1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice unmodified, this list of conditions, and the following 12 * disclaimer. 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. 16 * 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. 27 */ 28 29 /* Driver for VirtIO block devices. */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/bio.h> 38 #include <sys/malloc.h> 39 #include <sys/module.h> 40 #include <sys/msan.h> 41 #include <sys/sglist.h> 42 #include <sys/sysctl.h> 43 #include <sys/lock.h> 44 #include <sys/mutex.h> 45 #include <sys/queue.h> 46 47 #include <geom/geom.h> 48 #include <geom/geom_disk.h> 49 50 #include <machine/bus.h> 51 #include <machine/resource.h> 52 #include <sys/bus.h> 53 #include <sys/rman.h> 54 55 #include <dev/virtio/virtio.h> 56 #include <dev/virtio/virtqueue.h> 57 #include <dev/virtio/block/virtio_blk.h> 58 59 #include "virtio_if.h" 60 61 struct vtblk_request { 62 struct virtio_blk_outhdr vbr_hdr; 63 struct bio *vbr_bp; 64 uint8_t vbr_ack; 65 TAILQ_ENTRY(vtblk_request) vbr_link; 66 }; 67 68 enum vtblk_cache_mode { 69 VTBLK_CACHE_WRITETHROUGH, 70 VTBLK_CACHE_WRITEBACK, 71 VTBLK_CACHE_MAX 72 }; 73 74 struct vtblk_softc { 75 device_t vtblk_dev; 76 struct mtx vtblk_mtx; 77 uint64_t vtblk_features; 78 uint32_t vtblk_flags; 79 #define VTBLK_FLAG_INDIRECT 0x0001 80 #define VTBLK_FLAG_DETACH 0x0002 81 #define VTBLK_FLAG_SUSPEND 0x0004 82 #define VTBLK_FLAG_BARRIER 0x0008 83 #define VTBLK_FLAG_WCE_CONFIG 0x0010 84 85 struct virtqueue *vtblk_vq; 86 struct sglist *vtblk_sglist; 87 struct disk *vtblk_disk; 88 89 struct bio_queue_head vtblk_bioq; 90 TAILQ_HEAD(, vtblk_request) 91 vtblk_req_free; 92 TAILQ_HEAD(, vtblk_request) 93 vtblk_req_ready; 94 struct vtblk_request *vtblk_req_ordered; 95 96 int vtblk_max_nsegs; 97 int vtblk_request_count; 98 enum vtblk_cache_mode vtblk_write_cache; 99 100 struct bio_queue vtblk_dump_queue; 101 struct vtblk_request vtblk_dump_request; 102 }; 103 104 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" }, 118 119 { 0, NULL } 120 }; 121 122 static int vtblk_modevent(module_t, int, void *); 123 124 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_attach_completed(device_t); 131 static int vtblk_config_change(device_t); 132 133 static int vtblk_open(struct disk *); 134 static int vtblk_close(struct disk *); 135 static int vtblk_ioctl(struct disk *, u_long, void *, int, 136 struct thread *); 137 static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t); 138 static void vtblk_strategy(struct bio *); 139 140 static int vtblk_negotiate_features(struct vtblk_softc *); 141 static int vtblk_setup_features(struct vtblk_softc *); 142 static int vtblk_maximum_segments(struct vtblk_softc *, 143 struct virtio_blk_config *); 144 static int vtblk_alloc_virtqueue(struct vtblk_softc *); 145 static void vtblk_resize_disk(struct vtblk_softc *, uint64_t); 146 static void vtblk_alloc_disk(struct vtblk_softc *, 147 struct virtio_blk_config *); 148 static void vtblk_create_disk(struct vtblk_softc *); 149 150 static int vtblk_request_prealloc(struct vtblk_softc *); 151 static void vtblk_request_free(struct vtblk_softc *); 152 static struct vtblk_request * 153 vtblk_request_dequeue(struct vtblk_softc *); 154 static void vtblk_request_enqueue(struct vtblk_softc *, 155 struct vtblk_request *); 156 static struct vtblk_request * 157 vtblk_request_next_ready(struct vtblk_softc *); 158 static void vtblk_request_requeue_ready(struct vtblk_softc *, 159 struct vtblk_request *); 160 static struct vtblk_request * 161 vtblk_request_next(struct vtblk_softc *); 162 static struct vtblk_request * 163 vtblk_request_bio(struct vtblk_softc *); 164 static int vtblk_request_execute(struct vtblk_softc *, 165 struct vtblk_request *); 166 static int vtblk_request_error(struct vtblk_request *); 167 168 static void vtblk_queue_completed(struct vtblk_softc *, 169 struct bio_queue *); 170 static void vtblk_done_completed(struct vtblk_softc *, 171 struct bio_queue *); 172 static void vtblk_drain_vq(struct vtblk_softc *); 173 static void vtblk_drain(struct vtblk_softc *); 174 175 static void vtblk_startio(struct vtblk_softc *); 176 static void vtblk_bio_done(struct vtblk_softc *, struct bio *, int); 177 178 static void vtblk_read_config(struct vtblk_softc *, 179 struct virtio_blk_config *); 180 static void vtblk_ident(struct vtblk_softc *); 181 static int vtblk_poll_request(struct vtblk_softc *, 182 struct vtblk_request *); 183 static int vtblk_quiesce(struct vtblk_softc *); 184 static void vtblk_vq_intr(void *); 185 static void vtblk_stop(struct vtblk_softc *); 186 187 static void vtblk_dump_quiesce(struct vtblk_softc *); 188 static int vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t); 189 static int vtblk_dump_flush(struct vtblk_softc *); 190 static void vtblk_dump_complete(struct vtblk_softc *); 191 192 static void