 1;
761 	drive.d_free_align = 1;
762 	bd->d_ops.o_drive_info(bd->d_private, &drive);
763 
764 	/*
765 	 * Several checks to make sure o_drive_info() didn't return bad
766 	 * values:
767 	 *
768 	 * There must be at least one queue
769 	 */
770 	if (drive.d_qcount == 0)
771 		goto fail_drive_info;
772 
773 	/* FREE/UNMAP/TRIM alignment needs to be at least 1 block */
774 	if (drive.d_free_align == 0)
775 		goto fail_drive_info;
776 
777 	/*
778 	 * If d_max_free_blks is not unlimited (not 0), then we cannot allow
779 	 * an unlimited segment size. It is however permissible to not impose
780 	 * a limit on the total number of blocks freed while limiting the
781 	 * amount allowed in an individual segment.
782 	 */
783 	if ((drive.d_max_free_blks > 0 && drive.d_max_free_seg_blks == 0))
784 		goto fail_drive_info;
785 
786 	/*
787 	 * If a limit is set on d_max_free_blks (by the above check, we know
788 	 * if there's a limit on d_max_free_blks, d_max_free_seg_blks cannot
789 	 * be unlimited), it cannot be smaller than the limit on an individual
790 	 * segment.
791 	 */
792 	if ((drive.d_max_free_blks > 0 &&
793 	    drive.d_max_free_seg_blks > drive.d_max_free_blks)) {
794 		goto fail_drive_info;
795 	}
796 
797 	bd->d_qcount = drive.d_qcount;
798 	bd->d_removable = drive.d_removable;
799 	bd->d_hotpluggable = drive.d_hotpluggable;
800 
801 	if (drive.d_maxxfer && drive.d_maxxfer < bd->d_maxxfer)
802 		bd->d_maxxfer = drive.d_maxxfer;
803 
804 	bd->d_free_align = drive.d_free_align;
805 	bd->d_max_free_seg = drive.d_max_free_seg;
806 	bd->d_max_free_blks = drive.d_max_free_blks;
807 	bd->d_max_free_seg_blks = drive.d_max_free_seg_blks;
808 
809 	bd_create_inquiry_props(dip, &drive);
810 	bd_create_errstats(bd, inst, &drive);
811 	bd_update_state(bd);
812 
813 	bd->d_queues = kmem_alloc(sizeof (*bd->d_queues) * bd->d_qcount,
814 	    KM_SLEEP);
815 	for (i = 0; i < bd->d_qcount; i++) {
816 		bd_queue_t *bq = &bd->d_queues[i];
817 
818 		bq->q_qsize = drive.d_qsize;
819 		bq->q_qactive = 0;
820 		mutex_init(&bq->q_iomutex, NULL, MUTEX_DRIVER, NULL);
821 
822 		list_create(&bq->q_waitq, sizeof (bd_xfer_impl_t),
823 		    offsetof(struct bd_xfer_impl, i_linkage));
824 		list_create(&bq->q_runq, sizeof (bd_xfer_impl_t),
825 		    offsetof(struct bd_xfer_impl, i_linkage));
826 	}
827 
828 	if (*(uint64_t *)drive.d_eui64 != 0 ||
829 	    *(uint64_t *)drive.d_guid != 0 ||
830 	    *((uint64_t *)drive.d_guid + 1) != 0)
831 		node_type = DDI_NT_BLOCK_BLKDEV;
832 	else if (drive.d_lun >= 0)
833 		node_type = DDI_NT_BLOCK_CHAN;
834 	else
835 		node_type = DDI_NT_BLOCK;
836 
837 	rv = cmlb_attach(dip, &bd_tg_ops, DTYPE_DIRECT,
838 	    bd->d_removable, bd->d_hotpluggable, node_type,
839 	    CMLB_FAKE_LABEL_ONE_PARTITION, bd->d_cmlbh, 0);
840 	if (rv != 0) {
841 		goto fail_cmlb_attach;
842 	}
843 
844 	if (bd->d_ops.o_devid_init != NULL) {
845 		rv = bd->d_ops.o_devid_init(bd->d_private, dip, &bd->d_devid);
846 		if (rv == DDI_SUCCESS) {
847 			if (ddi_devid_register(dip, bd->d_devid) !=
848 			    DDI_SUCCESS) {
849 				cmn_err(CE_WARN,
850 				    "%s: unable to register devid", name);
851 			}
852 		}
853 	}
854 
855 	/*
856 	 * Add a zero-length attribute to tell the world we support
857 	 * kernel ioctls (for layered drivers).  Also set up properties
858 	 * used by HAL to identify removable media.
859 	 */
860 	(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
861 	    DDI_KERNEL_IOCTL, NULL, 0);
862 	if (bd->d_removable) {
863 		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
864 		    "removable-media", NULL, 0);
865 	}
866 	if (bd->d_hotpluggable) {
867 		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
868 		    "hotpluggable", NULL, 0);
869 	}
870 
871 	/*
872 	 * Before we proceed, we need to ensure that the geometry and labels on
873 	 * the cmlb disk are reasonable. When cmlb first attaches, it does not
874 	 * perform label validation and creates minor nodes based on the
875 	 * assumption of the size. This may not be correct and the rest of the
876 	 * system assumes that this will have been done before we allow opens
877 	 * to proceed. Otherwise, on first open, this'll all end up changing
878 	 * around on users. We do not care if it succeeds or not. It is totally
879 	 * acceptable for this device to be unlabeled or not to have anything on
880 	 * it.
881 	 */
882 	(void) cmlb_validate(bd->d_cmlbh, 0, 0);
883 
884 	hdl->h_bd = bd;
885 	ddi_report_dev(dip);
886 
887 	return (DDI_SUCCESS);
888 
889 fail_cmlb_attach:
890 	bd_queues_free(bd);
891 	bd_destroy_errstats(bd);
892 
893 fail_drive_info:
894 	cmlb_free_handle(&bd->d_cmlbh);
895 
896 	if (bd->d_ksp != NULL) {
897 		kstat_delete(bd->d_ksp);
898 		bd->d_ksp = NULL;
899 	} else {
900 		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
901 	}
902 
903 	kmem_cache_destroy(bd->d_cache);
904 	cv_destroy(&bd->d_statecv);
905 	mutex_destroy(&bd->d_statemutex);
906 	mutex_destroy(&bd->d_ocmutex);
907 	mutex_destroy(&bd->d_ksmutex);
908 	mutex_destroy(&bd->d_dle_mutex);
909 	ddi_soft_state_free(bd_state, inst);
910 	return (DDI_FAILURE);
911 }
912 
913 static int
914 bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
915 {
916 	bd_handle_t	hdl;
917 	bd_t		*bd;
918 
919 	bd = ddi_get_driver_private(dip);
920 	hdl = ddi_get_parent_data(dip);
921 
922 	switch (cmd) {
923 	case DDI_DETACH:
924 		break;
925 	case DDI_SUSPEND:
926 		/* We don't suspend, but our parent does */
927 		return (DDI_SUCCESS);
928 	default:
929 		return (DDI_FAILURE);
930 	}
931 
932 	hdl->h_bd = NULL;
933 
934 	if (bd->d_ksp != NULL) {
935 		kstat_delete(bd->d_ksp);
936 		bd->d_ksp = NULL;
937 	} else {
938 		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
939 	}
940 
941 	bd_destroy_errstats(bd);
942 	cmlb_detach(bd->d_cmlbh, 0);
943 	cmlb_free_handle(&bd->d_cmlbh);
944 	if (bd->d_devid)
945 		ddi_devid_free(bd->d_devid);
946 	kmem_cache_destroy(bd->d_cache);
947 	mutex_destroy(&bd->d_ksmutex);
948 	mutex_destroy(&bd->d_ocmutex);
949 	mutex_destroy(&bd->d_statemutex);
950 	cv_destroy(&bd->d_statecv);
951 	mutex_destroy(&bd->d_dle_mutex);
952 	bd_queues_free(bd);
953 	ddi_soft_state_free(bd_state, ddi_get_instance(dip));
