xref: /illumos-gate/usr/src/uts/common/io/blkdev/blkdev.c (revision ca0df52a1ec24f9498f363369c65d0e408d579b1)
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  */
30 
31 #include <sys/types.h>
32 #include <sys/ksynch.h>
33 #include <sys/kmem.h>
34 #include <sys/file.h>
35 #include <sys/errno.h>
36 #include <sys/open.h>
37 #include <sys/buf.h>
38 #include <sys/uio.h>
39 #include <sys/aio_req.h>
40 #include <sys/cred.h>
41 #include <sys/modctl.h>
42 #include <sys/cmlb.h>
43 #include <sys/conf.h>
44 #include <sys/devops.h>
45 #include <sys/list.h>
46 #include <sys/sysmacros.h>
47 #include <sys/dkio.h>
48 #include <sys/dkioc_free_util.h>
49 #include <sys/vtoc.h>
50 #include <sys/scsi/scsi.h>	/* for DTYPE_DIRECT */
51 #include <sys/kstat.h>
52 #include <sys/fs/dv_node.h>
53 #include <sys/ddi.h>
54 #include <sys/sunddi.h>
55 #include <sys/note.h>
56 #include <sys/blkdev.h>
57 #include <sys/scsi/impl/inquiry.h>
58 #include <sys/taskq.h>
59 #include <sys/taskq_impl.h>
60 #include <sys/disp.h>
61 #include <sys/sysevent/eventdefs.h>
62 #include <sys/sysevent/dev.h>
63 
64 /*
65  * blkdev is a driver which provides a lot of the common functionality
66  * a block device driver may need and helps by removing code which
67  * is frequently duplicated in block device drivers.
68  *
69  * Within this driver all the struct cb_ops functions required for a
70  * block device driver are written with appropriate call back functions
71  * to be provided by the parent driver.
72  *
73  * To use blkdev, a driver needs to:
74  *	1. Create a bd_ops_t structure which has the call back operations
75  *	   blkdev will use.
76  *	2. Create a handle by calling bd_alloc_handle(). One of the
77  *	   arguments to this function is the bd_ops_t.
78  *	3. Call bd_attach_handle(). This will instantiate a blkdev device
79  *	   as a child device node of the calling driver.
80  *
81  * A parent driver is not restricted to just allocating and attaching a
82  * single instance, it may attach as many as it wishes. For each handle
83  * attached, appropriate entries in /dev/[r]dsk are created.
84  *
85  * The bd_ops_t routines that a parent of blkdev need to provide are:
86  *
87  * o_drive_info: Provide information to blkdev such as how many I/O queues
88  *		 to create and the size of those queues. Also some device
89  *		 specifics such as EUI, vendor, product, model, serial
90  *		 number ....
91  *
92  * o_media_info: Provide information about the media. Eg size and block size.
93  *
94  * o_devid_init: Creates and initializes the device id. Typically calls
95  *		 ddi_devid_init().
96  *
97  * o_sync_cache: Issues a device appropriate command to flush any write
98  *		 caches.
99  *
100  * o_read:	 Read data as described by bd_xfer_t argument.
101  *
102  * o_write:	 Write data as described by bd_xfer_t argument.
103  *
104  * o_free_space: Free the space described by bd_xfer_t argument (optional).
105  *
106  * Queues
107  * ------
108  * Part of the drive_info data is a queue count. blkdev will create
109  * "queue count" number of waitq/runq pairs. Each waitq/runq pair
110  * operates independently. As an I/O is scheduled up to the parent
111  * driver via o_read or o_write its queue number is given. If the
112  * parent driver supports multiple hardware queues it can then select
113  * where to submit the I/O request.
114  *
115  * Currently blkdev uses a simplistic round-robin queue selection method.
116  * It has the advantage that it is lockless. In the future it will be
117  * worthwhile reviewing this strategy for something which prioritizes queues
118  * depending on how busy they are.
119  *
120  * Each waitq/runq pair is protected by its mutex (q_iomutex). Incoming
121  * I/O requests are initially added to the waitq. They are taken off the
122  * waitq, added to the runq and submitted, providing the runq is less
123  * than the qsize as specified in the drive_info. As an I/O request
124  * completes, the parent driver is required to call bd_xfer_done(), which
125  * will remove the I/O request from the runq and pass I/O completion
126  * status up the stack.
127  *
128  * Locks
129  * -----
130  * There are 5 instance global locks d_ocmutex, d_ksmutex, d_errmutex,
131  * d_statemutex and d_dle_mutex. As well a q_iomutex per waitq/runq pair.
132  *
133  * Lock Hierarchy
134  * --------------
135  * The only two locks which may be held simultaneously are q_iomutex and
136  * d_ksmutex. In all cases q_iomutex must be acquired before d_ksmutex.
137  */
138 
139 #define	BD_MAXPART	64
140 #define	BDINST(dev)	(getminor(dev) / BD_MAXPART)
141 #define	BDPART(dev)	(getminor(dev) % BD_MAXPART)
142 
143 typedef struct bd bd_t;
144 typedef struct bd_xfer_impl bd_xfer_impl_t;
145 typedef struct bd_queue bd_queue_t;
146 
147 typedef enum {
148 	BD_DLE_PENDING	= 1 << 0,
149 	BD_DLE_RUNNING	= 1 << 1
150 } bd_dle_state_t;
151 
152 struct bd {
153 	void		*d_private;
154 	dev_info_t	*d_dip;
155 	kmutex_t	d_ocmutex;	/* open/close */
156 	kmutex_t	d_ksmutex;	/* kstat */
157 	kmutex_t	d_errmutex;
158 	kmutex_t	d_statemutex;
159 	kcondvar_t	d_statecv;
160 	enum dkio_state	d_state;
161 	cmlb_handle_t	d_cmlbh;
162 	unsigned	d_open_lyr[BD_MAXPART];	/* open count */
163 	uint64_t	d_open_excl;	/* bit mask indexed by partition */
164 	uint64_t	d_open_reg[OTYPCNT];		/* bit mask */
165 	uint64_t	d_io_counter;
166 
167 	uint32_t	d_qcount;
168 	uint32_t	d_qactive;
169 	uint32_t	d_maxxfer;
170 	uint32_t	d_blkshift;
171 	uint32_t	d_pblkshift;
172 	uint64_t	d_numblks;
173 	ddi_devid_t	d_devid;
174 
175 	uint64_t	d_max_free_seg;
176 	uint64_t	d_max_free_blks;
177 	uint64_t	d_max_free_seg_blks;
178 	uint64_t	d_free_align;
179 
180 	kmem_cache_t	*d_cache;
181 	bd_queue_t	*d_queues;
182 	kstat_t		*d_ksp;
183 	kstat_io_t	*d_kiop;
184 	kstat_t		*d_errstats;
185 	struct bd_errstats *d_kerr;
186 
187 	boolean_t	d_rdonly;
188 	boolean_t	d_ssd;
189 	boolean_t	d_removable;
190 	boolean_t	d_hotpluggable;
191 	boolean_t	d_use_dma;
192 
193 	ddi_dma_attr_t	d_dma;
194 	bd_ops_t	d_ops;
195 	bd_handle_t	d_handle;
196 
197 	kmutex_t	d_dle_mutex;
198 	taskq_ent_t	d_dle_ent;
199 	bd_dle_state_t	d_dle_state;
200 };
201 
202 struct bd_handle {
203 	bd_ops_t	h_ops;
204 	ddi_dma_attr_t	*h_dma;
205 	dev_info_t	*h_parent;
206 	dev_info_t	*h_child;
207 	void		*h_private;
208 	bd_t		*h_bd;
209 	char		*h_name;
210 	char		h_addr[50];	/* enough for w%0.32x,%X */
211 };
212 
213 struct bd_xfer_impl {
214 	bd_xfer_t	i_public;
215 	list_node_t	i_linkage;
216 	bd_t		*i_bd;
217 	buf_t		*i_bp;
218 	bd_queue_t	*i_bq;
219 	uint_t		i_num_win;
220 	uint_t		i_cur_win;
221 	off_t		i_offset;
222 	int		(*i_func)(void *, bd_xfer_t *);
223 	uint32_t	i_blkshift;
224 	size_t		i_len;
225 	size_t		i_resid;
226 };
227 
228 struct bd_queue {
229 	kmutex_t	q_iomutex;
230 	uint32_t	q_qsize;
231 	uint32_t	q_qactive;
232 	list_t		q_runq;
233 	list_t		q_waitq;
234 };
235 
236 #define	i_dmah		i_public.x_dmah
237 #define	i_dmac		i_public.x_dmac
238 #define	i_ndmac		i_public.x_ndmac
239 #define	i_kaddr		i_public.x_kaddr
240 #define	i_nblks		i_public.x_nblks
241 #define	i_blkno		i_public.x_blkno
242 #define	i_flags		i_public.x_flags
243 #define	i_qnum		i_public.x_qnum
244 #define	i_dfl		i_public.x_dfl
245 
246 #define	CAN_FREESPACE(bd) \
247 	(((bd)->d_ops.o_free_space == NULL) ? B_FALSE : B_TRUE)
248 
249 /*
250  * Private prototypes.
