xref: /freebsd/sys/dev/xen/blkfront/blkfront.c (revision 63a938566d524836885917d95bd491aa4400b181)
1 /*
2  * XenBSD block device driver
3  *
4  * Copyright (c) 2010-2013 Spectra Logic Corporation
5  * Copyright (c) 2009 Scott Long, Yahoo!
6  * Copyright (c) 2009 Frank Suchomel, Citrix
7  * Copyright (c) 2009 Doug F. Rabson, Citrix
8  * Copyright (c) 2005 Kip Macy
9  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
10  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
11  *
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this software and associated documentation files (the "Software"), to
15  * deal in the Software without restriction, including without limitation the
16  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
17  * sell copies of the Software, and to permit persons to whom the Software is
18  * furnished to do so, subject to the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * all copies or substantial portions of the Software.
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
28  * DEALINGS IN THE SOFTWARE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
37 #include <sys/kernel.h>
38 #include <vm/vm.h>
39 #include <vm/pmap.h>
40 
41 #include <sys/bio.h>
42 #include <sys/bus.h>
43 #include <sys/conf.h>
44 #include <sys/module.h>
45 #include <sys/sysctl.h>
46 
47 #include <machine/bus.h>
48 #include <sys/rman.h>
49 #include <machine/resource.h>
50 #include <machine/intr_machdep.h>
51 #include <machine/vmparam.h>
52 
53 #include <xen/xen-os.h>
54 #include <xen/hypervisor.h>
55 #include <xen/xen_intr.h>
56 #include <xen/gnttab.h>
57 #include <xen/interface/grant_table.h>
58 #include <xen/interface/io/protocols.h>
59 #include <xen/xenbus/xenbusvar.h>
60 
61 #include <machine/_inttypes.h>
62 
63 #include <geom/geom_disk.h>
64 
65 #include <dev/xen/blkfront/block.h>
66 
67 #include "xenbus_if.h"
68 
69 /*--------------------------- Forward Declarations ---------------------------*/
70 static void xbd_closing(device_t);
71 static void xbd_startio(struct xbd_softc *sc);
72 
73 /*---------------------------------- Macros ----------------------------------*/
74 #if 0
75 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
76 #else
77 #define DPRINTK(fmt, args...)
78 #endif
79 
80 #define XBD_SECTOR_SHFT		9
81 
82 /*---------------------------- Global Static Data ----------------------------*/
83 static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
84 
85 static int xbd_enable_indirect = 1;
86 SYSCTL_NODE(_hw, OID_AUTO, xbd, CTLFLAG_RD, 0, "xbd driver parameters");
87 SYSCTL_INT(_hw_xbd, OID_AUTO, xbd_enable_indirect, CTLFLAG_RDTUN,
88     &xbd_enable_indirect, 0, "Enable xbd indirect segments");
89 
90 /*---------------------------- Command Processing ----------------------------*/
91 static void
92 xbd_freeze(struct xbd_softc *sc, xbd_flag_t xbd_flag)
93 {
94 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) != 0)
95 		return;
96 
97 	sc->xbd_flags |= xbd_flag;
98 	sc->xbd_qfrozen_cnt++;
99 }
100 
101 static void
102 xbd_thaw(struct xbd_softc *sc, xbd_flag_t xbd_flag)
103 {
104 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) == 0)
105 		return;
106 
107 	if (sc->xbd_qfrozen_cnt == 0)
108 		panic("%s: Thaw with flag 0x%x while not frozen.",
109 		    __func__, xbd_flag);
110 
111 	sc->xbd_flags &= ~xbd_flag;
112 	sc->xbd_qfrozen_cnt--;
113 }
114 
115 static void
116 xbd_cm_freeze(struct xbd_softc *sc, struct xbd_command *cm, xbdc_flag_t cm_flag)
117 {
118 	if ((cm->cm_flags & XBDCF_FROZEN) != 0)
119 		return;
120 
121 	cm->cm_flags |= XBDCF_FROZEN|cm_flag;
122 	xbd_freeze(sc, XBDF_NONE);
123 }
124 
125 static void
126 xbd_cm_thaw(struct xbd_softc *sc, struct xbd_command *cm)
127 {
128 	if ((cm->cm_flags & XBDCF_FROZEN) == 0)
129 		return;
130 
131 	cm->cm_flags &= ~XBDCF_FROZEN;
132 	xbd_thaw(sc, XBDF_NONE);
133 }
134 
135 static inline void
136 xbd_flush_requests(struct xbd_softc *sc)
137 {
138 	int notify;
139 
140 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->xbd_ring, notify);
141 
142 	if (notify)
143 		xen_intr_signal(sc->xen_intr_handle);
144 }
145 
146 static void
147 xbd_free_command(struct xbd_command *cm)
148 {
149 
150 	KASSERT((cm->cm_flags & XBDCF_Q_MASK) == XBD_Q_NONE,
151 	    ("Freeing command that is still on queue %d.",
152 	    cm->cm_flags & XBDCF_Q_MASK));
153 
154 	cm->cm_flags = XBDCF_INITIALIZER;
155 	cm->cm_bp = NULL;
156 	cm->cm_complete = NULL;
157 	xbd_enqueue_cm(cm, XBD_Q_FREE);
158 	xbd_thaw(cm->cm_sc, XBDF_CM_SHORTAGE);
159 }
160 
161 static void
162 xbd_mksegarray(bus_dma_segment_t *segs, int nsegs,
163     grant_ref_t * gref_head, int otherend_id, int readonly,
164     grant_ref_t * sg_ref, struct blkif_request_segment *sg)
165 {
166 	struct blkif_request_segment *last_block_sg = sg + nsegs;
167 	vm_paddr_t buffer_ma;
168 	uint64_t fsect, lsect;
169 	int ref;
170 
171 	while (sg < last_block_sg) {
172 		KASSERT(segs->ds_addr % (1 << XBD_SECTOR_SHFT) == 0,
173 		    ("XEN disk driver I/O must be sector aligned"));
174 		KASSERT(segs->ds_len % (1 << XBD_SECTOR_SHFT) == 0,
175 		    ("XEN disk driver I/Os must be a multiple of "
176 		    "the sector length"));
177 		buffer_ma = segs->ds_addr;
178 		fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
179 		lsect = fsect + (segs->ds_len  >> XBD_SECTOR_SHFT) - 1;
180 
181 		KASSERT(lsect <= 7, ("XEN disk driver data cannot "
182 		    "cross a page boundary"));
183 
184 		/* install a grant reference. */
185 		ref = gnttab_claim_grant_reference(gref_head);
186 
187 		/*
188 		 * GNTTAB_LIST_END == 0xffffffff, but it is private
189 		 * to gnttab.c.
