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_WC_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_WCE, "WriteCache" },
113 { VIRTIO_BLK_F_TOPOLOGY, "Topology" },
114 { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" },
119 static int vtblk_modevent(module_t, int, void *);
121 static int vtblk_probe(device_t);
122 static int vtblk_attach(device_t);
123 static int vtblk_detach(device_t);
124 static int vtblk_suspend(device_t);
125 static int vtblk_resume(device_t);
126 static int vtblk_shutdown(device_t);
127 static int vtblk_config_change(device_t);
129 static int vtblk_open(struct disk *);
130 static int vtblk_close(struct disk *);
131 static int vtblk_ioctl(struct disk *, u_long, void *, int,
133 static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t);
134 static void vtblk_strategy(struct bio *);
136 static void vtblk_negotiate_features(struct vtblk_softc *);
137 static void vtblk_setup_features(struct vtblk_softc *);
138 static int vtblk_maximum_segments(struct vtblk_softc *,
139 struct virtio_blk_config *);
140 static int vtblk_alloc_virtqueue(struct vtblk_softc *);
141 static void vtblk_resize_disk(struct vtblk_softc *, uint64_t);
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_request_prealloc(struct vtblk_softc *);
147 static void vtblk_request_free(struct vtblk_softc *);
148 static struct vtblk_request *
149 vtblk_request_dequeue(struct vtblk_softc *);
150 static void vtblk_request_enqueue(struct vtblk_softc *,
151 struct vtblk_request *);
152 static struct vtblk_request *
153 vtblk_request_next_ready(struct vtblk_softc *);
154 static void vtblk_request_requeue_ready(struct vtblk_softc *,
155 struct vtblk_request *);
156 static struct vtblk_request *
157 vtblk_request_next(struct vtblk_softc *);
158 static struct vtblk_request *
159 vtblk_request_bio(struct vtblk_softc *);
160 static int vtblk_request_execute(struct vtblk_softc *,
161 struct vtblk_request *);
162 static int vtblk_request_error(struct vtblk_request *);
164 static void vtblk_queue_completed(struct vtblk_softc *,
166 static void vtblk_done_completed(struct vtblk_softc *,
168 static void vtblk_drain_vq(struct vtblk_softc *);
169 static void vtblk_drain(struct vtblk_softc *);
171 static void vtblk_startio(struct vtblk_softc *);
172 static void vtblk_bio_done(struct vtblk_softc *, struct bio *, int);
174 static void vtblk_read_config(struct vtblk_softc *,
175 struct virtio_blk_config *);
176 static void vtblk_ident(struct vtblk_softc *);
177 static int vtblk_poll_request(struct vtblk_softc *,
178 struct vtblk_request *);
179 static int vtblk_quiesce(struct vtblk_softc *);
180 static void vtblk_vq_intr(void *);
181 static void vtblk_stop(struct vtblk_softc *);
183 static void vtblk_dump_quiesce(struct vtblk_softc *);
184 static int vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t);
185 static int vtblk_dump_flush(struct vtblk_softc *);
186 static void vtblk_dump_complete(struct vtblk_softc *);
188 static void vtblk_set_write_cache(struct vtblk_softc *, int);
189 static int vtblk_write_cache_enabled(struct vtblk_softc *sc,
190 struct virtio_blk_config *);
191 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS);
193 static void vtblk_setup_sysctl(struct vtblk_softc *);
194 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int);
197 static int vtblk_no_ident = 0;
198 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident);
199 static int vtblk_writecache_mode = -1;
200 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode);
202 /* Features desired/implemented by this driver. */
203 #define VTBLK_FEATURES \
204 (VIRTIO_BLK_F_BARRIER | \
205 VIRTIO_BLK_F_SIZE_MAX | \
206 VIRTIO_BLK_F_SEG_MAX | \
207 VIRTIO_BLK_F_GEOMETRY | \
209 VIRTIO_BLK_F_BLK_SIZE | \
211 VIRTIO_BLK_F_TOPOLOGY | \
212 VIRTIO_BLK_F_CONFIG_WCE | \
213 VIRTIO_RING_F_INDIRECT_DESC)
215 #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx
216 #define VTBLK_LOCK_INIT(_sc, _name) \
217 mtx_init(VTBLK_MTX((_sc)), (_name), \
218 "VirtIO Block Lock", MTX_DEF)
219 #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc)))
220 #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc)))
221 #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc)))
222 #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED)
223 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \
224 mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED)
226 #define VTBLK_DISK_NAME "vtbd"
227 #define VTBLK_QUIESCE_TIMEOUT (30 * hz)
230 * Each block request uses at least two segments - one for the header
231 * and one for the status.