251  */
252 
253 static void bd_prop_update_inqstring(dev_info_t *, char *, char *, size_t);
254 static void bd_create_inquiry_props(dev_info_t *, bd_drive_t *);
255 static void bd_create_errstats(bd_t *, int, bd_drive_t *);
256 static void bd_destroy_errstats(bd_t *);
257 static void bd_errstats_setstr(kstat_named_t *, char *, size_t, char *);
258 static void bd_init_errstats(bd_t *, bd_drive_t *);
259 static void bd_fini_errstats(bd_t *);
260 
261 static int bd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
262 static int bd_attach(dev_info_t *, ddi_attach_cmd_t);
263 static int bd_detach(dev_info_t *, ddi_detach_cmd_t);
264 
265 static int bd_open(dev_t *, int, int, cred_t *);
266 static int bd_close(dev_t, int, int, cred_t *);
267 static int bd_strategy(struct buf *);
268 static int bd_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
269 static int bd_dump(dev_t, caddr_t, daddr_t, int);
270 static int bd_read(dev_t, struct uio *, cred_t *);
271 static int bd_write(dev_t, struct uio *, cred_t *);
272 static int bd_aread(dev_t, struct aio_req *, cred_t *);
273 static int bd_awrite(dev_t, struct aio_req *, cred_t *);
274 static int bd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *,
275     caddr_t, int *);
276 
277 static int bd_tg_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t,
278     void *);
279 static int bd_tg_getinfo(dev_info_t *, int, void *, void *);
280 static int bd_xfer_ctor(void *, void *, int);
281 static void bd_xfer_dtor(void *, void *);
282 static void bd_sched(bd_t *, bd_queue_t *);
283 static void bd_submit(bd_t *, bd_xfer_impl_t *);
284 static void bd_runq_exit(bd_xfer_impl_t *, int);
285 static void bd_update_state(bd_t *);
286 static int bd_check_state(bd_t *, enum dkio_state *);
287 static int bd_flush_write_cache(bd_t *, struct dk_callback *);
288 static int bd_check_uio(dev_t, struct uio *);
289 static int bd_free_space(dev_t, bd_t *, dkioc_free_list_t *);
290 
291 struct cmlb_tg_ops bd_tg_ops = {
292 	TG_DK_OPS_VERSION_1,
293 	bd_tg_rdwr,
294 	bd_tg_getinfo,
295 };
296 
297 static struct cb_ops bd_cb_ops = {
298 	bd_open,		/* open */
299 	bd_close,		/* close */
300 	bd_strategy,		/* strategy */
301 	nodev,			/* print */
302 	bd_dump,		/* dump */
303 	bd_read,		/* read */
304 	bd_write,		/* write */
305 	bd_ioctl,		/* ioctl */
306 	nodev,			/* devmap */
307 	nodev,			/* mmap */
308 	nodev,			/* segmap */
309 	nochpoll,		/* poll */
310 	bd_prop_op,		/* cb_prop_op */
311 	0,			/* streamtab  */
312 	D_64BIT | D_MP,		/* Driver comaptibility flag */
313 	CB_REV,			/* cb_rev */
314 	bd_aread,		/* async read */
315 	bd_awrite		/* async write */
316 };
317 
318 struct dev_ops bd_dev_ops = {
319 	DEVO_REV,		/* devo_rev, */
320 	0,			/* refcnt  */
321 	bd_getinfo,		/* getinfo */
322 	nulldev,		/* identify */
323 	nulldev,		/* probe */
324 	bd_attach,		/* attach */
325 	bd_detach,		/* detach */
326 	nodev,			/* reset */
327 	&bd_cb_ops,		/* driver operations */
328 	NULL,			/* bus operations */
329 	NULL,			/* power */
330 	ddi_quiesce_not_needed,	/* quiesce */
331 };
332 
333 static struct modldrv modldrv = {
334 	&mod_driverops,
335 	"Generic Block Device",
336 	&bd_dev_ops,
337 };
338 
339 static struct modlinkage modlinkage = {
340 	MODREV_1, { &modldrv, NULL }
341 };
342 
343 static void *bd_state;
344 static krwlock_t bd_lock;
345 static taskq_t *bd_taskq;
346 
347 int
348 _init(void)
349 {
350 	char taskq_name[TASKQ_NAMELEN];
351 	const char *name;
352 	int rv;
353 
354 	rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2);
355 	if (rv != DDI_SUCCESS)
356 		return (rv);
357 
358 	name = mod_modname(&modlinkage);
359 	(void) snprintf(taskq_name, sizeof (taskq_name), "%s_taskq", name);
360 	bd_taskq = taskq_create(taskq_name, 1, minclsyspri, 0, 0, 0);
361 	if (bd_taskq == NULL) {
362 		cmn_err(CE_WARN, "%s: unable to create %s", name, taskq_name);
363 		ddi_soft_state_fini(&bd_state);
364 		return (DDI_FAILURE);
365 	}
366 
367 	rw_init(&bd_lock, NULL, RW_DRIVER, NULL);
368 
369 	rv = mod_install(&modlinkage);
370 	if (rv != DDI_SUCCESS) {
371 		rw_destroy(&bd_lock);
372 		taskq_destroy(bd_taskq);
373 		ddi_soft_state_fini(&bd_state);
374 	}
375 	return (rv);
376 }
377 
378 int
379 _fini(void)
380 {
381 	int	rv;
382 
383 	rv = mod_remove(&modlinkage);
384 	if (rv == DDI_SUCCESS) {
385 		rw_destroy(&bd_lock);
386 		taskq_destroy(bd_taskq);
387 		ddi_soft_state_fini(&bd_state);
388 	}
389 	return (rv);
390 }
391 
392 int
393 _info(struct modinfo *modinfop)
394 {
395 	return (mod_info(&modlinkage, modinfop));
396 }
397 
398 static int
399 bd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resultp)
400 {
401 	bd_t	*bd;
402 	minor_t	inst;
403 
404 	_NOTE(ARGUNUSED(dip));
405 
406 	inst = BDINST((dev_t)arg);
407 
408 	switch (cmd) {
409 	case DDI_INFO_DEVT2DEVINFO:
410 		bd = ddi_get_soft_state(bd_state, inst);
411 		if (bd == NULL) {
412 			return (DDI_FAILURE);
413 		}
414 		*resultp = (void *)bd->d_dip;
415 		break;
416 
417 	case DDI_INFO_DEVT2INSTANCE:
418 		*resultp = (void *)(intptr_t)inst;
419 		break;
420 
421 	default:
422 		return (DDI_FAILURE);
423 	}
424 	return (DDI_SUCCESS);
425 }
426 
427 static void
428 bd_prop_update_inqstring(dev_info_t *dip, char *name, char *data, size_t len)
429 {
430 	int	ilen;
431 	char	*data_string;
432 
433 	ilen = scsi_ascii_inquiry_len(data, len);
434 	ASSERT3U(ilen, <=, len);
435 	if (ilen <= 0)
436 		return;
437 	/* ensure null termination */
438 	data_string = kmem_zalloc(ilen + 1, KM_SLEEP);
439 	bcopy(data, data_string, ilen);
440 	(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip, name, data_string);
441 	kmem_free(data_string, ilen + 1);
442 }
443 
444 static void
445 bd_create_inquiry_props(dev_info_t *dip, bd_drive_t *drive)
446 {
447 	if (drive->d_vendor_len > 0)
448 		bd_prop_update_inqstring(dip, INQUIRY_VENDOR_ID,
449 		    drive->d_vendor, drive->d_vendor_len);
450 
451 	if (drive->d_product_len > 0)
452 		bd_prop_update_inqstring(dip, INQUIRY_PRODUCT_ID,
453 		    drive->d_product, drive->d_product_len);
454 
455 	if (drive->d_serial_len > 0)
456 		bd_prop_update_inqstring(dip, INQUIRY_SERIAL_NO,
457 		    drive->d_serial, drive->d_serial_len);
458 
459 	if (drive->d_revision_len > 0)
460 		bd_prop_update_inqstring(dip, INQUIRY_REVISION_ID,
461 		    drive->d_revision, drive->d_revision_len);
462 }
463 
464 static void
465 bd_create_errstats(bd_t *bd, int inst, bd_drive_t *drive)
466 {
467 	char	ks_module[KSTAT_STRLEN];
468 	char	ks_name[KSTAT_STRLEN];
469 	int	ndata = sizeof (struct bd_errstats) / sizeof (kstat_named_t);
470 
471 	if (bd->d_errstats != NULL)
472 		return;
473 
474 	(void) snprintf(ks_module, sizeof (ks_module), "%serr",
475 	    ddi_driver_name(bd->d_dip));
476 	(void) snprintf(ks_name, sizeof (ks_name), "%s%d,err",
477 	    ddi_driver_name(bd->d_dip), inst);
478 
479 	bd->d_errstats = kstat_create(ks_module, inst, ks_name, "device_error",
480 	    KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
481 
482 	mutex_init(&bd->d_errmutex, NULL, MUTEX_DRIVER, NULL);
483 	if (bd->d_errstats == NULL) {
484 		/*
485 		 * Even if we cannot create the kstat, we create a
486 		 * scratch kstat.  The reason for this is to ensure
487 		 * that we can update the kstat all of the time,
488 		 * without adding an extra branch instruction.
489 		 */
490 		bd->d_kerr = kmem_zalloc(sizeof (struct bd_errstats),
491 		    KM_SLEEP);
492 	} else {
493 		bd->d_errstats->ks_lock = &bd->d_errmutex;
494 		bd->d_kerr = (struct bd_errstats *)bd->d_errstats->ks_data;
495 	}
496 
497 	kstat_named_init(&bd->d_kerr->bd_softerrs,	"Soft Errors",
498 	    KSTAT_DATA_UINT32);
499 	kstat_named_init(&bd->d_kerr->bd_harderrs,	"Hard Errors",
500 	    KSTAT_DATA_UINT32);
501 	kstat_named_init(&bd->d_kerr->bd_transerrs,	"Transport Errors",
502 	    KSTAT_DATA_UINT32);
503 
504 	if (drive->d_model_len > 0) {
505 		kstat_named_init(&bd->d_kerr->bd_model,	"Model",
506 		    KSTAT_DATA_STRING);
507 	} else {
508 		kstat_named_init(&bd->d_kerr->bd_vid,	"Vendor",
509 		    KSTAT_DATA_STRING);
510 		kstat_named_init(&bd->d_kerr->bd_pid,	"Product",
511 		    KSTAT_DATA_STRING);
512 	}
513 
514 	kstat_named_init(&bd->d_kerr->bd_revision,	"Revision",
515 	    KSTAT_DATA_STRING);
516 	kstat_named_init(&bd->d_kerr->bd_serial,	"Serial No",
517 	    KSTAT_DATA_STRING);
518 	kstat_named_init(&bd->d_kerr->bd_capacity,	"Size",
519 	    KSTAT_DATA_ULONGLONG);
520 	kstat_named_init(&bd->d_kerr->bd_rq_media_err,	"Media Error",
521 	    KSTAT_DATA_UINT32);
522 	kstat_named_init(&bd->d_kerr->bd_rq_ntrdy_err,	"Device Not Ready",
523 	    KSTAT_DATA_UINT32);
524 	kstat_named_init(&bd->d_kerr->bd_rq_nodev_err,	"No Device",
525 	    KSTAT_DATA_UINT32);
526 	kstat_named_init(&bd->d_kerr->bd_rq_recov_err,	"Recoverable",
527 	    KSTAT_DATA_UINT32);
528 	kstat_named_init(&bd->d_kerr->bd_rq_illrq_err,	"Illegal Request",
529 	    KSTAT_DATA_UINT32);
530 	kstat_named_init(&bd->d_kerr->bd_rq_pfa_err,
531 	    "Predictive Failure Analysis", KSTAT_DATA_UINT32);
532 
533 	bd->d_errstats->ks_private = bd;
534 
535 	kstat_install(bd->d_errstats);
536 	bd_init_errstats(bd, drive);
537 }
538 
539 static void
540 bd_destroy_errstats(bd_t *bd)
541 {
542 	if (bd->d_errstats != NULL) {
543 		bd_fini_errstats(bd);
544 		kstat_delete(bd->d_errstats);
545 		bd->d_errstats = NULL;
546 	} else {
547 		kmem_free(bd->d_kerr, sizeof (struct bd_errstats));
548 		bd->d_kerr = NULL;
549 		mutex_destroy(&bd->d_errmutex);
550 	}
551 }
552 
553 static void
554 bd_errstats_setstr(kstat_named_t *k, char *str, size_t len, char *alt)
555 {
556 	char	*tmp;
557 	size_t	km_len;
558 
559 	if (KSTAT_NAMED_STR_PTR(k) == NULL) {
560 		if (len > 0)
561 			km_len = strnlen(str, len);
562 		else if (alt != NULL)
563 			km_len = strlen(alt);
564 		else
565 			return;
566 
567 		tmp = kmem_alloc(km_len + 1, KM_SLEEP);
568 		bcopy(len > 0 ? str : alt, tmp, km_len);
569 		tmp[km_len] = '\0';
570 
571 		kstat_named_setstr(k, tmp);
572 	}
573 }
574 
575 static void
576 bd_errstats_clrstr(kstat_named_t *k)
577 {
578 	if (KSTAT_NAMED_STR_PTR(k) == NULL)
579 		return;
580 
581 	kmem_free(KSTAT_NAMED_STR_PTR(k), KSTAT_NAMED_STR_BUFLEN(k));
582 	kstat_named_setstr(k, NULL);
583 }
584 
585 static void
586 bd_init_errstats(bd_t *bd, bd_drive_t *drive)
587 {
588 	struct bd_errstats	*est = bd->d_kerr;
589 
590 	mutex_enter(&bd->d_errmutex);
591 
592 	if (drive->d_model_len > 0 &&
593 	    KSTAT_NAMED_STR_PTR(&est->bd_model) == NULL) {
594 		bd_errstats_setstr(&est->bd_model, drive->d_model,
595 		    drive->d_model_len, NULL);
596 	} else {
597 		bd_errstats_setstr(&est->bd_vid, drive->d_vendor,
598 		    drive->d_vendor_len, "Unknown ");
599 		bd_errstats_setstr(&est->bd_pid, drive->d_product,
600 		    drive->d_product_len, "Unknown         ");
601 	}
602 
603 	bd_errstats_setstr(&est->bd_revision, drive->d_revision,
604 	    drive->d_revision_len, "0001");
605 	bd_errstats_setstr(&est->bd_serial, drive->d_serial,
606 	    drive->d_serial_len, "0               ");
607 
608 	mutex_exit(&bd->d_errmutex);
609 }
610 
611 static void
612 bd_fini_errstats(bd_t *bd)
613 {
614 	struct bd_errstats	*est = bd->d_kerr;
615 
616 	mutex_enter(&bd->d_errmutex);
617 
618 	bd_errstats_clrstr(&est->bd_model);
619 	bd_errstats_clrstr(&est->bd_vid);
620 	bd_errstats_clrstr(&est->bd_pid);
621 	bd_errstats_clrstr(&est->bd_revision);
622 	bd_errstats_clrstr(&est->bd_serial);
623 
624 	mutex_exit(&bd->d_errmutex);
625 }
626 
627 static void
628 bd_queues_free(bd_t *bd)
629 {
630 	uint32_t i;
631 
632 	for (i = 0; i < bd->d_qcount; i++) {
633 		bd_queue_t *bq = &bd->d_queues[i];
634 
635 		mutex_destroy(&bq->q_iomutex);
636 		list_destroy(&bq->q_waitq);
637 		list_destroy(&bq->q_runq);
