xref: /titanic_51/usr/src/uts/common/xen/io/xdf.c (revision 0cfb4e8b28ab16b0cb224619362e8135673714c3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * xdf.c - Xen Virtual Block Device Driver
29  * TODO:
30  *	- support alternate block size (currently only DEV_BSIZE supported)
31  *	- revalidate geometry for removable devices
32  */
33 
34 #pragma ident	"%Z%%M%	%I%	%E% SMI"
35 
36 #include <sys/ddi.h>
37 #include <sys/sunddi.h>
38 #include <sys/conf.h>
39 #include <sys/cmlb.h>
40 #include <sys/dkio.h>
41 #include <sys/promif.h>
42 #include <sys/sysmacros.h>
43 #include <sys/kstat.h>
44 #include <sys/mach_mmu.h>
45 #ifdef XPV_HVM_DRIVER
46 #include <sys/xpv_support.h>
47 #include <sys/sunndi.h>
48 #endif /* XPV_HVM_DRIVER */
49 #include <public/io/xenbus.h>
50 #include <xen/sys/xenbus_impl.h>
51 #include <xen/sys/xendev.h>
52 #include <sys/gnttab.h>
53 #include <sys/scsi/generic/inquiry.h>
54 #include <xen/io/blkif_impl.h>
55 #include <io/xdf.h>
56 
57 #define	FLUSH_DISKCACHE	0x1
58 #define	WRITE_BARRIER	0x2
59 #define	DEFAULT_FLUSH_BLOCK	156 /* block to write to cause a cache flush */
60 #define	USE_WRITE_BARRIER(vdp)				\
61 	((vdp)->xdf_feature_barrier && !(vdp)->xdf_flush_supported)
62 #define	USE_FLUSH_DISKCACHE(vdp)			\
63 	((vdp)->xdf_feature_barrier && (vdp)->xdf_flush_supported)
64 #define	IS_WRITE_BARRIER(vdp, bp)			\
65 	(!IS_READ(bp) && USE_WRITE_BARRIER(vdp) &&	\
66 	((bp)->b_un.b_addr == (vdp)->xdf_cache_flush_block))
67 #define	IS_FLUSH_DISKCACHE(bp)				\
68 	(!IS_READ(bp) && USE_FLUSH_DISKCACHE(vdp) && ((bp)->b_bcount == 0))
69 
70 static void *vbd_ss;
71 static kmem_cache_t *xdf_vreq_cache;
72 static kmem_cache_t *xdf_gs_cache;
73 static int xdf_maxphys = XB_MAXPHYS;
74 int xdfdebug = 0;
75 extern int do_polled_io;
76 diskaddr_t xdf_flush_block = DEFAULT_FLUSH_BLOCK;
77 int	xdf_barrier_flush_disable = 0;
78 
79 /*
80  * dev_ops and cb_ops entrypoints
81  */
82 static int xdf_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
83 static int xdf_attach(dev_info_t *, ddi_attach_cmd_t);
84 static int xdf_detach(dev_info_t *, ddi_detach_cmd_t);
85 static int xdf_reset(dev_info_t *, ddi_reset_cmd_t);
86 static int xdf_open(dev_t *, int, int, cred_t *);
87 static int xdf_close(dev_t, int, int, struct cred *);
88 static int xdf_strategy(struct buf *);
89 static int xdf_read(dev_t, struct uio *, cred_t *);
90 static int xdf_aread(dev_t, struct aio_req *, cred_t *);
91 static int xdf_write(dev_t, struct uio *, cred_t *);
92 static int xdf_awrite(dev_t, struct aio_req *, cred_t *);
93 static int xdf_dump(dev_t, caddr_t, daddr_t, int);
94 static int xdf_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
95 static uint_t xdf_intr(caddr_t);
96 static int xdf_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *,
97     caddr_t, int *);
98 
99 /*
100  * misc private functions
101  */
102 static int xdf_suspend(dev_info_t *);
103 static int xdf_resume(dev_info_t *);
104 static int xdf_start_connect(xdf_t *);
105 static int xdf_start_disconnect(xdf_t *);
106 static int xdf_post_connect(xdf_t *);
107 static void xdf_post_disconnect(xdf_t *);
108 static void xdf_oe_change(dev_info_t *, ddi_eventcookie_t, void *, void *);
109 static void xdf_iostart(xdf_t *);
110 static void xdf_iofini(xdf_t *, uint64_t, int);
111 static int xdf_prepare_rreq(xdf_t *, struct buf *, blkif_request_t *);
112 static int xdf_drain_io(xdf_t *);
113 static boolean_t xdf_isopen(xdf_t *, int);
114 static int xdf_check_state_transition(xdf_t *, XenbusState);
115 static int xdf_connect(xdf_t *, boolean_t);
116 static int xdf_dmacallback(caddr_t);
117 static void xdf_timeout_handler(void *);
118 static uint_t xdf_iorestart(caddr_t);
119 static v_req_t *vreq_get(xdf_t *, buf_t *);
120 static void vreq_free(xdf_t *, v_req_t *);
121 static int vreq_setup(xdf_t *, v_req_t *);
122 static ge_slot_t *gs_get(xdf_t *, int);
123 static void gs_free(xdf_t *, ge_slot_t *);
124 static grant_ref_t gs_grant(ge_slot_t *, mfn_t);
125 static void unexpectedie(xdf_t *);
126 static void xdfmin(struct buf *);
127 static void xdf_synthetic_pgeom(dev_info_t *, cmlb_geom_t *);
128 extern int xdf_kstat_create(dev_info_t *, char *, int);
129 extern void xdf_kstat_delete(dev_info_t *);
130 
131 #if defined(XPV_HVM_DRIVER)
132 static void xdf_hvm_add(dev_info_t *);
133 static void xdf_hvm_rm(dev_info_t *);
134 static void xdf_hvm_init(void);
135 static void xdf_hvm_fini(void);
136 #endif /* XPV_HVM_DRIVER */
137 
138 static 	struct cb_ops xdf_cbops = {
139 	xdf_open,
140 	xdf_close,
141 	xdf_strategy,
142 	nodev,
143 	xdf_dump,
144 	xdf_read,
145 	xdf_write,
146 	xdf_ioctl,
147 	nodev,
148 	nodev,
149 	nodev,
150 	nochpoll,
151 	xdf_prop_op,
152 	NULL,
153 	D_MP | D_NEW | D_64BIT,
154 	CB_REV,
155 	xdf_aread,
156 	xdf_awrite
157 };
158 
159 struct dev_ops xdf_devops = {
160 	DEVO_REV,		/* devo_rev */
161 	0,			/* devo_refcnt */
162 	xdf_getinfo,		/* devo_getinfo */
163 	nulldev,		/* devo_identify */
164 	nulldev,		/* devo_probe */
165 	xdf_attach,		/* devo_attach */
166 	xdf_detach,		/* devo_detach */
167 	xdf_reset,		/* devo_reset */
168 	&xdf_cbops,		/* devo_cb_ops */
169 	(struct bus_ops *)NULL	/* devo_bus_ops */
170 };
171 
172 static struct modldrv modldrv = {
173 	&mod_driverops,		/* Type of module.  This one is a driver */
174 	"virtual block driver %I%",	/* short description */
175 	&xdf_devops		/* driver specific ops */
176 };
177 
178 static struct modlinkage xdf_modlinkage = {
179 	MODREV_1, (void *)&modldrv, NULL
180 };
181 
182 /*
183  * I/O buffer DMA attributes
184  * Make sure: one DMA window contains BLKIF_MAX_SEGMENTS_PER_REQUEST at most
185  */
186 static ddi_dma_attr_t xb_dma_attr = {
187 	DMA_ATTR_V0,
188 	(uint64_t)0,			/* lowest address */
189 	(uint64_t)0xffffffffffffffff,	/* highest usable address */
190 	(uint64_t)0xffffff,		/* DMA counter limit max */
191 	(uint64_t)XB_BSIZE,		/* alignment in bytes */
192 	XB_BSIZE - 1,			/* bitmap of burst sizes */
193 	XB_BSIZE,			/* min transfer */
194 	(uint64_t)XB_MAX_XFER, 		/* maximum transfer */
195 	(uint64_t)PAGEOFFSET,		/* 1 page segment length  */
196 	BLKIF_MAX_SEGMENTS_PER_REQUEST,	/* maximum number of segments */
197 	XB_BSIZE,			/* granularity */
198 	0,				/* flags (reserved) */
199 };
200 
201 static ddi_device_acc_attr_t xc_acc_attr = {
202 	DDI_DEVICE_ATTR_V0,
203 	DDI_NEVERSWAP_ACC,
204 	DDI_STRICTORDER_ACC
205 };
206 
207 /* callbacks from commmon label */
208 
209 int xdf_lb_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t, void *);
210 int xdf_lb_getinfo(dev_info_t *, int, void *, void *);
211 
212 static cmlb_tg_ops_t xdf_lb_ops = {
213 	TG_DK_OPS_VERSION_1,
214 	xdf_lb_rdwr,
215 	xdf_lb_getinfo
216 };
217 
218 int
219 _init(void)
220 {
221 	int rc;
222 
223 	if ((rc = ddi_soft_state_init(&vbd_ss, sizeof (xdf_t), 0)) != 0)
224 		return (rc);
225 
226 	xdf_vreq_cache = kmem_cache_create("xdf_vreq_cache",
227 	    sizeof (v_req_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
228 	xdf_gs_cache = kmem_cache_create("xdf_gs_cache",
229 	    sizeof (ge_slot_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
230 
231 #if defined(XPV_HVM_DRIVER)
232 	xdf_hvm_init();
233 #endif /* XPV_HVM_DRIVER */
234 
235 	if ((rc = mod_install(&xdf_modlinkage)) != 0) {
236 #if defined(XPV_HVM_DRIVER)
237 		xdf_hvm_fini();
238 #endif /* XPV_HVM_DRIVER */
239 		kmem_cache_destroy(xdf_vreq_cache);
240 		kmem_cache_destroy(xdf_gs_cache);
241 		ddi_soft_state_fini(&vbd_ss);
242 		return (rc);
243 	}
244 
245 	return (rc);
246 }
247 
248 int
249 _fini(void)
250 {
251 
252 	int err;
253 	if ((err = mod_remove(&xdf_modlinkage)) != 0)
254 		return (err);
255 
256 #if defined(XPV_HVM_DRIVER)
257 	xdf_hvm_fini();
258 #endif /* XPV_HVM_DRIVER */
259 
260 	kmem_cache_destroy(xdf_vreq_cache);
261 	kmem_cache_destroy(xdf_gs_cache);
262 	ddi_soft_state_fini(&vbd_ss);
263 
264 	return (0);
265 }
266 
267 int
268 _info(struct modinfo *modinfop)
269 {
270 	return (mod_info(&xdf_modlinkage, modinfop));
271 }
272 
273 /*ARGSUSED*/
274 static int
275 xdf_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **rp)
276 {
277 	int instance;
278 	xdf_t *vbdp;
279 
280 	instance = XDF_INST(getminor((dev_t)arg));
281 
282 	switch (cmd) {
283 	case DDI_INFO_DEVT2DEVINFO:
284 		if ((vbdp = ddi_get_soft_state(vbd_ss, instance)) == NULL) {
285 			*rp = NULL;
286 			return (DDI_FAILURE);
287 		}
288 		*rp = vbdp->xdf_dip;
289 		return (DDI_SUCCESS);
290 
291 	case DDI_INFO_DEVT2INSTANCE:
292 		*rp = (void *)(uintptr_t)instance;
293 		return (DDI_SUCCESS);
294 
295 	default:
296 		return (DDI_FAILURE);
297 	}
298 }
299 
300 static int
301 xdf_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
302 	char *name, caddr_t valuep, int *lengthp)
303 {
304 	int instance = ddi_get_instance(dip);
305 	xdf_t *vdp;
306 	diskaddr_t p_blkcnt;
307 
308 	/*
309 	 * xdf dynamic properties are device specific and size oriented.
310 	 * Requests issued under conditions where size is valid are passed
311 	 * to ddi_prop_op_nblocks with the size information, otherwise the
312 	 * request is passed to ddi_prop_op.
313 	 */
314 	vdp = ddi_get_soft_state(vbd_ss, instance);
315 
316 	if ((dev == DDI_DEV_T_ANY) || (vdp == NULL))
317 		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
318 		    name, valuep, lengthp));
319 
320 	/* do cv_wait until connected or failed */
321 	mutex_enter(&vdp->xdf_dev_lk);
322 	if (xdf_connect(vdp, B_TRUE) != XD_READY) {
323 		mutex_exit(&vdp->xdf_dev_lk);
324 		goto out;
325 	}
326 	mutex_exit(&vdp->xdf_dev_lk);
327 
328 	if (cmlb_partinfo(vdp->xdf_vd_lbl, XDF_PART(getminor(dev)), &p_blkcnt,
329 	    NULL, NULL, NULL, NULL) == 0)
330 		return (ddi_prop_op_nblocks(dev, dip, prop_op, mod_flags,
331 		    name, valuep, lengthp, (uint64_t)p_blkcnt));
332 
333 out:
334 	return (ddi_prop_op(dev, dip, prop_op, mod_flags, name, valuep,
335 	    lengthp));
336 }
337 
338 static int
339 xdf_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
340 {
341 	xdf_t *vdp;
342 	ddi_iblock_cookie_t softibc;
343 	int instance;
344 
345 	xdfdebug = ddi_prop_get_int(DDI_DEV_T_ANY, devi, DDI_PROP_NOTPROM,
346 	    "xdfdebug", 0);
347 
348 	switch (cmd) {
349 		case DDI_ATTACH:
350 			break;
351 
352 		case DDI_RESUME:
353 			return (xdf_resume(devi));
354 
355 		default:
356 			return (DDI_FAILURE);
357 	}
358 
359 	instance = ddi_get_instance(devi);
360 	if (ddi_soft_state_zalloc(vbd_ss, instance) != DDI_SUCCESS)
361 		return (DDI_FAILURE);
362 
363 	DPRINTF(DDI_DBG, ("xdf%d: attaching\n", instance));
364 	vdp = ddi_get_soft_state(vbd_ss, instance);
365 	ddi_set_driver_private(devi, vdp);
366 	vdp->xdf_dip = devi;
367 	cv_init(&vdp->xdf_dev_cv, NULL, CV_DEFAULT, NULL);
368 
369 	if (ddi_get_iblock_cookie(devi, 0, &vdp->xdf_ibc) != DDI_SUCCESS) {
370 		cmn_err(CE_WARN, "xdf@%s: failed to get iblock cookie",
371 		    ddi_get_name_addr(devi));
372 		goto errout0;
373 	}
374 	mutex_init(&vdp->xdf_dev_lk, NULL, MUTEX_DRIVER, (void *)vdp->xdf_ibc);
375 	mutex_init(&vdp->xdf_cb_lk, NULL, MUTEX_DRIVER, (void *)vdp->xdf_ibc);
376 	mutex_init(&vdp->xdf_iostat_lk, NULL, MUTEX_DRIVER,
377 	    (void *)vdp->xdf_ibc);
378 
379 	if (ddi_get_soft_iblock_cookie(devi, DDI_SOFTINT_LOW, &softibc)
380 	    != DDI_SUCCESS) {
381 		cmn_err(CE_WARN, "xdf@%s: failed to get softintr iblock cookie",
382 		    ddi_get_name_addr(devi));
383 		goto errout0;
384 	}
385 	if (ddi_add_softintr(devi, DDI_SOFTINT_LOW, &vdp->xdf_softintr_id,
386 	    &softibc, NULL, xdf_iorestart, (caddr_t)vdp) != DDI_SUCCESS) {
387 		cmn_err(CE_WARN, "xdf@%s: failed to add softintr",
388 		    ddi_get_name_addr(devi));
389 		goto errout0;
390 	}
391 
392 #if !defined(XPV_HVM_DRIVER)
393 	/* create kstat for iostat(1M) */
394 	if (xdf_kstat_create(devi, "xdf", instance) != 0) {
395 		cmn_err(CE_WARN, "xdf@%s: failed to create kstat",
396 		    ddi_get_name_addr(devi));
397 		goto errout0;
398 	}
399 #endif /* !XPV_HVM_DRIVER */
400 
401 	/* driver handles kernel-issued IOCTLs */
402 	if (ddi_prop_create(DDI_DEV_T_NONE, devi, DDI_PROP_CANSLEEP,
403 	    DDI_KERNEL_IOCTL, NULL, 0) != DDI_PROP_SUCCESS) {
404 		cmn_err(CE_WARN, "xdf@%s: cannot create DDI_KERNEL_IOCTL prop",
405 		    ddi_get_name_addr(devi));
406 		goto errout0;
407 	}
408 
409 	/*
410 	 * Initialize the physical geometry stucture.  Note that currently
411 	 * we don't know the size of the backend device so the number
412 	 * of blocks on the device will be initialized to zero.  Once
413 	 * we connect to the backend device we'll update the physical
414 	 * geometry to reflect the real size of the device.
