xref: /freebsd/sys/dev/xen/blkfront/blkfront.c (revision 5d254c0494187e5508b8de1ebe9d740ee3bbc64a) 
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 <machine/xen/hypervisor.h>
44 #include <machine/xen/xen-os.h>
45 #include <machine/xen/xen_intr.h>
46 #include <machine/xen/xenbus.h>
47 #include <machine/xen/evtchn.h>
48 #include <xen/interface/grant_table.h>
49 
50 #include <geom/geom_disk.h>
51 #include <machine/xen/xenfunc.h>
52 #include <xen/gnttab.h>
53 
54 #include <dev/xen/blkfront/block.h>
55 
56 #define    ASSERT(S)       KASSERT(S, (#S))
57 /* prototypes */
58 struct xb_softc;
59 static void xb_startio(struct xb_softc *sc);
60 static void connect(struct blkfront_info *);
61 static void blkfront_closing(struct xenbus_device *);
62 static int blkfront_remove(struct xenbus_device *);
63 static int talk_to_backend(struct xenbus_device *, struct blkfront_info *);
64 static int setup_blkring(struct xenbus_device *, struct blkfront_info *);
65 static void blkif_int(void *);
66 #if 0
67 static void blkif_restart_queue(void *arg);
68 #endif
69 static void blkif_recover(struct blkfront_info *);
70 static void blkif_completion(struct blk_shadow *);
71 static void blkif_free(struct blkfront_info *, int);
72 
73 #define GRANT_INVALID_REF 0
74 #define BLK_RING_SIZE __RING_SIZE((blkif_sring_t *)0, PAGE_SIZE)
75 
76 LIST_HEAD(xb_softc_list_head, xb_softc) xbsl_head;
77 
78 /* Control whether runtime update of vbds is enabled. */
79 #define ENABLE_VBD_UPDATE 0
80 
81 #if ENABLE_VBD_UPDATE
82 static void vbd_update(void);
83 #endif
84 
85 
86 #define BLKIF_STATE_DISCONNECTED 0
87 #define BLKIF_STATE_CONNECTED    1
88 #define BLKIF_STATE_SUSPENDED    2
89 
90 #ifdef notyet
91 static char *blkif_state_name[] = {
92 	[BLKIF_STATE_DISCONNECTED] = "disconnected",
93 	[BLKIF_STATE_CONNECTED]    = "connected",
94 	[BLKIF_STATE_SUSPENDED]    = "closed",
95 };
96 
97 static char * blkif_status_name[] = {
98 	[BLKIF_INTERFACE_STATUS_CLOSED]       = "closed",
99 	[BLKIF_INTERFACE_STATUS_DISCONNECTED] = "disconnected",
100 	[BLKIF_INTERFACE_STATUS_CONNECTED]    = "connected",
101 	[BLKIF_INTERFACE_STATUS_CHANGED]      = "changed",
102 };
103 #endif
104 #define WPRINTK(fmt, args...) printf("[XEN] " fmt, ##args)
105 #if 0
106 #define DPRINTK(fmt, args...) printf("[XEN] %s:%d" fmt ".\n", __FUNCTION__, __LINE__,##args)
107 #else
108 #define DPRINTK(fmt, args...)