190 		 */
191 		KASSERT(ref != ~0, ("grant_reference failed"));
192 
193 		gnttab_grant_foreign_access_ref(
194 		    ref,
195 		    otherend_id,
196 		    buffer_ma >> PAGE_SHIFT,
197 		    readonly);
198 
199 		*sg_ref = ref;
200 		*sg = (struct blkif_request_segment) {
201 			.gref       = ref,
202 			.first_sect = fsect,
203 			.last_sect  = lsect
204 		};
205 		sg++;
206 		sg_ref++;
207 		segs++;
208 	}
209 }
210 
211 static void
212 xbd_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
213 {
214 	struct xbd_softc *sc;
215 	struct xbd_command *cm;
216 	int op;
217 
218 	cm = arg;
219 	sc = cm->cm_sc;
220 
221 	if (error) {
222 		cm->cm_bp->bio_error = EIO;
223 		biodone(cm->cm_bp);
224 		xbd_free_command(cm);
225 		return;
226 	}
227 
228 	KASSERT(nsegs <= sc->xbd_max_request_segments,
229 	    ("Too many segments in a blkfront I/O"));
230 
231 	if (nsegs <= BLKIF_MAX_SEGMENTS_PER_REQUEST) {
232 		blkif_request_t	*ring_req;
233 
234 		/* Fill out a blkif_request_t structure. */
235 		ring_req = (blkif_request_t *)
236 		    RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
237 		sc->xbd_ring.req_prod_pvt++;
238 		ring_req->id = cm->cm_id;
239 		ring_req->operation = cm->cm_operation;
240 		ring_req->sector_number = cm->cm_sector_number;
241 		ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
242 		ring_req->nr_segments = nsegs;
243 		cm->cm_nseg = nsegs;
244 		xbd_mksegarray(segs, nsegs, &cm->cm_gref_head,
245 		    xenbus_get_otherend_id(sc->xbd_dev),
246 		    cm->cm_operation == BLKIF_OP_WRITE,
247 		    cm->cm_sg_refs, ring_req->seg);
248 	} else {
249 		blkif_request_indirect_t *ring_req;
250 
251 		/* Fill out a blkif_request_indirect_t structure. */
252 		ring_req = (blkif_request_indirect_t *)
253 		    RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
254 		sc->xbd_ring.req_prod_pvt++;
255 		ring_req->id = cm->cm_id;
256 		ring_req->operation = BLKIF_OP_INDIRECT;
257 		ring_req->indirect_op = cm->cm_operation;
258 		ring_req->sector_number = cm->cm_sector_number;
259 		ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
260 		ring_req->nr_segments = nsegs;
261 		cm->cm_nseg = nsegs;
262 		xbd_mksegarray(segs, nsegs, &cm->cm_gref_head,
263 		    xenbus_get_otherend_id(sc->xbd_dev),
264 		    cm->cm_operation == BLKIF_OP_WRITE,
265 		    cm->cm_sg_refs, cm->cm_indirectionpages);
266 		memcpy(ring_req->indirect_grefs, &cm->cm_indirectionrefs,
267 		    sizeof(grant_ref_t) * sc->xbd_max_request_indirectpages);
268 	}
269 
270 	if (cm->cm_operation == BLKIF_OP_READ)
271 		op = BUS_DMASYNC_PREREAD;
272 	else if (cm->cm_operation == BLKIF_OP_WRITE)
273 		op = BUS_DMASYNC_PREWRITE;
274 	else
275 		op = 0;
276 	bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
277 
278 	gnttab_free_grant_references(cm->cm_gref_head);
279 
280 	xbd_enqueue_cm(cm, XBD_Q_BUSY);
281 
282 	/*
283 	 * If bus dma had to asynchronously call us back to dispatch
284 	 * this command, we are no longer executing in the context of
285 	 * xbd_startio().  Thus we cannot rely on xbd_startio()'s call to
286 	 * xbd_flush_requests() to publish this command to the backend
287 	 * along with any other commands that it could batch.
288 	 */
289 	if ((cm->cm_flags & XBDCF_ASYNC_MAPPING) != 0)
290 		xbd_flush_requests(sc);
291 
292 	return;
293 }
294 
295 static int
296 xbd_queue_request(struct xbd_softc *sc, struct xbd_command *cm)
297 {
298 	int error;
299 
300 	if (cm->cm_bp != NULL)
301 		error = bus_dmamap_load_bio(sc->xbd_io_dmat, cm->cm_map,
302 		    cm->cm_bp, xbd_queue_cb, cm, 0);
303 	else
304 		error = bus_dmamap_load(sc->xbd_io_dmat, cm->cm_map,
305 		    cm->cm_data, cm->cm_datalen, xbd_queue_cb, cm, 0);
306 	if (error == EINPROGRESS) {
307 		/*
308 		 * Maintain queuing order by freezing the queue.  The next
309 		 * command may not require as many resources as the command
310 		 * we just attempted to map, so we can't rely on bus dma
311 		 * blocking for it too.
312 		 */
313 		xbd_cm_freeze(sc, cm, XBDCF_ASYNC_MAPPING);
314 		return (0);
315 	}
316 
317 	return (error);
318 }
319 
320 static void
321 xbd_restart_queue_callback(void *arg)
322 {
323 	struct xbd_softc *sc = arg;
324 
325 	mtx_lock(&sc->xbd_io_lock);
326 
327 	xbd_thaw(sc, XBDF_GNT_SHORTAGE);
328 
329 	xbd_startio(sc);
330 
331 	mtx_unlock(&sc->xbd_io_lock);
332 }
333 
334 static struct xbd_command *
335 xbd_bio_command(struct xbd_softc *sc)
336 {
337 	struct xbd_command *cm;
338 	struct bio *bp;
339 
340 	if (__predict_false(sc->xbd_state != XBD_STATE_CONNECTED))
341 		return (NULL);
342 
343 	bp = xbd_dequeue_bio(sc);
344 	if (bp == NULL)
345 		return (NULL);
346 
347 	if ((cm = xbd_dequeue_cm(sc, XBD_Q_FREE)) == NULL) {
348 		xbd_freeze(sc, XBDF_CM_SHORTAGE);
349 		xbd_requeue_bio(sc, bp);
350 		return (NULL);
351 	}
352 
353 	if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
354 	    &cm->cm_gref_head) != 0) {
355 		gnttab_request_free_callback(&sc->xbd_callback,
356 		    xbd_restart_queue_callback, sc,
357 		    sc->xbd_max_request_segments);
358 		xbd_freeze(sc, XBDF_GNT_SHORTAGE);
359 		xbd_requeue_bio(sc, bp);
360 		xbd_enqueue_cm(cm, XBD_Q_FREE);
361 		return (NULL);
362 	}
363 
364 	cm->cm_bp = bp;
365 	cm->cm_sector_number = (blkif_sector_t)bp->bio_pblkno;
366 
367 	switch (bp->bio_cmd) {
368 	case BIO_READ:
369 		cm->cm_operation = BLKIF_OP_READ;
370 		break;
371 	case BIO_WRITE:
372 		cm->cm_operation = BLKIF_OP_WRITE;
373 		if ((bp->bio_flags & BIO_ORDERED) != 0) {
374 			if ((sc->xbd_flags & XBDF_BARRIER) != 0) {
375 				cm->cm_operation = BLKIF_OP_WRITE_BARRIER;
376 			} else {
377 				/*
378 				 * Single step this command.
379 				 */
380 				cm->cm_flags |= XBDCF_Q_FREEZE;
381 				if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
382 					/*
383 					 * Wait for in-flight requests to
384 					 * finish.
385 					 */
386 					xbd_freeze(sc, XBDF_WAIT_IDLE);
387 					xbd_requeue_cm(cm, XBD_Q_READY);
388 					return (NULL);
389 				}
390 			}
391 		}
392 		break;
393 	case BIO_FLUSH:
394 		if ((sc->xbd_flags & XBDF_FLUSH) != 0)
395 			cm->cm_operation = BLKIF_OP_FLUSH_DISKCACHE;
396 		else if ((sc->xbd_flags & XBDF_BARRIER) != 0)
397 			cm->cm_operation = BLKIF_OP_WRITE_BARRIER;
398 		else
399 			panic("flush request, but no flush support available");
400 		break;
401 	default:
402 		panic("unknown bio command %d", bp->bio_cmd);
403 	}
404 
405 	return (cm);
406 }
407 
408 /*
409  * Dequeue buffers and place them in the shared communication ring.
