xref: /titanic_51/usr/src/uts/common/io/blkdev/blkdev.c (revision 15e1afcd5e908ae29b1e6018838638befdc225a2)
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  */
24 
25 /*
26  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
27  */
28 
29 #include <sys/types.h>
30 #include <sys/ksynch.h>
31 #include <sys/kmem.h>
32 #include <sys/file.h>
33 #include <sys/errno.h>
34 #include <sys/open.h>
35 #include <sys/buf.h>
36 #include <sys/uio.h>
37 #include <sys/aio_req.h>
38 #include <sys/cred.h>
39 #include <sys/modctl.h>
40 #include <sys/cmlb.h>
41 #include <sys/conf.h>
42 #include <sys/devops.h>
43 #include <sys/list.h>
44 #include <sys/sysmacros.h>
45 #include <sys/dkio.h>
46 #include <sys/vtoc.h>
47 #include <sys/scsi/scsi.h>	/* for DTYPE_DIRECT */
48 #include <sys/kstat.h>
49 #include <sys/fs/dv_node.h>
50 #include <sys/ddi.h>
51 #include <sys/sunddi.h>
52 #include <sys/note.h>
53 #include <sys/blkdev.h>
54 
55 #define	BD_MAXPART	64
56 #define	BDINST(dev)	(getminor(dev) / BD_MAXPART)
57 #define	BDPART(dev)	(getminor(dev) % BD_MAXPART)
58 
59 typedef struct bd bd_t;
60 typedef struct bd_xfer_impl bd_xfer_impl_t;
61 
62 struct bd {
63 	void		*d_private;
64 	dev_info_t	*d_dip;
65 	kmutex_t	d_ocmutex;
66 	kmutex_t	d_iomutex;
67 	kmutex_t	d_statemutex;
68 	kcondvar_t	d_statecv;
69 	enum dkio_state	d_state;
70 	cmlb_handle_t	d_cmlbh;
71 	unsigned	d_open_lyr[BD_MAXPART];	/* open count */
72 	uint64_t	d_open_excl;	/* bit mask indexed by partition */
73 	uint64_t	d_open_reg[OTYPCNT];		/* bit mask */
74 
75 	uint32_t	d_qsize;
76 	uint32_t	d_qactive;
77 	uint32_t	d_maxxfer;
78 	uint32_t	d_blkshift;
79 	uint64_t	d_numblks;
80 	ddi_devid_t	d_devid;
81 
82 	kmem_cache_t	*d_cache;
83 	list_t		d_runq;
84 	list_t		d_waitq;
85 	kstat_t		*d_ksp;
86 	kstat_io_t	*d_kiop;
87 
88 	boolean_t	d_rdonly;
89 	boolean_t	d_removable;
90 	boolean_t	d_hotpluggable;
91 	boolean_t	d_use_dma;
92 
93 	ddi_dma_attr_t	d_dma;
94 	bd_ops_t	d_ops;
95 	bd_handle_t	d_handle;
96 };
97 
98 struct bd_handle {
99 	bd_ops_t	h_ops;
100 	ddi_dma_attr_t	*h_dma;
101 	dev_info_t	*h_parent;
102 	dev_info_t	*h_child;
103 	void		*h_private;
104 	bd_t		*h_bd;
105 	char		*h_name;
106 	char		h_addr[20];	/* enough for %X,%X */
107 };
108 
109 struct bd_xfer_impl {
110 	bd_xfer_t	i_public;
111 	list_node_t	i_linkage;
112 	bd_t		*i_bd;
113 	buf_t		*i_bp;
114 	uint_t		i_num_win;
115 	uint_t		i_cur_win;
116 	off_t		i_offset;
117 	int		(*i_func)(void *, bd_xfer_t *);
118 	uint32_t	i_blkshift;
119 	size_t		i_len;
120 	size_t		i_resid;
121 };
122 
123 #define	i_dmah		i_public.x_dmah
124 #define	i_dmac		i_public.x_dmac
125 #define	i_ndmac		i_public.x_ndmac
126 #define	i_kaddr		i_public.x_kaddr
127 #define	i_nblks		i_public.x_nblks
128 #define	i_blkno		i_public.x_blkno
129 #define	i_flags		i_public.x_flags
130 
131 
132 /*
133  * Private prototypes.
134  */
135 
136 static int bd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
137 static int bd_attach(dev_info_t *, ddi_attach_cmd_t);
138 static int bd_detach(dev_info_t *, ddi_detach_cmd_t);
139 
140 static int bd_open(dev_t *, int, int, cred_t *);
141 static int bd_close(dev_t, int, int, cred_t *);
142 static int bd_strategy(struct buf *);
143 static int bd_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
144 static int bd_dump(dev_t, caddr_t, daddr_t, int);
145 static int bd_read(dev_t, struct uio *, cred_t *);
146 static int bd_write(dev_t, struct uio *, cred_t *);
147 static int bd_aread(dev_t, struct aio_req *, cred_t *);
148 static int bd_awrite(dev_t, struct aio_req *, cred_t *);
149 static int bd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *,
150     caddr_t, int *);
151 
152 static int bd_tg_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t,
153     void *);
154 static int bd_tg_getinfo(dev_info_t *, int, void *, void *);
155 static int bd_xfer_ctor(void *, void *, int);
156 static void bd_xfer_dtor(void *, void *);
157 static void bd_sched(bd_t *);
158 static void bd_submit(bd_t *, bd_xfer_impl_t *);
159 static void bd_runq_exit(bd_xfer_impl_t *, int);
160 static void bd_update_state(bd_t *);
161 static int bd_check_state(bd_t *, enum dkio_state *);
162 static int bd_flush_write_cache(bd_t *, struct dk_callback *);
163 
164 struct cmlb_tg_ops bd_tg_ops = {
165 	TG_DK_OPS_VERSION_1,
166 	bd_tg_rdwr,
167 	bd_tg_getinfo,
168 };
169 
170 static struct cb_ops bd_cb_ops = {
171 	bd_open, 		/* open */
172 	bd_close, 		/* close */
173 	bd_strategy, 		/* strategy */
174 	nodev, 			/* print */
175 	bd_dump,		/* dump */
176 	bd_read, 		/* read */
177 	bd_write, 		/* write */
178 	bd_ioctl, 		/* ioctl */
179 	nodev, 			/* devmap */
180 	nodev, 			/* mmap */
181 	nodev, 			/* segmap */
182 	nochpoll, 		/* poll */
183 	bd_prop_op, 		/* cb_prop_op */
184 	0, 			/* streamtab  */
185 	D_64BIT | D_MP,		/* Driver comaptibility flag */
186 	CB_REV,			/* cb_rev */
187 	bd_aread,		/* async read */
188 	bd_awrite		/* async write */
189 };
190 
191 struct dev_ops bd_dev_ops = {
192 	DEVO_REV, 		/* devo_rev, */
193 	0, 			/* refcnt  */
194 	bd_getinfo,		/* getinfo */
195 	nulldev, 		/* identify */
196 	nulldev, 		/* probe */
197 	bd_attach, 		/* attach */
198 	bd_detach,		/* detach */
199 	nodev, 			/* reset */
200 	&bd_cb_ops, 		/* driver operations */
201 	NULL,			/* bus operations */
202 	NULL,			/* power */
203 	ddi_quiesce_not_needed,	/* quiesce */
204 };
205 
206 static struct modldrv modldrv = {
207 	&mod_driverops,
208 	"Generic Block Device",
209 	&bd_dev_ops,
210 };
211 
212 static struct modlinkage modlinkage = {
213 	MODREV_1, { &modldrv, NULL }
214 };
215 
216 static void *bd_state;
217 static krwlock_t bd_lock;
218 
219 int
220 _init(void)
221 {
222 	int	rv;
223 
224 	rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2);
225 	if (rv != DDI_SUCCESS) {
226 		return (rv);
227 	}
228 	rw_init(&bd_lock, NULL, RW_DRIVER, NULL);
229 	rv = mod_install(&modlinkage);
230 	if (rv != DDI_SUCCESS) {
231 		rw_destroy(&bd_lock);
232 		ddi_soft_state_fini(&bd_state);
233 	}
234 	return (rv);
235 }
236 
237 int
