xref: /freebsd/sys/dev/xen/blkfront/blkfront.c (revision 39beb93c3f8bdbf72a61fda42300b5ebed7390c8)
1 /*-
2  * All rights reserved.
3  *
4  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
5  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
6  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
7  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
8  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
9  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
10  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
11  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
12  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
13  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
14  * SUCH DAMAGE.
15  *
16  */
17 
18 /*
19  * XenoBSD block device driver
20  */
21 
22 #include <sys/cdefs.h>
23 __FBSDID("$FreeBSD$");
24 
25 #include <sys/param.h>
26 #include <sys/systm.h>
27 #include <sys/malloc.h>
28 #include <sys/kernel.h>
29 #include <vm/vm.h>
30 #include <vm/pmap.h>
31 
32 #include <sys/bio.h>
33 #include <sys/bus.h>
34 #include <sys/conf.h>
35 #include <sys/module.h>
36 
37 #include <machine/bus.h>
38 #include <sys/rman.h>
39 #include <machine/resource.h>
40 #include <machine/intr_machdep.h>
41 #include <machine/vmparam.h>
42 
43 #include <xen/hypervisor.h>
44 #include <machine/xen/xen-os.h>
45 #include <xen/xen_intr.h>
46 #include <xen/evtchn.h>
47 #include <xen/interface/grant_table.h>
48 #include <xen/interface/io/protocols.h>
49 #include <xen/xenbus/xenbusvar.h>
50 
51 #include <geom/geom_disk.h>
52 #include <machine/xen/xenfunc.h>
53 #include <xen/gnttab.h>
54 
55 #include <dev/xen/blkfront/block.h>
56 
57 #include "xenbus_if.h"
58 
59 #define    ASSERT(S)       KASSERT(S, (#S))
60 /* prototypes */
61 struct xb_softc;
62 static void xb_startio(struct xb_softc *sc);
63 static void connect(device_t, struct blkfront_info *);
64 static void blkfront_closing(device_t);
65 static int blkfront_detach(device_t);
66 static int talk_to_backend(device_t, struct blkfront_info *);
67 static int setup_blkring(device_t, struct blkfront_info *);
68 static void blkif_int(void *);
69 #if 0
70 static void blkif_restart_queue(void *arg);
71 #endif
72 static void blkif_recover(struct blkfront_info *);
73 static void blkif_completion(struct blk_shadow *);
74 static void blkif_free(struct blkfront_info *, int);
75 
76 #define GRANT_INVALID_REF 0
77 #define BLK_RING_SIZE __RING_SIZE((blkif_sring_t *)0, PAGE_SIZE)
78 
79 LIST_HEAD(xb_softc_list_head, xb_softc) xbsl_head;
80 
81 /* Control whether runtime update of vbds is enabled. */
82 #define ENABLE_VBD_UPDATE 0
83 
84 #if ENABLE_VBD_UPDATE
85 static void vbd_update(void);
86 #endif
87 
88 
89 #define BLKIF_STATE_DISCONNECTED 0
90 #define BLKIF_STATE_CONNECTED    1
91 #define BLKIF_STATE_SUSPENDED    2
92 
93 #ifdef notyet
94 static char *blkif_state_name[] = {
95 	[BLKIF_STATE_DISCONNECTED] = "disconnected",
96 	[BLKIF_STATE_CONNECTED]    = "connected",
97 	[BLKIF_STATE_SUSPENDED]    = "closed",
98 };
99 
100 static char * blkif_status_name[] = {
101 	[BLKIF_INTERFACE_STATUS_CLOSED]       = "closed",
102 	[BLKIF_INTERFACE_STATUS_DISCONNECTED] = "disconnected",
103 	[BLKIF_INTERFACE_STATUS_CONNECTED]    = "connected",
104 	[BLKIF_INTERFACE_STATUS_CHANGED]      = "changed",
105 };
106 #endif
107 #define WPRINTK(fmt, args...) printf("[XEN] " fmt, ##args)
108 #if 0
109 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d" fmt ".\n", __FUNCTION__, __LINE__,##args)
110 #else
111 #define DPRINTK(fmt, args...)
112 #endif
113 
114 static grant_ref_t gref_head;
115 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
116     (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
117 
118 static void kick_pending_request_queues(struct blkfront_info *);
119 static int blkif_open(struct disk *dp);
120 static int blkif_close(struct disk *dp);
121 static int blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td);
122 static int blkif_queue_request(struct bio *bp);
123 static void xb_strategy(struct bio *bp);
124 
125 
126 
127 /* XXX move to xb_vbd.c when VBD update support is added */
128 #define MAX_VBDS 64
129 
130 #define XBD_SECTOR_SIZE		512	/* XXX: assume for now */
131 #define XBD_SECTOR_SHFT		9
132 
133 static struct mtx blkif_io_lock;
134 
135 static vm_paddr_t
136 pfn_to_mfn(vm_paddr_t pfn)
137 {
138 	return (phystomach(pfn << PAGE_SHIFT) >> PAGE_SHIFT);
139 }
140 
141 
142 /*
143  * Translate Linux major/minor to an appropriate name and unit
144  * number. For HVM guests, this allows us to use the same drive names
145  * with blkfront as the emulated drives, easing transition slightly.