109 #endif
110 
111 static grant_ref_t gref_head;
112 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
113     (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
114 
115 static void kick_pending_request_queues(struct blkfront_info *);
116 static int blkif_open(struct disk *dp);
117 static int blkif_close(struct disk *dp);
118 static int blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td);
119 static int blkif_queue_request(struct bio *bp);
120 static void xb_strategy(struct bio *bp);
121 
122 
123 
124 /* XXX move to xb_vbd.c when VBD update support is added */
125 #define MAX_VBDS 64
126 
127 #define XBD_SECTOR_SIZE		512	/* XXX: assume for now */
128 #define XBD_SECTOR_SHFT		9
129 
130 static struct mtx blkif_io_lock;
131 
132 static vm_paddr_t
133 pfn_to_mfn(vm_paddr_t pfn)
134 {
135 	return (phystomach(pfn << PAGE_SHIFT) >> PAGE_SHIFT);
136 }
137 
138 
139 int
140 xlvbd_add(blkif_sector_t capacity, int unit, uint16_t vdisk_info, uint16_t sector_size,
141 	  struct blkfront_info *info)
142 {
143 	struct xb_softc	*sc;
144 	int			error = 0;
145 
146 	sc = (struct xb_softc *)malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO);
147 	sc->xb_unit = unit;
148 	sc->xb_info = info;
149 	info->sc = sc;
150 
151 	memset(&sc->xb_disk, 0, sizeof(sc->xb_disk));
152 	sc->xb_disk = disk_alloc();
153 	sc->xb_disk->d_unit = unit;
154 	sc->xb_disk->d_open = blkif_open;
155 	sc->xb_disk->d_close = blkif_close;
156 	sc->xb_disk->d_ioctl = blkif_ioctl;
157 	sc->xb_disk->d_strategy = xb_strategy;
158 	sc->xb_disk->d_name = "xbd";
159 	sc->xb_disk->d_drv1 = sc;
160 	sc->xb_disk->d_sectorsize = sector_size;
161 
162 	/* XXX */
163 	sc->xb_disk->d_mediasize = capacity << XBD_SECTOR_SHFT;
164 #if 0
165 	sc->xb_disk->d_maxsize = DFLTPHYS;
166 #else /* XXX: xen can't handle large single i/o requests */
167 	sc->xb_disk->d_maxsize = 4096;
168 #endif
169 #ifdef notyet
170 	XENPRINTF("attaching device 0x%x unit %d capacity %llu\n",
171 		  xb_diskinfo[sc->xb_unit].device, sc->xb_unit,
172 		  sc->xb_disk->d_mediasize);
173 #endif
174 	sc->xb_disk->d_flags = 0;
175 	disk_create(sc->xb_disk, DISK_VERSION_00);
176 	bioq_init(&sc->xb_bioq);
177 
178 	return error;
179 }
180 
181 void
182 xlvbd_del(struct blkfront_info *info)
183 {
184 	struct xb_softc	*sc;
185 
186 	sc = info->sc;
187 	disk_destroy(sc->xb_disk);
188 }
189 /************************ end VBD support *****************/
190 
191 /*
192  * Read/write routine for a buffer.  Finds the proper unit, place it on
193  * the sortq and kick the controller.
194  */
195 static void
196 xb_strategy(struct bio *bp)
197 {
198 	struct xb_softc	*sc = (struct xb_softc *)bp->bio_disk->d_drv1;
199 
200 	/* bogus disk? */
201 	if (sc == NULL) {
202 		bp->bio_error = EINVAL;
203 		bp->bio_flags |= BIO_ERROR;
204 		goto bad;
205 	}
206 
207 	DPRINTK("");
208 
209 	/*
210 	 * Place it in the queue of disk activities for this disk
211 	 */
212 	mtx_lock(&blkif_io_lock);
213 	bioq_disksort(&sc->xb_bioq, bp);
214 
215 	xb_startio(sc);
216 	mtx_unlock(&blkif_io_lock);
217 	return;
218 
219  bad:
220 	/*
221 	 * Correctly set the bio to indicate a failed tranfer.
222 	 */
223 	bp->bio_resid = bp->bio_bcount;
224 	biodone(bp);
225 	return;
226 }
227 
228 
229 /* Setup supplies the backend dir, virtual device.