415 	 */
416 	xdf_synthetic_pgeom(devi, &vdp->xdf_pgeom);
417 
418 	/*
419 	 * create default device minor nodes: non-removable disk
420 	 * we will adjust minor nodes after we are connected w/ backend
421 	 */
422 	cmlb_alloc_handle(&vdp->xdf_vd_lbl);
423 	if (cmlb_attach(devi, &xdf_lb_ops, DTYPE_DIRECT, 0, 1, DDI_NT_BLOCK,
424 #if defined(XPV_HVM_DRIVER)
425 	    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT |
426 	    CMLB_INTERNAL_MINOR_NODES,
427 #else /* !XPV_HVM_DRIVER */
428 	    CMLB_FAKE_LABEL_ONE_PARTITION,
429 #endif /* !XPV_HVM_DRIVER */
430 	    vdp->xdf_vd_lbl, NULL) != 0) {
431 		cmn_err(CE_WARN, "xdf@%s: default cmlb attach failed",
432 		    ddi_get_name_addr(devi));
433 		goto errout0;
434 	}
435 
436 	/*
437 	 * We ship with cache-enabled disks
438 	 */
439 	vdp->xdf_wce = 1;
440 
441 	mutex_enter(&vdp->xdf_cb_lk);
442 
443 	/* Watch backend XenbusState change */
444 	if (xvdi_add_event_handler(devi, XS_OE_STATE,
445 	    xdf_oe_change) != DDI_SUCCESS) {
446 		mutex_exit(&vdp->xdf_cb_lk);
447 		goto errout0;
448 	}
449 
450 	if (xdf_start_connect(vdp) != DDI_SUCCESS) {
451 		cmn_err(CE_WARN, "xdf@%s: start connection failed",
452 		    ddi_get_name_addr(devi));
453 		(void) xdf_start_disconnect(vdp);
454 		mutex_exit(&vdp->xdf_cb_lk);
455 		goto errout1;
456 	}
457 
458 	mutex_exit(&vdp->xdf_cb_lk);
459 
460 	list_create(&vdp->xdf_vreq_act, sizeof (v_req_t),
461 	    offsetof(v_req_t, v_link));
462 	list_create(&vdp->xdf_gs_act, sizeof (ge_slot_t),
463 	    offsetof(ge_slot_t, link));
464 
465 #if defined(XPV_HVM_DRIVER)
466 	xdf_hvm_add(devi);
467 
468 	(void) ddi_prop_update_int(DDI_DEV_T_NONE, devi, DDI_NO_AUTODETACH, 1);
469 
470 	/*
471 	 * Report our version to dom0.
472 	 */
473 	if (xenbus_printf(XBT_NULL, "hvmpv/xdf", "version", "%d",
474 	    HVMPV_XDF_VERS))
475 		cmn_err(CE_WARN, "xdf: couldn't write version\n");
476 #endif /* XPV_HVM_DRIVER */
477 
478 	ddi_report_dev(devi);
479 
480 	DPRINTF(DDI_DBG, ("xdf%d: attached\n", instance));
481 
482 	return (DDI_SUCCESS);
483 
484 errout1:
485 	xvdi_remove_event_handler(devi, XS_OE_STATE);
486 errout0:
487 	if (vdp->xdf_vd_lbl != NULL) {
488 		cmlb_detach(vdp->xdf_vd_lbl, NULL);
489 		cmlb_free_handle(&vdp->xdf_vd_lbl);
490 	}
491 #if !defined(XPV_HVM_DRIVER)
492 	xdf_kstat_delete(devi);
493 #endif /* !XPV_HVM_DRIVER */
494 	if (vdp->xdf_softintr_id != NULL)
495 		ddi_remove_softintr(vdp->xdf_softintr_id);
496 	if (vdp->xdf_ibc != NULL) {
497 		mutex_destroy(&vdp->xdf_cb_lk);
498 		mutex_destroy(&vdp->xdf_dev_lk);
499 	}
500 	cv_destroy(&vdp->xdf_dev_cv);
501 	ddi_soft_state_free(vbd_ss, instance);
502 	ddi_set_driver_private(devi, NULL);
503 	ddi_prop_remove_all(devi);
504 	cmn_err(CE_WARN, "xdf@%s: attach failed", ddi_get_name_addr(devi));
505 	return (DDI_FAILURE);
506 }
507 
508 static int
509 xdf_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
510 {
511 	xdf_t *vdp;
512 	int instance;
513 
514 	switch (cmd) {
515 
516 	case DDI_PM_SUSPEND:
517 		break;
518 
519 	case DDI_SUSPEND:
520 		return (xdf_suspend(devi));
521 
522 	case DDI_DETACH:
523 		break;
524 
525 	default:
526 		return (DDI_FAILURE);
527 	}
528 
529 	instance = ddi_get_instance(devi);
530 	DPRINTF(DDI_DBG, ("xdf%d: detaching\n", instance));
531 	vdp = ddi_get_soft_state(vbd_ss, instance);
532 
533 	if (vdp == NULL)
534 		return (DDI_FAILURE);
535 
536 	mutex_enter(&vdp->xdf_dev_lk);
537 	if (xdf_isopen(vdp, -1)) {
538 		mutex_exit(&vdp->xdf_dev_lk);
539 		return (DDI_FAILURE);
540 	}
541 
542 	if (vdp->xdf_status != XD_CLOSED) {
543 		mutex_exit(&vdp->xdf_dev_lk);
544 		return (DDI_FAILURE);
545 	}
546 
547 #if defined(XPV_HVM_DRIVER)
548 	xdf_hvm_rm(devi);
549 #endif /* XPV_HVM_DRIVER */
550 
551 	ASSERT(!ISDMACBON(vdp));
552 	mutex_exit(&vdp->xdf_dev_lk);
553 
554 	if (vdp->xdf_timeout_id != 0)
555 		(void) untimeout(vdp->xdf_timeout_id);
556 
557 	xvdi_remove_event_handler(devi, XS_OE_STATE);
558 
559 	/* we'll support backend running in domU later */
560 #ifdef	DOMU_BACKEND
561 	(void) xvdi_post_event(devi, XEN_HP_REMOVE);
562 #endif
563 
564 	list_destroy(&vdp->xdf_vreq_act);
565 	list_destroy(&vdp->xdf_gs_act);
566 	ddi_prop_remove_all(devi);
567 	xdf_kstat_delete(devi);
568 	ddi_remove_softintr(vdp->xdf_softintr_id);
569 	ddi_set_driver_private(devi, NULL);
570 	cv_destroy(&vdp->xdf_dev_cv);
571 	mutex_destroy(&vdp->xdf_cb_lk);
572 	mutex_destroy(&vdp->xdf_dev_lk);
573 	if (vdp->xdf_cache_flush_block != NULL)
574 		kmem_free(vdp->xdf_flush_mem, 2 * DEV_BSIZE);
575 	ddi_soft_state_free(vbd_ss, instance);
576 	return (DDI_SUCCESS);
577 }
578 
579 static int
580 xdf_suspend(dev_info_t *devi)
581 {
582 	xdf_t *vdp;
583 	int instance;
584 	enum xdf_state st;
585 
586 	instance = ddi_get_instance(devi);
587 
588 	if (xdfdebug & SUSRES_DBG)
589 		xen_printf("xdf_suspend: xdf#%d\n", instance);
590 
591 	if ((vdp = ddi_get_soft_state(vbd_ss, instance)) == NULL)
592 		return (DDI_FAILURE);
593 
594 	xvdi_suspend(devi);
595 
596 	mutex_enter(&vdp->xdf_cb_lk);
597 	mutex_enter(&vdp->xdf_dev_lk);
598 	st = vdp->xdf_status;
599 	/* change status to stop further I/O requests */
600 	if (st == XD_READY)
601 		vdp->xdf_status = XD_SUSPEND;
602 	mutex_exit(&vdp->xdf_dev_lk);
603 	mutex_exit(&vdp->xdf_cb_lk);
604 
605 	/* make sure no more I/O responses left in the ring buffer */
606 	if ((st == XD_INIT) || (st == XD_READY)) {
607 #ifdef XPV_HVM_DRIVER
608 		ec_unbind_evtchn(vdp->xdf_evtchn);
609 		xvdi_free_evtchn(devi);
610 #else /* !XPV_HVM_DRIVER */
611 		(void) ddi_remove_intr(devi, 0, NULL);
612 #endif /* !XPV_HVM_DRIVER */
613 		(void) xdf_drain_io(vdp);
614 		/*
615 		 * no need to teardown the ring buffer here
616 		 * it will be simply re-init'ed during resume when
617 		 * we call xvdi_alloc_ring
618 		 */
619 	}
620 
621 	if (xdfdebug & SUSRES_DBG)
622 		xen_printf("xdf_suspend: SUCCESS\n");
623 
624 	return (DDI_SUCCESS);
625 }
626 
627 /*ARGSUSED*/
628 static int
629 xdf_resume(dev_info_t *devi)
630 {
631 	xdf_t *vdp;
632 	int instance;
633 
634 	instance = ddi_get_instance(devi);
635 	if (xdfdebug & SUSRES_DBG)
636 		xen_printf("xdf_resume: xdf%d\n", instance);
637 
638 	if ((vdp = ddi_get_soft_state(vbd_ss, instance)) == NULL)
639 		return (DDI_FAILURE);
640 
641 	mutex_enter(&vdp->xdf_cb_lk);
642 
643 	if (xvdi_resume(devi) != DDI_SUCCESS) {
644 		mutex_exit(&vdp->xdf_cb_lk);
645 		return (DDI_FAILURE);
646 	}
647 
648 	mutex_enter(&vdp->xdf_dev_lk);
649 	ASSERT(vdp->xdf_status != XD_READY);
650 	vdp->xdf_status = XD_UNKNOWN;
651 	mutex_exit(&vdp->xdf_dev_lk);
652 
653 	if (xdf_start_connect(vdp) != DDI_SUCCESS) {
654 		mutex_exit(&vdp->xdf_cb_lk);
655 		return (DDI_FAILURE);
656 	}
657 
658 	mutex_exit(&vdp->xdf_cb_lk);
659 
660 	if (xdfdebug & SUSRES_DBG)
661 		xen_printf("xdf_resume: done\n");
662 	return (DDI_SUCCESS);
663 }
664 
665 /*ARGSUSED*/
666 static int
667 xdf_reset(dev_info_t *devi, ddi_reset_cmd_t cmd)
668 {
669 	xdf_t *vdp;
670 	int instance;
671 
672 	instance = ddi_get_instance(devi);
673 	DPRINTF(DDI_DBG, ("xdf%d: resetting\n", instance));
674 	if ((vdp = ddi_get_soft_state(vbd_ss, instance)) == NULL)
675 		return (DDI_FAILURE);
676 
677 	/*
678 	 * wait for any outstanding I/O to complete
679 	 */
680 	(void) xdf_drain_io(vdp);
681 
682 	DPRINTF(DDI_DBG, ("xdf%d: reset complete\n", instance));
683 	return (DDI_SUCCESS);
684 }
685 
686 static int
687 xdf_open(dev_t *devp, int flag, int otyp, cred_t *credp)
688 {
689 	minor_t	minor;
690 	xdf_t	*vdp;
691 	int part;
692 	ulong_t parbit;
693 	diskaddr_t p_blkct = 0;
694 	boolean_t firstopen;
695 	boolean_t nodelay;
696 
697 	minor = getminor(*devp);
698 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
699 		return (ENXIO);
700 
701 	nodelay = (flag & (FNDELAY | FNONBLOCK));
702 
703 	DPRINTF(DDI_DBG, ("xdf%d: opening\n", XDF_INST(minor)));
704 
705 	/* do cv_wait until connected or failed */
706 	mutex_enter(&vdp->xdf_dev_lk);
707 	if (!nodelay && (xdf_connect(vdp, B_TRUE) != XD_READY)) {
708 		mutex_exit(&vdp->xdf_dev_lk);
709 		return (ENXIO);
710 	}
711 
712 	if ((flag & FWRITE) && XD_IS_RO(vdp)) {
713 		mutex_exit(&vdp->xdf_dev_lk);
714 		return (EROFS);
715 	}
716 
717 	part = XDF_PART(minor);
718 	parbit = 1 << part;
719 	if ((vdp->xdf_vd_exclopen & parbit) ||
720 	    ((flag & FEXCL) && xdf_isopen(vdp, part))) {
721 		mutex_exit(&vdp->xdf_dev_lk);
722 		return (EBUSY);
723 	}
724 
725 	/* are we the first one to open this node? */
726 	firstopen = !xdf_isopen(vdp, -1);
727 
728 	if (otyp == OTYP_LYR)
729 		vdp->xdf_vd_lyropen[part]++;
730 
731 	vdp->xdf_vd_open[otyp] |= parbit;
732 
733 	if (flag & FEXCL)
734 		vdp->xdf_vd_exclopen |= parbit;
735 
736 	mutex_exit(&vdp->xdf_dev_lk);
737 
738 	/* force a re-validation */
739 	if (firstopen)
740 		cmlb_invalidate(vdp->xdf_vd_lbl, NULL);
741 
742 	/*
743 	 * check size
744 	 * ignore CD/DVD which contains a zero-sized s0
745 	 */
746 	if (!nodelay && !XD_IS_CD(vdp) &&
747 	    ((cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkct,
748 	    NULL, NULL, NULL, NULL) != 0) || (p_blkct == 0))) {
749 		(void) xdf_close(*devp, flag, otyp, credp);
750 		return (ENXIO);
751 	}
752 
753 	return (0);
754 }
755 
756 /*ARGSUSED*/
757 static int
758 xdf_close(dev_t dev, int flag, int otyp, struct cred *credp)
759 {
760 	minor_t	minor;
761 	xdf_t	*vdp;
762 	int part;
763 	ulong_t parbit;
764 
765 	minor = getminor(dev);
766 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
767 		return (ENXIO);
768 
769 	mutex_enter(&vdp->xdf_dev_lk);
770 	part = XDF_PART(minor);
771 	if (!xdf_isopen(vdp, part)) {
772 		mutex_exit(&vdp->xdf_dev_lk);
773 		return (ENXIO);
774 	}
775 	parbit = 1 << part;
776 
777 	ASSERT((vdp->xdf_vd_open[otyp] & parbit) != 0);
778 	if (otyp == OTYP_LYR) {
779 		ASSERT(vdp->xdf_vd_lyropen[part] > 0);
