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/sglist.h> 41 #include <sys/sysctl.h> 42 #include <sys/lock.h> 43 #include <sys/mutex.h> 44 #include <sys/queue.h> 45 46 #include <geom/geom.h> 47 #include <geom/geom_disk.h> 48 49 #include <machine/bus.h> 50 #include <machine/resource.h> 51 #include <sys/bus.h> 52 #include <sys/rman.h> 53 54 #include <dev/virtio/virtio.h> 55 #include <dev/virtio/virtqueue.h> 56 #include <dev/virtio/block/virtio_blk.h> 57 58 #include "virtio_if.h" 59 60 struct vtblk_request { 61 struct virtio_blk_outhdr vbr_hdr; 62 struct bio *vbr_bp; 63 uint8_t vbr_ack; 64 TAILQ_ENTRY(vtblk_request) vbr_link; 65 }; 66 67 enum vtblk_cache_mode { 68 VTBLK_CACHE_WRITETHROUGH, 69 VTBLK_CACHE_WRITEBACK, 70 VTBLK_CACHE_MAX 71 }; 72 73 struct vtblk_softc { 74 device_t vtblk_dev; 75 struct mtx vtblk_mtx; 76 uint64_t vtblk_features; 77 uint32_t vtblk_flags; 78 #define VTBLK_FLAG_INDIRECT 0x0001 79 #define VTBLK_FLAG_DETACH 0x0002 80 #define VTBLK_FLAG_SUSPEND 0x0004 81 #define VTBLK_FLAG_BARRIER 0x0008 82 #define VTBLK_FLAG_WCE_CONFIG 0x0010 83 84 struct virtqueue *vtblk_vq; 85 struct sglist *vtblk_sglist; 86 struct disk *vtblk_disk; 87 88 struct bio_queue_head vtblk_bioq; 89 TAILQ_HEAD(, vtblk_request) 90 vtblk_req_free; 91 TAILQ_HEAD(, vtblk_request) 92 vtblk_req_ready; 93 struct vtblk_request *vtblk_req_ordered; 94 95 int vtblk_max_nsegs; 96 int vtblk_request_count; 97 enum vtblk_cache_mode vtblk_write_cache; 98 99 struct bio_queue vtblk_dump_queue; 100 struct vtblk_request vtblk_dump_request; 101 }; 102 103 static struct virtio_feature_desc vtblk_feature_desc[] = { 104 { VIRTIO_BLK_F_BARRIER, "HostBarrier" }, 105 { VIRTIO_BLK_F_SIZE_MAX, "MaxSegSize" }, 106 { VIRTIO_BLK_F_SEG_MAX, "MaxNumSegs" }, 107 { VIRTIO_BLK_F_GEOMETRY, "DiskGeometry" }, 108 { VIRTIO_BLK_F_RO, "ReadOnly" }, 109 { VIRTIO_BLK_F_BLK_SIZE, "BlockSize" }, 110 { VIRTIO_BLK_F_SCSI, "SCSICmds" }, 111 { VIRTIO_BLK_F_FLUSH, "FlushCmd" }, 112 { VIRTIO_BLK_F_TOPOLOGY, "Topology" }, 113 { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" }, 114 { VIRTIO_BLK_F_MQ, "Multiqueue" }, 115 { VIRTIO_BLK_F_DISCARD, "Discard" }, 116 { VIRTIO_BLK_F_WRITE_ZEROES, "WriteZeros" }, 117 118 { 0, NULL } 119 }; 120 121 static int vtblk_modevent(module_t, int, void *); 122 123 static int vtblk_probe(device_t); 124 static int vtblk_attach(device_t); 125 static int vtblk_detach(device_t); 126 static int vtblk_suspend(device_t); 127 static int vtblk_resume(device_t); 128 static int vtblk_shutdown(device_t); 129 static int vtblk_config_change(device_t); 130 131 static int vtblk_open(struct disk *); 132 static int vtblk_close(struct disk *); 133 static int vtblk_ioctl(struct disk *, u_long, void *, int, 134 struct thread *); 135 static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t); 136 static void vtblk_strategy(struct bio *); 137 138 static int vtblk_negotiate_features(struct vtblk_softc *); 139 static int vtblk_setup_features(struct vtblk_softc *); 140 static int vtblk_maximum_segments(struct vtblk_softc *, 141 struct virtio_blk_config *); 142 static int vtblk_alloc_virtqueue(struct vtblk_softc *); 143 static void vtblk_resize_disk(struct vtblk_softc *, uint64_t); 144 static void vtblk_alloc_disk(struct vtblk_softc *, 145 struct virtio_blk_config *); 146 static void vtblk_create_disk(struct vtblk_softc *); 147 148 static int vtblk_request_prealloc(struct vtblk_softc *); 149 static void vtblk_request_free(struct vtblk_softc *); 150 static struct