2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 /* Driver for VirtIO block devices. */
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/sglist.h>
41 #include <sys/sysctl.h>
43 #include <sys/mutex.h>
44 #include <sys/queue.h>
46 #include <geom/geom.h>
47 #include <geom/geom_disk.h>
49 #include <machine/bus.h>
50 #include <machine/resource.h>
54 #include <dev/virtio/virtio.h>
55 #include <dev/virtio/virtqueue.h>
56 #include <dev/virtio/block/virtio_blk.h>
58 #include "virtio_if.h"
60 struct vtblk_request {
61 struct virtio_blk_outhdr vbr_hdr;
64 TAILQ_ENTRY(vtblk_request) vbr_link;
67 enum vtblk_cache_mode {
68 VTBLK_CACHE_WRITETHROUGH,
69 VTBLK_CACHE_WRITEBACK,
76 uint64_t vtblk_features;
78 #define VTBLK_FLAG_INDIRECT 0x0001
79 #define VTBLK_FLAG_READONLY 0x0002
80 #define VTBLK_FLAG_DETACH 0x0004
81 #define VTBLK_FLAG_SUSPEND 0x0008
82 #define VTBLK_FLAG_BARRIER 0x0010
83 #define VTBLK_FLAG_WCE_CONFIG 0x0020
85 struct virtqueue *vtblk_vq;
86 struct sglist *vtblk_sglist;
87 struct disk *vtblk_disk;
89 struct bio_queue_head vtblk_bioq;
90 TAILQ_HEAD(, vtblk_request)
92 TAILQ_HEAD(, vtblk_request)
94 struct vtblk_request *vtblk_req_ordered;
97 int vtblk_request_count;
98 enum vtblk_cache_mode vtblk_write_cache;
100 struct bio_queue vtblk_dump_queue;
101 struct vtblk_request vtblk_dump_request;
104 static struct virtio_feature_desc vtblk_feature_desc[] = {
105 { VIRTIO_BLK_F_BARRIER, "HostBarrier" },
106 { VIRTIO_BLK_F_SIZE_MAX, "MaxSegSize" },
107 { VIRTIO_BLK_F_SEG_MAX, "MaxNumSegs" },
108 { VIRTIO_BLK_F_GEOMETRY, "DiskGeometry" },
109 { VIRTIO_BLK_F_RO, "ReadOnly" },
110 { VIRTIO_BLK_F_BLK_SIZE, "BlockSize" },
111 { VIRTIO_BLK_F_SCSI, "SCSICmds" },
112 { VIRTIO_BLK_F_FLUSH, "FlushCmd" },
113 { VIRTIO_BLK_F_TOPOLOGY, "Topology" },
114 { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" },
115 { VIRTIO_BLK_F_MQ, "Multiqueue" },
116 { VIRTIO_BLK_F_DISCARD, "Discard" },
117 { VIRTIO_BLK_F_WRITE_ZEROES, "WriteZeros" },
122 static int vtblk_modevent(module_t, int, void *);
124 static int vtblk_probe(device_t);
125 static int vtblk_attach(device_t);
126 static int vtblk_detach(device_t);
127 static int vtblk_suspend(device_t);
128 static int vtblk_resume(device_t);
129 static int vtblk_shutdown(device_t);
130 static int vtblk_config_change(device_t);
132 static int vtblk_open(struct disk *);
133 static int vtblk_close(struct disk *);
134 static int vtblk_ioctl(struct disk *, u_long, void *, int,
136 static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t);
137 static void vtblk_strategy(struct bio *);
139 static void vtblk_negotiate_features(struct vtblk_softc *);
140 static void vtblk_setup_features(struct vtblk_softc *);
141 static int vtblk_maximum_segments(struct vtblk_softc *,
142 struct virtio_blk_config *);
143 static int vtblk_alloc_virtqueue(struct vtblk_softc *);
144 static void vtblk_resize_disk(struct vtblk_softc *, uint64_t);
145 static void vtblk_alloc_disk(struct vtblk_softc *,
146 struct virtio_blk_config *);
147 static void vtblk_create_disk(struct vtblk_softc *);
149 static int vtblk_request_prealloc(struct vtblk_softc *);
150 static void vtblk_request_free(struct vtblk_softc *);
151 static struct vtblk_request *
152 vtblk_request_dequeue(struct vtblk_softc *);
153 static void vtblk_request_enqueue(struct vtblk_softc *,
154 struct vtblk_request *);
155 static struct vtblk_request *
156 vtblk_request_next_ready(struct vtblk_softc *);
157 static void vtblk_request_requeue_ready(struct vtblk_softc *,
158 struct vtblk_request *);
159 static struct vtblk_request *