vtblk_set_write_cache(struct vtblk_softc *, int); 193 static int vtblk_write_cache_enabled(struct vtblk_softc *sc, 194 struct virtio_blk_config *); 195 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS); 196 197 static void vtblk_setup_sysctl(struct vtblk_softc *); 198 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int); 199 200 #define vtblk_modern(_sc) (((_sc)->vtblk_features & VIRTIO_F_VERSION_1) != 0) 201 #define vtblk_htog16(_sc, _val) virtio_htog16(vtblk_modern(_sc), _val) 202 #define vtblk_htog32(_sc, _val) virtio_htog32(vtblk_modern(_sc), _val) 203 #define vtblk_htog64(_sc, _val) virtio_htog64(vtblk_modern(_sc), _val) 204 #define vtblk_gtoh16(_sc, _val) virtio_gtoh16(vtblk_modern(_sc), _val) 205 #define vtblk_gtoh32(_sc, _val) virtio_gtoh32(vtblk_modern(_sc), _val) 206 #define vtblk_gtoh64(_sc, _val) virtio_gtoh64(vtblk_modern(_sc), _val) 207 208 /* Tunables. */ 209 static int vtblk_no_ident = 0; 210 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident); 211 static int vtblk_writecache_mode = -1; 212 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode); 213 214 #define VTBLK_COMMON_FEATURES \ 215 (VIRTIO_BLK_F_SIZE_MAX | \ 216 VIRTIO_BLK_F_SEG_MAX | \ 217 VIRTIO_BLK_F_GEOMETRY | \ 218 VIRTIO_BLK_F_RO | \ 219 VIRTIO_BLK_F_BLK_SIZE | \ 220 VIRTIO_BLK_F_FLUSH | \ 221 VIRTIO_BLK_F_TOPOLOGY | \ 222 VIRTIO_BLK_F_CONFIG_WCE | \ 223 VIRTIO_BLK_F_DISCARD | \ 224 VIRTIO_RING_F_INDIRECT_DESC) 225 226 #define VTBLK_MODERN_FEATURES (VTBLK_COMMON_FEATURES) 227 #define VTBLK_LEGACY_FEATURES (VIRTIO_BLK_F_BARRIER | VTBLK_COMMON_FEATURES) 228 229 #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx 230 #define VTBLK_LOCK_INIT(_sc, _name) \ 231 mtx_init(VTBLK_MTX((_sc)), (_name), \ 232 "VirtIO Block Lock", MTX_DEF) 233 #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc))) 234 #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc))) 235 #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc))) 236 #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED) 237 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \ 238 mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED) 239 240 #define VTBLK_DISK_NAME "vtbd" 241 #define VTBLK_QUIESCE_TIMEOUT (30 * hz) 242 #define VTBLK_BSIZE 512 243 244 /* 245 * Each block request uses at least two segments - one for the header 246 * and one for the status. 247 */ 248 #define VTBLK_MIN_SEGMENTS 2 249 250 static device_method_t vtblk_methods[] = { 251 /* Device methods. */ 252 DEVMETHOD(device_probe, vtblk_probe), 253 DEVMETHOD(device_attach, vtblk_attach), 254 DEVMETHOD(device_detach, vtblk_detach), 255 DEVMETHOD(device_suspend, vtblk_suspend), 256 DEVMETHOD(device_resume, vtblk_resume), 257 DEVMETHOD(device_shutdown, vtblk_shutdown), 258 259 /* VirtIO methods. */ 260 DEVMETHOD(virtio_attach_completed, vtblk_attach_completed), 261 DEVMETHOD(virtio_config_change, vtblk_config_change), 262 263 DEVMETHOD_END 264 }; 265 266 static driver_t vtblk_driver = { 267 "vtblk", 268 vtblk_methods, 269 sizeof(struct vtblk_softc) 270 }; 271 272 VIRTIO_DRIVER_MODULE(virtio_blk, vtblk_driver, vtblk_modevent, NULL); 273 MODULE_VERSION(virtio_blk, 1); 274 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1); 275 276 VIRTIO_SIMPLE_PNPINFO(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter"); 277 278 static int 279 vtblk_modevent(module_t mod, int type, void *unused) 280 { 281 int error; 282 283 error = 0; 284 285 switch (type) { 286 case MOD_LOAD: 287 case MOD_QUIESCE: 288 case MOD_UNLOAD: 289 case MOD_SHUTDOWN: 290 break; 291 default: 292 error = EOPNOTSUPP; 293 break; 294 } 295 296 return (error); 297 } 298 299 static int 300 vtblk_probe(device_t dev) 301 { 302 return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk)); 303 } 304 305 static int 306 vtblk_attach(device_t dev) 307 { 308 struct vtblk_softc *sc; 309 struct virtio_blk_config blkcfg; 310 int error; 311 312 sc = device_get_softc(dev); 313 sc->vtblk_dev = dev; 314 virtio_set_feature_desc(dev, vtblk_feature_desc); 315 316 VTBLK_LOCK_INIT(sc, device_get_nameunit(dev)); 317 bioq_init(&sc->vtblk_bioq); 318 TAILQ_INIT(&sc->vtblk_dump_queue); 319 TAILQ_INIT(&sc->vtblk_req_free); 320 TAILQ_INIT(&sc->vtblk_req_ready); 321 322 vtblk_setup_sysctl(sc); 323 324 error = vtblk_setup_features(sc); 325 if (error) { 326 device_printf(dev, "cannot setup features\n"); 327 goto fail; 328 } 329 330 vtblk_read_config(sc, &blkcfg); 331 332 /* 333 * With the current sglist(9) implementation, it is not easy 334 * for us to support a maximum segment size as adjacent 335 * segments are coalesced. For now, just make sure it's larger 336 * than the maximum supported transfer size. 337 */ 338 if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) { 339 if (blkcfg.size_max < maxphys) { 340 error = ENOTSUP; 341 device_printf(dev, "host requires unsupported " 342 "maximum segment size feature\n"); 343 goto fail; 344 } 345 } 346 347 sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg); 348 if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) { 349 error = EINVAL; 350 device_printf(dev, "fewer than minimum number of segments " 351 "allowed: %d\n", sc->vtblk_max_nsegs); 352 goto fail; 353 } 354 355 sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT); 356 if (sc->vtblk_sglist == NULL) { 357 error = ENOMEM; 358 device_printf(dev, "cannot allocate sglist\n"); 359 goto fail; 360 } 361 362 error = vtblk_alloc_virtqueue(sc); 363 if (error) { 364 device_printf(dev, "cannot allocate virtqueue\n"); 365 goto fail; 366 } 367 368 error = vtblk_request_prealloc(sc); 369 if (error) { 370 device_printf(dev, "cannot preallocate requests\n"); 371 goto fail; 372 } 373 374 vtblk_alloc_disk(sc, &blkcfg); 375 376 error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY); 377 if (error) { 378 device_printf(dev, "cannot setup virtqueue interrupt\n"); 379 goto fail; 380 } 381 382 virtqueue_enable_intr(sc->vtblk_vq); 383 384 fail: 385 if (error) 386 vtblk_detach(dev); 387 388 return (error); 389 } 390 391 static int 392 vtblk_detach(device_t dev) 393 { 394 struct vtblk_softc *sc; 395 396 sc = device_get_softc(dev); 397 398 VTBLK_LOCK(sc); 399 sc->vtblk_flags |= VTBLK_FLAG_DETACH; 400 if (device_is_attached(dev)) 401 vtblk_stop(sc); 402 VTBLK_UNLOCK(sc); 403 404 vtblk_drain(sc); 405 406 if (sc->vtblk_disk != NULL) { 407 disk_destroy(sc->vtblk_disk); 408 sc->vtblk_disk = NULL; 409 } 410 411 if (sc->vtblk_sglist != NULL) { 412 sglist_free(sc->vtblk_sglist); 413 sc->vtblk_sglist = NULL; 414 } 415 416 VTBLK_LOCK_DESTROY(sc); 417 418 return (0); 419 } 420 421 static int 422 vtblk_suspend(device_t dev) 423 { 424 struct vtblk_softc *sc; 425 int error; 426 427 sc = device_get_softc(dev); 428 429 VTBLK_LOCK(sc); 430 sc->vtblk_flags |= VTBLK_FLAG_SUSPEND; 431 /* XXX BMV: virtio_stop(), etc needed here? */ 432 error = vtblk_quiesce(sc); 433 if (error) 434 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; 435 VTBLK_UNLOCK(sc); 436 437 return (error); 438 } 439 440 static int 441 vtblk_resume(device_t dev) 442 { 443 struct vtblk_softc *sc; 444 445 sc = device_get_softc(dev); 446 447 VTBLK_LOCK(sc); 448 /* XXX BMV: virtio_reinit(), etc needed here? */ 449 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; 450 vtblk_startio(sc); 451 VTBLK_UNLOCK(sc); 452 453 return (0); 454 } 455 456 static int 457 vtblk_shutdown(device_t dev) 458 { 459 460 return (0); 461 } 462 463 static int 464 vtblk_attach_completed(device_t dev) 465 { 466 struct vtblk_softc *sc; 467 468 sc = device_get_softc(dev); 469 470 /* 471 * Create disk after attach as VIRTIO_BLK_T_GET_ID can only be 472 * processed after the device acknowledged 473 * VIRTIO_CONFIG_STATUS_DRIVER_OK. 474 */ 475 vtblk_create_disk(sc); 476 return (0); 477 } 478 479 static int 480 vtblk_config_change(device_t dev) 481 { 482 struct vtblk_softc *sc; 483 struct virtio_blk_config blkcfg; 484 uint64_t capacity; 485 486 sc = device_get_softc(dev); 487 488 vtblk_read_config(sc, &blkcfg); 489 490 /* Capacity is always in 512-byte units. */ 491 capacity = blkcfg.capacity * VTBLK_BSIZE; 492 493 if (sc->vtblk_disk->d_mediasize != capacity) 494 vtblk_resize_disk(sc, capacity); 495 496 return (0); 497 } 498 499 static int 500 vtblk_open(struct disk *dp) 501 { 502 struct vtblk_softc *sc; 503 504 if ((sc = dp->d_drv1) == NULL) 505 return (ENXIO); 506 507 return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0); 508 } 509 510 static int 511 vtblk_close(struct disk *dp) 512 { 513 struct vtblk_softc *sc; 514 515 if ((sc = dp->d_drv1) == NULL) 516 return (ENXIO); 517 518 return (0); 519 } 520 521 static int 522 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, 523 struct thread *td) 524 { 525 struct vtblk_softc *sc; 526 527 if ((sc = dp->d_drv1) == NULL) 528 return (ENXIO); 529 530 return (ENOTTY); 531 } 532 533 static int 534 vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset, 535 size_t length) 536 { 537 struct disk *dp; 538 struct vtblk_softc *sc; 539 int error; 540 541 dp = arg; 542 error = 0; 543 544 if ((sc = dp->d_drv1) == NULL) 545 return (ENXIO); 546 547 VTBLK_LOCK(sc); 548 549 vtblk_dump_quiesce(sc); 550 551 if (length > 0) 552 error = vtblk_dump_write(sc, virtual, offset, length); 553 if (error || (virtual == NULL && offset == 0)) 554 vtblk_dump_complete(sc); 555 556 VTBLK_UNLOCK(sc); 557 558 return (error); 559 } 560 561 static void 562 vtblk_strategy(struct bio *bp) 563 { 564 struct vtblk_softc *sc; 565 566 if ((sc = bp->bio_disk->d_drv1) == NULL) { 567 vtblk_bio_done(NULL, bp, EINVAL); 568 return; 569 } 570 571 if ((bp->bio_cmd != BIO_READ) && (bp->bio_cmd != BIO_WRITE) && 572 (bp->bio_cmd != BIO_FLUSH) && (bp->bio_cmd != BIO_DELETE)) { 573 vtblk_bio_done(sc, bp, EOPNOTSUPP); 