233 #define VTBLK_MIN_SEGMENTS 2
235 static device_method_t vtblk_methods[] = {
236 /* Device methods. */
237 DEVMETHOD(device_probe, vtblk_probe),
238 DEVMETHOD(device_attach, vtblk_attach),
239 DEVMETHOD(device_detach, vtblk_detach),
240 DEVMETHOD(device_suspend, vtblk_suspend),
241 DEVMETHOD(device_resume, vtblk_resume),
242 DEVMETHOD(device_shutdown, vtblk_shutdown),
244 /* VirtIO methods. */
245 DEVMETHOD(virtio_config_change, vtblk_config_change),
250 static driver_t vtblk_driver = {
253 sizeof(struct vtblk_softc)
255 static devclass_t vtblk_devclass;
257 DRIVER_MODULE(virtio_blk, virtio_mmio, vtblk_driver, vtblk_devclass,
259 DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass,
261 MODULE_VERSION(virtio_blk, 1);
262 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1);
265 vtblk_modevent(module_t mod, int type, void *unused)
286 vtblk_probe(device_t dev)
289 if (virtio_get_device_type(dev) != VIRTIO_ID_BLOCK)
292 device_set_desc(dev, "VirtIO Block Adapter");
294 return (BUS_PROBE_DEFAULT);
298 vtblk_attach(device_t dev)
300 struct vtblk_softc *sc;
301 struct virtio_blk_config blkcfg;
304 virtio_set_feature_desc(dev, vtblk_feature_desc);
306 sc = device_get_softc(dev);
308 VTBLK_LOCK_INIT(sc, device_get_nameunit(dev));
309 bioq_init(&sc->vtblk_bioq);
310 TAILQ_INIT(&sc->vtblk_dump_queue);
311 TAILQ_INIT(&sc->vtblk_req_free);
312 TAILQ_INIT(&sc->vtblk_req_ready);
314 vtblk_setup_sysctl(sc);
315 vtblk_setup_features(sc);
317 vtblk_read_config(sc, &blkcfg);
320 * With the current sglist(9) implementation, it is not easy
321 * for us to support a maximum segment size as adjacent
322 * segments are coalesced. For now, just make sure it's larger
323 * than the maximum supported transfer size.
325 if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) {
326 if (blkcfg.size_max < MAXPHYS) {
328 device_printf(dev, "host requires unsupported "
329 "maximum segment size feature\n");
334 sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg);
335 if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) {
337 device_printf(dev, "fewer than minimum number of segments "
338 "allowed: %d\n", sc->vtblk_max_nsegs);
342 sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT);
343 if (sc->vtblk_sglist == NULL) {
345 device_printf(dev, "cannot allocate sglist\n");
349 error = vtblk_alloc_virtqueue(sc);
351 device_printf(dev, "cannot allocate virtqueue\n");
355 error = vtblk_request_prealloc(sc);
357 device_printf(dev, "cannot preallocate requests\n");
361 vtblk_alloc_disk(sc, &blkcfg);
363 error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY);
365 device_printf(dev, "cannot setup virtqueue interrupt\n");
369 vtblk_create_disk(sc);
371 virtqueue_enable_intr(sc->vtblk_vq);
381 vtblk_detach(device_t dev)
383 struct vtblk_softc *sc;
385 sc = device_get_softc(dev);
388 sc->vtblk_flags |= VTBLK_FLAG_DETACH;
389 if (device_is_attached(dev))
395 if (sc->vtblk_disk != NULL) {
396 disk_destroy(sc->vtblk_disk);
397 sc->vtblk_disk = NULL;
400 if (sc->vtblk_sglist != NULL) {
401 sglist_free(sc->vtblk_sglist);
402 sc->vtblk_sglist = NULL;
405 VTBLK_LOCK_DESTROY(sc);
411 vtblk_suspend(device_t dev)
413 struct vtblk_softc *sc;
416 sc = device_get_softc(dev);
419 sc->vtblk_flags |= VTBLK_FLAG_SUSPEND;