410  * Return when no more requests can be accepted or all buffers have
411  * been queued.
412  *
413  * Signal XEN once the ring has been filled out.
414  */
415 static void
416 xbd_startio(struct xbd_softc *sc)
417 {
418 	struct xbd_command *cm;
419 	int error, queued = 0;
420 
421 	mtx_assert(&sc->xbd_io_lock, MA_OWNED);
422 
423 	if (sc->xbd_state != XBD_STATE_CONNECTED)
424 		return;
425 
426 	while (!RING_FULL(&sc->xbd_ring)) {
427 
428 		if (sc->xbd_qfrozen_cnt != 0)
429 			break;
430 
431 		cm = xbd_dequeue_cm(sc, XBD_Q_READY);
432 
433 		if (cm == NULL)
434 		    cm = xbd_bio_command(sc);
435 
436 		if (cm == NULL)
437 			break;
438 
439 		if ((cm->cm_flags & XBDCF_Q_FREEZE) != 0) {
440 			/*
441 			 * Single step command.  Future work is
442 			 * held off until this command completes.
443 			 */
444 			xbd_cm_freeze(sc, cm, XBDCF_Q_FREEZE);
445 		}
446 
447 		if ((error = xbd_queue_request(sc, cm)) != 0) {
448 			printf("xbd_queue_request returned %d\n", error);
449 			break;
450 		}
451 		queued++;
452 	}
453 
454 	if (queued != 0)
455 		xbd_flush_requests(sc);
456 }
457 
458 static void
459 xbd_bio_complete(struct xbd_softc *sc, struct xbd_command *cm)
460 {
461 	struct bio *bp;
462 
463 	bp = cm->cm_bp;
464 
465 	if (__predict_false(cm->cm_status != BLKIF_RSP_OKAY)) {
466 		disk_err(bp, "disk error" , -1, 0);
467 		printf(" status: %x\n", cm->cm_status);
468 		bp->bio_flags |= BIO_ERROR;
469 	}
470 
471 	if (bp->bio_flags & BIO_ERROR)
472 		bp->bio_error = EIO;
473 	else
474 		bp->bio_resid = 0;
475 
476 	xbd_free_command(cm);
477 	biodone(bp);
478 }
479 
480 static void
481 xbd_int(void *xsc)
482 {
483 	struct xbd_softc *sc = xsc;
484 	struct xbd_command *cm;
485 	blkif_response_t *bret;
486 	RING_IDX i, rp;
487 	int op;
488 
489 	mtx_lock(&sc->xbd_io_lock);
490 
491 	if (__predict_false(sc->xbd_state == XBD_STATE_DISCONNECTED)) {
492 		mtx_unlock(&sc->xbd_io_lock);
493 		return;
494 	}
495 
496  again:
497 	rp = sc->xbd_ring.sring->rsp_prod;
498 	rmb(); /* Ensure we see queued responses up to 'rp'. */
499 
500 	for (i = sc->xbd_ring.rsp_cons; i != rp;) {
501 		bret = RING_GET_RESPONSE(&sc->xbd_ring, i);
502 		cm   = &sc->xbd_shadow[bret->id];
503 
504 		xbd_remove_cm(cm, XBD_Q_BUSY);
505 		gnttab_end_foreign_access_references(cm->cm_nseg,
506 		    cm->cm_sg_refs);
507 		i++;
508 
509 		if (cm->cm_operation == BLKIF_OP_READ)
510 			op = BUS_DMASYNC_POSTREAD;
511 		else if (cm->cm_operation == BLKIF_OP_WRITE ||
512 		    cm->cm_operation == BLKIF_OP_WRITE_BARRIER)
513 			op = BUS_DMASYNC_POSTWRITE;
514 		else
515 			op = 0;
516 		bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
517 		bus_dmamap_unload(sc->xbd_io_dmat, cm->cm_map);
518 
519 		/*
520 		 * Release any hold this command has on future command
521 		 * dispatch.
522 		 */
523 		xbd_cm_thaw(sc, cm);
524 
525 		/*
526 		 * Directly call the i/o complete routine to save an
527 		 * an indirection in the common case.
528 		 */
529 		cm->cm_status = bret->status;
530 		if (cm->cm_bp)
531 			xbd_bio_complete(sc, cm);
532 		else if (cm->cm_complete != NULL)
533 			cm->cm_complete(cm);
534 		else
535 			xbd_free_command(cm);
536 	}
537 
538 	sc->xbd_ring.rsp_cons = i;
539 
540 	if (i != sc->xbd_ring.req_prod_pvt) {
541 		int more_to_do;
542 		RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, more_to_do);
543 		if (more_to_do)
544 			goto again;
545 	} else {
546 		sc->xbd_ring.sring->rsp_event = i + 1;
547 	}
548 
549 	if (xbd_queue_length(sc, XBD_Q_BUSY) == 0)
550 		xbd_thaw(sc, XBDF_WAIT_IDLE);
551 
552 	xbd_startio(sc);
553 
554 	if (__predict_false(sc->xbd_state == XBD_STATE_SUSPENDED))
555 		wakeup(&sc->xbd_cm_q[XBD_Q_BUSY]);
556 
557 	mtx_unlock(&sc->xbd_io_lock);
558 }
559 
560 /*------------------------------- Dump Support -------------------------------*/
561 /**
562  * Quiesce the disk writes for a dump file before allowing the next buffer.
563  */
564 static void
565 xbd_quiesce(struct xbd_softc *sc)
566 {
567 	int mtd;
568 
569 	// While there are outstanding requests
570 	while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
571 		RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, mtd);
572 		if (mtd) {
573 			/* Received request completions, update queue. */
574 			xbd_int(sc);
575 		}
576 		if (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
577 			/*
578 			 * Still pending requests, wait for the disk i/o
579 			 * to complete.
580 			 */
581 			HYPERVISOR_yield();
582 		}
583 	}
584 }
585 
586 /* Kernel dump function for a paravirtualized disk device */
587 static void
588 xbd_dump_complete(struct xbd_command *cm)
589 {
590 
591 	xbd_enqueue_cm(cm, XBD_Q_COMPLETE);
592 }
593 
594 static int
595 xbd_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
596     size_t length)
597 {
598 	struct disk *dp = arg;
599 	struct xbd_softc *sc = dp->d_drv1;
600 	struct xbd_command *cm;
601 	size_t chunk;
602 	int sbp;
603 	int rc = 0;
604 
605 	if (length <= 0)
606 		return (rc);
607 
608 	xbd_quiesce(sc);	/* All quiet on the western front. */
609 
610 	/*
611 	 * If this lock is held, then this module is failing, and a
612 	 * successful kernel dump is highly unlikely anyway.
613 	 */
614 	mtx_lock(&sc->xbd_io_lock);
615 
616 	/* Split the 64KB block as needed */
617 	for (sbp=0; length > 0; sbp++) {
618 		cm = xbd_dequeue_cm(sc, XBD_Q_FREE);
619 		if (cm == NULL) {
620 			mtx_unlock(&sc->xbd_io_lock);
621 			device_printf(sc->xbd_dev, "dump: no more commands?\n");
622 			return (EBUSY);
623 		}
624 
625 		if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
626 		    &cm->cm_gref_head) != 0) {
627 			xbd_free_command(cm);
628 			mtx_unlock(&sc->xbd_io_lock);
629 			device_printf(sc->xbd_dev, "no more grant allocs?\n");
630 			return (EBUSY);
631 		}
632 
633 		chunk = length > sc->xbd_max_request_size ?