230 
231 We place an event channel and shared frame entries.
232 We watch backend to wait if it's ok. */
233 static int blkfront_probe(struct xenbus_device *dev,
234 			  const struct xenbus_device_id *id)
235 {
236 	int err, vdevice, i;
237 	struct blkfront_info *info;
238 
239 	/* FIXME: Use dynamic device id if this is not set. */
240 	err = xenbus_scanf(XBT_NIL, dev->nodename,
241 			   "virtual-device", "%i", &vdevice);
242 	if (err != 1) {
243 		xenbus_dev_fatal(dev, err, "reading virtual-device");
244 		printf("couldn't find virtual device");
245 		return (err);
246 	}
247 
248 	info = malloc(sizeof(*info), M_DEVBUF, M_NOWAIT|M_ZERO);
249 	if (info == NULL) {
250 		xenbus_dev_fatal(dev, ENOMEM, "allocating info structure");
251 		return ENOMEM;
252 	}
253 
254 	/*
255 	 * XXX debug only
256 	 */
257 	for (i = 0; i < sizeof(*info); i++)
258 			if (((uint8_t *)info)[i] != 0)
259 					panic("non-null memory");
260 
261 	info->shadow_free = 0;
262 	info->xbdev = dev;
263 	info->vdevice = vdevice;
264 	info->connected = BLKIF_STATE_DISCONNECTED;
265 
266 	/* work queue needed ? */
267 	for (i = 0; i < BLK_RING_SIZE; i++)
268 		info->shadow[i].req.id = i+1;
269 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
270 
271 	/* Front end dir is a number, which is used as the id. */
272 	info->handle = strtoul(strrchr(dev->nodename,'/')+1, NULL, 0);
273 	dev->dev_driver_data = info;
274 
275 	err = talk_to_backend(dev, info);
276 	if (err) {
277 		free(info, M_DEVBUF);
278 		dev->dev_driver_data = NULL;
279 		return err;
280 	}
281 
282 	return 0;
283 }
284 
285 
286 static int blkfront_resume(struct xenbus_device *dev)
287 {
288 	struct blkfront_info *info = dev->dev_driver_data;
289 	int err;
290 
291 	DPRINTK("blkfront_resume: %s\n", dev->nodename);
292 
293 	blkif_free(info, 1);
294 
295 	err = talk_to_backend(dev, info);
296 	if (!err)
297 		blkif_recover(info);
298 
299 	return err;
300 }
301 
302 /* Common code used when first setting up, and when resuming. */
303 static int talk_to_backend(struct xenbus_device *dev,
304 			   struct blkfront_info *info)
305 {
306 	const char *message = NULL;
307 	struct xenbus_transaction xbt;
308 	int err;
309 
310 	/* Create shared ring, alloc event channel. */
311 	err = setup_blkring(dev, info);
312 	if (err)
313 		goto out;
314 
315  again:
316 	err = xenbus_transaction_start(&xbt);
317 	if (err) {
318 		xenbus_dev_fatal(dev, err, "starting transaction");
319 		goto destroy_blkring;
320 	}
321 
322 	err = xenbus_printf(xbt, dev->nodename,
323 			    "ring-ref","%u", info->ring_ref);
324 	if (err) {
325 		message = "writing ring-ref";
326 		goto abort_transaction;
327 	}
328 	err = xenbus_printf(xbt, dev->nodename,
329 		"event-channel", "%u", irq_to_evtchn_port(info->irq));
330 	if (err) {
331 		message = "writing event-channel";
332 		goto abort_transaction;
333 	}
334 
335 	err = xenbus_transaction_end(xbt, 0);
336 	if (err) {
337 		if (err == -EAGAIN)
338 			goto again;
339 		xenbus_dev_fatal(dev, err, "completing transaction");
340 		goto destroy_blkring;
341 	}
342 	xenbus_switch_state(dev, XenbusStateInitialised);
343 
344 	return 0;
345 
346  abort_transaction:
347 	xenbus_transaction_end(xbt, 1);
348 	if (message)
349 		xenbus_dev_fatal(dev, err, "%s", message);
350  destroy_blkring:
351 	blkif_free(info, 0);
352  out:
353 	return err;
354 }
355 
356 static int
357 setup_blkring(struct xenbus_device *dev, struct blkfront_info *info)
358 {
359 	blkif_sring_t *sring;
360 	int err;
361 
362 	info->ring_ref = GRANT_INVALID_REF;
363 
364 	sring = (blkif_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
365 	if (sring == NULL) {
366 		xenbus_dev_fatal(dev, ENOMEM, "allocating shared ring");
367 		return ENOMEM;
368 	}
369 	SHARED_RING_INIT(sring);
370 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
371 
372 	err = xenbus_grant_ring(dev, (vtomach(info->ring.sring) >> PAGE_SHIFT));
373 	if (err < 0) {
374 		free(sring, M_DEVBUF);
375 		info->ring.sring = NULL;
376 		goto fail;
377 	}
378 	info->ring_ref = err;
379 
380 	err = bind_listening_port_to_irqhandler(dev->otherend_id,
381 		"xbd", (driver_intr_t *)blkif_int, info,
382 					INTR_TYPE_BIO | INTR_MPSAFE, NULL);
383 	if (err <= 0) {
384 		xenbus_dev_fatal(dev, err,
385 				 "bind_evtchn_to_irqhandler failed");
386 		goto fail;
387 	}
388 	info->irq = err;
389 
390 	return 0;
391  fail:
392 	blkif_free(info, 0);
393 	return err;
394 }
395 
396 
397 /**
398  * Callback received when the backend's state changes.