780 		if (--vdp->xdf_vd_lyropen[part] == 0)
781 			vdp->xdf_vd_open[otyp] &= ~parbit;
782 	} else {
783 		vdp->xdf_vd_open[otyp] &= ~parbit;
784 	}
785 	vdp->xdf_vd_exclopen &= ~parbit;
786 
787 	mutex_exit(&vdp->xdf_dev_lk);
788 	return (0);
789 }
790 
791 static int
792 xdf_strategy(struct buf *bp)
793 {
794 	xdf_t	*vdp;
795 	minor_t minor;
796 	diskaddr_t p_blkct, p_blkst;
797 	ulong_t nblks;
798 	int part;
799 
800 	minor = getminor(bp->b_edev);
801 	part = XDF_PART(minor);
802 
803 	vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor));
804 	if ((vdp == NULL) || !xdf_isopen(vdp, part)) {
805 		bioerror(bp, ENXIO);
806 		bp->b_resid = bp->b_bcount;
807 		biodone(bp);
808 		return (0);
809 	}
810 
811 	/* Check for writes to a read only device */
812 	if (!IS_READ(bp) && XD_IS_RO(vdp)) {
813 		bioerror(bp, EROFS);
814 		bp->b_resid = bp->b_bcount;
815 		biodone(bp);
816 		return (0);
817 	}
818 
819 	/* Check if this I/O is accessing a partition or the entire disk */
820 	if ((long)bp->b_private == XB_SLICE_NONE) {
821 		/* This I/O is using an absolute offset */
822 		p_blkct = vdp->xdf_xdev_nblocks;
823 		p_blkst = 0;
824 	} else {
825 		/* This I/O is using a partition relative offset */
826 		if (cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkct,
827 		    &p_blkst, NULL, NULL, NULL)) {
828 			bioerror(bp, ENXIO);
829 			bp->b_resid = bp->b_bcount;
830 			biodone(bp);
831 			return (0);
832 		}
833 	}
834 
835 	/* check for a starting block beyond the disk or partition limit */
836 	if (bp->b_blkno > p_blkct) {
837 		DPRINTF(IO_DBG, ("xdf: block %lld exceeds VBD size %"PRIu64,
838 		    (longlong_t)bp->b_blkno, (uint64_t)p_blkct));
839 		bioerror(bp, EINVAL);
840 		bp->b_resid = bp->b_bcount;
841 		biodone(bp);
842 		return (0);
843 	}
844 
845 	/* Legacy: don't set error flag at this case */
846 	if (bp->b_blkno == p_blkct) {
847 		bp->b_resid = bp->b_bcount;
848 		biodone(bp);
849 		return (0);
850 	}
851 
852 	/* Adjust for partial transfer */
853 	nblks = bp->b_bcount >> XB_BSHIFT;
854 	if ((bp->b_blkno + nblks) > p_blkct) {
855 		bp->b_resid = ((bp->b_blkno + nblks) - p_blkct) << XB_BSHIFT;
856 		bp->b_bcount -= bp->b_resid;
857 	}
858 
859 	DPRINTF(IO_DBG, ("xdf: strategy blk %lld len %lu\n",
860 	    (longlong_t)bp->b_blkno, (ulong_t)bp->b_bcount));
861 
862 	/* Fix up the buf struct */
863 	bp->b_flags |= B_BUSY;
864 	bp->av_forw = bp->av_back = NULL; /* not tagged with a v_req */
865 	bp->b_private = (void *)(uintptr_t)p_blkst;
866 
867 	mutex_enter(&vdp->xdf_dev_lk);
868 	if (vdp->xdf_xdev_iostat != NULL)
869 		kstat_waitq_enter(KSTAT_IO_PTR(vdp->xdf_xdev_iostat));
870 	if (vdp->xdf_f_act == NULL) {
871 		vdp->xdf_f_act = vdp->xdf_l_act = bp;
872 	} else {
873 		vdp->xdf_l_act->av_forw = bp;
874 		vdp->xdf_l_act = bp;
875 	}
876 	mutex_exit(&vdp->xdf_dev_lk);
877 
878 	xdf_iostart(vdp);
879 	if (do_polled_io)
880 		(void) xdf_drain_io(vdp);
881 	return (0);
882 }
883 
884 /*ARGSUSED*/
885 static int
886 xdf_read(dev_t dev, struct uio *uiop, cred_t *credp)
887 {
888 
889 	xdf_t	*vdp;
890 	minor_t minor;
891 	diskaddr_t p_blkcnt;
892 	int part;
893 
894 	minor = getminor(dev);
895 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
896 		return (ENXIO);
897 
898 	DPRINTF(IO_DBG, ("xdf: read offset 0x%"PRIx64"\n",
899 	    (int64_t)uiop->uio_offset));
900 
901 	part = XDF_PART(minor);
902 	if (!xdf_isopen(vdp, part))
903 		return (ENXIO);
904 
905 	if (cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkcnt,
906 	    NULL, NULL, NULL, NULL))
907 		return (ENXIO);
908 
909 	if (U_INVAL(uiop))
910 		return (EINVAL);
911 
912 	return (physio(xdf_strategy, NULL, dev, B_READ, xdfmin, uiop));
913 }
914 
915 /*ARGSUSED*/
916 static int
917 xdf_write(dev_t dev, struct uio *uiop, cred_t *credp)
918 {
919 	xdf_t *vdp;
920 	minor_t minor;
921 	diskaddr_t p_blkcnt;
922 	int part;
923 
924 	minor = getminor(dev);
925 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
926 		return (ENXIO);
927 
928 	DPRINTF(IO_DBG, ("xdf: write offset 0x%"PRIx64"\n",
929 	    (int64_t)uiop->uio_offset));
930 
931 	part = XDF_PART(minor);
932 	if (!xdf_isopen(vdp, part))
933 		return (ENXIO);
934 
935 	if (cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkcnt,
936 	    NULL, NULL, NULL, NULL))
937 		return (ENXIO);
938 
939 	if (uiop->uio_loffset >= XB_DTOB(p_blkcnt))
940 		return (ENOSPC);
941 
942 	if (U_INVAL(uiop))
943 		return (EINVAL);
944 
945 	return (physio(xdf_strategy, NULL, dev, B_WRITE, minphys, uiop));
946 }
947 
948 /*ARGSUSED*/
949 static int
950 xdf_aread(dev_t dev, struct aio_req *aiop, cred_t *credp)
951 {
952 	xdf_t	*vdp;
953 	minor_t minor;
954 	struct uio *uiop = aiop->aio_uio;
955 	diskaddr_t p_blkcnt;
956 	int part;
957 
958 	minor = getminor(dev);
959 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
960 		return (ENXIO);
961 
962 	part = XDF_PART(minor);
963 	if (!xdf_isopen(vdp, part))
964 		return (ENXIO);
965 
966 	if (cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkcnt,
967 	    NULL, NULL, NULL, NULL))
968 		return (ENXIO);
969 
970 	if (uiop->uio_loffset >= XB_DTOB(p_blkcnt))
971 		return (ENOSPC);
972 
973 	if (U_INVAL(uiop))
974 		return (EINVAL);
975 
976 	return (aphysio(xdf_strategy, anocancel, dev, B_READ, minphys, aiop));
977 }
978 
979 /*ARGSUSED*/
980 static int
981 xdf_awrite(dev_t dev, struct aio_req *aiop, cred_t *credp)
982 {
983 	xdf_t *vdp;
984 	minor_t minor;
985 	struct uio *uiop = aiop->aio_uio;
986 	diskaddr_t p_blkcnt;
987 	int part;
988 
989 	minor = getminor(dev);
990 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
991 		return (ENXIO);
992 
993 	part = XDF_PART(minor);
994 	if (!xdf_isopen(vdp, part))
995 		return (ENXIO);
996 
997 	if (cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkcnt,
998 	    NULL, NULL, NULL, NULL))
999 		return (ENXIO);
1000 
1001 	if (uiop->uio_loffset >= XB_DTOB(p_blkcnt))
1002 		return (ENOSPC);
1003 
1004 	if (U_INVAL(uiop))
1005 		return (EINVAL);
1006 
1007 	return (aphysio(xdf_strategy, anocancel, dev, B_WRITE, minphys, aiop));
1008 }
1009 
1010 static int
1011 xdf_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk)
1012 {
1013 	struct buf dumpbuf, *dbp;
1014 	xdf_t	*vdp;
1015 	minor_t minor;
1016 	int err = 0;
1017 	int part;
1018 	diskaddr_t p_blkcnt, p_blkst;
1019 
1020 	minor = getminor(dev);
1021 	if ((vdp = ddi_get_soft_state(vbd_ss, XDF_INST(minor))) == NULL)
1022 		return (ENXIO);
1023 
1024 	DPRINTF(IO_DBG, ("xdf: dump addr (0x%p) blk (%ld) nblks (%d)\n",
1025 	    addr, blkno, nblk));
1026 
1027 	part = XDF_PART(minor);
1028 	if (!xdf_isopen(vdp, part))
1029 		return (ENXIO);
1030 
1031 	if (cmlb_partinfo(vdp->xdf_vd_lbl, part, &p_blkcnt, &p_blkst,
1032 	    NULL, NULL, NULL))
1033 		return (ENXIO);
1034 
1035 	if ((blkno + nblk) > p_blkcnt) {
1036 		cmn_err(CE_WARN, "xdf: block %ld exceeds VBD size %"PRIu64,
1037 		    blkno + nblk, (uint64_t)p_blkcnt);
1038 		return (EINVAL);
1039 	}
1040 
1041 	dbp = &dumpbuf;
1042 	bioinit(dbp);
1043 	dbp->b_flags = B_BUSY;
1044 	dbp->b_un.b_addr = addr;
1045 	dbp->b_bcount = nblk << DEV_BSHIFT;
1046 	dbp->b_blkno = blkno;
1047 	dbp->b_edev = dev;
1048 	dbp->b_private = (void *)(uintptr_t)p_blkst;
1049 
1050 	mutex_enter(&vdp->xdf_dev_lk);
1051 	if (vdp->xdf_xdev_iostat != NULL)
1052 		kstat_waitq_enter(KSTAT_IO_PTR(vdp->xdf_xdev_iostat));
1053 	if (vdp->xdf_f_act == NULL) {
1054 		vdp->xdf_f_act = vdp->xdf_l_act = dbp;
1055 	} else {
1056 		vdp->xdf_l_act->av_forw = dbp;
1057 		vdp->xdf_l_act = dbp;
1058 	}
1059 	dbp->av_forw = NULL;
1060 	dbp->av_back = NULL;
1061 	mutex_exit(&vdp->xdf_dev_lk);
1062 	xdf_iostart(vdp);
1063 	err = xdf_drain_io(vdp);
1064 	biofini(dbp);
1065 	return (err);
1066 }
1067 
1068 /*ARGSUSED*/
1069 static int
1070 xdf_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp,
1071     int *rvalp)
1072 {
1073 	int instance;
1074 	xdf_t	*vdp;
1075 	minor_t minor;
1076 	int part;
1077 
1078 	minor = getminor(dev);
1079 	instance = XDF_INST(minor);
1080 
1081 	if ((vdp = ddi_get_soft_state(vbd_ss, instance)) == NULL)
1082 		return (ENXIO);
1083 
1084 	DPRINTF(IOCTL_DBG, ("xdf%d:ioctl: cmd %d (0x%x)\n",
1085 	    instance, cmd, cmd));
1086 
1087 	part = XDF_PART(minor);
1088 	if (!xdf_isopen(vdp, part))
1089 		return (ENXIO);
1090 
1091 	switch (cmd) {
1092 	case DKIOCGMEDIAINFO: {
1093 		struct dk_minfo	media_info;
1094 
1095 		media_info.dki_lbsize = DEV_BSIZE;
1096 		media_info.dki_capacity = vdp->xdf_pgeom.g_capacity;
1097 		media_info.dki_media_type = DK_FIXED_DISK;
1098 
1099 		if (ddi_copyout(&media_info, (void *)arg,
1100 		    sizeof (struct dk_minfo), mode)) {
1101 			return (EFAULT);
1102 		} else {
1103 			return (0);
1104 		}
1105 	}
1106 
1107 	case DKIOCINFO: {
1108 		struct dk_cinfo info;
1109 
1110 		/* controller information */
1111 		if (XD_IS_CD(vdp))
1112 			info.dki_ctype = DKC_CDROM;
1113 		else
1114 			info.dki_ctype = DKC_VBD;
1115 
1116 		info.dki_cnum = 0;
1117 		(void) strncpy((char *)(&info.dki_cname), "xdf", 8);
1118 
1119 		/* unit information */
1120 		info.dki_unit = ddi_get_instance(vdp->xdf_dip);
1121 		(void) strncpy((char *)(&info.dki_dname), "xdf", 8);
1122 		info.dki_flags = DKI_FMTVOL;
1123 		info.dki_partition = part;
1124 		info.dki_maxtransfer = maxphys / DEV_BSIZE;
1125 		info.dki_addr = 0;
1126 		info.dki_space = 0;
1127 		info.dki_prio = 0;
1128 		info.dki_vec = 0;
1129 
1130 		if (ddi_copyout(&info, (void *)arg, sizeof (info), mode))
1131 			return (EFAULT);
1132 		else
1133 			return (0);
1134 	}
1135 
1136 	case DKIOCSTATE: {
1137 		enum dkio_state	dkstate = DKIO_INSERTED;
1138 		if (ddi_copyout(&dkstate, (void *)arg, sizeof (dkstate),
1139 		    mode) != 0)
1140 			return (EFAULT);
1141 		return (0);
1142 	}
1143 
1144 	/*
1145 	 * is media removable?