146  */
147 static void
148 blkfront_vdevice_to_unit(int vdevice, int *unit, const char **name)
149 {
150 	static struct vdev_info {
151 		int major;
152 		int shift;
153 		int base;
154 		const char *name;
155 	} info[] = {
156 		{3,	6,	0,	"ad"},	/* ide0 */
157 		{22,	6,	2,	"ad"},	/* ide1 */
158 		{33,	6,	4,	"ad"},	/* ide2 */
159 		{34,	6,	6,	"ad"},	/* ide3 */
160 		{56,	6,	8,	"ad"},	/* ide4 */
161 		{57,	6,	10,	"ad"},	/* ide5 */
162 		{88,	6,	12,	"ad"},	/* ide6 */
163 		{89,	6,	14,	"ad"},	/* ide7 */
164 		{90,	6,	16,	"ad"},	/* ide8 */
165 		{91,	6,	18,	"ad"},	/* ide9 */
166 
167 		{8,	4,	0,	"da"},	/* scsi disk0 */
168 		{65,	4,	16,	"da"},	/* scsi disk1 */
169 		{66,	4,	32,	"da"},	/* scsi disk2 */
170 		{67,	4,	48,	"da"},	/* scsi disk3 */
171 		{68,	4,	64,	"da"},	/* scsi disk4 */
172 		{69,	4,	80,	"da"},	/* scsi disk5 */
173 		{70,	4,	96,	"da"},	/* scsi disk6 */
174 		{71,	4,	112,	"da"},	/* scsi disk7 */
175 		{128,	4,	128,	"da"},	/* scsi disk8 */
176 		{129,	4,	144,	"da"},	/* scsi disk9 */
177 		{130,	4,	160,	"da"},	/* scsi disk10 */
178 		{131,	4,	176,	"da"},	/* scsi disk11 */
179 		{132,	4,	192,	"da"},	/* scsi disk12 */
180 		{133,	4,	208,	"da"},	/* scsi disk13 */
181 		{134,	4,	224,	"da"},	/* scsi disk14 */
182 		{135,	4,	240,	"da"},	/* scsi disk15 */
183 
184 		{202,	4,	0,	"xbd"},	/* xbd */
185 
186 		{0,	0,	0,	NULL},
187 	};
188 	int major = vdevice >> 8;
189 	int minor = vdevice & 0xff;
190 	int i;
191 
192 	if (vdevice & (1 << 28)) {
193 		*unit = (vdevice & ((1 << 28) - 1)) >> 8;
194 		*name = "xbd";
195 	}
196 
197 	for (i = 0; info[i].major; i++) {
198 		if (info[i].major == major) {
199 			*unit = info[i].base + (minor >> info[i].shift);
200 			*name = info[i].name;
201 			return;
202 		}
203 	}
204 
205 	*unit = minor >> 4;
206 	*name = "xbd";
207 }
208 
209 int
210 xlvbd_add(device_t dev, blkif_sector_t capacity,
211     int vdevice, uint16_t vdisk_info, uint16_t sector_size,
212     struct blkfront_info *info)
213 {
214 	struct xb_softc	*sc;
215 	int	unit, error = 0;
216 	const char *name;
217 
218 	blkfront_vdevice_to_unit(vdevice, &unit, &name);
219 
220 	sc = (struct xb_softc *)malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
221 	sc->xb_unit = unit;
222 	sc->xb_info = info;
223 	info->sc = sc;
224 
225 	if (strcmp(name, "xbd"))
226 		device_printf(dev, "attaching as %s%d\n", name, unit);
227 
228 	memset(&sc->xb_disk, 0, sizeof(sc->xb_disk));
229 	sc->xb_disk = disk_alloc();
230 	sc->xb_disk->d_unit = sc->xb_unit;
231 	sc->xb_disk->d_open = blkif_open;
232 	sc->xb_disk->d_close = blkif_close;
233 	sc->xb_disk->d_ioctl = blkif_ioctl;
234 	sc->xb_disk->d_strategy = xb_strategy;
235 	sc->xb_disk->d_name = name;
236 	sc->xb_disk->d_drv1 = sc;
237 	sc->xb_disk->d_sectorsize = sector_size;
238 
239 	/* XXX */
240 	sc->xb_disk->d_mediasize = capacity << XBD_SECTOR_SHFT;
241 #if 0
242 	sc->xb_disk->d_maxsize = DFLTPHYS;
243 #else /* XXX: xen can't handle large single i/o requests */
244 	sc->xb_disk->d_maxsize = 4096;
245 #endif
246 #ifdef notyet
247 	XENPRINTF("attaching device 0x%x unit %d capacity %llu\n",
248 		  xb_diskinfo[sc->xb_unit].device, sc->xb_unit,
249 		  sc->xb_disk->d_mediasize);
250 #endif
251 	sc->xb_disk->d_flags = 0;
252 	disk_create(sc->xb_disk, DISK_VERSION_00);
253 	bioq_init(&sc->xb_bioq);
254 
255 	return error;
256 }
257 
258 void
259 xlvbd_del(struct blkfront_info *info)
260 {
261 	struct xb_softc	*sc;
262 
263 	sc = info->sc;
264 	disk_destroy(sc->xb_disk);
265 }
266 /************************ end VBD support *****************/
267 
268 /*
269  * Read/write routine for a buffer.  Finds the proper unit, place it on
270  * the sortq and kick the controller.