399  */
400 static void backend_changed(struct xenbus_device *dev,
401 			    XenbusState backend_state)
402 {
403 	struct blkfront_info *info = dev->dev_driver_data;
404 
405 	DPRINTK("blkfront:backend_changed.\n");
406 
407 	switch (backend_state) {
408 	case XenbusStateUnknown:
409 	case XenbusStateInitialising:
410 	case XenbusStateInitWait:
411 	case XenbusStateInitialised:
412 	case XenbusStateClosed:
413 		break;
414 
415 	case XenbusStateConnected:
416 		connect(info);
417 		break;
418 
419 	case XenbusStateClosing:
420 		if (info->users > 0)
421 			xenbus_dev_error(dev, -EBUSY,
422 					 "Device in use; refusing to close");
423 		else
424 			blkfront_closing(dev);
425 #ifdef notyet
426 		bd = bdget(info->dev);
427 		if (bd == NULL)
428 			xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
429 
430 		down(&bd->bd_sem);
431 		if (info->users > 0)
432 			xenbus_dev_error(dev, -EBUSY,
433 					 "Device in use; refusing to close");
434 		else
435 			blkfront_closing(dev);
436 		up(&bd->bd_sem);
437 		bdput(bd);
438 #endif
439 	}
440 }
441 
442 /*
443 ** Invoked when the backend is finally 'ready' (and has told produced
444 ** the details about the physical device - #sectors, size, etc).
445 */
446 static void
447 connect(struct blkfront_info *info)
448 {
449 	unsigned long sectors, sector_size;
450 	unsigned int binfo;
451 	int err;
452 
453         if( (info->connected == BLKIF_STATE_CONNECTED) ||
454 	    (info->connected == BLKIF_STATE_SUSPENDED) )
455 		return;
456 
457 	DPRINTK("blkfront.c:connect:%s.\n", info->xbdev->otherend);
458 
459 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
460 			    "sectors", "%lu", &sectors,
461 			    "info", "%u", &binfo,
462 			    "sector-size", "%lu", &sector_size,
463 			    NULL);
464 	if (err) {
465 		xenbus_dev_fatal(info->xbdev, err,
466 				 "reading backend fields at %s",
467 				 info->xbdev->otherend);
468 		return;
469 	}
470 	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
471 			    "feature-barrier", "%lu", &info->feature_barrier,
472 			    NULL);
473 	if (err)
474 		info->feature_barrier = 0;
475 
476 	xlvbd_add(sectors, info->vdevice, binfo, sector_size, info);
477 
478 	(void)xenbus_switch_state(info->xbdev, XenbusStateConnected);
479 
480 	/* Kick pending requests. */
481 	mtx_lock(&blkif_io_lock);
482 	info->connected = BLKIF_STATE_CONNECTED;
483 	kick_pending_request_queues(info);
484 	mtx_unlock(&blkif_io_lock);
485 
486 #if 0
487 	add_disk(info->gd);
488 #endif
489 }
490 
491 /**
492  * Handle the change of state of the backend to Closing.  We must delete our
493  * device-layer structures now, to ensure that writes are flushed through to
494  * the backend.  Once is this done, we can switch to Closed in
495  * acknowledgement.