954 	return (DDI_SUCCESS);
955 }
956 
957 static int
958 bd_xfer_ctor(void *buf, void *arg, int kmflag)
959 {
960 	bd_xfer_impl_t	*xi;
961 	bd_t		*bd = arg;
962 	int		(*dcb)(caddr_t);
963 
964 	if (kmflag == KM_PUSHPAGE || kmflag == KM_SLEEP) {
965 		dcb = DDI_DMA_SLEEP;
966 	} else {
967 		dcb = DDI_DMA_DONTWAIT;
968 	}
969 
970 	xi = buf;
971 	bzero(xi, sizeof (*xi));
972 	xi->i_bd = bd;
973 
974 	if (bd->d_use_dma) {
975 		if (ddi_dma_alloc_handle(bd->d_dip, &bd->d_dma, dcb, NULL,
976 		    &xi->i_dmah) != DDI_SUCCESS) {
977 			return (-1);
978 		}
979 	}
980 
981 	return (0);
982 }
983 
984 static void
985 bd_xfer_dtor(void *buf, void *arg)
986 {
987 	bd_xfer_impl_t	*xi = buf;
988 
989 	_NOTE(ARGUNUSED(arg));
990 
991 	if (xi->i_dmah)
992 		ddi_dma_free_handle(&xi->i_dmah);
993 	xi->i_dmah = NULL;
994 }
995 
996 static bd_xfer_impl_t *
997 bd_xfer_alloc(bd_t *bd, struct buf *bp, int (*func)(void *, bd_xfer_t *),
998     int kmflag)
999 {
1000 	bd_xfer_impl_t		*xi;
1001 	int			rv = 0;
1002 	int			status;
1003 	unsigned		dir;
1004 	int			(*cb)(caddr_t);
1005 	size_t			len;
1006 	uint32_t		shift;
1007 
1008 	if (kmflag == KM_SLEEP) {
1009 		cb = DDI_DMA_SLEEP;
1010 	} else {
1011 		cb = DDI_DMA_DONTWAIT;
1012 	}
1013 
1014 	xi = kmem_cache_alloc(bd->d_cache, kmflag);
1015 	if (xi == NULL) {
1016 		bioerror(bp, ENOMEM);
1017 		return (NULL);
1018 	}
1019 
1020 	ASSERT(bp);
1021 
1022 	xi->i_bp = bp;
1023 	xi->i_func = func;
1024 	xi->i_blkno = bp->b_lblkno >> (bd->d_blkshift - DEV_BSHIFT);
1025 
1026 	if (bp->b_bcount == 0) {
1027 		xi->i_len = 0;
1028 		xi->i_nblks = 0;
1029 		xi->i_kaddr = NULL;
1030 		xi->i_resid = 0;
1031 		xi->i_num_win = 0;
1032 		goto done;
1033 	}
1034 
1035 	if (bp->b_flags & B_READ) {
1036 		dir = DDI_DMA_READ;
1037 		xi->i_func = bd->d_ops.o_read;
1038 	} else {
1039 		dir = DDI_DMA_WRITE;
1040 		xi->i_func = bd->d_ops.o_write;
1041 	}
1042 
1043 	shift = bd->d_blkshift;
1044 	xi->i_blkshift = shift;
1045 
1046 	if (!bd->d_use_dma) {
1047 		bp_mapin(bp);
1048 		rv = 0;
1049 		xi->i_offset = 0;
1050 		xi->i_num_win =
1051 		    (bp->b_bcount + (bd->d_maxxfer - 1)) / bd->d_maxxfer;
1052 		xi->i_cur_win = 0;
1053 		xi->i_len = min(bp->b_bcount, bd->d_maxxfer);
1054 		xi->i_nblks = xi->i_len >> shift;
1055 		xi->i_kaddr = bp->b_un.b_addr;
1056 		xi->i_resid = bp->b_bcount;
1057 	} else {
1058 
1059 		/*
1060 		 * We have to use consistent DMA if the address is misaligned.
1061 		 */
1062 		if (((bp->b_flags & (B_PAGEIO | B_REMAPPED)) != B_PAGEIO) &&
1063 		    ((uintptr_t)bp->b_un.b_addr & 0x7)) {
1064 			dir |= DDI_DMA_CONSISTENT | DDI_DMA_PARTIAL;
1065 		} else {
1066 			dir |= DDI_DMA_STREAMING | DDI_DMA_PARTIAL;
1067 		}
1068 
1069 		status = ddi_dma_buf_bind_handle(xi->i_dmah, bp, dir, cb,
1070 		    NULL, &xi->i_dmac, &xi->i_ndmac);
1071 		switch (status) {
1072 		case DDI_DMA_MAPPED:
1073 			xi->i_num_win = 1;
1074 			xi->i_cur_win = 0;
1075 			xi->i_offset = 0;
1076 			xi->i_len = bp->b_bcount;
1077 			xi->i_nblks = xi->i_len >> shift;
1078 			xi->i_resid = bp->b_bcount;
1079 			rv = 0;
1080 			break;
1081 		case DDI_DMA_PARTIAL_MAP:
1082 			xi->i_cur_win = 0;
1083 
1084 			if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) !=
1085 			    DDI_SUCCESS) ||
1086 			    (ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset,
1087 			    &len, &xi->i_dmac, &xi->i_ndmac) !=
1088 			    DDI_SUCCESS) ||
1089 			    (P2PHASE(len, (1U << shift)) != 0)) {
1090 				(void) ddi_dma_unbind_handle(xi->i_dmah);
1091 				rv = EFAULT;
1092 				goto done;
1093 			}
1094 			xi->i_len = len;
1095 			xi->i_nblks = xi->i_len >> shift;
1096 			xi->i_resid = bp->b_bcount;
1097 			rv = 0;
1098 			break;
1099 		case DDI_DMA_NORESOURCES:
1100 			rv = EAGAIN;
1101 			goto done;
1102 		case DDI_DMA_TOOBIG:
1103 			rv = EINVAL;
1104 			goto done;
1105 		case DDI_DMA_NOMAPPING:
1106 		case DDI_DMA_INUSE:
1107 		default:
1108 			rv = EFAULT;
1109 			goto done;
1110 		}
1111 	}
1112 
1113 done:
1114 	if (rv != 0) {
1115 		kmem_cache_free(bd->d_cache, xi);
1116 		bioerror(bp, rv);
1117 		return (NULL);
1118 	}
1119 
1120 	return (xi);
1121 }
1122 
1123 static void
1124 bd_xfer_free(bd_xfer_impl_t *xi)
1125 {
1126 	if (xi->i_dmah) {
1127 		(void) ddi_dma_unbind_handle(xi->i_dmah);
1128 	}
1129 	if (xi->i_dfl != NULL) {
1130 		dfl_free((dkioc_free_list_t *)xi->i_dfl);
1131 		xi->i_dfl = NULL;
1132 	}
1133 	kmem_cache_free(xi->i_bd->d_cache, xi);
1134 }
1135 
1136 static int
1137 bd_open(dev_t *devp, int flag, int otyp, cred_t *credp)
1138 {
1139 	dev_t		dev = *devp;
1140 	bd_t		*bd;
1141 	minor_t		part;
1142 	minor_t		inst;
1143 	uint64_t	mask;
1144 	boolean_t	ndelay;
1145 	int		rv;
1146 	diskaddr_t	nblks;
1147 	diskaddr_t	lba;
1148 
1149 	_NOTE(ARGUNUSED(credp));
1150 
1151 	part = BDPART(dev);
1152 	inst = BDINST(dev);
1153 
1154 	if (otyp >= OTYPCNT)
1155 		return (EINVAL);
1156 
1157 	ndelay = (flag & (FNDELAY | FNONBLOCK)) ? B_TRUE : B_FALSE;
1158 
1159 	/*
1160 	 * Block any DR events from changing the set of registered
1161 	 * devices while we function.
1162 	 */
1163 	rw_enter(&bd_lock, RW_READER);
1164 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1165 		rw_exit(&bd_lock);
1166 		return (ENXIO);
1167 	}
1168 
1169 	mutex_enter(&bd->d_ocmutex);
1170 
1171 	ASSERT(part < 64);
1172 	mask = (1U << part);
1173 
1174 	bd_update_state(bd);
1175 
1176 	if (cmlb_validate(bd->d_cmlbh, 0, 0) != 0) {
1177 
1178 		/* non-blocking opens are allowed to succeed */
1179 		if (!ndelay) {
1180 			rv = ENXIO;
1181 			goto done;
1182 		}
1183 	} else if (cmlb_partinfo(bd->d_cmlbh, part, &nblks, &lba,
1184 	    NULL, NULL, 0) == 0) {
1185 
1186 		/*
1187 		 * We read the partinfo, verify valid ranges.  If the
1188 		 * partition is invalid, and we aren't blocking or
1189 		 * doing a raw access, then fail. (Non-blocking and
1190 		 * raw accesses can still succeed to allow a disk with
1191 		 * bad partition data to opened by format and fdisk.)
1192 		 */
1193 		if ((!nblks) && ((!ndelay) || (otyp != OTYP_CHR))) {
1194 			rv = ENXIO;
1195 			goto done;
1196 		}
1197 	} else if (!ndelay) {
1198 		/*
1199 		 * cmlb_partinfo failed -- invalid partition or no
1200 		 * disk label.