271  */
272 static void
273 xb_strategy(struct bio *bp)
274 {
275 	struct xb_softc	*sc = (struct xb_softc *)bp->bio_disk->d_drv1;
276 
277 	/* bogus disk? */
278 	if (sc == NULL) {
279 		bp->bio_error = EINVAL;
280 		bp->bio_flags |= BIO_ERROR;
281 		goto bad;
282 	}
283 
284 	DPRINTK("");
285 
286 	/*
287 	 * Place it in the queue of disk activities for this disk
288 	 */
289 	mtx_lock(&blkif_io_lock);
290 	bioq_disksort(&sc->xb_bioq, bp);
291 
292 	xb_startio(sc);
293 	mtx_unlock(&blkif_io_lock);
294 	return;
295 
296  bad:
297 	/*
298 	 * Correctly set the bio to indicate a failed tranfer.
299 	 */
300 	bp->bio_resid = bp->bio_bcount;
301 	biodone(bp);
302 	return;
303 }
304 
305 static int
306 blkfront_probe(device_t dev)
307 {
308 
309 	if (!strcmp(xenbus_get_type(dev), "vbd")) {
310 		device_set_desc(dev, "Virtual Block Device");
311 		device_quiet(dev);
312 		return (0);
313 	}
314 
315 	return (ENXIO);
316 }
317 
318 /*
319  * Setup supplies the backend dir, virtual device.  We place an event
320  * channel and shared frame entries.  We watch backend to wait if it's
321  * ok.
322  */
323 static int
324 blkfront_attach(device_t dev)
325 {
326 	int err, vdevice, i, unit;
327 	struct blkfront_info *info;
328 	const char *name;
329 
330 	/* FIXME: Use dynamic device id if this is not set. */
331 	err = xenbus_scanf(XBT_NIL, xenbus_get_node(dev),
332 	    "virtual-device", NULL, "%i", &vdevice);
333 	if (err) {
334 		xenbus_dev_fatal(dev, err, "reading virtual-device");
335 		printf("couldn't find virtual device");
336 		return (err);
337 	}
338 
339 	blkfront_vdevice_to_unit(vdevice, &unit, &name);
340 	if (!strcmp(name, "xbd"))
341 		device_set_unit(dev, unit);
342 
343 	info = device_get_softc(dev);
344 
345 	/*
346 	 * XXX debug only
347 	 */
348 	for (i = 0; i < sizeof(*info); i++)
349 			if (((uint8_t *)info)[i] != 0)
350 					panic("non-null memory");
351 
352 	info->shadow_free = 0;
353 	info->xbdev = dev;
354 	info->vdevice = vdevice;
355 	info->connected = BLKIF_STATE_DISCONNECTED;
356 
357 	/* work queue needed ? */
358 	for (i = 0; i < BLK_RING_SIZE; i++)
359 		info->shadow[i].req.id = i+1;
360 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
361 
362 	/* Front end dir is a number, which is used as the id. */
363 	info->handle = strtoul(strrchr(xenbus_get_node(dev),'/')+1, NULL, 0);
364 
365 	err = talk_to_backend(dev, info);
366 	if (err)
367 		return (err);
368 
369 	return (0);
370 }
371 
372 static int
373 blkfront_resume(device_t dev)
374 {
375 	struct blkfront_info *info = device_get_softc(dev);
376 	int err;
377 
378 	DPRINTK("blkfront_resume: %s\n", dev->nodename);
379 
380 	blkif_free(info, 1);
381 
382 	err = talk_to_backend(dev, info);
383 
384 	if (info->connected == BLKIF_STATE_SUSPENDED && !err)
385 		blkif_recover(info);
386 
387 	return err;
388 }
389 
390 /* Common code used when first setting up, and when resuming. */
391 static int
392 talk_to_backend(device_t dev, struct blkfront_info *info)
393 {
394 	const char *message = NULL;
395 	struct xenbus_transaction xbt;
396 	int err;
397 
398 	/* Create shared ring, alloc event channel. */
399 	err = setup_blkring(dev, info);
400 	if (err)
401 		goto out;
402 
403  again:
404 	err = xenbus_transaction_start(&xbt);
405 	if (err) {
406 		xenbus_dev_fatal(dev, err, "starting transaction");
407 		goto destroy_blkring;
408 	}
409 
410 	err = xenbus_printf(xbt, xenbus_get_node(dev),