420 /* XXX BMV: virtio_stop(), etc needed here? */
421 error = vtblk_quiesce(sc);
423 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
430 vtblk_resume(device_t dev)
432 struct vtblk_softc *sc;
434 sc = device_get_softc(dev);
437 /* XXX BMV: virtio_reinit(), etc needed here? */
438 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
446 vtblk_shutdown(device_t dev)
453 vtblk_config_change(device_t dev)
455 struct vtblk_softc *sc;
456 struct virtio_blk_config blkcfg;
459 sc = device_get_softc(dev);
461 vtblk_read_config(sc, &blkcfg);
463 /* Capacity is always in 512-byte units. */
464 capacity = blkcfg.capacity * 512;
466 if (sc->vtblk_disk->d_mediasize != capacity)
467 vtblk_resize_disk(sc, capacity);
473 vtblk_open(struct disk *dp)
475 struct vtblk_softc *sc;
477 if ((sc = dp->d_drv1) == NULL)
480 return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0);
484 vtblk_close(struct disk *dp)
486 struct vtblk_softc *sc;
488 if ((sc = dp->d_drv1) == NULL)
495 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag,
498 struct vtblk_softc *sc;
500 if ((sc = dp->d_drv1) == NULL)
507 vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
511 struct vtblk_softc *sc;
517 if ((sc = dp->d_drv1) == NULL)
522 vtblk_dump_quiesce(sc);
525 error = vtblk_dump_write(sc, virtual, offset, length);
526 if (error || (virtual == NULL && offset == 0))
527 vtblk_dump_complete(sc);
535 vtblk_strategy(struct bio *bp)
537 struct vtblk_softc *sc;
539 if ((sc = bp->bio_disk->d_drv1) == NULL) {
540 vtblk_bio_done(NULL, bp, EINVAL);
545 * Fail any write if RO. Unfortunately, there does not seem to
546 * be a better way to report our readonly'ness to GEOM above.
548 if (sc->vtblk_flags & VTBLK_FLAG_READONLY &&
549 (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_FLUSH)) {
550 vtblk_bio_done(sc, bp, EROFS);
556 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) {
558 vtblk_bio_done(sc, bp, ENXIO);
562 bioq_insert_tail(&sc->vtblk_bioq, bp);
569 vtblk_negotiate_features(struct vtblk_softc *sc)
575 features = VTBLK_FEATURES;
577 sc->vtblk_features = virtio_negotiate_features(dev, features);
581 vtblk_setup_features(struct vtblk_softc *sc)
587 vtblk_negotiate_features(sc);
589 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
590 sc->vtblk_flags |= VTBLK_FLAG_INDIRECT;
591 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO))
592 sc->vtblk_flags |= VTBLK_FLAG_READONLY;
593 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER))
594 sc->vtblk_flags |= VTBLK_FLAG_BARRIER;
595 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE))
596 sc->vtblk_flags |= VTBLK_FLAG_WC_CONFIG;
600 vtblk_maximum_segments(struct vtblk_softc *sc,
601 struct virtio_blk_config *blkcfg)
607 nsegs = VTBLK_MIN_SEGMENTS;
609 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) {
610 nsegs += MIN(blkcfg->seg_max, MAXPHYS / PAGE_SIZE + 1);
611 if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT)
612 nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT);
620 vtblk_alloc_virtqueue(struct vtblk_softc *sc)
623 struct vq_alloc_info vq_info;
627 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs,
628 vtblk_vq_intr, sc, &sc->vtblk_vq,
629 "%s request", device_get_nameunit(dev));
631 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info));
