2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions, and the following
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 /* Driver for VirtIO block devices. */
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/sglist.h>
39 #include <sys/sysctl.h>
41 #include <sys/mutex.h>
42 #include <sys/queue.h>
44 #include <geom/geom_disk.h>
46 #include <machine/bus.h>
47 #include <machine/resource.h>
51 #include <dev/virtio/virtio.h>
52 #include <dev/virtio/virtqueue.h>
53 #include <dev/virtio/block/virtio_blk.h>
55 #include "virtio_if.h"
57 struct vtblk_request {
58 struct virtio_blk_outhdr vbr_hdr;
62 TAILQ_ENTRY(vtblk_request) vbr_link;
65 enum vtblk_cache_mode {
66 VTBLK_CACHE_WRITETHROUGH,
67 VTBLK_CACHE_WRITEBACK,
74 uint64_t vtblk_features;
76 #define VTBLK_FLAG_INDIRECT 0x0001
77 #define VTBLK_FLAG_READONLY 0x0002
78 #define VTBLK_FLAG_DETACH 0x0004
79 #define VTBLK_FLAG_SUSPEND 0x0008
80 #define VTBLK_FLAG_DUMPING 0x0010
81 #define VTBLK_FLAG_BARRIER 0x0020
82 #define VTBLK_FLAG_WC_CONFIG 0x0040
84 struct virtqueue *vtblk_vq;
85 struct sglist *vtblk_sglist;
86 struct disk *vtblk_disk;
88 struct bio_queue_head vtblk_bioq;
89 TAILQ_HEAD(, vtblk_request)
91 TAILQ_HEAD(, vtblk_request)
93 struct vtblk_request *vtblk_req_ordered;
96 int vtblk_request_count;
97 enum vtblk_cache_mode vtblk_write_cache;
99 struct vtblk_request vtblk_dump_request;
102 static struct virtio_feature_desc vtblk_feature_desc[] = {
103 { VIRTIO_BLK_F_BARRIER, "HostBarrier" },
104 { VIRTIO_BLK_F_SIZE_MAX, "MaxSegSize" },
105 { VIRTIO_BLK_F_SEG_MAX, "MaxNumSegs" },
106 { VIRTIO_BLK_F_GEOMETRY, "DiskGeometry" },
107 { VIRTIO_BLK_F_RO, "ReadOnly" },
108 { VIRTIO_BLK_F_BLK_SIZE, "BlockSize" },
109 { VIRTIO_BLK_F_SCSI, "SCSICmds" },
110 { VIRTIO_BLK_F_WCE, "WriteCache" },
111 { VIRTIO_BLK_F_TOPOLOGY, "Topology" },
112 { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" },
117 static int vtblk_modevent(module_t, int, void *);
119 static int vtblk_probe(device_t);
120 static int vtblk_attach(device_t);
121 static int vtblk_detach(device_t);
122 static int vtblk_suspend(device_t);
123 static int vtblk_resume(device_t);
124 static int vtblk_shutdown(device_t);
125 static int vtblk_config_change(device_t);
127 static int vtblk_open(struct disk *);
128 static int vtblk_close(struct disk *);
129 static int vtblk_ioctl(struct disk *, u_long, void *, int,
131 static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t);
132 static void vtblk_strategy(struct bio *);
134 static void vtblk_negotiate_features(struct vtblk_softc *);
135 static int vtblk_maximum_segments(struct vtblk_softc *,
136 struct virtio_blk_config *);
137 static int vtblk_alloc_virtqueue(struct vtblk_softc *);
138 static void vtblk_set_write_cache(struct vtblk_softc *, int);
139 static int vtblk_write_cache_enabled(struct vtblk_softc *sc,
140 struct virtio_blk_config *);
141 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS);
142 static void vtblk_alloc_disk(struct vtblk_softc *,