238 _fini(void)
239 {
240 	int	rv;
241 
242 	rv = mod_remove(&modlinkage);
243 	if (rv == DDI_SUCCESS) {
244 		rw_destroy(&bd_lock);
245 		ddi_soft_state_fini(&bd_state);
246 	}
247 	return (rv);
248 }
249 
250 int
251 _info(struct modinfo *modinfop)
252 {
253 	return (mod_info(&modlinkage, modinfop));
254 }
255 
256 static int
257 bd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resultp)
258 {
259 	bd_t	*bd;
260 	minor_t	inst;
261 
262 	_NOTE(ARGUNUSED(dip));
263 
264 	inst = BDINST((dev_t)arg);
265 
266 	switch (cmd) {
267 	case DDI_INFO_DEVT2DEVINFO:
268 		bd = ddi_get_soft_state(bd_state, inst);
269 		if (bd == NULL) {
270 			return (DDI_FAILURE);
271 		}
272 		*resultp = (void *)bd->d_dip;
273 		break;
274 
275 	case DDI_INFO_DEVT2INSTANCE:
276 		*resultp = (void *)(intptr_t)inst;
277 		break;
278 
279 	default:
280 		return (DDI_FAILURE);
281 	}
282 	return (DDI_SUCCESS);
283 }
284 
285 static int
286 bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
287 {
288 	int		inst;
289 	bd_handle_t	hdl;
290 	bd_t		*bd;
291 	bd_drive_t	drive;
292 	int		rv;
293 	char		name[16];
294 	char		kcache[32];
295 
296 	switch (cmd) {
297 	case DDI_ATTACH:
298 		break;
299 	case DDI_RESUME:
300 		/* We don't do anything native for suspend/resume */
301 		return (DDI_SUCCESS);
302 	default:
303 		return (DDI_FAILURE);
304 	}
305 
306 	inst = ddi_get_instance(dip);
307 	hdl = ddi_get_parent_data(dip);
308 
309 	(void) snprintf(name, sizeof (name), "%s%d",
310 	    ddi_driver_name(dip), ddi_get_instance(dip));
311 	(void) snprintf(kcache, sizeof (kcache), "%s_xfer", name);
312 
313 	if (hdl == NULL) {
314 		cmn_err(CE_WARN, "%s: missing parent data!", name);
315 		return (DDI_FAILURE);
316 	}
317 
318 	if (ddi_soft_state_zalloc(bd_state, inst) != DDI_SUCCESS) {
319 		cmn_err(CE_WARN, "%s: unable to zalloc soft state!", name);
320 		return (DDI_FAILURE);
321 	}
322 	bd = ddi_get_soft_state(bd_state, inst);
323 
324 	if (hdl->h_dma) {
325 		bd->d_dma = *(hdl->h_dma);
326 		bd->d_dma.dma_attr_granular =
327 		    max(DEV_BSIZE, bd->d_dma.dma_attr_granular);
328 		bd->d_use_dma = B_TRUE;
329 
330 		if (bd->d_maxxfer &&
331 		    (bd->d_maxxfer != bd->d_dma.dma_attr_maxxfer)) {
332 			cmn_err(CE_WARN,
333 			    "%s: inconsistent maximum transfer size!",
334 			    name);
335 			/* We force it */
336 			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
337 		} else {
338 			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
339 		}
340 	} else {
341 		bd->d_use_dma = B_FALSE;
342 		if (bd->d_maxxfer == 0) {
343 			bd->d_maxxfer = 1024 * 1024;
344 		}
345 	}
346 	bd->d_ops = hdl->h_ops;
347 	bd->d_private = hdl->h_private;
348 	bd->d_blkshift = 9;	/* 512 bytes, to start */
349 
350 	if (bd->d_maxxfer % DEV_BSIZE) {
351 		cmn_err(CE_WARN, "%s: maximum transfer misaligned!", name);
352 		bd->d_maxxfer &= ~(DEV_BSIZE - 1);
353 	}
354 	if (bd->d_maxxfer < DEV_BSIZE) {
355 		cmn_err(CE_WARN, "%s: maximum transfer size too small!", name);
356 		ddi_soft_state_free(bd_state, inst);
357 		return (DDI_FAILURE);
358 	}
359 
360 	bd->d_dip = dip;
361 	bd->d_handle = hdl;
362 	hdl->h_bd = bd;
363 	ddi_set_driver_private(dip, bd);
364 
365 	mutex_init(&bd->d_iomutex, NULL, MUTEX_DRIVER, NULL);
366 	mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL);
367 	mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL);
368 	cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL);
369 
370 	list_create(&bd->d_waitq, sizeof (bd_xfer_impl_t),
371 	    offsetof(struct bd_xfer_impl, i_linkage));
372 	list_create(&bd->d_runq, sizeof (bd_xfer_impl_t),
373 	    offsetof(struct bd_xfer_impl, i_linkage));
374 
375 	bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8,
376 	    bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0);
377 
378 	bd->d_ksp = kstat_create(ddi_driver_name(dip), inst, NULL, "disk",
379 	    KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
380 	if (bd->d_ksp != NULL) {
381 		bd->d_ksp->ks_lock = &bd->d_iomutex;
382 		kstat_install(bd->d_ksp);
383 		bd->d_kiop = bd->d_ksp->ks_data;
384 	} else {
385 		/*
386 		 * Even if we cannot create the kstat, we create a
387 		 * scratch kstat.  The reason for this is to ensure
388 		 * that we can update the kstat all of the time,
389 		 * without adding an extra branch instruction.
390 		 */
391 		bd->d_kiop = kmem_zalloc(sizeof (kstat_io_t), KM_SLEEP);
392 	}
393 
394 	cmlb_alloc_handle(&bd->d_cmlbh);
395 
396 	bd->d_state = DKIO_NONE;
397 
398 	bzero(&drive, sizeof (drive));
399 	bd->d_ops.o_drive_info(bd->d_private, &drive);
400 	bd->d_qsize = drive.d_qsize;
401 	bd->d_maxxfer = drive.d_maxxfer;
402 	bd->d_removable = drive.d_removable;
403 	bd->d_hotpluggable = drive.d_hotpluggable;
404 
405 	rv = cmlb_attach(dip, &bd_tg_ops, DTYPE_DIRECT,
406 	    bd->d_removable, bd->d_hotpluggable,
407 	    drive.d_lun >= 0 ? DDI_NT_BLOCK_CHAN : DDI_NT_BLOCK,
408 	    CMLB_FAKE_LABEL_ONE_PARTITION, bd->d_cmlbh, 0);
409 	if (rv != 0) {
410 		cmlb_free_handle(&bd->d_cmlbh);
411 		kmem_cache_destroy(bd->d_cache);
412 		mutex_destroy(&bd->d_iomutex);
413 		mutex_destroy(&bd->d_ocmutex);
414 		mutex_destroy(&bd->d_statemutex);
415 		cv_destroy(&bd->d_statecv);
416 		list_destroy(&bd->d_waitq);
417 		list_destroy(&bd->d_runq);
418 		if (bd->d_ksp != NULL) {
419 			kstat_delete(bd->d_ksp);
420 			bd->d_ksp = NULL;
421 		} else {
422 			kmem_free(bd->d_kiop, sizeof (kstat_io_t));
423 		}
424 		ddi_soft_state_free(bd_state, inst);
425 		return (DDI_FAILURE);
426 	}
427 
428 	if (bd->d_ops.o_devid_init != NULL) {
429 		rv = bd->d_ops.o_devid_init(bd->d_private, dip, &bd->d_devid);
430 		if (rv == DDI_SUCCESS) {
431 			if (ddi_devid_register(dip, bd->d_devid) !=
432 			    DDI_SUCCESS) {
433 				cmn_err(CE_WARN,
434 				    "%s: unable to register devid", name);
435 			}
436 		}
437 	}
438 
439 	/*
440 	 * Add a zero-length attribute to tell the world we support
441 	 * kernel ioctls (for layered drivers).  Also set up properties
442 	 * used by HAL to identify removable media.