1146 	 */
1147 	case DKIOCREMOVABLE: {
1148 		int i = XD_IS_RM(vdp) ? 1 : 0;
1149 		if (ddi_copyout(&i, (caddr_t)arg, sizeof (int), mode))
1150 			return (EFAULT);
1151 		return (0);
1152 	}
1153 
1154 	case DKIOCG_PHYGEOM:
1155 	case DKIOCG_VIRTGEOM:
1156 	case DKIOCGGEOM:
1157 	case DKIOCSGEOM:
1158 	case DKIOCGAPART:
1159 	case DKIOCSAPART:
1160 	case DKIOCGVTOC:
1161 	case DKIOCSVTOC:
1162 	case DKIOCPARTINFO:
1163 	case DKIOCGMBOOT:
1164 	case DKIOCSMBOOT:
1165 	case DKIOCGETEFI:
1166 	case DKIOCSETEFI:
1167 	case DKIOCPARTITION: {
1168 		int rc;
1169 
1170 		rc = cmlb_ioctl(vdp->xdf_vd_lbl, dev, cmd, arg, mode, credp,
1171 		    rvalp, NULL);
1172 		return (rc);
1173 	}
1174 
1175 	case DKIOCGETWCE:
1176 		if (ddi_copyout(&vdp->xdf_wce, (void *)arg,
1177 		    sizeof (vdp->xdf_wce), mode))
1178 			return (EFAULT);
1179 		return (0);
1180 	case DKIOCSETWCE:
1181 		if (ddi_copyin((void *)arg, &vdp->xdf_wce,
1182 		    sizeof (vdp->xdf_wce), mode))
1183 			return (EFAULT);
1184 		return (0);
1185 	case DKIOCFLUSHWRITECACHE: {
1186 		int rc;
1187 		struct dk_callback *dkc = (struct dk_callback *)arg;
1188 
1189 		if (vdp->xdf_flush_supported) {
1190 			rc = xdf_lb_rdwr(vdp->xdf_dip, TG_WRITE,
1191 			    NULL, 0, 0, (void *)dev);
1192 		} else if (vdp->xdf_feature_barrier &&
1193 		    !xdf_barrier_flush_disable) {
1194 			rc = xdf_lb_rdwr(vdp->xdf_dip, TG_WRITE,
1195 			    vdp->xdf_cache_flush_block, xdf_flush_block,
1196 			    DEV_BSIZE, (void *)dev);
1197 		} else {
1198 			return (ENOTTY);
1199 		}
1200 		if ((mode & FKIOCTL) && (dkc != NULL) &&
1201 		    (dkc->dkc_callback != NULL)) {
1202 			(*dkc->dkc_callback)(dkc->dkc_cookie, rc);
1203 			/* need to return 0 after calling callback */
1204 			rc = 0;
1205 		}
1206 		return (rc);
1207 	}
1208 
1209 	default:
1210 		return (ENOTTY);
1211 	}
1212 }
1213 
1214 /*
1215  * xdf interrupt handler
1216  */
1217 static uint_t
1218 xdf_intr(caddr_t arg)
1219 {
1220 	xdf_t *vdp = (xdf_t *)arg;
1221 	xendev_ring_t *xbr;
1222 	blkif_response_t *resp;
1223 	int bioerr;
1224 	uint64_t id;
1225 	extern int do_polled_io;
1226 	uint8_t op;
1227 	uint16_t status;
1228 	ddi_acc_handle_t acchdl;
1229 
1230 	mutex_enter(&vdp->xdf_dev_lk);
1231 
1232 	if ((xbr = vdp->xdf_xb_ring) == NULL) {
1233 		mutex_exit(&vdp->xdf_dev_lk);
1234 		return (DDI_INTR_UNCLAIMED);
1235 	}
1236 
1237 	acchdl = vdp->xdf_xb_ring_hdl;
1238 
1239 	/*
1240 	 * complete all requests which have a response
1241 	 */
1242 	while (resp = xvdi_ring_get_response(xbr)) {
1243 		id = ddi_get64(acchdl, &resp->id);
1244 		op = ddi_get8(acchdl, &resp->operation);
1245 		status = ddi_get16(acchdl, (uint16_t *)&resp->status);
1246 		DPRINTF(INTR_DBG, ("resp: op %d id %"PRIu64" status %d\n",
1247 		    op, id, status));
1248 
1249 		/*
1250 		 * XXPV - close connection to the backend and restart
1251 		 */
1252 		if (status != BLKIF_RSP_OKAY) {
1253 			DPRINTF(IO_DBG, ("xdf@%s: I/O error while %s",
1254 			    ddi_get_name_addr(vdp->xdf_dip),
1255 			    (op == BLKIF_OP_READ) ? "reading" : "writing"));
1256 			bioerr = EIO;
1257 		} else {
1258 			bioerr = 0;
1259 		}
1260 
1261 		xdf_iofini(vdp, id, bioerr);
1262 	}
1263 
1264 	mutex_exit(&vdp->xdf_dev_lk);
1265 
1266 	if (!do_polled_io)
1267 		xdf_iostart(vdp);
1268 
1269 	return (DDI_INTR_CLAIMED);
1270 }
1271 
1272 int xdf_fbrewrites;	/* how many times was our flush block rewritten */
1273 
1274 /*
1275  * Snarf new data if our flush block was re-written
1276  */
1277 static void
1278 check_fbwrite(xdf_t *vdp, buf_t *bp, daddr_t blkno)
1279 {
1280 	int nblks;
1281 	boolean_t mapin;
1282 
1283 	if (IS_WRITE_BARRIER(vdp, bp))
1284 		return; /* write was a flush write */
1285 
1286 	mapin = B_FALSE;
1287 	nblks = bp->b_bcount >> DEV_BSHIFT;
1288 	if (xdf_flush_block >= blkno && xdf_flush_block < (blkno + nblks)) {
1289 		xdf_fbrewrites++;
1290 		if (bp->b_flags & (B_PAGEIO | B_PHYS)) {
1291 			mapin = B_TRUE;
1292 			bp_mapin(bp);
1293 		}
1294 		bcopy(bp->b_un.b_addr +
1295 		    ((xdf_flush_block - blkno) << DEV_BSHIFT),
1296 		    vdp->xdf_cache_flush_block, DEV_BSIZE);
1297 		if (mapin)
1298 			bp_mapout(bp);
1299 	}
1300 }
1301 
1302 static void
1303 xdf_iofini(xdf_t *vdp, uint64_t id, int bioerr)
1304 {
1305 	ge_slot_t *gs = (ge_slot_t *)(uintptr_t)id;
1306 	v_req_t *vreq = gs->vreq;
1307 	buf_t *bp = vreq->v_buf;
1308 
1309 	gs_free(vdp, gs);
1310 	if (bioerr)
1311 		bioerror(bp, bioerr);
1312 	vreq->v_nslots--;
1313 	if (vreq->v_nslots != 0)
1314 		return;
1315 
1316 	XDF_UPDATE_IO_STAT(vdp, bp);
1317 	if (vdp->xdf_xdev_iostat != NULL)
1318 		kstat_runq_exit(KSTAT_IO_PTR(vdp->xdf_xdev_iostat));
1319 
1320 	if (IS_ERROR(bp))
1321 		bp->b_resid = bp->b_bcount;
1322 
1323 	vreq_free(vdp, vreq);
1324 	biodone(bp);
1325 }
1326 
1327 /*
1328  * return value of xdf_prepare_rreq()
1329  * used in xdf_iostart()
1330  */
1331 #define	XF_PARTIAL	0 /* rreq is full, not all I/O in buf transferred */
1332 #define	XF_COMP		1 /* no more I/O left in buf */
1333 
1334 static void
1335 xdf_iostart(xdf_t *vdp)
1336 {
1337 	xendev_ring_t *xbr;
1338 	struct buf *bp;
1339 	blkif_request_t *rreq;
1340 	int retval;
1341 	int rreqready = 0;
1342 
1343 	xbr = vdp->xdf_xb_ring;
1344 
1345 	/*
1346 	 * populate the ring request(s)
1347 	 *
1348 	 * loop until there is no buf to transfer or no free slot
1349 	 * available in I/O ring
1350 	 */
1351 	mutex_enter(&vdp->xdf_dev_lk);
1352 
1353 	for (;;) {
1354 		if (vdp->xdf_status != XD_READY)
1355 			break;
1356 
1357 		/* active buf queue empty? */
1358 		if ((bp = vdp->xdf_f_act) == NULL)
1359 			break;
1360 
1361 		/* try to grab a vreq for this bp */
1362 		if ((BP2VREQ(bp) == NULL) && (vreq_get(vdp, bp) == NULL))
1363 				break;
1364 		/* alloc DMA/GTE resources */
1365 		if (vreq_setup(vdp, BP2VREQ(bp)) != DDI_SUCCESS)
1366 			break;
1367 
1368 		/* get next blkif_request in the ring */
1369 		if ((rreq = xvdi_ring_get_request(xbr)) == NULL)
1370 			break;
1371 		bzero(rreq, sizeof (blkif_request_t));
1372 
1373 		/* populate blkif_request with this buf */
1374 		rreqready++;
1375 		retval = xdf_prepare_rreq(vdp, bp, rreq);
1376 		if (retval == XF_COMP) {
1377 			/* finish this bp, switch to next one */
1378 			if (vdp->xdf_xdev_iostat != NULL)
1379 				kstat_waitq_to_runq(
1380 				    KSTAT_IO_PTR(vdp->xdf_xdev_iostat));
1381 			vdp->xdf_f_act = bp->av_forw;
1382 			bp->av_forw = NULL;
1383 		}
1384 	}
1385 
1386 	/*
1387 	 * Send the request(s) to the backend
1388 	 */
1389 	if (rreqready) {
1390 		if (xvdi_ring_push_request(xbr)) {
1391 			DPRINTF(IO_DBG, ("xdf_iostart: "
1392 			    "sent request(s) to backend\n"));
1393 			xvdi_notify_oe(vdp->xdf_dip);
1394 		}
1395 	}
1396 
1397 	mutex_exit(&vdp->xdf_dev_lk);
1398 }
1399 
1400 /*
1401  * populate a single blkif_request_t w/ a buf
1402  */
1403 static int
1404 xdf_prepare_rreq(xdf_t *vdp, struct buf *bp, blkif_request_t *rreq)
1405 {
1406 	int		rval;
1407 	grant_ref_t	gr;
1408 	uint8_t		fsect, lsect;
1409 	size_t		bcnt;
1410 	paddr_t		dma_addr;
1411 	off_t		blk_off;
1412 	dev_info_t	*dip = vdp->xdf_dip;
1413 	blkif_vdev_t	vdev = xvdi_get_vdevnum(dip);
1414 	v_req_t		*vreq = BP2VREQ(bp);
1415 	uint64_t	blkno = vreq->v_blkno;
1416 	uint_t		ndmacs = vreq->v_ndmacs;
1417 	ddi_acc_handle_t acchdl = vdp->xdf_xb_ring_hdl;
1418 	int		seg = 0;
1419 	int		isread = IS_READ(bp);
1420 
1421 	if (isread)
1422 		ddi_put8(acchdl, &rreq->operation, BLKIF_OP_READ);
1423 	else {
1424 		switch (vreq->v_flush_diskcache) {
1425 		case FLUSH_DISKCACHE:
1426 			ddi_put8(acchdl, &rreq->operation,
1427 			    BLKIF_OP_FLUSH_DISKCACHE);
1428 			ddi_put16(acchdl, &rreq->handle, vdev);
1429 			ddi_put64(acchdl, &rreq->id,
1430 			    (uint64_t)(uintptr_t)(vreq->v_gs));
1431 			ddi_put8(acchdl, &rreq->nr_segments, 0);
1432 			return (XF_COMP);
1433 		case WRITE_BARRIER:
1434 			ddi_put8(acchdl, &rreq->operation,
1435 			    BLKIF_OP_WRITE_BARRIER);
1436 			break;
1437 		default:
1438 			if (!vdp->xdf_wce)
1439 				ddi_put8(acchdl, &rreq->operation,
1440 				    BLKIF_OP_WRITE_BARRIER);
1441 			else
1442 				ddi_put8(acchdl, &rreq->operation,
1443 				    BLKIF_OP_WRITE);
1444 			break;
1445 		}
1446 	}
1447 
1448 	ddi_put16(acchdl, &rreq->handle, vdev);
1449 	ddi_put64(acchdl, &rreq->sector_number, blkno);
1450 	ddi_put64(acchdl, &rreq->id, (uint64_t)(uintptr_t)(vreq->v_gs));
1451 
1452 	/*
1453 	 * loop until all segments are populated or no more dma cookie in buf
1454 	 */
1455 	for (;;) {
1456 	/*
1457 	 * Each segment of a blkif request can transfer up to
1458 	 * one 4K page of data.