496  */
497 static void blkfront_closing(struct xenbus_device *dev)
498 {
499 	struct blkfront_info *info = dev->dev_driver_data;
500 
501 	DPRINTK("blkfront_closing: %s removed\n", dev->nodename);
502 
503 	if (info->mi) {
504 		DPRINTK("Calling xlvbd_del\n");
505 		xlvbd_del(info);
506 		info->mi = NULL;
507 	}
508 
509 	xenbus_switch_state(dev, XenbusStateClosed);
510 }
511 
512 
513 static int blkfront_remove(struct xenbus_device *dev)
514 {
515 	struct blkfront_info *info = dev->dev_driver_data;
516 
517 	DPRINTK("blkfront_remove: %s removed\n", dev->nodename);
518 
519 	blkif_free(info, 0);
520 
521 	free(info, M_DEVBUF);
522 
523 	return 0;
524 }
525 
526 
527 static inline int
528 GET_ID_FROM_FREELIST(struct blkfront_info *info)
529 {
530 	unsigned long nfree = info->shadow_free;
531 
532 	KASSERT(nfree <= BLK_RING_SIZE, ("free %lu > RING_SIZE", nfree));
533 	info->shadow_free = info->shadow[nfree].req.id;
534 	info->shadow[nfree].req.id = 0x0fffffee; /* debug */
535 	return nfree;
536 }
537 
538 static inline void
539 ADD_ID_TO_FREELIST(struct blkfront_info *info, unsigned long id)
540 {
541 	info->shadow[id].req.id  = info->shadow_free;
542 	info->shadow[id].request = 0;
543 	info->shadow_free = id;
544 }
545 
546 static inline void
547 flush_requests(struct blkfront_info *info)
548 {
549 	int notify;
550 
551 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
552 
553 	if (notify)
554 		notify_remote_via_irq(info->irq);
555 }
556 
557 static void
558 kick_pending_request_queues(struct blkfront_info *info)
559 {
560 	/* XXX check if we can't simplify */
561 #if 0
562 	if (!RING_FULL(&info->ring)) {
563 		/* Re-enable calldowns. */
564 		blk_start_queue(info->rq);
565 		/* Kick things off immediately. */
566 		do_blkif_request(info->rq);
567 	}
568 #endif
569 	if (!RING_FULL(&info->ring)) {
570 #if 0
571 		sc = LIST_FIRST(&xbsl_head);
572 		LIST_REMOVE(sc, entry);
573 		/* Re-enable calldowns. */
574 		blk_start_queue(di->rq);
575 #endif
576 		/* Kick things off immediately. */
577 		xb_startio(info->sc);
578 	}
579 }
580 
581 #if 0
582 /* XXX */
583 static void blkif_restart_queue(void *arg)
584 {
585 	struct blkfront_info *info = (struct blkfront_info *)arg;
586 
587 	mtx_lock(&blkif_io_lock);
588 	kick_pending_request_queues(info);
589 	mtx_unlock(&blkif_io_lock);
590 }
591 #endif
592 
593 static void blkif_restart_queue_callback(void *arg)
594 {
595 #if 0
596 	struct blkfront_info *info = (struct blkfront_info *)arg;
597 	/* XXX BSD equiv ? */
598 
599 	schedule_work(&info->work);
600 #endif
601 }
602 
603 static int
604 blkif_open(struct disk *dp)
605 {
606 	struct xb_softc	*sc = (struct xb_softc *)dp->d_drv1;
607 
608 	if (sc == NULL) {
609 		printk("xb%d: not found", sc->xb_unit);
610 		return (ENXIO);
611 	}
612 
613 	sc->xb_flags |= XB_OPEN;
614 	sc->xb_info->users++;
615 	return (0);
616 }
617 
618 static int
619 blkif_close(struct disk *dp)
620 {
621 	struct xb_softc	*sc = (struct xb_softc *)dp->d_drv1;
622 
623 	if (sc == NULL)
624 		return (ENXIO);
625 	sc->xb_flags &= ~XB_OPEN;
626 	if (--(sc->xb_info->users) == 0) {
627 		/* Check whether we have been instructed to close.  We will
628 		   have ignored this request initially, as the device was
629 		   still mounted. */
630 		struct xenbus_device * dev = sc->xb_info->xbdev;
631 		XenbusState state = xenbus_read_driver_state(dev->otherend);
632 
633 		if (state == XenbusStateClosing)
634 			blkfront_closing(dev);
635 	}
636 	return (0);
637 }
638 
639 static int
640 blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
641 {
642 	struct xb_softc	*sc = (struct xb_softc *)dp->d_drv1;
643 
644 	if (sc == NULL)
645 		return (ENXIO);
646 
647 	return (ENOTTY);
648 }
649 
650 
651 /*
652  * blkif_queue_request
653  *
654  * request block io
655  *
656  * id: for guest use only.
657  * operation: BLKIF_OP_{READ,WRITE,PROBE}
658  * buffer: buffer to read/write into. this should be a
659  *   virtual address in the guest os.
