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
84 #define VTBLK_FLAG_DISCARD 0x0040
86 struct virtqueue *vtblk_vq;
87 struct sglist *vtblk_sglist;
88 struct disk *vtblk_disk;
90 struct bio_queue_head vtblk_bioq;
91 TAILQ_HEAD(, vtblk_request)
93 TAILQ_HEAD(, vtblk_request)
95 struct vtblk_request *vtblk_req_ordered;
98 int vtblk_request_count;
99 enum vtblk_cache_mode vtblk_write_cache;
101 struct bio_queue vtblk_dump_queue;
102 struct vtblk_request vtblk_dump_request;
105 static struct virtio_feature_desc vtblk_feature_desc[] = {
106 { VIRTIO_BLK_F_BARRIER, "HostBarrier" },
107 { VIRTIO_BLK_F_SIZE_MAX, "MaxSegSize" },
108 { VIRTIO_BLK_F_SEG_MAX, "MaxNumSegs" },
109 { VIRTIO_BLK_F_GEOMETRY, "DiskGeometry" },
110 { VIRTIO_BLK_F_RO, "ReadOnly" },
111 { VIRTIO_BLK_F_BLK_SIZE, "BlockSize" },
112 { VIRTIO_BLK_F_SCSI, "SCSICmds" },
113 { VIRTIO_BLK_F_WCE, "WriteCache" },
114 { VIRTIO_BLK_F_TOPOLOGY, "Topology" },
115 { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" },
116 { VIRTIO_BLK_F_DISCARD, "Discard" },
120 static int vtblk_modevent(module_t, int, void *);
122 static int vtblk_probe(device_t);
123 static int vtblk_attach(device_t);
124 static int vtblk_detach(device_t);
125 static int vtblk_suspend(device_t);
126 static int vtblk_resume(device_t);
127 static int vtblk_shutdown(device_t);
128 static int vtblk_config_change(device_t);
130 static int vtblk_open(struct disk *);
131 static int vtblk_close(struct disk *);
132 static int vtblk_ioctl(struct disk *, u_long, void *, int,
134 static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t);
135 static void vtblk_strategy(struct bio *);
137 static void vtblk_negotiate_features(struct vtblk_softc *);
138 static void vtblk_setup_features(struct vtblk_softc *);
139 static int vtblk_maximum_segments(struct vtblk_softc *,
140 struct virtio_blk_config *);
141 static int vtblk_alloc_virtqueue(struct vtblk_softc *);
142 static void vtblk_resize_disk(struct vtblk_softc *, uint64_t);
143 static void vtblk_alloc_disk(struct vtblk_softc *,
144 struct virtio_blk_config *);
145 static void vtblk_create_disk(struct vtblk_softc *);
147 static int vtblk_request_prealloc(struct vtblk_softc *);
148 static void vtblk_request_free(struct vtblk_softc *);
149 static struct vtblk_request *
150 vtblk_request_dequeue(struct vtblk_softc *);
151 static void vtblk_request_enqueue(struct vtblk_softc *,
152 struct vtblk_request *);
153 static struct vtblk_request *
154 vtblk_request_next_ready(struct vtblk_softc *);
155 static void vtblk_request_requeue_ready(struct vtblk_softc *,
156 struct vtblk_request *);
157 static struct vtblk_request *
158 vtblk_request_next(struct vtblk_softc *);
159 static struct vtblk_request *
160 vtblk_request_bio(struct vtblk_softc *);
161 static int vtblk_request_execute(struct vtblk_softc *,
162 struct vtblk_request *);
163 static int vtblk_request_error(struct vtblk_request *);
165 static void vtblk_queue_completed(struct vtblk_softc *,
167 static void vtblk_done_completed(struct vtblk_softc *,
169 static void vtblk_drain_vq(struct vtblk_softc *);
170 static void vtblk_drain(struct vtblk_softc *);
172 static void vtblk_startio(struct vtblk_softc *);
173 static void vtblk_bio_done(struct vtblk_softc *, struct bio *, int);
175 static void vtblk_read_config(struct vtblk_softc *,
176 struct virtio_blk_config *);
177 static void vtblk_ident(struct vtblk_softc *);
178 static int vtblk_poll_request(struct vtblk_softc *,
179 struct vtblk_request *);
180 static int vtblk_quiesce(struct vtblk_softc *);
