xref: /freebsd/sys/dev/xen/blkfront/blkfront.c (revision 3b1f7d9e5d6f44b50ff07fde6fd0e1135f213762)
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 #include <sys/bus_dma.h>
53 
54 #include <machine/_inttypes.h>
55 #include <machine/xen/xen-os.h>
56 #include <machine/xen/xenvar.h>
57 #include <machine/xen/xenfunc.h>
58 
59 #include <xen/hypervisor.h>
60 #include <xen/xen_intr.h>
61 #include <xen/evtchn.h>
62 #include <xen/gnttab.h>
63 #include <xen/interface/grant_table.h>
64 #include <xen/interface/io/protocols.h>
65 #include <xen/xenbus/xenbusvar.h>
66 
67 #include <geom/geom_disk.h>
68 
69 #include <dev/xen/blkfront/block.h>
70 
71 #include "xenbus_if.h"
72 
73 /*--------------------------- Forward Declarations ---------------------------*/
74 static void xbd_closing(device_t);
75 static void xbd_startio(struct xbd_softc *sc);
76 
77 /*---------------------------------- Macros ----------------------------------*/
78 #if 0
79 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
80 #else
81 #define DPRINTK(fmt, args...)
82 #endif
83 
84 #define XBD_SECTOR_SHFT		9
85 
86 /*---------------------------- Global Static Data ----------------------------*/
87 static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
88 
89 /*---------------------------- Command Processing ----------------------------*/
90 static void
91 xbd_freeze(struct xbd_softc *sc, xbd_flag_t xbd_flag)
92 {
93 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) != 0)
94 		return;
95 
96 	sc->xbd_flags |= xbd_flag;
97 	sc->xbd_qfrozen_cnt++;
98 }
99 
100 static void
101 xbd_thaw(struct xbd_softc *sc, xbd_flag_t xbd_flag)
102 {
103 	if (xbd_flag != XBDF_NONE && (sc->xbd_flags & xbd_flag) == 0)
104 		return;
105 
106 	if (sc->xbd_qfrozen_cnt == 0)
107 		panic("%s: Thaw with flag 0x%x while not frozen.",
108 		    __func__, xbd_flag);
109 
110 	sc->xbd_flags &= ~xbd_flag;
111 	sc->xbd_qfrozen_cnt--;
112 }
113 
114 static inline void
115 xbd_flush_requests(struct xbd_softc *sc)
116 {
117 	int notify;
118 
119 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->xbd_ring, notify);
120 
121 	if (notify)
122 		notify_remote_via_irq(sc->xbd_irq);
123 }
124 
125 static void
126 xbd_free_command(struct xbd_command *cm)
127 {
128 
129 	KASSERT((cm->cm_flags & XBDCF_Q_MASK) == XBD_Q_NONE,
130 	    ("Freeing command that is still on queue %d.",
131 	    cm->cm_flags & XBDCF_Q_MASK));
132 
133 	cm->cm_flags = XBDCF_INITIALIZER;
134 	cm->cm_bp = NULL;
135 	cm->cm_complete = NULL;
136 	xbd_enqueue_cm(cm, XBD_Q_FREE);
137 	xbd_thaw(cm->cm_sc, XBDF_CM_SHORTAGE);
138 }
139 
140 static void
141 xbd_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
142 {
143 	struct xbd_softc *sc;
144 	struct xbd_command *cm;
145 	blkif_request_t	*ring_req;
146 	struct blkif_request_segment *sg;
147 	struct blkif_request_segment *last_block_sg;
148 	grant_ref_t *sg_ref;
149 	vm_paddr_t buffer_ma;
150 	uint64_t fsect, lsect;
151 	int ref;
152 	int op;
153 	int block_segs;
154 
155 	cm = arg;
156 	sc = cm->cm_sc;
157 
158 	if (error) {
159 		printf("error %d in xbd_queue_cb\n", error);
160 		cm->cm_bp->bio_error = EIO;
161 		biodone(cm->cm_bp);
162 		xbd_free_command(cm);
163 		return;
164 	}
165 
166 	/* Fill out a communications ring structure. */
167 	ring_req = RING_GET_REQUEST(&sc->xbd_ring, sc->xbd_ring.req_prod_pvt);
168 	sc->xbd_ring.req_prod_pvt++;
169 	ring_req->id = cm->cm_id;
170 	ring_req->operation = cm->cm_operation;
171 	ring_req->sector_number = cm->cm_sector_number;
172 	ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xbd_disk;
173 	ring_req->nr_segments = nsegs;
174 	cm->cm_nseg = nsegs;
175 
176 	block_segs    = MIN(nsegs, BLKIF_MAX_SEGMENTS_PER_HEADER_BLOCK);
177 	sg            = ring_req->seg;
178 	last_block_sg = sg + block_segs;
179 	sg_ref        = cm->cm_sg_refs;
180 
181 	while (1) {
182 
183 		while (sg < last_block_sg) {
184 			buffer_ma = segs->ds_addr;
185 			fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
186 			lsect = fsect + (segs->ds_len  >> XBD_SECTOR_SHFT) - 1;
187 
188 			KASSERT(lsect <= 7, ("XEN disk driver data cannot "
189 			    "cross a page boundary"));
190 
191 			/* install a grant reference. */
192 			ref = gnttab_claim_grant_reference(&cm->cm_gref_head);
193 
194 			/*
195 			 * GNTTAB_LIST_END == 0xffffffff, but it is private
196 			 * to gnttab.c.
197 			 */
198 			KASSERT(ref != ~0, ("grant_reference failed"));
199 
200 			gnttab_grant_foreign_access_ref(
201 			    ref,
202 			    xenbus_get_otherend_id(sc->xbd_dev),
203 			    buffer_ma >> PAGE_SHIFT,
204 			    ring_req->operation == BLKIF_OP_WRITE);
205 
206 			*sg_ref = ref;
207 			*sg = (struct blkif_request_segment) {
208 				.gref       = ref,
209 				.first_sect = fsect,
210 				.last_sect  = lsect
211 			};
212 			sg++;
213 			sg_ref++;
214 			segs++;
215 			nsegs--;
216 		}
217 		block_segs = MIN(nsegs, BLKIF_MAX_SEGMENTS_PER_SEGMENT_BLOCK);
218 		if (block_segs == 0)
219 			break;
220 
221 		sg = BLKRING_GET_SEG_BLOCK(&sc->xbd_ring,
222 		    sc->xbd_ring.req_prod_pvt);
223 		sc->xbd_ring.req_prod_pvt++;
224 		last_block_sg = sg + block_segs;
225 	}
226 
227 	if (cm->cm_operation == BLKIF_OP_READ)
228 		op = BUS_DMASYNC_PREREAD;
229 	else if (cm->cm_operation == BLKIF_OP_WRITE)
230 		op = BUS_DMASYNC_PREWRITE;
231 	else
232 		op = 0;
233 	bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
234 
235 	gnttab_free_grant_references(cm->cm_gref_head);
236 
237 	xbd_enqueue_cm(cm, XBD_Q_BUSY);
238 
239 	/*
240 	 * If bus dma had to asynchronously call us back to dispatch
241 	 * this command, we are no longer executing in the context of
242 	 * xbd_startio().  Thus we cannot rely on xbd_startio()'s call to
243 	 * xbd_flush_requests() to publish this command to the backend
244 	 * along with any other commands that it could batch.