160 vtblk_request_next(struct vtblk_softc *);
161 static struct vtblk_request *
162 vtblk_request_bio(struct vtblk_softc *);
163 static int vtblk_request_execute(struct vtblk_softc *,
164 struct vtblk_request *);
165 static int vtblk_request_error(struct vtblk_request *);
167 static void vtblk_queue_completed(struct vtblk_softc *,
169 static void vtblk_done_completed(struct vtblk_softc *,
171 static void vtblk_drain_vq(struct vtblk_softc *);
172 static void vtblk_drain(struct vtblk_softc *);
174 static void vtblk_startio(struct vtblk_softc *);
175 static void vtblk_bio_done(struct vtblk_softc *, struct bio *, int);
177 static void vtblk_read_config(struct vtblk_softc *,
178 struct virtio_blk_config *);
179 static void vtblk_ident(struct vtblk_softc *);
180 static int vtblk_poll_request(struct vtblk_softc *,
181 struct vtblk_request *);
182 static int vtblk_quiesce(struct vtblk_softc *);
183 static void vtblk_vq_intr(void *);
184 static void vtblk_stop(struct vtblk_softc *);
186 static void vtblk_dump_quiesce(struct vtblk_softc *);
187 static int vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t);
188 static int vtblk_dump_flush(struct vtblk_softc *);
189 static void vtblk_dump_complete(struct vtblk_softc *);
191 static void vtblk_set_write_cache(struct vtblk_softc *, int);
192 static int vtblk_write_cache_enabled(struct vtblk_softc *sc,
193 struct virtio_blk_config *);
194 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS);
196 static void vtblk_setup_sysctl(struct vtblk_softc *);
197 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int);
199 #define vtblk_modern(_sc) (((_sc)->vtblk_features & VIRTIO_F_VERSION_1) != 0)
200 #define vtblk_htog16(_sc, _val) virtio_htog16(vtblk_modern(_sc), _val)
201 #define vtblk_htog32(_sc, _val) virtio_htog32(vtblk_modern(_sc), _val)
202 #define vtblk_htog64(_sc, _val) virtio_htog64(vtblk_modern(_sc), _val)
203 #define vtblk_gtoh16(_sc, _val) virtio_gtoh16(vtblk_modern(_sc), _val)
204 #define vtblk_gtoh32(_sc, _val) virtio_gtoh32(vtblk_modern(_sc), _val)
205 #define vtblk_gtoh64(_sc, _val) virtio_gtoh64(vtblk_modern(_sc), _val)
208 static int vtblk_no_ident = 0;
209 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident);
210 static int vtblk_writecache_mode = -1;
211 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode);
213 #define VTBLK_COMMON_FEATURES \
214 (VIRTIO_BLK_F_SIZE_MAX | \
215 VIRTIO_BLK_F_SEG_MAX | \
216 VIRTIO_BLK_F_GEOMETRY | \
218 VIRTIO_BLK_F_BLK_SIZE | \
219 VIRTIO_BLK_F_FLUSH | \
220 VIRTIO_BLK_F_TOPOLOGY | \
221 VIRTIO_BLK_F_CONFIG_WCE | \
222 VIRTIO_BLK_F_DISCARD | \
223 VIRTIO_RING_F_INDIRECT_DESC)
225 #define VTBLK_MODERN_FEATURES (VTBLK_COMMON_FEATURES)
226 #define VTBLK_LEGACY_FEATURES (VIRTIO_BLK_F_BARRIER | VTBLK_COMMON_FEATURES)
228 #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx
229 #define VTBLK_LOCK_INIT(_sc, _name) \
230 mtx_init(VTBLK_MTX((_sc)), (_name), \
231 "VirtIO Block Lock", MTX_DEF)
232 #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc)))
233 #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc)))
234 #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc)))
235 #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED)
236 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \
237 mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED)
239 #define VTBLK_DISK_NAME "vtbd"
240 #define VTBLK_QUIESCE_TIMEOUT (30 * hz)
241 #define VTBLK_BSIZE 512
244 * Each block request uses at least two segments - one for the header
245 * and one for the status.