181 static void vtblk_vq_intr(void *);
182 static void vtblk_stop(struct vtblk_softc *);
184 static void vtblk_dump_quiesce(struct vtblk_softc *);
185 static int vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t);
186 static int vtblk_dump_flush(struct vtblk_softc *);
187 static void vtblk_dump_complete(struct vtblk_softc *);
189 static void vtblk_set_write_cache(struct vtblk_softc *, int);
190 static int vtblk_write_cache_enabled(struct vtblk_softc *sc,
191 struct virtio_blk_config *);
192 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS);
194 static void vtblk_setup_sysctl(struct vtblk_softc *);
195 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int);
198 static int vtblk_no_ident = 0;
199 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident);
200 static int vtblk_writecache_mode = -1;
201 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode);
203 /* Features desired/implemented by this driver. */
204 #define VTBLK_FEATURES \
205 (VIRTIO_BLK_F_BARRIER | \
206 VIRTIO_BLK_F_SIZE_MAX | \
207 VIRTIO_BLK_F_SEG_MAX | \
208 VIRTIO_BLK_F_GEOMETRY | \
210 VIRTIO_BLK_F_BLK_SIZE | \
212 VIRTIO_BLK_F_TOPOLOGY | \
213 VIRTIO_BLK_F_CONFIG_WCE | \
214 VIRTIO_BLK_F_DISCARD | \
215 VIRTIO_RING_F_INDIRECT_DESC)
217 #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx
218 #define VTBLK_LOCK_INIT(_sc, _name) \
219 mtx_init(VTBLK_MTX((_sc)), (_name), \
220 "VirtIO Block Lock", MTX_DEF)
221 #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc)))
222 #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc)))
223 #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc)))
224 #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED)
225 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \
226 mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED)
228 #define VTBLK_DISK_NAME "vtbd"
229 #define VTBLK_QUIESCE_TIMEOUT (30 * hz)
232 * Each block request uses at least two segments - one for the header
233 * and one for the status.
235 #define VTBLK_MIN_SEGMENTS 2
237 static device_method_t vtblk_methods[] = {
238 /* Device methods. */
239 DEVMETHOD(device_probe, vtblk_probe),
240 DEVMETHOD(device_attach, vtblk_attach),
241 DEVMETHOD(device_detach, vtblk_detach),
242 DEVMETHOD(device_suspend, vtblk_suspend),
243 DEVMETHOD(device_resume, vtblk_resume),
244 DEVMETHOD(device_shutdown, vtblk_shutdown),
246 /* VirtIO methods. */
247 DEVMETHOD(virtio_config_change, vtblk_config_change),
252 static driver_t vtblk_driver = {
255 sizeof(struct vtblk_softc)
257 static devclass_t vtblk_devclass;
259 DRIVER_MODULE(virtio_blk, virtio_mmio, vtblk_driver, vtblk_devclass,
261 DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass,
263 MODULE_VERSION(virtio_blk, 1);
264 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1);
266 VIRTIO_SIMPLE_PNPTABLE(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter");
267 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, virtio_blk);
268 VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_blk);
271 vtblk_modevent(module_t mod, int type, void *unused)
292 vtblk_probe(device_t dev)
294 return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk));
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 * VTBLK_BSIZE;
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 bp->bio_cmd == BIO_DELETE)) {
551 vtblk_bio_done(sc, bp, EROFS);
555 if ((bp->bio_cmd != BIO_READ) && (bp->bio_cmd != BIO_WRITE) &&
556 (bp->bio_cmd != BIO_FLUSH) && (bp->bio_cmd != BIO_DELETE)) {