443 	 */
444 	(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
445 	    DDI_KERNEL_IOCTL, NULL, 0);
446 	if (bd->d_removable) {
447 		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
448 		    "removable-media", NULL, 0);
449 	}
450 	if (bd->d_hotpluggable) {
451 		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
452 		    "hotpluggable", NULL, 0);
453 	}
454 
455 	ddi_report_dev(dip);
456 
457 	return (DDI_SUCCESS);
458 }
459 
460 static int
461 bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
462 {
463 	bd_t	*bd;
464 
465 	bd = ddi_get_driver_private(dip);
466 
467 	switch (cmd) {
468 	case DDI_DETACH:
469 		break;
470 	case DDI_SUSPEND:
471 		/* We don't suspend, but our parent does */
472 		return (DDI_SUCCESS);
473 	default:
474 		return (DDI_FAILURE);
475 	}
476 	if (bd->d_ksp != NULL) {
477 		kstat_delete(bd->d_ksp);
478 		bd->d_ksp = NULL;
479 	} else {
480 		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
481 	}
482 	cmlb_detach(bd->d_cmlbh, 0);
483 	cmlb_free_handle(&bd->d_cmlbh);
484 	if (bd->d_devid)
485 		ddi_devid_free(bd->d_devid);
486 	kmem_cache_destroy(bd->d_cache);
487 	mutex_destroy(&bd->d_iomutex);
488 	mutex_destroy(&bd->d_ocmutex);
489 	mutex_destroy(&bd->d_statemutex);
490 	cv_destroy(&bd->d_statecv);
491 	list_destroy(&bd->d_waitq);
492 	list_destroy(&bd->d_runq);
493 	ddi_soft_state_free(bd_state, ddi_get_instance(dip));
494 	return (DDI_SUCCESS);
495 }
496 
497 static int
498 bd_xfer_ctor(void *buf, void *arg, int kmflag)
499 {
500 	bd_xfer_impl_t	*xi;
501 	bd_t		*bd = arg;
502 	int		(*dcb)(caddr_t);
503 
504 	if (kmflag == KM_SLEEP) {
505 		dcb = DDI_DMA_SLEEP;
506 	} else {
507 		dcb = DDI_DMA_DONTWAIT;
508 	}
509 
510 	xi = buf;
511 	bzero(xi, sizeof (*xi));
512 	xi->i_bd = bd;
513 
514 	if (bd->d_use_dma) {
515 		if (ddi_dma_alloc_handle(bd->d_dip, &bd->d_dma, dcb, NULL,
516 		    &xi->i_dmah) != DDI_SUCCESS) {
517 			return (-1);
518 		}
519 	}
520 
521 	return (0);
522 }
523 
524 static void
525 bd_xfer_dtor(void *buf, void *arg)
526 {
527 	bd_xfer_impl_t	*xi = buf;
528 
529 	_NOTE(ARGUNUSED(arg));
530 
531 	if (xi->i_dmah)
532 		ddi_dma_free_handle(&xi->i_dmah);
533 	xi->i_dmah = NULL;
534 }
535 
536 static bd_xfer_impl_t *
537 bd_xfer_alloc(bd_t *bd, struct buf *bp, int (*func)(void *, bd_xfer_t *),
538     int kmflag)
539 {
540 	bd_xfer_impl_t		*xi;
541 	int			rv;
542 	int			status;
543 	unsigned		dir;
544 	int			(*cb)(caddr_t);
545 	size_t			len;
546 	uint32_t		shift;
547 
548 	if (kmflag == KM_SLEEP) {
549 		cb = DDI_DMA_SLEEP;
550 	} else {
551 		cb = DDI_DMA_DONTWAIT;
552 	}
553 
554 	xi = kmem_cache_alloc(bd->d_cache, kmflag);
555 	if (xi == NULL) {
556 		bioerror(bp, ENOMEM);
557 		return (NULL);
558 	}
559 
560 	ASSERT(bp);
561 	ASSERT(bp->b_bcount);
562 
563 	xi->i_bp = bp;
564 	xi->i_func = func;
565 	xi->i_blkno = bp->b_lblkno;
566 
567 	if (bp->b_bcount == 0) {
568 		xi->i_len = 0;
569 		xi->i_nblks = 0;
570 		xi->i_kaddr = NULL;
571 		xi->i_resid = 0;
572 		xi->i_num_win = 0;
573 		goto done;
574 	}
575 
576 	if (bp->b_flags & B_READ) {
577 		dir = DDI_DMA_READ;
578 		xi->i_func = bd->d_ops.o_read;
579 	} else {
580 		dir = DDI_DMA_WRITE;
581 		xi->i_func = bd->d_ops.o_write;
582 	}
583 
584 	shift = bd->d_blkshift;
585 	xi->i_blkshift = shift;
586 
587 	if (!bd->d_use_dma) {
588 		bp_mapin(bp);
589 		rv = 0;
590 		xi->i_offset = 0;
591 		xi->i_num_win =
592 		    (bp->b_bcount + (bd->d_maxxfer - 1)) / bd->d_maxxfer;
593 		xi->i_cur_win = 0;
594 		xi->i_len = min(bp->b_bcount, bd->d_maxxfer);
595 		xi->i_nblks = xi->i_len >> shift;
596 		xi->i_kaddr = bp->b_un.b_addr;
597 		xi->i_resid = bp->b_bcount;
598 	} else {
599 
600 		/*
601 		 * We have to use consistent DMA if the address is misaligned.