411 			    "ring-ref","%u", info->ring_ref);
412 	if (err) {
413 		message = "writing ring-ref";
414 		goto abort_transaction;
415 	}
416 	err = xenbus_printf(xbt, xenbus_get_node(dev),
417 		"event-channel", "%u", irq_to_evtchn_port(info->irq));
418 	if (err) {
419 		message = "writing event-channel";
420 		goto abort_transaction;
421 	}
422 	err = xenbus_printf(xbt, xenbus_get_node(dev),
423 		"protocol", "%s", XEN_IO_PROTO_ABI_NATIVE);
424 	if (err) {
425 		message = "writing protocol";
426 		goto abort_transaction;
427 	}
428 	err = xenbus_transaction_end(xbt, 0);
429 	if (err) {
430 		if (err == EAGAIN)
431 			goto again;
432 		xenbus_dev_fatal(dev, err, "completing transaction");
433 		goto destroy_blkring;
434 	}
435 	xenbus_set_state(dev, XenbusStateInitialised);
436 
437 	return 0;
438 
439  abort_transaction:
440 	xenbus_transaction_end(xbt, 1);
441 	if (message)
442 		xenbus_dev_fatal(dev, err, "%s", message);
443  destroy_blkring:
444 	blkif_free(info, 0);
445  out:
446 	return err;
447 }
448 
449 static int
450 setup_blkring(device_t dev, struct blkfront_info *info)
451 {
452 	blkif_sring_t *sring;
453 	int error;
454 
455 	info->ring_ref = GRANT_INVALID_REF;
456 
457 	sring = (blkif_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
458 	if (sring == NULL) {
459 		xenbus_dev_fatal(dev, ENOMEM, "allocating shared ring");
460 		return ENOMEM;
461 	}
462 	SHARED_RING_INIT(sring);
463 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
464 
465 	error = xenbus_grant_ring(dev, (vtomach(info->ring.sring) >> PAGE_SHIFT),
466 		&info->ring_ref);
467 	if (error) {
468 		free(sring, M_DEVBUF);
469 		info->ring.sring = NULL;
470 		goto fail;
471 	}
472 
473 	error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
474 		"xbd", (driver_intr_t *)blkif_int, info,
475 					INTR_TYPE_BIO | INTR_MPSAFE, &info->irq);
476 	if (error) {
477 		xenbus_dev_fatal(dev, error,
478 				 "bind_evtchn_to_irqhandler failed");
479 		goto fail;
480 	}
481 
482 	return (0);
483  fail:
484 	blkif_free(info, 0);
485 	return (error);
486 }
487 
488 
489 /**
490  * Callback received when the backend's state changes.
491  */
492 static void
493 blkfront_backend_changed(device_t dev, XenbusState backend_state)
494 {
495 	struct blkfront_info *info = device_get_softc(dev);
496 
497 	DPRINTK("blkfront:backend_changed.\n");
498 
499 	switch (backend_state) {
500 	case XenbusStateUnknown:
501 	case XenbusStateInitialising:
502 	case XenbusStateInitWait:
503 	case XenbusStateInitialised:
504 	case XenbusStateClosed:
505 	case XenbusStateReconfigured:
506 	case XenbusStateReconfiguring:
507 		break;
508 
509 	case XenbusStateConnected:
510 		connect(dev, info);
511 		break;
512 
513 	case XenbusStateClosing:
514 		if (info->users > 0)
515 			xenbus_dev_error(dev, -EBUSY,
516 					 "Device in use; refusing to close");
517 		else
518 			blkfront_closing(dev);
519 #ifdef notyet
520 		bd = bdget(info->dev);
521 		if (bd == NULL)
522 			xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
523 
524 		down(&bd->bd_sem);
525 		if (info->users > 0)
526 			xenbus_dev_error(dev, -EBUSY,
527 					 "Device in use; refusing to close");
528 		else
529 			blkfront_closing(dev);
530 		up(&bd->bd_sem);
531 		bdput(bd);
532 #endif
533 	}
534 }
535 
536 /*
537 ** Invoked when the backend is finally 'ready' (and has told produced
538 ** the details about the physical device - #sectors, size, etc).