245 	 */
246 	if ((cm->cm_flags & XBDCF_ASYNC_MAPPING) != 0)
247 		xbd_flush_requests(sc);
248 
249 	return;
250 }
251 
252 static int
253 xbd_queue_request(struct xbd_softc *sc, struct xbd_command *cm)
254 {
255 	int error;
256 
257 	error = bus_dmamap_load(sc->xbd_io_dmat, cm->cm_map, cm->cm_data,
258 	    cm->cm_datalen, xbd_queue_cb, cm, 0);
259 	if (error == EINPROGRESS) {
260 		/*
261 		 * Maintain queuing order by freezing the queue.  The next
262 		 * command may not require as many resources as the command
263 		 * we just attempted to map, so we can't rely on bus dma
264 		 * blocking for it too.
265 		 */
266 		xbd_freeze(sc, XBDF_NONE);
267 		cm->cm_flags |= XBDCF_FROZEN|XBDCF_ASYNC_MAPPING;
268 		return (0);
269 	}
270 
271 	return (error);
272 }
273 
274 static void
275 xbd_restart_queue_callback(void *arg)
276 {
277 	struct xbd_softc *sc = arg;
278 
279 	mtx_lock(&sc->xbd_io_lock);
280 
281 	xbd_thaw(sc, XBDF_GNT_SHORTAGE);
282 
283 	xbd_startio(sc);
284 
285 	mtx_unlock(&sc->xbd_io_lock);
286 }
287 
288 static struct xbd_command *
289 xbd_bio_command(struct xbd_softc *sc)
290 {
291 	struct xbd_command *cm;
292 	struct bio *bp;
293 
294 	if (unlikely(sc->xbd_state != XBD_STATE_CONNECTED))
295 		return (NULL);
296 
297 	bp = xbd_dequeue_bio(sc);
298 	if (bp == NULL)
299 		return (NULL);
300 
301 	if ((cm = xbd_dequeue_cm(sc, XBD_Q_FREE)) == NULL) {
302 		xbd_freeze(sc, XBDF_CM_SHORTAGE);
303 		xbd_requeue_bio(sc, bp);
304 		return (NULL);
305 	}
306 
307 	if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
308 	    &cm->cm_gref_head) != 0) {
309 		gnttab_request_free_callback(&sc->xbd_callback,
310 		    xbd_restart_queue_callback, sc,
311 		    sc->xbd_max_request_segments);
312 		xbd_freeze(sc, XBDF_GNT_SHORTAGE);
313 		xbd_requeue_bio(sc, bp);
314 		xbd_enqueue_cm(cm, XBD_Q_FREE);
315 		return (NULL);
316 	}
317 
318 	cm->cm_bp = bp;
319 	cm->cm_data = bp->bio_data;
320 	cm->cm_datalen = bp->bio_bcount;
321 	cm->cm_operation = (bp->bio_cmd == BIO_READ) ?
322 	    BLKIF_OP_READ : BLKIF_OP_WRITE;
323 	cm->cm_sector_number = (blkif_sector_t)bp->bio_pblkno;
324 
325 	return (cm);
326 }
327 
328 /*
329  * Dequeue buffers and place them in the shared communication ring.
330  * Return when no more requests can be accepted or all buffers have
331  * been queued.
332  *
333  * Signal XEN once the ring has been filled out.
334  */
335 static void
336 xbd_startio(struct xbd_softc *sc)
337 {
338 	struct xbd_command *cm;
339 	int error, queued = 0;
340 
341 	mtx_assert(&sc->xbd_io_lock, MA_OWNED);
342 
343 	if (sc->xbd_state != XBD_STATE_CONNECTED)
344 		return;
345 
346 	while (RING_FREE_REQUESTS(&sc->xbd_ring) >=
347 	    sc->xbd_max_request_blocks) {
348 		if (sc->xbd_qfrozen_cnt != 0)
349 			break;
350 
351 		cm = xbd_dequeue_cm(sc, XBD_Q_READY);
352 
353 		if (cm == NULL)
354 		    cm = xbd_bio_command(sc);
355 
356 		if (cm == NULL)
357 			break;
358 
359 		if ((error = xbd_queue_request(sc, cm)) != 0) {
360 			printf("xbd_queue_request returned %d\n", error);
361 			break;
362 		}
363 		queued++;
364 	}
365 
366 	if (queued != 0)
367 		xbd_flush_requests(sc);
368 }
369 
370 static void
371 xbd_bio_complete(struct xbd_softc *sc, struct xbd_command *cm)
372 {
373 	struct bio *bp;
374 
375 	bp = cm->cm_bp;
376 
377 	if (unlikely(cm->cm_status != BLKIF_RSP_OKAY)) {
378 		disk_err(bp, "disk error" , -1, 0);
379 		printf(" status: %x\n", cm->cm_status);
380 		bp->bio_flags |= BIO_ERROR;
381 	}
382 
383 	if (bp->bio_flags & BIO_ERROR)
384 		bp->bio_error = EIO;
385 	else
386 		bp->bio_resid = 0;
387 
388 	xbd_free_command(cm);
389 	biodone(bp);
390 }
391 
392 static int
393 xbd_completion(struct xbd_command *cm)
394 {
395 	gnttab_end_foreign_access_references(cm->cm_nseg, cm->cm_sg_refs);
396 	return (BLKIF_SEGS_TO_BLOCKS(cm->cm_nseg));
397 }
398 
399 static void
400 xbd_int(void *xsc)
401 {
402 	struct xbd_softc *sc = xsc;
403 	struct xbd_command *cm;
404 	blkif_response_t *bret;
405 	RING_IDX i, rp;
406 	int op;
407 
408 	mtx_lock(&sc->xbd_io_lock);
409 
410 	if (unlikely(sc->xbd_state == XBD_STATE_DISCONNECTED)) {
411 		mtx_unlock(&sc->xbd_io_lock);
412 		return;
413 	}
414 
415  again:
416 	rp = sc->xbd_ring.sring->rsp_prod;
417 	rmb(); /* Ensure we see queued responses up to 'rp'. */
418 
419 	for (i = sc->xbd_ring.rsp_cons; i != rp;) {
420 		bret = RING_GET_RESPONSE(&sc->xbd_ring, i);
421 		cm   = &sc->xbd_shadow[bret->id];
422 
423 		xbd_remove_cm(cm, XBD_Q_BUSY);
424 		i += xbd_completion(cm);
425 
426 		if (cm->cm_operation == BLKIF_OP_READ)
427 			op = BUS_DMASYNC_POSTREAD;
428 		else if (cm->cm_operation == BLKIF_OP_WRITE)
429 			op = BUS_DMASYNC_POSTWRITE;
430 		else
431 			op = 0;
432 		bus_dmamap_sync(sc->xbd_io_dmat, cm->cm_map, op);
433 		bus_dmamap_unload(sc->xbd_io_dmat, cm->cm_map);
434 
435 		/*
436 		 * Release any hold this command has on future command
437 		 * dispatch.