557 vtblk_bio_done(sc, bp, EOPNOTSUPP);
563 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) {
565 vtblk_bio_done(sc, bp, ENXIO);
569 if ((bp->bio_cmd == BIO_DELETE) &&
570 !(sc->vtblk_flags & VTBLK_FLAG_DISCARD)) {
572 vtblk_bio_done(sc, bp, EOPNOTSUPP);
576 bioq_insert_tail(&sc->vtblk_bioq, bp);
583 vtblk_negotiate_features(struct vtblk_softc *sc)
589 features = VTBLK_FEATURES;
591 sc->vtblk_features = virtio_negotiate_features(dev, features);
595 vtblk_setup_features(struct vtblk_softc *sc)
601 vtblk_negotiate_features(sc);
603 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
604 sc->vtblk_flags |= VTBLK_FLAG_INDIRECT;
605 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO))
606 sc->vtblk_flags |= VTBLK_FLAG_READONLY;
607 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER))
608 sc->vtblk_flags |= VTBLK_FLAG_BARRIER;
609 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE))
610 sc->vtblk_flags |= VTBLK_FLAG_WC_CONFIG;
611 if (virtio_with_feature(dev, VIRTIO_BLK_F_DISCARD))
612 sc->vtblk_flags |= VTBLK_FLAG_DISCARD;
616 vtblk_maximum_segments(struct vtblk_softc *sc,
617 struct virtio_blk_config *blkcfg)
623 nsegs = VTBLK_MIN_SEGMENTS;
625 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) {
626 nsegs += MIN(blkcfg->seg_max, maxphys / PAGE_SIZE + 1);
627 if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT)
628 nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT);
636 vtblk_alloc_virtqueue(struct vtblk_softc *sc)
639 struct vq_alloc_info vq_info;
643 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs,
644 vtblk_vq_intr, sc, &sc->vtblk_vq,
645 "%s request", device_get_nameunit(dev));
647 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info));
651 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity)
660 dp->d_mediasize = new_capacity;
662 device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n",
663 (uintmax_t) dp->d_mediasize >> 20,
664 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
668 error = disk_resize(dp, M_NOWAIT);
671 "disk_resize(9) failed, error: %d\n", error);
676 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
683 sc->vtblk_disk = dp = disk_alloc();
684 dp->d_open = vtblk_open;
685 dp->d_close = vtblk_close;
686 dp->d_ioctl = vtblk_ioctl;
687 dp->d_strategy = vtblk_strategy;
688 dp->d_name = VTBLK_DISK_NAME;
689 dp->d_unit = device_get_unit(dev);
691 dp->d_flags = DISKFLAG_CANFLUSHCACHE | DISKFLAG_UNMAPPED_BIO |
692 DISKFLAG_DIRECT_COMPLETION;
693 dp->d_hba_vendor = virtio_get_vendor(dev);
694 dp->d_hba_device = virtio_get_device(dev);
695 dp->d_hba_subvendor = virtio_get_subvendor(dev);
696 dp->d_hba_subdevice = virtio_get_subdevice(dev);
698 if ((sc->vtblk_flags & VTBLK_FLAG_READONLY) == 0)
699 dp->d_dump = vtblk_dump;
701 /* Capacity is always in 512-byte units. */
702 dp->d_mediasize = blkcfg->capacity * VTBLK_BSIZE;
704 if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE))
705 dp->d_sectorsize = blkcfg->blk_size;
707 dp->d_sectorsize = VTBLK_BSIZE;
710 * The VirtIO maximum I/O size is given in terms of segments.
711 * However, FreeBSD limits I/O size by logical buffer size, not
712 * by physically contiguous pages. Therefore, we have to assume
713 * no pages are contiguous. This may impose an artificially low
714 * maximum I/O size. But in practice, since QEMU advertises 128
715 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE,
716 * which is typically greater than maxphys. Eventually we should
717 * just advertise maxphys and split buffers that are too big.