602 		 */
603 		if (((bp->b_flags & (B_PAGEIO | B_REMAPPED)) != B_PAGEIO) &&
604 		    ((uintptr_t)bp->b_un.b_addr & 0x7)) {
605 			dir |= DDI_DMA_CONSISTENT | DDI_DMA_PARTIAL;
606 		} else {
607 			dir |= DDI_DMA_STREAMING | DDI_DMA_PARTIAL;
608 		}
609 
610 		status = ddi_dma_buf_bind_handle(xi->i_dmah, bp, dir, cb,
611 		    NULL, &xi->i_dmac, &xi->i_ndmac);
612 		switch (status) {
613 		case DDI_DMA_MAPPED:
614 			xi->i_num_win = 1;
615 			xi->i_cur_win = 0;
616 			xi->i_offset = 0;
617 			xi->i_len = bp->b_bcount;
618 			xi->i_nblks = xi->i_len >> shift;
619 			xi->i_resid = bp->b_bcount;
620 			rv = 0;
621 			break;
622 		case DDI_DMA_PARTIAL_MAP:
623 			xi->i_cur_win = 0;
624 
625 			if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) !=
626 			    DDI_SUCCESS) ||
627 			    (ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset,
628 			    &len, &xi->i_dmac, &xi->i_ndmac) !=
629 			    DDI_SUCCESS) ||
630 			    (P2PHASE(len, shift) != 0)) {
631 				(void) ddi_dma_unbind_handle(xi->i_dmah);
632 				rv = EFAULT;
633 				goto done;
634 			}
635 			xi->i_len = len;
636 			xi->i_nblks = xi->i_len >> shift;
637 			xi->i_resid = bp->b_bcount;
638 			rv = 0;
639 			break;
640 		case DDI_DMA_NORESOURCES:
641 			rv = EAGAIN;
642 			goto done;
643 		case DDI_DMA_TOOBIG:
644 			rv = EINVAL;
645 			goto done;
646 		case DDI_DMA_NOMAPPING:
647 		case DDI_DMA_INUSE:
648 		default:
649 			rv = EFAULT;
650 			goto done;
651 		}
652 	}
653 
654 done:
655 	if (rv != 0) {
656 		kmem_cache_free(bd->d_cache, xi);
657 		bioerror(bp, rv);
658 		return (NULL);
659 	}
660 
661 	return (xi);
662 }
663 
664 static void
665 bd_xfer_free(bd_xfer_impl_t *xi)
666 {
667 	if (xi->i_dmah) {
668 		(void) ddi_dma_unbind_handle(xi->i_dmah);
669 	}
670 	kmem_cache_free(xi->i_bd->d_cache, xi);
671 }
672 
673 static int
674 bd_open(dev_t *devp, int flag, int otyp, cred_t *credp)
675 {
676 	dev_t		dev = *devp;
677 	bd_t		*bd;
678 	minor_t		part;
679 	minor_t		inst;
680 	uint64_t	mask;
681 	boolean_t	ndelay;
682 	int		rv;
683 	diskaddr_t	nblks;
684 	diskaddr_t	lba;
685 
686 	_NOTE(ARGUNUSED(credp));
687 
688 	part = BDPART(dev);
689 	inst = BDINST(dev);
690 
691 	if (otyp >= OTYPCNT)
692 		return (EINVAL);
693 
694 	ndelay = (flag & (FNDELAY | FNONBLOCK)) ? B_TRUE : B_FALSE;
695 
696 	/*
697 	 * Block any DR events from changing the set of registered
698 	 * devices while we function.
699 	 */
700 	rw_enter(&bd_lock, RW_READER);
701 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
702 		rw_exit(&bd_lock);
703 		return (ENXIO);
704 	}
705 
706 	mutex_enter(&bd->d_ocmutex);
707 
708 	ASSERT(part < 64);
709 	mask = (1U << part);
710 
711 	bd_update_state(bd);
712 
713 	if (cmlb_validate(bd->d_cmlbh, 0, 0) != 0) {
714 
715 		/* non-blocking opens are allowed to succeed */
716 		if (!ndelay) {
717 			rv = ENXIO;
718 			goto done;
719 		}
720 	} else if (cmlb_partinfo(bd->d_cmlbh, part, &nblks, &lba,
721 	    NULL, NULL, 0) == 0) {
722 
723 		/*
724 		 * We read the partinfo, verify valid ranges.  If the
725 		 * partition is invalid, and we aren't blocking or
726 		 * doing a raw access, then fail. (Non-blocking and
727 		 * raw accesses can still succeed to allow a disk with
728 		 * bad partition data to opened by format and fdisk.)
729 		 */
730 		if ((!nblks) && ((!ndelay) || (otyp != OTYP_CHR))) {
731 			rv = ENXIO;
732 			goto done;
733 		}
734 	} else if (!ndelay) {
735 		/*
736 		 * cmlb_partinfo failed -- invalid partition or no
737 		 * disk label.
738 		 */
739 		rv = ENXIO;
740 		goto done;
741 	}
742 
743 	if ((flag & FWRITE) && bd->d_rdonly) {
744 		rv = EROFS;
745 		goto done;
746 	}
747 
748 	if ((bd->d_open_excl) & (mask)) {
749 		rv = EBUSY;
750 		goto done;
751 	}
752 	if (flag & FEXCL) {
753 		if (bd->d_open_lyr[part]) {
754 			rv = EBUSY;
755 			goto done;
756 		}
757 		for (int i = 0; i < OTYP_LYR; i++) {
758 			if (bd->d_open_reg[i] & mask) {
759 				rv = EBUSY;
760 				goto done;
761 			}
762 		}
763 	}
764 
765 	if (otyp == OTYP_LYR) {
766 		bd->d_open_lyr[part]++;
767 	} else {
768 		bd->d_open_reg[otyp] |= mask;
769 	}
770 	if (flag & FEXCL) {
771 		bd->d_open_excl |= mask;
772 	}
773 
774 	rv = 0;
775 done:
776 	mutex_exit(&bd->d_ocmutex);
777 	rw_exit(&bd_lock);
778 
779 	return (rv);
780 }
781 
782 static int
783 bd_close(dev_t dev, int flag, int otyp, cred_t *credp)
784 {
785 	bd_t		*bd;
786 	minor_t		inst;
787 	minor_t		part;
788 	uint64_t	mask;
789 	boolean_t	last = B_TRUE;
790 
791 	_NOTE(ARGUNUSED(flag));
792 	_NOTE(ARGUNUSED(credp));
793 
794 	part = BDPART(dev);
795 	inst = BDINST(dev);
796 
797 	ASSERT(part < 64);
798 	mask = (1U << part);
799 
800 	rw_enter(&bd_lock, RW_READER);
801 
802 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
803 		rw_exit(&bd_lock);
804 		return (ENXIO);
805 	}
806 
807 	mutex_enter(&bd->d_ocmutex);
808 	if (bd->d_open_excl & mask) {
809 		bd->d_open_excl &= ~mask;
810 	}
811 	if (otyp == OTYP_LYR) {
812 		bd->d_open_lyr[part]--;
813 	} else {
814 		bd->d_open_reg[otyp] &= ~mask;
815 	}
816 	for (int i = 0; i < 64; i++) {
817 		if (bd->d_open_lyr[part]) {
818 			last = B_FALSE;
819 		}
820 	}
821 	for (int i = 0; last && (i < OTYP_LYR); i++) {
822 		if (bd->d_open_reg[i]) {
823 			last = B_FALSE;
824 		}
825 	}
826 	mutex_exit(&bd->d_ocmutex);
827 
828 	if (last) {
829 		cmlb_invalidate(bd->d_cmlbh, 0);
830 	}
831 	rw_exit(&bd_lock);
832 
833 	return (0);
834 }
835 
836 static int
837 bd_dump(dev_t dev, caddr_t caddr, daddr_t blkno, int nblk)
838 {
839 	minor_t		inst;
840 	minor_t		part;
841 	diskaddr_t	pstart;
842 	diskaddr_t	psize;
843 	bd_t		*bd;
844 	bd_xfer_impl_t	*xi;
845 	buf_t		*bp;
846 	int		rv;
847 
848 	rw_enter(&bd_lock, RW_READER);
849 
850 	part = BDPART(dev);
851 	inst = BDINST(dev);
852 
853 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
854 		rw_exit(&bd_lock);
855 		return (ENXIO);
856 	}
857 	/*
858 	 * do cmlb, but do it synchronously unless we already have the
859 	 * partition (which we probably should.)
860 	 */
861 	if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL,
862 	    (void *)1)) {
863 		rw_exit(&bd_lock);
864 		return (ENXIO);
865 	}
866 
867 	if ((blkno + nblk) > psize) {
868 		rw_exit(&bd_lock);
869 		return (EINVAL);
870 	}
871 	bp = getrbuf(KM_NOSLEEP);
872 	if (bp == NULL) {
873 		rw_exit(&bd_lock);
874 		return (ENOMEM);
875 	}
876 
877 	bp->b_bcount = nblk << bd->d_blkshift;
878 	bp->b_resid = bp->b_bcount;
879 	bp->b_lblkno = blkno;
880 	bp->b_un.b_addr = caddr;
881 
882 	xi = bd_xfer_alloc(bd, bp,  bd->d_ops.o_write, KM_NOSLEEP);
883 	if (xi == NULL) {
884 		rw_exit(&bd_lock);
885 		freerbuf(bp);
886 		return (ENOMEM);
887 	}
888 	xi->i_blkno = blkno + pstart;
889 	xi->i_flags = BD_XFER_POLL;
890 	bd_submit(bd, xi);
891 	rw_exit(&bd_lock);
892 
893 	/*
894 	 * Generally, we should have run this entirely synchronously
895 	 * at this point and the biowait call should be a no-op.  If
896 	 * it didn't happen this way, it's a bug in the underlying
897 	 * driver not honoring BD_XFER_POLL.