1459 	 */
1460 		bcnt = vreq->v_dmac.dmac_size;
1461 		ASSERT(bcnt <= PAGESIZE);
1462 		ASSERT((bcnt % XB_BSIZE) == 0);
1463 		dma_addr = vreq->v_dmac.dmac_laddress;
1464 		blk_off = (uint_t)((paddr_t)XB_SEGOFFSET & dma_addr);
1465 		ASSERT((blk_off & XB_BMASK) == 0);
1466 		fsect = blk_off >> XB_BSHIFT;
1467 		lsect = fsect + (bcnt >> XB_BSHIFT) - 1;
1468 		ASSERT(fsect < XB_MAX_SEGLEN / XB_BSIZE &&
1469 		    lsect < XB_MAX_SEGLEN / XB_BSIZE);
1470 		DPRINTF(IO_DBG, ("  ""seg%d: dmacS %lu blk_off %ld\n",
1471 		    seg, vreq->v_dmac.dmac_size, blk_off));
1472 		gr = gs_grant(vreq->v_gs, PATOMA(dma_addr) >> PAGESHIFT);
1473 		ddi_put32(acchdl, &rreq->seg[seg].gref, gr);
1474 		ddi_put8(acchdl, &rreq->seg[seg].first_sect, fsect);
1475 		ddi_put8(acchdl, &rreq->seg[seg].last_sect, lsect);
1476 		DPRINTF(IO_DBG, ("  ""seg%d: fs %d ls %d gr %d dma 0x%"PRIx64
1477 		    "\n", seg, fsect, lsect, gr, dma_addr));
1478 
1479 		blkno += (bcnt >> XB_BSHIFT);
1480 		seg++;
1481 		ASSERT(seg <= BLKIF_MAX_SEGMENTS_PER_REQUEST);
1482 		if (--ndmacs) {
1483 			ddi_dma_nextcookie(vreq->v_dmahdl, &vreq->v_dmac);
1484 			continue;
1485 		}
1486 
1487 		vreq->v_status = VREQ_DMAWIN_DONE;
1488 		vreq->v_blkno = blkno;
1489 		if (vreq->v_dmaw + 1 == vreq->v_ndmaws)
1490 			/* last win */
1491 			rval = XF_COMP;
1492 		else
1493 			rval = XF_PARTIAL;
1494 		break;
1495 	}
1496 	ddi_put8(acchdl,  &rreq->nr_segments, seg);
1497 	DPRINTF(IO_DBG, ("xdf_prepare_rreq: request id=%"PRIx64" ready\n",
1498 	    rreq->id));
1499 
1500 	return (rval);
1501 }
1502 
1503 #define	XDF_QSEC	50000	/* .005 second */
1504 #define	XDF_POLLCNT	12	/* loop for 12 times before time out */
1505 
1506 static int
1507 xdf_drain_io(xdf_t *vdp)
1508 {
1509 	int pollc, rval;
1510 	xendev_ring_t *xbr;
1511 
1512 	if (xdfdebug & SUSRES_DBG)
1513 		xen_printf("xdf_drain_io: start\n");
1514 
1515 	mutex_enter(&vdp->xdf_dev_lk);
1516 
1517 	if ((vdp->xdf_status != XD_READY) && (vdp->xdf_status != XD_SUSPEND))
1518 		goto out;
1519 
1520 	rval = 0;
1521 	xbr = vdp->xdf_xb_ring;
1522 	ASSERT(xbr != NULL);
1523 
1524 	for (pollc = 0; pollc < XDF_POLLCNT; pollc++) {
1525 		if (xvdi_ring_has_unconsumed_responses(xbr)) {
1526 			mutex_exit(&vdp->xdf_dev_lk);
1527 			(void) xdf_intr((caddr_t)vdp);
1528 			mutex_enter(&vdp->xdf_dev_lk);
1529 		}
1530 		if (!xvdi_ring_has_incomp_request(xbr))
1531 			goto out;
1532 
1533 #ifndef	XPV_HVM_DRIVER
1534 		(void) HYPERVISOR_yield();
1535 #endif /* XPV_HVM_DRIVER */
1536 		/*
1537 		 * file-backed devices can be slow
1538 		 */
1539 		drv_usecwait(XDF_QSEC << pollc);
1540 	}
1541 	cmn_err(CE_WARN, "xdf_polled_io: timeout");
1542 	rval = EIO;
1543 out:
1544 	mutex_exit(&vdp->xdf_dev_lk);
1545 	if (xdfdebug & SUSRES_DBG)
1546 		xen_printf("xdf_drain_io: end, err=%d\n", rval);
1547 	return (rval);
1548 }
1549 
1550 /* ARGSUSED5 */
1551 int
1552 xdf_lb_rdwr(dev_info_t *devi, uchar_t cmd, void *bufp,
1553     diskaddr_t start, size_t reqlen, void *tg_cookie)
1554 {
1555 	xdf_t *vdp;
1556 	struct buf *bp;
1557 	int err = 0;
1558 
1559 	vdp = ddi_get_soft_state(vbd_ss, ddi_get_instance(devi));
1560 	if (vdp == NULL)
1561 		return (ENXIO);
1562 
1563 	if ((start + (reqlen >> DEV_BSHIFT)) > vdp->xdf_pgeom.g_capacity)
1564 		return (EINVAL);
1565 
1566 	bp = getrbuf(KM_SLEEP);
1567 	if (cmd == TG_READ)
1568 		bp->b_flags = B_BUSY | B_READ;
1569 	else
1570 		bp->b_flags = B_BUSY | B_WRITE;
1571 	bp->b_un.b_addr = bufp;
1572 	bp->b_bcount = reqlen;
1573 	bp->b_blkno = start;
1574 	bp->b_edev = DDI_DEV_T_NONE; /* don't have dev_t */
1575 
1576 	mutex_enter(&vdp->xdf_dev_lk);
1577 	if (vdp->xdf_xdev_iostat != NULL)
1578 		kstat_waitq_enter(KSTAT_IO_PTR(vdp->xdf_xdev_iostat));
1579 	if (vdp->xdf_f_act == NULL) {
1580 		vdp->xdf_f_act = vdp->xdf_l_act = bp;
1581 	} else {
1582 		vdp->xdf_l_act->av_forw = bp;
1583 		vdp->xdf_l_act = bp;
1584 	}
1585 	mutex_exit(&vdp->xdf_dev_lk);
1586 	xdf_iostart(vdp);
1587 	err = biowait(bp);
1588 
1589 	ASSERT(bp->b_flags & B_DONE);
1590 
1591 	freerbuf(bp);
1592 	return (err);
1593 }
1594 
1595 /*
1596  * synthetic geometry
1597  */
1598 #define	XDF_NSECTS	256
1599 #define	XDF_NHEADS	16
1600 
1601 static void
1602 xdf_synthetic_pgeom(dev_info_t *devi, cmlb_geom_t *geomp)
1603 {
1604 	xdf_t *vdp;
1605 	uint_t ncyl;
1606 
1607 	vdp = ddi_get_soft_state(vbd_ss, ddi_get_instance(devi));
1608 
1609 	ncyl = vdp->xdf_xdev_nblocks / (XDF_NHEADS * XDF_NSECTS);
1610 
1611 	geomp->g_ncyl = ncyl == 0 ? 1 : ncyl;
1612 	geomp->g_acyl = 0;
1613 	geomp->g_nhead = XDF_NHEADS;
1614 	geomp->g_secsize = XB_BSIZE;
1615 	geomp->g_nsect = XDF_NSECTS;
1616 	geomp->g_intrlv = 0;
1617 	geomp->g_rpm = 7200;
1618 	geomp->g_capacity = vdp->xdf_xdev_nblocks;
1619 }
1620 
1621 static int
1622 xdf_lb_getcap(dev_info_t *devi, diskaddr_t *capp)
1623 {
1624 	xdf_t *vdp;
1625 
1626 	vdp = ddi_get_soft_state(vbd_ss, ddi_get_instance(devi));
1627 
1628 	if (vdp == NULL)
1629 		return (ENXIO);
1630 
1631 	mutex_enter(&vdp->xdf_dev_lk);
1632 	*capp = vdp->xdf_pgeom.g_capacity;
1633 	DPRINTF(LBL_DBG, ("capacity %llu\n", *capp));
1634 	mutex_exit(&vdp->xdf_dev_lk);
1635 	return (0);
1636 }
1637 
1638 static int
1639 xdf_lb_getpgeom(dev_info_t *devi, cmlb_geom_t *geomp)
1640 {
1641 	xdf_t *vdp;
1642 
1643 	if ((vdp = ddi_get_soft_state(vbd_ss, ddi_get_instance(devi))) == NULL)
1644 		return (ENXIO);
1645 	*geomp = vdp->xdf_pgeom;
1646 	return (0);
1647 }
1648 
1649 /*
1650  * No real HBA, no geometry available from it
1651  */
1652 /*ARGSUSED*/
1653 static int
1654 xdf_lb_getvgeom(dev_info_t *devi, cmlb_geom_t *geomp)
1655 {
1656 	return (EINVAL);
1657 }
1658 
1659 static int
1660 xdf_lb_getattribute(dev_info_t *devi, tg_attribute_t *tgattributep)
1661 {
1662 	xdf_t *vdp;
1663 
1664 	if (!(vdp = ddi_get_soft_state(vbd_ss, ddi_get_instance(devi))))
1665 		return (ENXIO);
1666 
1667 	if (XD_IS_RO(vdp))
1668 		tgattributep->media_is_writable = 0;
1669 	else
1670 		tgattributep->media_is_writable = 1;
1671 	return (0);
1672 }
1673 
1674 /* ARGSUSED3 */
1675 int
1676 xdf_lb_getinfo(dev_info_t *devi, int cmd, void *arg, void *tg_cookie)
1677 {
1678 	switch (cmd) {
1679 	case TG_GETPHYGEOM:
1680 		return (xdf_lb_getpgeom(devi, (cmlb_geom_t *)arg));
1681 	case TG_GETVIRTGEOM:
1682 		return (xdf_lb_getvgeom(devi, (cmlb_geom_t *)arg));
1683 	case TG_GETCAPACITY:
1684 		return (xdf_lb_getcap(devi, (diskaddr_t *)arg));
1685 	case TG_GETBLOCKSIZE:
1686 		*(uint32_t *)arg = XB_BSIZE;
1687 		return (0);
1688 	case TG_GETATTR:
1689 		return (xdf_lb_getattribute(devi, (tg_attribute_t *)arg));
1690 	default:
1691 		return (ENOTTY);
1692 	}
1693 }
1694 
1695 /*
1696  * Kick-off connect process
1697  * Status should be XD_UNKNOWN or XD_CLOSED
1698  * On success, status will be changed to XD_INIT
1699  * On error, status won't be changed
1700  */
1701 static int
1702 xdf_start_connect(xdf_t *vdp)
1703 {
1704 	char *xsnode;
1705 	grant_ref_t gref;
1706 	xenbus_transaction_t xbt;
1707 	int rv;
1708 	dev_info_t *dip = vdp->xdf_dip;
1709 
1710 	if ((vdp->xdf_peer = xvdi_get_oeid(dip)) == (domid_t)-1)
1711 		goto errout;
1712 
1713 	if (xvdi_alloc_evtchn(dip) != DDI_SUCCESS) {
1714 		cmn_err(CE_WARN, "xdf@%s: failed to alloc event channel",
1715 		    ddi_get_name_addr(dip));
1716 		goto errout;
1717 	}
1718 	vdp->xdf_evtchn = xvdi_get_evtchn(dip);
1719 #ifdef XPV_HVM_DRIVER
1720 	ec_bind_evtchn_to_handler(vdp->xdf_evtchn, IPL_VBD, xdf_intr, vdp);
1721 #else /* !XPV_HVM_DRIVER */
1722 	if (ddi_add_intr(dip, 0, NULL, NULL, xdf_intr, (caddr_t)vdp) !=
1723 	    DDI_SUCCESS) {
1724 		cmn_err(CE_WARN, "xdf_start_connect: xdf@%s: "
1725 		    "failed to add intr handler", ddi_get_name_addr(dip));
1726 		goto errout1;
1727 	}
1728 #endif /* !XPV_HVM_DRIVER */
1729 
1730 	if (xvdi_alloc_ring(dip, BLKIF_RING_SIZE,
1731 	    sizeof (union blkif_sring_entry), &gref, &vdp->xdf_xb_ring) !=
1732 	    DDI_SUCCESS) {
1733 		cmn_err(CE_WARN, "xdf@%s: failed to alloc comm ring",
1734 		    ddi_get_name_addr(dip));
1735 		goto errout2;
1736 	}
1737 	vdp->xdf_xb_ring_hdl = vdp->xdf_xb_ring->xr_acc_hdl; /* ugly!! */
1738 
1739 	/*
1740 	 * Write into xenstore the info needed by backend
1741 	 */
1742 	if ((xsnode = xvdi_get_xsname(dip)) == NULL) {
1743 		cmn_err(CE_WARN, "xdf@%s: "
1744 		    "failed to get xenstore node path",
1745 		    ddi_get_name_addr(dip));
1746 		goto fail_trans;
1747 	}
1748 trans_retry:
1749 	if (xenbus_transaction_start(&xbt)) {
1750 		cmn_err(CE_WARN, "xdf@%s: failed to start transaction",
1751 		    ddi_get_name_addr(dip));
1752 		xvdi_fatal_error(dip, EIO, "transaction start");
1753 		goto fail_trans;
1754 	}
1755 
1756 	if (rv = xenbus_printf(xbt, xsnode, "ring-ref", "%u", gref)) {
1757 		cmn_err(CE_WARN, "xdf@%s: failed to write ring-ref",
1758 		    ddi_get_name_addr(dip));
1759 		xvdi_fatal_error(dip, rv, "writing ring-ref");
1760 		goto abort_trans;
1761 	}
1762 
1763 	if (rv = xenbus_printf(xbt, xsnode, "event-channel", "%u",
1764 	    vdp->xdf_evtchn)) {
1765 		cmn_err(CE_WARN, "xdf@%s: failed to write event-channel",
1766 		    ddi_get_name_addr(dip));
1767 		xvdi_fatal_error(dip, rv, "writing event-channel");
1768 		goto abort_trans;
1769 	}
1770 
1771 	/*
1772 	 * "protocol" is written by the domain builder in the case of PV
1773 	 * domains. However, it is not written for HVM domains, so let's
1774 	 * write it here.
1775 	 */
1776 	if (rv = xenbus_printf(xbt, xsnode, "protocol", "%s",
1777 	    XEN_IO_PROTO_ABI_NATIVE)) {
1778 		cmn_err(CE_WARN, "xdf@%s: failed to write protocol",
1779 		    ddi_get_name_addr(dip));
1780 		xvdi_fatal_error(dip, rv, "writing protocol");
1781 		goto abort_trans;
1782 	}
1783 
1784 	if ((rv = xvdi_switch_state(dip, xbt, XenbusStateInitialised)) > 0) {
1785 		cmn_err(CE_WARN, "xdf@%s: "
1786 		    "failed to switch state to XenbusStateInitialised",
1787 		    ddi_get_name_addr(dip));
1788 		xvdi_fatal_error(dip, rv, "writing state");
1789 		goto abort_trans;
1790 	}
1791 
1792 	/* kick-off connect process */
1793 	if (rv = xenbus_transaction_end(xbt, 0)) {
1794 		if (rv == EAGAIN)
1795 			goto trans_retry;
1796 		cmn_err(CE_WARN, "xdf@%s: failed to end transaction",
1797 		    ddi_get_name_addr(dip));
1798 		xvdi_fatal_error(dip, rv, "completing transaction");
1799 		goto fail_trans;
1800 	}
1801 
1802 	ASSERT(mutex_owned(&vdp->xdf_cb_lk));
1803 	mutex_enter(&vdp->xdf_dev_lk);
1804 	vdp->xdf_status = XD_INIT;
1805 	mutex_exit(&vdp->xdf_dev_lk);
1806 
1807 	return (DDI_SUCCESS);
1808 
1809 abort_trans:
1810 	(void) xenbus_transaction_end(xbt, 1);
1811 fail_trans:
1812 	xvdi_free_ring(vdp->xdf_xb_ring);
1813 errout2:
1814 #ifdef XPV_HVM_DRIVER
1815 	ec_unbind_evtchn(vdp->xdf_evtchn);
1816 #else /* !XPV_HVM_DRIVER */
1817 	(void) ddi_remove_intr(vdp->xdf_dip, 0, NULL);
1818 #endif /* !XPV_HVM_DRIVER */
1819 errout1:
1820 	xvdi_free_evtchn(dip);
1821 errout:
1822 	cmn_err(CE_WARN, "xdf@%s: fail to kick-off connecting",
1823 	    ddi_get_name_addr(dip));
1824 	return (DDI_FAILURE);
1825 }
1826 
1827 /*
1828  * Kick-off disconnect process
1829  * Status won't be changed
1830  */
1831 static int
1832 xdf_start_disconnect(xdf_t *vdp)
1833 {
1834 	if (xvdi_switch_state(vdp->xdf_dip, XBT_NULL, XenbusStateClosed) > 0) {
1835 		cmn_err(CE_WARN, "xdf@%s: fail to kick-off disconnecting",
1836 		    ddi_get_name_addr(vdp->xdf_dip));
1837 		return (DDI_FAILURE);
1838 	}
1839 
1840 	return (DDI_SUCCESS);
1841 }
1842 
1843 int
1844 xdf_get_flush_block(xdf_t *vdp)
1845 {
1846 	/*
1847 	 * Get a DEV_BSIZE aligned bufer
1848 	 */
1849 	vdp->xdf_flush_mem = kmem_alloc(DEV_BSIZE * 2, KM_SLEEP);
1850 	vdp->xdf_cache_flush_block =
1851 	    (char *)P2ROUNDUP((uintptr_t)(vdp->xdf_flush_mem), DEV_BSIZE);
1852 	if (xdf_lb_rdwr(vdp->xdf_dip, TG_READ, vdp->xdf_cache_flush_block,
1853 	    xdf_flush_block, DEV_BSIZE, NULL) != 0)
1854 		return (DDI_FAILURE);
1855 	return (DDI_SUCCESS);
1856 }
1857 
1858 /*
1859  * Finish other initialization after we've connected to backend
1860  * Status should be XD_INIT before calling this routine
1861  * On success, status should be changed to XD_READY
1862  * On error, status should stay XD_INIT
1863  */
1864 static int
1865 xdf_post_connect(xdf_t *vdp)
1866 {
1867 	int rv;
1868 	uint_t len;
1869 	char *type;
1870 	char *barrier;
1871 	dev_info_t *devi = vdp->xdf_dip;
1872 
1873 	/*
1874 	 * Determine if feature barrier is supported by backend
1875 	 */
1876 	if (xenbus_read(XBT_NULL, xvdi_get_oename(devi),
1877 	    "feature-barrier", (void **)&barrier, &len) == 0) {
1878 		vdp->xdf_feature_barrier = 1;
1879 		kmem_free(barrier, len);
1880 	} else {
1881 		cmn_err(CE_NOTE, "xdf@%s: failed to read feature-barrier",
1882 		    ddi_get_name_addr(vdp->xdf_dip));
1883 		vdp->xdf_feature_barrier = 0;
1884 	}
1885 
1886 	/* probe backend */
1887 	if (rv = xenbus_gather(XBT_NULL, xvdi_get_oename(devi),
1888 	    "sectors", "%"SCNu64, &vdp->xdf_xdev_nblocks,
1889 	    "info", "%u", &vdp->xdf_xdev_info, NULL)) {
1890 		cmn_err(CE_WARN, "xdf_post_connect: xdf@%s: "
1891 		    "cannot read backend info", ddi_get_name_addr(devi));
1892 		xvdi_fatal_error(devi, rv, "reading backend info");
1893 		return (DDI_FAILURE);
1894 	}
1895 
1896 	/*
1897 	 * Make sure that the device we're connecting isn't smaller than
1898 	 * the old connected device.