247 #define VTBLK_MIN_SEGMENTS 2
249 static device_method_t vtblk_methods[] = {
250 /* Device methods. */
251 DEVMETHOD(device_probe, vtblk_probe),
252 DEVMETHOD(device_attach, vtblk_attach),
253 DEVMETHOD(device_detach, vtblk_detach),
254 DEVMETHOD(device_suspend, vtblk_suspend),
255 DEVMETHOD(device_resume, vtblk_resume),
256 DEVMETHOD(device_shutdown, vtblk_shutdown),
258 /* VirtIO methods. */
259 DEVMETHOD(virtio_config_change, vtblk_config_change),
264 static driver_t vtblk_driver = {
267 sizeof(struct vtblk_softc)
269 static devclass_t vtblk_devclass;
271 DRIVER_MODULE(virtio_blk, virtio_mmio, vtblk_driver, vtblk_devclass,
273 DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass,
275 MODULE_VERSION(virtio_blk, 1);
276 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1);
278 VIRTIO_SIMPLE_PNPTABLE(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter");
279 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, virtio_blk);
280 VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_blk);
283 vtblk_modevent(module_t mod, int type, void *unused)
304 vtblk_probe(device_t dev)
306 return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk));
310 vtblk_attach(device_t dev)
312 struct vtblk_softc *sc;
313 struct virtio_blk_config blkcfg;
316 virtio_set_feature_desc(dev, vtblk_feature_desc);
318 sc = device_get_softc(dev);
320 VTBLK_LOCK_INIT(sc, device_get_nameunit(dev));
321 bioq_init(&sc->vtblk_bioq);
322 TAILQ_INIT(&sc->vtblk_dump_queue);
323 TAILQ_INIT(&sc->vtblk_req_free);
324 TAILQ_INIT(&sc->vtblk_req_ready);
326 vtblk_setup_sysctl(sc);
327 vtblk_setup_features(sc);
329 vtblk_read_config(sc, &blkcfg);
332 * With the current sglist(9) implementation, it is not easy
333 * for us to support a maximum segment size as adjacent
334 * segments are coalesced. For now, just make sure it's larger
335 * than the maximum supported transfer size.
337 if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) {
338 if (blkcfg.size_max < maxphys) {
340 device_printf(dev, "host requires unsupported "
341 "maximum segment size feature\n");
346 sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg);
347 if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) {
349 device_printf(dev, "fewer than minimum number of segments "
350 "allowed: %d\n", sc->vtblk_max_nsegs);
354 sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT);
355 if (sc->vtblk_sglist == NULL) {
357 device_printf(dev, "cannot allocate sglist\n");
361 error = vtblk_alloc_virtqueue(sc);
363 device_printf(dev, "cannot allocate virtqueue\n");
367 error = vtblk_request_prealloc(sc);
369 device_printf(dev, "cannot preallocate requests\n");
373 vtblk_alloc_disk(sc, &blkcfg);
375 error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY);
377 device_printf(dev, "cannot setup virtqueue interrupt\n");
381 vtblk_create_disk(sc);
383 virtqueue_enable_intr(sc->vtblk_vq);
393 vtblk_detach(device_t dev)
395 struct vtblk_softc *sc;
397 sc = device_get_softc(dev);
400 sc->vtblk_flags |= VTBLK_FLAG_DETACH;
401 if (device_is_attached(dev))
407 if (sc->vtblk_disk != NULL) {
408 disk_destroy(sc->vtblk_disk);
409 sc->vtblk_disk = NULL;
412 if (sc->vtblk_sglist != NULL) {
413 sglist_free(sc->vtblk_sglist);
414 sc->vtblk_sglist = NULL;
417 VTBLK_LOCK_DESTROY(sc);
423 vtblk_suspend(device_t dev)
425 struct vtblk_softc *sc;
428 sc = device_get_softc(dev);
431 sc->vtblk_flags |= VTBLK_FLAG_SUSPEND;
432 /* XXX BMV: virtio_stop(), etc needed here? */
433 error = vtblk_quiesce(sc);
435 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
442 vtblk_resume(device_t dev)
444 struct vtblk_softc *sc;
446 sc = device_get_softc(dev);
449 /* XXX BMV: virtio_reinit(), etc needed here? */
450 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
458 vtblk_shutdown(device_t dev)
465 vtblk_config_change(device_t dev)
467 struct vtblk_softc *sc;
468 struct virtio_blk_config blkcfg;
471 sc = device_get_softc(dev);
473 vtblk_read_config(sc, &blkcfg);
475 /* Capacity is always in 512-byte units. */
476 capacity = blkcfg.capacity * VTBLK_BSIZE;
478 if (sc->vtblk_disk->d_mediasize != capacity)
479 vtblk_resize_disk(sc, capacity);
485 vtblk_open(struct disk *dp)
487 struct vtblk_softc *sc;