143 struct virtio_blk_config *);
144 static void vtblk_create_disk(struct vtblk_softc *);
146 static int vtblk_quiesce(struct vtblk_softc *);
147 static void vtblk_startio(struct vtblk_softc *);
148 static struct vtblk_request * vtblk_bio_request(struct vtblk_softc *);
149 static int vtblk_execute_request(struct vtblk_softc *,
150 struct vtblk_request *);
152 static void vtblk_vq_intr(void *);
154 static void vtblk_stop(struct vtblk_softc *);
156 static void vtblk_read_config(struct vtblk_softc *,
157 struct virtio_blk_config *);
158 static void vtblk_get_ident(struct vtblk_softc *);
159 static void vtblk_prepare_dump(struct vtblk_softc *);
160 static int vtblk_write_dump(struct vtblk_softc *, void *, off_t, size_t);
161 static int vtblk_flush_dump(struct vtblk_softc *);
162 static int vtblk_poll_request(struct vtblk_softc *,
163 struct vtblk_request *);
165 static void vtblk_finish_completed(struct vtblk_softc *);
166 static void vtblk_drain_vq(struct vtblk_softc *, int);
167 static void vtblk_drain(struct vtblk_softc *);
169 static int vtblk_alloc_requests(struct vtblk_softc *);
170 static void vtblk_free_requests(struct vtblk_softc *);
171 static struct vtblk_request * vtblk_dequeue_request(struct vtblk_softc *);
172 static void vtblk_enqueue_request(struct vtblk_softc *,
173 struct vtblk_request *);
175 static struct vtblk_request * vtblk_dequeue_ready(struct vtblk_softc *);
176 static void vtblk_enqueue_ready(struct vtblk_softc *,
177 struct vtblk_request *);
179 static int vtblk_request_error(struct vtblk_request *);
180 static void vtblk_finish_bio(struct bio *, int);
182 static void vtblk_setup_sysctl(struct vtblk_softc *);
183 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int);
186 static int vtblk_no_ident = 0;
187 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident);
188 static int vtblk_writecache_mode = -1;
189 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode);
191 /* Features desired/implemented by this driver. */
192 #define VTBLK_FEATURES \
193 (VIRTIO_BLK_F_BARRIER | \
194 VIRTIO_BLK_F_SIZE_MAX | \
195 VIRTIO_BLK_F_SEG_MAX | \
196 VIRTIO_BLK_F_GEOMETRY | \
198 VIRTIO_BLK_F_BLK_SIZE | \
200 VIRTIO_BLK_F_CONFIG_WCE | \
201 VIRTIO_RING_F_INDIRECT_DESC)
203 #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx
204 #define VTBLK_LOCK_INIT(_sc, _name) \
205 mtx_init(VTBLK_MTX((_sc)), (_name), \
206 "VirtIO Block Lock", MTX_DEF)
207 #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc)))
208 #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc)))
209 #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc)))
210 #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED)
211 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \
212 mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED)
214 #define VTBLK_DISK_NAME "vtbd"
215 #define VTBLK_QUIESCE_TIMEOUT (30 * hz)
218 * Each block request uses at least two segments - one for the header
219 * and one for the status.