1899 	 */
1900 	if (vdp->xdf_xdev_nblocks < vdp->xdf_pgeom.g_capacity) {
1901 		cmn_err(CE_WARN, "xdf_post_connect: xdf@%s: "
1902 		    "backend disk device shrank", ddi_get_name_addr(devi));
1903 		/* XXX:  call xvdi_fatal_error() here? */
1904 		xvdi_fatal_error(devi, rv, "reading backend info");
1905 		return (DDI_FAILURE);
1906 	}
1907 
1908 	/*
1909 	 * Only update the physical geometry to reflect the new device
1910 	 * size if this is the first time we're connecting to the backend
1911 	 * device.  Once we assign a physical geometry to a device it stays
1912 	 * fixed until:
1913 	 *	- we get detach and re-attached (at which point we
1914 	 *	  automatically assign a new physical geometry).
1915 	 *	- someone calls TG_SETPHYGEOM to explicity set the
1916 	 *	  physical geometry.
1917 	 */
1918 	if (vdp->xdf_pgeom.g_capacity == 0)
1919 		xdf_synthetic_pgeom(devi, &vdp->xdf_pgeom);
1920 
1921 	/* fix disk type */
1922 	if (xenbus_read(XBT_NULL, xvdi_get_xsname(devi), "device-type",
1923 	    (void **)&type, &len) != 0) {
1924 		cmn_err(CE_WARN, "xdf_post_connect: xdf@%s: "
1925 		    "cannot read device-type", ddi_get_name_addr(devi));
1926 		xvdi_fatal_error(devi, rv, "reading device-type");
1927 		return (DDI_FAILURE);
1928 	}
1929 	if (strcmp(type, "cdrom") == 0)
1930 		vdp->xdf_xdev_info |= VDISK_CDROM;
1931 	kmem_free(type, len);
1932 
1933 	/*
1934 	 * We've created all the minor nodes via cmlb_attach() using default
1935 	 * value in xdf_attach() to make it possible to block in xdf_open(),
1936 	 * in case there's anyone (say, booting thread) ever trying to open
1937 	 * it before connected to backend. We will refresh all those minor
1938 	 * nodes w/ latest info we've got now when we are almost connected.
1939 	 *
1940 	 * Don't do this when xdf is already opened by someone (could happen
1941 	 * during resume), for that cmlb_attach() will invalid the label info
1942 	 * and confuse those who has already opened the node, which is bad.
1943 	 */
1944 	if (!xdf_isopen(vdp, -1) && (XD_IS_CD(vdp) || XD_IS_RM(vdp))) {
1945 		/* re-init cmlb w/ latest info we got from backend */
1946 		if (cmlb_attach(devi, &xdf_lb_ops,
1947 		    XD_IS_CD(vdp) ? DTYPE_RODIRECT : DTYPE_DIRECT,
1948 		    XD_IS_RM(vdp), 1, DDI_NT_BLOCK,
1949 #if defined(XPV_HVM_DRIVER)
1950 		    CMLB_CREATE_ALTSLICE_VTOC_16_DTYPE_DIRECT |
1951 		    CMLB_INTERNAL_MINOR_NODES,
1952 #else /* !XPV_HVM_DRIVER */
1953 		    CMLB_FAKE_LABEL_ONE_PARTITION,
1954 #endif /* !XPV_HVM_DRIVER */
1955 		    vdp->xdf_vd_lbl, NULL) != 0) {
1956 			cmn_err(CE_WARN, "xdf@%s: cmlb attach failed",
1957 			    ddi_get_name_addr(devi));
1958 			return (DDI_FAILURE);
1959 		}
1960 	}
1961 
1962 	/* mark vbd is ready for I/O */
1963 	ASSERT(mutex_owned(&vdp->xdf_cb_lk));
1964 	mutex_enter(&vdp->xdf_dev_lk);
1965 	vdp->xdf_status = XD_READY;
1966 	mutex_exit(&vdp->xdf_dev_lk);
1967 	/*
1968 	 * If backend has feature-barrier, see if it supports disk
1969 	 * cache flush op.
1970 	 */
1971 	vdp->xdf_flush_supported = 0;
1972 	if (vdp->xdf_feature_barrier) {
1973 		/*
1974 		 * Pretend we already know flush is supported so probe
1975 		 * will attempt the correct op.
1976 		 */
1977 		vdp->xdf_flush_supported = 1;
1978 		if (xdf_lb_rdwr(vdp->xdf_dip, TG_WRITE, NULL, 0, 0, 0) == 0) {
1979 			vdp->xdf_flush_supported = 1;
1980 		} else {
1981 			vdp->xdf_flush_supported = 0;
1982 			/*
1983 			 * If the other end does not support the cache flush op
1984 			 * then we must use a barrier-write to force disk
1985 			 * cache flushing.  Barrier writes require that a data
1986 			 * block actually be written.
1987 			 * Cache a block to barrier-write when we are
1988 			 * asked to perform a flush.
1989 			 * XXX - would it be better to just copy 1 block
1990 			 * (512 bytes) from whatever write we did last
1991 			 * and rewrite that block?
1992 			 */
1993 			if (xdf_get_flush_block(vdp) != DDI_SUCCESS)
1994 				return (DDI_FAILURE);
1995 		}
1996 	}
1997 
1998 	cmn_err(CE_CONT, "?xdf@%s: %"PRIu64" blocks", ddi_get_name_addr(devi),
1999 	    (uint64_t)vdp->xdf_xdev_nblocks);
2000 
2001 	return (DDI_SUCCESS);
2002 }
2003 
2004 /*
2005  * Finish other uninitialization after we've disconnected from backend
2006  * when status is XD_CLOSING or XD_INIT. After returns, status is XD_CLOSED
2007  */
2008 static void
2009 xdf_post_disconnect(xdf_t *vdp)
2010 {
2011 #ifdef XPV_HVM_DRIVER
2012 	ec_unbind_evtchn(vdp->xdf_evtchn);
2013 #else /* !XPV_HVM_DRIVER */
2014 	(void) ddi_remove_intr(vdp->xdf_dip, 0, NULL);
2015 #endif /* !XPV_HVM_DRIVER */
2016 	xvdi_free_evtchn(vdp->xdf_dip);
2017 	xvdi_free_ring(vdp->xdf_xb_ring);
2018 	vdp->xdf_xb_ring = NULL;
2019 	vdp->xdf_xb_ring_hdl = NULL;
2020 	vdp->xdf_peer = (domid_t)-1;
2021 
2022 	ASSERT(mutex_owned(&vdp->xdf_cb_lk));
2023 	mutex_enter(&vdp->xdf_dev_lk);
2024 	vdp->xdf_status = XD_CLOSED;
2025 	mutex_exit(&vdp->xdf_dev_lk);
2026 }
2027 
2028 /*ARGSUSED*/
2029 static void
2030 xdf_oe_change(dev_info_t *dip, ddi_eventcookie_t id, void *arg, void *impl_data)
2031 {
2032 	XenbusState new_state = *(XenbusState *)impl_data;
2033 	xdf_t *vdp = (xdf_t *)ddi_get_driver_private(dip);
2034 	boolean_t unexpect_die = B_FALSE;
2035 	int status;
2036 
2037 	DPRINTF(DDI_DBG, ("xdf@%s: otherend state change to %d!\n",
2038 	    ddi_get_name_addr(dip), new_state));
2039 
2040 	mutex_enter(&vdp->xdf_cb_lk);
2041 
2042 	if (xdf_check_state_transition(vdp, new_state) == DDI_FAILURE) {
2043 		mutex_exit(&vdp->xdf_cb_lk);
2044 		return;
2045 	}
2046 
2047 	switch (new_state) {
2048 	case XenbusStateInitialising:
2049 		ASSERT(vdp->xdf_status == XD_CLOSED);
2050 		/*
2051 		 * backend recovered from a previous failure,
2052 		 * kick-off connect process again
2053 		 */
2054 		if (xdf_start_connect(vdp) != DDI_SUCCESS) {
2055 			cmn_err(CE_WARN, "xdf@%s:"
2056 			    " failed to start reconnecting to backend",
2057 			    ddi_get_name_addr(dip));
2058 		}
2059 		break;
2060 	case XenbusStateConnected:
2061 		ASSERT(vdp->xdf_status == XD_INIT);
2062 		(void) xvdi_switch_state(dip, XBT_NULL, XenbusStateConnected);
2063 		/* finish final init after connect */
2064 		if (xdf_post_connect(vdp) != DDI_SUCCESS)
2065 			(void) xdf_start_disconnect(vdp);
2066 		break;
2067 	case XenbusStateClosing:
2068 		if (vdp->xdf_status == XD_READY) {
2069 			mutex_enter(&vdp->xdf_dev_lk);
2070 			if (xdf_isopen(vdp, -1)) {
2071 				cmn_err(CE_NOTE, "xdf@%s: hot-unplug failed, "
2072 				    "still in use", ddi_get_name_addr(dip));
2073 				mutex_exit(&vdp->xdf_dev_lk);
2074 				break;
2075 			} else {
2076 				vdp->xdf_status = XD_CLOSING;
2077 			}
2078 			mutex_exit(&vdp->xdf_dev_lk);
2079 		}
2080 		(void) xdf_start_disconnect(vdp);
2081 		break;
2082 	case XenbusStateClosed:
2083 		/* first check if BE closed unexpectedly */
2084 		mutex_enter(&vdp->xdf_dev_lk);
2085 		if (xdf_isopen(vdp, -1)) {
2086 			unexpect_die = B_TRUE;
2087 			unexpectedie(vdp);
2088 			cmn_err(CE_WARN, "xdf@%s: backend closed, "
2089 			    "reconnecting...", ddi_get_name_addr(dip));
2090 		}
2091 		mutex_exit(&vdp->xdf_dev_lk);
2092 
2093 		if (vdp->xdf_status == XD_READY) {
2094 			mutex_enter(&vdp->xdf_dev_lk);
2095 			vdp->xdf_status = XD_CLOSING;
2096 			mutex_exit(&vdp->xdf_dev_lk);
2097 
2098 #ifdef	DOMU_BACKEND
2099 			(void) xvdi_post_event(dip, XEN_HP_REMOVE);
2100 #endif
2101 
2102 			xdf_post_disconnect(vdp);
2103 			(void) xvdi_switch_state(dip, XBT_NULL,
2104 			    XenbusStateClosed);
2105 		} else if ((vdp->xdf_status == XD_INIT) ||
2106 		    (vdp->xdf_status == XD_CLOSING)) {
2107 			xdf_post_disconnect(vdp);
2108 		} else {
2109 			mutex_enter(&vdp->xdf_dev_lk);
2110 			vdp->xdf_status = XD_CLOSED;
2111 			mutex_exit(&vdp->xdf_dev_lk);
2112 		}
2113 	}
2114 
2115 	/* notify anybody waiting for oe state change */
2116 	mutex_enter(&vdp->xdf_dev_lk);
2117 	cv_broadcast(&vdp->xdf_dev_cv);
2118 	mutex_exit(&vdp->xdf_dev_lk);
2119 
2120 	status = vdp->xdf_status;
2121 	mutex_exit(&vdp->xdf_cb_lk);
2122 
2123 	if (status == XD_READY) {
2124 		xdf_iostart(vdp);
2125 	} else if ((status == XD_CLOSED) && !unexpect_die) {
2126 		/* interface is closed successfully, remove all minor nodes */
2127 		cmlb_detach(vdp->xdf_vd_lbl, NULL);
2128 		cmlb_free_handle(&vdp->xdf_vd_lbl);
2129 	}
2130 }
2131 
2132 /* check if partition is open, -1 - check all partitions on the disk */
2133 static boolean_t
2134 xdf_isopen(xdf_t *vdp, int partition)
2135 {
2136 	int i;
2137 	ulong_t parbit;
2138 	boolean_t rval = B_FALSE;
2139 
2140 	ASSERT((partition == -1) ||
2141 	    ((partition >= 0) || (partition < XDF_PEXT)));
2142 
2143 	if (partition == -1)
2144 		parbit = (ulong_t)-1;
2145 	else
2146 		parbit = 1 << partition;
2147 
2148 	for (i = 0; i < OTYPCNT; i++) {
2149 		if (vdp->xdf_vd_open[i] & parbit)
2150 			rval = B_TRUE;
2151 	}
2152 
2153 	return (rval);
2154 }
2155 
2156 /*
2157  * Xdf_check_state_transition will check the XenbusState change to see
2158  * if the change is a valid transition or not.