vtblk_request * 151 vtblk_request_dequeue(struct vtblk_softc *); 152 static void vtblk_request_enqueue(struct vtblk_softc *, 153 struct vtblk_request *); 154 static struct vtblk_request * 155 vtblk_request_next_ready(struct vtblk_softc *); 156 static void vtblk_request_requeue_ready(struct vtblk_softc *, 157 struct vtblk_request *); 158 static struct vtblk_request * 159 vtblk_request_next(struct vtblk_softc *); 160 static struct vtblk_request * 161 vtblk_request_bio(struct vtblk_softc *); 162 static int vtblk_request_execute(struct vtblk_softc *, 163 struct vtblk_request *); 164 static int vtblk_request_error(struct vtblk_request *); 165 166 static void vtblk_queue_completed(struct vtblk_softc *, 167 struct bio_queue *); 168 static void vtblk_done_completed(struct vtblk_softc *, 169 struct bio_queue *); 170 static void vtblk_drain_vq(struct vtblk_softc *); 171 static void vtblk_drain(struct vtblk_softc *); 172 173 static void vtblk_startio(struct vtblk_softc *); 174 static void vtblk_bio_done(struct vtblk_softc *, struct bio *, int); 175 176 static void vtblk_read_config(struct vtblk_softc *, 177 struct virtio_blk_config *); 178 static void vtblk_ident(struct vtblk_softc *); 179 static int vtblk_poll_request(struct vtblk_softc *, 180 struct vtblk_request *); 181 static int vtblk_quiesce(struct vtblk_softc *); 182 static void vtblk_vq_intr(void *); 183 static void vtblk_stop(struct vtblk_softc *); 184 185 static void vtblk_dump_quiesce(struct vtblk_softc *); 186 static int vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t); 187 static int vtblk_dump_flush(struct vtblk_softc *); 188 static void vtblk_dump_complete(struct vtblk_softc *); 189 190 static void vtblk_set_write_cache(struct vtblk_softc *, int); 191 static int vtblk_write_cache_enabled(struct vtblk_softc *sc, 192 struct virtio_blk_config *); 193 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS); 194 195 static void vtblk_setup_sysctl(struct vtblk_softc *); 196 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int); 197 198 #define vtblk_modern(_sc) (((_sc)->vtblk_features & VIRTIO_F_VERSION_1) != 0) 199 #define vtblk_htog16(_sc, _val) virtio_htog16(vtblk_modern(_sc), _val) 200 #define vtblk_htog32(_sc, _val) virtio_htog32(vtblk_modern(_sc), _val) 201 #define vtblk_htog64(_sc, _val) virtio_htog64(vtblk_modern(_sc), _val) 202 #define vtblk_gtoh16(_sc, _val) virtio_gtoh16(vtblk_modern(_sc), _val) 203 #define vtblk_gtoh32(_sc, _val) virtio_gtoh32(vtblk_modern(_sc), _val) 204 #define vtblk_gtoh64(_sc, _val) virtio_gtoh64(vtblk_modern(_sc), _val) 205 206 /* Tunables. */ 207 static int vtblk_no_ident = 0; 208 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident); 209 static int vtblk_writecache_mode = -1; 210 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode); 211 212 #define VTBLK_COMMON_FEATURES \ 213 (VIRTIO_BLK_F_SIZE_MAX | \ 214 VIRTIO_BLK_F_SEG_MAX | \ 215 VIRTIO_BLK_F_GEOMETRY | \ 216 VIRTIO_BLK_F_RO | \ 217 VIRTIO_BLK_F_BLK_SIZE | \ 218 VIRTIO_BLK_F_FLUSH | \ 219 VIRTIO_BLK_F_TOPOLOGY | \ 220 VIRTIO_BLK_F_CONFIG_WCE | \ 221 VIRTIO_BLK_F_DISCARD | \ 222 VIRTIO_RING_F_INDIRECT_DESC) 223 224 #define VTBLK_MODERN_FEATURES (VTBLK_COMMON_FEATURES) 225 #define VTBLK_LEGACY_FEATURES (VIRTIO_BLK_F_BARRIER | VTBLK_COMMON_FEATURES) 226 227 #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx 228 #define VTBLK_LOCK_INIT(_sc, _name) \ 229 mtx_init(VTBLK_MTX((_sc)), (_name), \ 230 "VirtIO Block Lock", MTX_DEF) 231 #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc))) 232 #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc))) 233 #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc))) 234 #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED) 235 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \ 236 mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED) 237 238 #define VTBLK_DISK_NAME "vtbd" 239 #define VTBLK_QUIESCE_TIMEOUT (30 * hz) 240 #define VTBLK_BSIZE 512 241 242 /* 243 * Each block request uses at least two segments - one for the header 244 * and one for the status. 245 */ 246 #define VTBLK_MIN_SEGMENTS 2 247 248 static device_method_t vtblk_methods[] = { 249 /* Device methods. */ 250 DEVMETHOD(device_probe, vtblk_probe), 251 DEVMETHOD(device_attach, vtblk_attach), 252 DEVMETHOD(device_detach, vtblk_detach), 253 DEVMETHOD(device_suspend, vtblk_suspend), 254 DEVMETHOD(device_resume, vtblk_resume), 255 DEVMETHOD(device_shutdown, vtblk_shutdown), 256 257 /* VirtIO methods. */ 258 DEVMETHOD(virtio_config_change, vtblk_config_change), 259 260 DEVMETHOD_END 261 }; 262 263 static driver_t vtblk_driver = { 264 "vtblk", 265 vtblk_methods, 266 sizeof(struct vtblk_softc) 267 }; 268 static devclass_t vtblk_devclass; 269 270 VIRTIO_DRIVER_MODULE(virtio_blk, vtblk_driver, vtblk_devclass, 271 vtblk_modevent, 0); 272 MODULE_VERSION(virtio_blk, 1); 273 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1); 274 275 VIRTIO_SIMPLE_PNPINFO(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter"); 276 277 static int 278 vtblk_modevent(module_t mod, int type, void *unused) 279 { 280 int error; 281 282 error = 0; 283 284 switch (type) { 285 case MOD_LOAD: 286 case MOD_QUIESCE: 287 case MOD_UNLOAD: 288 case MOD_SHUTDOWN: 289 break; 290 default: 291 error = EOPNOTSUPP; 292 break; 293 } 294 295 return (error); 296 } 297 298 static int 299 vtblk_probe(device_t dev) 300 { 301 return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk)); 302 } 303 304 static int 305 vtblk_attach(device_t dev) 306 { 307 struct vtblk_softc *sc; 308 struct virtio_blk_config blkcfg; 309 int error; 310 311 sc = device_get_softc(dev); 312 sc->vtblk_dev = dev; 313 virtio_set_feature_desc(dev, vtblk_feature_desc); 314 315 VTBLK_LOCK_INIT(sc, device_get_nameunit(dev)); 316 bioq_init(&sc->vtblk_bioq); 317 TAILQ_INIT(&sc->vtblk_dump_queue); 318 TAILQ_INIT(&sc->vtblk_req_free); 319 TAILQ_INIT(&sc->vtblk_req_ready); 320 321 vtblk_setup_sysctl(sc); 322 323 error = vtblk_setup_features(sc); 324 if (error) { 325 device_printf(dev, "cannot setup features\n"); 326 goto fail; 327 } 328 329 vtblk_read_config(sc, &blkcfg); 330 331 /* 332 * With the current sglist(9) implementation, it is not easy 333 * for us to support a maximum segment size as adjacent 334 * segments are coalesced. For now, just make sure it's larger 335 * than the maximum supported transfer size. 336 */ 337 if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) { 338 if (blkcfg.size_max < maxphys) { 339 error = ENOTSUP; 340 device_printf(dev, "host requires unsupported " 341 "maximum segment size feature\n"); 342 goto fail; 343 } 344 } 345 346 sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg); 347 if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) { 348 error = EINVAL; 349 device_printf(dev, "fewer than minimum number of segments " 350 "allowed: %d\n", sc->vtblk_max_nsegs); 351 goto fail; 352 } 353 354 sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT); 355 if (sc->vtblk_sglist == NULL) { 