634 		    sc->xbd_max_request_size : length;
635 		cm->cm_data = virtual;
636 		cm->cm_datalen = chunk;
637 		cm->cm_operation = BLKIF_OP_WRITE;
638 		cm->cm_sector_number = offset / dp->d_sectorsize;
639 		cm->cm_complete = xbd_dump_complete;
640 
641 		xbd_enqueue_cm(cm, XBD_Q_READY);
642 
643 		length -= chunk;
644 		offset += chunk;
645 		virtual = (char *) virtual + chunk;
646 	}
647 
648 	/* Tell DOM0 to do the I/O */
649 	xbd_startio(sc);
650 	mtx_unlock(&sc->xbd_io_lock);
651 
652 	/* Poll for the completion. */
653 	xbd_quiesce(sc);	/* All quite on the eastern front */
654 
655 	/* If there were any errors, bail out... */
656 	while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) {
657 		if (cm->cm_status != BLKIF_RSP_OKAY) {
658 			device_printf(sc->xbd_dev,
659 			    "Dump I/O failed at sector %jd\n",
660 			    cm->cm_sector_number);
661 			rc = EIO;
662 		}
663 		xbd_free_command(cm);
664 	}
665 
666 	return (rc);
667 }
668 
669 /*----------------------------- Disk Entrypoints -----------------------------*/
670 static int
671 xbd_open(struct disk *dp)
672 {
673 	struct xbd_softc *sc = dp->d_drv1;
674 
675 	if (sc == NULL) {
676 		printf("xbd%d: not found", dp->d_unit);
677 		return (ENXIO);
678 	}
679 
680 	sc->xbd_flags |= XBDF_OPEN;
681 	sc->xbd_users++;
682 	return (0);
683 }
684 
685 static int
686 xbd_close(struct disk *dp)
687 {
688 	struct xbd_softc *sc = dp->d_drv1;
689 
690 	if (sc == NULL)
691 		return (ENXIO);
692 	sc->xbd_flags &= ~XBDF_OPEN;
693 	if (--(sc->xbd_users) == 0) {
694 		/*
695 		 * Check whether we have been instructed to close.  We will
696 		 * have ignored this request initially, as the device was
697 		 * still mounted.
698 		 */
699 		if (xenbus_get_otherend_state(sc->xbd_dev) ==
700 		    XenbusStateClosing)
701 			xbd_closing(sc->xbd_dev);
702 	}
703 	return (0);
704 }
705 
706 static int
707 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
708 {
709 	struct xbd_softc *sc = dp->d_drv1;
710 
711 	if (sc == NULL)
712 		return (ENXIO);
713 
714 	return (ENOTTY);
715 }
716 
717 /*
718  * Read/write routine for a buffer.  Finds the proper unit, place it on
719  * the sortq and kick the controller.
720  */
721 static void
722 xbd_strategy(struct bio *bp)
723 {
724 	struct xbd_softc *sc = bp->bio_disk->d_drv1;
725 
726 	/* bogus disk? */
727 	if (sc == NULL) {
728 		bp->bio_error = EINVAL;
729 		bp->bio_flags |= BIO_ERROR;
730 		bp->bio_resid = bp->bio_bcount;
731 		biodone(bp);
732 		return;
733 	}
734 
735 	/*
736 	 * Place it in the queue of disk activities for this disk
737 	 */
738 	mtx_lock(&sc->xbd_io_lock);
739 
740 	xbd_enqueue_bio(sc, bp);
741 	xbd_startio(sc);
742 
743 	mtx_unlock(&sc->xbd_io_lock);
744 	return;
745 }
746 
747 /*------------------------------ Ring Management -----------------------------*/
748 static int
749 xbd_alloc_ring(struct xbd_softc *sc)
750 {
751 	blkif_sring_t *sring;
752 	uintptr_t sring_page_addr;
753 	int error;
754 	int i;
755 
756 	sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
757 	    M_NOWAIT|M_ZERO);
758 	if (sring == NULL) {
759 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring");
760 		return (ENOMEM);
761 	}
762 	SHARED_RING_INIT(sring);
763 	FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE);
764 
765 	for (i = 0, sring_page_addr = (uintptr_t)sring;
766 	     i < sc->xbd_ring_pages;
767 	     i++, sring_page_addr += PAGE_SIZE) {
768 
769 		error = xenbus_grant_ring(sc->xbd_dev,
770 		    (vtophys(sring_page_addr) >> PAGE_SHIFT),
771 		    &sc->xbd_ring_ref[i]);
772 		if (error) {
773 			xenbus_dev_fatal(sc->xbd_dev, error,
774 			    "granting ring_ref(%d)", i);
775 			return (error);
776 		}
777 	}
778 	if (sc->xbd_ring_pages == 1) {
779 		error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
780 		    "ring-ref", "%u", sc->xbd_ring_ref[0]);
781 		if (error) {
782 			xenbus_dev_fatal(sc->xbd_dev, error,
783 			    "writing %s/ring-ref",
784 			    xenbus_get_node(sc->xbd_dev));
785 			return (error);
786 		}
787 	} else {
788 		for (i = 0; i < sc->xbd_ring_pages; i++) {
789 			char ring_ref_name[]= "ring_refXX";
790 
791 			snprintf(ring_ref_name, sizeof(ring_ref_name),
792 			    "ring-ref%u", i);
793 			error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
794 			     ring_ref_name, "%u", sc->xbd_ring_ref[i]);
795 			if (error) {
796 				xenbus_dev_fatal(sc->xbd_dev, error,
797 				    "writing %s/%s",
798 				    xenbus_get_node(sc->xbd_dev),
799 				    ring_ref_name);
800 				return (error);
801 			}
802 		}
803 	}
804 
805 	error = xen_intr_alloc_and_bind_local_port(sc->xbd_dev,
806 	    xenbus_get_otherend_id(sc->xbd_dev), NULL, xbd_int, sc,
807 	    INTR_TYPE_BIO | INTR_MPSAFE, &sc->xen_intr_handle);
808 	if (error) {
809 		xenbus_dev_fatal(sc->xbd_dev, error,
810 		    "xen_intr_alloc_and_bind_local_port failed");
811 		return (error);
812 	}
813 
814 	return (0);
815 }
816 
817 static void
818 xbd_free_ring(struct xbd_softc *sc)
819 {
820 	int i;
821 
822 	if (sc->xbd_ring.sring == NULL)
823 		return;
824 
825 	for (i = 0; i < sc->xbd_ring_pages; i++) {
826 		if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) {
827 			gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]);
828 			sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
829 		}
830 	}
831 	free(sc->xbd_ring.sring, M_XENBLOCKFRONT);
832 	sc->xbd_ring.sring = NULL;
833 }
834 
835 /*-------------------------- Initialization/Teardown -------------------------*/
836 static int
837 xbd_feature_string(struct xbd_softc *sc, char *features, size_t len)
838 {
839 	struct sbuf sb;
840 	int feature_cnt;
841 
842 	sbuf_new(&sb, features, len, SBUF_FIXEDLEN);
843 
844 	feature_cnt = 0;
845 	if ((sc->xbd_flags & XBDF_FLUSH) != 0) {
846 		sbuf_printf(&sb, "flush");
847 		feature_cnt++;
848 	}
849 
850 	if ((sc->xbd_flags & XBDF_BARRIER) != 0) {