1201 		 */
1202 		rv = ENXIO;
1203 		goto done;
1204 	}
1205 
1206 	if ((flag & FWRITE) && bd->d_rdonly) {
1207 		rv = EROFS;
1208 		goto done;
1209 	}
1210 
1211 	if ((bd->d_open_excl) & (mask)) {
1212 		rv = EBUSY;
1213 		goto done;
1214 	}
1215 	if (flag & FEXCL) {
1216 		if (bd->d_open_lyr[part]) {
1217 			rv = EBUSY;
1218 			goto done;
1219 		}
1220 		for (int i = 0; i < OTYP_LYR; i++) {
1221 			if (bd->d_open_reg[i] & mask) {
1222 				rv = EBUSY;
1223 				goto done;
1224 			}
1225 		}
1226 	}
1227 
1228 	if (otyp == OTYP_LYR) {
1229 		bd->d_open_lyr[part]++;
1230 	} else {
1231 		bd->d_open_reg[otyp] |= mask;
1232 	}
1233 	if (flag & FEXCL) {
1234 		bd->d_open_excl |= mask;
1235 	}
1236 
1237 	rv = 0;
1238 done:
1239 	mutex_exit(&bd->d_ocmutex);
1240 	rw_exit(&bd_lock);
1241 
1242 	return (rv);
1243 }
1244 
1245 static int
1246 bd_close(dev_t dev, int flag, int otyp, cred_t *credp)
1247 {
1248 	bd_t		*bd;
1249 	minor_t		inst;
1250 	minor_t		part;
1251 	uint64_t	mask;
1252 	boolean_t	last = B_TRUE;
1253 
1254 	_NOTE(ARGUNUSED(flag));
1255 	_NOTE(ARGUNUSED(credp));
1256 
1257 	part = BDPART(dev);
1258 	inst = BDINST(dev);
1259 
1260 	ASSERT(part < 64);
1261 	mask = (1U << part);
1262 
1263 	rw_enter(&bd_lock, RW_READER);
1264 
1265 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1266 		rw_exit(&bd_lock);
1267 		return (ENXIO);
1268 	}
1269 
1270 	mutex_enter(&bd->d_ocmutex);
1271 	if (bd->d_open_excl & mask) {
1272 		bd->d_open_excl &= ~mask;
1273 	}
1274 	if (otyp == OTYP_LYR) {
1275 		bd->d_open_lyr[part]--;
1276 	} else {
1277 		bd->d_open_reg[otyp] &= ~mask;
1278 	}
1279 	for (int i = 0; i < 64; i++) {
1280 		if (bd->d_open_lyr[part]) {
1281 			last = B_FALSE;
1282 		}
1283 	}
1284 	for (int i = 0; last && (i < OTYP_LYR); i++) {
1285 		if (bd->d_open_reg[i]) {
1286 			last = B_FALSE;
1287 		}
1288 	}
1289 	mutex_exit(&bd->d_ocmutex);
1290 
1291 	if (last) {
1292 		cmlb_invalidate(bd->d_cmlbh, 0);
1293 	}
1294 	rw_exit(&bd_lock);
1295 
1296 	return (0);
1297 }
1298 
1299 static int
1300 bd_dump(dev_t dev, caddr_t caddr, daddr_t blkno, int nblk)
1301 {
1302 	minor_t		inst;
1303 	minor_t		part;
1304 	diskaddr_t	pstart;
1305 	diskaddr_t	psize;
1306 	bd_t		*bd;
1307 	bd_xfer_impl_t	*xi;
1308 	buf_t		*bp;
1309 	int		rv;
1310 	uint32_t	shift;
1311 	daddr_t		d_blkno;
1312 	int	d_nblk;
1313 
1314 	rw_enter(&bd_lock, RW_READER);
1315 
1316 	part = BDPART(dev);
1317 	inst = BDINST(dev);
1318 
1319 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1320 		rw_exit(&bd_lock);
1321 		return (ENXIO);
1322 	}
1323 	shift = bd->d_blkshift;
1324 	d_blkno = blkno >> (shift - DEV_BSHIFT);
1325 	d_nblk = nblk >> (shift - DEV_BSHIFT);
1326 	/*
1327 	 * do cmlb, but do it synchronously unless we already have the
1328 	 * partition (which we probably should.)
1329 	 */
1330 	if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL,
1331 	    (void *)1)) {
1332 		rw_exit(&bd_lock);
1333 		return (ENXIO);
1334 	}
1335 
1336 	if ((d_blkno + d_nblk) > psize) {
1337 		rw_exit(&bd_lock);
1338 		return (EINVAL);
1339 	}
1340 	bp = getrbuf(KM_NOSLEEP);
1341 	if (bp == NULL) {
1342 		rw_exit(&bd_lock);
1343 		return (ENOMEM);
1344 	}
1345 
1346 	bp->b_bcount = nblk << DEV_BSHIFT;
1347 	bp->b_resid = bp->b_bcount;
1348 	bp->b_lblkno = blkno;
1349 	bp->b_un.b_addr = caddr;
1350 
1351 	xi = bd_xfer_alloc(bd, bp,  bd->d_ops.o_write, KM_NOSLEEP);
1352 	if (xi == NULL) {
1353 		rw_exit(&bd_lock);
1354 		freerbuf(bp);
1355 		return (ENOMEM);
1356 	}
1357 	xi->i_blkno = d_blkno + pstart;
1358 	xi->i_flags = BD_XFER_POLL;
1359 	bd_submit(bd, xi);
1360 	rw_exit(&bd_lock);
1361 
1362 	/*
1363 	 * Generally, we should have run this entirely synchronously
1364 	 * at this point and the biowait call should be a no-op.  If
1365 	 * it didn't happen this way, it's a bug in the underlying
1366 	 * driver not honoring BD_XFER_POLL.
1367 	 */
1368 	(void) biowait(bp);
1369 	rv = geterror(bp);
1370 	freerbuf(bp);
1371 	return (rv);
1372 }
1373 
1374 void
1375 bd_minphys(struct buf *bp)
1376 {
1377 	minor_t inst;
1378 	bd_t	*bd;
1379 	inst = BDINST(bp->b_edev);
1380 
1381 	bd = ddi_get_soft_state(bd_state, inst);
1382 
1383 	/*
1384 	 * In a non-debug kernel, bd_strategy will catch !bd as
1385 	 * well, and will fail nicely.
1386 	 */
1387 	ASSERT(bd);
1388 
1389 	if (bp->b_bcount > bd->d_maxxfer)
1390 		bp->b_bcount = bd->d_maxxfer;
1391 }
1392 
1393 static int
1394 bd_check_uio(dev_t dev, struct uio *uio)
1395 {
1396 	bd_t		*bd;
1397 	uint32_t	shift;
1398 
1399 	if ((bd = ddi_get_soft_state(bd_state, BDINST(dev))) == NULL) {
1400 		return (ENXIO);
1401 	}
1402 
1403 	shift = bd->d_blkshift;
1404 	if ((P2PHASE(uio->uio_loffset, (1U << shift)) != 0) ||
1405 	    (P2PHASE(uio->uio_iov->iov_len, (1U << shift)) != 0)) {
1406 		return (EINVAL);
1407 	}
1408 
1409 	return (0);
1410 }
1411 
1412 static int
1413 bd_read(dev_t dev, struct uio *uio, cred_t *credp)
1414 {
1415 	_NOTE(ARGUNUSED(credp));
1416 	int	ret = bd_check_uio(dev, uio);
1417 	if (ret != 0) {
1418 		return (ret);
1419 	}
1420 	return (physio(bd_strategy, NULL, dev, B_READ, bd_minphys, uio));
1421 }
1422 
1423 static int
1424 bd_write(dev_t dev, struct uio *uio, cred_t *credp)
1425 {
1426 	_NOTE(ARGUNUSED(credp));
1427 	int	ret = bd_check_uio(dev, uio);
1428 	if (ret != 0) {
1429 		return (ret);
1430 	}
1431 	return (physio(bd_strategy, NULL, dev, B_WRITE, bd_minphys, uio));
1432 }
1433 
1434 static int
1435 bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp)
1436 {
1437 	_NOTE(ARGUNUSED(credp));
1438 	int	ret = bd_check_uio(dev, aio->aio_uio);
1439 	if (ret != 0) {
1440 		return (ret);
1441 	}
1442 	return (aphysio(bd_strategy, anocancel, dev, B_READ, bd_minphys, aio));
1443 }
1444 
1445 static int
1446 bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp)
1447 {
1448 	_NOTE(ARGUNUSED(credp));
1449 	int	ret = bd_check_uio(dev, aio->aio_uio);
1450 	if (ret != 0) {
1451 		return (ret);
1452 	}
1453 	return (aphysio(bd_strategy, anocancel, dev, B_WRITE, bd_minphys, aio));
1454 }
1455 
1456 static int
1457 bd_strategy(struct buf *bp)
1458 {
1459 	minor_t		inst;
1460 	minor_t		part;
1461 	bd_t		*bd;
1462 	diskaddr_t	p_lba;
1463 	diskaddr_t	p_nblks;
1464 	diskaddr_t	b_nblks;
1465 	bd_xfer_impl_t	*xi;
1466 	uint32_t	shift;
1467 	int		(*func)(void *, bd_xfer_t *);
1468 	diskaddr_t	lblkno;
1469 
1470 	part = BDPART(bp->b_edev);
1471 	inst = BDINST(bp->b_edev);
1472 
1473 	ASSERT(bp);
1474 
1475 	bp->b_resid = bp->b_bcount;
1476 