356 error = ENOMEM; 357 device_printf(dev, "cannot allocate sglist\n"); 358 goto fail; 359 } 360 361 error = vtblk_alloc_virtqueue(sc); 362 if (error) { 363 device_printf(dev, "cannot allocate virtqueue\n"); 364 goto fail; 365 } 366 367 error = vtblk_request_prealloc(sc); 368 if (error) { 369 device_printf(dev, "cannot preallocate requests\n"); 370 goto fail; 371 } 372 373 vtblk_alloc_disk(sc, &blkcfg); 374 375 error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY); 376 if (error) { 377 device_printf(dev, "cannot setup virtqueue interrupt\n"); 378 goto fail; 379 } 380 381 vtblk_create_disk(sc); 382 383 virtqueue_enable_intr(sc->vtblk_vq); 384 385 fail: 386 if (error) 387 vtblk_detach(dev); 388 389 return (error); 390 } 391 392 static int 393 vtblk_detach(device_t dev) 394 { 395 struct vtblk_softc *sc; 396 397 sc = device_get_softc(dev); 398 399 VTBLK_LOCK(sc); 400 sc->vtblk_flags |= VTBLK_FLAG_DETACH; 401 if (device_is_attached(dev)) 402 vtblk_stop(sc); 403 VTBLK_UNLOCK(sc); 404 405 vtblk_drain(sc); 406 407 if (sc->vtblk_disk != NULL) { 408 disk_destroy(sc->vtblk_disk); 409 sc->vtblk_disk = NULL; 410 } 411 412 if (sc->vtblk_sglist != NULL) { 413 sglist_free(sc->vtblk_sglist); 414 sc->vtblk_sglist = NULL; 415 } 416 417 VTBLK_LOCK_DESTROY(sc); 418 419 return (0); 420 } 421 422 static int 423 vtblk_suspend(device_t dev) 424 { 425 struct vtblk_softc *sc; 426 int error; 427 428 sc = device_get_softc(dev); 429 430 VTBLK_LOCK(sc); 431 sc->vtblk_flags |= VTBLK_FLAG_SUSPEND; 432 /* XXX BMV: virtio_stop(), etc needed here? */ 433 error = vtblk_quiesce(sc); 434 if (error) 435 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; 436 VTBLK_UNLOCK(sc); 437 438 return (error); 439 } 440 441 static int 442 vtblk_resume(device_t dev) 443 { 444 struct vtblk_softc *sc; 445 446 sc = device_get_softc(dev); 447 448 VTBLK_LOCK(sc); 449 /* XXX BMV: virtio_reinit(), etc needed here? */ 450 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; 451 vtblk_startio(sc); 452 VTBLK_UNLOCK(sc); 453 454 return (0); 455 } 456 457 static int 458 vtblk_shutdown(device_t dev) 459 { 460 461 return (0); 462 } 463 464 static int 465 vtblk_config_change(device_t dev) 466 { 467 struct vtblk_softc *sc; 468 struct virtio_blk_config blkcfg; 469 uint64_t capacity; 470 471 sc = device_get_softc(dev); 472 473 vtblk_read_config(sc, &blkcfg); 474 475 /* Capacity is always in 512-byte units. */ 476 capacity = blkcfg.capacity * VTBLK_BSIZE; 477 478 if (sc->vtblk_disk->d_mediasize != capacity) 479 vtblk_resize_disk(sc, capacity); 480 481 return (0); 482 } 483 484 static int 485 vtblk_open(struct disk *dp) 486 { 487 struct vtblk_softc *sc; 488 489 if ((sc = dp->d_drv1) == NULL) 490 return (ENXIO); 491 492 return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0); 493 } 494 495 static int 496 vtblk_close(struct disk *dp) 497 { 498 struct vtblk_softc *sc; 499 500 if ((sc = dp->d_drv1) == NULL) 501 return (ENXIO); 502 503 return (0); 504 } 505 506 static int 507 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, 508 struct thread *td) 509 { 510 struct vtblk_softc *sc; 511 512 if ((sc = dp->d_drv1) == NULL) 513 return (ENXIO); 514 515 return (ENOTTY); 516 } 517 518 static int 519 vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset, 520 size_t length) 521 { 522 struct disk *dp; 523 struct vtblk_softc *sc; 524 int error; 525 526 dp = arg; 527 error = 0; 528 529 if ((sc = dp->d_drv1) == NULL) 530 return (ENXIO); 531 532 VTBLK_LOCK(sc); 533 534 