438 		 */
439 		if ((cm->cm_flags & XBDCF_FROZEN) != 0) {
440 			xbd_thaw(sc, XBDF_NONE);
441 			cm->cm_flags &= ~XBDCF_FROZEN;
442 		}
443 
444 		/*
445 		 * Directly call the i/o complete routine to save an
446 		 * an indirection in the common case.
447 		 */
448 		cm->cm_status = bret->status;
449 		if (cm->cm_bp)
450 			xbd_bio_complete(sc, cm);
451 		else if (cm->cm_complete != NULL)
452 			cm->cm_complete(cm);
453 		else
454 			xbd_free_command(cm);
455 	}
456 
457 	sc->xbd_ring.rsp_cons = i;
458 
459 	if (i != sc->xbd_ring.req_prod_pvt) {
460 		int more_to_do;
461 		RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, more_to_do);
462 		if (more_to_do)
463 			goto again;
464 	} else {
465 		sc->xbd_ring.sring->rsp_event = i + 1;
466 	}
467 
468 	xbd_startio(sc);
469 
470 	if (unlikely(sc->xbd_state == XBD_STATE_SUSPENDED))
471 		wakeup(&sc->xbd_cm_q[XBD_Q_BUSY]);
472 
473 	mtx_unlock(&sc->xbd_io_lock);
474 }
475 
476 /*------------------------------- Dump Support -------------------------------*/
477 /**
478  * Quiesce the disk writes for a dump file before allowing the next buffer.
479  */
480 static void
481 xbd_quiesce(struct xbd_softc *sc)
482 {
483 	int mtd;
484 
485 	// While there are outstanding requests
486 	while (!TAILQ_EMPTY(&sc->xbd_cm_q[XBD_Q_BUSY].q_tailq)) {
487 		RING_FINAL_CHECK_FOR_RESPONSES(&sc->xbd_ring, mtd);
488 		if (mtd) {
489 			/* Recieved request completions, update queue. */
490 			xbd_int(sc);
491 		}
492 		if (!TAILQ_EMPTY(&sc->xbd_cm_q[XBD_Q_BUSY].q_tailq)) {
493 			/*
494 			 * Still pending requests, wait for the disk i/o
495 			 * to complete.
496 			 */
497 			HYPERVISOR_yield();
498 		}
499 	}
500 }
501 
502 /* Kernel dump function for a paravirtualized disk device */
503 static void
504 xbd_dump_complete(struct xbd_command *cm)
505 {
506 
507 	xbd_enqueue_cm(cm, XBD_Q_COMPLETE);
508 }
509 
510 static int
511 xbd_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
512     size_t length)
513 {
514 	struct disk *dp = arg;
515 	struct xbd_softc *sc = dp->d_drv1;
516 	struct xbd_command *cm;
517 	size_t chunk;
518 	int sbp;
519 	int rc = 0;
520 
521 	if (length <= 0)
522 		return (rc);
523 
524 	xbd_quiesce(sc);	/* All quiet on the western front. */
525 
526 	/*
527 	 * If this lock is held, then this module is failing, and a
528 	 * successful kernel dump is highly unlikely anyway.
529 	 */
530 	mtx_lock(&sc->xbd_io_lock);
531 
532 	/* Split the 64KB block as needed */
533 	for (sbp=0; length > 0; sbp++) {
534 		cm = xbd_dequeue_cm(sc, XBD_Q_FREE);
535 		if (cm == NULL) {
536 			mtx_unlock(&sc->xbd_io_lock);
537 			device_printf(sc->xbd_dev, "dump: no more commands?\n");
538 			return (EBUSY);
539 		}
540 
541 		if (gnttab_alloc_grant_references(sc->xbd_max_request_segments,
542 		    &cm->cm_gref_head) != 0) {
543 			xbd_free_command(cm);
544 			mtx_unlock(&sc->xbd_io_lock);
545 			device_printf(sc->xbd_dev, "no more grant allocs?\n");
546 			return (EBUSY);
547 		}
548 
549 		chunk = length > sc->xbd_max_request_size ?
550 		    sc->xbd_max_request_size : length;
551 		cm->cm_data = virtual;
552 		cm->cm_datalen = chunk;
553 		cm->cm_operation = BLKIF_OP_WRITE;
554 		cm->cm_sector_number = offset / dp->d_sectorsize;
555 		cm->cm_complete = xbd_dump_complete;
556 
557 		xbd_enqueue_cm(cm, XBD_Q_READY);
558 
559 		length -= chunk;
560 		offset += chunk;
561 		virtual = (char *) virtual + chunk;
562 	}
563 
564 	/* Tell DOM0 to do the I/O */
565 	xbd_startio(sc);
566 	mtx_unlock(&sc->xbd_io_lock);
567 
568 	/* Poll for the completion. */
569 	xbd_quiesce(sc);	/* All quite on the eastern front */
570 
571 	/* If there were any errors, bail out... */
572 	while ((cm = xbd_dequeue_cm(sc, XBD_Q_COMPLETE)) != NULL) {
573 		if (cm->cm_status != BLKIF_RSP_OKAY) {
574 			device_printf(sc->xbd_dev,
575 			    "Dump I/O failed at sector %jd\n",
576 			    cm->cm_sector_number);
577 			rc = EIO;
578 		}
579 		xbd_free_command(cm);
580 	}
581 
582 	return (rc);
583 }
584 
585 /*----------------------------- Disk Entrypoints -----------------------------*/
586 static int
587 xbd_open(struct disk *dp)
588 {
589 	struct xbd_softc *sc = dp->d_drv1;
590 
591 	if (sc == NULL) {
592 		printf("xb%d: not found", sc->xbd_unit);
593 		return (ENXIO);
594 	}
595 
596 	sc->xbd_flags |= XBDF_OPEN;
597 	sc->xbd_users++;
598 	return (0);
599 }
600 
601 static int
602 xbd_close(struct disk *dp)
603 {
604 	struct xbd_softc *sc = dp->d_drv1;
605 
606 	if (sc == NULL)
607 		return (ENXIO);
608 	sc->xbd_flags &= ~XBDF_OPEN;
609 	if (--(sc->xbd_users) == 0) {
610 		/*
611 		 * Check whether we have been instructed to close.  We will
612 		 * have ignored this request initially, as the device was
613 		 * still mounted.