574 return; 575 } 576 577 VTBLK_LOCK(sc); 578 579 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) { 580 VTBLK_UNLOCK(sc); 581 vtblk_bio_done(sc, bp, ENXIO); 582 return; 583 } 584 585 bioq_insert_tail(&sc->vtblk_bioq, bp); 586 vtblk_startio(sc); 587 588 VTBLK_UNLOCK(sc); 589 } 590 591 static int 592 vtblk_negotiate_features(struct vtblk_softc *sc) 593 { 594 device_t dev; 595 uint64_t features; 596 597 dev = sc->vtblk_dev; 598 features = virtio_bus_is_modern(dev) ? VTBLK_MODERN_FEATURES : 599 VTBLK_LEGACY_FEATURES; 600 601 sc->vtblk_features = virtio_negotiate_features(dev, features); 602 return (virtio_finalize_features(dev)); 603 } 604 605 static int 606 vtblk_setup_features(struct vtblk_softc *sc) 607 { 608 device_t dev; 609 int error; 610 611 dev = sc->vtblk_dev; 612 613 error = vtblk_negotiate_features(sc); 614 if (error) 615 return (error); 616 617 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 618 sc->vtblk_flags |= VTBLK_FLAG_INDIRECT; 619 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE)) 620 sc->vtblk_flags |= VTBLK_FLAG_WCE_CONFIG; 621 622 /* Legacy. */ 623 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER)) 624 sc->vtblk_flags |= VTBLK_FLAG_BARRIER; 625 626 return (0); 627 } 628 629 static int 630 vtblk_maximum_segments(struct vtblk_softc *sc, 631 struct virtio_blk_config *blkcfg) 632 { 633 device_t dev; 634 int nsegs; 635 636 dev = sc->vtblk_dev; 637 nsegs = VTBLK_MIN_SEGMENTS; 638 639 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) { 640 nsegs += MIN(blkcfg->seg_max, maxphys / PAGE_SIZE + 1); 641 if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT) 642 nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT); 643 } else 644 nsegs += 1; 645 646 return (nsegs); 647 } 648 649 static int 650 vtblk_alloc_virtqueue(struct vtblk_softc *sc) 651 { 652 device_t dev; 653 struct vq_alloc_info vq_info; 654 655 dev = sc->vtblk_dev; 656 657 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs, 658 vtblk_vq_intr, sc, &sc->vtblk_vq, 659 "%s request", device_get_nameunit(dev)); 660 661 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info)); 662 } 663 664 static void 665 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity) 666 { 667 device_t dev; 668 struct disk *dp; 669 int error; 670 671 dev = sc->vtblk_dev; 672 dp = sc->vtblk_disk; 673 674 dp->d_mediasize = new_capacity; 675 if (bootverbose) { 676 device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n", 677 (uintmax_t) dp->d_mediasize >> 20, 678 (uintmax_t) dp->d_mediasize / dp->d_sectorsize, 679 dp->d_sectorsize); 680 } 681 682 error = disk_resize(dp, M_NOWAIT); 683 if (error) { 684 device_printf(dev, 685 "disk_resize(9) failed, error: %d\n", error); 686 } 687 } 688 689 static void 690 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) 691 { 692 device_t dev; 693 struct disk *dp; 694 695 dev = sc->vtblk_dev; 696 697 sc->vtblk_disk = dp = disk_alloc(); 698 dp->d_open = vtblk_open; 699 dp->d_close = vtblk_close; 700 dp->d_ioctl = vtblk_ioctl; 701 dp->d_strategy = vtblk_strategy; 702 dp->d_name = VTBLK_DISK_NAME; 703 dp->d_unit = device_get_unit(dev); 704 dp->d_drv1 = sc; 705 dp->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION; 706 dp->d_hba_vendor = virtio_get_vendor(dev); 707 dp->d_hba_device = virtio_get_device(dev); 708 dp->d_hba_subvendor = virtio_get_subvendor(dev); 709 dp->d_hba_subdevice = virtio_get_subdevice(dev); 710 711 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO)) 712 dp->d_flags |= DISKFLAG_WRITE_PROTECT; 713 else { 714 if (virtio_with_feature(dev, VIRTIO_BLK_F_FLUSH)) 715 dp->d_flags |= DISKFLAG_CANFLUSHCACHE; 716 dp->d_dump = vtblk_dump; 717 } 718 719 /* Capacity is always in 512-byte units. */ 720 dp->d_mediasize = blkcfg->capacity * VTBLK_BSIZE; 721 722 if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE)) 723 dp->d_sectorsize = blkcfg->blk_size; 724 else 725 dp->d_sectorsize = VTBLK_BSIZE; 726 727 /* 728 * The VirtIO maximum I/O size is given in terms of segments. 729 * However, FreeBSD limits I/O size by logical buffer size, not 730 * by physically contiguous pages. Therefore, we have to assume 731 * no pages are contiguous. This may impose an artificially low 732 * maximum I/O size. But in practice, since QEMU advertises 128 733 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE, 734 * which is typically greater than maxphys. Eventually we should 735 * just advertise maxphys and split buffers that are too big. 736 * 737 * Note we must subtract one additional segment in case of non 738 * page aligned buffers. 