vtblk_dump_quiesce(sc); 535 536 if (length > 0) 537 error = vtblk_dump_write(sc, virtual, offset, length); 538 if (error || (virtual == NULL && offset == 0)) 539 vtblk_dump_complete(sc); 540 541 VTBLK_UNLOCK(sc); 542 543 return (error); 544 } 545 546 static void 547 vtblk_strategy(struct bio *bp) 548 { 549 struct vtblk_softc *sc; 550 551 if ((sc = bp->bio_disk->d_drv1) == NULL) { 552 vtblk_bio_done(NULL, bp, EINVAL); 553 return; 554 } 555 556 if ((bp->bio_cmd != BIO_READ) && (bp->bio_cmd != BIO_WRITE) && 557 (bp->bio_cmd != BIO_FLUSH) && (bp->bio_cmd != BIO_DELETE)) { 558 vtblk_bio_done(sc, bp, EOPNOTSUPP); 559 return; 560 } 561 562 VTBLK_LOCK(sc); 563 564 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) { 565 VTBLK_UNLOCK(sc); 566 vtblk_bio_done(sc, bp, ENXIO); 567 return; 568 } 569 570 bioq_insert_tail(&sc->vtblk_bioq, bp); 571 vtblk_startio(sc); 572 573 VTBLK_UNLOCK(sc); 574 } 575 576 static int 577 vtblk_negotiate_features(struct vtblk_softc *sc) 578 { 579 device_t dev; 580 uint64_t features; 581 582 dev = sc->vtblk_dev; 583 features = virtio_bus_is_modern(dev) ? VTBLK_MODERN_FEATURES : 584 VTBLK_LEGACY_FEATURES; 585 586 sc->vtblk_features = virtio_negotiate_features(dev, features); 587 return (virtio_finalize_features(dev)); 588 } 589 590 static int 591 vtblk_setup_features(struct vtblk_softc *sc) 592 { 593 device_t dev; 594 int error; 595 596 dev = sc->vtblk_dev; 597 598 error = vtblk_negotiate_features(sc); 599 if (error) 600 return (error); 601 602 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 603 sc->vtblk_flags |= VTBLK_FLAG_INDIRECT; 604 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE)) 605 sc->vtblk_flags |= VTBLK_FLAG_WCE_CONFIG; 606 607 /* Legacy. */ 608 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER)) 609 sc->vtblk_flags |= VTBLK_FLAG_BARRIER; 610 611 return (0); 612 } 613 614 static int 615 vtblk_maximum_segments(struct vtblk_softc *sc, 616 struct virtio_blk_config *blkcfg) 617 { 618 device_t dev; 619 int nsegs; 620 621 dev = sc->vtblk_dev; 622 nsegs = VTBLK_MIN_SEGMENTS; 623 624 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) { 625 nsegs += MIN(blkcfg->seg_max, maxphys / PAGE_SIZE + 1); 626 if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT) 627 nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT); 628 } else 629 nsegs += 1; 630 631 return (nsegs); 632 } 633 634 static int 635 vtblk_alloc_virtqueue(struct vtblk_softc *sc) 636 { 637 device_t dev; 638 struct vq_alloc_info vq_info; 639 640 dev = sc->vtblk_dev; 641 642 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs, 643 vtblk_vq_intr, sc, &sc->vtblk_vq, 644 "%s request", device_get_nameunit(dev)); 645 646 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info)); 647 } 648 649 static void 650 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity) 651 { 652 device_t dev; 653 struct disk *dp; 654 int error; 655 656 dev = sc->vtblk_dev; 657 dp = sc->vtblk_disk; 658 659 dp->d_mediasize = new_capacity; 660 if (bootverbose) { 661 device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n", 662 (uintmax_t) dp->d_mediasize >> 20, 663 (uintmax_t) dp->d_mediasize / dp->d_sectorsize, 664 dp->d_sectorsize); 665 } 666 667 error = disk_resize(dp, M_NOWAIT); 668 if (error) { 669 device_printf(dev, 670 "disk_resize(9) failed, error: %d\n", error); 671 } 672 } 673 674 static void 675 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) 676 { 677 device_t dev; 678 struct disk *dp; 679 680 dev = sc->vtblk_dev; 