614 		 */
615 		if (xenbus_get_otherend_state(sc->xbd_dev) ==
616 		    XenbusStateClosing)
617 			xbd_closing(sc->xbd_dev);
618 	}
619 	return (0);
620 }
621 
622 static int
623 xbd_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
624 {
625 	struct xbd_softc *sc = dp->d_drv1;
626 
627 	if (sc == NULL)
628 		return (ENXIO);
629 
630 	return (ENOTTY);
631 }
632 
633 /*
634  * Read/write routine for a buffer.  Finds the proper unit, place it on
635  * the sortq and kick the controller.
636  */
637 static void
638 xbd_strategy(struct bio *bp)
639 {
640 	struct xbd_softc *sc = bp->bio_disk->d_drv1;
641 
642 	/* bogus disk? */
643 	if (sc == NULL) {
644 		bp->bio_error = EINVAL;
645 		bp->bio_flags |= BIO_ERROR;
646 		bp->bio_resid = bp->bio_bcount;
647 		biodone(bp);
648 		return;
649 	}
650 
651 	/*
652 	 * Place it in the queue of disk activities for this disk
653 	 */
654 	mtx_lock(&sc->xbd_io_lock);
655 
656 	xbd_enqueue_bio(sc, bp);
657 	xbd_startio(sc);
658 
659 	mtx_unlock(&sc->xbd_io_lock);
660 	return;
661 }
662 
663 /*------------------------------ Ring Management -----------------------------*/
664 static int
665 xbd_alloc_ring(struct xbd_softc *sc)
666 {
667 	blkif_sring_t *sring;
668 	uintptr_t sring_page_addr;
669 	int error;
670 	int i;
671 
672 	sring = malloc(sc->xbd_ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
673 	    M_NOWAIT|M_ZERO);
674 	if (sring == NULL) {
675 		xenbus_dev_fatal(sc->xbd_dev, ENOMEM, "allocating shared ring");
676 		return (ENOMEM);
677 	}
678 	SHARED_RING_INIT(sring);
679 	FRONT_RING_INIT(&sc->xbd_ring, sring, sc->xbd_ring_pages * PAGE_SIZE);
680 
681 	for (i = 0, sring_page_addr = (uintptr_t)sring;
682 	     i < sc->xbd_ring_pages;
683 	     i++, sring_page_addr += PAGE_SIZE) {
684 
685 		error = xenbus_grant_ring(sc->xbd_dev,
686 		    (vtomach(sring_page_addr) >> PAGE_SHIFT),
687 		    &sc->xbd_ring_ref[i]);
688 		if (error) {
689 			xenbus_dev_fatal(sc->xbd_dev, error,
690 			    "granting ring_ref(%d)", i);
691 			return (error);
692 		}
693 	}
694 	if (sc->xbd_ring_pages == 1) {
695 		error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
696 		    "ring-ref", "%u", sc->xbd_ring_ref[0]);
697 		if (error) {
698 			xenbus_dev_fatal(sc->xbd_dev, error,
699 			    "writing %s/ring-ref",
700 			    xenbus_get_node(sc->xbd_dev));
701 			return (error);
702 		}
703 	} else {
704 		for (i = 0; i < sc->xbd_ring_pages; i++) {
705 			char ring_ref_name[]= "ring_refXX";
706 
707 			snprintf(ring_ref_name, sizeof(ring_ref_name),
708 			    "ring-ref%u", i);
709 			error = xs_printf(XST_NIL, xenbus_get_node(sc->xbd_dev),
710 			     ring_ref_name, "%u", sc->xbd_ring_ref[i]);
711 			if (error) {
712 				xenbus_dev_fatal(sc->xbd_dev, error,
713 				    "writing %s/%s",
714 				    xenbus_get_node(sc->xbd_dev),
715 				    ring_ref_name);
716 				return (error);
717 			}
718 		}
719 	}
720 
721 	error = bind_listening_port_to_irqhandler(
722 	    xenbus_get_otherend_id(sc->xbd_dev),
723 	    "xbd", (driver_intr_t *)xbd_int, sc,
724 	    INTR_TYPE_BIO | INTR_MPSAFE, &sc->xbd_irq);
725 	if (error) {
726 		xenbus_dev_fatal(sc->xbd_dev, error,
727 		    "bind_evtchn_to_irqhandler failed");
728 		return (error);
729 	}
730 
731 	return (0);
732 }
733 
734 static void
735 xbd_free_ring(struct xbd_softc *sc)
736 {
737 	int i;
738 
739 	if (sc->xbd_ring.sring == NULL)
740 		return;
741 
742 	for (i = 0; i < sc->xbd_ring_pages; i++) {
743 		if (sc->xbd_ring_ref[i] != GRANT_REF_INVALID) {
744 			gnttab_end_foreign_access_ref(sc->xbd_ring_ref[i]);
745 			sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
746 		}
747 	}
748 	free(sc->xbd_ring.sring, M_XENBLOCKFRONT);
749 	sc->xbd_ring.sring = NULL;
750 }
751 
752 /*-------------------------- Initialization/Teardown -------------------------*/
753 static void
754 xbd_setup_sysctl(struct xbd_softc *xbd)
755 {
756 	struct sysctl_ctx_list *sysctl_ctx = NULL;
757 	struct sysctl_oid *sysctl_tree = NULL;
758 
759 	sysctl_ctx = device_get_sysctl_ctx(xbd->xbd_dev);
760 	if (sysctl_ctx == NULL)
761 		return;
762 
763 	sysctl_tree = device_get_sysctl_tree(xbd->xbd_dev);
764 	if (sysctl_tree == NULL)
765 		return;
766 
767 	SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
768 	    "max_requests", CTLFLAG_RD, &xbd->xbd_max_requests, -1,
769 	    "maximum outstanding requests (negotiated)");
770 
771 	SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
772 	    "max_request_segments", CTLFLAG_RD,
773 	    &xbd->xbd_max_request_segments, 0,
774 	    "maximum number of pages per requests (negotiated)");
775 
776 	SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
777 	    "max_request_size", CTLFLAG_RD, &xbd->xbd_max_request_size, 0,
778 	    "maximum size in bytes of a request (negotiated)");
779 
780 	SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
781 	    "ring_pages", CTLFLAG_RD, &xbd->xbd_ring_pages, 0,
782 	    "communication channel pages (negotiated)");
783 }
784 
785 /*
786  * Translate Linux major/minor to an appropriate name and unit
787  * number. For HVM guests, this allows us to use the same drive names
788  * with blkfront as the emulated drives, easing transition slightly.