898 	 */
899 	(void) biowait(bp);
900 	rv = geterror(bp);
901 	freerbuf(bp);
902 	return (rv);
903 }
904 
905 static int
906 bd_read(dev_t dev, struct uio *uio, cred_t *credp)
907 {
908 	_NOTE(ARGUNUSED(credp));
909 	return (physio(bd_strategy, NULL, dev, B_READ, minphys, uio));
910 }
911 
912 static int
913 bd_write(dev_t dev, struct uio *uio, cred_t *credp)
914 {
915 	_NOTE(ARGUNUSED(credp));
916 	return (physio(bd_strategy, NULL, dev, B_WRITE, minphys, uio));
917 }
918 
919 static int
920 bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp)
921 {
922 	_NOTE(ARGUNUSED(credp));
923 	return (aphysio(bd_strategy, anocancel, dev, B_READ, minphys, aio));
924 }
925 
926 static int
927 bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp)
928 {
929 	_NOTE(ARGUNUSED(credp));
930 	return (aphysio(bd_strategy, anocancel, dev, B_WRITE, minphys, aio));
931 }
932 
933 static int
934 bd_strategy(struct buf *bp)
935 {
936 	minor_t		inst;
937 	minor_t		part;
938 	bd_t		*bd;
939 	diskaddr_t	p_lba;
940 	diskaddr_t	p_nblks;
941 	diskaddr_t	b_nblks;
942 	bd_xfer_impl_t	*xi;
943 	uint32_t	shift;
944 	int		(*func)(void *, bd_xfer_t *);
945 
946 	part = BDPART(bp->b_edev);
947 	inst = BDINST(bp->b_edev);
948 
949 	ASSERT(bp);
950 
951 	bp->b_resid = bp->b_bcount;
952 
953 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
954 		bioerror(bp, ENXIO);
955 		biodone(bp);
956 		return (0);
957 	}
958 
959 	if (cmlb_partinfo(bd->d_cmlbh, part, &p_nblks, &p_lba,
960 	    NULL, NULL, 0)) {
961 		bioerror(bp, ENXIO);
962 		biodone(bp);
963 		return (0);
964 	}
965 
966 	shift = bd->d_blkshift;
967 
968 	if ((P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
969 	    (bp->b_lblkno > p_nblks)) {
970 		bioerror(bp, ENXIO);
971 		biodone(bp);
972 		return (0);
973 	}
974 	b_nblks = bp->b_bcount >> shift;
975 	if ((bp->b_lblkno == p_nblks) || (bp->b_bcount == 0)) {
976 		biodone(bp);
977 		return (0);
978 	}
979 
980 	if ((b_nblks + bp->b_lblkno) > p_nblks) {
981 		bp->b_resid = ((bp->b_lblkno + b_nblks - p_nblks) << shift);
982 		bp->b_bcount -= bp->b_resid;
983 	} else {
984 		bp->b_resid = 0;
985 	}
986 	func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write;
987 
988 	xi = bd_xfer_alloc(bd, bp, func, KM_NOSLEEP);
989 	if (xi == NULL) {
990 		xi = bd_xfer_alloc(bd, bp, func, KM_PUSHPAGE);
991 	}
992 	if (xi == NULL) {
993 		/* bd_request_alloc will have done bioerror */
994 		biodone(bp);
995 		return (0);
996 	}
997 	xi->i_blkno = bp->b_lblkno + p_lba;
998 
999 	bd_submit(bd, xi);
1000 
1001 	return (0);
1002 }
1003 
1004 static int
1005 bd_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp, int *rvalp)
1006 {
1007 	minor_t		inst;
1008 	uint16_t	part;
1009 	bd_t		*bd;
1010 	void		*ptr = (void *)arg;
1011 	int		rv;
1012 
1013 	part = BDPART(dev);
1014 	inst = BDINST(dev);
1015 
1016 	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
1017 		return (ENXIO);
1018 	}
1019 
1020 	rv = cmlb_ioctl(bd->d_cmlbh, dev, cmd, arg, flag, credp, rvalp, 0);
1021 	if (rv != ENOTTY)
1022 		return (rv);
1023 
1024 	switch (cmd) {
1025 	case DKIOCGMEDIAINFO: {
1026 		struct dk_minfo minfo;
1027 
1028 		/* make sure our state information is current */
1029 		bd_update_state(bd);
1030 		bzero(&minfo, sizeof (minfo));
1031 		minfo.dki_media_type = DK_FIXED_DISK;
1032 		minfo.dki_lbsize = (1U << bd->d_blkshift);
1033 		minfo.dki_capacity = bd->d_numblks;
1034 		if (ddi_copyout(&minfo, ptr, sizeof (minfo), flag))  {
1035 			return (EFAULT);
1036 		}
1037 		return (0);
1038 	}
1039 	case DKIOCINFO: {
1040 		struct dk_cinfo cinfo;
1041 		bzero(&cinfo, sizeof (cinfo));
1042 		cinfo.dki_ctype = DKC_BLKDEV;
1043 		cinfo.dki_cnum = ddi_get_instance(ddi_get_parent(bd->d_dip));
1044 		(void) snprintf(cinfo.dki_cname, sizeof (cinfo.dki_cname),
1045 		    "%s", ddi_driver_name(ddi_get_parent(bd->d_dip)));
1046 		(void) snprintf(cinfo.dki_dname, sizeof (cinfo.dki_dname),
1047 		    "%s", ddi_driver_name(bd->d_dip));
1048 		cinfo.dki_unit = inst;
1049 		cinfo.dki_flags = DKI_FMTVOL;
1050 		cinfo.dki_partition = part;
1051 		cinfo.dki_maxtransfer = bd->d_maxxfer / DEV_BSIZE;
1052 		cinfo.dki_addr = 0;
1053 		cinfo.dki_slave = 0;
1054 		cinfo.dki_space = 0;
1055 		cinfo.dki_prio = 0;
1056 		cinfo.dki_vec = 0;
1057 		if (ddi_copyout(&cinfo, ptr, sizeof (cinfo), flag))  {
1058 			return (EFAULT);
1059 		}
1060 		return (0);
1061 	}
1062 	case DKIOCREMOVABLE: {
1063 		int i;
1064 		i = bd->d_removable ? 1 : 0;
1065 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1066 			return (EFAULT);
1067 		}
1068 		return (0);
1069 	}
1070 	case DKIOCHOTPLUGGABLE: {
1071 		int i;
1072 		i = bd->d_hotpluggable ? 1 : 0;
1073 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1074 			return (EFAULT);
1075 		}
1076 		return (0);
1077 	}
1078 	case DKIOCREADONLY: {
1079 		int i;
1080 		i = bd->d_rdonly ? 1 : 0;
1081 		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
1082 			return (EFAULT);
1083 		}
1084 		return (0);
1085 	}
1086 	case DKIOCSTATE: {
1087 		enum dkio_state	state;
1088 		if (ddi_copyin(ptr, &state, sizeof (state), flag)) {
1089 			return (EFAULT);
1090 		}
1091 		if ((rv = bd_check_state(bd, &state)) != 0) {
1092 			return (rv);
1093 		}
1094 		if (ddi_copyout(&state, ptr, sizeof (state), flag)) {
1095 			return (EFAULT);
1096 		}
1097 		return (0);
1098 	}
1099 	case DKIOCFLUSHWRITECACHE: {
1100 		struct dk_callback *dkc;
1101 
1102 		dkc = flag & FKIOCTL ? (void *)arg : NULL;
1103 		rv = bd_flush_write_cache(bd, dkc);
1104 		return (rv);
1105 	}
1106 
1107 	default:
1108 		break;
1109 
1110 	}
1111 	return (ENOTTY);
1112 }
1113 
1114 static int
1115 bd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
1116     char *name, caddr_t valuep, int *lengthp)
1117 {
1118 	bd_t	*bd;
1119 
1120 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1121 	if (bd == NULL)
1122 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
1123 		    name, valuep, lengthp));
1124 
1125 	return (cmlb_prop_op(bd->d_cmlbh, dev, dip, prop_op, mod_flags, name,
1126 	    valuep, lengthp, BDPART(dev), 0));
1127 }
1128 
1129 
1130 static int
1131 bd_tg_rdwr(dev_info_t *dip, uchar_t cmd, void *bufaddr, diskaddr_t start,
1132     size_t length, void *tg_cookie)
1133 {
1134 	bd_t		*bd;
1135 	buf_t		*bp;
1136 	bd_xfer_impl_t	*xi;
1137 	int		rv;
1138 	int		(*func)(void *, bd_xfer_t *);
1139 	int		kmflag;
1140 
1141 	/*
1142 	 * If we are running in polled mode (such as during dump(9e)
1143 	 * execution), then we cannot sleep for kernel allocations.