1477 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1478 		bioerror(bp, ENXIO);
1479 		biodone(bp);
1480 		return (0);
1481 	}
1482 
1483 	if (cmlb_partinfo(bd->d_cmlbh, part, &p_nblks, &p_lba,
1484 	    NULL, NULL, 0)) {
1485 		bioerror(bp, ENXIO);
1486 		biodone(bp);
1487 		return (0);
1488 	}
1489 
1490 	shift = bd->d_blkshift;
1491 	lblkno = bp->b_lblkno >> (shift - DEV_BSHIFT);
1492 	if ((P2PHASE(bp->b_lblkno, (1U << (shift - DEV_BSHIFT))) != 0) ||
1493 	    (P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
1494 	    (lblkno > p_nblks)) {
1495 		bioerror(bp, EINVAL);
1496 		biodone(bp);
1497 		return (0);
1498 	}
1499 	b_nblks = bp->b_bcount >> shift;
1500 	if ((lblkno == p_nblks) || (bp->b_bcount == 0)) {
1501 		biodone(bp);
1502 		return (0);
1503 	}
1504 
1505 	if ((b_nblks + lblkno) > p_nblks) {
1506 		bp->b_resid = ((lblkno + b_nblks - p_nblks) << shift);
1507 		bp->b_bcount -= bp->b_resid;
1508 	} else {
1509 		bp->b_resid = 0;
1510 	}
1511 	func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write;
1512 
1513 	xi = bd_xfer_alloc(bd, bp, func, KM_NOSLEEP);
1514 	if (xi == NULL) {
1515 		xi = bd_xfer_alloc(bd, bp, func, KM_PUSHPAGE);
1516 	}
1517 	if (xi == NULL) {
1518 		/* bd_request_alloc will have done bioerror */
1519 		biodone(bp);
1520 		return (0);
1521 	}
1522 	xi->i_blkno = lblkno + p_lba;
1523 
1524 	bd_submit(bd, xi);
1525 
1526 	return (0);
1527 }
1528 
1529 static int
1530 bd_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp, int *rvalp)
1531 {
1532 	minor_t		inst;
1533 	uint16_t	part;
1534 	bd_t		*bd;
1535 	void		*ptr = (void *)arg;
1536 	int		rv;
1537 
1538 	part = BDPART(dev);
1539 	inst = BDINST(dev);
1540 
1541 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1542 		return (ENXIO);
1543 	}
1544 
1545 	rv = cmlb_ioctl(bd->d_cmlbh, dev, cmd, arg, flag, credp, rvalp, 0);
1546 	if (rv != ENOTTY)
1547 		return (rv);
1548 
1549 	if (rvalp != NULL) {
1550 		/* the return value of the ioctl is 0 by default */
1551 		*rvalp = 0;
1552 	}
1553 
1554 	switch (cmd) {
1555 	case DKIOCGMEDIAINFO: {
1556 		struct dk_minfo minfo;
1557 
1558 		/* make sure our state information is current */
1559 		bd_update_state(bd);
1560 		bzero(&minfo, sizeof (minfo));
1561 		minfo.dki_media_type = DK_FIXED_DISK;
1562 		minfo.dki_lbsize = (1U << bd->d_blkshift);
1563 		minfo.dki_capacity = bd->d_numblks;
1564 		if (ddi_copyout(&minfo, ptr, sizeof (minfo), flag)) {
1565 			return (EFAULT);
1566 		}
1567 		return (0);
1568 	}
1569 	case DKIOCGMEDIAINFOEXT: {
1570 		struct dk_minfo_ext miext;
1571 		size_t len;
1572 
1573 		/* make sure our state information is current */
1574 		bd_update_state(bd);
1575 		bzero(&miext, sizeof (miext));
1576 		miext.dki_media_type = DK_FIXED_DISK;
1577 		miext.dki_lbsize = (1U << bd->d_blkshift);
1578 		miext.dki_pbsize = (1U << bd->d_pblkshift);
1579 		miext.dki_capacity = bd->d_numblks;
1580 
1581 		switch (ddi_model_convert_from(flag & FMODELS)) {
1582 		case DDI_MODEL_ILP32:
1583 			len = sizeof (struct dk_minfo_ext32);
1584 			break;
1585 		default:
1586 			len = sizeof (struct dk_minfo_ext);
1587 			break;
1588 		}
1589 
1590 		if (ddi_copyout(&miext, ptr, len, flag)) {
1591 			return (EFAULT);
1592 		}
1593 		return (0);
1594 	}
1595 	case DKIOCINFO: {
1596 		struct dk_cinfo cinfo;
1597 		bzero(&cinfo, sizeof (cinfo));
1598 		cinfo.dki_ctype = DKC_BLKDEV;
1599 		cinfo.dki_cnum = ddi_get_instance(ddi_get_parent(bd->d_dip));
1600 		(void) snprintf(cinfo.dki_cname, sizeof (cinfo.dki_cname),
1601 		    "%s", ddi_driver_name(ddi_get_parent(bd->d_dip)));
1602 		(void) snprintf(cinfo.dki_dname, sizeof (cinfo.dki_dname),
1603 		    "%s", ddi_driver_name(bd->d_dip));
1604 		cinfo.dki_unit = inst;
1605 		cinfo.dki_flags = DKI_FMTVOL;
1606 		cinfo.dki_partition = part;
1607 		cinfo.dki_maxtransfer = bd->d_maxxfer / DEV_BSIZE;
1608 		cinfo.dki_addr = 0;
1609 		cinfo.dki_slave = 0;
1610 		cinfo.dki_space = 0;
1611 		cinfo.dki_prio = 0;
1612 		cinfo.dki_vec = 0;
1613 		if (ddi_copyout(&cinfo, ptr, sizeof (cinfo), flag)) {
1614 			return (EFAULT);
1615 		}
1616 		return (0);
1617 	}
1618 	case DKIOCREMOVABLE: {
1619 		int i;
1620 		i = bd->d_removable ? 1 : 0;
1621 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1622 			return (EFAULT);
1623 		}
1624 		return (0);
1625 	}
1626 	case DKIOCHOTPLUGGABLE: {
1627 		int i;
1628 		i = bd->d_hotpluggable ? 1 : 0;
1629 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1630 			return (EFAULT);
1631 		}
1632 		return (0);
1633 	}
1634 	case DKIOCREADONLY: {
1635 		int i;
1636 		i = bd->d_rdonly ? 1 : 0;
1637 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1638 			return (EFAULT);
1639 		}
1640 		return (0);
1641 	}
1642 	case DKIOCSOLIDSTATE: {
1643 		int i;
1644 		i = bd->d_ssd ? 1 : 0;
1645 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1646 			return (EFAULT);
1647 		}
1648 		return (0);
1649 	}
1650 	case DKIOCSTATE: {
1651 		enum dkio_state	state;
1652 		if (ddi_copyin(ptr, &state, sizeof (state), flag)) {
1653 			return (EFAULT);
1654 		}
1655 		if ((rv = bd_check_state(bd, &state)) != 0) {
1656 			return (rv);
1657 		}
1658 		if (ddi_copyout(&state, ptr, sizeof (state), flag)) {
1659 			return (EFAULT);
1660 		}
1661 		return (0);
1662 	}
1663 	case DKIOCFLUSHWRITECACHE: {
1664 		struct dk_callback *dkc = NULL;
1665 
1666 		if (flag & FKIOCTL)
1667 			dkc = (void *)arg;
1668 
1669 		rv = bd_flush_write_cache(bd, dkc);
1670 		return (rv);
1671 	}
1672 	case DKIOCFREE: {
1673 		dkioc_free_list_t *dfl = NULL;
1674 
1675 		/*
1676 		 * Check free space support early to avoid copyin/allocation
1677 		 * when unnecessary.
1678 		 */
1679 		if (!CAN_FREESPACE(bd))
1680 			return (ENOTSUP);
1681 
1682 		rv = dfl_copyin(ptr, &dfl, flag, KM_SLEEP);
1683 		if (rv != 0)
1684 			return (rv);
1685 
1686 		/*
1687 		 * bd_free_space() consumes 'dfl'. bd_free_space() will
1688 		 * call dfl_iter() which will normally try to pass dfl through
1689 		 * to bd_free_space_cb() which attaches dfl to the bd_xfer_t
1690 		 * that is then queued for the underlying driver. Once the
1691 		 * driver processes the request, the bd_xfer_t instance is
1692 		 * disposed of, including any attached dkioc_free_list_t.
1693 		 *
1694 		 * If dfl cannot be processed by the underlying driver due to
1695 		 * size or alignment requirements of the driver, dfl_iter()
1696 		 * will replace dfl with one or more new dkioc_free_list_t
1697 		 * instances with the correct alignment and sizes for the driver
1698 		 * (and free the original dkioc_free_list_t).