638 	}
639 
640 	kmem_free(bd->d_queues, sizeof (*bd->d_queues) * bd->d_qcount);
641 }
642 
643 static int
644 bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
645 {
646 	int		inst;
647 	bd_handle_t	hdl;
648 	bd_t		*bd;
649 	bd_drive_t	drive;
650 	uint32_t	i;
651 	int		rv;
652 	char		name[16];
653 	char		kcache[32];
654 	char		*node_type;
655 
656 	switch (cmd) {
657 	case DDI_ATTACH:
658 		break;
659 	case DDI_RESUME:
660 		/* We don't do anything native for suspend/resume */
661 		return (DDI_SUCCESS);
662 	default:
663 		return (DDI_FAILURE);
664 	}
665 
666 	inst = ddi_get_instance(dip);
667 	hdl = ddi_get_parent_data(dip);
668 
669 	(void) snprintf(name, sizeof (name), "%s%d",
670 	    ddi_driver_name(dip), ddi_get_instance(dip));
671 	(void) snprintf(kcache, sizeof (kcache), "%s_xfer", name);
672 
673 	if (hdl == NULL) {
674 		cmn_err(CE_WARN, "%s: missing parent data!", name);
675 		return (DDI_FAILURE);
676 	}
677 
678 	if (ddi_soft_state_zalloc(bd_state, inst) != DDI_SUCCESS) {
679 		cmn_err(CE_WARN, "%s: unable to zalloc soft state!", name);
680 		return (DDI_FAILURE);
681 	}
682 	bd = ddi_get_soft_state(bd_state, inst);
683 
684 	if (hdl->h_dma) {
685 		bd->d_dma = *(hdl->h_dma);
686 		bd->d_dma.dma_attr_granular =
687 		    max(DEV_BSIZE, bd->d_dma.dma_attr_granular);
688 		bd->d_use_dma = B_TRUE;
689 
690 		if (bd->d_maxxfer &&
691 		    (bd->d_maxxfer != bd->d_dma.dma_attr_maxxfer)) {
692 			cmn_err(CE_WARN,
693 			    "%s: inconsistent maximum transfer size!",
694 			    name);
695 			/* We force it */
696 			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
697 		} else {
698 			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
699 		}
700 	} else {
701 		bd->d_use_dma = B_FALSE;
702 		if (bd->d_maxxfer == 0) {
703 			bd->d_maxxfer = 1024 * 1024;
704 		}
705 	}
706 	bd->d_ops = hdl->h_ops;
707 	bd->d_private = hdl->h_private;
708 	bd->d_blkshift = DEV_BSHIFT;	/* 512 bytes, to start */
709 
710 	if (bd->d_maxxfer % DEV_BSIZE) {
711 		cmn_err(CE_WARN, "%s: maximum transfer misaligned!", name);
712 		bd->d_maxxfer &= ~(DEV_BSIZE - 1);
713 	}
714 	if (bd->d_maxxfer < DEV_BSIZE) {
715 		cmn_err(CE_WARN, "%s: maximum transfer size too small!", name);
716 		ddi_soft_state_free(bd_state, inst);
717 		return (DDI_FAILURE);
718 	}
719 
720 	bd->d_dip = dip;
721 	bd->d_handle = hdl;
722 	ddi_set_driver_private(dip, bd);
723 
724 	mutex_init(&bd->d_ksmutex, NULL, MUTEX_DRIVER, NULL);
725 	mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL);
726 	mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL);
727 	cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL);
728 	mutex_init(&bd->d_dle_mutex, NULL, MUTEX_DRIVER, NULL);
729 	bd->d_dle_state = 0;
730 
731 	bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8,
732 	    bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0);
733 
734 	bd->d_ksp = kstat_create(ddi_driver_name(dip), inst, NULL, "disk",
735 	    KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
736 	if (bd->d_ksp != NULL) {
737 		bd->d_ksp->ks_lock = &bd->d_ksmutex;
738 		kstat_install(bd->d_ksp);
739 		bd->d_kiop = bd->d_ksp->ks_data;
740 	} else {
741 		/*
742 		 * Even if we cannot create the kstat, we create a
743 		 * scratch kstat.  The reason for this is to ensure
744 		 * that we can update the kstat all of the time,
745 		 * without adding an extra branch instruction.
746 		 */
747 		bd->d_kiop = kmem_zalloc(sizeof (kstat_io_t), KM_SLEEP);
748 	}
749 
750 	cmlb_alloc_handle(&bd->d_cmlbh);
751 
752 	bd->d_state = DKIO_NONE;
753 
754 	bzero(&drive, sizeof (drive));
755 	/*
756 	 * Default to one queue, and no restrictions on free space requests
757 	 * (if driver provides method) parent driver can override.
758 	 */
759 	drive.d_qcount = 1;
760 	drive.d_free_align = 1;
761 	bd->d_ops.o_drive_info(bd->d_private, &drive);
762 
763 	/*
764 	 * Several checks to make sure o_drive_info() didn't return bad
765 	 * values:
766 	 *
767 	 * There must be at least one queue
768 	 */
769 	if (drive.d_qcount == 0)
770 		goto fail_drive_info;
771 
772 	/* FREE/UNMAP/TRIM alignment needs to be at least 1 block */
773 	if (drive.d_free_align == 0)
774 		goto fail_drive_info;
775 
776 	/*
777 	 * If d_max_free_blks is not unlimited (not 0), then we cannot allow
778 	 * an unlimited segment size. It is however permissible to not impose
779 	 * a limit on the total number of blocks freed while limiting the
780 	 * amount allowed in an individual segment.
781 	 */
782 	if ((drive.d_max_free_blks > 0 && drive.d_max_free_seg_blks == 0))
783 		goto fail_drive_info;
784 
785 	/*
786 	 * If a limit is set on d_max_free_blks (by the above check, we know
787 	 * if there's a limit on d_max_free_blks, d_max_free_seg_blks cannot
788 	 * be unlimited), it cannot be smaller than the limit on an individual
789 	 * segment.
790 	 */
791 	if ((drive.d_max_free_blks > 0 &&
792 	    drive.d_max_free_seg_blks > drive.d_max_free_blks)) {
793 		goto fail_drive_info;
794 	}
795 
796 	bd->d_qcount = drive.d_qcount;
797 	bd->d_removable = drive.d_removable;
798 	bd->d_hotpluggable = drive.d_hotpluggable;
799 
800 	if (drive.d_maxxfer && drive.d_maxxfer < bd->d_maxxfer)
801 		bd->d_maxxfer = drive.d_maxxfer;
802 
803 	bd->d_free_align = drive.d_free_align;
804 	bd->d_max_free_seg = drive.d_max_free_seg;
805 	bd->d_max_free_blks = drive.d_max_free_blks;
806 	bd->d_max_free_seg_blks = drive.d_max_free_seg_blks;
807 
808 	bd_create_inquiry_props(dip, &drive);
809 	bd_create_errstats(bd, inst, &drive);
810 	bd_update_state(bd);
811 
812 	bd->d_queues = kmem_alloc(sizeof (*bd->d_queues) * bd->d_qcount,
813 	    KM_SLEEP);
814 	for (i = 0; i < bd->d_qcount; i++) {
815 		bd_queue_t *bq = &bd->d_queues[i];
816 
817 		bq->q_qsize = drive.d_qsize;
818 		bq->q_qactive = 0;
819 		mutex_init(&bq->q_iomutex, NULL, MUTEX_DRIVER, NULL);
820 
821 		list_create(&bq->q_waitq, sizeof (bd_xfer_impl_t),
822 		    offsetof(struct bd_xfer_impl, i_linkage));
823 		list_create(&bq->q_runq, sizeof (bd_xfer_impl_t),
824 		    offsetof(struct bd_xfer_impl, i_linkage));
825 	}
826 
827 	if (*(uint64_t *)drive.d_eui64 != 0 ||
828 	    *(uint64_t *)drive.d_guid != 0 ||
829 	    *((uint64_t *)drive.d_guid + 1) != 0)
830 		node_type = DDI_NT_BLOCK_BLKDEV;
831 	else if (drive.d_lun >= 0)
832 		node_type = DDI_NT_BLOCK_CHAN;
833 	else
834 		node_type = DDI_NT_BLOCK;
835 
836 	rv = cmlb_attach(dip, &bd_tg_ops, DTYPE_DIRECT,
837 	    bd->d_removable, bd->d_hotpluggable, node_type,
838 	    CMLB_FAKE_LABEL_ONE_PARTITION, bd->d_cmlbh, 0);
839 	if (rv != 0) {
840 		goto fail_cmlb_attach;
841 	}
842 
843 	if (bd->d_ops.o_devid_init != NULL) {
844 		rv = bd->d_ops.o_devid_init(bd->d_private, dip, &bd->d_devid);
845 		if (rv == DDI_SUCCESS) {
846 			if (ddi_devid_register(dip, bd->d_devid) !=
847 			    DDI_SUCCESS) {
848 				cmn_err(CE_WARN,
849 				    "%s: unable to register devid", name);
850 			}
851 		}
852 	}
853 
854 	/*
855 	 * Add a zero-length attribute to tell the world we support
856 	 * kernel ioctls (for layered drivers).  Also set up properties
857 	 * used by HAL to identify removable media.
858 	 */
859 	(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
860 	    DDI_KERNEL_IOCTL, NULL, 0);
861 	if (bd->d_removable) {
862 		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
863 		    "removable-media", NULL, 0);
864 	}
865 	if (bd->d_hotpluggable) {
866 		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
867 		    "hotpluggable", NULL, 0);
868 	}
869 
870 	hdl->h_bd = bd;
871 	ddi_report_dev(dip);
872 
873 	return (DDI_SUCCESS);
874 
875 fail_cmlb_attach:
876 	bd_queues_free(bd);
877 	bd_destroy_errstats(bd);
878 
879 fail_drive_info:
880 	cmlb_free_handle(&bd->d_cmlbh);
881 
882 	if (bd->d_ksp != NULL) {
883 		kstat_delete(bd->d_ksp);
884 		bd->d_ksp = NULL;
885 	} else {
886 		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
887 	}
888 
889 	kmem_cache_destroy(bd->d_cache);
890 	cv_destroy(&bd->d_statecv);
891 	mutex_destroy(&bd->d_statemutex);
892 	mutex_destroy(&bd->d_ocmutex);
893 	mutex_destroy(&bd->d_ksmutex);
894 	mutex_destroy(&bd->d_dle_mutex);
895 	ddi_soft_state_free(bd_state, inst);
896 	return (DDI_FAILURE);
897 }
898 
899 static int
900 bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
901 {
902 	bd_handle_t	hdl;
903 	bd_t		*bd;
904 
905 	bd = ddi_get_driver_private(dip);
906 	hdl = ddi_get_parent_data(dip);
907 
908 	switch (cmd) {
909 	case DDI_DETACH:
910 		break;
911 	case DDI_SUSPEND:
912 		/* We don't suspend, but our parent does */
913 		return (DDI_SUCCESS);
914 	default:
915 		return (DDI_FAILURE);
916 	}
917 
918 	hdl->h_bd = NULL;
919 
920 	if (bd->d_ksp != NULL) {
921 		kstat_delete(bd->d_ksp);
922 		bd->d_ksp = NULL;
923 	} else {
924 		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
925 	}
926 
927 	bd_destroy_errstats(bd);
928 	cmlb_detach(bd->d_cmlbh, 0);
929 	cmlb_free_handle(&bd->d_cmlbh);
930 	if (bd->d_devid)
931 		ddi_devid_free(bd->d_devid);
932 	kmem_cache_destroy(bd->d_cache);
933 	mutex_destroy(&bd->d_ksmutex);
934 	mutex_destroy(&bd->d_ocmutex);
935 	mutex_destroy(&bd->d_statemutex);
936 	cv_destroy(&bd->d_statecv);
937 	mutex_destroy(&bd->d_dle_mutex);
938 	bd_queues_free(bd);
939 	ddi_soft_state_free(bd_state, ddi_get_instance(dip));
940 	return (DDI_SUCCESS);
941 }
942 
943 static int
944 bd_xfer_ctor(void *buf, void *arg, int kmflag)
945 {
946 	bd_xfer_impl_t	*xi;
947 	bd_t		*bd = arg;
948 	int		(*dcb)(caddr_t);
949 
950 	if (kmflag == KM_PUSHPAGE || kmflag == KM_SLEEP) {
951 		dcb = DDI_DMA_SLEEP;
952 	} else {
953 		dcb = DDI_DMA_DONTWAIT;
954 	}
955 
956 	xi = buf;
957 	bzero(xi, sizeof (*xi));
958 	xi->i_bd = bd;
959 
960 	if (bd->d_use_dma) {
961 		if (ddi_dma_alloc_handle(bd->d_dip, &bd->d_dma, dcb, NULL,
962 		    &xi->i_dmah) != DDI_SUCCESS) {
963 			return (-1);
964 		}
965 	}
966 
967 	return (0);
968 }
969 
970 static void
971 bd_xfer_dtor(void *buf, void *arg)
972 {
973 	bd_xfer_impl_t	*xi = buf;
974 
975 	_NOTE(ARGUNUSED(arg));
976 
977 	if (xi->i_dmah)
978 		ddi_dma_free_handle(&xi->i_dmah);
979 	xi->i_dmah = NULL;
980 }
981 
982 static bd_xfer_impl_t *
983 bd_xfer_alloc(bd_t *bd, struct buf *bp, int (*func)(void *, bd_xfer_t *),
984     int kmflag)
985 {
986 	bd_xfer_impl_t		*xi;
987 	int			rv = 0;
988 	int			status;
989 	unsigned		dir;
990 	int			(*cb)(caddr_t);
991 	size_t			len;
992 	uint32_t		shift;
993 
994 	if (kmflag == KM_SLEEP) {
995 		cb = DDI_DMA_SLEEP;
996 	} else {
997 		cb = DDI_DMA_DONTWAIT;
998 	}
999 
1000 	xi = kmem_cache_alloc(bd->d_cache, kmflag);
1001 	if (xi == NULL) {
1002 		bioerror(bp, ENOMEM);
1003 		return (NULL);
1004 	}
1005 
1006 	ASSERT(bp);
1007 
1008 	xi->i_bp = bp;
1009 	xi->i_func = func;
1010 	xi->i_blkno = bp->b_lblkno >> (bd->d_blkshift - DEV_BSHIFT);
1011 
1012 	if (bp->b_bcount == 0) {
1013 		xi->i_len = 0;
1014 		xi->i_nblks = 0;
1015 		xi->i_kaddr = NULL;
1016 		xi->i_resid = 0;
1017 		xi->i_num_win = 0;
1018 		goto done;
1019 	}
1020 
1021 	if (bp->b_flags & B_READ) {
1022 		dir = DDI_DMA_READ;
1023 		xi->i_func = bd->d_ops.o_read;
1024 	} else {
1025 		dir = DDI_DMA_WRITE;
1026 		xi->i_func = bd->d_ops.o_write;
1027 	}
1028 
1029 	shift = bd->d_blkshift;
1030 	xi->i_blkshift = shift;
1031 
1032 	if (!bd->d_use_dma) {
1033 		bp_mapin(bp);
1034 		rv = 0;
1035 		xi->i_offset = 0;
1036 		xi->i_num_win =
1037 		    (bp->b_bcount + (bd->d_maxxfer - 1)) / bd->d_maxxfer;
1038 		xi->i_cur_win = 0;
1039 		xi->i_len = min(bp->b_bcount, bd->d_maxxfer);
1040 		xi->i_nblks = xi->i_len >> shift;
1041 		xi->i_kaddr = bp->b_un.b_addr;
1042 		xi->i_resid = bp->b_bcount;
1043 	} else {
1044 
1045 		/*
1046 		 * We have to use consistent DMA if the address is misaligned.