1144 	 */
1145 	kmflag = tg_cookie ? KM_NOSLEEP : KM_SLEEP;
1146 
1147 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1148 
1149 	if (P2PHASE(length, (1U << bd->d_blkshift)) != 0) {
1150 		/* We can only transfer whole blocks at a time! */
1151 		return (EINVAL);
1152 	}
1153 
1154 	if ((bp = getrbuf(kmflag)) == NULL) {
1155 		return (ENOMEM);
1156 	}
1157 
1158 	switch (cmd) {
1159 	case TG_READ:
1160 		bp->b_flags = B_READ;
1161 		func = bd->d_ops.o_read;
1162 		break;
1163 	case TG_WRITE:
1164 		bp->b_flags = B_WRITE;
1165 		func = bd->d_ops.o_write;
1166 		break;
1167 	default:
1168 		freerbuf(bp);
1169 		return (EINVAL);
1170 	}
1171 
1172 	bp->b_un.b_addr = bufaddr;
1173 	bp->b_bcount = length;
1174 	xi = bd_xfer_alloc(bd, bp, func, kmflag);
1175 	if (xi == NULL) {
1176 		rv = geterror(bp);
1177 		freerbuf(bp);
1178 		return (rv);
1179 	}
1180 	xi->i_flags = tg_cookie ? BD_XFER_POLL : 0;
1181 	xi->i_blkno = start;
1182 	bd_submit(bd, xi);
1183 	(void) biowait(bp);
1184 	rv = geterror(bp);
1185 	freerbuf(bp);
1186 
1187 	return (rv);
1188 }
1189 
1190 static int
1191 bd_tg_getinfo(dev_info_t *dip, int cmd, void *arg, void *tg_cookie)
1192 {
1193 	bd_t		*bd;
1194 
1195 	_NOTE(ARGUNUSED(tg_cookie));
1196 	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
1197 
1198 	switch (cmd) {
1199 	case TG_GETPHYGEOM:
1200 	case TG_GETVIRTGEOM:
1201 		/*
1202 		 * We don't have any "geometry" as such, let cmlb
1203 		 * fabricate something.
1204 		 */
1205 		return (ENOTTY);
1206 
1207 	case TG_GETCAPACITY:
1208 		bd_update_state(bd);
1209 		*(diskaddr_t *)arg = bd->d_numblks;
1210 		return (0);
1211 
1212 	case TG_GETBLOCKSIZE:
1213 		*(uint32_t *)arg = (1U << bd->d_blkshift);
1214 		return (0);
1215 
1216 	case TG_GETATTR:
1217 		/*
1218 		 * It turns out that cmlb really doesn't do much for
1219 		 * non-writable media, but lets make the information
1220 		 * available for it in case it does more in the
1221 		 * future.  (The value is currently used for
1222 		 * triggering special behavior for CD-ROMs.)
1223 		 */
1224 		bd_update_state(bd);
1225 		((tg_attribute_t *)arg)->media_is_writable =
1226 		    bd->d_rdonly ? B_FALSE : B_TRUE;
1227 		return (0);
1228 
1229 	default:
1230 		return (EINVAL);
1231 	}
1232 }
1233 
1234 
1235 static void
1236 bd_sched(bd_t *bd)
1237 {
1238 	bd_xfer_impl_t	*xi;
1239 	struct buf	*bp;
1240 	int		rv;
1241 
1242 	mutex_enter(&bd->d_iomutex);
1243 
1244 	while ((bd->d_qactive < bd->d_qsize) &&
1245 	    ((xi = list_remove_head(&bd->d_waitq)) != NULL)) {
1246 		bd->d_qactive++;
1247 		kstat_waitq_to_runq(bd->d_kiop);
1248 		list_insert_tail(&bd->d_runq, xi);
1249 
1250 		/*
1251 		 * Submit the job to the driver.  We drop the I/O mutex
1252 		 * so that we can deal with the case where the driver
1253 		 * completion routine calls back into us synchronously.
1254 		 */
1255 
1256 		mutex_exit(&bd->d_iomutex);
1257 
1258 		rv = xi->i_func(bd->d_private, &xi->i_public);
1259 		if (rv != 0) {
1260 			bp = xi->i_bp;
1261 			bd_xfer_free(xi);
1262 			bioerror(bp, rv);
1263 			biodone(bp);
1264 
1265 			mutex_enter(&bd->d_iomutex);
1266 			bd->d_qactive--;
1267 			kstat_runq_exit(bd->d_kiop);
1268 			list_remove(&bd->d_runq, xi);
1269 		} else {
1270 			mutex_enter(&bd->d_iomutex);
1271 		}
1272 	}
1273 
1274 	mutex_exit(&bd->d_iomutex);
1275 }
1276 
1277 static void
1278 bd_submit(bd_t *bd, bd_xfer_impl_t *xi)
1279 {
1280 	mutex_enter(&bd->d_iomutex);
1281 	list_insert_tail(&bd->d_waitq, xi);
1282 	kstat_waitq_enter(bd->d_kiop);
1283 	mutex_exit(&bd->d_iomutex);
1284 
1285 	bd_sched(bd);
1286 }
1287 
1288 static void
1289 bd_runq_exit(bd_xfer_impl_t *xi, int err)
1290 {
1291 	bd_t	*bd = xi->i_bd;
1292 	buf_t	*bp = xi->i_bp;
1293 
1294 	mutex_enter(&bd->d_iomutex);
1295 	bd->d_qactive--;
1296 	kstat_runq_exit(bd->d_kiop);
1297 	list_remove(&bd->d_runq, xi);
1298 	mutex_exit(&bd->d_iomutex);
1299 
1300 	if (err == 0) {
1301 		if (bp->b_flags & B_READ) {
1302 			bd->d_kiop->reads++;
1303 			bd->d_kiop->nread += (bp->b_bcount - xi->i_resid);
1304 		} else {
1305 			bd->d_kiop->writes++;
1306 			bd->d_kiop->nwritten += (bp->b_bcount - xi->i_resid);
1307 		}
1308 	}
1309 	bd_sched(bd);
1310 }
1311 
1312 static void
1313 bd_update_state(bd_t *bd)
1314 {
1315 	enum	dkio_state	state;
1316 	bd_media_t		media;
1317 	boolean_t		docmlb = B_FALSE;
1318 
1319 	bzero(&media, sizeof (media));
1320 
1321 	mutex_enter(&bd->d_statemutex);
1322 	if (bd->d_ops.o_media_info(bd->d_private, &media) == 0) {
1323 		if ((1U << bd->d_blkshift) != media.m_blksize) {
1324 			if ((media.m_blksize < 512) ||
1325 			    (!ISP2(media.m_blksize)) ||
1326 			    (P2PHASE(bd->d_maxxfer, media.m_blksize))) {
1327 				cmn_err(CE_WARN,
1328 				    "%s%d: Invalid media block size (%d)",
1329 				    ddi_driver_name(bd->d_dip),
1330 				    ddi_get_instance(bd->d_dip),
1331 				    media.m_blksize);
1332 				/*
1333 				 * We can't use the media, treat it as
1334 				 * not present.