221 #define VTBLK_MIN_SEGMENTS 2
223 static device_method_t vtblk_methods[] = {
224 /* Device methods. */
225 DEVMETHOD(device_probe, vtblk_probe),
226 DEVMETHOD(device_attach, vtblk_attach),
227 DEVMETHOD(device_detach, vtblk_detach),
228 DEVMETHOD(device_suspend, vtblk_suspend),
229 DEVMETHOD(device_resume, vtblk_resume),
230 DEVMETHOD(device_shutdown, vtblk_shutdown),
232 /* VirtIO methods. */
233 DEVMETHOD(virtio_config_change, vtblk_config_change),
238 static driver_t vtblk_driver = {
241 sizeof(struct vtblk_softc)
243 static devclass_t vtblk_devclass;
245 DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass,
247 MODULE_VERSION(virtio_blk, 1);
248 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1);
251 vtblk_modevent(module_t mod, int type, void *unused)
272 vtblk_probe(device_t dev)
275 if (virtio_get_device_type(dev) != VIRTIO_ID_BLOCK)
278 device_set_desc(dev, "VirtIO Block Adapter");
280 return (BUS_PROBE_DEFAULT);
284 vtblk_attach(device_t dev)
286 struct vtblk_softc *sc;
287 struct virtio_blk_config blkcfg;
290 sc = device_get_softc(dev);
293 VTBLK_LOCK_INIT(sc, device_get_nameunit(dev));
295 bioq_init(&sc->vtblk_bioq);
296 TAILQ_INIT(&sc->vtblk_req_free);
297 TAILQ_INIT(&sc->vtblk_req_ready);
299 virtio_set_feature_desc(dev, vtblk_feature_desc);
300 vtblk_negotiate_features(sc);
302 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
303 sc->vtblk_flags |= VTBLK_FLAG_INDIRECT;
304 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO))
305 sc->vtblk_flags |= VTBLK_FLAG_READONLY;
306 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER))
307 sc->vtblk_flags |= VTBLK_FLAG_BARRIER;
308 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE))
309 sc->vtblk_flags |= VTBLK_FLAG_WC_CONFIG;
311 vtblk_setup_sysctl(sc);
313 /* Get local copy of config. */
314 vtblk_read_config(sc, &blkcfg);
317 * With the current sglist(9) implementation, it is not easy
318 * for us to support a maximum segment size as adjacent
319 * segments are coalesced. For now, just make sure it's larger
320 * than the maximum supported transfer size.
322 if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) {
323 if (blkcfg.size_max < MAXPHYS) {
325 device_printf(dev, "host requires unsupported "
326 "maximum segment size feature\n");
331 sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg);
332 if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) {
334 device_printf(dev, "fewer than minimum number of segments "
335 "allowed: %d\n", sc->vtblk_max_nsegs);
339 sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT);
340 if (sc->vtblk_sglist == NULL) {
342 device_printf(dev, "cannot allocate sglist\n");
346 error = vtblk_alloc_virtqueue(sc);
348 device_printf(dev, "cannot allocate virtqueue\n");
352 error = vtblk_alloc_requests(sc);
354 device_printf(dev, "cannot preallocate requests\n");
358 vtblk_alloc_disk(sc, &blkcfg);
360 error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY);
362 device_printf(dev, "cannot setup virtqueue interrupt\n");
366 vtblk_create_disk(sc);
368 virtqueue_enable_intr(sc->vtblk_vq);
378 vtblk_detach(device_t dev)
380 struct vtblk_softc *sc;
382 sc = device_get_softc(dev);
385 sc->vtblk_flags |= VTBLK_FLAG_DETACH;
386 if (device_is_attached(dev))
392 if (sc->vtblk_disk != NULL) {
393 disk_destroy(sc->vtblk_disk);
394 sc->vtblk_disk = NULL;
397 if (sc->vtblk_sglist != NULL) {
398 sglist_free(sc->vtblk_sglist);
399 sc->vtblk_sglist = NULL;
402 VTBLK_LOCK_DESTROY(sc);
408 vtblk_suspend(device_t dev)
410 struct vtblk_softc *sc;
413 sc = device_get_softc(dev);
416 sc->vtblk_flags |= VTBLK_FLAG_SUSPEND;
417 /* XXX BMV: virtio_stop(), etc needed here? */
418 error = vtblk_quiesce(sc);
420 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
427 vtblk_resume(device_t dev)
429 struct vtblk_softc *sc;
431 sc = device_get_softc(dev);
434 /* XXX BMV: virtio_reinit(), etc needed here? */
435 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