1047 		 */
1048 		if (((bp->b_flags & (B_PAGEIO | B_REMAPPED)) != B_PAGEIO) &&
1049 		    ((uintptr_t)bp->b_un.b_addr & 0x7)) {
1050 			dir |= DDI_DMA_CONSISTENT | DDI_DMA_PARTIAL;
1051 		} else {
1052 			dir |= DDI_DMA_STREAMING | DDI_DMA_PARTIAL;
1053 		}
1054 
1055 		status = ddi_dma_buf_bind_handle(xi->i_dmah, bp, dir, cb,
1056 		    NULL, &xi->i_dmac, &xi->i_ndmac);
1057 		switch (status) {
1058 		case DDI_DMA_MAPPED:
1059 			xi->i_num_win = 1;
1060 			xi->i_cur_win = 0;
1061 			xi->i_offset = 0;
1062 			xi->i_len = bp->b_bcount;
1063 			xi->i_nblks = xi->i_len >> shift;
1064 			xi->i_resid = bp->b_bcount;
1065 			rv = 0;
1066 			break;
1067 		case DDI_DMA_PARTIAL_MAP:
1068 			xi->i_cur_win = 0;
1069 
1070 			if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) !=
1071 			    DDI_SUCCESS) ||
1072 			    (ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset,
1073 			    &len, &xi->i_dmac, &xi->i_ndmac) !=
1074 			    DDI_SUCCESS) ||
1075 			    (P2PHASE(len, (1U << shift)) != 0)) {
1076 				(void) ddi_dma_unbind_handle(xi->i_dmah);
1077 				rv = EFAULT;
1078 				goto done;
1079 			}
1080 			xi->i_len = len;
1081 			xi->i_nblks = xi->i_len >> shift;
1082 			xi->i_resid = bp->b_bcount;
1083 			rv = 0;
1084 			break;
1085 		case DDI_DMA_NORESOURCES:
1086 			rv = EAGAIN;
1087 			goto done;
1088 		case DDI_DMA_TOOBIG:
1089 			rv = EINVAL;
1090 			goto done;
1091 		case DDI_DMA_NOMAPPING:
1092 		case DDI_DMA_INUSE:
1093 		default:
1094 			rv = EFAULT;
1095 			goto done;
1096 		}
1097 	}
1098 
1099 done:
1100 	if (rv != 0) {
1101 		kmem_cache_free(bd->d_cache, xi);
1102 		bioerror(bp, rv);
1103 		return (NULL);
1104 	}
1105 
1106 	return (xi);
1107 }
1108 
1109 static void
1110 bd_xfer_free(bd_xfer_impl_t *xi)
1111 {
1112 	if (xi->i_dmah) {
1113 		(void) ddi_dma_unbind_handle(xi->i_dmah);
1114 	}
1115 	if (xi->i_dfl != NULL) {
1116 		dfl_free((dkioc_free_list_t *)xi->i_dfl);
1117 		xi->i_dfl = NULL;
1118 	}
1119 	kmem_cache_free(xi->i_bd->d_cache, xi);
1120 }
1121 
1122 static int
1123 bd_open(dev_t *devp, int flag, int otyp, cred_t *credp)
1124 {
1125 	dev_t		dev = *devp;
1126 	bd_t		*bd;
1127 	minor_t		part;
1128 	minor_t		inst;
1129 	uint64_t	mask;
1130 	boolean_t	ndelay;
1131 	int		rv;
1132 	diskaddr_t	nblks;
1133 	diskaddr_t	lba;
1134 
1135 	_NOTE(ARGUNUSED(credp));
1136 
1137 	part = BDPART(dev);
1138 	inst = BDINST(dev);
1139 
1140 	if (otyp >= OTYPCNT)
1141 		return (EINVAL);
1142 
1143 	ndelay = (flag & (FNDELAY | FNONBLOCK)) ? B_TRUE : B_FALSE;
1144 
1145 	/*
1146 	 * Block any DR events from changing the set of registered
1147 	 * devices while we function.
1148 	 */
1149 	rw_enter(&bd_lock, RW_READER);
1150 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1151 		rw_exit(&bd_lock);
1152 		return (ENXIO);
1153 	}
1154 
1155 	mutex_enter(&bd->d_ocmutex);
1156 
1157 	ASSERT(part < 64);
1158 	mask = (1U << part);
1159 
1160 	bd_update_state(bd);
1161 
1162 	if (cmlb_validate(bd->d_cmlbh, 0, 0) != 0) {
1163 
1164 		/* non-blocking opens are allowed to succeed */
1165 		if (!ndelay) {
1166 			rv = ENXIO;
1167 			goto done;
1168 		}
1169 	} else if (cmlb_partinfo(bd->d_cmlbh, part, &nblks, &lba,
1170 	    NULL, NULL, 0) == 0) {
1171 
1172 		/*
1173 		 * We read the partinfo, verify valid ranges.  If the
1174 		 * partition is invalid, and we aren't blocking or
1175 		 * doing a raw access, then fail. (Non-blocking and
1176 		 * raw accesses can still succeed to allow a disk with
1177 		 * bad partition data to opened by format and fdisk.)
1178 		 */
1179 		if ((!nblks) && ((!ndelay) || (otyp != OTYP_CHR))) {
1180 			rv = ENXIO;
1181 			goto done;
1182 		}
1183 	} else if (!ndelay) {
1184 		/*
1185 		 * cmlb_partinfo failed -- invalid partition or no
1186 		 * disk label.
1187 		 */
1188 		rv = ENXIO;
1189 		goto done;
1190 	}
1191 
1192 	if ((flag & FWRITE) && bd->d_rdonly) {
1193 		rv = EROFS;
1194 		goto done;
1195 	}
1196 
1197 	if ((bd->d_open_excl) & (mask)) {
1198 		rv = EBUSY;
1199 		goto done;
1200 	}
1201 	if (flag & FEXCL) {
1202 		if (bd->d_open_lyr[part]) {
1203 			rv = EBUSY;
1204 			goto done;
1205 		}
1206 		for (int i = 0; i < OTYP_LYR; i++) {
1207 			if (bd->d_open_reg[i] & mask) {
1208 				rv = EBUSY;
1209 				goto done;
1210 			}
1211 		}
1212 	}
1213 
1214 	if (otyp == OTYP_LYR) {
1215 		bd->d_open_lyr[part]++;
1216 	} else {
1217 		bd->d_open_reg[otyp] |= mask;
1218 	}
1219 	if (flag & FEXCL) {
1220 		bd->d_open_excl |= mask;
1221 	}
1222 
1223 	rv = 0;
1224 done:
1225 	mutex_exit(&bd->d_ocmutex);
1226 	rw_exit(&bd_lock);
1227 
1228 	return (rv);
1229 }
1230 
1231 static int
1232 bd_close(dev_t dev, int flag, int otyp, cred_t *credp)
1233 {
1234 	bd_t		*bd;
1235 	minor_t		inst;
1236 	minor_t		part;
1237 	uint64_t	mask;
1238 	boolean_t	last = B_TRUE;
1239 
1240 	_NOTE(ARGUNUSED(flag));
1241 	_NOTE(ARGUNUSED(credp));
1242 
1243 	part = BDPART(dev);
1244 	inst = BDINST(dev);
1245 
1246 	ASSERT(part < 64);
1247 	mask = (1U << part);
1248 
1249 	rw_enter(&bd_lock, RW_READER);
1250 
1251 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1252 		rw_exit(&bd_lock);
1253 		return (ENXIO);
1254 	}
1255 
1256 	mutex_enter(&bd->d_ocmutex);
1257 	if (bd->d_open_excl & mask) {
1258 		bd->d_open_excl &= ~mask;
1259 	}
1260 	if (otyp == OTYP_LYR) {
1261 		bd->d_open_lyr[part]--;
1262 	} else {
1263 		bd->d_open_reg[otyp] &= ~mask;
1264 	}
1265 	for (int i = 0; i < 64; i++) {
1266 		if (bd->d_open_lyr[part]) {
1267 			last = B_FALSE;
1268 		}
1269 	}
1270 	for (int i = 0; last && (i < OTYP_LYR); i++) {
1271 		if (bd->d_open_reg[i]) {
1272 			last = B_FALSE;
1273 		}
1274 	}
1275 	mutex_exit(&bd->d_ocmutex);
1276 
1277 	if (last) {
1278 		cmlb_invalidate(bd->d_cmlbh, 0);
1279 	}
1280 	rw_exit(&bd_lock);
1281 
1282 	return (0);
1283 }
1284 
1285 static int
1286 bd_dump(dev_t dev, caddr_t caddr, daddr_t blkno, int nblk)
1287 {
1288 	minor_t		inst;
1289 	minor_t		part;
1290 	diskaddr_t	pstart;
1291 	diskaddr_t	psize;
1292 	bd_t		*bd;
1293 	bd_xfer_impl_t	*xi;
1294 	buf_t		*bp;
1295 	int		rv;
1296 	uint32_t	shift;
1297 	daddr_t		d_blkno;
1298 	int	d_nblk;
1299 
1300 	rw_enter(&bd_lock, RW_READER);
1301 
1302 	part = BDPART(dev);
1303 	inst = BDINST(dev);
1304 
1305 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1306 		rw_exit(&bd_lock);
1307 		return (ENXIO);
1308 	}
1309 	shift = bd->d_blkshift;
1310 	d_blkno = blkno >> (shift - DEV_BSHIFT);
1311 	d_nblk = nblk >> (shift - DEV_BSHIFT);
1312 	/*
1313 	 * do cmlb, but do it synchronously unless we already have the
1314 	 * partition (which we probably should.)