789  */
790 static void
791 xbd_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name)
792 {
793 	static struct vdev_info {
794 		int major;
795 		int shift;
796 		int base;
797 		const char *name;
798 	} info[] = {
799 		{3,	6,	0,	"ada"},	/* ide0 */
800 		{22,	6,	2,	"ada"},	/* ide1 */
801 		{33,	6,	4,	"ada"},	/* ide2 */
802 		{34,	6,	6,	"ada"},	/* ide3 */
803 		{56,	6,	8,	"ada"},	/* ide4 */
804 		{57,	6,	10,	"ada"},	/* ide5 */
805 		{88,	6,	12,	"ada"},	/* ide6 */
806 		{89,	6,	14,	"ada"},	/* ide7 */
807 		{90,	6,	16,	"ada"},	/* ide8 */
808 		{91,	6,	18,	"ada"},	/* ide9 */
809 
810 		{8,	4,	0,	"da"},	/* scsi disk0 */
811 		{65,	4,	16,	"da"},	/* scsi disk1 */
812 		{66,	4,	32,	"da"},	/* scsi disk2 */
813 		{67,	4,	48,	"da"},	/* scsi disk3 */
814 		{68,	4,	64,	"da"},	/* scsi disk4 */
815 		{69,	4,	80,	"da"},	/* scsi disk5 */
816 		{70,	4,	96,	"da"},	/* scsi disk6 */
817 		{71,	4,	112,	"da"},	/* scsi disk7 */
818 		{128,	4,	128,	"da"},	/* scsi disk8 */
819 		{129,	4,	144,	"da"},	/* scsi disk9 */
820 		{130,	4,	160,	"da"},	/* scsi disk10 */
821 		{131,	4,	176,	"da"},	/* scsi disk11 */
822 		{132,	4,	192,	"da"},	/* scsi disk12 */
823 		{133,	4,	208,	"da"},	/* scsi disk13 */
824 		{134,	4,	224,	"da"},	/* scsi disk14 */
825 		{135,	4,	240,	"da"},	/* scsi disk15 */
826 
827 		{202,	4,	0,	"xbd"},	/* xbd */
828 
829 		{0,	0,	0,	NULL},
830 	};
831 	int major = vdevice >> 8;
832 	int minor = vdevice & 0xff;
833 	int i;
834 
835 	if (vdevice & (1 << 28)) {
836 		*unit = (vdevice & ((1 << 28) - 1)) >> 8;
837 		*name = "xbd";
838 		return;
839 	}
840 
841 	for (i = 0; info[i].major; i++) {
842 		if (info[i].major == major) {
843 			*unit = info[i].base + (minor >> info[i].shift);
844 			*name = info[i].name;
845 			return;
846 		}
847 	}
848 
849 	*unit = minor >> 4;
850 	*name = "xbd";
851 }
852 
853 int
854 xbd_instance_create(struct xbd_softc *sc, blkif_sector_t sectors,
855     int vdevice, uint16_t vdisk_info, unsigned long sector_size)
856 {
857 	int unit, error = 0;
858 	const char *name;
859 
860 	xbd_vdevice_to_unit(vdevice, &unit, &name);
861 
862 	sc->xbd_unit = unit;
863 
864 	if (strcmp(name, "xbd"))
865 		device_printf(sc->xbd_dev, "attaching as %s%d\n", name, unit);
866 
867 	sc->xbd_disk = disk_alloc();
868 	sc->xbd_disk->d_unit = sc->xbd_unit;
869 	sc->xbd_disk->d_open = xbd_open;
870 	sc->xbd_disk->d_close = xbd_close;
871 	sc->xbd_disk->d_ioctl = xbd_ioctl;
872 	sc->xbd_disk->d_strategy = xbd_strategy;
873 	sc->xbd_disk->d_dump = xbd_dump;
874 	sc->xbd_disk->d_name = name;
875 	sc->xbd_disk->d_drv1 = sc;
876 	sc->xbd_disk->d_sectorsize = sector_size;
877 
878 	sc->xbd_disk->d_mediasize = sectors * sector_size;
879 	sc->xbd_disk->d_maxsize = sc->xbd_max_request_size;
880 	sc->xbd_disk->d_flags = 0;
881 	disk_create(sc->xbd_disk, DISK_VERSION);
882 
883 	return error;
884 }
885 
886 static void
887 xbd_free(struct xbd_softc *sc)
888 {
889 	int i;
890 
891 	/* Prevent new requests being issued until we fix things up. */
892 	mtx_lock(&sc->xbd_io_lock);
893 	sc->xbd_state = XBD_STATE_DISCONNECTED;
894 	mtx_unlock(&sc->xbd_io_lock);
895 
896 	/* Free resources associated with old device channel. */
897 	xbd_free_ring(sc);
898 	if (sc->xbd_shadow) {
899 
900 		for (i = 0; i < sc->xbd_max_requests; i++) {
901 			struct xbd_command *cm;
902 
903 			cm = &sc->xbd_shadow[i];
904 			if (cm->cm_sg_refs != NULL) {
905 				free(cm->cm_sg_refs, M_XENBLOCKFRONT);
906 				cm->cm_sg_refs = NULL;
907 			}
908 
909 			bus_dmamap_destroy(sc->xbd_io_dmat, cm->cm_map);
910 		}
911 		free(sc->xbd_shadow, M_XENBLOCKFRONT);
912 		sc->xbd_shadow = NULL;
913 
914 		bus_dma_tag_destroy(sc->xbd_io_dmat);
915 
916 		xbd_initq_cm(sc, XBD_Q_FREE);
917 		xbd_initq_cm(sc, XBD_Q_READY);
918 		xbd_initq_cm(sc, XBD_Q_COMPLETE);
919 	}
920 
921 	if (sc->xbd_irq) {
922 		unbind_from_irqhandler(sc->xbd_irq);
923 		sc->xbd_irq = 0;
924 	}
925 }
926 
927 /*--------------------------- State Change Handlers --------------------------*/
928 static void
929 xbd_initialize(struct xbd_softc *sc)
930 {
931 	const char *otherend_path;
932 	const char *node_path;
933 	uint32_t max_ring_page_order;
934 	int error;
935 	int i;
936 
937 	if (xenbus_get_state(sc->xbd_dev) != XenbusStateInitialising) {
938 		/* Initialization has already been performed. */
939 		return;
940 	}
941 
942 	/*
943 	 * Protocol defaults valid even if negotiation for a
944 	 * setting fails.