660  */
661 static int blkif_queue_request(struct bio *bp)
662 {
663 	caddr_t alignbuf;
664 	vm_paddr_t buffer_ma;
665 	blkif_request_t     *ring_req;
666 	unsigned long id;
667 	uint64_t fsect, lsect;
668 	struct xb_softc *sc = (struct xb_softc *)bp->bio_disk->d_drv1;
669 	struct blkfront_info *info = sc->xb_info;
670 	int ref;
671 
672 	if (unlikely(sc->xb_info->connected != BLKIF_STATE_CONNECTED))
673 		return 1;
674 
675 	if (gnttab_alloc_grant_references(
676 		    BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
677 		gnttab_request_free_callback(
678 			&info->callback,
679 			blkif_restart_queue_callback,
680 			info,
681 			BLKIF_MAX_SEGMENTS_PER_REQUEST);
682 		return 1;
683 	}
684 
685 	/* Check if the buffer is properly aligned */
686 	if ((vm_offset_t)bp->bio_data & PAGE_MASK) {
687 		int align = (bp->bio_bcount < PAGE_SIZE/2) ? XBD_SECTOR_SIZE :
688 			PAGE_SIZE;
689 		caddr_t newbuf = malloc(bp->bio_bcount + align, M_DEVBUF,
690 					M_NOWAIT);
691 
692 		alignbuf = (char *)roundup2((u_long)newbuf, align);
693 
694 		/* save a copy of the current buffer */
695 		bp->bio_driver1 = newbuf;
696 		bp->bio_driver2 = alignbuf;
697 
698 		/* Copy the data for a write */
699 		if (bp->bio_cmd == BIO_WRITE)
700 			bcopy(bp->bio_data, alignbuf, bp->bio_bcount);
701 	} else
702 		alignbuf = bp->bio_data;
703 
704 	/* Fill out a communications ring structure. */
705 	ring_req 	         = RING_GET_REQUEST(&info->ring,
706 						    info->ring.req_prod_pvt);
707 	id		         = GET_ID_FROM_FREELIST(info);
708 	info->shadow[id].request = (unsigned long)bp;
709 
710 	ring_req->id 	         = id;
711 	ring_req->operation 	 = (bp->bio_cmd == BIO_READ) ? BLKIF_OP_READ :
712 		BLKIF_OP_WRITE;
713 
714 	ring_req->sector_number= (blkif_sector_t)bp->bio_pblkno;
715 	ring_req->handle 	  = (blkif_vdev_t)(uintptr_t)sc->xb_disk;
716 
717 	ring_req->nr_segments  = 0;	/* XXX not doing scatter/gather since buffer
718 					 * chaining is not supported.
719 					 */
720 
721 	buffer_ma = vtomach(alignbuf);
722 	fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
723 	lsect = fsect + (bp->bio_bcount >> XBD_SECTOR_SHFT) - 1;
724 	/* install a grant reference. */
725 	ref = gnttab_claim_grant_reference(&gref_head);
726 	KASSERT( ref != -ENOSPC, ("grant_reference failed") );
727 
728 	gnttab_grant_foreign_access_ref(
729 		ref,
730 		info->xbdev->otherend_id,
731 		buffer_ma >> PAGE_SHIFT,
732 		ring_req->operation & 1 ); /* ??? */
733 	info->shadow[id].frame[ring_req->nr_segments] =
734 		buffer_ma >> PAGE_SHIFT;
735 
736 	ring_req->seg[ring_req->nr_segments] =
737 		(struct blkif_request_segment) {
738 			.gref       = ref,
739 			.first_sect = fsect,
740 			.last_sect  = lsect };
741 
742 	ring_req->nr_segments++;
743 	KASSERT((buffer_ma & (XBD_SECTOR_SIZE-1)) == 0,
744 		("XEN buffer must be sector aligned"));
745 	KASSERT(lsect <= 7,
746 		("XEN disk driver data cannot cross a page boundary"));
747 
748 	buffer_ma &= ~PAGE_MASK;
749 
750 	info->ring.req_prod_pvt++;
751 
752 	/* Keep a private copy so we can reissue requests when recovering. */
753 	info->shadow[id].req = *ring_req;
754 
755 	gnttab_free_grant_references(gref_head);
756 
757 	return 0;
758 }
759 
760 
761 
762 /*
763  * Dequeue buffers and place them in the shared communication ring.
764  * Return when no more requests can be accepted or all buffers have
765  * been queued.