1315 	 */
1316 	if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL,
1317 	    (void *)1)) {
1318 		rw_exit(&bd_lock);
1319 		return (ENXIO);
1320 	}
1321 
1322 	if ((d_blkno + d_nblk) > psize) {
1323 		rw_exit(&bd_lock);
1324 		return (EINVAL);
1325 	}
1326 	bp = getrbuf(KM_NOSLEEP);
1327 	if (bp == NULL) {
1328 		rw_exit(&bd_lock);
1329 		return (ENOMEM);
1330 	}
1331 
1332 	bp->b_bcount = nblk << DEV_BSHIFT;
1333 	bp->b_resid = bp->b_bcount;
1334 	bp->b_lblkno = blkno;
1335 	bp->b_un.b_addr = caddr;
1336 
1337 	xi = bd_xfer_alloc(bd, bp,  bd->d_ops.o_write, KM_NOSLEEP);
1338 	if (xi == NULL) {
1339 		rw_exit(&bd_lock);
1340 		freerbuf(bp);
1341 		return (ENOMEM);
1342 	}
1343 	xi->i_blkno = d_blkno + pstart;
1344 	xi->i_flags = BD_XFER_POLL;
1345 	bd_submit(bd, xi);
1346 	rw_exit(&bd_lock);
1347 
1348 	/*
1349 	 * Generally, we should have run this entirely synchronously
1350 	 * at this point and the biowait call should be a no-op.  If
1351 	 * it didn't happen this way, it's a bug in the underlying
1352 	 * driver not honoring BD_XFER_POLL.
1353 	 */
1354 	(void) biowait(bp);
1355 	rv = geterror(bp);
1356 	freerbuf(bp);
1357 	return (rv);
1358 }
1359 
1360 void
1361 bd_minphys(struct buf *bp)
1362 {
1363 	minor_t inst;
1364 	bd_t	*bd;
1365 	inst = BDINST(bp->b_edev);
1366 
1367 	bd = ddi_get_soft_state(bd_state, inst);
1368 
1369 	/*
1370 	 * In a non-debug kernel, bd_strategy will catch !bd as
1371 	 * well, and will fail nicely.
1372 	 */
1373 	ASSERT(bd);
1374 
1375 	if (bp->b_bcount > bd->d_maxxfer)
1376 		bp->b_bcount = bd->d_maxxfer;
1377 }
1378 
1379 static int
1380 bd_check_uio(dev_t dev, struct uio *uio)
1381 {
1382 	bd_t		*bd;
1383 	uint32_t	shift;
1384 
1385 	if ((bd = ddi_get_soft_state(bd_state, BDINST(dev))) == NULL) {
1386 		return (ENXIO);
1387 	}
1388 
1389 	shift = bd->d_blkshift;
1390 	if ((P2PHASE(uio->uio_loffset, (1U << shift)) != 0) ||
1391 	    (P2PHASE(uio->uio_iov->iov_len, (1U << shift)) != 0)) {
1392 		return (EINVAL);
1393 	}
1394 
1395 	return (0);
1396 }
1397 
1398 static int
1399 bd_read(dev_t dev, struct uio *uio, cred_t *credp)
1400 {
1401 	_NOTE(ARGUNUSED(credp));
1402 	int	ret = bd_check_uio(dev, uio);
1403 	if (ret != 0) {
1404 		return (ret);
1405 	}
1406 	return (physio(bd_strategy, NULL, dev, B_READ, bd_minphys, uio));
1407 }
1408 
1409 static int
1410 bd_write(dev_t dev, struct uio *uio, cred_t *credp)
1411 {
1412 	_NOTE(ARGUNUSED(credp));
1413 	int	ret = bd_check_uio(dev, uio);
1414 	if (ret != 0) {
1415 		return (ret);
1416 	}
1417 	return (physio(bd_strategy, NULL, dev, B_WRITE, bd_minphys, uio));
1418 }
1419 
1420 static int
1421 bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp)
1422 {
1423 	_NOTE(ARGUNUSED(credp));
1424 	int	ret = bd_check_uio(dev, aio->aio_uio);
1425 	if (ret != 0) {
1426 		return (ret);
1427 	}
1428 	return (aphysio(bd_strategy, anocancel, dev, B_READ, bd_minphys, aio));
1429 }
1430 
1431 static int
1432 bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp)
1433 {
1434 	_NOTE(ARGUNUSED(credp));
1435 	int	ret = bd_check_uio(dev, aio->aio_uio);
1436 	if (ret != 0) {
1437 		return (ret);
1438 	}
1439 	return (aphysio(bd_strategy, anocancel, dev, B_WRITE, bd_minphys, aio));
1440 }
1441 
1442 static int
1443 bd_strategy(struct buf *bp)
1444 {
1445 	minor_t		inst;
1446 	minor_t		part;
1447 	bd_t		*bd;
1448 	diskaddr_t	p_lba;
1449 	diskaddr_t	p_nblks;
1450 	diskaddr_t	b_nblks;
1451 	bd_xfer_impl_t	*xi;
1452 	uint32_t	shift;
1453 	int		(*func)(void *, bd_xfer_t *);
1454 	diskaddr_t	lblkno;
1455 
1456 	part = BDPART(bp->b_edev);
1457 	inst = BDINST(bp->b_edev);
1458 
1459 	ASSERT(bp);
1460 
1461 	bp->b_resid = bp->b_bcount;
1462 
1463 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1464 		bioerror(bp, ENXIO);
1465 		biodone(bp);
1466 		return (0);
1467 	}
1468 
1469 	if (cmlb_partinfo(bd->d_cmlbh, part, &p_nblks, &p_lba,
1470 	    NULL, NULL, 0)) {
1471 		bioerror(bp, ENXIO);
1472 		biodone(bp);
1473 		return (0);
1474 	}
1475 
1476 	shift = bd->d_blkshift;
1477 	lblkno = bp->b_lblkno >> (shift - DEV_BSHIFT);
1478 	if ((P2PHASE(bp->b_lblkno, (1U << (shift - DEV_BSHIFT))) != 0) ||
1479 	    (P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
1480 	    (lblkno > p_nblks)) {
1481 		bioerror(bp, EINVAL);
1482 		biodone(bp);
1483 		return (0);
1484 	}
1485 	b_nblks = bp->b_bcount >> shift;
1486 	if ((lblkno == p_nblks) || (bp->b_bcount == 0)) {
1487 		biodone(bp);
1488 		return (0);
1489 	}
1490 
1491 	if ((b_nblks + lblkno) > p_nblks) {
1492 		bp->b_resid = ((lblkno + b_nblks - p_nblks) << shift);
1493 		bp->b_bcount -= bp->b_resid;
1494 	} else {
1495 		bp->b_resid = 0;
1496 	}
1497 	func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write;
1498 
1499 	xi = bd_xfer_alloc(bd, bp, func, KM_NOSLEEP);
1500 	if (xi == NULL) {
1501 		xi = bd_xfer_alloc(bd, bp, func, KM_PUSHPAGE);
1502 	}
1503 	if (xi == NULL) {
1504 		/* bd_request_alloc will have done bioerror */
1505 		biodone(bp);
1506 		return (0);
1507 	}
1508 	xi->i_blkno = lblkno + p_lba;
1509 
1510 	bd_submit(bd, xi);
1511 
1512 	return (0);
1513 }
1514 
1515 static int
1516 bd_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp, int *rvalp)
1517 {
1518 	minor_t		inst;
1519 	uint16_t	part;
1520 	bd_t		*bd;
1521 	void		*ptr = (void *)arg;
1522 	int		rv;
1523 
1524 	part = BDPART(dev);
1525 	inst = BDINST(dev);
1526 
1527 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1528 		return (ENXIO);
1529 	}
1530 
1531 	rv = cmlb_ioctl(bd->d_cmlbh, dev, cmd, arg, flag, credp, rvalp, 0);
1532 	if (rv != ENOTTY)
1533 		return (rv);
1534 
1535 	if (rvalp != NULL) {
1536 		/* the return value of the ioctl is 0 by default */
1537 		*rvalp = 0;
1538 	}
1539 
1540 	switch (cmd) {
1541 	case DKIOCGMEDIAINFO: {
1542 		struct dk_minfo minfo;
1543 
1544 		/* make sure our state information is current */
1545 		bd_update_state(bd);
1546 		bzero(&minfo, sizeof (minfo));
1547 		minfo.dki_media_type = DK_FIXED_DISK;
1548 		minfo.dki_lbsize = (1U << bd->d_blkshift);
1549 		minfo.dki_capacity = bd->d_numblks;
1550 		if (ddi_copyout(&minfo, ptr, sizeof (minfo), flag)) {
1551 			return (EFAULT);
1552 		}
1553 		return (0);
1554 	}
1555 	case DKIOCGMEDIAINFOEXT: {
1556 		struct dk_minfo_ext miext;
1557 		size_t len;
1558 
1559 		/* make sure our state information is current */
1560 		bd_update_state(bd);
1561 		bzero(&miext, sizeof (miext));
1562 		miext.dki_media_type = DK_FIXED_DISK;
1563 		miext.dki_lbsize = (1U << bd->d_blkshift);
1564 		miext.dki_pbsize = (1U << bd->d_pblkshift);
1565 		miext.dki_capacity = bd->d_numblks;
1566 
1567 		switch (ddi_model_convert_from(flag & FMODELS)) {
1568 		case DDI_MODEL_ILP32:
1569 			len = sizeof (struct dk_minfo_ext32);
1570 			break;
1571 		default:
1572 			len = sizeof (struct dk_minfo_ext);
1573 			break;
1574 		}
1575 
1576 		if (ddi_copyout(&miext, ptr, len, flag)) {
1577 			return (EFAULT);
1578 		}
1579 		return (0);
1580 	}
1581 	case DKIOCINFO: {
1582 		struct dk_cinfo cinfo;
1583 		bzero(&cinfo, sizeof (cinfo));
1584 		cinfo.dki_ctype = DKC_BLKDEV;
1585 		cinfo.dki_cnum = ddi_get_instance(ddi_get_parent(bd->d_dip));
1586 		(void) snprintf(cinfo.dki_cname, sizeof (cinfo.dki_cname),
1587 		    "%s", ddi_driver_name(ddi_get_parent(bd->d_dip)));
1588 		(void) snprintf(cinfo.dki_dname, sizeof (cinfo.dki_dname),
1589 		    "%s", ddi_driver_name(bd->d_dip));
1590 		cinfo.dki_unit = inst;
1591 		cinfo.dki_flags = DKI_FMTVOL;
1592 		cinfo.dki_partition = part;
1593 		cinfo.dki_maxtransfer = bd->d_maxxfer / DEV_BSIZE;
1594 		cinfo.dki_addr = 0;
1595 		cinfo.dki_slave = 0;
1596 		cinfo.dki_space = 0;
1597 		cinfo.dki_prio = 0;
1598 		cinfo.dki_vec = 0;
1599 		if (ddi_copyout(&cinfo, ptr, sizeof (cinfo), flag)) {
1600 			return (EFAULT);
1601 		}
1602 		return (0);
1603 	}
1604 	case DKIOCREMOVABLE: {
1605 		int i;
1606 		i = bd->d_removable ? 1 : 0;
1607 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1608 			return (EFAULT);
1609 		}
1610 		return (0);
1611 	}
1612 	case DKIOCHOTPLUGGABLE: {
1613 		int i;
1614 		i = bd->d_hotpluggable ? 1 : 0;
1615 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1616 			return (EFAULT);
1617 		}
1618 		return (0);
1619 	}
1620 	case DKIOCREADONLY: {
1621 		int i;
1622 		i = bd->d_rdonly ? 1 : 0;
1623 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1624 			return (EFAULT);
1625 		}
1626 		return (0);
1627 	}
1628 	case DKIOCSOLIDSTATE: {
1629 		int i;
1630 		i = bd->d_ssd ? 1 : 0;
1631 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1632 			return (EFAULT);
1633 		}
1634 		return (0);
1635 	}
1636 	case DKIOCSTATE: {
1637 		enum dkio_state	state;
1638 		if (ddi_copyin(ptr, &state, sizeof (state), flag)) {
1639 			return (EFAULT);
1640 		}
1641 		if ((rv = bd_check_state(bd, &state)) != 0) {
1642 			return (rv);
1643 		}
1644 		if (ddi_copyout(&state, ptr, sizeof (state), flag)) {
1645 			return (EFAULT);
1646 		}
1647 		return (0);
1648 	}
1649 	case DKIOCFLUSHWRITECACHE: {
1650 		struct dk_callback *dkc = NULL;
1651 
1652 		if (flag & FKIOCTL)
1653 			dkc = (void *)arg;
1654 
1655 		rv = bd_flush_write_cache(bd, dkc);
1656 		return (rv);
1657 	}
1658 	case DKIOCFREE: {
1659 		dkioc_free_list_t *dfl = NULL;
1660 
1661 		/*
1662 		 * Check free space support early to avoid copyin/allocation
1663 		 * when unnecessary.