719 * Note we must subtract one additional segment in case of non
720 * page aligned buffers.
722 dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) *
724 if (dp->d_maxsize < PAGE_SIZE)
725 dp->d_maxsize = PAGE_SIZE; /* XXX */
727 if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) {
728 dp->d_fwsectors = blkcfg->geometry.sectors;
729 dp->d_fwheads = blkcfg->geometry.heads;
732 if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) &&
733 blkcfg->topology.physical_block_exp > 0) {
734 dp->d_stripesize = dp->d_sectorsize *
735 (1 << blkcfg->topology.physical_block_exp);
736 dp->d_stripeoffset = (dp->d_stripesize -
737 blkcfg->topology.alignment_offset * dp->d_sectorsize) %
741 if (virtio_with_feature(dev, VIRTIO_BLK_F_DISCARD)) {
742 dp->d_flags |= DISKFLAG_CANDELETE;
743 dp->d_delmaxsize = blkcfg->max_discard_sectors * VTBLK_BSIZE;
746 if (vtblk_write_cache_enabled(sc, blkcfg) != 0)
747 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK;
749 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH;
753 vtblk_create_disk(struct vtblk_softc *sc)
761 device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n",
762 (uintmax_t) dp->d_mediasize >> 20,
763 (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
766 disk_create(dp, DISK_VERSION);
770 vtblk_request_prealloc(struct vtblk_softc *sc)
772 struct vtblk_request *req;
775 nreqs = virtqueue_size(sc->vtblk_vq);
778 * Preallocate sufficient requests to keep the virtqueue full. Each
779 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce
780 * the number allocated when indirect descriptors are not available.
782 if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0)
783 nreqs /= VTBLK_MIN_SEGMENTS;
785 for (i = 0; i < nreqs; i++) {
786 req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT);
790 MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1);
791 MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1);
793 sc->vtblk_request_count++;
794 vtblk_request_enqueue(sc, req);
801 vtblk_request_free(struct vtblk_softc *sc)
803 struct vtblk_request *req;
805 MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready));
807 while ((req = vtblk_request_dequeue(sc)) != NULL) {
808 sc->vtblk_request_count--;
812 KASSERT(sc->vtblk_request_count == 0,
813 ("%s: leaked %d requests", __func__, sc->vtblk_request_count));
816 static struct vtblk_request *
817 vtblk_request_dequeue(struct vtblk_softc *sc)
819 struct vtblk_request *req;
821 req = TAILQ_FIRST(&sc->vtblk_req_free);
823 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link);
824 bzero(req, sizeof(struct vtblk_request));
831 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req)
834 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link);
837 static struct vtblk_request *
838 vtblk_request_next_ready(struct vtblk_softc *sc)
840 struct vtblk_request *req;
842 req = TAILQ_FIRST(&sc->vtblk_req_ready);
844 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link);
850 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req)
853 /* NOTE: Currently, there will be at most one request in the queue. */
854 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link);
857 static struct vtblk_request *
858 vtblk_request_next(struct vtblk_softc *sc)
860 struct vtblk_request *req;
862 req = vtblk_request_next_ready(sc);
866 return (vtblk_request_bio(sc));
869 static struct vtblk_request *
870 vtblk_request_bio(struct vtblk_softc *sc)
872 struct bio_queue_head *bioq;
873 struct vtblk_request *req;
876 bioq = &sc->vtblk_bioq;
878 if (bioq_first(bioq) == NULL)
881 req = vtblk_request_dequeue(sc);
885 bp = bioq_takefirst(bioq);
888 req->vbr_hdr.ioprio = 1;
890 switch (bp->bio_cmd) {
892 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
895 req->vbr_hdr.type = VIRTIO_BLK_T_IN;
896 req->vbr_hdr.sector = bp->bio_offset / VTBLK_BSIZE;
899 req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
900 req->vbr_hdr.sector = bp->bio_offset / VTBLK_BSIZE;
903 req->vbr_hdr.type = VIRTIO_BLK_T_DISCARD;
904 req->vbr_hdr.sector = bp->bio_offset / VTBLK_BSIZE;
907 panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd);
910 if (bp->bio_flags & BIO_ORDERED)
911 req->vbr_hdr.type |= VIRTIO_BLK_T_BARRIER;
917 vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req)
919 struct virtqueue *vq;
922 int ordered, readable, writable, error;
925 sg = sc->vtblk_sglist;
931 * Some hosts (such as bhyve) do not implement the barrier feature,
932 * so we emulate it in the driver by allowing the barrier request
933 * to be the only one in flight.