635 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity)
644 dp->d_mediasize = new_capacity;
646 device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n",
647 (uintmax_t) dp->d_mediasize >> 20,
648 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
652 error = disk_resize(dp, M_NOWAIT);
655 "disk_resize(9) failed, error: %d\n", error);
660 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
667 sc->vtblk_disk = dp = disk_alloc();
668 dp->d_open = vtblk_open;
669 dp->d_close = vtblk_close;
670 dp->d_ioctl = vtblk_ioctl;
671 dp->d_strategy = vtblk_strategy;
672 dp->d_name = VTBLK_DISK_NAME;
673 dp->d_unit = device_get_unit(dev);
675 dp->d_flags = DISKFLAG_CANFLUSHCACHE | DISKFLAG_UNMAPPED_BIO |
676 DISKFLAG_DIRECT_COMPLETION;
677 dp->d_hba_vendor = virtio_get_vendor(dev);
678 dp->d_hba_device = virtio_get_device(dev);
679 dp->d_hba_subvendor = virtio_get_subvendor(dev);
680 dp->d_hba_subdevice = virtio_get_subdevice(dev);
682 if ((sc->vtblk_flags & VTBLK_FLAG_READONLY) == 0)
683 dp->d_dump = vtblk_dump;
685 /* Capacity is always in 512-byte units. */
686 dp->d_mediasize = blkcfg->capacity * 512;
688 if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE))
689 dp->d_sectorsize = blkcfg->blk_size;
691 dp->d_sectorsize = 512;
694 * The VirtIO maximum I/O size is given in terms of segments.
695 * However, FreeBSD limits I/O size by logical buffer size, not
696 * by physically contiguous pages. Therefore, we have to assume
697 * no pages are contiguous. This may impose an artificially low
698 * maximum I/O size. But in practice, since QEMU advertises 128
699 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE,
700 * which is typically greater than MAXPHYS. Eventually we should
701 * just advertise MAXPHYS and split buffers that are too big.
703 * Note we must subtract one additional segment in case of non
704 * page aligned buffers.
706 dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) *
708 if (dp->d_maxsize < PAGE_SIZE)
709 dp->d_maxsize = PAGE_SIZE; /* XXX */
711 if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) {
712 dp->d_fwsectors = blkcfg->geometry.sectors;
713 dp->d_fwheads = blkcfg->geometry.heads;
716 if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) &&
717 blkcfg->topology.physical_block_exp > 0) {
718 dp->d_stripesize = dp->d_sectorsize *
719 (1 << blkcfg->topology.physical_block_exp);
720 dp->d_stripeoffset = (dp->d_stripesize -
721 blkcfg->topology.alignment_offset * dp->d_sectorsize) %
725 if (vtblk_write_cache_enabled(sc, blkcfg) != 0)
726 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK;
728 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH;
732 vtblk_create_disk(struct vtblk_softc *sc)
740 device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n",
741 (uintmax_t) dp->d_mediasize >> 20,
742 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
745 disk_create(dp, DISK_VERSION);
749 vtblk_request_prealloc(struct vtblk_softc *sc)
751 struct vtblk_request *req;
754 nreqs = virtqueue_size(sc->vtblk_vq);
757 * Preallocate sufficient requests to keep the virtqueue full. Each
758 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce
759 * the number allocated when indirect descriptors are not available.