1664 		 */
1665 		if (!CAN_FREESPACE(bd))
1666 			return (ENOTSUP);
1667 
1668 		rv = dfl_copyin(ptr, &dfl, flag, KM_SLEEP);
1669 		if (rv != 0)
1670 			return (rv);
1671 
1672 		/*
1673 		 * bd_free_space() consumes 'dfl'. bd_free_space() will
1674 		 * call dfl_iter() which will normally try to pass dfl through
1675 		 * to bd_free_space_cb() which attaches dfl to the bd_xfer_t
1676 		 * that is then queued for the underlying driver. Once the
1677 		 * driver processes the request, the bd_xfer_t instance is
1678 		 * disposed of, including any attached dkioc_free_list_t.
1679 		 *
1680 		 * If dfl cannot be processed by the underlying driver due to
1681 		 * size or alignment requirements of the driver, dfl_iter()
1682 		 * will replace dfl with one or more new dkioc_free_list_t
1683 		 * instances with the correct alignment and sizes for the driver
1684 		 * (and free the original dkioc_free_list_t).
1685 		 */
1686 		rv = bd_free_space(dev, bd, dfl);
1687 		return (rv);
1688 	}
1689 
1690 	case DKIOC_CANFREE: {
1691 		boolean_t supported = CAN_FREESPACE(bd);
1692 
1693 		if (ddi_copyout(&supported, (void *)arg, sizeof (supported),
1694 		    flag) != 0) {
1695 			return (EFAULT);
1696 		}
1697 
1698 		return (0);
1699 	}
1700 
1701 	default:
1702 		break;
1703 
1704 	}
1705 	return (ENOTTY);
1706 }
1707 
1708 static int
1709 bd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1710     char *name, caddr_t valuep, int *lengthp)
1711 {
1712 	bd_t	*bd;
1713 
1714 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1715 	if (bd == NULL)
1716 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1717 		    name, valuep, lengthp));
1718 
1719 	return (cmlb_prop_op(bd->d_cmlbh, dev, dip, prop_op, mod_flags, name,
1720 	    valuep, lengthp, BDPART(dev), 0));
1721 }
1722 
1723 
1724 static int
1725 bd_tg_rdwr(dev_info_t *dip, uchar_t cmd, void *bufaddr, diskaddr_t start,
1726     size_t length, void *tg_cookie)
1727 {
1728 	bd_t		*bd;
1729 	buf_t		*bp;
1730 	bd_xfer_impl_t	*xi;
1731 	int		rv;
1732 	int		(*func)(void *, bd_xfer_t *);
1733 	int		kmflag;
1734 
1735 	/*
1736 	 * If we are running in polled mode (such as during dump(9e)
1737 	 * execution), then we cannot sleep for kernel allocations.
1738 	 */
1739 	kmflag = tg_cookie ? KM_NOSLEEP : KM_SLEEP;
1740 
1741 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1742 
1743 	if (P2PHASE(length, (1U << bd->d_blkshift)) != 0) {
1744 		/* We can only transfer whole blocks at a time! */
1745 		return (EINVAL);
1746 	}
1747 
1748 	if ((bp = getrbuf(kmflag)) == NULL) {
1749 		return (ENOMEM);
1750 	}
1751 
1752 	switch (cmd) {
1753 	case TG_READ:
1754 		bp->b_flags = B_READ;
1755 		func = bd->d_ops.o_read;
1756 		break;
1757 	case TG_WRITE:
1758 		bp->b_flags = B_WRITE;
1759 		func = bd->d_ops.o_write;
1760 		break;
1761 	default:
1762 		freerbuf(bp);
1763 		return (EINVAL);
1764 	}
1765 
1766 	bp->b_un.b_addr = bufaddr;
1767 	bp->b_bcount = length;
1768 	xi = bd_xfer_alloc(bd, bp, func, kmflag);
1769 	if (xi == NULL) {
1770 		rv = geterror(bp);
1771 		freerbuf(bp);
1772 		return (rv);
1773 	}
1774 	xi->i_flags = tg_cookie ? BD_XFER_POLL : 0;
1775 	xi->i_blkno = start;
1776 	bd_submit(bd, xi);
1777 	(void) biowait(bp);
1778 	rv = geterror(bp);
1779 	freerbuf(bp);
1780 
1781 	return (rv);
1782 }
1783 
1784 static int
1785 bd_tg_getinfo(dev_info_t *dip, int cmd, void *arg, void *tg_cookie)
1786 {
1787 	bd_t		*bd;
1788 
1789 	_NOTE(ARGUNUSED(tg_cookie));
1790 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1791 
1792 	switch (cmd) {
1793 	case TG_GETPHYGEOM:
1794 	case TG_GETVIRTGEOM:
1795 		/*
1796 		 * We don't have any "geometry" as such, let cmlb
1797 		 * fabricate something.
1798 		 */
1799 		return (ENOTTY);
1800 
1801 	case TG_GETCAPACITY:
1802 		bd_update_state(bd);
1803 		*(diskaddr_t *)arg = bd->d_numblks;
1804 		return (0);
1805 
1806 	case TG_GETBLOCKSIZE:
1807 		*(uint32_t *)arg = (1U << bd->d_blkshift);
1808 		return (0);
1809 
1810 	case TG_GETATTR:
1811 		/*
1812 		 * It turns out that cmlb really doesn't do much for
1813 		 * non-writable media, but lets make the information
1814 		 * available for it in case it does more in the
1815 		 * future.  (The value is currently used for
1816 		 * triggering special behavior for CD-ROMs.)
1817 		 */
1818 		bd_update_state(bd);
1819 		((tg_attribute_t *)arg)->media_is_writable =
1820 		    bd->d_rdonly ? B_FALSE : B_TRUE;
1821 		((tg_attribute_t *)arg)->media_is_solid_state = bd->d_ssd;
1822 		((tg_attribute_t *)arg)->media_is_rotational = B_FALSE;
1823 		return (0);
1824 
1825 	default:
1826 		return (EINVAL);
1827 	}
1828 }
1829 
1830 
1831 static void
1832 bd_sched(bd_t *bd, bd_queue_t *bq)
1833 {
1834 	bd_xfer_impl_t	*xi;
1835 	struct buf	*bp;
1836 	int		rv;
1837 
1838 	mutex_enter(&bq->q_iomutex);
1839 
1840 	while ((bq->q_qactive < bq->q_qsize) &&
1841 	    ((xi = list_remove_head(&bq->q_waitq)) != NULL)) {
1842 		mutex_enter(&bd->d_ksmutex);
1843 		kstat_waitq_to_runq(bd->d_kiop);
1844 		mutex_exit(&bd->d_ksmutex);
1845 
1846 		bq->q_qactive++;
1847 		list_insert_tail(&bq->q_runq, xi);
1848 
1849 		/*
1850 		 * Submit the job to the driver.  We drop the I/O mutex
1851 		 * so that we can deal with the case where the driver
1852 		 * completion routine calls back into us synchronously.
1853 		 */
1854 
1855 		mutex_exit(&bq->q_iomutex);
1856 
1857 		rv = xi->i_func(bd->d_private, &xi->i_public);
1858 		if (rv != 0) {
1859 			bp = xi->i_bp;
1860 			bioerror(bp, rv);
1861 			biodone(bp);
1862 
1863 			atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32);
1864 
1865 			mutex_enter(&bq->q_iomutex);
1866 
1867 			mutex_enter(&bd->d_ksmutex);
1868 			kstat_runq_exit(bd->d_kiop);
1869 			mutex_exit(&bd->d_ksmutex);
1870 
1871 			bq->q_qactive--;
1872 			list_remove(&bq->q_runq, xi);
1873 			bd_xfer_free(xi);
1874 		} else {
1875 			mutex_enter(&bq->q_iomutex);
1876 		}
1877 	}
1878 
1879 	mutex_exit(&bq->q_iomutex);
1880 }
1881 
1882 static void
1883 bd_submit(bd_t *bd, bd_xfer_impl_t *xi)
1884 {
1885 	uint64_t	nv = atomic_inc_64_nv(&bd->d_io_counter);
1886 	unsigned	q = nv % bd->d_qcount;
1887 	bd_queue_t	*bq = &bd->d_queues[q];
1888 
1889 	xi->i_bq = bq;
1890 	xi->i_qnum = q;
1891 
1892 	mutex_enter(&bq->q_iomutex);
1893 
1894 	list_insert_tail(&bq->q_waitq, xi);
1895 
1896 	mutex_enter(&bd->d_ksmutex);
1897 	kstat_waitq_enter(bd->d_kiop);
1898 	mutex_exit(&bd->d_ksmutex);
1899 
1900 	mutex_exit(&bq->q_iomutex);
1901 
1902 	bd_sched(bd, bq);
1903 }
1904 
1905 static void
1906 bd_runq_exit(bd_xfer_impl_t *xi, int err)
1907 {
1908 	bd_t		*bd = xi->i_bd;
1909 	buf_t		*bp = xi->i_bp;
1910 	bd_queue_t	*bq = xi->i_bq;
1911 
1912 	mutex_enter(&bq->q_iomutex);
1913 	bq->q_qactive--;
1914 
1915 	mutex_enter(&bd->d_ksmutex);
1916 	kstat_runq_exit(bd->d_kiop);
1917 	mutex_exit(&bd->d_ksmutex);
1918 
1919 	list_remove(&bq->q_runq, xi);
1920 	mutex_exit(&bq->q_iomutex);
1921 
1922 	if (err == 0) {
1923 		if (bp->b_flags & B_READ) {
1924 			atomic_inc_uint(&bd->d_kiop->reads);
1925 			atomic_add_64((uint64_t *)&bd->d_kiop->nread,
1926 			    bp->b_bcount - xi->i_resid);
1927 		} else {
1928 			atomic_inc_uint(&bd->d_kiop->writes);
1929 			atomic_add_64((uint64_t *)&bd->d_kiop->nwritten,
1930 			    bp->b_bcount - xi->i_resid);
1931 		}
1932 	}
1933 	bd_sched(bd, bq);
1934 }
1935 
1936 static void
1937 bd_dle_sysevent_task(void *arg)
1938 {
1939 	nvlist_t *attr = NULL;
1940 	char *path = NULL;
1941 	bd_t *bd = arg;
1942 	dev_info_t *dip = bd->d_dip;
1943 	size_t n;
1944 
1945 	mutex_enter(&bd->d_dle_mutex);
1946 	bd->d_dle_state &= ~BD_DLE_PENDING;
1947 	bd->d_dle_state |= BD_DLE_RUNNING;
1948 	mutex_exit(&bd->d_dle_mutex);
1949 
1950 	dev_err(dip, CE_NOTE, "!dynamic LUN expansion");
1951 
1952 	if (nvlist_alloc(&attr, NV_UNIQUE_NAME_TYPE, KM_SLEEP) != 0) {
1953 		mutex_enter(&bd->d_dle_mutex);
1954 		bd->d_dle_state &= ~(BD_DLE_RUNNING|BD_DLE_PENDING);
1955 		mutex_exit(&bd->d_dle_mutex);
1956 		return;
1957 	}
1958 
1959 	path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1960 
1961 	n = snprintf(path, MAXPATHLEN, "/devices");
1962 	(void) ddi_pathname(dip, path + n);
1963 	n = strlen(path);
1964 	n += snprintf(path + n, MAXPATHLEN - n, ":x");
1965 
1966 	for (;;) {
1967 		/*
1968 		 * On receipt of this event, the ZFS sysevent module will scan
1969 		 * active zpools for child vdevs matching this physical path.
1970 		 * In order to catch both whole disk pools and those with an
1971 		 * EFI boot partition, generate separate sysevents for minor
1972 		 * node 'a' and 'b'.