489 if ((sc = dp->d_drv1) == NULL)
492 return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0);
496 vtblk_close(struct disk *dp)
498 struct vtblk_softc *sc;
500 if ((sc = dp->d_drv1) == NULL)
507 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag,
510 struct vtblk_softc *sc;
512 if ((sc = dp->d_drv1) == NULL)
519 vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
523 struct vtblk_softc *sc;
529 if ((sc = dp->d_drv1) == NULL)
534 vtblk_dump_quiesce(sc);
537 error = vtblk_dump_write(sc, virtual, offset, length);
538 if (error || (virtual == NULL && offset == 0))
539 vtblk_dump_complete(sc);
547 vtblk_strategy(struct bio *bp)
549 struct vtblk_softc *sc;
551 if ((sc = bp->bio_disk->d_drv1) == NULL) {
552 vtblk_bio_done(NULL, bp, EINVAL);
557 * Fail any write if RO. Unfortunately, there does not seem to
558 * be a better way to report our readonly'ness to GEOM above.
560 if (sc->vtblk_flags & VTBLK_FLAG_READONLY &&
561 (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_FLUSH ||
562 bp->bio_cmd == BIO_DELETE)) {
563 vtblk_bio_done(sc, bp, EROFS);
567 if ((bp->bio_cmd != BIO_READ) && (bp->bio_cmd != BIO_WRITE) &&
568 (bp->bio_cmd != BIO_FLUSH) && (bp->bio_cmd != BIO_DELETE)) {
569 vtblk_bio_done(sc, bp, EOPNOTSUPP);
575 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) {
577 vtblk_bio_done(sc, bp, ENXIO);
581 bioq_insert_tail(&sc->vtblk_bioq, bp);
588 vtblk_negotiate_features(struct vtblk_softc *sc)
594 features = virtio_bus_is_modern(dev) ? VTBLK_MODERN_FEATURES :
595 VTBLK_LEGACY_FEATURES;
597 sc->vtblk_features = virtio_negotiate_features(dev, features);
598 virtio_finalize_features(dev);
602 vtblk_setup_features(struct vtblk_softc *sc)
608 vtblk_negotiate_features(sc);
610 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
611 sc->vtblk_flags |= VTBLK_FLAG_INDIRECT;
612 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO))
613 sc->vtblk_flags |= VTBLK_FLAG_READONLY;
614 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE))
615 sc->vtblk_flags |= VTBLK_FLAG_WCE_CONFIG;
618 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER))
619 sc->vtblk_flags |= VTBLK_FLAG_BARRIER;
623 vtblk_maximum_segments(struct vtblk_softc *sc,
624 struct virtio_blk_config *blkcfg)
630 nsegs = VTBLK_MIN_SEGMENTS;
632 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) {
633 nsegs += MIN(blkcfg->seg_max, maxphys / PAGE_SIZE + 1);
634 if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT)
635 nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT);
643 vtblk_alloc_virtqueue(struct vtblk_softc *sc)
646 struct vq_alloc_info vq_info;
650 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs,
651 vtblk_vq_intr, sc, &sc->vtblk_vq,
652 "%s request", device_get_nameunit(dev));
654 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info));
658 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity)
667 dp->d_mediasize = new_capacity;
669 device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n",
670 (uintmax_t) dp->d_mediasize >> 20,
671 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
675 error = disk_resize(dp, M_NOWAIT);
678 "disk_resize(9) failed, error: %d\n", error);
683 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
690 sc->vtblk_disk = dp = disk_alloc();
691 dp->d_open = vtblk_open;
692 dp->d_close = vtblk_close;
693 dp->d_ioctl = vtblk_ioctl;
694 dp->d_strategy = vtblk_strategy;
695 dp->d_name = VTBLK_DISK_NAME;
696 dp->d_unit = device_get_unit(dev);
698 dp->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION;
699 dp->d_hba_vendor = virtio_get_vendor(dev);
700 dp->d_hba_device = virtio_get_device(dev);
701 dp->d_hba_subvendor = virtio_get_subvendor(dev);
702 dp->d_hba_subdevice = virtio_get_subdevice(dev);
704 if (virtio_with_feature(dev, VIRTIO_BLK_F_FLUSH))
705 dp->d_flags |= DISKFLAG_CANFLUSHCACHE;
707 if ((sc->vtblk_flags & VTBLK_FLAG_READONLY) == 0)
708 dp->d_dump = vtblk_dump;
710 /* Capacity is always in 512-byte units. */
711 dp->d_mediasize = blkcfg->capacity * VTBLK_BSIZE;
713 if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE))
714 dp->d_sectorsize = blkcfg->blk_size;
716 dp->d_sectorsize = VTBLK_BSIZE;
719 * The VirtIO maximum I/O size is given in terms of segments.