443 vtblk_shutdown(device_t dev)
450 vtblk_config_change(device_t dev)
457 vtblk_open(struct disk *dp)
459 struct vtblk_softc *sc;
461 if ((sc = dp->d_drv1) == NULL)
464 return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0);
468 vtblk_close(struct disk *dp)
470 struct vtblk_softc *sc;
472 if ((sc = dp->d_drv1) == NULL)
479 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag,
482 struct vtblk_softc *sc;
484 if ((sc = dp->d_drv1) == NULL)
491 vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
495 struct vtblk_softc *sc;
500 if ((sc = dp->d_drv1) == NULL)
505 if ((sc->vtblk_flags & VTBLK_FLAG_DUMPING) == 0) {
506 vtblk_prepare_dump(sc);
507 sc->vtblk_flags |= VTBLK_FLAG_DUMPING;
511 error = vtblk_write_dump(sc, virtual, offset, length);
512 else if (virtual == NULL && offset == 0)
513 error = vtblk_flush_dump(sc);
516 sc->vtblk_flags &= ~VTBLK_FLAG_DUMPING;
525 vtblk_strategy(struct bio *bp)
527 struct vtblk_softc *sc;
529 if ((sc = bp->bio_disk->d_drv1) == NULL) {
530 vtblk_finish_bio(bp, EINVAL);
535 * Fail any write if RO. Unfortunately, there does not seem to
536 * be a better way to report our readonly'ness to GEOM above.
538 if (sc->vtblk_flags & VTBLK_FLAG_READONLY &&
539 (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_FLUSH)) {
540 vtblk_finish_bio(bp, EROFS);
546 * Prevent read/write buffers spanning too many segments from
547 * getting into the queue. This should only trip if d_maxsize
548 * was incorrectly set.
550 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
551 int nsegs, max_nsegs;
553 nsegs = sglist_count(bp->bio_data, bp->bio_bcount);
554 max_nsegs = sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS;
556 KASSERT(nsegs <= max_nsegs,
557 ("%s: bio %p spanned too many segments: %d, max: %d",
558 __func__, bp, nsegs, max_nsegs));
563 if (sc->vtblk_flags & VTBLK_FLAG_DETACH)
564 vtblk_finish_bio(bp, ENXIO);
566 bioq_disksort(&sc->vtblk_bioq, bp);
568 if ((sc->vtblk_flags & VTBLK_FLAG_SUSPEND) == 0)
575 vtblk_negotiate_features(struct vtblk_softc *sc)
581 features = VTBLK_FEATURES;
583 sc->vtblk_features = virtio_negotiate_features(dev, features);
587 vtblk_maximum_segments(struct vtblk_softc *sc,
588 struct virtio_blk_config *blkcfg)
594 nsegs = VTBLK_MIN_SEGMENTS;
596 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) {
597 nsegs += MIN(blkcfg->seg_max, MAXPHYS / PAGE_SIZE + 1);
598 if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT)
599 nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT);
607 vtblk_alloc_virtqueue(struct vtblk_softc *sc)
610 struct vq_alloc_info vq_info;
614 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs,
615 vtblk_vq_intr, sc, &sc->vtblk_vq,
616 "%s request", device_get_nameunit(dev));
618 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info));
622 vtblk_set_write_cache(struct vtblk_softc *sc, int wc)
625 /* Set either writeback (1) or writethrough (0) mode. */
626 virtio_write_dev_config_1(sc->vtblk_dev,
627 offsetof(struct virtio_blk_config, writeback), wc);
631 vtblk_write_cache_enabled(struct vtblk_softc *sc,
632 struct virtio_blk_config *blkcfg)
636 if (sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) {
637 wc = vtblk_tunable_int(sc, "writecache_mode",
638 vtblk_writecache_mode);
639 if (wc >= 0 && wc < VTBLK_CACHE_MAX)
640 vtblk_set_write_cache(sc, wc);
642 wc = blkcfg->writeback;
644 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_WCE);
650 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS)
652 struct vtblk_softc *sc;
656 wc = sc->vtblk_write_cache;
658 error = sysctl_handle_int(oidp, &wc, 0, req);
659 if (error || req->newptr == NULL)
661 if ((sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) == 0)
663 if (wc < 0 || wc >= VTBLK_CACHE_MAX)
667 sc->vtblk_write_cache = wc;
668 vtblk_set_write_cache(sc, sc->vtblk_write_cache);
675 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
682 sc->vtblk_disk = dp = disk_alloc();
683 dp->d_open = vtblk_open;
684 dp->d_close = vtblk_close;