1973 		 */
1974 		for (char c = 'a'; c < 'c'; c++) {
1975 			path[n - 1] = c;
1976 
1977 			if (nvlist_add_string(attr, DEV_PHYS_PATH, path) != 0)
1978 				break;
1979 
1980 			(void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW,
1981 			    EC_DEV_STATUS, ESC_DEV_DLE, attr, NULL, DDI_SLEEP);
1982 		}
1983 
1984 		mutex_enter(&bd->d_dle_mutex);
1985 		if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
1986 			bd->d_dle_state &= ~BD_DLE_RUNNING;
1987 			mutex_exit(&bd->d_dle_mutex);
1988 			break;
1989 		}
1990 		bd->d_dle_state &= ~BD_DLE_PENDING;
1991 		mutex_exit(&bd->d_dle_mutex);
1992 	}
1993 
1994 	nvlist_free(attr);
1995 	kmem_free(path, MAXPATHLEN);
1996 }
1997 
1998 static void
1999 bd_update_state(bd_t *bd)
2000 {
2001 	enum	dkio_state	state = DKIO_INSERTED;
2002 	boolean_t		docmlb = B_FALSE;
2003 	bd_media_t		media;
2004 
2005 	bzero(&media, sizeof (media));
2006 
2007 	mutex_enter(&bd->d_statemutex);
2008 	if (bd->d_ops.o_media_info(bd->d_private, &media) != 0) {
2009 		bd->d_numblks = 0;
2010 		state = DKIO_EJECTED;
2011 		goto done;
2012 	}
2013 
2014 	if ((media.m_blksize < 512) ||
2015 	    (!ISP2(media.m_blksize)) ||
2016 	    (P2PHASE(bd->d_maxxfer, media.m_blksize))) {
2017 		dev_err(bd->d_dip, CE_WARN, "Invalid media block size (%d)",
2018 		    media.m_blksize);
2019 		/*
2020 		 * We can't use the media, treat it as not present.
2021 		 */
2022 		state = DKIO_EJECTED;
2023 		bd->d_numblks = 0;
2024 		goto done;
2025 	}
2026 
2027 	if (((1U << bd->d_blkshift) != media.m_blksize) ||
2028 	    (bd->d_numblks != media.m_nblks)) {
2029 		/* Device size changed */
2030 		docmlb = B_TRUE;
2031 	}
2032 
2033 	bd->d_blkshift = ddi_ffs(media.m_blksize) - 1;
2034 	bd->d_pblkshift = bd->d_blkshift;
2035 	bd->d_numblks = media.m_nblks;
2036 	bd->d_rdonly = media.m_readonly;
2037 	bd->d_ssd = media.m_solidstate;
2038 
2039 	/*
2040 	 * Only use the supplied physical block size if it is non-zero,
2041 	 * greater or equal to the block size, and a power of 2. Ignore it
2042 	 * if not, it's just informational and we can still use the media.
2043 	 */
2044 	if ((media.m_pblksize != 0) &&
2045 	    (media.m_pblksize >= media.m_blksize) &&
2046 	    (ISP2(media.m_pblksize)))
2047 		bd->d_pblkshift = ddi_ffs(media.m_pblksize) - 1;
2048 
2049 done:
2050 	if (state != bd->d_state) {
2051 		bd->d_state = state;
2052 		cv_broadcast(&bd->d_statecv);
2053 		docmlb = B_TRUE;
2054 	}
2055 	mutex_exit(&bd->d_statemutex);
2056 
2057 	bd->d_kerr->bd_capacity.value.ui64 = bd->d_numblks << bd->d_blkshift;
2058 
2059 	if (docmlb) {
2060 		if (state == DKIO_INSERTED) {
2061 			(void) cmlb_validate(bd->d_cmlbh, 0, 0);
2062 
2063 			mutex_enter(&bd->d_dle_mutex);
2064 			/*
2065 			 * If there is already an event pending, there's
2066 			 * nothing to do; we coalesce multiple events.
2067 			 */
2068 			if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
2069 				if ((bd->d_dle_state & BD_DLE_RUNNING) == 0) {
2070 					taskq_dispatch_ent(bd_taskq,
2071 					    bd_dle_sysevent_task, bd, 0,
2072 					    &bd->d_dle_ent);
2073 				}
2074 				bd->d_dle_state |= BD_DLE_PENDING;
2075 			}
2076 			mutex_exit(&bd->d_dle_mutex);
2077 		} else {
2078 			cmlb_invalidate(bd->d_cmlbh, 0);
2079 		}
2080 	}
2081 }
2082 
2083 static int
2084 bd_check_state(bd_t *bd, enum dkio_state *state)
2085 {
2086 	clock_t		when;
2087 
2088 	for (;;) {
2089 
2090 		bd_update_state(bd);
2091 
2092 		mutex_enter(&bd->d_statemutex);
2093 
2094 		if (bd->d_state != *state) {
2095 			*state = bd->d_state;
2096 			mutex_exit(&bd->d_statemutex);
2097 			break;
2098 		}
2099 
2100 		when = drv_usectohz(1000000);
2101 		if (cv_reltimedwait_sig(&bd->d_statecv, &bd->d_statemutex,
2102 		    when, TR_CLOCK_TICK) == 0) {
2103 			mutex_exit(&bd->d_statemutex);
2104 			return (EINTR);
2105 		}
2106 
2107 		mutex_exit(&bd->d_statemutex);
2108 	}
2109 
2110 	return (0);
2111 }
2112 
2113 static int
2114 bd_flush_write_cache_done(struct buf *bp)
2115 {
2116 	struct dk_callback *dc = (void *)bp->b_private;
2117 
2118 	(*dc->dkc_callback)(dc->dkc_cookie, geterror(bp));
2119 	kmem_free(dc, sizeof (*dc));
2120 	freerbuf(bp);
2121 	return (0);
2122 }
2123 
2124 static int
2125 bd_flush_write_cache(bd_t *bd, struct dk_callback *dkc)
2126 {
2127 	buf_t			*bp;
2128 	struct dk_callback	*dc;
2129 	bd_xfer_impl_t		*xi;
2130 	int			rv;
2131 
2132 	if (bd->d_ops.o_sync_cache == NULL) {
2133 		return (ENOTSUP);
2134 	}
2135 	if ((bp = getrbuf(KM_SLEEP)) == NULL) {
2136 		return (ENOMEM);
2137 	}
2138 	bp->b_resid = 0;
2139 	bp->b_bcount = 0;
2140 
2141 	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_sync_cache, KM_SLEEP);
2142 	if (xi == NULL) {
2143 		rv = geterror(bp);
2144 		freerbuf(bp);
2145 		return (rv);
2146 	}
2147 
2148 	/* Make an asynchronous flush, but only if there is a callback */
2149 	if (dkc != NULL && dkc->dkc_callback != NULL) {
2150 		/* Make a private copy of the callback structure */
2151 		dc = kmem_alloc(sizeof (*dc), KM_SLEEP);
2152 		*dc = *dkc;
2153 		bp->b_private = dc;
2154 		bp->b_iodone = bd_flush_write_cache_done;
2155 
2156 		bd_submit(bd, xi);
2157 		return (0);
2158 	}
2159 
2160 	/* In case there is no callback, perform a synchronous flush */
2161 	bd_submit(bd, xi);
2162 	(void) biowait(bp);
2163 	rv = geterror(bp);
2164 	freerbuf(bp);
2165 
2166 	return (rv);
2167 }
2168 
2169 static int
2170 bd_free_space_done(struct buf *bp)
2171 {
2172 	freerbuf(bp);
2173 	return (0);
2174 }
2175 
2176 static int
2177 bd_free_space_cb(dkioc_free_list_t *dfl, void *arg, int kmflag)
2178 {
2179 	bd_t		*bd = arg;
2180 	buf_t		*bp = NULL;
2181 	bd_xfer_impl_t	*xi = NULL;
2182 	boolean_t	sync = DFL_ISSYNC(dfl) ?  B_TRUE : B_FALSE;
2183 	int		rv = 0;
2184 
2185 	bp = getrbuf(KM_SLEEP);
2186 	bp->b_resid = 0;
2187 	bp->b_bcount = 0;
2188 	bp->b_lblkno = 0;
2189 
2190 	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_free_space, kmflag);
2191 	xi->i_dfl = dfl;
2192 
2193 	if (!sync) {
2194 		bp->b_iodone = bd_free_space_done;
2195 		bd_submit(bd, xi);
2196 		return (0);
2197 	}
2198 
2199 	xi->i_flags |= BD_XFER_POLL;
2200 	bd_submit(bd, xi);
2201 
2202 	(void) biowait(bp);
2203 	rv = geterror(bp);
2204 	freerbuf(bp);
2205 
2206 	return (rv);
2207 }
2208 
2209 static int
2210 bd_free_space(dev_t dev, bd_t *bd, dkioc_free_list_t *dfl)
2211 {
2212 	diskaddr_t p_len, p_offset;
2213 	uint64_t offset_bytes, len_bytes;
2214 	minor_t part = BDPART(dev);
2215 	const uint_t bshift = bd->d_blkshift;
2216 	dkioc_free_info_t dfi = {
2217 		.dfi_bshift = bshift,
2218 		.dfi_align = bd->d_free_align << bshift,
2219 		.dfi_max_bytes = bd->d_max_free_blks << bshift,
2220 		.dfi_max_ext = bd->d_max_free_seg,
2221 		.dfi_max_ext_bytes = bd->d_max_free_seg_blks << bshift,
2222 	};
2223 
2224 	if (cmlb_partinfo(bd->d_cmlbh, part, &p_len, &p_offset, NULL,
2225 	    NULL, 0) != 0) {
2226 		dfl_free(dfl);
2227 		return (ENXIO);
2228 	}
2229 
2230 	/*
2231 	 * bd_ioctl created our own copy of dfl, so we can modify as
2232 	 * necessary
2233 	 */
2234 	offset_bytes = (uint64_t)p_offset << bshift;
2235 	len_bytes = (uint64_t)p_len << bshift;
2236 
2237 	dfl->dfl_offset += offset_bytes;
2238 	if (dfl->dfl_offset < offset_bytes) {
2239 		dfl_free(dfl);
2240 		return (EOVERFLOW);
2241 	}
2242 
2243 	return (dfl_iter(dfl, &dfi, offset_bytes + len_bytes, bd_free_space_cb,
2244 	    bd, KM_SLEEP));
2245 }
2246 
2247 /*
2248  * Nexus support.
2249  */
2250 int
2251 bd_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop,
2252     void *arg, void *result)
2253 {
2254 	bd_handle_t	hdl;
2255 
2256 	switch (ctlop) {
2257 	case DDI_CTLOPS_REPORTDEV:
2258 		cmn_err(CE_CONT, "?Block device: %s@%s, %s%d\n",
2259 		    ddi_node_name(rdip), ddi_get_name_addr(rdip),
2260 		    ddi_driver_name(rdip), ddi_get_instance(rdip));
2261 		return (DDI_SUCCESS);
2262 
2263 	case DDI_CTLOPS_INITCHILD:
2264 		hdl = ddi_get_parent_data((dev_info_t *)arg);
2265 		if (hdl == NULL) {
2266 			return (DDI_NOT_WELL_FORMED);
2267 		}
2268 		ddi_set_name_addr((dev_info_t *)arg, hdl->h_addr);
2269 		return (DDI_SUCCESS);
2270 
2271 	case DDI_CTLOPS_UNINITCHILD:
2272 		ddi_set_name_addr((dev_info_t *)arg, NULL);
2273 		ndi_prop_remove_all((dev_info_t *)arg);
2274 		return (DDI_SUCCESS);
2275 
2276 	default:
2277 		return (ddi_ctlops(dip, rdip, ctlop, arg, result));
2278 	}
2279 }
2280 
2281 /*
2282  * Functions for device drivers.
2283  */
2284 bd_handle_t
2285 bd_alloc_handle(void *private, bd_ops_t *ops, ddi_dma_attr_t *dma, int kmflag)
2286 {
2287 	bd_handle_t	hdl;
2288 
2289 	switch (ops->o_version) {
2290 	case BD_OPS_VERSION_0:
2291 	case BD_OPS_VERSION_1:
2292 	case BD_OPS_VERSION_2:
2293 		break;
2294 
2295 	default:
2296 		/* Unsupported version */
2297 		return (NULL);
2298 	}
2299 
2300 	hdl = kmem_zalloc(sizeof (*hdl), kmflag);
2301 	if (hdl == NULL) {
2302 		return (NULL);
2303 	}
2304 
2305 	switch (ops->o_version) {
2306 	case BD_OPS_VERSION_2:
2307 		hdl->h_ops.o_free_space = ops->o_free_space;
2308 		/*FALLTHRU*/
2309 	case BD_OPS_VERSION_1:
2310 	case BD_OPS_VERSION_0:
2311 		hdl->h_ops.o_drive_info = ops->o_drive_info;
2312 		hdl->h_ops.o_media_info = ops->o_media_info;
2313 		hdl->h_ops.o_devid_init = ops->o_devid_init;
2314 		hdl->h_ops.o_sync_cache = ops->o_sync_cache;
2315 		hdl->h_ops.o_read = ops->o_read;
2316 		hdl->h_ops.o_write = ops->o_write;
2317 		break;
2318 	}
2319 
2320 	hdl->h_dma = dma;
2321 	hdl->h_private = private;
2322 
2323 	return (hdl);
2324 }
2325 
2326 void
2327 bd_free_handle(bd_handle_t hdl)
2328 {
2329 	kmem_free(hdl, sizeof (*hdl));
2330 }
2331 
2332 int
2333 bd_attach_handle(dev_info_t *dip, bd_handle_t hdl)
2334 {
2335 	bd_drive_t	drive = { 0 };
2336 	dev_info_t	*child;
2337 	size_t		len;
2338 
2339 	/*
2340 	 * It's not an error if bd_attach_handle() is called on a handle that
2341 	 * already is attached. We just ignore the request to attach and return.