720 * However, FreeBSD limits I/O size by logical buffer size, not
721 * by physically contiguous pages. Therefore, we have to assume
722 * no pages are contiguous. This may impose an artificially low
723 * maximum I/O size. But in practice, since QEMU advertises 128
724 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE,
725 * which is typically greater than maxphys. Eventually we should
726 * just advertise maxphys and split buffers that are too big.
728 * Note we must subtract one additional segment in case of non
729 * page aligned buffers.
731 dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) *
733 if (dp->d_maxsize < PAGE_SIZE)
734 dp->d_maxsize = PAGE_SIZE; /* XXX */
736 if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) {
737 dp->d_fwsectors = blkcfg->geometry.sectors;
738 dp->d_fwheads = blkcfg->geometry.heads;
741 if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) &&
742 blkcfg->topology.physical_block_exp > 0) {
743 dp->d_stripesize = dp->d_sectorsize *
744 (1 << blkcfg->topology.physical_block_exp);
745 dp->d_stripeoffset = (dp->d_stripesize -
746 blkcfg->topology.alignment_offset * dp->d_sectorsize) %
750 if (virtio_with_feature(dev, VIRTIO_BLK_F_DISCARD)) {
751 dp->d_flags |= DISKFLAG_CANDELETE;
752 dp->d_delmaxsize = blkcfg->max_discard_sectors * VTBLK_BSIZE;
755 if (vtblk_write_cache_enabled(sc, blkcfg) != 0)
756 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK;
758 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH;
762 vtblk_create_disk(struct vtblk_softc *sc)
770 device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n",
771 (uintmax_t) dp->d_mediasize >> 20,
772 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
775 disk_create(dp, DISK_VERSION);
779 vtblk_request_prealloc(struct vtblk_softc *sc)
781 struct vtblk_request *req;
784 nreqs = virtqueue_size(sc->vtblk_vq);
787 * Preallocate sufficient requests to keep the virtqueue full. Each
788 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce
789 * the number allocated when indirect descriptors are not available.