851 		if (feature_cnt != 0)
852 			sbuf_printf(&sb, ", ");
853 		sbuf_printf(&sb, "write_barrier");
854 		feature_cnt++;
855 	}
856 
857 	if ((sc->xbd_flags & XBDF_DISCARD) != 0) {
858 		if (feature_cnt != 0)
859 			sbuf_printf(&sb, ", ");
860 		sbuf_printf(&sb, "discard");
861 		feature_cnt++;
862 	}
863 
864 	if ((sc->xbd_flags & XBDF_PERSISTENT) != 0) {
865 		if (feature_cnt != 0)
866 			sbuf_printf(&sb, ", ");
867 		sbuf_printf(&sb, "persistent_grants");
868 		feature_cnt++;
869 	}
870 
871 	(void) sbuf_finish(&sb);
872 	return (sbuf_len(&sb));
873 }
874 
875 static int
876 xbd_sysctl_features(SYSCTL_HANDLER_ARGS)
877 {
878 	char features[80];
879 	struct xbd_softc *sc = arg1;
880 	int error;
881 	int len;
882 
883 	error = sysctl_wire_old_buffer(req, 0);
884 	if (error != 0)
885 		return (error);
886 
887 	len = xbd_feature_string(sc, features, sizeof(features));
888 
889 	/* len is -1 on error, which will make the SYSCTL_OUT a no-op. */
890 	return (SYSCTL_OUT(req, features, len + 1/*NUL*/));
891 }
892 
893 static void
894 xbd_setup_sysctl(struct xbd_softc *xbd)
895 {
896 	struct sysctl_ctx_list *sysctl_ctx = NULL;
897 	struct sysctl_oid *sysctl_tree = NULL;
898 	struct sysctl_oid_list *children;
899 
900 	sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev);
901 	if (sysctl_ctx == NULL)
902 		return;
903 
904 	sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev);
905 	if (sysctl_tree == NULL)
906 		return;
907 
908 	children = SYSCTL_CHILDREN(sysctl_tree);
909 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
910 	    "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1,
911 	    "maximum outstanding requests (negotiated)");
912 
913 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
914 	    "max_request_segments", CTLFLAG_RD,
915 	    &xbd->xbd_max_request_segments, 0,
916 	    "maximum number of pages per requests (negotiated)");
917 
918 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
919 	    "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0,
920 	    "maximum size in bytes of a request (negotiated)");
921 
922 	SYSCTL_ADD_UINT(sysctl_ctx, children, OID_AUTO,
923 	    "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0,
924 	    "communication channel pages (negotiated)");
925 
926 	SYSCTL_ADD_PROC(sysctl_ctx, children, OID_AUTO,
927 	    "features", CTLTYPE_STRING|CTLFLAG_RD, xbd, 0,
928 	    xbd_sysctl_features, "A", "protocol features (negotiated)");
929 }
930 
931 /*
932  * Translate Linux major/minor to an appropriate name and unit
933  * number. For HVM guests, this allows us to use the same drive names
934  * with blkfront as the emulated drives, easing transition slightly.
935  */
936 static void
937 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name)
938 {
939 	static struct vdev_info {
940 		int major;
941 		int shift;
942 		int base;
943 		const char *name;
944 	} info[] = {
945 		{3,	6,	0,	"ada"},	/* ide0 */
946 		{22,	6,	2,	"ada"},	/* ide1 */
947 		{33,	6,	4,	"ada"},	/* ide2 */
948 		{34,	6,	6,	"ada"},	/* ide3 */
949 		{56,	6,	8,	"ada"},	/* ide4 */
950 		{57,	6,	10,	"ada"},	/* ide5 */
951 		{88,	6,	12,	"ada"},	/* ide6 */
952 		{89,	6,	14,	"ada"},	/* ide7 */
953 		{90,	6,	16,	"ada"},	/* ide8 */
954 		{91,	6,	18,	"ada"},	/* ide9 */
955 
956 		{8,	4,	0,	"da"},	/* scsi disk0 */
957 		{65,	4,	16,	"da"},	/* scsi disk1 */
958 		{66,	4,	32,	"da"},	/* scsi disk2 */
959 		{67,	4,	48,	"da"},	/* scsi disk3 */
960 		{68,	4,	64,	"da"},	/* scsi disk4 */
961 		{69,	4,	80,	"da"},	/* scsi disk5 */
962 		{70,	4,	96,	"da"},	/* scsi disk6 */
963 		{71,	4,	112,	"da"},	/* scsi disk7 */
964 		{128,	4,	128,	"da"},	/* scsi disk8 */
965 		{129,	4,	144,	"da"},	/* scsi disk9 */
966 		{130,	4,	160,	"da"},	/* scsi disk10 */
967 		{131,	4,	176,	"da"},	/* scsi disk11 */
968 		{132,	4,	192,	"da"},	/* scsi disk12 */
969 		{133,	4,	208,	"da"},	/* scsi disk13 */
970 		{134,	4,	224,	"da"},	/* scsi disk14 */
971 		{135,	4,	240,	"da"},	/* scsi disk15 */
972 
973 		{202,	4,	0,	"xbd"},	/* xbd */
974 
975 		{0,	0,	0,	NULL},
976 	};
977 	int major = vdevice >> 8;
978 	int minor = vdevice & 0xff;
979 	int i;
980 
981 	if (vdevice & (1 << 28)) {
982 		*unit = (vdevice & ((1 << 28) - 1)) >> 8;
983 		*name = "xbd";
984 		return;
985 	}
986 
987 	for (i = 0; info[i].major; i++) {
988 		if (info[i].major == major) {
989 			*unit = info[i].base + (minor >> info[i].shift);
990 			*name = info[i].name;
991 			return;
992 		}
993 	}
994 
995 	*unit = minor >> 4;
996 	*name = "xbd";
997 }
998 
999 int
1000 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors,
1001     int vdevice, uint16_t vdisk_info, unsigned long sector_size,
1002     unsigned long phys_sector_size)
1003 {
1004 	char features[80];
1005 	int unit, error = 0;
1006 	const char *name;
1007 
1008 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1009 
1010 	sc->xbd_unit = unit;
1011 
1012 	if (strcmp(name, "xbd") != 0)
1013 		device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit);
1014 
1015 	if (xbd_feature_string(sc, features, sizeof(features)) > 0) {
1016 		device_printf(sc->xbd_dev, "features: %s\n",
1017 		    features);
1018 	}
1019 
1020 	sc->xbd_disk = disk_alloc();
1021 	sc->xbd_disk->d_unit = sc->xbd_unit;
1022 	sc->xbd_disk->d_open = xbd_open;
1023 	sc->xbd_disk->d_close = xbd_close;
1024 	sc->xbd_disk->d_ioctl = xbd_ioctl;
1025 	sc->xbd_disk->d_strategy = xbd_strategy;
1026 	sc->xbd_disk->d_dump = xbd_dump;
1027 	sc->xbd_disk->d_name = name;
1028 	sc->xbd_disk->d_drv1 = sc;
1029 	sc->xbd_disk->d_sectorsize = sector_size;
1030 	sc->xbd_disk->d_stripesize = phys_sector_size;
1031 	sc->xbd_disk->d_stripeoffset = 0;
1032 
1033 	sc->xbd_disk->d_mediasize = sectors * sector_size;
1034 	sc->xbd_disk->d_maxsize = sc->xbd_max_request_size;
1035 	sc->xbd_disk->d_flags = DISKFLAG_UNMAPPED_BIO;
1036 	if ((sc->xbd_flags & (XBDF_FLUSH|XBDF_BARRIER)) != 0) {