1699 		 */
1700 		rv = bd_free_space(dev, bd, dfl);
1701 		return (rv);
1702 	}
1703 
1704 	case DKIOC_CANFREE: {
1705 		boolean_t supported = CAN_FREESPACE(bd);
1706 
1707 		if (ddi_copyout(&supported, (void *)arg, sizeof (supported),
1708 		    flag) != 0) {
1709 			return (EFAULT);
1710 		}
1711 
1712 		return (0);
1713 	}
1714 
1715 	default:
1716 		break;
1717 
1718 	}
1719 	return (ENOTTY);
1720 }
1721 
1722 static int
1723 bd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1724     char *name, caddr_t valuep, int *lengthp)
1725 {
1726 	bd_t	*bd;
1727 
1728 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1729 	if (bd == NULL)
1730 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1731 		    name, valuep, lengthp));
1732 
1733 	return (cmlb_prop_op(bd->d_cmlbh, dev, dip, prop_op, mod_flags, name,
1734 	    valuep, lengthp, BDPART(dev), 0));
1735 }
1736 
1737 
1738 static int
1739 bd_tg_rdwr(dev_info_t *dip, uchar_t cmd, void *bufaddr, diskaddr_t start,
1740     size_t length, void *tg_cookie)
1741 {
1742 	bd_t		*bd;
1743 	buf_t		*bp;
1744 	bd_xfer_impl_t	*xi;
1745 	int		rv;
1746 	int		(*func)(void *, bd_xfer_t *);
1747 	int		kmflag;
1748 
1749 	/*
1750 	 * If we are running in polled mode (such as during dump(9e)
1751 	 * execution), then we cannot sleep for kernel allocations.
1752 	 */
1753 	kmflag = tg_cookie ? KM_NOSLEEP : KM_SLEEP;
1754 
1755 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1756 
1757 	if (P2PHASE(length, (1U << bd->d_blkshift)) != 0) {
1758 		/* We can only transfer whole blocks at a time! */
1759 		return (EINVAL);
1760 	}
1761 
1762 	if ((bp = getrbuf(kmflag)) == NULL) {
1763 		return (ENOMEM);
1764 	}
1765 
1766 	switch (cmd) {
1767 	case TG_READ:
1768 		bp->b_flags = B_READ;
1769 		func = bd->d_ops.o_read;
1770 		break;
1771 	case TG_WRITE:
1772 		bp->b_flags = B_WRITE;
1773 		func = bd->d_ops.o_write;
1774 		break;
1775 	default:
1776 		freerbuf(bp);
1777 		return (EINVAL);
1778 	}
1779 
1780 	bp->b_un.b_addr = bufaddr;
1781 	bp->b_bcount = length;
1782 	xi = bd_xfer_alloc(bd, bp, func, kmflag);
1783 	if (xi == NULL) {
1784 		rv = geterror(bp);
1785 		freerbuf(bp);
1786 		return (rv);
1787 	}
1788 	xi->i_flags = tg_cookie ? BD_XFER_POLL : 0;
1789 	xi->i_blkno = start;
1790 	bd_submit(bd, xi);
1791 	(void) biowait(bp);
1792 	rv = geterror(bp);
1793 	freerbuf(bp);
1794 
1795 	return (rv);
1796 }
1797 
1798 static int
1799 bd_tg_getinfo(dev_info_t *dip, int cmd, void *arg, void *tg_cookie)
1800 {
1801 	bd_t		*bd;
1802 
1803 	_NOTE(ARGUNUSED(tg_cookie));
1804 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1805 
1806 	switch (cmd) {
1807 	case TG_GETPHYGEOM:
1808 	case TG_GETVIRTGEOM:
1809 		/*
1810 		 * We don't have any "geometry" as such, let cmlb
1811 		 * fabricate something.
1812 		 */
1813 		return (ENOTTY);
1814 
1815 	case TG_GETCAPACITY:
1816 		bd_update_state(bd);
1817 		*(diskaddr_t *)arg = bd->d_numblks;
1818 		return (0);
1819 
1820 	case TG_GETBLOCKSIZE:
1821 		*(uint32_t *)arg = (1U << bd->d_blkshift);
1822 		return (0);
1823 
1824 	case TG_GETATTR:
1825 		/*
1826 		 * It turns out that cmlb really doesn't do much for
1827 		 * non-writable media, but lets make the information
1828 		 * available for it in case it does more in the
1829 		 * future.  (The value is currently used for
1830 		 * triggering special behavior for CD-ROMs.)
1831 		 */
1832 		bd_update_state(bd);
1833 		((tg_attribute_t *)arg)->media_is_writable =
1834 		    bd->d_rdonly ? B_FALSE : B_TRUE;
1835 		((tg_attribute_t *)arg)->media_is_solid_state = bd->d_ssd;
1836 		((tg_attribute_t *)arg)->media_is_rotational = B_FALSE;
1837 		return (0);
1838 
1839 	default:
1840 		return (EINVAL);
1841 	}
1842 }
1843 
1844 
1845 static void
1846 bd_sched(bd_t *bd, bd_queue_t *bq)
1847 {
1848 	bd_xfer_impl_t	*xi;
1849 	struct buf	*bp;
1850 	int		rv;
1851 
1852 	mutex_enter(&bq->q_iomutex);
1853 
1854 	while ((bq->q_qactive < bq->q_qsize) &&
1855 	    ((xi = list_remove_head(&bq->q_waitq)) != NULL)) {
1856 		mutex_enter(&bd->d_ksmutex);
1857 		kstat_waitq_to_runq(bd->d_kiop);
1858 		mutex_exit(&bd->d_ksmutex);
1859 
1860 		bq->q_qactive++;
1861 		list_insert_tail(&bq->q_runq, xi);
1862 
1863 		/*
1864 		 * Submit the job to the driver.  We drop the I/O mutex
1865 		 * so that we can deal with the case where the driver
1866 		 * completion routine calls back into us synchronously.
1867 		 */
1868 
1869 		mutex_exit(&bq->q_iomutex);
1870 
1871 		rv = xi->i_func(bd->d_private, &xi->i_public);
1872 		if (rv != 0) {
1873 			bp = xi->i_bp;
1874 			bioerror(bp, rv);
1875 			biodone(bp);
1876 
1877 			atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32);
1878 
1879 			mutex_enter(&bq->q_iomutex);
1880 
1881 			mutex_enter(&bd->d_ksmutex);
1882 			kstat_runq_exit(bd->d_kiop);
1883 			mutex_exit(&bd->d_ksmutex);
1884 
1885 			bq->q_qactive--;
1886 			list_remove(&bq->q_runq, xi);
1887 			bd_xfer_free(xi);
1888 		} else {
1889 			mutex_enter(&bq->q_iomutex);
1890 		}
1891 	}
1892 
1893 	mutex_exit(&bq->q_iomutex);
1894 }
1895 
1896 static void
1897 bd_submit(bd_t *bd, bd_xfer_impl_t *xi)
1898 {
1899 	uint64_t	nv = atomic_inc_64_nv(&bd->d_io_counter);
1900 	unsigned	q = nv % bd->d_qcount;
1901 	bd_queue_t	*bq = &bd->d_queues[q];
1902 
1903 	xi->i_bq = bq;
1904 	xi->i_qnum = q;
1905 
1906 	mutex_enter(&bq->q_iomutex);
1907 
1908 	list_insert_tail(&bq->q_waitq, xi);
1909 
1910 	mutex_enter(&bd->d_ksmutex);
1911 	kstat_waitq_enter(bd->d_kiop);
1912 	mutex_exit(&bd->d_ksmutex);
1913 
1914 	mutex_exit(&bq->q_iomutex);
1915 
1916 	bd_sched(bd, bq);
1917 }
1918 
1919 static void
1920 bd_runq_exit(bd_xfer_impl_t *xi, int err)
1921 {
1922 	bd_t		*bd = xi->i_bd;
1923 	buf_t		*bp = xi->i_bp;
1924 	bd_queue_t	*bq = xi->i_bq;
1925 
1926 	mutex_enter(&bq->q_iomutex);
1927 	bq->q_qactive--;
1928 
1929 	mutex_enter(&bd->d_ksmutex);
1930 	kstat_runq_exit(bd->d_kiop);
1931 	mutex_exit(&bd->d_ksmutex);
1932 
1933 	list_remove(&bq->q_runq, xi);
1934 	mutex_exit(&bq->q_iomutex);
1935 
1936 	if (err == 0) {
1937 		if (bp->b_flags & B_READ) {
1938 			atomic_inc_uint(&bd->d_kiop->reads);
1939 			atomic_add_64((uint64_t *)&bd->d_kiop->nread,
1940 			    bp->b_bcount - xi->i_resid);
1941 		} else {
1942 			atomic_inc_uint(&bd->d_kiop->writes);
1943 			atomic_add_64((uint64_t *)&bd->d_kiop->nwritten,
1944 			    bp->b_bcount - xi->i_resid);
1945 		}
1946 	}
1947 	bd_sched(bd, bq);
1948 }
1949 
1950 static void
1951 bd_dle_sysevent_task(void *arg)
1952 {
1953 	nvlist_t *attr = NULL;
1954 	char *path = NULL;
1955 	bd_t *bd = arg;
1956 	dev_info_t *dip = bd->d_dip;
1957 	size_t n;
1958 
1959 	mutex_enter(&bd->d_dle_mutex);
1960 	bd->d_dle_state &= ~BD_DLE_PENDING;
1961 	bd->d_dle_state |= BD_DLE_RUNNING;
1962 	mutex_exit(&bd->d_dle_mutex);
1963 
1964 	dev_err(dip, CE_NOTE, "!dynamic LUN expansion");
1965 
1966 	if (nvlist_alloc(&attr, NV_UNIQUE_NAME_TYPE, KM_SLEEP) != 0) {
1967 		mutex_enter(&bd->d_dle_mutex);
1968 		bd->d_dle_state &= ~(BD_DLE_RUNNING|BD_DLE_PENDING);
1969 		mutex_exit(&bd->d_dle_mutex);
1970 		return;
1971 	}
1972 
1973 	path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1974 
1975 	n = snprintf(path, MAXPATHLEN, "/devices");
1976 	(void) ddi_pathname(dip, path + n);
1977 	n = strlen(path);
1978 	n += snprintf(path + n, MAXPATHLEN - n, ":x");
1979 
1980 	for (;;) {
1981 		/*
1982 		 * On receipt of this event, the ZFS sysevent module will scan
1983 		 * active zpools for child vdevs matching this physical path.