739 */ 740 dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) * 741 PAGE_SIZE; 742 if (dp->d_maxsize < PAGE_SIZE) 743 dp->d_maxsize = PAGE_SIZE; /* XXX */ 744 745 if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) { 746 dp->d_fwsectors = blkcfg->geometry.sectors; 747 dp->d_fwheads = blkcfg->geometry.heads; 748 } 749 750 if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) && 751 blkcfg->topology.physical_block_exp > 0) { 752 dp->d_stripesize = dp->d_sectorsize * 753 (1 << blkcfg->topology.physical_block_exp); 754 dp->d_stripeoffset = (dp->d_stripesize - 755 blkcfg->topology.alignment_offset * dp->d_sectorsize) % 756 dp->d_stripesize; 757 } 758 759 if (virtio_with_feature(dev, VIRTIO_BLK_F_DISCARD)) { 760 dp->d_flags |= DISKFLAG_CANDELETE; 761 dp->d_delmaxsize = blkcfg->max_discard_sectors * VTBLK_BSIZE; 762 } 763 764 if (vtblk_write_cache_enabled(sc, blkcfg) != 0) 765 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK; 766 else 767 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH; 768 } 769 770 static void 771 vtblk_create_disk(struct vtblk_softc *sc) 772 { 773 struct disk *dp; 774 775 dp = sc->vtblk_disk; 776 777 vtblk_ident(sc); 778 779 device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n", 780 (uintmax_t) dp->d_mediasize >> 20, 781 (uintmax_t) dp->d_mediasize / dp->d_sectorsize, 782 dp->d_sectorsize); 783 784 disk_create(dp, DISK_VERSION); 785 } 786 787 static int 788 vtblk_request_prealloc(struct vtblk_softc *sc) 789 { 790 struct vtblk_request *req; 791 int i, nreqs; 792 793 nreqs = virtqueue_size(sc->vtblk_vq); 794 795 /* 796 * Preallocate sufficient requests to keep the virtqueue full. Each 797 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce 798 * the number allocated when indirect descriptors are not available. 799 */ 800 if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0) 801 nreqs /= VTBLK_MIN_SEGMENTS; 802 803 for (i = 0; i < nreqs; i++) { 804 req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT); 805 if (req == NULL) 806 return (ENOMEM); 807 808 MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1); 809 MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1); 810 811 sc->vtblk_request_count++; 812 vtblk_request_enqueue(sc, req); 813 } 814 815 return (0); 816 } 817 818 static void 819 vtblk_request_free(struct vtblk_softc *sc) 820 { 821 struct vtblk_request *req; 822 823 MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready)); 824 825 while ((req = vtblk_request_dequeue(sc)) != NULL) { 826 sc->vtblk_request_count--; 827 free(req, M_DEVBUF); 828 } 829 830 KASSERT(sc->vtblk_request_count == 0, 831 ("%s: leaked %d requests", __func__, sc->vtblk_request_count)); 832 } 833 834 static struct vtblk_request * 835 vtblk_request_dequeue(struct vtblk_softc *sc) 836 { 837 struct vtblk_request *req; 838 839 req = TAILQ_FIRST(&sc->vtblk_req_free); 840 if (req != NULL) { 841 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link); 842 bzero(req, sizeof(struct vtblk_request)); 843 } 844 845 return (req); 846 } 847 848 static void 849 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req) 850 { 851 852 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link); 853 } 854 855 static struct vtblk_request * 856 vtblk_request_next_ready(struct vtblk_softc *sc) 857 { 858 struct vtblk_request *req; 859 860 req = TAILQ_FIRST(&sc->vtblk_req_ready); 861 if (req != NULL) 862 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link); 863 864 return (req); 865 } 866 867 static void 868 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req) 869 { 870 871 /* NOTE: Currently, there will be at most one request in the queue. */ 872 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link); 873 } 874 875 static struct vtblk_request * 876 vtblk_request_next(struct vtblk_softc *sc) 877 { 878 struct vtblk_request *req; 879 880 req = vtblk_request_next_ready(sc); 881 if (req != NULL) 882 return (req); 883 884 return (vtblk_request_bio(sc)); 885 } 886 887 static struct vtblk_request * 888 vtblk_request_bio(struct vtblk_softc *sc) 889 { 890 struct bio_queue_head *bioq; 891 struct vtblk_request *req; 892 struct bio *bp; 893 894 bioq = &sc->vtblk_bioq; 895 896 if (bioq_first(bioq) == NULL) 897 return (NULL); 898 899 req = vtblk_request_dequeue(sc); 900 if (req == NULL) 901 return (NULL); 902 903 bp = bioq_takefirst(bioq); 904 req->vbr_bp = bp; 905 req->vbr_ack = -1; 906 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 907 908 switch (bp->bio_cmd) { 909 case BIO_FLUSH: 910 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH); 911 req->vbr_hdr.sector = 0; 912 break; 913 case BIO_READ: 914 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_IN); 915 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 916 break; 917 case BIO_WRITE: 918 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT); 919 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 920 break; 921 case BIO_DELETE: 922 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_DISCARD); 923 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 924 break; 925 default: 926 panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd); 927 } 928 929 if (bp->bio_flags & BIO_ORDERED) 930 req->vbr_hdr.type |= vtblk_gtoh32(sc, VIRTIO_BLK_T_BARRIER); 931 932 return (req); 933 } 934 935 static int 936 vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req) 937 { 938 struct virtqueue *vq; 939 struct sglist *sg; 940 struct bio *bp; 941 int ordered, readable, writable, error; 942 943 vq = sc->vtblk_vq; 944 sg = sc->vtblk_sglist; 945 bp = req->vbr_bp; 946 ordered = 0; 947 writable = 0; 948 949 /* 950 * Some hosts (such as bhyve) do not implement the barrier feature, 951 * so we emulate it in the driver by allowing the barrier request 952 * to be the only one in flight. 