2159  * The new state is written by backend domain, or by running xenstore-write
2160  * to change it manually in dom0
2161  */
2162 static int
2163 xdf_check_state_transition(xdf_t *vdp, XenbusState oestate)
2164 {
2165 	int status;
2166 	int stcheck;
2167 #define	STOK	0 /* need further process */
2168 #define	STNOP	1 /* no action need taking */
2169 #define	STBUG	2 /* unexpected state change, could be a bug */
2170 
2171 	status = vdp->xdf_status;
2172 	stcheck = STOK;
2173 
2174 	switch (status) {
2175 	case XD_UNKNOWN:
2176 		if ((oestate == XenbusStateUnknown)		||
2177 		    (oestate == XenbusStateConnected))
2178 			stcheck = STBUG;
2179 		else if ((oestate == XenbusStateInitialising)	||
2180 		    (oestate == XenbusStateInitWait)		||
2181 		    (oestate == XenbusStateInitialised))
2182 			stcheck = STNOP;
2183 		break;
2184 	case XD_INIT:
2185 		if (oestate == XenbusStateUnknown)
2186 			stcheck = STBUG;
2187 		else if ((oestate == XenbusStateInitialising)	||
2188 		    (oestate == XenbusStateInitWait)		||
2189 		    (oestate == XenbusStateInitialised))
2190 			stcheck = STNOP;
2191 		break;
2192 	case XD_READY:
2193 		if ((oestate == XenbusStateUnknown)		||
2194 		    (oestate == XenbusStateInitialising)	||
2195 		    (oestate == XenbusStateInitWait)		||
2196 		    (oestate == XenbusStateInitialised))
2197 			stcheck = STBUG;
2198 		else if (oestate == XenbusStateConnected)
2199 			stcheck = STNOP;
2200 		break;
2201 	case XD_CLOSING:
2202 		if ((oestate == XenbusStateUnknown)		||
2203 		    (oestate == XenbusStateInitialising)	||
2204 		    (oestate == XenbusStateInitWait)		||
2205 		    (oestate == XenbusStateInitialised)		||
2206 		    (oestate == XenbusStateConnected))
2207 			stcheck = STBUG;
2208 		else if (oestate == XenbusStateClosing)
2209 			stcheck = STNOP;
2210 		break;
2211 	case XD_CLOSED:
2212 		if ((oestate == XenbusStateUnknown)		||
2213 		    (oestate == XenbusStateConnected))
2214 			stcheck = STBUG;
2215 		else if ((oestate == XenbusStateInitWait)	||
2216 		    (oestate == XenbusStateInitialised)		||
2217 		    (oestate == XenbusStateClosing)		||
2218 		    (oestate == XenbusStateClosed))
2219 			stcheck = STNOP;
2220 		break;
2221 	case XD_SUSPEND:
2222 	default:
2223 			stcheck = STBUG;
2224 	}
2225 
2226 	if (stcheck == STOK)
2227 		return (DDI_SUCCESS);
2228 
2229 	if (stcheck == STBUG)
2230 		cmn_err(CE_NOTE, "xdf@%s: unexpected otherend "
2231 		    "state change to %d!, when status is %d",
2232 		    ddi_get_name_addr(vdp->xdf_dip), oestate, status);
2233 
2234 	return (DDI_FAILURE);
2235 }
2236 
2237 static int
2238 xdf_connect(xdf_t *vdp, boolean_t wait)
2239 {
2240 	ASSERT(mutex_owned(&vdp->xdf_dev_lk));
2241 	while (vdp->xdf_status != XD_READY) {
2242 		if (!wait || (vdp->xdf_status > XD_READY))
2243 			break;
2244 
2245 		if (cv_wait_sig(&vdp->xdf_dev_cv, &vdp->xdf_dev_lk) == 0)
2246 			break;
2247 	}
2248 
2249 	return (vdp->xdf_status);
2250 }
2251 
2252 /*
2253  * callback func when DMA/GTE resources is available
2254  *
2255  * Note: we only register one callback function to grant table subsystem
2256  * since we only have one 'struct gnttab_free_callback' in xdf_t.
2257  */
2258 static int
2259 xdf_dmacallback(caddr_t arg)
2260 {
2261 	xdf_t *vdp = (xdf_t *)arg;
2262 	ASSERT(vdp != NULL);
2263 
2264 	DPRINTF(DMA_DBG, ("xdf@%s: DMA callback started\n",
2265 	    ddi_get_name_addr(vdp->xdf_dip)));
2266 
2267 	ddi_trigger_softintr(vdp->xdf_softintr_id);
2268 	return (DDI_DMA_CALLBACK_DONE);
2269 }
2270 
2271 static uint_t
2272 xdf_iorestart(caddr_t arg)
2273 {
2274 	xdf_t *vdp = (xdf_t *)arg;
2275 
2276 	ASSERT(vdp != NULL);
2277 
2278 	mutex_enter(&vdp->xdf_dev_lk);
2279 	ASSERT(ISDMACBON(vdp));
2280 	SETDMACBOFF(vdp);
2281 	mutex_exit(&vdp->xdf_dev_lk);
2282 
2283 	xdf_iostart(vdp);
2284 
2285 	return (DDI_INTR_CLAIMED);
2286 }
2287 
2288 static void
2289 xdf_timeout_handler(void *arg)
2290 {
2291 	xdf_t *vdp = arg;
2292 
2293 	mutex_enter(&vdp->xdf_dev_lk);
2294 	vdp->xdf_timeout_id = 0;
2295 	mutex_exit(&vdp->xdf_dev_lk);
2296 
2297 	/* new timeout thread could be re-scheduled */
2298 	xdf_iostart(vdp);
2299 }
2300 
2301 /*
2302  * Alloc a vreq for this bp
2303  * bp->av_back contains the pointer to the vreq upon return
2304  */
2305 static v_req_t *
2306 vreq_get(xdf_t *vdp, buf_t *bp)
2307 {
2308 	v_req_t *vreq = NULL;
2309 
2310 	ASSERT(BP2VREQ(bp) == NULL);
2311 
2312 	vreq = kmem_cache_alloc(xdf_vreq_cache, KM_NOSLEEP);
2313 	if (vreq == NULL) {
2314 		if (vdp->xdf_timeout_id == 0)
2315 			/* restart I/O after one second */
2316 			vdp->xdf_timeout_id =
2317 			    timeout(xdf_timeout_handler, vdp, hz);
2318 		return (NULL);
2319 	}
2320 	bzero(vreq, sizeof (v_req_t));
2321 
2322 	list_insert_head(&vdp->xdf_vreq_act, (void *)vreq);
2323 	bp->av_back = (buf_t *)vreq;
2324 	vreq->v_buf = bp;
2325 	vreq->v_status = VREQ_INIT;
2326 	/* init of other fields in vreq is up to the caller */
2327 
2328 	return (vreq);
2329 }
2330 
2331 static void
2332 vreq_free(xdf_t *vdp, v_req_t *vreq)
2333 {
2334 	buf_t *bp = vreq->v_buf;
2335 
2336 	list_remove(&vdp->xdf_vreq_act, (void *)vreq);
2337 
2338 	if (vreq->v_flush_diskcache == FLUSH_DISKCACHE)
2339 		goto done;
2340 
2341 	switch (vreq->v_status) {
2342 	case VREQ_DMAWIN_DONE:
2343 	case VREQ_GS_ALLOCED:
2344 	case VREQ_DMABUF_BOUND:
2345 		(void) ddi_dma_unbind_handle(vreq->v_dmahdl);
2346 		/*FALLTHRU*/
2347 	case VREQ_DMAMEM_ALLOCED:
2348 		if (!ALIGNED_XFER(bp)) {
2349 			ASSERT(vreq->v_abuf != NULL);
2350 			if (!IS_ERROR(bp) && IS_READ(bp))
2351 				bcopy(vreq->v_abuf, bp->b_un.b_addr,
2352 				    bp->b_bcount);
2353 			ddi_dma_mem_free(&vreq->v_align);
2354 		}
2355 		/*FALLTHRU*/
2356 	case VREQ_MEMDMAHDL_ALLOCED:
2357 		if (!ALIGNED_XFER(bp))
2358 			ddi_dma_free_handle(&vreq->v_memdmahdl);
2359 		/*FALLTHRU*/
2360 	case VREQ_DMAHDL_ALLOCED:
2361 		ddi_dma_free_handle(&vreq->v_dmahdl);
2362 		break;
2363 	default:
2364 		break;
2365 	}
2366 done:
2367 	vreq->v_buf->av_back = NULL;
2368 	kmem_cache_free(xdf_vreq_cache, vreq);
2369 }
2370 
2371 /*
2372  * Initalize the DMA and grant table resources for the buf
2373  */
2374 static int
2375 vreq_setup(xdf_t *vdp, v_req_t *vreq)
2376 {
2377 	int rc;
2378 	ddi_dma_attr_t dmaattr;
2379 	uint_t ndcs, ndws;
2380 	ddi_dma_handle_t dh;
2381 	ddi_dma_handle_t mdh;
2382 	ddi_dma_cookie_t dc;
2383 	ddi_acc_handle_t abh;
2384 	caddr_t	aba;
2385 	ge_slot_t *gs;
2386 	size_t bufsz;
2387 	off_t off;
2388 	size_t sz;
2389 	buf_t *bp = vreq->v_buf;
2390 	int dma_flags = (IS_READ(bp) ? DDI_DMA_READ : DDI_DMA_WRITE) |
2391 	    DDI_DMA_STREAMING | DDI_DMA_PARTIAL;
2392 
2393 	switch (vreq->v_status) {
2394 	case VREQ_INIT:
2395 		if (IS_FLUSH_DISKCACHE(bp)) {
2396 			if ((gs = gs_get(vdp, IS_READ(bp))) == NULL) {
2397 				DPRINTF(DMA_DBG, (
2398 				    "xdf@%s: get ge_slotfailed\n",
2399 				    ddi_get_name_addr(vdp->xdf_dip)));
2400 				return (DDI_FAILURE);
2401 			}
2402 			vreq->v_blkno = 0;
2403 			vreq->v_nslots = 1;
2404 			vreq->v_gs = gs;
2405 			vreq->v_flush_diskcache = FLUSH_DISKCACHE;
2406 			vreq->v_status = VREQ_GS_ALLOCED;
2407 			gs->vreq = vreq;
2408 			return (DDI_SUCCESS);
2409 		}
2410 
2411 		if (IS_WRITE_BARRIER(vdp, bp))
2412 			vreq->v_flush_diskcache = WRITE_BARRIER;
2413 		vreq->v_blkno = bp->b_blkno +
2414 		    (diskaddr_t)(uintptr_t)bp->b_private;
2415 		bp->b_private = NULL;
2416 		/* See if we wrote new data to our flush block */
2417 		if (!IS_READ(bp) && USE_WRITE_BARRIER(vdp))
2418 			check_fbwrite(vdp, bp, vreq->v_blkno);
2419 		vreq->v_status = VREQ_INIT_DONE;
2420 		/*FALLTHRU*/
2421 
2422 	case VREQ_INIT_DONE:
2423 		/*
2424 		 * alloc DMA handle
2425 		 */
2426 		rc = ddi_dma_alloc_handle(vdp->xdf_dip, &xb_dma_attr,
2427 		    xdf_dmacallback, (caddr_t)vdp, &dh);
2428 		if (rc != DDI_SUCCESS) {
2429 			SETDMACBON(vdp);
2430 			DPRINTF(DMA_DBG, ("xdf@%s: DMA handle alloc failed\n",
2431 			    ddi_get_name_addr(vdp->xdf_dip)));
2432 			return (DDI_FAILURE);
2433 		}
2434 
2435 		vreq->v_dmahdl = dh;
2436 		vreq->v_status = VREQ_DMAHDL_ALLOCED;
2437 		/*FALLTHRU*/
2438 
2439 	case VREQ_DMAHDL_ALLOCED:
2440 		/*
2441 		 * alloc dma handle for 512-byte aligned buf
2442 		 */
2443 		if (!ALIGNED_XFER(bp)) {
2444 			/*
2445 			 * XXPV: we need to temporarily enlarge the seg
2446 			 * boundary and s/g length to work round CR6381968
2447 			 */
2448 			dmaattr = xb_dma_attr;
2449 			dmaattr.dma_attr_seg = (uint64_t)-1;
2450 			dmaattr.dma_attr_sgllen = INT_MAX;
2451 			rc = ddi_dma_alloc_handle(vdp->xdf_dip, &dmaattr,
2452 			    xdf_dmacallback, (caddr_t)vdp, &mdh);
2453 			if (rc != DDI_SUCCESS) {
2454 				SETDMACBON(vdp);
2455 				DPRINTF(DMA_DBG, ("xdf@%s: unaligned buf DMA"
2456 				    "handle alloc failed\n",
2457 				    ddi_get_name_addr(vdp->xdf_dip)));
2458 				return (DDI_FAILURE);
2459 			}
2460 			vreq->v_memdmahdl = mdh;
2461 			vreq->v_status = VREQ_MEMDMAHDL_ALLOCED;
2462 		}
2463 		/*FALLTHRU*/
2464 
2465 	case VREQ_MEMDMAHDL_ALLOCED:
2466 		/*
2467 		 * alloc 512-byte aligned buf
2468 		 */
2469 		if (!ALIGNED_XFER(bp)) {
2470 			if (bp->b_flags & (B_PAGEIO | B_PHYS))
2471 				bp_mapin(bp);
2472 
2473 			rc = ddi_dma_mem_alloc(vreq->v_memdmahdl,
2474 			    roundup(bp->b_bcount, XB_BSIZE), &xc_acc_attr,
2475 			    DDI_DMA_STREAMING, xdf_dmacallback, (caddr_t)vdp,
2476 			    &aba, &bufsz, &abh);
2477 			if (rc != DDI_SUCCESS) {
2478 				SETDMACBON(vdp);
2479 				DPRINTF(DMA_DBG, (
2480 				    "xdf@%s: DMA mem allocation failed\n",
2481 				    ddi_get_name_addr(vdp->xdf_dip)));
2482 				return (DDI_FAILURE);
2483 			}
2484 
2485 			vreq->v_abuf = aba;
2486 			vreq->v_align = abh;
2487 			vreq->v_status = VREQ_DMAMEM_ALLOCED;
2488 
2489 			ASSERT(bufsz >= bp->b_bcount);
2490 			if (!IS_READ(bp))
2491 				bcopy(bp->b_un.b_addr, vreq->v_abuf,
2492 				    bp->b_bcount);
2493 		}
2494 		/*FALLTHRU*/
2495 
2496 	case VREQ_DMAMEM_ALLOCED:
2497 		/*
2498 		 * dma bind
2499 		 */
2500 		if (ALIGNED_XFER(bp)) {
2501 			rc = ddi_dma_buf_bind_handle(vreq->v_dmahdl, bp,
2502 			    dma_flags, xdf_dmacallback, (caddr_t)vdp,
2503 			    &dc, &ndcs);
2504 		} else {
2505 			rc = ddi_dma_addr_bind_handle(vreq->v_dmahdl,
2506 			    NULL, vreq->v_abuf, bp->b_bcount, dma_flags,
2507 			    xdf_dmacallback, (caddr_t)vdp, &dc, &ndcs);
2508 		}
2509 		if (rc == DDI_DMA_MAPPED || rc == DDI_DMA_PARTIAL_MAP) {
2510 			/* get num of dma windows */
2511 			if (rc == DDI_DMA_PARTIAL_MAP) {
2512 				rc = ddi_dma_numwin(vreq->v_dmahdl, &ndws);
2513 				ASSERT(rc == DDI_SUCCESS);
2514 			} else {
2515 				ndws = 1;
2516 			}
2517 		} else {
2518 			SETDMACBON(vdp);
2519 			DPRINTF(DMA_DBG, ("xdf@%s: DMA bind failed\n",
2520 			    ddi_get_name_addr(vdp->xdf_dip)));
2521 			return (DDI_FAILURE);
2522 		}
2523 
2524 		vreq->v_dmac = dc;