1984 		 * In order to catch both whole disk pools and those with an
1985 		 * EFI boot partition, generate separate sysevents for minor
1986 		 * node 'a' and 'b'.
1987 		 */
1988 		for (char c = 'a'; c < 'c'; c++) {
1989 			path[n - 1] = c;
1990 
1991 			if (nvlist_add_string(attr, DEV_PHYS_PATH, path) != 0)
1992 				break;
1993 
1994 			(void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW,
1995 			    EC_DEV_STATUS, ESC_DEV_DLE, attr, NULL, DDI_SLEEP);
1996 		}
1997 
1998 		mutex_enter(&bd->d_dle_mutex);
1999 		if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
2000 			bd->d_dle_state &= ~BD_DLE_RUNNING;
2001 			mutex_exit(&bd->d_dle_mutex);
2002 			break;
2003 		}
2004 		bd->d_dle_state &= ~BD_DLE_PENDING;
2005 		mutex_exit(&bd->d_dle_mutex);
2006 	}
2007 
2008 	nvlist_free(attr);
2009 	kmem_free(path, MAXPATHLEN);
2010 }
2011 
2012 static void
2013 bd_update_state(bd_t *bd)
2014 {
2015 	enum	dkio_state	state = DKIO_INSERTED;
2016 	boolean_t		docmlb = B_FALSE;
2017 	bd_media_t		media;
2018 
2019 	bzero(&media, sizeof (media));
2020 
2021 	mutex_enter(&bd->d_statemutex);
2022 	if (bd->d_ops.o_media_info(bd->d_private, &media) != 0) {
2023 		bd->d_numblks = 0;
2024 		state = DKIO_EJECTED;
2025 		goto done;
2026 	}
2027 
2028 	if ((media.m_blksize < 512) ||
2029 	    (!ISP2(media.m_blksize)) ||
2030 	    (P2PHASE(bd->d_maxxfer, media.m_blksize))) {
2031 		dev_err(bd->d_dip, CE_WARN, "Invalid media block size (%d)",
2032 		    media.m_blksize);
2033 		/*
2034 		 * We can't use the media, treat it as not present.
2035 		 */
2036 		state = DKIO_EJECTED;
2037 		bd->d_numblks = 0;
2038 		goto done;
2039 	}
2040 
2041 	if (((1U << bd->d_blkshift) != media.m_blksize) ||
2042 	    (bd->d_numblks != media.m_nblks)) {
2043 		/* Device size changed */
2044 		docmlb = B_TRUE;
2045 	}
2046 
2047 	bd->d_blkshift = ddi_ffs(media.m_blksize) - 1;
2048 	bd->d_pblkshift = bd->d_blkshift;
2049 	bd->d_numblks = media.m_nblks;
2050 	bd->d_rdonly = media.m_readonly;
2051 	bd->d_ssd = media.m_solidstate;
2052 
2053 	/*
2054 	 * Only use the supplied physical block size if it is non-zero,
2055 	 * greater or equal to the block size, and a power of 2. Ignore it
2056 	 * if not, it's just informational and we can still use the media.
2057 	 */
2058 	if ((media.m_pblksize != 0) &&
2059 	    (media.m_pblksize >= media.m_blksize) &&
2060 	    (ISP2(media.m_pblksize)))
2061 		bd->d_pblkshift = ddi_ffs(media.m_pblksize) - 1;
2062 
2063 done:
2064 	if (state != bd->d_state) {
2065 		bd->d_state = state;
2066 		cv_broadcast(&bd->d_statecv);
2067 		docmlb = B_TRUE;
2068 	}
2069 	mutex_exit(&bd->d_statemutex);
2070 
2071 	bd->d_kerr->bd_capacity.value.ui64 = bd->d_numblks << bd->d_blkshift;
2072 
2073 	if (docmlb) {
2074 		if (state == DKIO_INSERTED) {
2075 			(void) cmlb_validate(bd->d_cmlbh, 0, 0);
2076 
2077 			mutex_enter(&bd->d_dle_mutex);
2078 			/*
2079 			 * If there is already an event pending, there's
2080 			 * nothing to do; we coalesce multiple events.
2081 			 */
2082 			if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
2083 				if ((bd->d_dle_state & BD_DLE_RUNNING) == 0) {
2084 					taskq_dispatch_ent(bd_taskq,
2085 					    bd_dle_sysevent_task, bd, 0,
2086 					    &bd->d_dle_ent);
2087 				}
2088 				bd->d_dle_state |= BD_DLE_PENDING;
2089 			}
2090 			mutex_exit(&bd->d_dle_mutex);
2091 		} else {
2092 			cmlb_invalidate(bd->d_cmlbh, 0);
2093 		}
2094 	}
2095 }
2096 
2097 static int
2098 bd_check_state(bd_t *bd, enum dkio_state *state)
2099 {
2100 	clock_t		when;
2101 
2102 	for (;;) {
2103 
2104 		bd_update_state(bd);
2105 
2106 		mutex_enter(&bd->d_statemutex);
2107 
2108 		if (bd->d_state != *state) {
2109 			*state = bd->d_state;
2110 			mutex_exit(&bd->d_statemutex);
2111 			break;
2112 		}
2113 
2114 		when = drv_usectohz(1000000);
2115 		if (cv_reltimedwait_sig(&bd->d_statecv, &bd->d_statemutex,
2116 		    when, TR_CLOCK_TICK) == 0) {
2117 			mutex_exit(&bd->d_statemutex);
2118 			return (EINTR);
2119 		}
2120 
2121 		mutex_exit(&bd->d_statemutex);
2122 	}
2123 
2124 	return (0);
2125 }
2126 
2127 static int
2128 bd_flush_write_cache_done(struct buf *bp)
2129 {
2130 	struct dk_callback *dc = (void *)bp->b_private;
2131 
2132 	(*dc->dkc_callback)(dc->dkc_cookie, geterror(bp));
2133 	kmem_free(dc, sizeof (*dc));
2134 	freerbuf(bp);
2135 	return (0);
2136 }
2137 
2138 static int
2139 bd_flush_write_cache(bd_t *bd, struct dk_callback *dkc)
2140 {
2141 	buf_t			*bp;
2142 	struct dk_callback	*dc;
2143 	bd_xfer_impl_t		*xi;
2144 	int			rv;
2145 
2146 	if (bd->d_ops.o_sync_cache == NULL) {
2147 		return (ENOTSUP);
2148 	}
2149 	if ((bp = getrbuf(KM_SLEEP)) == NULL) {
2150 		return (ENOMEM);
2151 	}
2152 	bp->b_resid = 0;
2153 	bp->b_bcount = 0;
2154 
2155 	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_sync_cache, KM_SLEEP);
2156 	if (xi == NULL) {
2157 		rv = geterror(bp);
2158 		freerbuf(bp);
2159 		return (rv);
2160 	}
2161 
2162 	/* Make an asynchronous flush, but only if there is a callback */
2163 	if (dkc != NULL && dkc->dkc_callback != NULL) {
2164 		/* Make a private copy of the callback structure */
2165 		dc = kmem_alloc(sizeof (*dc), KM_SLEEP);
2166 		*dc = *dkc;
2167 		bp->b_private = dc;
2168 		bp->b_iodone = bd_flush_write_cache_done;
2169 
2170 		bd_submit(bd, xi);
2171 		return (0);
2172 	}
2173 
2174 	/* In case there is no callback, perform a synchronous flush */
2175 	bd_submit(bd, xi);
2176 	(void) biowait(bp);
2177 	rv = geterror(bp);
2178 	freerbuf(bp);
2179 
2180 	return (rv);
2181 }
2182 
2183 static int
2184 bd_free_space_done(struct buf *bp)
2185 {
2186 	freerbuf(bp);
2187 	return (0);
2188 }
2189 
2190 static int
2191 bd_free_space_cb(dkioc_free_list_t *dfl, void *arg, int kmflag)
2192 {
2193 	bd_t		*bd = arg;
2194 	buf_t		*bp = NULL;
2195 	bd_xfer_impl_t	*xi = NULL;
2196 	boolean_t	sync = DFL_ISSYNC(dfl) ?  B_TRUE : B_FALSE;
2197 	int		rv = 0;
2198 
2199 	bp = getrbuf(KM_SLEEP);
2200 	bp->b_resid = 0;
2201 	bp->b_bcount = 0;
2202 	bp->b_lblkno = 0;
2203 
2204 	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_free_space, kmflag);
2205 	xi->i_dfl = dfl;
2206 
2207 	if (!sync) {
2208 		bp->b_iodone = bd_free_space_done;
2209 		bd_submit(bd, xi);
2210 		return (0);
2211 	}
2212 
2213 	xi->i_flags |= BD_XFER_POLL;
2214 	bd_submit(bd, xi);
2215 
2216 	(void) biowait(bp);
2217 	rv = geterror(bp);
2218 	freerbuf(bp);
2219 
2220 	return (rv);
2221 }
2222 
2223 static int