761 if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0)
762 nreqs /= VTBLK_MIN_SEGMENTS;
764 for (i = 0; i < nreqs; i++) {
765 req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT);
769 MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1);
770 MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1);
772 sc->vtblk_request_count++;
773 vtblk_request_enqueue(sc, req);
780 vtblk_request_free(struct vtblk_softc *sc)
782 struct vtblk_request *req;
784 MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready));
786 while ((req = vtblk_request_dequeue(sc)) != NULL) {
787 sc->vtblk_request_count--;
791 KASSERT(sc->vtblk_request_count == 0,
792 ("%s: leaked %d requests", __func__, sc->vtblk_request_count));
795 static struct vtblk_request *
796 vtblk_request_dequeue(struct vtblk_softc *sc)
798 struct vtblk_request *req;
800 req = TAILQ_FIRST(&sc->vtblk_req_free);
802 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link);
803 bzero(req, sizeof(struct vtblk_request));
810 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req)
813 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link);
816 static struct vtblk_request *
817 vtblk_request_next_ready(struct vtblk_softc *sc)
819 struct vtblk_request *req;
821 req = TAILQ_FIRST(&sc->vtblk_req_ready);
823 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link);
829 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req)
832 /* NOTE: Currently, there will be at most one request in the queue. */
833 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link);
836 static struct vtblk_request *
837 vtblk_request_next(struct vtblk_softc *sc)
839 struct vtblk_request *req;
841 req = vtblk_request_next_ready(sc);
845 return (vtblk_request_bio(sc));
848 static struct vtblk_request *
849 vtblk_request_bio(struct vtblk_softc *sc)
851 struct bio_queue_head *bioq;
852 struct vtblk_request *req;
855 bioq = &sc->vtblk_bioq;
857 if (bioq_first(bioq) == NULL)
860 req = vtblk_request_dequeue(sc);
864 bp = bioq_takefirst(bioq);
867 req->vbr_hdr.ioprio = 1;
869 switch (bp->bio_cmd) {
871 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
874 req->vbr_hdr.type = VIRTIO_BLK_T_IN;
875 req->vbr_hdr.sector = bp->bio_offset / 512;
878 req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
879 req->vbr_hdr.sector = bp->bio_offset / 512;
882 panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd);
885 if (bp->bio_flags & BIO_ORDERED)
886 req->vbr_hdr.type |= VIRTIO_BLK_T_BARRIER;
892 vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req)
894 struct virtqueue *vq;
897 int ordered, readable, writable, error;
900 sg = sc->vtblk_sglist;
906 * Some hosts (such as bhyve) do not implement the barrier feature,
907 * so we emulate it in the driver by allowing the barrier request
908 * to be the only one in flight.
910 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) {
911 if (sc->vtblk_req_ordered != NULL)
913 if (bp->bio_flags & BIO_ORDERED) {
914 if (!virtqueue_empty(vq))
917 req->vbr_hdr.type &= ~VIRTIO_BLK_T_BARRIER;
922 sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr));
924 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
925 error = sglist_append_bio(sg, bp);
926 if (error || sg->sg_nseg == sg->sg_maxseg) {
927 panic("%s: bio %p data buffer too big %d",
928 __func__, bp, error);
931 /* BIO_READ means the host writes into our buffer. */
932 if (bp->bio_cmd == BIO_READ)
933 writable = sg->sg_nseg - 1;
937 sglist_append(sg, &req->vbr_ack, sizeof(uint8_t));
938 readable = sg->sg_nseg - writable;
940 error = virtqueue_enqueue(vq, req, sg, readable, writable);
941 if (error == 0 && ordered)
942 sc->vtblk_req_ordered = req;
948 vtblk_request_error(struct vtblk_request *req)
952 switch (req->vbr_ack) {
953 case VIRTIO_BLK_S_OK:
956 case VIRTIO_BLK_S_UNSUPP:
968 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue)
970 struct vtblk_request *req;
973 while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) {
974 if (sc->vtblk_req_ordered != NULL) {
975 MPASS(sc->vtblk_req_ordered == req);
976 sc->vtblk_req_ordered = NULL;
980 bp->bio_error = vtblk_request_error(req);
981 TAILQ_INSERT_TAIL(queue, bp, bio_queue);
983 vtblk_request_enqueue(sc, req);
988 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue)
990 struct bio *bp, *tmp;
992 TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) {
993 if (bp->bio_error != 0)
994 disk_err(bp, "hard error", -1, 1);
995 vtblk_bio_done(sc, bp, bp->bio_error);
1000 vtblk_drain_vq(struct vtblk_softc *sc)
1002 struct virtqueue *vq;
1003 struct vtblk_request *req;