2525 		vreq->v_dmaw = 0;
2526 		vreq->v_ndmacs = ndcs;
2527 		vreq->v_ndmaws = ndws;
2528 		vreq->v_nslots = ndws;
2529 		vreq->v_status = VREQ_DMABUF_BOUND;
2530 		/*FALLTHRU*/
2531 
2532 	case VREQ_DMABUF_BOUND:
2533 		/*
2534 		 * get ge_slot, callback is set upon failure from gs_get(),
2535 		 * if not set previously
2536 		 */
2537 		if ((gs = gs_get(vdp, IS_READ(bp))) == NULL) {
2538 			DPRINTF(DMA_DBG, ("xdf@%s: get ge_slot failed\n",
2539 			    ddi_get_name_addr(vdp->xdf_dip)));
2540 			return (DDI_FAILURE);
2541 		}
2542 
2543 		vreq->v_gs = gs;
2544 		gs->vreq = vreq;
2545 		vreq->v_status = VREQ_GS_ALLOCED;
2546 		break;
2547 
2548 	case VREQ_GS_ALLOCED:
2549 		/* nothing need to be done */
2550 		break;
2551 
2552 	case VREQ_DMAWIN_DONE:
2553 		/*
2554 		 * move to the next dma window
2555 		 */
2556 		ASSERT((vreq->v_dmaw + 1) < vreq->v_ndmaws);
2557 
2558 		/* get a ge_slot for this DMA window */
2559 		if ((gs = gs_get(vdp, IS_READ(bp))) == NULL) {
2560 			DPRINTF(DMA_DBG, ("xdf@%s: get ge_slot failed\n",
2561 			    ddi_get_name_addr(vdp->xdf_dip)));
2562 			return (DDI_FAILURE);
2563 		}
2564 
2565 		vreq->v_gs = gs;
2566 		gs->vreq = vreq;
2567 		vreq->v_dmaw++;
2568 		rc = ddi_dma_getwin(vreq->v_dmahdl, vreq->v_dmaw, &off, &sz,
2569 		    &vreq->v_dmac, &vreq->v_ndmacs);
2570 		ASSERT(rc == DDI_SUCCESS);
2571 		vreq->v_status = VREQ_GS_ALLOCED;
2572 		break;
2573 
2574 	default:
2575 		return (DDI_FAILURE);
2576 	}
2577 
2578 	return (DDI_SUCCESS);
2579 }
2580 
2581 static ge_slot_t *
2582 gs_get(xdf_t *vdp, int isread)
2583 {
2584 	grant_ref_t gh;
2585 	ge_slot_t *gs;
2586 
2587 	/* try to alloc GTEs needed in this slot, first */
2588 	if (gnttab_alloc_grant_references(
2589 	    BLKIF_MAX_SEGMENTS_PER_REQUEST, &gh) == -1) {
2590 		if (vdp->xdf_gnt_callback.next == NULL) {
2591 			SETDMACBON(vdp);
2592 			gnttab_request_free_callback(
2593 			    &vdp->xdf_gnt_callback,
2594 			    (void (*)(void *))xdf_dmacallback,
2595 			    (void *)vdp,
2596 			    BLKIF_MAX_SEGMENTS_PER_REQUEST);
2597 		}
2598 		return (NULL);
2599 	}
2600 
2601 	gs = kmem_cache_alloc(xdf_gs_cache, KM_NOSLEEP);
2602 	if (gs == NULL) {
2603 		gnttab_free_grant_references(gh);
2604 		if (vdp->xdf_timeout_id == 0)
2605 			/* restart I/O after one second */
2606 			vdp->xdf_timeout_id =
2607 			    timeout(xdf_timeout_handler, vdp, hz);
2608 		return (NULL);
2609 	}
2610 
2611 	/* init gs_slot */
2612 	list_insert_head(&vdp->xdf_gs_act, (void *)gs);
2613 	gs->oeid = vdp->xdf_peer;
2614 	gs->isread = isread;
2615 	gs->ghead = gh;
2616 	gs->ngrefs = 0;
2617 
2618 	return (gs);
2619 }
2620 
2621 static void
2622 gs_free(xdf_t *vdp, ge_slot_t *gs)
2623 {
2624 	int i;
2625 	grant_ref_t *gp = gs->ge;
2626 	int ngrefs = gs->ngrefs;
2627 	boolean_t isread = gs->isread;
2628 
2629 	list_remove(&vdp->xdf_gs_act, (void *)gs);
2630 
2631 	/* release all grant table entry resources used in this slot */
2632 	for (i = 0; i < ngrefs; i++, gp++)
2633 		gnttab_end_foreign_access(*gp, !isread, 0);
2634 	gnttab_free_grant_references(gs->ghead);
2635 
2636 	kmem_cache_free(xdf_gs_cache, (void *)gs);
2637 }
2638 
2639 static grant_ref_t
2640 gs_grant(ge_slot_t *gs, mfn_t mfn)
2641 {
2642 	grant_ref_t gr = gnttab_claim_grant_reference(&gs->ghead);
2643 
2644 	ASSERT(gr != -1);
2645 	ASSERT(gs->ngrefs < BLKIF_MAX_SEGMENTS_PER_REQUEST);
2646 	gs->ge[gs->ngrefs++] = gr;
2647 	gnttab_grant_foreign_access_ref(gr, gs->oeid, mfn, !gs->isread);
2648 
2649 	return (gr);
2650 }
2651 
2652 static void
2653 unexpectedie(xdf_t *vdp)
2654 {
2655 	/* clean up I/Os in ring that have responses */
2656 	if (xvdi_ring_has_unconsumed_responses(vdp->xdf_xb_ring)) {
2657 		mutex_exit(&vdp->xdf_dev_lk);
2658 		(void) xdf_intr((caddr_t)vdp);
2659 		mutex_enter(&vdp->xdf_dev_lk);
2660 	}
2661 
2662 	/* free up all grant table entries */
2663 	while (!list_is_empty(&vdp->xdf_gs_act))
2664 		gs_free(vdp, list_head(&vdp->xdf_gs_act));
2665 
2666 	/*
2667 	 * move bp back to active list orderly
2668 	 * vreq_busy is updated in vreq_free()
2669 	 */
2670 	while (!list_is_empty(&vdp->xdf_vreq_act)) {
2671 		v_req_t *vreq = list_head(&vdp->xdf_vreq_act);
2672 		buf_t *bp = vreq->v_buf;
2673 
2674 		bp->av_back = NULL;
2675 		bp->b_resid = bp->b_bcount;
2676 		if (vdp->xdf_f_act == NULL) {
2677 			vdp->xdf_f_act = vdp->xdf_l_act = bp;
2678 		} else {
2679 			/* move to the head of list */
2680 			bp->av_forw = vdp->xdf_f_act;
2681 			vdp->xdf_f_act = bp;
2682 		}
2683 		if (vdp->xdf_xdev_iostat != NULL)
2684 			kstat_runq_back_to_waitq(
2685 			    KSTAT_IO_PTR(vdp->xdf_xdev_iostat));
2686 		vreq_free(vdp, vreq);
2687 	}
2688 }
2689 
2690 static void
2691 xdfmin(struct buf *bp)
2692 {
2693 	if (bp->b_bcount > xdf_maxphys)
2694 		bp->b_bcount = xdf_maxphys;
2695 }
2696 
2697 void
2698 xdf_kstat_delete(dev_info_t *dip)
2699 {
2700 	xdf_t	*vdp = (xdf_t *)ddi_get_driver_private(dip);
2701 	kstat_t	*kstat;
2702 
2703 	/*
2704 	 * The locking order here is xdf_iostat_lk and then xdf_dev_lk.
2705 	 * xdf_dev_lk is used to protect the xdf_xdev_iostat pointer
2706 	 * and the contents of the our kstat.  xdf_iostat_lk is used
2707 	 * to protect the allocation and freeing of the actual kstat.
2708 	 * xdf_dev_lk can't be used for this purpose because kstat
2709 	 * readers use it to access the contents of the kstat and
2710 	 * hence it can't be held when calling kstat_delete().
2711 	 */
2712 	mutex_enter(&vdp->xdf_iostat_lk);
2713 	mutex_enter(&vdp->xdf_dev_lk);
2714 
2715 	if (vdp->xdf_xdev_iostat == NULL) {
2716 		mutex_exit(&vdp->xdf_dev_lk);
2717 		mutex_exit(&vdp->xdf_iostat_lk);
2718 		return;
2719 	}
2720 
2721 	kstat = vdp->xdf_xdev_iostat;
2722 	vdp->xdf_xdev_iostat = NULL;
2723 	mutex_exit(&vdp->xdf_dev_lk);
2724 
2725 	kstat_delete(kstat);
2726 	mutex_exit(&vdp->xdf_iostat_lk);
2727 }
2728 
2729 int
2730 xdf_kstat_create(dev_info_t *dip, char *ks_module, int ks_instance)
2731 {
2732 	xdf_t	*vdp = (xdf_t *)ddi_get_driver_private(dip);
2733 
2734 	/* See comment about locking in xdf_kstat_delete(). */
2735 	mutex_enter(&vdp->xdf_iostat_lk);
2736 	mutex_enter(&vdp->xdf_dev_lk);
2737 
2738 	if (vdp->xdf_xdev_iostat != NULL) {
2739 		mutex_exit(&vdp->xdf_dev_lk);
2740 		mutex_exit(&vdp->xdf_iostat_lk);
2741 		return (-1);
2742 	}
2743 
2744 	if ((vdp->xdf_xdev_iostat = kstat_create(
2745 	    ks_module, ks_instance, NULL, "disk",
2746 	    KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT)) == NULL) {
2747 		mutex_exit(&vdp->xdf_dev_lk);
2748 		mutex_exit(&vdp->xdf_iostat_lk);
2749 		return (-1);
2750 	}
2751 
2752 	vdp->xdf_xdev_iostat->ks_lock = &vdp->xdf_dev_lk;
2753 	kstat_install(vdp->xdf_xdev_iostat);
2754 	mutex_exit(&vdp->xdf_dev_lk);
2755 	mutex_exit(&vdp->xdf_iostat_lk);
2756 
2757 	return (0);
2758 }
2759 
2760 #if defined(XPV_HVM_DRIVER)
2761 
2762 typedef struct xdf_hvm_entry {
2763 	list_node_t	xdf_he_list;
2764 	char		*xdf_he_path;
2765 	dev_info_t	*xdf_he_dip;
2766 } xdf_hvm_entry_t;
2767 
2768 static list_t xdf_hvm_list;
2769 static kmutex_t xdf_hvm_list_lock;
2770 
2771 static xdf_hvm_entry_t *
2772 i_xdf_hvm_find(char *path, dev_info_t *dip)
2773 {
2774 	xdf_hvm_entry_t	*i;
2775 
2776 	ASSERT((path != NULL) || (dip != NULL));
2777 	ASSERT(MUTEX_HELD(&xdf_hvm_list_lock));
2778 
2779 	i = list_head(&xdf_hvm_list);
2780 	while (i != NULL) {
2781 		if ((path != NULL) && strcmp(i->xdf_he_path, path) != 0) {
2782 			i = list_next(&xdf_hvm_list, i);
2783 			continue;
2784 		}
2785 		if ((dip != NULL) && (i->xdf_he_dip != dip)) {
2786 			i = list_next(&xdf_hvm_list, i);
2787 			continue;
2788 		}
2789 		break;
2790 	}
2791 	return (i);
2792 }
2793 
2794 dev_info_t *
2795 xdf_hvm_hold(char *path)
2796 {
2797 	xdf_hvm_entry_t	*i;
2798 	dev_info_t	*dip;
2799 
2800 	mutex_enter(&xdf_hvm_list_lock);
2801 	i = i_xdf_hvm_find(path, NULL);
2802 	if (i == NULL) {
2803 		mutex_exit(&xdf_hvm_list_lock);
2804 		return (B_FALSE);
2805 	}
2806 	ndi_hold_devi(dip = i->xdf_he_dip);
2807 	mutex_exit(&xdf_hvm_list_lock);
2808 	return (dip);
2809 }
2810 
2811 static void
2812 xdf_hvm_add(dev_info_t *dip)
2813 {
2814 	xdf_hvm_entry_t	*i;
2815 	char		*path;
2816 
2817 	/* figure out the path for the dip */
2818 	path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
2819 	(void) ddi_pathname(dip, path);
2820 
2821 	i = kmem_alloc(sizeof (*i), KM_SLEEP);
2822 	i->xdf_he_dip = dip;
2823 	i->xdf_he_path = i_ddi_strdup(path, KM_SLEEP);
2824 
2825 	mutex_enter(&xdf_hvm_list_lock);
2826 	ASSERT(i_xdf_hvm_find(path, NULL) == NULL);
2827 	ASSERT(i_xdf_hvm_find(NULL, dip) == NULL);
2828 	list_insert_head(&xdf_hvm_list, i);
2829 	mutex_exit(&xdf_hvm_list_lock);
2830 
2831 	kmem_free(path, MAXPATHLEN);
2832 }
2833 
2834 static void
2835 xdf_hvm_rm(dev_info_t *dip)
2836 {
2837 	xdf_hvm_entry_t	*i;
2838 
2839 	mutex_enter(&xdf_hvm_list_lock);
2840 	VERIFY((i = i_xdf_hvm_find(NULL, dip)) != NULL);
2841 	list_remove(&xdf_hvm_list, i);
2842 	mutex_exit(&xdf_hvm_list_lock);
2843 
2844 	kmem_free(i->xdf_he_path, strlen(i->xdf_he_path) + 1);
2845 	kmem_free(i, sizeof (*i));
2846 }
2847 
2848 static void
2849 xdf_hvm_init(void)
2850 {
2851 	list_create(&xdf_hvm_list, sizeof (xdf_hvm_entry_t),
2852 	    offsetof(xdf_hvm_entry_t, xdf_he_list));
2853 	mutex_init(&xdf_hvm_list_lock, NULL, MUTEX_DEFAULT, NULL);
2854 }
2855 
2856 static void
2857 xdf_hvm_fini(void)
2858 {
2859 	ASSERT(list_head(&xdf_hvm_list) == NULL);
2860 	list_destroy(&xdf_hvm_list);
2861 	mutex_destroy(&xdf_hvm_list_lock);
2862 }
2863 
2864 int
2865 xdf_hvm_connect(dev_info_t *dip)
2866 {
2867 	xdf_t	*vdp = (xdf_t *)ddi_get_driver_private(dip);
2868 	int	rv;
2869 
2870 	/* do cv_wait until connected or failed */
2871 	mutex_enter(&vdp->xdf_dev_lk);
2872 	rv = xdf_connect(vdp, B_TRUE);
2873 	mutex_exit(&vdp->xdf_dev_lk);
2874 	return ((rv == XD_READY) ? 0 : -1);
2875 }
2876 
2877 int
2878 xdf_hvm_setpgeom(dev_info_t *dip, cmlb_geom_t *geomp)
2879 {
2880 	xdf_t	*vdp = (xdf_t *)ddi_get_driver_private(dip);
2881 
2882 	/* sanity check the requested physical geometry */
2883 	mutex_enter(&vdp->xdf_dev_lk);
2884 	if ((geomp->g_secsize != XB_BSIZE) ||
2885 	    (geomp->g_capacity == 0)) {
2886 		mutex_exit(&vdp->xdf_dev_lk);
2887 		return (EINVAL);
2888 	}
2889 
2890 	/*
2891 	 * If we've already connected to the backend device then make sure
2892 	 * we're not defining a physical geometry larger than our backend
2893 	 * device.
2894 	 */
2895 	if ((vdp->xdf_xdev_nblocks != 0) &&
2896 	    (geomp->g_capacity > vdp->xdf_xdev_nblocks)) {
2897 		mutex_exit(&vdp->xdf_dev_lk);
2898 		return (EINVAL);
2899 	}
2900 
2901 	vdp->xdf_pgeom = *geomp;
2902 	mutex_exit(&vdp->xdf_dev_lk);
2903 
2904 	/* force a re-validation */
2905 	cmlb_invalidate(vdp->xdf_vd_lbl, NULL);
2906 
2907 	return (0);
2908 }
2909 
2910 #endif /* XPV_HVM_DRIVER */
2911