766  *
767  * Signal XEN once the ring has been filled out.
768  */
769 static void
770 xb_startio(struct xb_softc *sc)
771 {
772 	struct bio		*bp;
773 	int			queued = 0;
774 	struct blkfront_info *info = sc->xb_info;
775 	DPRINTK("");
776 
777 	mtx_assert(&blkif_io_lock, MA_OWNED);
778 
779 	while ((bp = bioq_takefirst(&sc->xb_bioq)) != NULL) {
780 
781 		if (RING_FULL(&info->ring))
782 			goto wait;
783 
784 		if (blkif_queue_request(bp)) {
785 		wait:
786 			bioq_insert_head(&sc->xb_bioq, bp);
787 			break;
788 		}
789 		queued++;
790 	}
791 
792 	if (queued != 0)
793 		flush_requests(sc->xb_info);
794 }
795 
796 static void
797 blkif_int(void *xsc)
798 {
799 	struct xb_softc *sc = NULL;
800 	struct bio *bp;
801 	blkif_response_t *bret;
802 	RING_IDX i, rp;
803 	struct blkfront_info *info = xsc;
804 	DPRINTK("");
805 
806 	TRACE_ENTER;
807 
808 	mtx_lock(&blkif_io_lock);
809 
810 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
811 		mtx_unlock(&blkif_io_lock);
812 		return;
813 	}
814 
815  again:
816 	rp = info->ring.sring->rsp_prod;
817 	rmb(); /* Ensure we see queued responses up to 'rp'. */
818 
819 	for (i = info->ring.rsp_cons; i != rp; i++) {
820 		unsigned long id;
821 
822 		bret = RING_GET_RESPONSE(&info->ring, i);
823 		id   = bret->id;
824 		bp   = (struct bio *)info->shadow[id].request;
825 
826 		blkif_completion(&info->shadow[id]);
827 
828 		ADD_ID_TO_FREELIST(info, id);
829 
830 		switch (bret->operation) {
831 		case BLKIF_OP_READ:
832 			/* had an unaligned buffer that needs to be copied */
833 			if (bp->bio_driver1)
834 				bcopy(bp->bio_driver2, bp->bio_data, bp->bio_bcount);
835 			/* FALLTHROUGH */
836 		case BLKIF_OP_WRITE:
837 
838 			/* free the copy buffer */
839 			if (bp->bio_driver1) {
840 				free(bp->bio_driver1, M_DEVBUF);
841 				bp->bio_driver1 = NULL;
842 			}
843 
844 			if ( unlikely(bret->status != BLKIF_RSP_OKAY) ) {
845 					printf("Bad return from blkdev data request: %x\n",
846 					  bret->status);
847 				bp->bio_flags |= BIO_ERROR;
848 			}
849 
850 			sc = (struct xb_softc *)bp->bio_disk->d_drv1;
851 
852 			if (bp->bio_flags & BIO_ERROR)
853 				bp->bio_error = EIO;
854 			else
855 				bp->bio_resid = 0;
856 
857 			biodone(bp);
858 			break;
859 		default:
860 			panic("received invalid operation");
861 			break;
862 		}
863 	}
864 
865 	info->ring.rsp_cons = i;
866 
867 	if (i != info->ring.req_prod_pvt) {
868 		int more_to_do;
869 		RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
870 		if (more_to_do)
871 			goto again;
872 	} else {
873 		info->ring.sring->rsp_event = i + 1;
874 	}
875 
876 	kick_pending_request_queues(info);
877 
878 	mtx_unlock(&blkif_io_lock);
879 }
880 
881 static void
882 blkif_free(struct blkfront_info *info, int suspend)
883 {
884 
885 /* Prevent new requests being issued until we fix things up. */
886 	mtx_lock(&blkif_io_lock);
887 	info->connected = suspend ?