791 if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0)
792 nreqs /= VTBLK_MIN_SEGMENTS;
794 for (i = 0; i < nreqs; i++) {
795 req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT);
799 MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1);
800 MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1);
802 sc->vtblk_request_count++;
803 vtblk_request_enqueue(sc, req);
810 vtblk_request_free(struct vtblk_softc *sc)
812 struct vtblk_request *req;
814 MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready));
816 while ((req = vtblk_request_dequeue(sc)) != NULL) {
817 sc->vtblk_request_count--;
821 KASSERT(sc->vtblk_request_count == 0,
822 ("%s: leaked %d requests", __func__, sc->vtblk_request_count));
825 static struct vtblk_request *
826 vtblk_request_dequeue(struct vtblk_softc *sc)
828 struct vtblk_request *req;
830 req = TAILQ_FIRST(&sc->vtblk_req_free);
832 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link);
833 bzero(req, sizeof(struct vtblk_request));
840 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req)
843 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link);
846 static struct vtblk_request *
847 vtblk_request_next_ready(struct vtblk_softc *sc)
849 struct vtblk_request *req;
851 req = TAILQ_FIRST(&sc->vtblk_req_ready);
853 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link);
859 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req)
862 /* NOTE: Currently, there will be at most one request in the queue. */
863 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link);
866 static struct vtblk_request *
867 vtblk_request_next(struct vtblk_softc *sc)
869 struct vtblk_request *req;
871 req = vtblk_request_next_ready(sc);
875 return (vtblk_request_bio(sc));
878 static struct vtblk_request *
879 vtblk_request_bio(struct vtblk_softc *sc)
881 struct bio_queue_head *bioq;
882 struct vtblk_request *req;
885 bioq = &sc->vtblk_bioq;
887 if (bioq_first(bioq) == NULL)
890 req = vtblk_request_dequeue(sc);
894 bp = bioq_takefirst(bioq);
897 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
899 switch (bp->bio_cmd) {
901 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH);
902 req->vbr_hdr.sector = 0;
905 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_IN);
906 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
909 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT);
910 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
913 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_DISCARD);
914 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
917 panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd);
920 if (bp->bio_flags & BIO_ORDERED)
921 req->vbr_hdr.type |= vtblk_gtoh32(sc, VIRTIO_BLK_T_BARRIER);
927 vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req)
929 struct virtqueue *vq;
932 int ordered, readable, writable, error;
935 sg = sc->vtblk_sglist;
941 * Some hosts (such as bhyve) do not implement the barrier feature,
942 * so we emulate it in the driver by allowing the barrier request
943 * to be the only one in flight.
945 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) {
946 if (sc->vtblk_req_ordered != NULL)
948 if (bp->bio_flags & BIO_ORDERED) {
949 if (!virtqueue_empty(vq))
952 req->vbr_hdr.type &= vtblk_gtoh32(sc,
953 ~VIRTIO_BLK_T_BARRIER);
958 sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr));
960 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
961 error = sglist_append_bio(sg, bp);
962 if (error || sg->sg_nseg == sg->sg_maxseg) {
963 panic("%s: bio %p data buffer too big %d",
964 __func__, bp, error);
967 /* BIO_READ means the host writes into our buffer. */
968 if (bp->bio_cmd == BIO_READ)
969 writable = sg->sg_nseg - 1;
970 } else if (bp->bio_cmd == BIO_DELETE) {
971 struct virtio_blk_discard_write_zeroes *discard;
973 discard = malloc(sizeof(*discard), M_DEVBUF, M_NOWAIT | M_ZERO);
977 bp->bio_driver1 = discard;
978 discard->sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
979 discard->num_sectors = vtblk_gtoh32(sc, bp->bio_bcount / VTBLK_BSIZE);
980 error = sglist_append(sg, discard, sizeof(*discard));
981 if (error || sg->sg_nseg == sg->sg_maxseg) {
982 panic("%s: bio %p data buffer too big %d",
983 __func__, bp, error);
988 sglist_append(sg, &req->vbr_ack, sizeof(uint8_t));
989 readable = sg->sg_nseg - writable;
991 error = virtqueue_enqueue(vq, req, sg, readable, writable);
992 if (error == 0 && ordered)
993 sc->vtblk_req_ordered = req;
999 vtblk_request_error(struct vtblk_request *req)
1003 switch (req->vbr_ack) {
1004 case VIRTIO_BLK_S_OK:
1007 case VIRTIO_BLK_S_UNSUPP:
1019 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue)
1021 struct vtblk_request *req;
1024 while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) {
1025 if (sc->vtblk_req_ordered != NULL) {
1026 MPASS(sc->vtblk_req_ordered == req);
1027 sc->vtblk_req_ordered = NULL;
1031 bp->bio_error = vtblk_request_error(req);
1032 TAILQ_INSERT_TAIL(queue, bp, bio_queue);
1034 vtblk_request_enqueue(sc, req);
1039 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue)
1041 struct bio *bp, *tmp;
1043 TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) {
1044 if (bp->bio_error != 0)
1045 disk_err(bp, "hard error", -1, 1);
1046 vtblk_bio_done(sc, bp, bp->bio_error);
1051 vtblk_drain_vq(struct vtblk_softc *sc)
1053 struct virtqueue *vq;
1054 struct vtblk_request *req;
1060 while ((req = virtqueue_drain(vq, &last)) != NULL) {