1037 		sc->xbd_disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
1038 		device_printf(sc->xbd_dev,
1039 		    "synchronize cache commands enabled.\n");
1040 	}
1041 	disk_create(sc->xbd_disk, DISK_VERSION);
1042 
1043 	return error;
1044 }
1045 
1046 static void
1047 xbd_free(struct xbd_softc *sc)
1048 {
1049 	int i;
1050 
1051 	/* Prevent new requests being issued until we fix things up. */
1052 	mtx_lock(&sc->xbd_io_lock);
1053 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1054 	mtx_unlock(&sc->xbd_io_lock);
1055 
1056 	/* Free resources associated with old device channel. */
1057 	xbd_free_ring(sc);
1058 	if (sc->xbd_shadow) {
1059 
1060 		for (i = 0; i < sc->xbd_max_requests; i++) {
1061 			struct xbd_command *cm;
1062 
1063 			cm = &sc->xbd_shadow[i];
1064 			if (cm->cm_sg_refs != NULL) {
1065 				free(cm->cm_sg_refs, M_XENBLOCKFRONT);
1066 				cm->cm_sg_refs = NULL;
1067 			}
1068 
1069 			if (cm->cm_indirectionpages != NULL) {
1070 				gnttab_end_foreign_access_references(
1071 				    sc->xbd_max_request_indirectpages,
1072 				    &cm->cm_indirectionrefs[0]);
1073 				contigfree(cm->cm_indirectionpages, PAGE_SIZE *
1074 				    sc->xbd_max_request_indirectpages,
1075 				    M_XENBLOCKFRONT);
1076 				cm->cm_indirectionpages = NULL;
1077 			}
1078 
1079 			bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map);
1080 		}
1081 		free(sc->xbd_shadow, M_XENBLOCKFRONT);
1082 		sc->xbd_shadow = NULL;
1083 
1084 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1085 
1086 		xbd_initq_cm(sc, XBD_Q_FREE);
1087 		xbd_initq_cm(sc, XBD_Q_READY);
1088 		xbd_initq_cm(sc, XBD_Q_COMPLETE);
1089 	}
1090 
1091 	xen_intr_unbind(&sc->xen_intr_handle);
1092 
1093 }
1094 
1095 /*--------------------------- State Change Handlers --------------------------*/
1096 static void
1097 xbd_initialize(struct xbd_softc *sc)
1098 {
1099 	const char *otherend_path;
1100 	const char *node_path;
1101 	uint32_t max_ring_page_order;
1102 	int error;
1103 
1104 	if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) {
1105 		/* Initialization has already been performed. */
1106 		return;
1107 	}
1108 
1109 	/*
1110 	 * Protocol defaults valid even if negotiation for a
1111 	 * setting fails.
1112 	 */
1113 	max_ring_page_order = 0;
1114 	sc->xbd_ring_pages = 1;
1115 
1116 	/*
1117 	 * Protocol negotiation.
1118 	 *
1119 	 * \note xs_gather() returns on the first encountered error, so
1120 	 *       we must use independent calls in order to guarantee
1121 	 *       we don't miss information in a sparsly populated back-end
1122 	 *       tree.
1123 	 *
1124 	 * \note xs_scanf() does not update variables for unmatched
1125 	 *	 fields.
1126 	 */
1127 	otherend_path = xenbus_get_otherend_path(sc->xbd_dev);
1128 	node_path = xenbus_get_node(sc->xbd_dev);
1129 
1130 	/* Support both backend schemes for relaying ring page limits. */
1131 	(void)xs_scanf(XST_NIL, otherend_path,
1132 	    "max-ring-page-order", NULL, "%" PRIu32,
1133 	    &max_ring_page_order);
1134 	sc->xbd_ring_pages = 1 << max_ring_page_order;
1135 	(void)xs_scanf(XST_NIL, otherend_path,
1136 	    "max-ring-pages", NULL, "%" PRIu32,
1137 	    &sc->xbd_ring_pages);
1138 	if (sc->xbd_ring_pages < 1)
1139 		sc->xbd_ring_pages = 1;
1140 
1141 	if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) {
1142 		device_printf(sc->xbd_dev,
1143 		    "Back-end specified ring-pages of %u "
1144 		    "limited to front-end limit of %u.\n",
1145 		    sc->xbd_ring_pages, XBD_MAX_RING_PAGES);
1146 		sc->xbd_ring_pages = XBD_MAX_RING_PAGES;
1147 	}
1148 
1149 	if (powerof2(sc->xbd_ring_pages) == 0) {
1150 		uint32_t new_page_limit;
1151 
1152 		new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1);
1153 		device_printf(sc->xbd_dev,
1154 		    "Back-end specified ring-pages of %u "
1155 		    "is not a power of 2. Limited to %u.\n",
1156 		    sc->xbd_ring_pages, new_page_limit);
1157 		sc->xbd_ring_pages = new_page_limit;
1158 	}
1159 
1160 	sc->xbd_max_requests =
1161 	    BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE);
1162 	if (sc->xbd_max_requests > XBD_MAX_REQUESTS) {
1163 		device_printf(sc->xbd_dev,
1164 		    "Back-end specified max_requests of %u "
1165 		    "limited to front-end limit of %zu.\n",
1166 		    sc->xbd_max_requests, XBD_MAX_REQUESTS);
1167 		sc->xbd_max_requests = XBD_MAX_REQUESTS;
1168 	}
1169 
1170 	if (xbd_alloc_ring(sc) != 0)
1171 		return;
1172 
1173 	/* Support both backend schemes for relaying ring page limits. */
1174 	if (sc->xbd_ring_pages > 1) {
1175 		error = xs_printf(XST_NIL, node_path,
1176 		    "num-ring-pages","%u",
1177 		    sc->xbd_ring_pages);
1178 		if (error) {
1179 			xenbus_dev_fatal(sc->xbd_dev, error,
1180 			    "writing %s/num-ring-pages",
1181 			    node_path);
1182 			return;
1183 		}
1184 
1185 		error = xs_printf(XST_NIL, node_path,
1186 		    "ring-page-order", "%u",
1187 		    fls(sc->xbd_ring_pages) - 1);
1188 		if (error) {
1189 			xenbus_dev_fatal(sc->xbd_dev, error,
1190 			    "writing %s/ring-page-order",
1191 			    node_path);
1192 			return;
1193 		}
1194 	}
1195 
1196 	error = xs_printf(XST_NIL, node_path, "event-channel",
1197 	    "%u", xen_intr_port(sc->xen_intr_handle));
1198 	if (error) {
1199 		xenbus_dev_fatal(sc->xbd_dev, error,
1200 		    "writing %s/event-channel",
1201 		    node_path);
1202 		return;
1203 	}
1204 
1205 	error = xs_printf(XST_NIL, node_path, "protocol",
1206 	    "%s", XEN_IO_PROTO_ABI_NATIVE);
1207 	if (error) {
1208 		xenbus_dev_fatal(sc->xbd_dev, error,
1209 		    "writing %s/protocol",
1210 		    node_path);
1211 		return;
1212 	}
1213 
1214 	xenbus_set_state(sc->xbd_dev, XenbusStateInitialised);
1215 }
1216 
1217 /*
1218  * Invoked when the backend is finally 'ready' (and has published
1219  * the details about the physical device - #sectors, size, etc).