945 	 */
946 	max_ring_page_order = 0;
947 	sc->xbd_ring_pages = 1;
948 	sc->xbd_max_request_segments = BLKIF_MAX_SEGMENTS_PER_HEADER_BLOCK;
949 	sc->xbd_max_request_size =
950 	    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments);
951 	sc->xbd_max_request_blocks =
952 	    BLKIF_SEGS_TO_BLOCKS(sc->xbd_max_request_segments);
953 
954 	/*
955 	 * Protocol negotiation.
956 	 *
957 	 * \note xs_gather() returns on the first encountered error, so
958 	 *       we must use independant calls in order to guarantee
959 	 *       we don't miss information in a sparsly populated back-end
960 	 *       tree.
961 	 *
962 	 * \note xs_scanf() does not update variables for unmatched
963 	 *	 fields.
964 	 */
965 	otherend_path = xenbus_get_otherend_path(sc->xbd_dev);
966 	node_path = xenbus_get_node(sc->xbd_dev);
967 
968 	/* Support both backend schemes for relaying ring page limits. */
969 	(void)xs_scanf(XST_NIL, otherend_path,
970 	    "max-ring-page-order", NULL, "%" PRIu32,
971 	    &max_ring_page_order);
972 	sc->xbd_ring_pages = 1 << max_ring_page_order;
973 	(void)xs_scanf(XST_NIL, otherend_path,
974 	    "max-ring-pages", NULL, "%" PRIu32,
975 	    &sc->xbd_ring_pages);
976 	if (sc->xbd_ring_pages < 1)
977 		sc->xbd_ring_pages = 1;
978 
979 	sc->xbd_max_requests =
980 	    BLKIF_MAX_RING_REQUESTS(sc->xbd_ring_pages * PAGE_SIZE);
981 	(void)xs_scanf(XST_NIL, otherend_path,
982 	    "max-requests", NULL, "%" PRIu32,
983 	    &sc->xbd_max_requests);
984 
985 	(void)xs_scanf(XST_NIL, otherend_path,
986 	    "max-request-segments", NULL, "%" PRIu32,
987 	    &sc->xbd_max_request_segments);
988 
989 	(void)xs_scanf(XST_NIL, otherend_path,
990 	    "max-request-size", NULL, "%" PRIu32,
991 	    &sc->xbd_max_request_size);
992 
993 	if (sc->xbd_ring_pages > XBD_MAX_RING_PAGES) {
994 		device_printf(sc->xbd_dev,
995 		    "Back-end specified ring-pages of %u "
996 		    "limited to front-end limit of %zu.\n",
997 		    sc->xbd_ring_pages, XBD_MAX_RING_PAGES);
998 		sc->xbd_ring_pages = XBD_MAX_RING_PAGES;
999 	}
1000 
1001 	if (powerof2(sc->xbd_ring_pages) == 0) {
1002 		uint32_t new_page_limit;
1003 
1004 		new_page_limit = 0x01 << (fls(sc->xbd_ring_pages) - 1);
1005 		device_printf(sc->xbd_dev,
1006 		    "Back-end specified ring-pages of %u "
1007 		    "is not a power of 2. Limited to %u.\n",
1008 		    sc->xbd_ring_pages, new_page_limit);
1009 		sc->xbd_ring_pages = new_page_limit;
1010 	}
1011 
1012 	if (sc->xbd_max_requests > XBD_MAX_REQUESTS) {
1013 		device_printf(sc->xbd_dev,
1014 		    "Back-end specified max_requests of %u "
1015 		    "limited to front-end limit of %u.\n",
1016 		    sc->xbd_max_requests, XBD_MAX_REQUESTS);
1017 		sc->xbd_max_requests = XBD_MAX_REQUESTS;
1018 	}
1019 
1020 	if (sc->xbd_max_request_segments > XBD_MAX_SEGMENTS_PER_REQUEST) {
1021 		device_printf(sc->xbd_dev,
1022 		    "Back-end specified max_request_segments of %u "
1023 		    "limited to front-end limit of %u.\n",
1024 		    sc->xbd_max_request_segments,
1025 		    XBD_MAX_SEGMENTS_PER_REQUEST);
1026 		sc->xbd_max_request_segments = XBD_MAX_SEGMENTS_PER_REQUEST;
1027 	}
1028 
1029 	if (sc->xbd_max_request_size > XBD_MAX_REQUEST_SIZE) {
1030 		device_printf(sc->xbd_dev,
1031 		    "Back-end specified max_request_size of %u "
1032 		    "limited to front-end limit of %u.\n",
1033 		    sc->xbd_max_request_size,
1034 		    XBD_MAX_REQUEST_SIZE);
1035 		sc->xbd_max_request_size = XBD_MAX_REQUEST_SIZE;
1036 	}
1037 
1038  	if (sc->xbd_max_request_size >
1039 	    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments)) {
1040  		device_printf(sc->xbd_dev,
1041 		    "Back-end specified max_request_size of %u "
1042 		    "limited to front-end limit of %u.  (Too few segments.)\n",
1043 		    sc->xbd_max_request_size,
1044 		    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments));
1045  		sc->xbd_max_request_size =
1046  		    XBD_SEGS_TO_SIZE(sc->xbd_max_request_segments);
1047  	}
1048 
1049 	sc->xbd_max_request_blocks =
1050 	    BLKIF_SEGS_TO_BLOCKS(sc->xbd_max_request_segments);
1051 
1052 	/* Allocate datastructures based on negotiated values. */
1053 	error = bus_dma_tag_create(
1054 	    bus_get_dma_tag(sc->xbd_dev),	/* parent */
1055 	    512, PAGE_SIZE,			/* algnmnt, boundary */
1056 	    BUS_SPACE_MAXADDR,			/* lowaddr */
1057 	    BUS_SPACE_MAXADDR,			/* highaddr */
1058 	    NULL, NULL,				/* filter, filterarg */
1059 	    sc->xbd_max_request_size,
1060 	    sc->xbd_max_request_segments,
1061 	    PAGE_SIZE,				/* maxsegsize */
1062 	    BUS_DMA_ALLOCNOW,			/* flags */
1063 	    busdma_lock_mutex,			/* lockfunc */
1064 	    &sc->xbd_io_lock,			/* lockarg */
1065 	    &sc->xbd_io_dmat);
1066 	if (error != 0) {
1067 		xenbus_dev_fatal(sc->xbd_dev, error,
1068 		    "Cannot allocate parent DMA tag\n");
1069 		return;
1070 	}
1071 
1072 	/* Per-transaction data allocation. */
1073 	sc->xbd_shadow = malloc(sizeof(*sc->xbd_shadow) * sc->xbd_max_requests,