539 */
540 static void
541 connect(device_t dev, struct blkfront_info *info)
542 {
543 	unsigned long sectors, sector_size;
544 	unsigned int binfo;
545 	int err;
546 
547         if( (info->connected == BLKIF_STATE_CONNECTED) ||
548 	    (info->connected == BLKIF_STATE_SUSPENDED) )
549 		return;
550 
551 	DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
552 
553 	err = xenbus_gather(XBT_NIL, xenbus_get_otherend_path(dev),
554 			    "sectors", "%lu", &sectors,
555 			    "info", "%u", &binfo,
556 			    "sector-size", "%lu", &sector_size,
557 			    NULL);
558 	if (err) {
559 		xenbus_dev_fatal(dev, err,
560 		    "reading backend fields at %s",
561 		    xenbus_get_otherend_path(dev));
562 		return;
563 	}
564 	err = xenbus_gather(XBT_NIL, xenbus_get_otherend_path(dev),
565 			    "feature-barrier", "%lu", &info->feature_barrier,
566 			    NULL);
567 	if (err)
568 		info->feature_barrier = 0;
569 
570 	device_printf(dev, "%juMB <%s> at %s",
571 	    (uintmax_t) sectors / (1048576 / sector_size),
572 	    device_get_desc(dev),
573 	    xenbus_get_node(dev));
574 	bus_print_child_footer(device_get_parent(dev), dev);
575 
576 	xlvbd_add(dev, sectors, info->vdevice, binfo, sector_size, info);
577 
578 	(void)xenbus_set_state(dev, XenbusStateConnected);
579 
580 	/* Kick pending requests. */
581 	mtx_lock(&blkif_io_lock);
582 	info->connected = BLKIF_STATE_CONNECTED;
583 	kick_pending_request_queues(info);
584 	mtx_unlock(&blkif_io_lock);
585 	info->is_ready = 1;
586 
587 #if 0
588 	add_disk(info->gd);
589 #endif
590 }
591 
592 /**
593  * Handle the change of state of the backend to Closing.  We must delete our
594  * device-layer structures now, to ensure that writes are flushed through to
595  * the backend.  Once is this done, we can switch to Closed in
596  * acknowledgement.
597  */
598 static void
599 blkfront_closing(device_t dev)
600 {
601 	struct blkfront_info *info = device_get_softc(dev);
602 
603 	DPRINTK("blkfront_closing: %s removed\n", xenbus_get_node(dev));
604 
605 	if (info->mi) {
606 		DPRINTK("Calling xlvbd_del\n");
607 		xlvbd_del(info);
608 		info->mi = NULL;
609 	}
610 
611 	xenbus_set_state(dev, XenbusStateClosed);
612 }
613 
614 
615 static int
616 blkfront_detach(device_t dev)
617 {
618 	struct blkfront_info *info = device_get_softc(dev);
619 
620 	DPRINTK("blkfront_remove: %s removed\n", xenbus_get_node(dev));
621 
622 	blkif_free(info, 0);
623 
624 	return 0;
625 }
626 
627 
628 static inline int
629 GET_ID_FROM_FREELIST(struct blkfront_info *info)
630 {
631 	unsigned long nfree = info->shadow_free;
632 
633 	KASSERT(nfree <= BLK_RING_SIZE, ("free %lu > RING_SIZE", nfree));
634 	info->shadow_free = info->shadow[nfree].req.id;
635 	info->shadow[nfree].req.id = 0x0fffffee; /* debug */
636 	return nfree;
637 }
638 
639 static inline void
640 ADD_ID_TO_FREELIST(struct blkfront_info *info, unsigned long id)
641 {
642 	info->shadow[id].req.id  = info->shadow_free;
643 	info->shadow[id].request = 0;
644 	info->shadow_free = id;
645 }
646 
647 static inline void
648 flush_requests(struct blkfront_info *info)
649 {
650 	int notify;
651 
652 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
653 
654 	if (notify)
655 		notify_remote_via_irq(info->irq);
656 }
657 
658 static void
659 kick_pending_request_queues(struct blkfront_info *info)
660 {
661 	/* XXX check if we can't simplify */
662 #if 0
663 	if (!RING_FULL(&info->ring)) {
664 		/* Re-enable calldowns. */
665 		blk_start_queue(info->rq);
666 		/* Kick things off immediately. */
667 		do_blkif_request(info->rq);
668 	}
669 #endif
670 	if (!RING_FULL(&info->ring)) {
671 #if 0
672 		sc = LIST_FIRST(&xbsl_head);
673 		LIST_REMOVE(sc, entry);
674 		/* Re-enable calldowns. */
675 		blk_start_queue(di->rq);
676 #endif
677 		/* Kick things off immediately. */
678 		xb_startio(info->sc);
679 	}
680 }
681 
682 #if 0