681 682 sc->vtblk_disk = dp = disk_alloc(); 683 dp->d_open = vtblk_open; 684 dp->d_close = vtblk_close; 685 dp->d_ioctl = vtblk_ioctl; 686 dp->d_strategy = vtblk_strategy; 687 dp->d_name = VTBLK_DISK_NAME; 688 dp->d_unit = device_get_unit(dev); 689 dp->d_drv1 = sc; 690 dp->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION; 691 dp->d_hba_vendor = virtio_get_vendor(dev); 692 dp->d_hba_device = virtio_get_device(dev); 693 dp->d_hba_subvendor = virtio_get_subvendor(dev); 694 dp->d_hba_subdevice = virtio_get_subdevice(dev); 695 696 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO)) 697 dp->d_flags |= DISKFLAG_WRITE_PROTECT; 698 else { 699 if (virtio_with_feature(dev, VIRTIO_BLK_F_FLUSH)) 700 dp->d_flags |= DISKFLAG_CANFLUSHCACHE; 701 dp->d_dump = vtblk_dump; 702 } 703 704 /* Capacity is always in 512-byte units. */ 705 dp->d_mediasize = blkcfg->capacity * VTBLK_BSIZE; 706 707 if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE)) 708 dp->d_sectorsize = blkcfg->blk_size; 709 else 710 dp->d_sectorsize = VTBLK_BSIZE; 711 712 /* 713 * The VirtIO maximum I/O size is given in terms of segments. 714 * However, FreeBSD limits I/O size by logical buffer size, not 715 * by physically contiguous pages. Therefore, we have to assume 716 * no pages are contiguous. This may impose an artificially low 717 * maximum I/O size. But in practice, since QEMU advertises 128 718 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE, 719 * which is typically greater than maxphys. Eventually we should 720 * just advertise maxphys and split buffers that are too big. 721 * 722 * Note we must subtract one additional segment in case of non 723 * page aligned buffers. 724 */ 725 dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) * 726 PAGE_SIZE; 727 if (dp->d_maxsize < PAGE_SIZE) 728 dp->d_maxsize = PAGE_SIZE; /* XXX */ 729 730 if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) { 731 dp->d_fwsectors = blkcfg->geometry.sectors; 732 dp->d_fwheads = blkcfg->geometry.heads; 733 } 734 735 if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) && 736 blkcfg->topology.physical_block_exp > 0) { 737 dp->d_stripesize = dp->d_sectorsize * 738 (1 << blkcfg->topology.physical_block_exp); 739 dp->d_stripeoffset = (dp->d_stripesize - 740 blkcfg->topology.alignment_offset * dp->d_sectorsize) % 741 dp->d_stripesize; 742 } 743 744 if (virtio_with_feature(dev, VIRTIO_BLK_F_DISCARD)) { 745 dp->d_flags |= DISKFLAG_CANDELETE; 746 dp->d_delmaxsize = blkcfg->max_discard_sectors * VTBLK_BSIZE; 747 } 748 749 if (vtblk_write_cache_enabled(sc, blkcfg) != 0) 750 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK; 751 else 752 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH; 753 } 754 755 static void 756 vtblk_create_disk(struct vtblk_softc *sc) 757 { 758 struct disk *dp; 759 760 dp = sc->vtblk_disk; 761 762 vtblk_ident(sc); 763 764 device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n", 765 (uintmax_t) dp->d_mediasize >> 20, 766 (uintmax_t) dp->d_mediasize / dp->d_sectorsize, 767 dp->d_sectorsize); 768 769 disk_create(dp, DISK_VERSION); 770 } 771 772 static int 773 vtblk_request_prealloc(struct vtblk_softc *sc) 774 { 775 struct vtblk_request *req; 776 int i, nreqs; 777 778 nreqs = virtqueue_size(sc->vtblk_vq); 779 780 /* 781 * Preallocate sufficient requests to keep the virtqueue full. Each 782 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce 783 * the number allocated when indirect descriptors are not available. 