685 dp->d_ioctl = vtblk_ioctl;
686 dp->d_strategy = vtblk_strategy;
687 dp->d_name = VTBLK_DISK_NAME;
688 dp->d_unit = device_get_unit(dev);
690 dp->d_flags = DISKFLAG_CANFLUSHCACHE;
691 dp->d_hba_vendor = virtio_get_vendor(dev);
692 dp->d_hba_device = virtio_get_device(dev);
693 dp->d_hba_subvendor = virtio_get_subvendor(dev);
694 dp->d_hba_subdevice = virtio_get_subdevice(dev);
696 if ((sc->vtblk_flags & VTBLK_FLAG_READONLY) == 0)
697 dp->d_dump = vtblk_dump;
699 /* Capacity is always in 512-byte units. */
700 dp->d_mediasize = blkcfg->capacity * 512;
702 if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE))
703 dp->d_sectorsize = blkcfg->blk_size;
705 dp->d_sectorsize = 512;
708 * The VirtIO maximum I/O size is given in terms of segments.
709 * However, FreeBSD limits I/O size by logical buffer size, not
710 * by physically contiguous pages. Therefore, we have to assume
711 * no pages are contiguous. This may impose an artificially low
712 * maximum I/O size. But in practice, since QEMU advertises 128
713 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE,
714 * which is typically greater than MAXPHYS. Eventually we should
715 * just advertise MAXPHYS and split buffers that are too big.
717 * Note we must subtract one additional segment in case of non
718 * page aligned buffers.
720 dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) *
722 if (dp->d_maxsize < PAGE_SIZE)
723 dp->d_maxsize = PAGE_SIZE; /* XXX */
725 if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) {
726 dp->d_fwsectors = blkcfg->geometry.sectors;
727 dp->d_fwheads = blkcfg->geometry.heads;
730 if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY)) {
731 dp->d_stripesize = dp->d_sectorsize *
732 (1 << blkcfg->topology.physical_block_exp);
733 dp->d_stripeoffset = (dp->d_stripesize -
734 blkcfg->topology.alignment_offset * dp->d_sectorsize) %
738 if (vtblk_write_cache_enabled(sc, blkcfg) != 0)
739 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK;
741 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH;
745 vtblk_create_disk(struct vtblk_softc *sc)
752 * Retrieving the identification string must be done after
753 * the virtqueue interrupt is setup otherwise it will hang.
757 device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n",
758 (uintmax_t) dp->d_mediasize >> 20,
759 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
762 disk_create(dp, DISK_VERSION);
766 vtblk_quiesce(struct vtblk_softc *sc)
772 VTBLK_LOCK_ASSERT(sc);
774 while (!virtqueue_empty(sc->vtblk_vq)) {
775 if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq",
776 VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) {
786 vtblk_startio(struct vtblk_softc *sc)
788 struct virtqueue *vq;
789 struct vtblk_request *req;
795 VTBLK_LOCK_ASSERT(sc);
797 while (!virtqueue_full(vq)) {
798 if ((req = vtblk_dequeue_ready(sc)) == NULL)
799 req = vtblk_bio_request(sc);
803 if (vtblk_execute_request(sc, req) != 0) {
804 vtblk_enqueue_ready(sc, req);
812 virtqueue_notify(vq);
815 static struct vtblk_request *
816 vtblk_bio_request(struct vtblk_softc *sc)
818 struct bio_queue_head *bioq;
819 struct vtblk_request *req;
822 bioq = &sc->vtblk_bioq;
824 if (bioq_first(bioq) == NULL)
827 req = vtblk_dequeue_request(sc);
831 bp = bioq_takefirst(bioq);
834 req->vbr_hdr.ioprio = 1;
836 switch (bp->bio_cmd) {
838 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
841 req->vbr_hdr.type = VIRTIO_BLK_T_IN;
842 req->vbr_hdr.sector = bp->bio_offset / 512;
845 req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
846 req->vbr_hdr.sector = bp->bio_offset / 512;
849 panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd);
856 vtblk_execute_request(struct vtblk_softc *sc, struct vtblk_request *req)
858 struct virtqueue *vq;
861 int ordered, readable, writable, error;
864 sg = sc->vtblk_sglist;
869 VTBLK_LOCK_ASSERT(sc);
872 * Wait until the ordered request completes before
873 * executing subsequent requests.