888 		BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
889 	mtx_unlock(&blkif_io_lock);
890 
891 	/* Free resources associated with old device channel. */
892 	if (info->ring_ref != GRANT_INVALID_REF) {
893 		gnttab_end_foreign_access(info->ring_ref, 0,
894 					  info->ring.sring);
895 		info->ring_ref = GRANT_INVALID_REF;
896 		info->ring.sring = NULL;
897 	}
898 	if (info->irq)
899 		unbind_from_irqhandler(info->irq, info);
900 	info->irq = 0;
901 
902 }
903 
904 static void
905 blkif_completion(struct blk_shadow *s)
906 {
907 	int i;
908 
909 	for (i = 0; i < s->req.nr_segments; i++)
910 		gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
911 }
912 
913 static void
914 blkif_recover(struct blkfront_info *info)
915 {
916 	int i, j;
917 	blkif_request_t *req;
918 	struct blk_shadow *copy;
919 
920 	/* Stage 1: Make a safe copy of the shadow state. */
921 	copy = (struct blk_shadow *)malloc(sizeof(info->shadow), M_DEVBUF, M_NOWAIT|M_ZERO);
922 	PANIC_IF(copy == NULL);
923 	memcpy(copy, info->shadow, sizeof(info->shadow));
924 
925 	/* Stage 2: Set up free list. */
926 	memset(&info->shadow, 0, sizeof(info->shadow));
927 	for (i = 0; i < BLK_RING_SIZE; i++)
928 		info->shadow[i].req.id = i+1;
929 	info->shadow_free = info->ring.req_prod_pvt;
930 	info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
931 
932 	/* Stage 3: Find pending requests and requeue them. */
933 	for (i = 0; i < BLK_RING_SIZE; i++) {
934 		/* Not in use? */
935 		if (copy[i].request == 0)
936 			continue;
937 
938 		/* Grab a request slot and copy shadow state into it. */
939 		req = RING_GET_REQUEST(
940 			&info->ring, info->ring.req_prod_pvt);
941 		*req = copy[i].req;
942 
943 		/* We get a new request id, and must reset the shadow state. */
944 		req->id = GET_ID_FROM_FREELIST(info);
945 		memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
946 
947 		/* Rewrite any grant references invalidated by suspend/resume. */
948 		for (j = 0; j < req->nr_segments; j++)
949 			gnttab_grant_foreign_access_ref(
950 				req->seg[j].gref,
951 				info->xbdev->otherend_id,
952 				pfn_to_mfn(info->shadow[req->id].frame[j]),
953 				0 /* assume not readonly */);
954 
955 		info->shadow[req->id].req = *req;
956 
957 		info->ring.req_prod_pvt++;
958 	}
959 
960 	free(copy, M_DEVBUF);
961 
962 	xenbus_switch_state(info->xbdev, XenbusStateConnected);
963 
964 	/* Now safe for us to use the shared ring */
965 	mtx_lock(&blkif_io_lock);
966 	info->connected = BLKIF_STATE_CONNECTED;
967 	mtx_unlock(&blkif_io_lock);
968 
969 	/* Send off requeued requests */
970 	mtx_lock(&blkif_io_lock);
971 	flush_requests(info);
972 
973 	/* Kick any other new requests queued since we resumed */
974 	kick_pending_request_queues(info);
975 	mtx_unlock(&blkif_io_lock);
976 }
977 
978 static int
979 blkfront_is_ready(struct xenbus_device *dev)
980 {
981 	struct blkfront_info *info = dev->dev_driver_data;
982 
983 	return info->is_ready;
984 }
985 
986 static struct xenbus_device_id blkfront_ids[] = {
987 	{ "vbd" },
988 	{ "" }
989 };
990 
991 
992 static struct xenbus_driver blkfront = {
993 	.name             = "vbd",
994 	.ids              = blkfront_ids,
995 	.probe            = blkfront_probe,
996 	.remove           = blkfront_remove,
997 	.resume           = blkfront_resume,
998 	.otherend_changed = backend_changed,
999 	.is_ready		  = blkfront_is_ready,
1000 };
1001 
1002 
1003 
1004 static void
1005 xenbus_init(void)
1006 {
1007 	xenbus_register_frontend(&blkfront);
1008 }
1009 
1010 MTX_SYSINIT(ioreq, &blkif_io_lock, "BIO LOCK", MTX_NOWITNESS); /* XXX how does one enroll a lock? */
1011 SYSINIT(xbdev, SI_SUB_PSEUDO, SI_ORDER_SECOND, xenbus_init, NULL);
1012 
1013 
1014 /*
1015  * Local variables:
1016  * mode: C
1017  * c-set-style: "BSD"
1018  * c-basic-offset: 8
1019  * tab-width: 4
1020  * indent-tabs-mode: t
1021  * End:
1022  */
1023