1009 while ((req = virtqueue_drain(vq, &last)) != NULL) {
1010 vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1011 vtblk_request_enqueue(sc, req);
1014 sc->vtblk_req_ordered = NULL;
1015 KASSERT(virtqueue_empty(vq), ("virtqueue not empty"));
1019 vtblk_drain(struct vtblk_softc *sc)
1021 struct bio_queue queue;
1022 struct bio_queue_head *bioq;
1023 struct vtblk_request *req;
1026 bioq = &sc->vtblk_bioq;
1029 if (sc->vtblk_vq != NULL) {
1030 vtblk_queue_completed(sc, &queue);
1031 vtblk_done_completed(sc, &queue);
1036 while ((req = vtblk_request_next_ready(sc)) != NULL) {
1037 vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1038 vtblk_request_enqueue(sc, req);
1041 while (bioq_first(bioq) != NULL) {
1042 bp = bioq_takefirst(bioq);
1043 vtblk_bio_done(sc, bp, ENXIO);
1046 vtblk_request_free(sc);
1050 vtblk_startio(struct vtblk_softc *sc)
1052 struct virtqueue *vq;
1053 struct vtblk_request *req;
1056 VTBLK_LOCK_ASSERT(sc);
1060 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1063 while (!virtqueue_full(vq)) {
1064 req = vtblk_request_next(sc);
1068 if (vtblk_request_execute(sc, req) != 0) {
1069 vtblk_request_requeue_ready(sc, req);
1077 virtqueue_notify(vq);
1081 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error)
1084 /* Because of GEOM direct dispatch, we cannot hold any locks. */
1086 VTBLK_LOCK_ASSERT_NOTOWNED(sc);
1089 bp->bio_resid = bp->bio_bcount;
1090 bp->bio_error = error;
1091 bp->bio_flags |= BIO_ERROR;
1097 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \
1098 if (virtio_with_feature(_dev, _feature)) { \
1099 virtio_read_device_config(_dev, \
1100 offsetof(struct virtio_blk_config, _field), \
1101 &(_cfg)->_field, sizeof((_cfg)->_field)); \
1105 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
1109 dev = sc->vtblk_dev;
1111 bzero(blkcfg, sizeof(struct virtio_blk_config));
1113 /* The capacity is always available. */
1114 virtio_read_device_config(dev, offsetof(struct virtio_blk_config,
1115 capacity), &blkcfg->capacity, sizeof(blkcfg->capacity));
1117 /* Read the configuration if the feature was negotiated. */
1118 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg);
1119 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg);
1120 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, geometry, blkcfg);
1121 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg);
1122 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, topology, blkcfg);
1123 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, writeback, blkcfg);
1126 #undef VTBLK_GET_CONFIG
1129 vtblk_ident(struct vtblk_softc *sc)
1133 struct vtblk_request *req;
1136 dp = sc->vtblk_disk;
1137 len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE);
1139 if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0)
1142 req = vtblk_request_dequeue(sc);
1147 req->vbr_hdr.type = VIRTIO_BLK_T_GET_ID;
1148 req->vbr_hdr.ioprio = 1;
1149 req->vbr_hdr.sector = 0;
1154 buf.bio_cmd = BIO_READ;
1155 buf.bio_data = dp->d_ident;
1156 buf.bio_bcount = len;
1159 error = vtblk_poll_request(sc, req);
1162 vtblk_request_enqueue(sc, req);
1165 device_printf(sc->vtblk_dev,
1166 "error getting device identifier: %d\n", error);
1171 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req)
1173 struct virtqueue *vq;
1178 if (!virtqueue_empty(vq))
1181 error = vtblk_request_execute(sc, req);
1185 virtqueue_notify(vq);
1186 virtqueue_poll(vq, NULL);
1188 error = vtblk_request_error(req);
1189 if (error && bootverbose) {
1190 device_printf(sc->vtblk_dev,
1191 "%s: IO error: %d\n", __func__, error);
1198 vtblk_quiesce(struct vtblk_softc *sc)
1202 VTBLK_LOCK_ASSERT(sc);
1205 while (!virtqueue_empty(sc->vtblk_vq)) {
1206 if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq",
1207 VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) {
1217 vtblk_vq_intr(void *xsc)
1219 struct vtblk_softc *sc;
1220 struct virtqueue *vq;
1221 struct bio_queue queue;
1230 if (sc->vtblk_flags & VTBLK_FLAG_DETACH)
1233 vtblk_queue_completed(sc, &queue);
1236 if (virtqueue_enable_intr(vq) != 0) {
1237 virtqueue_disable_intr(vq);
1241 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1242 wakeup(&sc->vtblk_vq);
1246 vtblk_done_completed(sc, &queue);
1250 vtblk_stop(struct vtblk_softc *sc)
1253 virtqueue_disable_intr(sc->vtblk_vq);
1254 virtio_stop(sc->vtblk_dev);
1258 vtblk_dump_quiesce(struct vtblk_softc *sc)
1262 * Spin here until all the requests in-flight at the time of the
1263 * dump are completed and queued. The queued requests will be
1264 * biodone'd once the dump is finished.