784 */ 785 if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0) 786 nreqs /= VTBLK_MIN_SEGMENTS; 787 788 for (i = 0; i < nreqs; i++) { 789 req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT); 790 if (req == NULL) 791 return (ENOMEM); 792 793 MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1); 794 MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1); 795 796 sc->vtblk_request_count++; 797 vtblk_request_enqueue(sc, req); 798 } 799 800 return (0); 801 } 802 803 static void 804 vtblk_request_free(struct vtblk_softc *sc) 805 { 806 struct vtblk_request *req; 807 808 MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready)); 809 810 while ((req = vtblk_request_dequeue(sc)) != NULL) { 811 sc->vtblk_request_count--; 812 free(req, M_DEVBUF); 813 } 814 815 KASSERT(sc->vtblk_request_count == 0, 816 ("%s: leaked %d requests", __func__, sc->vtblk_request_count)); 817 } 818 819 static struct vtblk_request * 820 vtblk_request_dequeue(struct vtblk_softc *sc) 821 { 822 struct vtblk_request *req; 823 824 req = TAILQ_FIRST(&sc->vtblk_req_free); 825 if (req != NULL) { 826 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link); 827 bzero(req, sizeof(struct vtblk_request)); 828 } 829 830 return (req); 831 } 832 833 static void 834 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req) 835 { 836 837 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link); 838 } 839 840 static struct vtblk_request * 841 vtblk_request_next_ready(struct vtblk_softc *sc) 842 { 843 struct vtblk_request *req; 844 845 req = TAILQ_FIRST(&sc->vtblk_req_ready); 846 if (req != NULL) 847 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link); 848 849 return (req); 850 } 851 852 static void 853 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req) 854 { 855 856 /* NOTE: Currently, there will be at most one request in the queue. */ 857 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link); 858 } 859 860 static struct vtblk_request * 861 vtblk_request_next(struct vtblk_softc *sc) 862 { 863 struct vtblk_request *req; 864 865 req = vtblk_request_next_ready(sc); 866 if (req != NULL) 867 return (req); 868 869 return (vtblk_request_bio(sc)); 870 } 871 872 static struct vtblk_request * 873 vtblk_request_bio(struct vtblk_softc *sc) 874 { 875 struct bio_queue_head *bioq; 876 struct vtblk_request *req; 877 struct bio *bp; 878 879 bioq = &sc->vtblk_bioq; 880 881 if (bioq_first(bioq) == NULL) 882 return (NULL); 883 884 req = vtblk_request_dequeue(sc); 885 if (req == NULL) 886 return (NULL); 887 888 bp = bioq_takefirst(bioq); 889 req->vbr_bp = bp; 890 req->vbr_ack = -1; 891 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 892 893 switch (bp->bio_cmd) { 894 case BIO_FLUSH: 895 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH); 896 req->vbr_hdr.sector = 0; 897 break; 898 case BIO_READ: 899 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_IN); 900 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 901 break; 902 case BIO_WRITE: 903 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT); 904 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 905 break; 906 case BIO_DELETE: 907 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_DISCARD); 908 req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 909 break; 910 default: 911 panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd); 912 } 913 914 if (bp->bio_flags & BIO_ORDERED) 915 req->vbr_hdr.type |= vtblk_gtoh32(sc, VIRTIO_BLK_T_BARRIER); 916 917 return (req); 918 } 919 920 static int 921 vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req) 922 { 923 struct virtqueue *vq; 924 struct sglist *sg; 925 struct bio *bp; 926 int ordered, readable, writable, error; 927 928 vq = sc->vtblk_vq; 929 sg = sc->vtblk_sglist; 930 bp = req->vbr_bp; 931 ordered = 0; 932 writable = 0; 933 934 /* 935 * Some hosts (such as bhyve) do not implement the barrier feature, 936 * so we emulate it in the driver by allowing the barrier request 937 * to be the only one in flight. 