1074 	    M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
1075 	if (sc->xbd_shadow == NULL) {
1076 		bus_dma_tag_destroy(sc->xbd_io_dmat);
1077 		xenbus_dev_fatal(sc->xbd_dev, error,
1078 		    "Cannot allocate request structures\n");
1079 		return;
1080 	}
1081 
1082 	for (i = 0; i < sc->xbd_max_requests; i++) {
1083 		struct xbd_command *cm;
1084 
1085 		cm = &sc->xbd_shadow[i];
1086 		cm->cm_sg_refs = malloc(
1087 		    sizeof(grant_ref_t) * sc->xbd_max_request_segments,
1088 		    M_XENBLOCKFRONT, M_NOWAIT);
1089 		if (cm->cm_sg_refs == NULL)
1090 			break;
1091 		cm->cm_id = i;
1092 		cm->cm_flags = XBDCF_INITIALIZER;
1093 		cm->cm_sc = sc;
1094 		if (bus_dmamap_create(sc->xbd_io_dmat, 0, &cm->cm_map) != 0)
1095 			break;
1096 		xbd_free_command(cm);
1097 	}
1098 
1099 	if (xbd_alloc_ring(sc) != 0)
1100 		return;
1101 
1102 	/* Support both backend schemes for relaying ring page limits. */
1103 	if (sc->xbd_ring_pages > 1) {
1104 		error = xs_printf(XST_NIL, node_path,
1105 		    "num-ring-pages","%u",
1106 		    sc->xbd_ring_pages);
1107 		if (error) {
1108 			xenbus_dev_fatal(sc->xbd_dev, error,
1109 			    "writing %s/num-ring-pages",
1110 			    node_path);
1111 			return;
1112 		}
1113 
1114 		error = xs_printf(XST_NIL, node_path,
1115 		    "ring-page-order", "%u",
1116 		    fls(sc->xbd_ring_pages) - 1);
1117 		if (error) {
1118 			xenbus_dev_fatal(sc->xbd_dev, error,
1119 			    "writing %s/ring-page-order",
1120 			    node_path);
1121 			return;
1122 		}
1123 	}
1124 
1125 	error = xs_printf(XST_NIL, node_path,
1126 	    "max-requests","%u",
1127 	    sc->xbd_max_requests);
1128 	if (error) {
1129 		xenbus_dev_fatal(sc->xbd_dev, error,
1130 		    "writing %s/max-requests",
1131 		    node_path);
1132 		return;
1133 	}
1134 
1135 	error = xs_printf(XST_NIL, node_path,
1136 	    "max-request-segments","%u",
1137 	    sc->xbd_max_request_segments);
1138 	if (error) {
1139 		xenbus_dev_fatal(sc->xbd_dev, error,
1140 		    "writing %s/max-request-segments",
1141 		    node_path);
1142 		return;
1143 	}
1144 
1145 	error = xs_printf(XST_NIL, node_path,
1146 	    "max-request-size","%u",
1147 	    sc->xbd_max_request_size);
1148 	if (error) {
1149 		xenbus_dev_fatal(sc->xbd_dev, error,
1150 		    "writing %s/max-request-size",
1151 		    node_path);
1152 		return;
1153 	}
1154 
1155 	error = xs_printf(XST_NIL, node_path, "event-channel",
1156 	    "%u", irq_to_evtchn_port(sc->xbd_irq));
1157 	if (error) {
1158 		xenbus_dev_fatal(sc->xbd_dev, error,
1159 		    "writing %s/event-channel",
1160 		    node_path);
1161 		return;
1162 	}
1163 
1164 	error = xs_printf(XST_NIL, node_path, "protocol",
1165 	    "%s", XEN_IO_PROTO_ABI_NATIVE);
1166 	if (error) {
1167 		xenbus_dev_fatal(sc->xbd_dev, error,
1168 		    "writing %s/protocol",
1169 		    node_path);
1170 		return;
1171 	}
1172 
1173 	xenbus_set_state(sc->xbd_dev, XenbusStateInitialised);
1174 }
1175 
1176 /*
1177  * Invoked when the backend is finally 'ready' (and has published
1178  * the details about the physical device - #sectors, size, etc).
1179  */
1180 static void
1181 xbd_connect(struct xbd_softc *sc)
1182 {
1183 	device_t dev = sc->xbd_dev;
1184 	unsigned long sectors, sector_size;
1185 	unsigned int binfo;
1186 	int err, feature_barrier;
1187 
1188 	if (sc->xbd_state == XBD_STATE_CONNECTED ||
1189 	    sc->xbd_state == XBD_STATE_SUSPENDED)
1190 		return;
1191 
1192 	DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
1193 
1194 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1195 	    "sectors", "%lu", &sectors,
1196 	    "info", "%u", &binfo,
1197 	    "sector-size", "%lu", &sector_size,
1198 	    NULL);
1199 	if (err) {
1200 		xenbus_dev_fatal(dev, err,
1201 		    "reading backend fields at %s",
1202 		    xenbus_get_otherend_path(dev));
1203 		return;
1204 	}
1205 	err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
1206 	     "feature-barrier", "%lu", &feature_barrier,
1207 	     NULL);
1208 	if (!err || feature_barrier)
1209 		sc->xbd_flags |= XBDF_BARRIER;
1210 
1211 	if (sc->xbd_disk == NULL) {
1212 		device_printf(dev, "%juMB <%s> at %s",
1213 		    (uintmax_t) sectors / (1048576 / sector_size),
1214 		    device_get_desc(dev),
1215 		    xenbus_get_node(dev));
1216 		bus_print_child_footer(device_get_parent(dev), dev);
1217 
1218 		xbd_instance_create(sc, sectors, sc->xbd_vdevice, binfo,
1219 		    sector_size);
1220 	}
1221 
1222 	(void)xenbus_set_state(dev, XenbusStateConnected);
1223 
1224 	/* Kick pending requests. */
1225 	mtx_lock(&sc->xbd_io_lock);
1226 	sc->xbd_state = XBD_STATE_CONNECTED;
1227 	xbd_startio(sc);
1228 	sc->xbd_flags |= XBDF_READY;
1229 	mtx_unlock(&sc->xbd_io_lock);
1230 }
1231 
1232 /**
1233  * Handle the change of state of the backend to Closing.  We must delete our
1234  * device-layer structures now, to ensure that writes are flushed through to
1235  * the backend.  Once this is done, we can switch to Closed in
1236  * acknowledgement.