683 /* XXX */
684 static void blkif_restart_queue(void *arg)
685 {
686 	struct blkfront_info *info = (struct blkfront_info *)arg;
687 
688 	mtx_lock(&blkif_io_lock);
689 	kick_pending_request_queues(info);
690 	mtx_unlock(&blkif_io_lock);
691 }
692 #endif
693 
694 static void blkif_restart_queue_callback(void *arg)
695 {
696 #if 0
697 	struct blkfront_info *info = (struct blkfront_info *)arg;
698 	/* XXX BSD equiv ? */
699 
700 	schedule_work(&info->work);
701 #endif
702 }
703 
704 static int
705 blkif_open(struct disk *dp)
706 {
707 	struct xb_softc	*sc = (struct xb_softc *)dp->d_drv1;
708 
709 	if (sc == NULL) {
710 		printk("xb%d: not found", sc->xb_unit);
711 		return (ENXIO);
712 	}
713 
714 	sc->xb_flags |= XB_OPEN;
715 	sc->xb_info->users++;
716 	return (0);
717 }
718 
719 static int
720 blkif_close(struct disk *dp)
721 {
722 	struct xb_softc	*sc = (struct xb_softc *)dp->d_drv1;
723 
724 	if (sc == NULL)
725 		return (ENXIO);
726 	sc->xb_flags &= ~XB_OPEN;
727 	if (--(sc->xb_info->users) == 0) {
728 		/* Check whether we have been instructed to close.  We will
729 		   have ignored this request initially, as the device was
730 		   still mounted. */
731 		device_t dev = sc->xb_info->xbdev;
732 		XenbusState state =
733 			xenbus_read_driver_state(xenbus_get_otherend_path(dev));
734 
735 		if (state == XenbusStateClosing)
736 			blkfront_closing(dev);
737 	}
738 	return (0);
739 }
740 
741 static int
742 blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
743 {
744 	struct xb_softc	*sc = (struct xb_softc *)dp->d_drv1;
745 
746 	if (sc == NULL)
747 		return (ENXIO);
748 
749 	return (ENOTTY);
750 }
751 
752 
753 /*
754  * blkif_queue_request
755  *
756  * request block io
757  *
758  * id: for guest use only.
759  * operation: BLKIF_OP_{READ,WRITE,PROBE}
760  * buffer: buffer to read/write into. this should be a
761  *   virtual address in the guest os.
762  */
763 static int blkif_queue_request(struct bio *bp)
764 {
765 	caddr_t alignbuf;
766 	vm_paddr_t buffer_ma;
767 	blkif_request_t     *ring_req;
768 	unsigned long id;
769 	uint64_t fsect, lsect;
770 	struct xb_softc *sc = (struct xb_softc *)bp->bio_disk->d_drv1;
771 	struct blkfront_info *info = sc->xb_info;
772 	int ref;
773 
774 	if (unlikely(sc->xb_info->connected != BLKIF_STATE_CONNECTED))
775 		return 1;
776 
777 	if (gnttab_alloc_grant_references(
778 		    BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
779 		gnttab_request_free_callback(
780 			&info->callback,
781 			blkif_restart_queue_callback,
782 			info,
783 			BLKIF_MAX_SEGMENTS_PER_REQUEST);
784 		return 1;
785 	}
786 
787 	/* Check if the buffer is properly aligned */
788 	if ((vm_offset_t)bp->bio_data & PAGE_MASK) {
789 		int align = (bp->bio_bcount < PAGE_SIZE/2) ? XBD_SECTOR_SIZE :
790 			PAGE_SIZE;
791 		caddr_t newbuf = malloc(bp->bio_bcount + align, M_DEVBUF,
792 					M_NOWAIT);
793 
794 		alignbuf = (char *)roundup2((u_long)newbuf, align);
795 
796 		/* save a copy of the current buffer */
797 		bp->bio_driver1 = newbuf;
798 		bp->bio_driver2 = alignbuf;
799 
800 		/* Copy the data for a write */
801 		if (bp->bio_cmd == BIO_WRITE)
802 			bcopy(bp->bio_data, alignbuf, bp->bio_bcount);
803 	} else
804 		alignbuf = bp->bio_data;
805 
806 	/* Fill out a communications ring structure. */
807 	ring_req 	         = RING_GET_REQUEST(&info->ring,
808 						    info->ring.req_prod_pvt);
809 	id		         = GET_ID_FROM_FREELIST(info);
810 	info->shadow[id].request = (unsigned long)bp;
811 
812 	ring_req->id 	         = id;
813 	ring_req->operation 	 = (bp->bio_cmd == BIO_READ) ? BLKIF_OP_READ :
814 		BLKIF_OP_WRITE;
815 
816 	ring_req->sector_number= (blkif_sector_t)bp->bio_pblkno;
817 	ring_req->handle 	  = (blkif_vdev_t)(uintptr_t)sc->xb_disk;
818 
819 	ring_req->nr_segments  = 0;	/* XXX not doing scatter/gather since buffer
820 					 * chaining is not supported.