935 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) {
936 if (sc->vtblk_req_ordered != NULL)
938 if (bp->bio_flags & BIO_ORDERED) {
939 if (!virtqueue_empty(vq))
942 req->vbr_hdr.type &= ~VIRTIO_BLK_T_BARRIER;
947 sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr));
949 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
950 error = sglist_append_bio(sg, bp);
951 if (error || sg->sg_nseg == sg->sg_maxseg) {
952 panic("%s: bio %p data buffer too big %d",
953 __func__, bp, error);
956 /* BIO_READ means the host writes into our buffer. */
957 if (bp->bio_cmd == BIO_READ)
958 writable = sg->sg_nseg - 1;
959 } else if (bp->bio_cmd == BIO_DELETE) {
960 struct virtio_blk_discard_write_zeroes *discard;
962 discard = malloc(sizeof(*discard), M_DEVBUF, M_NOWAIT | M_ZERO);
965 discard->sector = bp->bio_offset / VTBLK_BSIZE;
966 discard->num_sectors = bp->bio_bcount / VTBLK_BSIZE;
967 bp->bio_driver1 = discard;
968 error = sglist_append(sg, discard, sizeof(*discard));
969 if (error || sg->sg_nseg == sg->sg_maxseg) {
970 panic("%s: bio %p data buffer too big %d",
971 __func__, bp, error);
976 sglist_append(sg, &req->vbr_ack, sizeof(uint8_t));
977 readable = sg->sg_nseg - writable;
979 error = virtqueue_enqueue(vq, req, sg, readable, writable);
980 if (error == 0 && ordered)
981 sc->vtblk_req_ordered = req;
987 vtblk_request_error(struct vtblk_request *req)
991 switch (req->vbr_ack) {
992 case VIRTIO_BLK_S_OK:
995 case VIRTIO_BLK_S_UNSUPP:
1007 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue)
1009 struct vtblk_request *req;
1012 while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) {
1013 if (sc->vtblk_req_ordered != NULL) {
1014 MPASS(sc->vtblk_req_ordered == req);
1015 sc->vtblk_req_ordered = NULL;
1019 bp->bio_error = vtblk_request_error(req);
1020 TAILQ_INSERT_TAIL(queue, bp, bio_queue);
1022 vtblk_request_enqueue(sc, req);
1027 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue)
1029 struct bio *bp, *tmp;
1031 TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) {
1032 if (bp->bio_error != 0)
1033 disk_err(bp, "hard error", -1, 1);
1034 vtblk_bio_done(sc, bp, bp->bio_error);
1039 vtblk_drain_vq(struct vtblk_softc *sc)
1041 struct virtqueue *vq;
1042 struct vtblk_request *req;
1048 while ((req = virtqueue_drain(vq, &last)) != NULL) {
1049 vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1050 vtblk_request_enqueue(sc, req);
1053 sc->vtblk_req_ordered = NULL;
1054 KASSERT(virtqueue_empty(vq), ("virtqueue not empty"));
1058 vtblk_drain(struct vtblk_softc *sc)
1060 struct bio_queue queue;
1061 struct bio_queue_head *bioq;
1062 struct vtblk_request *req;
1065 bioq = &sc->vtblk_bioq;
1068 if (sc->vtblk_vq != NULL) {
1069 vtblk_queue_completed(sc, &queue);
1070 vtblk_done_completed(sc, &queue);
1075 while ((req = vtblk_request_next_ready(sc)) != NULL) {
1076 vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1077 vtblk_request_enqueue(sc, req);
1080 while (bioq_first(bioq) != NULL) {
1081 bp = bioq_takefirst(bioq);
1082 vtblk_bio_done(sc, bp, ENXIO);
1085 vtblk_request_free(sc);
1089 vtblk_startio(struct vtblk_softc *sc)
1091 struct virtqueue *vq;
1092 struct vtblk_request *req;
1095 VTBLK_LOCK_ASSERT(sc);
1099 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1102 while (!virtqueue_full(vq)) {
1103 req = vtblk_request_next(sc);
1107 if (vtblk_request_execute(sc, req) != 0) {
1108 vtblk_request_requeue_ready(sc, req);
1116 virtqueue_notify(vq);
1120 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error)
1123 /* Because of GEOM direct dispatch, we cannot hold any locks. */