1061 vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1062 vtblk_request_enqueue(sc, req);
1065 sc->vtblk_req_ordered = NULL;
1066 KASSERT(virtqueue_empty(vq), ("virtqueue not empty"));
1070 vtblk_drain(struct vtblk_softc *sc)
1072 struct bio_queue_head *bioq;
1073 struct vtblk_request *req;
1076 bioq = &sc->vtblk_bioq;
1078 if (sc->vtblk_vq != NULL) {
1079 struct bio_queue queue;
1082 vtblk_queue_completed(sc, &queue);
1083 vtblk_done_completed(sc, &queue);
1088 while ((req = vtblk_request_next_ready(sc)) != NULL) {
1089 vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1090 vtblk_request_enqueue(sc, req);
1093 while (bioq_first(bioq) != NULL) {
1094 bp = bioq_takefirst(bioq);
1095 vtblk_bio_done(sc, bp, ENXIO);
1098 vtblk_request_free(sc);
1102 vtblk_startio(struct vtblk_softc *sc)
1104 struct virtqueue *vq;
1105 struct vtblk_request *req;
1108 VTBLK_LOCK_ASSERT(sc);
1112 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1115 while (!virtqueue_full(vq)) {
1116 req = vtblk_request_next(sc);
1120 if (vtblk_request_execute(sc, req) != 0) {
1121 vtblk_request_requeue_ready(sc, req);
1129 virtqueue_notify(vq);
1133 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error)
1136 /* Because of GEOM direct dispatch, we cannot hold any locks. */
1138 VTBLK_LOCK_ASSERT_NOTOWNED(sc);
1141 bp->bio_resid = bp->bio_bcount;
1142 bp->bio_error = error;
1143 bp->bio_flags |= BIO_ERROR;
1146 if (bp->bio_driver1 != NULL) {
1147 free(bp->bio_driver1, M_DEVBUF);
1148 bp->bio_driver1 = NULL;
1154 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \
1155 if (virtio_with_feature(_dev, _feature)) { \
1156 virtio_read_device_config(_dev, \
1157 offsetof(struct virtio_blk_config, _field), \
1158 &(_cfg)->_field, sizeof((_cfg)->_field)); \
1162 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
1166 dev = sc->vtblk_dev;
1168 bzero(blkcfg, sizeof(struct virtio_blk_config));
1170 /* The capacity is always available. */
1171 virtio_read_device_config(dev, offsetof(struct virtio_blk_config,
1172 capacity), &blkcfg->capacity, sizeof(blkcfg->capacity));
1174 /* Read the configuration if the feature was negotiated. */
1175 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg);
1176 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg);
1177 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY,
1178 geometry.cylinders, blkcfg);
1179 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY,
1180 geometry.heads, blkcfg);
1181 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY,
1182 geometry.sectors, blkcfg);
1183 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg);
1184 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1185 topology.physical_block_exp, blkcfg);
1186 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1187 topology.alignment_offset, blkcfg);
1188 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1189 topology.min_io_size, blkcfg);
1190 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1191 topology.opt_io_size, blkcfg);
1192 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, wce, blkcfg);
1193 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_sectors,
1195 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_seg, blkcfg);
1196 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, discard_sector_alignment,
1200 #undef VTBLK_GET_CONFIG
1203 vtblk_ident(struct vtblk_softc *sc)
1207 struct vtblk_request *req;
1210 dp = sc->vtblk_disk;
1211 len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE);
1213 if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0)
1216 req = vtblk_request_dequeue(sc);
1221 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_GET_ID);
1222 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
1223 req->vbr_hdr.sector = 0;
1228 buf.bio_cmd = BIO_READ;
1229 buf.bio_data = dp->d_ident;
1230 buf.bio_bcount = len;
1233 error = vtblk_poll_request(sc, req);
1236 vtblk_request_enqueue(sc, req);
1239 device_printf(sc->vtblk_dev,
1240 "error getting device identifier: %d\n", error);
1245 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req)
1247 struct virtqueue *vq;
1252 if (!virtqueue_empty(vq))
1255 error = vtblk_request_execute(sc, req);
1259 virtqueue_notify(vq);
1260 virtqueue_poll(vq, NULL);
1262 error = vtblk_request_error(req);
1263 if (error && bootverbose) {
1264 device_printf(sc->vtblk_dev,
1265 "%s: IO error: %d\n", __func__, error);
1272 vtblk_quiesce(struct vtblk_softc *sc)
1276 VTBLK_LOCK_ASSERT(sc);
1279 while (!virtqueue_empty(sc->vtblk_vq)) {
1280 if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq",
1281 VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) {
1291 vtblk_vq_intr(void *xsc)
1293 struct vtblk_softc *sc;
1294 struct virtqueue *vq;
1295 struct bio_queue queue;
1304 if (sc->vtblk_flags & VTBLK_FLAG_DETACH)
1307 vtblk_queue_completed(sc, &queue);
1310 if (virtqueue_enable_intr(vq) != 0) {
1311 virtqueue_disable_intr(vq);
1315 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1316 wakeup(&sc->vtblk_vq);
1320 vtblk_done_completed(sc, &queue);
1324 vtblk_stop(struct vtblk_softc *sc)
1327 virtqueue_disable_intr(sc->vtblk_vq);
1328 virtio_stop(sc->vtblk_dev);
1332 vtblk_dump_quiesce(struct vtblk_softc *sc)
1336 * Spin here until all the requests in-flight at the time of the
1337 * dump are completed and queued. The queued requests will be
1338 * biodone'd once the dump is finished.