1335 				 */
1336 				state = DKIO_EJECTED;
1337 				bd->d_numblks = 0;
1338 			} else {
1339 				bd->d_blkshift = ddi_ffs(media.m_blksize) - 1;
1340 				bd->d_numblks = media.m_nblks;
1341 				bd->d_rdonly = media.m_readonly;
1342 				state = DKIO_INSERTED;
1343 			}
1344 
1345 			/* Device size changed */
1346 			docmlb = B_TRUE;
1347 
1348 		} else {
1349 			if (bd->d_numblks != media.m_nblks) {
1350 				/* Device size changed */
1351 				docmlb = B_TRUE;
1352 			}
1353 			bd->d_numblks = media.m_nblks;
1354 			bd->d_rdonly = media.m_readonly;
1355 			state = DKIO_INSERTED;
1356 		}
1357 
1358 	} else {
1359 		bd->d_numblks = 0;
1360 		state = DKIO_EJECTED;
1361 	}
1362 	if (state != bd->d_state) {
1363 		bd->d_state = state;
1364 		cv_broadcast(&bd->d_statecv);
1365 		docmlb = B_TRUE;
1366 	}
1367 	mutex_exit(&bd->d_statemutex);
1368 
1369 	if (docmlb) {
1370 		if (state == DKIO_INSERTED) {
1371 			(void) cmlb_validate(bd->d_cmlbh, 0, 0);
1372 		} else {
1373 			cmlb_invalidate(bd->d_cmlbh, 0);
1374 		}
1375 	}
1376 }
1377 
1378 static int
1379 bd_check_state(bd_t *bd, enum dkio_state *state)
1380 {
1381 	clock_t		when;
1382 
1383 	for (;;) {
1384 
1385 		bd_update_state(bd);
1386 
1387 		mutex_enter(&bd->d_statemutex);
1388 
1389 		if (bd->d_state != *state) {
1390 			*state = bd->d_state;
1391 			mutex_exit(&bd->d_statemutex);
1392 			break;
1393 		}
1394 
1395 		when = drv_usectohz(1000000);
1396 		if (cv_reltimedwait_sig(&bd->d_statecv, &bd->d_statemutex,
1397 		    when, TR_CLOCK_TICK) == 0) {
1398 			mutex_exit(&bd->d_statemutex);
1399 			return (EINTR);
1400 		}
1401 
1402 		mutex_exit(&bd->d_statemutex);
1403 	}
1404 
1405 	return (0);
1406 }
1407 
1408 static int
1409 bd_flush_write_cache_done(struct buf *bp)
1410 {
1411 	struct dk_callback *dc = (void *)bp->b_private;
1412 
1413 	(*dc->dkc_callback)(dc->dkc_cookie, geterror(bp));
1414 	kmem_free(dc, sizeof (*dc));
1415 	freerbuf(bp);
1416 	return (0);
1417 }
1418 
1419 static int
1420 bd_flush_write_cache(bd_t *bd, struct dk_callback *dkc)
1421 {
1422 	buf_t			*bp;
1423 	struct dk_callback	*dc;
1424 	bd_xfer_impl_t		*xi;
1425 	int			rv;
1426 
1427 	if (bd->d_ops.o_sync_cache == NULL) {
1428 		return (ENOTSUP);
1429 	}
1430 	if ((bp = getrbuf(KM_SLEEP)) == NULL) {
1431 		return (ENOMEM);
1432 	}
1433 	bp->b_resid = 0;
1434 	bp->b_bcount = 0;
1435 
1436 	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_sync_cache, KM_SLEEP);
1437 	if (xi == NULL) {
1438 		rv = geterror(bp);
1439 		freerbuf(bp);
1440 		return (rv);
1441 	}
1442 
1443 	if (dkc != NULL) {
1444 		/* Make a private copy of the callback structure */
1445 		dc = kmem_alloc(sizeof (*dc), KM_SLEEP);
1446 		*dc = *dkc;
1447 		bp->b_private = dc;
1448 		bp->b_iodone = bd_flush_write_cache_done;
1449 	}
1450 
1451 	bd_submit(bd, xi);
1452 	if (dkc == NULL) {
1453 		/* wait synchronously */
1454 		(void) biowait(bp);
1455 		rv = geterror(bp);
1456 		freerbuf(bp);
1457 	} else {
1458 		/* deferred via callback */
1459 		rv = 0;
1460 	}
1461 	return (rv);
1462 }
1463 
1464 /*
1465  * Nexus support.
1466  */
1467 int
1468 bd_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop,
1469     void *arg, void *result)
1470 {
1471 	bd_handle_t	hdl;
1472 
1473 	switch (ctlop) {
1474 	case DDI_CTLOPS_REPORTDEV:
1475 		cmn_err(CE_CONT, "?Block device: %s@%s, %s%d\n",
1476 		    ddi_node_name(rdip), ddi_get_name_addr(rdip),
1477 		    ddi_driver_name(rdip), ddi_get_instance(rdip));
1478 		return (DDI_SUCCESS);
1479 
1480 	case DDI_CTLOPS_INITCHILD:
1481 		hdl = ddi_get_parent_data((dev_info_t *)arg);
1482 		if (hdl == NULL) {
1483 			return (DDI_NOT_WELL_FORMED);
1484 		}
1485 		ddi_set_name_addr((dev_info_t *)arg, hdl->h_addr);
1486 		return (DDI_SUCCESS);
1487 
1488 	case DDI_CTLOPS_UNINITCHILD:
1489 		ddi_set_name_addr((dev_info_t *)arg, NULL);
1490 		ndi_prop_remove_all((dev_info_t *)arg);
1491 		return (DDI_SUCCESS);
1492 
1493 	default:
1494 		return (ddi_ctlops(dip, rdip, ctlop, arg, result));
1495 	}
1496 }
1497 
1498 /*
1499  * Functions for device drivers.