2224 bd_free_space(dev_t dev, bd_t *bd, dkioc_free_list_t *dfl)
2225 {
2226 	diskaddr_t p_len, p_offset;
2227 	uint64_t offset_bytes, len_bytes;
2228 	minor_t part = BDPART(dev);
2229 	const uint_t bshift = bd->d_blkshift;
2230 	dkioc_free_info_t dfi = {
2231 		.dfi_bshift = bshift,
2232 		.dfi_align = bd->d_free_align << bshift,
2233 		.dfi_max_bytes = bd->d_max_free_blks << bshift,
2234 		.dfi_max_ext = bd->d_max_free_seg,
2235 		.dfi_max_ext_bytes = bd->d_max_free_seg_blks << bshift,
2236 	};
2237 
2238 	if (cmlb_partinfo(bd->d_cmlbh, part, &p_len, &p_offset, NULL,
2239 	    NULL, 0) != 0) {
2240 		dfl_free(dfl);
2241 		return (ENXIO);
2242 	}
2243 
2244 	/*
2245 	 * bd_ioctl created our own copy of dfl, so we can modify as
2246 	 * necessary
2247 	 */
2248 	offset_bytes = (uint64_t)p_offset << bshift;
2249 	len_bytes = (uint64_t)p_len << bshift;
2250 
2251 	dfl->dfl_offset += offset_bytes;
2252 	if (dfl->dfl_offset < offset_bytes) {
2253 		dfl_free(dfl);
2254 		return (EOVERFLOW);
2255 	}
2256 
2257 	return (dfl_iter(dfl, &dfi, offset_bytes + len_bytes, bd_free_space_cb,
2258 	    bd, KM_SLEEP));
2259 }
2260 
2261 /*
2262  * Nexus support.
2263  */
2264 int
2265 bd_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop,
2266     void *arg, void *result)
2267 {
2268 	bd_handle_t	hdl;
2269 
2270 	switch (ctlop) {
2271 	case DDI_CTLOPS_REPORTDEV:
2272 		cmn_err(CE_CONT, "?Block device: %s@%s, %s%d\n",
2273 		    ddi_node_name(rdip), ddi_get_name_addr(rdip),
2274 		    ddi_driver_name(rdip), ddi_get_instance(rdip));
2275 		return (DDI_SUCCESS);
2276 
2277 	case DDI_CTLOPS_INITCHILD:
2278 		hdl = ddi_get_parent_data((dev_info_t *)arg);
2279 		if (hdl == NULL) {
2280 			return (DDI_NOT_WELL_FORMED);
2281 		}
2282 		ddi_set_name_addr((dev_info_t *)arg, hdl->h_addr);
2283 		return (DDI_SUCCESS);
2284 
2285 	case DDI_CTLOPS_UNINITCHILD:
2286 		ddi_set_name_addr((dev_info_t *)arg, NULL);
2287 		ndi_prop_remove_all((dev_info_t *)arg);
2288 		return (DDI_SUCCESS);
2289 
2290 	default:
2291 		return (ddi_ctlops(dip, rdip, ctlop, arg, result));
2292 	}
2293 }
2294 
2295 /*
2296  * Functions for device drivers.
2297  */
2298 bd_handle_t
2299 bd_alloc_handle(void *private, bd_ops_t *ops, ddi_dma_attr_t *dma, int kmflag)
2300 {
2301 	bd_handle_t	hdl;
2302 
2303 	switch (ops->o_version) {
2304 	case BD_OPS_VERSION_0:
2305 	case BD_OPS_VERSION_1:
2306 	case BD_OPS_VERSION_2:
2307 		break;
2308 
2309 	default:
2310 		/* Unsupported version */
2311 		return (NULL);
2312 	}
2313 
2314 	hdl = kmem_zalloc(sizeof (*hdl), kmflag);
2315 	if (hdl == NULL) {
2316 		return (NULL);
2317 	}
2318 
2319 	switch (ops->o_version) {
2320 	case BD_OPS_VERSION_2:
2321 		hdl->h_ops.o_free_space = ops->o_free_space;
2322 		/*FALLTHRU*/
2323 	case BD_OPS_VERSION_1:
2324 	case BD_OPS_VERSION_0:
2325 		hdl->h_ops.o_drive_info = ops->o_drive_info;
2326 		hdl->h_ops.o_media_info = ops->o_media_info;
2327 		hdl->h_ops.o_devid_init = ops->o_devid_init;
2328 		hdl->h_ops.o_sync_cache = ops->o_sync_cache;
2329 		hdl->h_ops.o_read = ops->o_read;
2330 		hdl->h_ops.o_write = ops->o_write;
2331 		break;
2332 	}
2333 
2334 	hdl->h_dma = dma;
2335 	hdl->h_private = private;
2336 
2337 	return (hdl);
2338 }
2339 
2340 void
2341 bd_free_handle(bd_handle_t hdl)
2342 {
2343 	kmem_free(hdl, sizeof (*hdl));
2344 }
2345 
2346 int
2347 bd_attach_handle(dev_info_t *dip, bd_handle_t hdl)
2348 {
2349 	bd_drive_t	drive = { 0 };
2350 	dev_info_t	*child;
2351 	size_t		len;
2352 
2353 	/*
2354 	 * It's not an error if bd_attach_handle() is called on a handle that
2355 	 * already is attached. We just ignore the request to attach and return.
2356 	 * This way drivers using blkdev don't have to keep track about blkdev
2357 	 * state, they can just call this function to make sure it attached.
2358 	 */
2359 	if (hdl->h_child != NULL) {
2360 		return (DDI_SUCCESS);
2361 	}
2362 
2363 	/* if drivers don't override this, make it assume none */
2364 	drive.d_lun = -1;
2365 	hdl->h_ops.o_drive_info(hdl->h_private, &drive);
2366 
2367 	hdl->h_parent = dip;
2368 	hdl->h_name = "blkdev";
2369 
2370 	/*
2371 	 * Prefer the GUID over the EUI64.
2372 	 */
2373 	if (*(uint64_t *)drive.d_guid != 0 ||
2374 	    *((uint64_t *)drive.d_guid + 1) != 0) {
2375 		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
2376 		    "w%02X%02X%02X%02X%02X%02X%02X%02X"
2377 		    "%02X%02X%02X%02X%02X%02X%02X%02X",
2378 		    drive.d_guid[0], drive.d_guid[1], drive.d_guid[2],
2379 		    drive.d_guid[3], drive.d_guid[4], drive.d_guid[5],
2380 		    drive.d_guid[6], drive.d_guid[7], drive.d_guid[8],
2381 		    drive.d_guid[9], drive.d_guid[10], drive.d_guid[11],
2382 		    drive.d_guid[12], drive.d_guid[13], drive.d_guid[14],
2383 		    drive.d_guid[15]);
2384 	} else if (*(uint64_t *)drive.d_eui64 != 0) {
2385 		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
2386 		    "w%02X%02X%02X%02X%02X%02X%02X%02X",
2387 		    drive.d_eui64[0], drive.d_eui64[1],
2388 		    drive.d_eui64[2], drive.d_eui64[3],
2389 		    drive.d_eui64[4], drive.d_eui64[5],
2390 		    drive.d_eui64[6], drive.d_eui64[7]);
2391 	} else {
2392 		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
2393 		    "%X", drive.d_target);
2394 	}
2395 
2396 	VERIFY(len <= sizeof (hdl->h_addr));
2397 
2398 	if (drive.d_lun >= 0) {
2399 		(void) snprintf(hdl->h_addr + len, sizeof (hdl->h_addr) - len,
2400 		    ",%X", drive.d_lun);
2401 	}
2402 
2403 	if (ndi_devi_alloc(dip, hdl->h_name, (pnode_t)DEVI_SID_NODEID,
2404 	    &child) != NDI_SUCCESS) {
2405 		cmn_err(CE_WARN, "%s%d: unable to allocate node %s@%s",
2406 		    ddi_driver_name(dip), ddi_get_instance(dip),
2407 		    "blkdev", hdl->h_addr);
2408 		return (DDI_FAILURE);
2409 	}
2410 
2411 	ddi_set_parent_data(child, hdl);
2412 	hdl->h_child = child;
2413 
2414 	if (ndi_devi_online(child, 0) != NDI_SUCCESS) {
2415 		cmn_err(CE_WARN, "%s%d: failed bringing node %s@%s online",
2416 		    ddi_driver_name(dip), ddi_get_instance(dip),
2417 		    hdl->h_name, hdl->h_addr);
2418 		(void) ndi_devi_free(child);
2419 		hdl->h_child = NULL;
2420 		return (DDI_FAILURE);
2421 	}
2422 
2423 	return (DDI_SUCCESS);
2424 }
2425 
2426 int
2427 bd_detach_handle(bd_handle_t hdl)
2428 {
2429 	int	rv;
2430 	char	*devnm;
2431 
2432 	/*
2433 	 * It's not an error if bd_detach_handle() is called on a handle that
2434 	 * already is detached. We just ignore the request to detach and return.