1125 VTBLK_LOCK_ASSERT_NOTOWNED(sc);
1128 bp->bio_resid = bp->bio_bcount;
1129 bp->bio_error = error;
1130 bp->bio_flags |= BIO_ERROR;
1133 if (bp->bio_driver1 != NULL) {
1134 free(bp->bio_driver1, M_DEVBUF);
1135 bp->bio_driver1 = NULL;
1141 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \
1142 if (virtio_with_feature(_dev, _feature)) { \
1143 virtio_read_device_config(_dev, \
1144 offsetof(struct virtio_blk_config, _field), \
1145 &(_cfg)->_field, sizeof((_cfg)->_field)); \
1149 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
1153 dev = sc->vtblk_dev;
1155 bzero(blkcfg, sizeof(struct virtio_blk_config));
1157 /* The capacity is always available. */
1158 virtio_read_device_config(dev, offsetof(struct virtio_blk_config,
1159 capacity), &blkcfg->capacity, sizeof(blkcfg->capacity));
1161 /* Read the configuration if the feature was negotiated. */
1162 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg);
1163 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg);
1164 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, geometry, blkcfg);
1165 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg);
1166 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, topology, blkcfg);
1167 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, wce, blkcfg);
1168 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_sectors,
1170 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_seg, blkcfg);
1171 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, discard_sector_alignment,
1175 #undef VTBLK_GET_CONFIG
1178 vtblk_ident(struct vtblk_softc *sc)
1182 struct vtblk_request *req;
1185 dp = sc->vtblk_disk;
1186 len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE);
1188 if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0)
1191 req = vtblk_request_dequeue(sc);
1196 req->vbr_hdr.type = VIRTIO_BLK_T_GET_ID;
1197 req->vbr_hdr.ioprio = 1;
1198 req->vbr_hdr.sector = 0;
1203 buf.bio_cmd = BIO_READ;
1204 buf.bio_data = dp->d_ident;
1205 buf.bio_bcount = len;
1208 error = vtblk_poll_request(sc, req);
1211 vtblk_request_enqueue(sc, req);
1214 device_printf(sc->vtblk_dev,
1215 "error getting device identifier: %d\n", error);
1220 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req)
1222 struct virtqueue *vq;
1227 if (!virtqueue_empty(vq))
1230 error = vtblk_request_execute(sc, req);
1234 virtqueue_notify(vq);
1235 virtqueue_poll(vq, NULL);
1237 error = vtblk_request_error(req);
1238 if (error && bootverbose) {
1239 device_printf(sc->vtblk_dev,
1240 "%s: IO error: %d\n", __func__, error);
1247 vtblk_quiesce(struct vtblk_softc *sc)
1251 VTBLK_LOCK_ASSERT(sc);
1254 while (!virtqueue_empty(sc->vtblk_vq)) {
1255 if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq",
1256 VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) {
1266 vtblk_vq_intr(void *xsc)
1268 struct vtblk_softc *sc;
1269 struct virtqueue *vq;
1270 struct bio_queue queue;
1279 if (sc->vtblk_flags & VTBLK_FLAG_DETACH)
1282 vtblk_queue_completed(sc, &queue);
1285 if (virtqueue_enable_intr(vq) != 0) {
1286 virtqueue_disable_intr(vq);
1290 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1291 wakeup(&sc->vtblk_vq);
1295 vtblk_done_completed(sc, &queue);
1299 vtblk_stop(struct vtblk_softc *sc)
1302 virtqueue_disable_intr(sc->vtblk_vq);
1303 virtio_stop(sc->vtblk_dev);
1307 vtblk_dump_quiesce(struct vtblk_softc *sc)
1311 * Spin here until all the requests in-flight at the time of the
1312 * dump are completed and queued. The queued requests will be
1313 * biodone'd once the dump is finished.