821 					 */
822 
823 	buffer_ma = vtomach(alignbuf);
824 	fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
825 	lsect = fsect + (bp->bio_bcount >> XBD_SECTOR_SHFT) - 1;
826 	/* install a grant reference. */
827 	ref = gnttab_claim_grant_reference(&gref_head);
828 	KASSERT( ref != -ENOSPC, ("grant_reference failed") );
829 
830 	gnttab_grant_foreign_access_ref(
831 		ref,
832 		xenbus_get_otherend_id(info->xbdev),
833 		buffer_ma >> PAGE_SHIFT,
834 		ring_req->operation & 1 ); /* ??? */
835 	info->shadow[id].frame[ring_req->nr_segments] =
836 		buffer_ma >> PAGE_SHIFT;
837 
838 	ring_req->seg[ring_req->nr_segments] =
839 		(struct blkif_request_segment) {
840 			.gref       = ref,
841 			.first_sect = fsect,
842 			.last_sect  = lsect };
843 
844 	ring_req->nr_segments++;
845 	KASSERT((buffer_ma & (XBD_SECTOR_SIZE-1)) == 0,
846 		("XEN buffer must be sector aligned"));
847 	KASSERT(lsect <= 7,
848 		("XEN disk driver data cannot cross a page boundary"));
849 
850 	buffer_ma &= ~PAGE_MASK;
851 
852 	info->ring.req_prod_pvt++;
853 
854 	/* Keep a private copy so we can reissue requests when recovering. */
855 	info->shadow[id].req = *ring_req;
856 
857 	gnttab_free_grant_references(gref_head);
858 
859 	return 0;
860 }
861 
862 
863 
864 /*
865  * Dequeue buffers and place them in the shared communication ring.
866  * Return when no more requests can be accepted or all buffers have
867  * been queued.
868  *
869  * Signal XEN once the ring has been filled out.
870  */
871 static void
872 xb_startio(struct xb_softc *sc)
873 {
874 	struct bio		*bp;
875 	int			queued = 0;
876 	struct blkfront_info *info = sc->xb_info;
877 	DPRINTK("");
878 
879 	mtx_assert(&blkif_io_lock, MA_OWNED);
880 
881 	while ((bp = bioq_takefirst(&sc->xb_bioq)) != NULL) {
882 
883 		if (RING_FULL(&info->ring))
884 			goto wait;
885 
886 		if (blkif_queue_request(bp)) {
887 		wait:
888 			bioq_insert_head(&sc->xb_bioq, bp);
889 			break;
890 		}
891 		queued++;
892 	}
893 
894 	if (queued != 0)
895 		flush_requests(sc->xb_info);
896 }
897 
898 static void
899 blkif_int(void *xsc)
900 {
901 	struct xb_softc *sc = NULL;
902 	struct bio *bp;
903 	blkif_response_t *bret;
904 	RING_IDX i, rp;
905 	struct blkfront_info *info = xsc;
906 	DPRINTK("");
907 
908 	TRACE_ENTER;
909 
910 	mtx_lock(&blkif_io_lock);
911 
912 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
913 		mtx_unlock(&blkif_io_lock);
914 		return;
915 	}
916 
917  again:
918 	rp = info->ring.sring->rsp_prod;
919 	rmb(); /* Ensure we see queued responses up to 'rp'. */
920 
921 	for (i = info->ring.rsp_cons; i != rp; i++) {
922 		unsigned long id;
923 
924 		bret = RING_GET_RESPONSE(&info->ring, i);
925 		id   = bret->id;
926 		bp   = (struct bio *)info->shadow[id].request;
927 
928 		blkif_completion(&info->shadow[id]);
929 
930 		ADD_ID_TO_FREELIST(info, id);
931 
932 		switch (bret->operation) {
933 		case BLKIF_OP_READ:
934 			/* had an unaligned buffer that needs to be copied */
935 			if (bp->bio_driver1)
936 				bcopy(bp->bio_driver2, bp->bio_data, bp->bio_bcount);
937 			/* FALLTHROUGH */
938 		case BLKIF_OP_WRITE:
939 
940 			/* free the copy buffer */
941 			if (bp->bio_driver1) {
942 				free(bp->bio_driver1, M_DEVBUF);
943 				bp->bio_driver1 = NULL;
944 			}
945 
946 			if ( unlikely(bret->status != BLKIF_RSP_OKAY) ) {
947 					printf("Bad return from blkdev data request: %x\n",
948 					  bret->status);
949 				bp->bio_flags |= BIO_ERROR;
950 			}
951 
952 			sc = (struct xb_softc *)bp->bio_disk->d_drv1;
953 
954 			if (bp->bio_flags & BIO_ERROR)
955 				bp->bio_error = EIO;
956 			else
957 				bp->bio_resid = 0;
958 
959 			biodone(bp);
960 			break;
961 		default:
962 			panic("received invalid operation");
963 			break;
964 		}
965 	}
966 
967 	info->ring.rsp_cons = i;
968 
969 	if (i != info->ring.req_prod_pvt) {
970 		int more_to_do;
971 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
972 		if (more_to_do)
973 			goto again;
974 	} else {
975 		info->ring.sring->rsp_event = i + 1;
976 	}
977 
978 	kick_pending_request_queues(info);
979 
980 	mtx_unlock(&blkif_io_lock);
981 }
982 
983 static void
984 blkif_free(struct blkfront_info *info, int suspend)
985 {
986 
987 /* Prevent new requests being issued until we fix things up. */
988 	mtx_lock(&blkif_io_lock);
989 	info->connected = suspend ?