1500  */
1501 bd_handle_t
1502 bd_alloc_handle(void *private, bd_ops_t *ops, ddi_dma_attr_t *dma, int kmflag)
1503 {
1504 	bd_handle_t	hdl;
1505 
1506 	hdl = kmem_zalloc(sizeof (*hdl), kmflag);
1507 	if (hdl != NULL) {
1508 		hdl->h_ops = *ops;
1509 		hdl->h_dma = dma;
1510 		hdl->h_private = private;
1511 	}
1512 
1513 	return (hdl);
1514 }
1515 
1516 void
1517 bd_free_handle(bd_handle_t hdl)
1518 {
1519 	kmem_free(hdl, sizeof (*hdl));
1520 }
1521 
1522 int
1523 bd_attach_handle(dev_info_t *dip, bd_handle_t hdl)
1524 {
1525 	dev_info_t	*child;
1526 	bd_drive_t	drive;
1527 
1528 	/* if drivers don't override this, make it assume none */
1529 	drive.d_lun = -1;
1530 	hdl->h_ops.o_drive_info(hdl->h_private, &drive);
1531 
1532 	hdl->h_parent = dip;
1533 	hdl->h_name = "blkdev";
1534 
1535 	if (drive.d_lun >= 0) {
1536 		(void) snprintf(hdl->h_addr, sizeof (hdl->h_addr), "%X,%X",
1537 		    drive.d_target, drive.d_lun);
1538 	} else {
1539 		(void) snprintf(hdl->h_addr, sizeof (hdl->h_addr), "%X",
1540 		    drive.d_target);
1541 	}
1542 	if (ndi_devi_alloc(dip, hdl->h_name, (pnode_t)DEVI_SID_NODEID,
1543 	    &child) != NDI_SUCCESS) {
1544 		cmn_err(CE_WARN, "%s%d: unable to allocate node %s@%s",
1545 		    ddi_driver_name(dip), ddi_get_instance(dip),
1546 		    "blkdev", hdl->h_addr);
1547 		return (DDI_FAILURE);
1548 	}
1549 
1550 	ddi_set_parent_data(child, hdl);
1551 	hdl->h_child = child;
1552 
1553 	if (ndi_devi_online(child, 0) == NDI_FAILURE) {
1554 		cmn_err(CE_WARN, "%s%d: failed bringing node %s@%s online",
1555 		    ddi_driver_name(dip), ddi_get_instance(dip),
1556 		    hdl->h_name, hdl->h_addr);
1557 		(void) ndi_devi_free(child);
1558 		return (DDI_FAILURE);
1559 	}
1560 
1561 	return (DDI_SUCCESS);
1562 }
1563 
1564 int
1565 bd_detach_handle(bd_handle_t hdl)
1566 {
1567 	int	circ;
1568 	int	rv;
1569 	char	*devnm;
1570 
1571 	if (hdl->h_child == NULL) {
1572 		return (DDI_SUCCESS);
1573 	}
1574 	ndi_devi_enter(hdl->h_parent, &circ);
1575 	if (i_ddi_node_state(hdl->h_child) < DS_INITIALIZED) {
1576 		rv = ddi_remove_child(hdl->h_child, 0);
1577 	} else {
1578 		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
1579 		(void) ddi_deviname(hdl->h_child, devnm);
1580 		(void) devfs_clean(hdl->h_parent, devnm + 1, DV_CLEAN_FORCE);
1581 		rv = ndi_devi_unconfig_one(hdl->h_parent, devnm + 1, NULL,
1582 		    NDI_DEVI_REMOVE | NDI_UNCONFIG);
1583 		kmem_free(devnm, MAXNAMELEN + 1);
1584 	}
1585 	if (rv == 0) {
1586 		hdl->h_child = NULL;
1587 	}
1588 
1589 	ndi_devi_exit(hdl->h_parent, circ);
1590 	return (rv = NDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
1591 }
1592 
1593 void
1594 bd_xfer_done(bd_xfer_t *xfer, int err)
1595 {
1596 	bd_xfer_impl_t	*xi = (void *)xfer;
1597 	buf_t		*bp = xi->i_bp;
1598 	int		rv;
1599 	bd_t		*bd = xi->i_bd;
1600 	size_t		len;
1601 
1602 	if (err != 0) {
1603 		bd_runq_exit(xi, err);
1604 
1605 		bp->b_resid += xi->i_resid;
1606 		bd_xfer_free(xi);
1607 		bioerror(bp, err);
1608 		biodone(bp);
1609 		return;
1610 	}
1611 
1612 	xi->i_cur_win++;
1613 	xi->i_resid -= xi->i_len;
1614 
1615 	if (xi->i_resid == 0) {
1616 		/* Job completed succcessfully! */
1617 		bd_runq_exit(xi, 0);
1618 
1619 		bd_xfer_free(xi);
1620 		biodone(bp);
1621 		return;
1622 	}
1623 
1624 	xi->i_blkno += xi->i_nblks;
1625 
1626 	if (bd->d_use_dma) {
1627 		/* More transfer still pending... advance to next DMA window. */
1628 		rv = ddi_dma_getwin(xi->i_dmah, xi->i_cur_win,
1629 		    &xi->i_offset, &len, &xi->i_dmac, &xi->i_ndmac);
1630 	} else {
1631 		/* Advance memory window. */
1632 		xi->i_kaddr += xi->i_len;
1633 		xi->i_offset += xi->i_len;
1634 		len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer);
1635 	}
1636 
1637 
1638 	if ((rv != DDI_SUCCESS) ||
1639 	    (P2PHASE(len, (1U << xi->i_blkshift) != 0))) {
1640 		bd_runq_exit(xi, EFAULT);
1641 
1642 		bp->b_resid += xi->i_resid;
1643 		bd_xfer_free(xi);
1644 		bioerror(bp, EFAULT);
1645 		biodone(bp);
1646 		return;
1647 	}
1648 	xi->i_len = len;
1649 	xi->i_nblks = len >> xi->i_blkshift;
1650 
1651 	/* Submit next window to hardware. */
1652 	rv = xi->i_func(bd->d_private, &xi->i_public);
1653 	if (rv != 0) {
1654 		bd_runq_exit(xi, rv);
1655 
1656 		bp->b_resid += xi->i_resid;
1657 		bd_xfer_free(xi);
1658 		bioerror(bp, rv);
1659 		biodone(bp);
1660 	}
1661 }
1662 
1663 void
1664 bd_state_change(bd_handle_t hdl)
1665 {
1666 	bd_t		*bd;
1667 
1668 	if ((bd = hdl->h_bd) != NULL) {
1669 		bd_update_state(bd);
1670 	}
1671 }
1672 
1673 void
1674 bd_mod_init(struct dev_ops *devops)
1675 {
1676 	static struct bus_ops bd_bus_ops = {
1677 		BUSO_REV,		/* busops_rev */
1678 		nullbusmap,		/* bus_map */
1679 		NULL,			/* bus_get_intrspec (OBSOLETE) */
1680 		NULL,			/* bus_add_intrspec (OBSOLETE) */
1681 		NULL,			/* bus_remove_intrspec (OBSOLETE) */
1682 		i_ddi_map_fault,	/* bus_map_fault */
1683 		ddi_dma_map,		/* bus_dma_map */
1684 		ddi_dma_allochdl,	/* bus_dma_allochdl */
1685 		ddi_dma_freehdl,	/* bus_dma_freehdl */
1686 		ddi_dma_bindhdl,	/* bus_dma_bindhdl */
1687 		ddi_dma_unbindhdl,	/* bus_dma_unbindhdl */
1688 		ddi_dma_flush,		/* bus_dma_flush */
1689 		ddi_dma_win,		/* bus_dma_win */
1690 		ddi_dma_mctl,		/* bus_dma_ctl */
1691 		bd_bus_ctl,		/* bus_ctl */
1692 		ddi_bus_prop_op,	/* bus_prop_op */
1693 		NULL,			/* bus_get_eventcookie */
1694 		NULL,			/* bus_add_eventcall */
1695 		NULL,			/* bus_remove_eventcall */
1696 		NULL,			/* bus_post_event */
1697 		NULL,			/* bus_intr_ctl (OBSOLETE) */
1698 		NULL,			/* bus_config */
1699 		NULL,			/* bus_unconfig */
1700 		NULL,			/* bus_fm_init */
1701 		NULL,			/* bus_fm_fini */
1702 		NULL,			/* bus_fm_access_enter */
1703 		NULL,			/* bus_fm_access_exit */
1704 		NULL,			/* bus_power */
1705 		NULL,			/* bus_intr_op */
1706 	};
1707 
1708 	devops->devo_bus_ops = &bd_bus_ops;
1709 
1710 	/*
1711 	 * NB: The device driver is free to supply its own
1712 	 * character entry device support.
1713 	 */
1714 }
1715 
1716 void
1717 bd_mod_fini(struct dev_ops *devops)
1718 {
1719 	devops->devo_bus_ops = NULL;
1720 }
1721