1315 while (!virtqueue_empty(sc->vtblk_vq))
1316 vtblk_queue_completed(sc, &sc->vtblk_dump_queue);
1320 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset,
1324 struct vtblk_request *req;
1326 req = &sc->vtblk_dump_request;
1328 req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
1329 req->vbr_hdr.ioprio = 1;
1330 req->vbr_hdr.sector = offset / VTBLK_BSIZE;
1335 buf.bio_cmd = BIO_WRITE;
1336 buf.bio_data = virtual;
1337 buf.bio_bcount = length;
1339 return (vtblk_poll_request(sc, req));
1343 vtblk_dump_flush(struct vtblk_softc *sc)
1346 struct vtblk_request *req;
1348 req = &sc->vtblk_dump_request;
1350 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
1351 req->vbr_hdr.ioprio = 1;
1352 req->vbr_hdr.sector = 0;
1357 buf.bio_cmd = BIO_FLUSH;
1359 return (vtblk_poll_request(sc, req));
1363 vtblk_dump_complete(struct vtblk_softc *sc)
1366 vtblk_dump_flush(sc);
1369 vtblk_done_completed(sc, &sc->vtblk_dump_queue);
1374 vtblk_set_write_cache(struct vtblk_softc *sc, int wc)
1377 /* Set either writeback (1) or writethrough (0) mode. */
1378 virtio_write_dev_config_1(sc->vtblk_dev,
1379 offsetof(struct virtio_blk_config, wce), wc);
1383 vtblk_write_cache_enabled(struct vtblk_softc *sc,
1384 struct virtio_blk_config *blkcfg)
1388 if (sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) {
1389 wc = vtblk_tunable_int(sc, "writecache_mode",
1390 vtblk_writecache_mode);
1391 if (wc >= 0 && wc < VTBLK_CACHE_MAX)
1392 vtblk_set_write_cache(sc, wc);
1396 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_WCE);
1402 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS)
1404 struct vtblk_softc *sc;
1407 sc = oidp->oid_arg1;
1408 wc = sc->vtblk_write_cache;
1410 error = sysctl_handle_int(oidp, &wc, 0, req);
1411 if (error || req->newptr == NULL)
1413 if ((sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) == 0)
1415 if (wc < 0 || wc >= VTBLK_CACHE_MAX)
1419 sc->vtblk_write_cache = wc;
1420 vtblk_set_write_cache(sc, sc->vtblk_write_cache);
1427 vtblk_setup_sysctl(struct vtblk_softc *sc)
1430 struct sysctl_ctx_list *ctx;
1431 struct sysctl_oid *tree;
1432 struct sysctl_oid_list *child;
1434 dev = sc->vtblk_dev;
1435 ctx = device_get_sysctl_ctx(dev);
1436 tree = device_get_sysctl_tree(dev);
1437 child = SYSCTL_CHILDREN(tree);
1439 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode",
1440 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
1441 vtblk_write_cache_sysctl, "I",
1442 "Write cache mode (writethrough (0) or writeback (1))");
1446 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def)
1450 snprintf(path, sizeof(path),
1451 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob);
1452 TUNABLE_INT_FETCH(path, &def);