990 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
991 	mtx_unlock(&blkif_io_lock);
992 
993 	/* Free resources associated with old device channel. */
994 	if (info->ring_ref != GRANT_INVALID_REF) {
995 		gnttab_end_foreign_access(info->ring_ref,
996 					  info->ring.sring);
997 		info->ring_ref = GRANT_INVALID_REF;
998 		info->ring.sring = NULL;
999 	}
1000 	if (info->irq)
1001 		unbind_from_irqhandler(info->irq);
1002 	info->irq = 0;
1003 
1004 }
1005 
1006 static void
1007 blkif_completion(struct blk_shadow *s)
1008 {
1009 	int i;
1010 
1011 	for (i = 0; i < s->req.nr_segments; i++)
1012 		gnttab_end_foreign_access(s->req.seg[i].gref, 0UL);
1013 }
1014 
1015 static void
1016 blkif_recover(struct blkfront_info *info)
1017 {
1018 	int i, j;
1019 	blkif_request_t *req;
1020 	struct blk_shadow *copy;
1021 
1022 	/* Stage 1: Make a safe copy of the shadow state. */
1023 	copy = (struct blk_shadow *)malloc(sizeof(info->shadow), M_DEVBUF, M_NOWAIT|M_ZERO);
1024 	PANIC_IF(copy == NULL);
1025 	memcpy(copy, info->shadow, sizeof(info->shadow));
1026 
1027 	/* Stage 2: Set up free list. */
1028 	memset(&info->shadow, 0, sizeof(info->shadow));
1029 	for (i = 0; i < BLK_RING_SIZE; i++)
1030 		info->shadow[i].req.id = i+1;
1031 	info->shadow_free = info->ring.req_prod_pvt;
1032 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
1033 
1034 	/* Stage 3: Find pending requests and requeue them. */
1035 	for (i = 0; i < BLK_RING_SIZE; i++) {
1036 		/* Not in use? */
1037 		if (copy[i].request == 0)
1038 			continue;
1039 
1040 		/* Grab a request slot and copy shadow state into it. */
1041 		req = RING_GET_REQUEST(
1042 			&info->ring, info->ring.req_prod_pvt);
1043 		*req = copy[i].req;
1044 
1045 		/* We get a new request id, and must reset the shadow state. */
1046 		req->id = GET_ID_FROM_FREELIST(info);
1047 		memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
1048 
1049 		/* Rewrite any grant references invalidated by suspend/resume. */
1050 		for (j = 0; j < req->nr_segments; j++)
1051 			gnttab_grant_foreign_access_ref(
1052 				req->seg[j].gref,
1053 				xenbus_get_otherend_id(info->xbdev),
1054 				pfn_to_mfn(info->shadow[req->id].frame[j]),
1055 				0 /* assume not readonly */);
1056 
1057 		info->shadow[req->id].req = *req;
1058 
1059 		info->ring.req_prod_pvt++;
1060 	}
1061 
1062 	free(copy, M_DEVBUF);
1063 
1064 	xenbus_set_state(info->xbdev, XenbusStateConnected);
1065 
1066 	/* Now safe for us to use the shared ring */
1067 	mtx_lock(&blkif_io_lock);
1068 	info->connected = BLKIF_STATE_CONNECTED;
1069 	mtx_unlock(&blkif_io_lock);
1070 
1071 	/* Send off requeued requests */
1072 	mtx_lock(&blkif_io_lock);
1073 	flush_requests(info);
1074 
1075 	/* Kick any other new requests queued since we resumed */
1076 	kick_pending_request_queues(info);
1077 	mtx_unlock(&blkif_io_lock);
1078 }
1079 
1080 /* ** Driver registration ** */
1081 static device_method_t blkfront_methods[] = {
1082 	/* Device interface */
1083 	DEVMETHOD(device_probe,         blkfront_probe),
1084 	DEVMETHOD(device_attach,        blkfront_attach),
1085 	DEVMETHOD(device_detach,        blkfront_detach),
1086 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
1087 	DEVMETHOD(device_suspend,       bus_generic_suspend),
1088 	DEVMETHOD(device_resume,        blkfront_resume),
1089 
1090 	/* Xenbus interface */
1091 	DEVMETHOD(xenbus_backend_changed, blkfront_backend_changed),
1092 
1093 	{ 0, 0 }
1094 };
1095 
1096 static driver_t blkfront_driver = {
1097 	"xbd",
1098 	blkfront_methods,
1099 	sizeof(struct blkfront_info),
1100 };
1101 devclass_t blkfront_devclass;
1102 
1103 DRIVER_MODULE(xbd, xenbus, blkfront_driver, blkfront_devclass, 0, 0);
1104 
1105 MTX_SYSINIT(ioreq, &blkif_io_lock, "BIO LOCK", MTX_NOWITNESS); /* XXX how does one enroll a lock? */
1106 
1107