1220  */
1221 static void
1222 xbd_connect(struct xbd_softc *sc)
1223 {
1224 	device_t dev = sc->xbd_dev;
1225 	unsigned long sectors, sector_size, phys_sector_size;
1226 	unsigned int binfo;
1227 	int err, feature_barrier, feature_flush;
1228 	int i, j;
1229 
1230 	if (sc->xbd_state == XBD_STATE_CONNECTED ||
1231 	    sc->xbd_state == XBD_STATE_SUSPENDED)
1232 		return;
1233 
1234 	DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
1235 
1236 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1237 	    "sectors", "%lu", &sectors,
1238 	    "info", "%u", &binfo,
1239 	    "sector-size", "%lu", &sector_size,
1240 	    NULL);
1241 	if (err) {
1242 		xenbus_dev_fatal(dev, err,
1243 		    "reading backend fields at %s",
1244 		    xenbus_get_otherend_path(dev));
1245 		return;
1246 	}
1247 	if ((sectors == 0) || (sector_size == 0)) {
1248 		xenbus_dev_fatal(dev, 0,
1249 		    "invalid parameters from %s:"
1250 		    " sectors = %lu, sector_size = %lu",
1251 		    xenbus_get_otherend_path(dev),
1252 		    sectors, sector_size);
1253 		return;
1254 	}
1255 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1256 	     "physical-sector-size", "%lu", &phys_sector_size,
1257 	     NULL);
1258 	if (err || phys_sector_size <= sector_size)
1259 		phys_sector_size = 0;
1260 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1261 	     "feature-barrier", "%d", &feature_barrier,
1262 	     NULL);
1263 	if (err == 0 && feature_barrier != 0)
1264 		sc->xbd_flags |= XBDF_BARRIER;
1265 
1266 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1267 	     "feature-flush-cache", "%d", &feature_flush,
1268 	     NULL);
1269 	if (err == 0 && feature_flush != 0)
1270 		sc->xbd_flags |= XBDF_FLUSH;
1271 
1272 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1273 	    "feature-max-indirect-segments", "%" PRIu32,
1274 	    &sc->xbd_max_request_segments, NULL);
1275 	if ((err != 0) || (xbd_enable_indirect == 0))
1276 		sc->xbd_max_request_segments = 0;
1277 	if (sc->xbd_max_request_segments > XBD_MAX_INDIRECT_SEGMENTS)
1278 		sc->xbd_max_request_segments = XBD_MAX_INDIRECT_SEGMENTS;
1279 	if (sc->xbd_max_request_segments > XBD_SIZE_TO_SEGS(MAXPHYS))
1280 		sc->xbd_max_request_segments = XBD_SIZE_TO_SEGS(MAXPHYS);
1281 	sc->xbd_max_request_indirectpages =
1282 	    XBD_INDIRECT_SEGS_TO_PAGES(sc->xbd_max_request_segments);
1283 	if (sc->xbd_max_request_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
1284 		sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1285 	sc->xbd_max_request_size =
1286 	    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments);
1287 
1288 	/* Allocate datastructures based on negotiated values. */
1289 	err = bus_dma_tag_create(
1290 	    bus_get_dma_tag(sc->xbd_dev),	/* parent */
1291 	    512, PAGE_SIZE,			/* algnmnt, boundary */
1292 	    BUS_SPACE_MAXADDR,			/* lowaddr */
1293 	    BUS_SPACE_MAXADDR,			/* highaddr */
1294 	    NULL, NULL,				/* filter, filterarg */
1295 	    sc->xbd_max_request_size,
1296 	    sc->xbd_max_request_segments,
1297 	    PAGE_SIZE,				/* maxsegsize */
1298 	    BUS_DMA_ALLOCNOW,			/* flags */
1299 	    busdma_lock_mutex,			/* lockfunc */
1300 	    &sc->xbd_io_lock,			/* lockarg */
1301 	    &sc->xbd_io_dmat);
1302 	if (err != 0) {
1303 		xenbus_dev_fatal(sc->xbd_dev, err,
1304 		    "Cannot allocate parent DMA tag\n");
1305 		return;
1306 	}
1307 
1308 	/* Per-transaction data allocation. */
1309 	sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests,
1310 	    M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
1311 	if (sc->xbd_shadow == NULL) {
1312 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1313 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM,
1314 		    "Cannot allocate request structures\n");
1315 		return;
1316 	}
1317 
1318 	for (i = 0; i < sc->xbd_max_requests; i++) {
1319 		struct xbd_command *cm;
1320 		void * indirectpages;
1321 
1322 		cm = &sc->xbd_shadow[i];
1323 		cm->cm_sg_refs = malloc(
1324 		    sizeof(grant_ref_t) * sc->xbd_max_request_segments,
1325 		    M_XENBLOCKFRONT, M_NOWAIT);
1326 		if (cm->cm_sg_refs == NULL)
1327 			break;
1328 		cm->cm_id = i;
1329 		cm->cm_flags = XBDCF_INITIALIZER;
1330 		cm->cm_sc = sc;
1331 		if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0)
1332 			break;
1333 		if (sc->xbd_max_request_indirectpages > 0) {
1334 			indirectpages = contigmalloc(
1335 			    PAGE_SIZE * sc->xbd_max_request_indirectpages,
1336 			    M_XENBLOCKFRONT, M_ZERO, 0, ~0, PAGE_SIZE, 0);
1337 		} else {
1338 			indirectpages = NULL;
1339 		}
1340 		for (j = 0; j < sc->xbd_max_request_indirectpages; j++) {
1341 			if (gnttab_grant_foreign_access(
1342 			    xenbus_get_otherend_id(sc->xbd_dev),
1343 			    (vtophys(indirectpages) >> PAGE_SHIFT) + j,
1344 			    1 /* grant read-only access */,
1345 			    &cm->cm_indirectionrefs[j]))
1346 				break;
1347 		}
1348 		if (j < sc->xbd_max_request_indirectpages)
1349 			break;
1350 		cm->cm_indirectionpages = indirectpages;
1351 		xbd_free_command(cm);
1352 	}
1353 
1354 	if (sc->xbd_disk == NULL) {
1355 		device_printf(dev, "%juMB <%s> at %s",
1356 		    (uintmax_t) sectors / (1048576 / sector_size),
1357 		    device_get_desc(dev),
1358 		    xenbus_get_node(dev));
1359 		bus_print_child_footer(device_get_parent(dev), dev);
1360 
1361 		xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo,
1362 		    sector_size, phys_sector_size);
1363 	}
1364 
1365 	(void)xenbus_set_state(dev, XenbusStateConnected);
1366 
1367 	/* Kick pending requests. */
1368 	mtx_lock(&sc->xbd_io_lock);
1369 	sc->xbd_state = XBD_STATE_CONNECTED;
1370 	xbd_startio(sc);
1371 	sc->xbd_flags |= XBDF_READY;
1372 	mtx_unlock(&sc->xbd_io_lock);
1373 }
1374 
1375 /**
1376  * Handle the change of state of the backend to Closing.  We must delete our
1377  * device-layer structures now, to ensure that writes are flushed through to
1378  * the backend.  Once this is done, we can switch to Closed in
1379  * acknowledgement.