1237  */
1238 static void
1239 xbd_closing(device_t dev)
1240 {
1241 	struct xbd_softc *sc = device_get_softc(dev);
1242 
1243 	xenbus_set_state(dev, XenbusStateClosing);
1244 
1245 	DPRINTK("xbd_closing: %s removed\n", xenbus_get_node(dev));
1246 
1247 	if (sc->xbd_disk != NULL) {
1248 		disk_destroy(sc->xbd_disk);
1249 		sc->xbd_disk = NULL;
1250 	}
1251 
1252 	xenbus_set_state(dev, XenbusStateClosed);
1253 }
1254 
1255 /*---------------------------- NewBus Entrypoints ----------------------------*/
1256 static int
1257 xbd_probe(device_t dev)
1258 {
1259 
1260 	if (!strcmp(xenbus_get_type(dev), "vbd")) {
1261 		device_set_desc(dev, "Virtual Block Device");
1262 		device_quiet(dev);
1263 		return (0);
1264 	}
1265 
1266 	return (ENXIO);
1267 }
1268 
1269 /*
1270  * Setup supplies the backend dir, virtual device.  We place an event
1271  * channel and shared frame entries.  We watch backend to wait if it's
1272  * ok.
1273  */
1274 static int
1275 xbd_attach(device_t dev)
1276 {
1277 	struct xbd_softc *sc;
1278 	const char *name;
1279 	uint32_t vdevice;
1280 	int error;
1281 	int i;
1282 	int unit;
1283 
1284 	/* FIXME: Use dynamic device id if this is not set. */
1285 	error = xs_scanf(XST_NIL, xenbus_get_node(dev),
1286 	    "virtual-device", NULL, "%" PRIu32, &vdevice);
1287 	if (error) {
1288 		xenbus_dev_fatal(dev, error, "reading virtual-device");
1289 		device_printf(dev, "Couldn't determine virtual device.\n");
1290 		return (error);
1291 	}
1292 
1293 	xbd_vdevice_to_unit(vdevice, &unit, &name);
1294 	if (!strcmp(name, "xbd"))
1295 		device_set_unit(dev, unit);
1296 
1297 	sc = device_get_softc(dev);
1298 	mtx_init(&sc->xbd_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
1299 	xbd_initqs(sc);
1300 	for (i = 0; i < XBD_MAX_RING_PAGES; i++)
1301 		sc->xbd_ring_ref[i] = GRANT_REF_INVALID;
1302 
1303 	sc->xbd_dev = dev;
1304 	sc->xbd_vdevice = vdevice;
1305 	sc->xbd_state = XBD_STATE_DISCONNECTED;
1306 
1307 	xbd_setup_sysctl(sc);
1308 
1309 	/* Wait for backend device to publish its protocol capabilities. */
1310 	xenbus_set_state(dev, XenbusStateInitialising);
1311 
1312 	return (0);
1313 }
1314 
1315 static int
1316 xbd_detach(device_t dev)
1317 {
1318 	struct xbd_softc *sc = device_get_softc(dev);
1319 
1320 	DPRINTK("%s: %s removed\n", __func__, xenbus_get_node(dev));
1321 
1322 	xbd_free(sc);
1323 	mtx_destroy(&sc->xbd_io_lock);
1324 
1325 	return 0;
1326 }
1327 
1328 static int
1329 xbd_suspend(device_t dev)
1330 {
1331 	struct xbd_softc *sc = device_get_softc(dev);
1332 	int retval;
1333 	int saved_state;
1334 
1335 	/* Prevent new requests being issued until we fix things up. */
1336 	mtx_lock(&sc->xbd_io_lock);
1337 	saved_state = sc->xbd_state;
1338 	sc->xbd_state = XBD_STATE_SUSPENDED;
1339 
1340 	/* Wait for outstanding I/O to drain. */
1341 	retval = 0;
1342 	while (TAILQ_EMPTY(&sc->xbd_cm_q[XBD_Q_BUSY].q_tailq) == 0) {
1343 		if (msleep(&sc->xbd_cm_q[XBD_Q_BUSY], &sc->xbd_io_lock,
1344 		    PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
1345 			retval = EBUSY;
1346 			break;
1347 		}
1348 	}
1349 	mtx_unlock(&sc->xbd_io_lock);
1350 
1351 	if (retval != 0)
1352 		sc->xbd_state = saved_state;
1353 
1354 	return (retval);
1355 }
1356 
1357 static int
1358 xbd_resume(device_t dev)
1359 {
1360 	struct xbd_softc *sc = device_get_softc(dev);
1361 
1362 	DPRINTK("xbd_resume: %s\n", xenbus_get_node(dev));
1363 
1364 	xbd_free(sc);
1365 	xbd_initialize(sc);
1366 	return (0);
1367 }
1368 
1369 /**
1370  * Callback received when the backend's state changes.
1371  */
1372 static void
1373 xbd_backend_changed(device_t dev, XenbusState backend_state)
1374 {
1375 	struct xbd_softc *sc = device_get_softc(dev);
1376 
1377 	DPRINTK("backend_state=%d\n", backend_state);
1378 
1379 	switch (backend_state) {
1380 	case XenbusStateUnknown:
1381 	case XenbusStateInitialising:
1382 	case XenbusStateReconfigured:
1383 	case XenbusStateReconfiguring:
1384 	case XenbusStateClosed:
1385 		break;
1386 
1387 	case XenbusStateInitWait:
1388 	case XenbusStateInitialised:
1389 		xbd_initialize(sc);
1390 		break;
1391 
1392 	case XenbusStateConnected:
1393 		xbd_initialize(sc);
1394 		xbd_connect(sc);
1395 		break;
1396 
1397 	case XenbusStateClosing:
1398 		if (sc->xbd_users > 0)
1399 			xenbus_dev_error(dev, -EBUSY,
1400 			    "Device in use; refusing to close");
1401 		else
1402 			xbd_closing(dev);
1403 		break;
1404 	}
1405 }
1406 
1407 /*---------------------------- NewBus Registration ---------------------------*/
1408 static device_method_t xbd_methods[] = {
1409 	/* Device interface */
1410 	DEVMETHOD(device_probe,         xbd_probe),
1411 	DEVMETHOD(device_attach,        xbd_attach),
1412 	DEVMETHOD(device_detach,        xbd_detach),
1413 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
1414 	DEVMETHOD(device_suspend,       xbd_suspend),
1415 	DEVMETHOD(device_resume,        xbd_resume),
1416 
1417 	/* Xenbus interface */
1418 	DEVMETHOD(xenbus_otherend_changed, xbd_backend_changed),
1419 
1420 	{ 0, 0 }
1421 };
1422 
1423 static driver_t xbd_driver = {
1424 	"xbd",
1425 	xbd_methods,
1426 	sizeof(struct xbd_softc),
1427 };
1428 devclass_t xbd_devclass;
1429 
1430 DRIVER_MODULE(xbd, xenbusb_front, xbd_driver, xbd_devclass, 0, 0);
1431