875 if (sc->vtblk_req_ordered != NULL)
878 if (bp->bio_flags & BIO_ORDERED) {
879 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) {
881 * This request will be executed once all
882 * the in-flight requests are completed.
884 if (!virtqueue_empty(vq))
888 req->vbr_hdr.type |= VIRTIO_BLK_T_BARRIER;
892 sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr));
894 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
895 error = sglist_append(sg, bp->bio_data, bp->bio_bcount);
896 if (error || sg->sg_nseg == sg->sg_maxseg)
897 panic("%s: data buffer too big bio:%p error:%d",
898 __func__, bp, error);
900 /* BIO_READ means the host writes into our buffer. */
901 if (bp->bio_cmd == BIO_READ)
902 writable = sg->sg_nseg - 1;
906 sglist_append(sg, &req->vbr_ack, sizeof(uint8_t));
907 readable = sg->sg_nseg - writable;
909 error = virtqueue_enqueue(vq, req, sg, readable, writable);
910 if (error == 0 && ordered)
911 sc->vtblk_req_ordered = req;
917 vtblk_vq_intr(void *xsc)
919 struct vtblk_softc *sc;
920 struct virtqueue *vq;
927 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) {
932 vtblk_finish_completed(sc);
934 if ((sc->vtblk_flags & VTBLK_FLAG_SUSPEND) == 0)
937 wakeup(&sc->vtblk_vq);
939 if (virtqueue_enable_intr(vq) != 0) {
940 virtqueue_disable_intr(vq);
949 vtblk_stop(struct vtblk_softc *sc)
952 virtqueue_disable_intr(sc->vtblk_vq);
953 virtio_stop(sc->vtblk_dev);
956 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \
957 if (virtio_with_feature(_dev, _feature)) { \
958 virtio_read_device_config(_dev, \
959 offsetof(struct virtio_blk_config, _field), \
960 &(_cfg)->_field, sizeof((_cfg)->_field)); \
964 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
970 bzero(blkcfg, sizeof(struct virtio_blk_config));
972 /* The capacity is always available. */
973 virtio_read_device_config(dev, offsetof(struct virtio_blk_config,
974 capacity), &blkcfg->capacity, sizeof(blkcfg->capacity));
976 /* Read the configuration if the feature was negotiated. */
977 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg);
978 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg);
979 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, geometry, blkcfg);
980 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg);
981 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, topology, blkcfg);
982 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, writeback, blkcfg);
985 #undef VTBLK_GET_CONFIG
988 vtblk_get_ident(struct vtblk_softc *sc)
992 struct vtblk_request *req;
996 len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE);
998 if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0)
1001 req = vtblk_dequeue_request(sc);
1006 req->vbr_hdr.type = VIRTIO_BLK_T_GET_ID;
1007 req->vbr_hdr.ioprio = 1;
1008 req->vbr_hdr.sector = 0;
1011 bzero(&buf, sizeof(struct bio));
1013 buf.bio_cmd = BIO_READ;
1014 buf.bio_data = dp->d_ident;
1015 buf.bio_bcount = len;
1018 error = vtblk_poll_request(sc, req);
1021 vtblk_enqueue_request(sc, req);
1024 device_printf(sc->vtblk_dev,
1025 "error getting device identifier: %d\n", error);
1030 vtblk_prepare_dump(struct vtblk_softc *sc)
1033 struct virtqueue *vq;
1035 dev = sc->vtblk_dev;
1041 * Drain all requests caught in-flight in the virtqueue,
1042 * skipping biodone(). When dumping, only one request is
1043 * outstanding at a time, and we just poll the virtqueue
1046 vtblk_drain_vq(sc, 1);