2342 	 * This way drivers using blkdev don't have to keep track about blkdev
2343 	 * state, they can just call this function to make sure it attached.
2344 	 */
2345 	if (hdl->h_child != NULL) {
2346 		return (DDI_SUCCESS);
2347 	}
2348 
2349 	/* if drivers don't override this, make it assume none */
2350 	drive.d_lun = -1;
2351 	hdl->h_ops.o_drive_info(hdl->h_private, &drive);
2352 
2353 	hdl->h_parent = dip;
2354 	hdl->h_name = "blkdev";
2355 
2356 	/*
2357 	 * Prefer the GUID over the EUI64.
2358 	 */
2359 	if (*(uint64_t *)drive.d_guid != 0 ||
2360 	    *((uint64_t *)drive.d_guid + 1) != 0) {
2361 		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
2362 		    "w%02X%02X%02X%02X%02X%02X%02X%02X"
2363 		    "%02X%02X%02X%02X%02X%02X%02X%02X",
2364 		    drive.d_guid[0], drive.d_guid[1], drive.d_guid[2],
2365 		    drive.d_guid[3], drive.d_guid[4], drive.d_guid[5],
2366 		    drive.d_guid[6], drive.d_guid[7], drive.d_guid[8],
2367 		    drive.d_guid[9], drive.d_guid[10], drive.d_guid[11],
2368 		    drive.d_guid[12], drive.d_guid[13], drive.d_guid[14],
2369 		    drive.d_guid[15]);
2370 	} else if (*(uint64_t *)drive.d_eui64 != 0) {
2371 		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
2372 		    "w%02X%02X%02X%02X%02X%02X%02X%02X",
2373 		    drive.d_eui64[0], drive.d_eui64[1],
2374 		    drive.d_eui64[2], drive.d_eui64[3],
2375 		    drive.d_eui64[4], drive.d_eui64[5],
2376 		    drive.d_eui64[6], drive.d_eui64[7]);
2377 	} else {
2378 		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
2379 		    "%X", drive.d_target);
2380 	}
2381 
2382 	VERIFY(len <= sizeof (hdl->h_addr));
2383 
2384 	if (drive.d_lun >= 0) {
2385 		(void) snprintf(hdl->h_addr + len, sizeof (hdl->h_addr) - len,
2386 		    ",%X", drive.d_lun);
2387 	}
2388 
2389 	if (ndi_devi_alloc(dip, hdl->h_name, (pnode_t)DEVI_SID_NODEID,
2390 	    &child) != NDI_SUCCESS) {
2391 		cmn_err(CE_WARN, "%s%d: unable to allocate node %s@%s",
2392 		    ddi_driver_name(dip), ddi_get_instance(dip),
2393 		    "blkdev", hdl->h_addr);
2394 		return (DDI_FAILURE);
2395 	}
2396 
2397 	ddi_set_parent_data(child, hdl);
2398 	hdl->h_child = child;
2399 
2400 	if (ndi_devi_online(child, 0) != NDI_SUCCESS) {
2401 		cmn_err(CE_WARN, "%s%d: failed bringing node %s@%s online",
2402 		    ddi_driver_name(dip), ddi_get_instance(dip),
2403 		    hdl->h_name, hdl->h_addr);
2404 		(void) ndi_devi_free(child);
2405 		hdl->h_child = NULL;
2406 		return (DDI_FAILURE);
2407 	}
2408 
2409 	return (DDI_SUCCESS);
2410 }
2411 
2412 int
2413 bd_detach_handle(bd_handle_t hdl)
2414 {
2415 	int	circ;
2416 	int	rv;
2417 	char	*devnm;
2418 
2419 	/*
2420 	 * It's not an error if bd_detach_handle() is called on a handle that
2421 	 * already is detached. We just ignore the request to detach and return.
2422 	 * This way drivers using blkdev don't have to keep track about blkdev
2423 	 * state, they can just call this function to make sure it detached.
2424 	 */
2425 	if (hdl->h_child == NULL) {
2426 		return (DDI_SUCCESS);
2427 	}
2428 	ndi_devi_enter(hdl->h_parent, &circ);
2429 	if (i_ddi_node_state(hdl->h_child) < DS_INITIALIZED) {
2430 		rv = ddi_remove_child(hdl->h_child, 0);
2431 	} else {
2432 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
2433 		(void) ddi_deviname(hdl->h_child, devnm);
2434 		(void) devfs_clean(hdl->h_parent, devnm + 1, DV_CLEAN_FORCE);
2435 		rv = ndi_devi_unconfig_one(hdl->h_parent, devnm + 1, NULL,
2436 		    NDI_DEVI_REMOVE | NDI_UNCONFIG);
2437 		kmem_free(devnm, MAXNAMELEN + 1);
2438 	}
2439 	if (rv == 0) {
2440 		hdl->h_child = NULL;
2441 	}
2442 
2443 	ndi_devi_exit(hdl->h_parent, circ);
2444 	return (rv == NDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
2445 }
2446 
2447 void
2448 bd_xfer_done(bd_xfer_t *xfer, int err)
2449 {
2450 	bd_xfer_impl_t	*xi = (void *)xfer;
2451 	buf_t		*bp = xi->i_bp;
2452 	int		rv = DDI_SUCCESS;
2453 	bd_t		*bd = xi->i_bd;
2454 	size_t		len;
2455 
2456 	if (err != 0) {
2457 		bd_runq_exit(xi, err);
2458 		atomic_inc_32(&bd->d_kerr->bd_harderrs.value.ui32);
2459 
2460 		bp->b_resid += xi->i_resid;
2461 		bd_xfer_free(xi);
2462 		bioerror(bp, err);
2463 		biodone(bp);
2464 		return;
2465 	}
2466 
2467 	xi->i_cur_win++;
2468 	xi->i_resid -= xi->i_len;
2469 
2470 	if (xi->i_resid == 0) {
2471 		/* Job completed succcessfully! */
2472 		bd_runq_exit(xi, 0);
2473 
2474 		bd_xfer_free(xi);
2475 		biodone(bp);
2476 		return;
2477 	}
2478 
2479 	xi->i_blkno += xi->i_nblks;
2480 
2481 	if (bd->d_use_dma) {
2482 		/* More transfer still pending... advance to next DMA window. */
2483 		rv = ddi_dma_getwin(xi->i_dmah, xi->i_cur_win,
2484 		    &xi->i_offset, &len, &xi->i_dmac, &xi->i_ndmac);
2485 	} else {
2486 		/* Advance memory window. */
2487 		xi->i_kaddr += xi->i_len;
2488 		xi->i_offset += xi->i_len;
2489 		len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer);
2490 	}
2491 
2492 
2493 	if ((rv != DDI_SUCCESS) ||
2494 	    (P2PHASE(len, (1U << xi->i_blkshift)) != 0)) {
2495 		bd_runq_exit(xi, EFAULT);
2496 
2497 		bp->b_resid += xi->i_resid;
2498 		bd_xfer_free(xi);
2499 		bioerror(bp, EFAULT);
2500 		biodone(bp);
2501 		return;
2502 	}
2503 	xi->i_len = len;
2504 	xi->i_nblks = len >> xi->i_blkshift;
2505 
2506 	/* Submit next window to hardware. */
2507 	rv = xi->i_func(bd->d_private, &xi->i_public);
2508 	if (rv != 0) {
2509 		bd_runq_exit(xi, rv);
2510 
2511 		atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32);
2512 
2513 		bp->b_resid += xi->i_resid;
2514 		bd_xfer_free(xi);
2515 		bioerror(bp, rv);
2516 		biodone(bp);
2517 	}
2518 }
2519 
2520 void
2521 bd_error(bd_xfer_t *xfer, int error)
2522 {
2523 	bd_xfer_impl_t	*xi = (void *)xfer;
2524 	bd_t		*bd = xi->i_bd;
2525 
2526 	switch (error) {
2527 	case BD_ERR_MEDIA:
2528 		atomic_inc_32(&bd->d_kerr->bd_rq_media_err.value.ui32);
2529 		break;
2530 	case BD_ERR_NTRDY:
2531 		atomic_inc_32(&bd->d_kerr->bd_rq_ntrdy_err.value.ui32);
2532 		break;
2533 	case BD_ERR_NODEV:
2534 		atomic_inc_32(&bd->d_kerr->bd_rq_nodev_err.value.ui32);
2535 		break;
2536 	case BD_ERR_RECOV:
2537 		atomic_inc_32(&bd->d_kerr->bd_rq_recov_err.value.ui32);
2538 		break;
2539 	case BD_ERR_ILLRQ:
2540 		atomic_inc_32(&bd->d_kerr->bd_rq_illrq_err.value.ui32);
2541 		break;
2542 	case BD_ERR_PFA:
2543 		atomic_inc_32(&bd->d_kerr->bd_rq_pfa_err.value.ui32);
2544 		break;
2545 	default:
2546 		cmn_err(CE_PANIC, "bd_error: unknown error type %d", error);
2547 		break;
2548 	}
2549 }
2550 
2551 void
2552 bd_state_change(bd_handle_t hdl)
2553 {
2554 	bd_t		*bd;
2555 
2556 	if ((bd = hdl->h_bd) != NULL) {
2557 		bd_update_state(bd);
2558 	}
2559 }
2560 
2561 void
2562 bd_mod_init(struct dev_ops *devops)
2563 {
2564 	static struct bus_ops bd_bus_ops = {
2565 		BUSO_REV,		/* busops_rev */
2566 		nullbusmap,		/* bus_map */
2567 		NULL,			/* bus_get_intrspec (OBSOLETE) */
2568 		NULL,			/* bus_add_intrspec (OBSOLETE) */
2569 		NULL,			/* bus_remove_intrspec (OBSOLETE) */
2570 		i_ddi_map_fault,	/* bus_map_fault */
2571 		NULL,			/* bus_dma_map (OBSOLETE) */
2572 		ddi_dma_allochdl,	/* bus_dma_allochdl */
2573 		ddi_dma_freehdl,	/* bus_dma_freehdl */
2574 		ddi_dma_bindhdl,	/* bus_dma_bindhdl */
2575 		ddi_dma_unbindhdl,	/* bus_dma_unbindhdl */
2576 		ddi_dma_flush,		/* bus_dma_flush */
2577 		ddi_dma_win,		/* bus_dma_win */
2578 		ddi_dma_mctl,		/* bus_dma_ctl */
2579 		bd_bus_ctl,		/* bus_ctl */
2580 		ddi_bus_prop_op,	/* bus_prop_op */
2581 		NULL,			/* bus_get_eventcookie */
2582 		NULL,			/* bus_add_eventcall */
2583 		NULL,			/* bus_remove_eventcall */
2584 		NULL,			/* bus_post_event */
2585 		NULL,			/* bus_intr_ctl (OBSOLETE) */
2586 		NULL,			/* bus_config */
2587 		NULL,			/* bus_unconfig */
2588 		NULL,			/* bus_fm_init */
2589 		NULL,			/* bus_fm_fini */
2590 		NULL,			/* bus_fm_access_enter */
2591 		NULL,			/* bus_fm_access_exit */
2592 		NULL,			/* bus_power */
2593 		NULL,			/* bus_intr_op */
2594 	};
2595 
2596 	devops->devo_bus_ops = &bd_bus_ops;
2597 
2598 	/*
2599 	 * NB: The device driver is free to supply its own
2600 	 * character entry device support.
2601 	 */
2602 }
2603 
2604 void
2605 bd_mod_fini(struct dev_ops *devops)
2606 {
2607 	devops->devo_bus_ops = NULL;
2608 }
2609