1266 while (!virtqueue_empty(sc->vtblk_vq))
1267 vtblk_queue_completed(sc, &sc->vtblk_dump_queue);
1271 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset,
1275 struct vtblk_request *req;
1277 req = &sc->vtblk_dump_request;
1279 req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
1280 req->vbr_hdr.ioprio = 1;
1281 req->vbr_hdr.sector = offset / 512;
1286 buf.bio_cmd = BIO_WRITE;
1287 buf.bio_data = virtual;
1288 buf.bio_bcount = length;
1290 return (vtblk_poll_request(sc, req));
1294 vtblk_dump_flush(struct vtblk_softc *sc)
1297 struct vtblk_request *req;
1299 req = &sc->vtblk_dump_request;
1301 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
1302 req->vbr_hdr.ioprio = 1;
1303 req->vbr_hdr.sector = 0;
1308 buf.bio_cmd = BIO_FLUSH;
1310 return (vtblk_poll_request(sc, req));
1314 vtblk_dump_complete(struct vtblk_softc *sc)
1317 vtblk_dump_flush(sc);
1320 vtblk_done_completed(sc, &sc->vtblk_dump_queue);
1325 vtblk_set_write_cache(struct vtblk_softc *sc, int wc)
1328 /* Set either writeback (1) or writethrough (0) mode. */
1329 virtio_write_dev_config_1(sc->vtblk_dev,
1330 offsetof(struct virtio_blk_config, writeback), wc);
1334 vtblk_write_cache_enabled(struct vtblk_softc *sc,
1335 struct virtio_blk_config *blkcfg)
1339 if (sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) {
1340 wc = vtblk_tunable_int(sc, "writecache_mode",
1341 vtblk_writecache_mode);
1342 if (wc >= 0 && wc < VTBLK_CACHE_MAX)
1343 vtblk_set_write_cache(sc, wc);
1345 wc = blkcfg->writeback;
1347 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_WCE);
1353 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS)
1355 struct vtblk_softc *sc;
1358 sc = oidp->oid_arg1;
1359 wc = sc->vtblk_write_cache;
1361 error = sysctl_handle_int(oidp, &wc, 0, req);
1362 if (error || req->newptr == NULL)
1364 if ((sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) == 0)
1366 if (wc < 0 || wc >= VTBLK_CACHE_MAX)
1370 sc->vtblk_write_cache = wc;
1371 vtblk_set_write_cache(sc, sc->vtblk_write_cache);
1378 vtblk_setup_sysctl(struct vtblk_softc *sc)
1381 struct sysctl_ctx_list *ctx;
1382 struct sysctl_oid *tree;
1383 struct sysctl_oid_list *child;
1385 dev = sc->vtblk_dev;
1386 ctx = device_get_sysctl_ctx(dev);
1387 tree = device_get_sysctl_tree(dev);
1388 child = SYSCTL_CHILDREN(tree);
1390 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode",
1391 CTLTYPE_INT | CTLFLAG_RW, sc, 0, vtblk_write_cache_sysctl,
1392 "I", "Write cache mode (writethrough (0) or writeback (1))");
1396 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def)
1400 snprintf(path, sizeof(path),
1401 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob);
1402 TUNABLE_INT_FETCH(path, &def);