1048 if (virtio_reinit(dev, sc->vtblk_features) != 0) {
1049 panic("%s: cannot reinit VirtIO block device during dump",
1050 device_get_nameunit(dev));
1053 virtqueue_disable_intr(vq);
1054 virtio_reinit_complete(dev);
1058 vtblk_write_dump(struct vtblk_softc *sc, void *virtual, off_t offset,
1062 struct vtblk_request *req;
1064 req = &sc->vtblk_dump_request;
1066 req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
1067 req->vbr_hdr.ioprio = 1;
1068 req->vbr_hdr.sector = offset / 512;
1071 bzero(&buf, sizeof(struct bio));
1073 buf.bio_cmd = BIO_WRITE;
1074 buf.bio_data = virtual;
1075 buf.bio_bcount = length;
1077 return (vtblk_poll_request(sc, req));
1081 vtblk_flush_dump(struct vtblk_softc *sc)
1084 struct vtblk_request *req;
1086 req = &sc->vtblk_dump_request;
1088 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
1089 req->vbr_hdr.ioprio = 1;
1090 req->vbr_hdr.sector = 0;
1093 bzero(&buf, sizeof(struct bio));
1095 buf.bio_cmd = BIO_FLUSH;
1097 return (vtblk_poll_request(sc, req));
1101 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req)
1103 struct virtqueue *vq;
1108 if (!virtqueue_empty(vq))
1111 error = vtblk_execute_request(sc, req);
1115 virtqueue_notify(vq);
1116 virtqueue_poll(vq, NULL);
1118 error = vtblk_request_error(req);
1119 if (error && bootverbose) {
1120 device_printf(sc->vtblk_dev,
1121 "%s: IO error: %d\n", __func__, error);
1128 vtblk_finish_completed(struct vtblk_softc *sc)
1130 struct vtblk_request *req;
1134 while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) {
1137 if (sc->vtblk_req_ordered != NULL) {
1138 /* This should be the only outstanding request. */
1139 MPASS(sc->vtblk_req_ordered == req);
1140 sc->vtblk_req_ordered = NULL;
1143 error = vtblk_request_error(req);
1145 disk_err(bp, "hard error", -1, 1);
1147 vtblk_finish_bio(bp, error);
1148 vtblk_enqueue_request(sc, req);
1153 vtblk_drain_vq(struct vtblk_softc *sc, int skip_done)
1155 struct virtqueue *vq;
1156 struct vtblk_request *req;
1162 while ((req = virtqueue_drain(vq, &last)) != NULL) {
1164 vtblk_finish_bio(req->vbr_bp, ENXIO);
1166 vtblk_enqueue_request(sc, req);
1169 sc->vtblk_req_ordered = NULL;
1170 KASSERT(virtqueue_empty(vq), ("virtqueue not empty"));
1174 vtblk_drain(struct vtblk_softc *sc)
1176 struct bio_queue_head *bioq;
1177 struct vtblk_request *req;
1180 bioq = &sc->vtblk_bioq;
1182 if (sc->vtblk_vq != NULL) {
1183 vtblk_finish_completed(sc);
1184 vtblk_drain_vq(sc, 0);
1187 while ((req = vtblk_dequeue_ready(sc)) != NULL) {
1188 vtblk_finish_bio(req->vbr_bp, ENXIO);
1189 vtblk_enqueue_request(sc, req);
1192 while (bioq_first(bioq) != NULL) {
1193 bp = bioq_takefirst(bioq);
1194 vtblk_finish_bio(bp, ENXIO);
1197 vtblk_free_requests(sc);
1202 vtblk_request_invariants(struct vtblk_request *req)
1204 int hdr_nsegs, ack_nsegs;
1206 hdr_nsegs = sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr));
1207 ack_nsegs = sglist_count(&req->vbr_ack, sizeof(req->vbr_ack));
1209 KASSERT(hdr_nsegs == 1, ("request header crossed page boundary"));
1210 KASSERT(ack_nsegs == 1, ("request ack crossed page boundary"));
1215 vtblk_alloc_requests(struct vtblk_softc *sc)
1217 struct vtblk_request *req;
1220 nreqs = virtqueue_size(sc->vtblk_vq);
1223 * Preallocate sufficient requests to keep the virtqueue full. Each
1224 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce
1225 * the number allocated when indirect descriptors are not available.