1340 while (!virtqueue_empty(sc->vtblk_vq))
1341 vtblk_queue_completed(sc, &sc->vtblk_dump_queue);
1345 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset,
1349 struct vtblk_request *req;
1351 req = &sc->vtblk_dump_request;
1353 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT);
1354 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
1355 req->vbr_hdr.sector = vtblk_gtoh64(sc, offset / VTBLK_BSIZE);
1360 buf.bio_cmd = BIO_WRITE;
1361 buf.bio_data = virtual;
1362 buf.bio_bcount = length;
1364 return (vtblk_poll_request(sc, req));
1368 vtblk_dump_flush(struct vtblk_softc *sc)
1371 struct vtblk_request *req;
1373 req = &sc->vtblk_dump_request;
1375 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH);
1376 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
1377 req->vbr_hdr.sector = 0;
1382 buf.bio_cmd = BIO_FLUSH;
1384 return (vtblk_poll_request(sc, req));
1388 vtblk_dump_complete(struct vtblk_softc *sc)
1391 vtblk_dump_flush(sc);
1394 vtblk_done_completed(sc, &sc->vtblk_dump_queue);
1399 vtblk_set_write_cache(struct vtblk_softc *sc, int wc)
1402 /* Set either writeback (1) or writethrough (0) mode. */
1403 virtio_write_dev_config_1(sc->vtblk_dev,
1404 offsetof(struct virtio_blk_config, wce), wc);
1408 vtblk_write_cache_enabled(struct vtblk_softc *sc,
1409 struct virtio_blk_config *blkcfg)
1413 if (sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) {
1414 wc = vtblk_tunable_int(sc, "writecache_mode",
1415 vtblk_writecache_mode);
1416 if (wc >= 0 && wc < VTBLK_CACHE_MAX)
1417 vtblk_set_write_cache(sc, wc);
1421 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_FLUSH);
1427 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS)
1429 struct vtblk_softc *sc;
1432 sc = oidp->oid_arg1;
1433 wc = sc->vtblk_write_cache;
1435 error = sysctl_handle_int(oidp, &wc, 0, req);
1436 if (error || req->newptr == NULL)
1438 if ((sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) == 0)
1440 if (wc < 0 || wc >= VTBLK_CACHE_MAX)
1444 sc->vtblk_write_cache = wc;
1445 vtblk_set_write_cache(sc, sc->vtblk_write_cache);
1452 vtblk_setup_sysctl(struct vtblk_softc *sc)
1455 struct sysctl_ctx_list *ctx;
1456 struct sysctl_oid *tree;
1457 struct sysctl_oid_list *child;
1459 dev = sc->vtblk_dev;
1460 ctx = device_get_sysctl_ctx(dev);
1461 tree = device_get_sysctl_tree(dev);
1462 child = SYSCTL_CHILDREN(tree);
1464 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode",
1465 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
1466 vtblk_write_cache_sysctl, "I",
1467 "Write cache mode (writethrough (0) or writeback (1))");
1471 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def)
1475 snprintf(path, sizeof(path),
1476 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob);
1477 TUNABLE_INT_FETCH(path, &def);