1380  */
1381 static void
1382 xbd_closing(device_t dev)
1383 {
1384 	struct xbd_softc *sc = device_get_softc(dev);
1385 
1386 	xenbus_set_state(dev, XenbusStateClosing);
1387 
1388 	DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev));
1389 
1390 	if (sc->xbd_disk != NULL) {
1391 		disk_destroy(sc->xbd_disk);
1392 		sc->xbd_disk = NULL;
1393 	}
1394 
1395 	xenbus_set_state(dev, XenbusStateClosed);
1396 }
1397 
1398 /*---------------------------- NewBus Entrypoints ----------------------------*/
1399 static int
1400 xbd_probe(device_t dev)
1401 {
1402 	if (strcmp(xenbus_get_type(dev), "vbd") != 0)
1403 		return (ENXIO);
1404 
1405 	if (xen_hvm_domain() && xen_disable_pv_disks != 0)
1406 		return (ENXIO);
1407 
1408 	if (xen_hvm_domain()) {
1409 		int error;
1410 		char *type;
1411 
1412 		/*
1413 		 * When running in an HVM domain, IDE disk emulation is
1414 		 * disabled early in boot so that native drivers will
1415 		 * not see emulated hardware.  However, CDROM device
1416 		 * emulation cannot be disabled.
1417 		 *
1418 		 * Through use of FreeBSD's vm_guest and xen_hvm_domain()
1419 		 * APIs, we could modify the native CDROM driver to fail its
1420 		 * probe when running under Xen.  Unfortunatlely, the PV
1421 		 * CDROM support in XenServer (up through at least version
1422 		 * 6.2) isn't functional, so we instead rely on the emulated
1423 		 * CDROM instance, and fail to attach the PV one here in
1424 		 * the blkfront driver.
1425 		 */
1426 		error = xs_read(XST_NIL, xenbus_get_node(dev),
1427 		    "device-type", NULL, (void **) &type);
1428 		if (error)
1429 			return (ENXIO);
1430 
1431 		if (strncmp(type, "cdrom", 5) == 0) {
1432 			free(type, M_XENSTORE);
1433 			return (ENXIO);
1434 		}
1435 		free(type, M_XENSTORE);
1436 	}
1437 
1438 	device_set_desc(dev, "Virtual Block Device");
1439 	device_quiet(dev);
1440 	return (0);
1441 }
1442 
1443 /*
1444  * Setup supplies the backend dir, virtual device.  We place an event
1445  * channel and shared frame entries.  We watch backend to wait if it's
1446  * ok.
1447  */
1448 static int
1449 xbd_attach(device_t dev)
1450 {
1451 	struct xbd_softc *sc;
1452 	const char *name;
1453 	uint32_t vdevice;
1454 	int error;
1455 	int i;
1456 	int unit;
1457 
1458 	/* FIXME: Use dynamic device id if this is not set. */
1459 	error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1460 	    "virtual-device", NULL, "%" PRIu32, &vdevice);
1461 	if (error)
1462 		error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1463 		    "virtual-device-ext", NULL, "%" PRIu32, &vdevice);
1464 	if (error) {
1465 		xenbus_dev_fatal(dev, error, "reading virtual-device");
1466 		device_printf(dev, "Couldn't determine virtual device.\n");
1467 		return (error);
1468 	}
1469 
1470 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1471 	if (!strcmp(name, "xbd"))
1472 		device_set_unit(dev, unit);
1473 
1474 	sc = device_get_softc(dev);
1475 	mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
1476 	xbd_initqs(sc);
1477 	for (i = 0; i < XBD_MAX_RING_PAGES; i++)
1478 		sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
1479 
1480 	sc->xbd_dev = dev;
1481 	sc->xbd_vdevice = vdevice;
1482 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1483 
1484 	xbd_setup_sysctl(sc);
1485 
1486 	/* Wait for backend device to publish its protocol capabilities. */
1487 	xenbus_set_state(dev, XenbusStateInitialising);
1488 
1489 	return (0);
1490 }
1491 
1492 static int
1493 xbd_detach(device_t dev)
1494 {
1495 	struct xbd_softc *sc = device_get_softc(dev);
1496 
1497 	DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev));
1498 
1499 	xbd_free(sc);
1500 	mtx_destroy(&sc->xbd_io_lock);
1501 
1502 	return 0;
1503 }
1504 
1505 static int
1506 xbd_suspend(device_t dev)
1507 {
1508 	struct xbd_softc *sc = device_get_softc(dev);
1509 	int retval;
1510 	int saved_state;
1511 
1512 	/* Prevent new requests being issued until we fix things up. */
1513 	mtx_lock(&sc->xbd_io_lock);
1514 	saved_state = sc->xbd_state;
1515 	sc->xbd_state = XBD_STATE_SUSPENDED;
1516 
1517 	/* Wait for outstanding I/O to drain. */
1518 	retval = 0;
1519 	while (xbd_queue_length(sc, XBD_Q_BUSY) != 0) {
1520 		if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock,
1521 		    PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
1522 			retval = EBUSY;
1523 			break;
1524 		}
1525 	}
1526 	mtx_unlock(&sc->xbd_io_lock);
1527 
1528 	if (retval != 0)
1529 		sc->xbd_state = saved_state;
1530 
1531 	return (retval);
1532 }
1533 
1534 static int
1535 xbd_resume(device_t dev)
1536 {
1537 	struct xbd_softc *sc = device_get_softc(dev);
1538 
1539 	if (xen_suspend_cancelled) {
1540 		sc->xbd_state = XBD_STATE_CONNECTED;
1541 		return (0);
1542 	}
1543 
1544 	DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev));
1545 
1546 	xbd_free(sc);
1547 	xbd_initialize(sc);
1548 	return (0);
1549 }
1550 
1551 /**
1552  * Callback received when the backend's state changes.
1553  */
1554 static void
1555 xbd_backend_changed(device_t dev, XenbusState backend_state)
1556 {
1557 	struct xbd_softc *sc = device_get_softc(dev);
1558 
1559 	DPRINTK("backend_state=%d\n", backend_state);
1560 
1561 	switch (backend_state) {
1562 	case XenbusStateUnknown:
1563 	case XenbusStateInitialising:
1564 	case XenbusStateReconfigured:
1565 	case XenbusStateReconfiguring:
1566 	case XenbusStateClosed:
1567 		break;
1568 
1569 	case XenbusStateInitWait:
1570 	case XenbusStateInitialised:
1571 		xbd_initialize(sc);
1572 		break;
1573 
1574 	case XenbusStateConnected:
1575 		xbd_initialize(sc);
1576 		xbd_connect(sc);
1577 		break;
1578 
1579 	case XenbusStateClosing:
1580 		if (sc->xbd_users > 0) {
1581 			device_printf(dev, "detaching with pending users\n");
1582 			KASSERT(sc->xbd_disk != NULL,
1583 			    ("NULL disk with pending users\n"));
1584 			disk_gone(sc->xbd_disk);
1585 		} else {
1586 			xbd_closing(dev);
1587 		}
1588 		break;
1589 	}
1590 }
1591 
1592 /*---------------------------- NewBus Registration ---------------------------*/
1593 static device_method_t xbd_methods[] = {
1594 	/* Device interface */
1595 	DEVMETHOD(device_probe,         xbd_probe),
1596 	DEVMETHOD(device_attach,        xbd_attach),
1597 	DEVMETHOD(device_detach,        xbd_detach),
1598 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
1599 	DEVMETHOD(device_suspend,       xbd_suspend),
1600 	DEVMETHOD(device_resume,        xbd_resume),
1601 
1602 	/* Xenbus interface */
1603 	DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed),
1604 
1605 	{ 0, 0 }
1606 };
1607 
1608 static driver_t xbd_driver = {
1609 	"xbd",
1610 	xbd_methods,
1611 	sizeof(struct xbd_softc),
1612 };
1613 devclass_t xbd_devclass;
1614 
1615 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, xbd_devclass, 0, 0);
1616