1227 if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0)
1228 nreqs /= VTBLK_MIN_SEGMENTS;
1230 for (i = 0; i < nreqs; i++) {
1231 req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT);
1236 vtblk_request_invariants(req);
1239 sc->vtblk_request_count++;
1240 vtblk_enqueue_request(sc, req);
1247 vtblk_free_requests(struct vtblk_softc *sc)
1249 struct vtblk_request *req;
1251 KASSERT(TAILQ_EMPTY(&sc->vtblk_req_ready),
1252 ("%s: ready requests left on queue", __func__));
1254 while ((req = vtblk_dequeue_request(sc)) != NULL) {
1255 sc->vtblk_request_count--;
1256 free(req, M_DEVBUF);
1259 KASSERT(sc->vtblk_request_count == 0,
1260 ("%s: leaked %d requests", __func__, sc->vtblk_request_count));
1263 static struct vtblk_request *
1264 vtblk_dequeue_request(struct vtblk_softc *sc)
1266 struct vtblk_request *req;
1268 req = TAILQ_FIRST(&sc->vtblk_req_free);
1270 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link);
1276 vtblk_enqueue_request(struct vtblk_softc *sc, struct vtblk_request *req)
1279 bzero(req, sizeof(struct vtblk_request));
1280 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link);
1283 static struct vtblk_request *
1284 vtblk_dequeue_ready(struct vtblk_softc *sc)
1286 struct vtblk_request *req;
1288 req = TAILQ_FIRST(&sc->vtblk_req_ready);
1290 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link);
1296 vtblk_enqueue_ready(struct vtblk_softc *sc, struct vtblk_request *req)
1299 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link);
1303 vtblk_request_error(struct vtblk_request *req)
1307 switch (req->vbr_ack) {
1308 case VIRTIO_BLK_S_OK:
1311 case VIRTIO_BLK_S_UNSUPP:
1323 vtblk_finish_bio(struct bio *bp, int error)
1327 bp->bio_resid = bp->bio_bcount;
1328 bp->bio_error = error;
1329 bp->bio_flags |= BIO_ERROR;
1336 vtblk_setup_sysctl(struct vtblk_softc *sc)
1339 struct sysctl_ctx_list *ctx;
1340 struct sysctl_oid *tree;
1341 struct sysctl_oid_list *child;
1343 dev = sc->vtblk_dev;
1344 ctx = device_get_sysctl_ctx(dev);
1345 tree = device_get_sysctl_tree(dev);
1346 child = SYSCTL_CHILDREN(tree);
1348 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode",
1349 CTLTYPE_INT | CTLFLAG_RW, sc, 0, vtblk_write_cache_sysctl,
1350 "I", "Write cache mode (writethrough (0) or writeback (1))");
1354 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def)
1358 snprintf(path, sizeof(path),
1359 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob);
1360 TUNABLE_INT_FETCH(path, &def);