953 */ 954 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) { 955 if (sc->vtblk_req_ordered != NULL) 956 return (EBUSY); 957 if (bp->bio_flags & BIO_ORDERED) { 958 if (!virtqueue_empty(vq)) 959 return (EBUSY); 960 ordered = 1; 961 req->vbr_hdr.type &= vtblk_gtoh32(sc, 962 ~VIRTIO_BLK_T_BARRIER); 963 } 964 } 965 966 sglist_reset(sg); 967 sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr)); 968 969 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) { 970 error = sglist_append_bio(sg, bp); 971 if (error || sg->sg_nseg == sg->sg_maxseg) { 972 panic("%s: bio %p data buffer too big %d", 973 __func__, bp, error); 974 } 975 976 /* BIO_READ means the host writes into our buffer. */ 977 if (bp->bio_cmd == BIO_READ) 978 writable = sg->sg_nseg - 1; 979 } else if (bp->bio_cmd == BIO_DELETE) { 980 struct virtio_blk_discard_write_zeroes *discard; 981 982 discard = malloc(sizeof(*discard), M_DEVBUF, M_NOWAIT | M_ZERO); 983 if (discard == NULL) 984 return (ENOMEM); 985 986 bp->bio_driver1 = discard; 987 discard->sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 988 discard->num_sectors = vtblk_gtoh32(sc, bp->bio_bcount / VTBLK_BSIZE); 989 error = sglist_append(sg, discard, sizeof(*discard)); 990 if (error || sg->sg_nseg == sg->sg_maxseg) { 991 panic("%s: bio %p data buffer too big %d", 992 __func__, bp, error); 993 } 994 } 995 996 writable++; 997 sglist_append(sg, &req->vbr_ack, sizeof(uint8_t)); 998 readable = sg->sg_nseg - writable; 999 1000 error = virtqueue_enqueue(vq, req, sg, readable, writable); 1001 if (error == 0 && ordered) 1002 sc->vtblk_req_ordered = req; 1003 1004 return (error); 1005 } 1006 1007 static int 1008 vtblk_request_error(struct vtblk_request *req) 1009 { 1010 int error; 1011 1012 switch (req->vbr_ack) { 1013 case VIRTIO_BLK_S_OK: 1014 error = 0; 1015 break; 1016 case VIRTIO_BLK_S_UNSUPP: 1017 error = ENOTSUP; 1018 break; 1019 default: 1020 error = EIO; 1021 break; 1022 } 1023 1024 return (error); 1025 } 1026 1027 static void 1028 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue) 1029 { 1030 struct vtblk_request *req; 1031 struct bio *bp; 1032 1033 while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) { 1034 if (sc->vtblk_req_ordered != NULL) { 1035 MPASS(sc->vtblk_req_ordered == req); 1036 sc->vtblk_req_ordered = NULL; 1037 } 1038 1039 bp = req->vbr_bp; 1040 bp->bio_error = vtblk_request_error(req); 1041 TAILQ_INSERT_TAIL(queue, bp, bio_queue); 1042 1043 vtblk_request_enqueue(sc, req); 1044 } 1045 } 1046 1047 static void 1048 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue) 1049 { 1050 struct bio *bp, *tmp; 1051 1052 TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) { 1053 if (bp->bio_error != 0) 1054 disk_err(bp, "hard error", -1, 1); 1055 vtblk_bio_done(sc, bp, bp->bio_error); 1056 } 1057 } 1058 1059 static void 1060 vtblk_drain_vq(struct vtblk_softc *sc) 1061 { 1062 struct virtqueue *vq; 1063 struct vtblk_request *req; 1064 int last; 1065 1066 vq = sc->vtblk_vq; 1067 last = 0; 1068 1069 while ((req = virtqueue_drain(vq, &last)) != NULL) { 1070 vtblk_bio_done(sc, req->vbr_bp, ENXIO); 1071 vtblk_request_enqueue(sc, req); 1072 } 1073 1074 sc->vtblk_req_ordered = NULL; 1075 KASSERT(virtqueue_empty(vq), ("virtqueue not empty")); 1076 } 1077 1078 static void 1079 vtblk_drain(struct vtblk_softc *sc) 1080 { 1081 struct bio_queue_head *bioq; 1082 struct vtblk_request *req; 1083 struct bio *bp; 1084 1085 bioq = &sc->vtblk_bioq; 1086 1087 if (sc->vtblk_vq != NULL) { 1088 struct bio_queue queue; 1089 1090 TAILQ_INIT(&queue); 1091 vtblk_queue_completed(sc, &queue); 1092 vtblk_done_completed(sc, &queue); 1093 1094 vtblk_drain_vq(sc); 1095 } 1096 1097 while ((req = vtblk_request_next_ready(sc)) != NULL) { 1098 vtblk_bio_done(sc, req->vbr_bp, ENXIO); 1099 vtblk_request_enqueue(sc, req); 1100 } 1101 1102 while (bioq_first(bioq) != NULL) { 1103 bp = bioq_takefirst(bioq); 1104 vtblk_bio_done(sc, bp, ENXIO); 1105 } 1106 1107 vtblk_request_free(sc); 1108 } 1109 1110 static void 1111 vtblk_startio(struct vtblk_softc *sc) 1112 { 1113 struct virtqueue *vq; 1114 struct vtblk_request *req; 1115 int enq; 1116 1117 VTBLK_LOCK_ASSERT(sc); 1118 vq = sc->vtblk_vq; 1119 enq = 0; 1120 1121 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) 1122 return; 1123 1124 while (!virtqueue_full(vq)) { 1125 req = vtblk_request_next(sc); 1126 if (req == NULL) 1127 break; 1128 1129 if (vtblk_request_execute(sc, req) != 0) { 1130 vtblk_request_requeue_ready(sc, req); 1131 break; 1132 } 1133 1134 enq++; 1135 } 1136 1137 if (enq > 0) 1138 virtqueue_notify(vq); 1139 } 1140 1141 static void 1142 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error) 1143 { 1144 1145 /* Because of GEOM direct dispatch, we cannot hold any locks. */ 1146 if (sc != NULL) 1147 VTBLK_LOCK_ASSERT_NOTOWNED(sc); 1148 1149 if (error) { 1150 bp->bio_resid = bp->bio_bcount; 1151 bp->bio_error = error; 1152 bp->bio_flags |= BIO_ERROR; 1153 } else { 1154 kmsan_mark_bio(bp, KMSAN_STATE_INITED); 1155 } 1156 1157 if (bp->bio_driver1 != NULL) { 1158 free(bp->bio_driver1, M_DEVBUF); 1159 bp->bio_driver1 = NULL; 1160 } 1161 1162 biodone(bp); 1163 } 1164 1165 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \ 1166 if (virtio_with_feature(_dev, _feature)) { \ 1167 virtio_read_device_config(_dev, \ 1168 offsetof(struct virtio_blk_config, _field), \ 1169 &(_cfg)->_field, sizeof((_cfg)->_field)); \ 1170 } 1171 1172 static void 1173 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) 1174 { 1175 device_t dev; 1176 1177 dev = sc->vtblk_dev; 1178 1179 bzero(blkcfg, sizeof(struct virtio_blk_config)); 1180 1181 /* The capacity is always available. */ 1182 virtio_read_device_config(dev, offsetof(struct virtio_blk_config, 1183 capacity), &blkcfg->capacity, sizeof(blkcfg->capacity)); 1184 1185 /* Read the configuration if the feature was negotiated. */ 1186 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg); 1187 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg); 1188 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, 1189 geometry.cylinders, blkcfg); 1190 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, 1191 geometry.heads, blkcfg); 1192 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, 1193 geometry.sectors, blkcfg); 1194 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg); 1195 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1196 topology.physical_block_exp, blkcfg); 1197 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1198 topology.alignment_offset, blkcfg); 1199 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1200 topology.min_io_size, blkcfg); 1201 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1202 topology.opt_io_size, blkcfg); 1203 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, wce, blkcfg); 1204 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_sectors, 1205 blkcfg); 1206 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_seg, blkcfg); 1207 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, discard_sector_alignment, 1208 blkcfg); 1209 } 1210 1211 #undef VTBLK_GET_CONFIG 1212 1213 static void 1214 vtblk_ident(struct vtblk_softc *sc) 1215 { 1216 struct bio buf; 1217 struct disk *dp; 1218 struct vtblk_request *req; 1219 int len, error; 1220 1221 dp = sc->vtblk_disk; 1222 len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE); 1223 1224 if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0) 1225 return; 1226 1227 req = vtblk_request_dequeue(sc); 1228 if (req == NULL) 1229 return; 1230 1231 req->vbr_ack = -1; 1232 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_GET_ID); 1233 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 1234 req->vbr_hdr.sector = 0; 1235 1236 req->vbr_bp = &buf; 1237 g_reset_bio(&buf); 1238 1239 buf.bio_cmd = BIO_READ; 1240 buf.bio_data = dp->d_ident; 1241 buf.bio_bcount = len; 1242 1243 VTBLK_LOCK(sc); 1244 error = vtblk_poll_request(sc, req); 1245 VTBLK_UNLOCK(sc); 1246 1247 vtblk_request_enqueue(sc, req); 1248 1249 if (error) { 1250 device_printf(sc->vtblk_dev, 1251 "error getting device identifier: %d\n", error); 1252 } 1253 } 1254 1255 static int 1256 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req) 1257 { 1258 struct virtqueue *vq; 1259 int error; 1260 1261 vq = sc->vtblk_vq; 1262 1263 if (!virtqueue_empty(vq)) 1264 return (EBUSY); 1265 1266 error = vtblk_request_execute(sc, req); 1267 if (error) 1268 return (error); 1269 1270 virtqueue_notify(vq); 1271 virtqueue_poll(vq, NULL); 1272 1273 error = vtblk_request_error(req); 1274 if (error && bootverbose) { 1275 device_printf(sc->vtblk_dev, 1276 "%s: IO error: %d\n", __func__, error); 1277 } 1278 1279 return (error); 1280 } 1281 1282 static int 1283 vtblk_quiesce(struct vtblk_softc *sc) 1284 { 1285 int error; 1286 1287 VTBLK_LOCK_ASSERT(sc); 1288 error = 0; 1289 1290 while (!virtqueue_empty(sc->vtblk_vq)) { 1291 if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq", 1292 VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) { 1293 error = EBUSY; 1294 break; 1295 } 1296 } 1297 1298 return (error); 1299 } 1300 1301 static void 1302 vtblk_vq_intr(void *xsc) 1303 { 1304 struct vtblk_softc *sc; 1305 struct virtqueue *vq; 1306 struct bio_queue queue; 1307 1308 sc = xsc; 1309 vq = sc->vtblk_vq; 1310 TAILQ_INIT(&queue); 1311 1312 VTBLK_LOCK(sc); 1313 1314 again: 1315 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) 1316 goto out; 1317 1318 vtblk_queue_completed(sc, &queue); 1319 vtblk_startio(sc); 1320 1321 if (virtqueue_enable_intr(vq) != 0) { 1322 virtqueue_disable_intr(vq); 1323 goto again; 1324 } 1325 1326 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) 1327 wakeup(&sc->vtblk_vq); 1328 1329 out: 1330 VTBLK_UNLOCK(sc); 1331 vtblk_done_completed(sc, &queue); 1332 } 1333 1334 static void 1335 vtblk_stop(struct vtblk_softc *sc) 1336 { 1337 1338 virtqueue_disable_intr(sc->vtblk_vq); 1339 virtio_stop(sc->vtblk_dev); 1340 } 1341 1342 static void 1343 vtblk_dump_quiesce(struct vtblk_softc *sc) 1344 { 1345 1346 /* 1347 * Spin here until all the requests in-flight at the time of the 1348 * dump are completed and queued. The queued requests will be 1349 * biodone'd once the dump is finished. 1350 */ 1351 while (!virtqueue_empty(sc->vtblk_vq)) 1352 vtblk_queue_completed(sc, &sc->vtblk_dump_queue); 1353 } 1354 1355 static int 1356 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset, 1357 size_t length) 1358 { 1359 struct bio buf; 1360 struct vtblk_request *req; 1361 1362 req = &sc->vtblk_dump_request; 1363 req->vbr_ack = -1; 1364 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT); 1365 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 1366 req->vbr_hdr.sector = vtblk_gtoh64(sc, offset / VTBLK_BSIZE); 1367 1368 req->vbr_bp = &buf; 1369 g_reset_bio(&buf); 1370 1371 buf.bio_cmd = BIO_WRITE; 1372 buf.bio_data = virtual; 1373 buf.bio_bcount = length; 1374 1375 return (vtblk_poll_request(sc, req)); 1376 } 1377 1378 static int 1379 vtblk_dump_flush(struct vtblk_softc *sc) 1380 { 1381 struct bio buf; 1382 struct vtblk_request *req; 1383 1384 req = &sc->vtblk_dump_request; 1385 req->vbr_ack = -1; 1386 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH); 1387 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 1388 req->vbr_hdr.sector = 0; 1389 1390 req->vbr_bp = &buf; 1391 g_reset_bio(&buf); 1392 1393 buf.bio_cmd = BIO_FLUSH; 1394 1395 return (vtblk_poll_request(sc, req)); 1396 } 1397 1398 static void 1399 vtblk_dump_complete(struct vtblk_softc *sc) 1400 { 1401 1402 vtblk_dump_flush(sc); 1403 1404 VTBLK_UNLOCK(sc); 1405 vtblk_done_completed(sc, &sc->vtblk_dump_queue); 1406 VTBLK_LOCK(sc); 1407 } 1408 1409 static void 1410 vtblk_set_write_cache(struct vtblk_softc *sc, int wc) 1411 { 1412 1413 /* Set either writeback (1) or writethrough (0) mode. */ 1414 virtio_write_dev_config_1(sc->vtblk_dev, 1415 offsetof(struct virtio_blk_config, wce), wc); 1416 } 1417 1418 static int 1419 vtblk_write_cache_enabled(struct vtblk_softc *sc, 1420 struct virtio_blk_config *blkcfg) 1421 { 1422 int wc; 1423 1424 if (sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) { 1425 wc = vtblk_tunable_int(sc, "writecache_mode", 1426 vtblk_writecache_mode); 1427 if (wc >= 0 && wc < VTBLK_CACHE_MAX) 1428 vtblk_set_write_cache(sc, wc); 1429 else 1430 wc = blkcfg->wce; 1431 } else 1432 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_FLUSH); 1433 1434 return (wc); 1435 } 1436 1437 static int 1438 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS) 1439 { 1440 struct vtblk_softc *sc; 1441 int wc, error; 1442 1443 sc = oidp->oid_arg1; 1444 wc = sc->vtblk_write_cache; 1445 1446 error = sysctl_handle_int(oidp, &wc, 0, req); 1447 if (error || req->newptr == NULL) 1448 return (error); 1449 if ((sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) == 0) 1450 return (EPERM); 1451 if (wc < 0 || wc >= VTBLK_CACHE_MAX) 1452 return (EINVAL); 1453 1454 VTBLK_LOCK(sc); 1455 sc->vtblk_write_cache = wc; 1456 vtblk_set_write_cache(sc, sc->vtblk_write_cache); 1457 VTBLK_UNLOCK(sc); 1458 1459 return (0); 1460 } 1461 1462 static void 1463 vtblk_setup_sysctl(struct vtblk_softc *sc) 1464 { 1465 device_t dev; 1466 struct sysctl_ctx_list *ctx; 1467 struct sysctl_oid *tree; 1468 struct sysctl_oid_list *child; 1469 1470 dev = sc->vtblk_dev; 1471 ctx = device_get_sysctl_ctx(dev); 1472 tree = device_get_sysctl_tree(dev); 1473 child = SYSCTL_CHILDREN(tree); 1474 1475 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode", 1476 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0, 1477 vtblk_write_cache_sysctl, "I", 1478 "Write cache mode (writethrough (0) or writeback (1))"); 1479 } 1480 1481 static int 1482 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def) 1483 { 1484 char path[64]; 1485 1486 snprintf(path, sizeof(path), 1487 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob); 1488 TUNABLE_INT_FETCH(path, &def); 1489 1490 return (def); 1491 } 1492