2435 	 * This way drivers using blkdev don't have to keep track about blkdev
2436 	 * state, they can just call this function to make sure it detached.
2437 	 */
2438 	if (hdl->h_child == NULL) {
2439 		return (DDI_SUCCESS);
2440 	}
2441 	ndi_devi_enter(hdl->h_parent);
2442 	if (i_ddi_node_state(hdl->h_child) < DS_INITIALIZED) {
2443 		rv = ddi_remove_child(hdl->h_child, 0);
2444 	} else {
2445 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2446 		(void) ddi_deviname(hdl->h_child, devnm);
2447 		(void) devfs_clean(hdl->h_parent, devnm + 1, DV_CLEAN_FORCE);
2448 		rv = ndi_devi_unconfig_one(hdl->h_parent, devnm + 1, NULL,
2449 		    NDI_DEVI_REMOVE | NDI_UNCONFIG);
2450 		kmem_free(devnm, MAXNAMELEN + 1);
2451 	}
2452 	if (rv == 0) {
2453 		hdl->h_child = NULL;
2454 	}
2455 
2456 	ndi_devi_exit(hdl->h_parent);
2457 	return (rv == NDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
2458 }
2459 
2460 void
2461 bd_xfer_done(bd_xfer_t *xfer, int err)
2462 {
2463 	bd_xfer_impl_t	*xi = (void *)xfer;
2464 	buf_t		*bp = xi->i_bp;
2465 	int		rv = DDI_SUCCESS;
2466 	bd_t		*bd = xi->i_bd;
2467 	size_t		len;
2468 
2469 	if (err != 0) {
2470 		bd_runq_exit(xi, err);
2471 		atomic_inc_32(&bd->d_kerr->bd_harderrs.value.ui32);
2472 
2473 		bp->b_resid += xi->i_resid;
2474 		bd_xfer_free(xi);
2475 		bioerror(bp, err);
2476 		biodone(bp);
2477 		return;
2478 	}
2479 
2480 	xi->i_cur_win++;
2481 	xi->i_resid -= xi->i_len;
2482 
2483 	if (xi->i_resid == 0) {
2484 		/* Job completed succcessfully! */
2485 		bd_runq_exit(xi, 0);
2486 
2487 		bd_xfer_free(xi);
2488 		biodone(bp);
2489 		return;
2490 	}
2491 
2492 	xi->i_blkno += xi->i_nblks;
2493 
2494 	if (bd->d_use_dma) {
2495 		/* More transfer still pending... advance to next DMA window. */
2496 		rv = ddi_dma_getwin(xi->i_dmah, xi->i_cur_win,
2497 		    &xi->i_offset, &len, &xi->i_dmac, &xi->i_ndmac);
2498 	} else {
2499 		/* Advance memory window. */
2500 		xi->i_kaddr += xi->i_len;
2501 		xi->i_offset += xi->i_len;
2502 		len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer);
2503 	}
2504 
2505 
2506 	if ((rv != DDI_SUCCESS) ||
2507 	    (P2PHASE(len, (1U << xi->i_blkshift)) != 0)) {
2508 		bd_runq_exit(xi, EFAULT);
2509 
2510 		bp->b_resid += xi->i_resid;
2511 		bd_xfer_free(xi);
2512 		bioerror(bp, EFAULT);
2513 		biodone(bp);
2514 		return;
2515 	}
2516 	xi->i_len = len;
2517 	xi->i_nblks = len >> xi->i_blkshift;
2518 
2519 	/* Submit next window to hardware. */
2520 	rv = xi->i_func(bd->d_private, &xi->i_public);
2521 	if (rv != 0) {
2522 		bd_runq_exit(xi, rv);
2523 
2524 		atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32);
2525 
2526 		bp->b_resid += xi->i_resid;
2527 		bd_xfer_free(xi);
2528 		bioerror(bp, rv);
2529 		biodone(bp);
2530 	}
2531 }
2532 
2533 void
2534 bd_error(bd_xfer_t *xfer, int error)
2535 {
2536 	bd_xfer_impl_t	*xi = (void *)xfer;
2537 	bd_t		*bd = xi->i_bd;
2538 
2539 	switch (error) {
2540 	case BD_ERR_MEDIA:
2541 		atomic_inc_32(&bd->d_kerr->bd_rq_media_err.value.ui32);
2542 		break;
2543 	case BD_ERR_NTRDY:
2544 		atomic_inc_32(&bd->d_kerr->bd_rq_ntrdy_err.value.ui32);
2545 		break;
2546 	case BD_ERR_NODEV:
2547 		atomic_inc_32(&bd->d_kerr->bd_rq_nodev_err.value.ui32);
2548 		break;
2549 	case BD_ERR_RECOV:
2550 		atomic_inc_32(&bd->d_kerr->bd_rq_recov_err.value.ui32);
2551 		break;
2552 	case BD_ERR_ILLRQ:
2553 		atomic_inc_32(&bd->d_kerr->bd_rq_illrq_err.value.ui32);
2554 		break;
2555 	case BD_ERR_PFA:
2556 		atomic_inc_32(&bd->d_kerr->bd_rq_pfa_err.value.ui32);
2557 		break;
2558 	default:
2559 		cmn_err(CE_PANIC, "bd_error: unknown error type %d", error);
2560 		break;
2561 	}
2562 }
2563 
2564 void
2565 bd_state_change(bd_handle_t hdl)
2566 {
2567 	bd_t		*bd;
2568 
2569 	if ((bd = hdl->h_bd) != NULL) {
2570 		bd_update_state(bd);
2571 	}
2572 }
2573 
2574 const char *
2575 bd_address(bd_handle_t hdl)
2576 {
2577 	return (hdl->h_addr);
2578 }
2579 
2580 void
2581 bd_mod_init(struct dev_ops *devops)
2582 {
2583 	static struct bus_ops bd_bus_ops = {
2584 		BUSO_REV,		/* busops_rev */
2585 		nullbusmap,		/* bus_map */
2586 		NULL,			/* bus_get_intrspec (OBSOLETE) */
2587 		NULL,			/* bus_add_intrspec (OBSOLETE) */
2588 		NULL,			/* bus_remove_intrspec (OBSOLETE) */
2589 		i_ddi_map_fault,	/* bus_map_fault */
2590 		NULL,			/* bus_dma_map (OBSOLETE) */
2591 		ddi_dma_allochdl,	/* bus_dma_allochdl */
2592 		ddi_dma_freehdl,	/* bus_dma_freehdl */
2593 		ddi_dma_bindhdl,	/* bus_dma_bindhdl */
2594 		ddi_dma_unbindhdl,	/* bus_dma_unbindhdl */
2595 		ddi_dma_flush,		/* bus_dma_flush */
2596 		ddi_dma_win,		/* bus_dma_win */
2597 		ddi_dma_mctl,		/* bus_dma_ctl */
2598 		bd_bus_ctl,		/* bus_ctl */
2599 		ddi_bus_prop_op,	/* bus_prop_op */
2600 		NULL,			/* bus_get_eventcookie */
2601 		NULL,			/* bus_add_eventcall */
2602 		NULL,			/* bus_remove_eventcall */
2603 		NULL,			/* bus_post_event */
2604 		NULL,			/* bus_intr_ctl (OBSOLETE) */
2605 		NULL,			/* bus_config */
2606 		NULL,			/* bus_unconfig */
2607 		NULL,			/* bus_fm_init */
2608 		NULL,			/* bus_fm_fini */
2609 		NULL,			/* bus_fm_access_enter */
2610 		NULL,			/* bus_fm_access_exit */
2611 		NULL,			/* bus_power */
2612 		NULL,			/* bus_intr_op */
2613 	};
2614 
2615 	devops->devo_bus_ops = &bd_bus_ops;
2616 
2617 	/*
2618 	 * NB: The device driver is free to supply its own
2619 	 * character entry device support.
2620 	 */
2621 }
2622 
2623 void
2624 bd_mod_fini(struct dev_ops *devops)
2625 {
2626 	devops->devo_bus_ops = NULL;
2627 }
2628