938 */ 939 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) { 940 if (sc->vtblk_req_ordered != NULL) 941 return (EBUSY); 942 if (bp->bio_flags & BIO_ORDERED) { 943 if (!virtqueue_empty(vq)) 944 return (EBUSY); 945 ordered = 1; 946 req->vbr_hdr.type &= vtblk_gtoh32(sc, 947 ~VIRTIO_BLK_T_BARRIER); 948 } 949 } 950 951 sglist_reset(sg); 952 sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr)); 953 954 if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) { 955 error = sglist_append_bio(sg, bp); 956 if (error || sg->sg_nseg == sg->sg_maxseg) { 957 panic("%s: bio %p data buffer too big %d", 958 __func__, bp, error); 959 } 960 961 /* BIO_READ means the host writes into our buffer. */ 962 if (bp->bio_cmd == BIO_READ) 963 writable = sg->sg_nseg - 1; 964 } else if (bp->bio_cmd == BIO_DELETE) { 965 struct virtio_blk_discard_write_zeroes *discard; 966 967 discard = malloc(sizeof(*discard), M_DEVBUF, M_NOWAIT | M_ZERO); 968 if (discard == NULL) 969 return (ENOMEM); 970 971 bp->bio_driver1 = discard; 972 discard->sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE); 973 discard->num_sectors = vtblk_gtoh32(sc, bp->bio_bcount / VTBLK_BSIZE); 974 error = sglist_append(sg, discard, sizeof(*discard)); 975 if (error || sg->sg_nseg == sg->sg_maxseg) { 976 panic("%s: bio %p data buffer too big %d", 977 __func__, bp, error); 978 } 979 } 980 981 writable++; 982 sglist_append(sg, &req->vbr_ack, sizeof(uint8_t)); 983 readable = sg->sg_nseg - writable; 984 985 error = virtqueue_enqueue(vq, req, sg, readable, writable); 986 if (error == 0 && ordered) 987 sc->vtblk_req_ordered = req; 988 989 return (error); 990 } 991 992 static int 993 vtblk_request_error(struct vtblk_request *req) 994 { 995 int error; 996 997 switch (req->vbr_ack) { 998 case VIRTIO_BLK_S_OK: 999 error = 0; 1000 break; 1001 case VIRTIO_BLK_S_UNSUPP: 1002 error = ENOTSUP; 1003 break; 1004 default: 1005 error = EIO; 1006 break; 1007 } 1008 1009 return (error); 1010 } 1011 1012 static void 1013 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue) 1014 { 1015 struct vtblk_request *req; 1016 struct bio *bp; 1017 1018 while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) { 1019 if (sc->vtblk_req_ordered != NULL) { 1020 MPASS(sc->vtblk_req_ordered == req); 1021 sc->vtblk_req_ordered = NULL; 1022 } 1023 1024 bp = req->vbr_bp; 1025 bp->bio_error = vtblk_request_error(req); 1026 TAILQ_INSERT_TAIL(queue, bp, bio_queue); 1027 1028 vtblk_request_enqueue(sc, req); 1029 } 1030 } 1031 1032 static void 1033 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue) 1034 { 1035 struct bio *bp, *tmp; 1036 1037 TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) { 1038 if (bp->bio_error != 0) 1039 disk_err(bp, "hard error", -1, 1); 1040 vtblk_bio_done(sc, bp, bp->bio_error); 1041 } 1042 } 1043 1044 static void 1045 vtblk_drain_vq(struct vtblk_softc *sc) 1046 { 1047 struct virtqueue *vq; 1048 struct vtblk_request *req; 1049 int last; 1050 1051 vq = sc->vtblk_vq; 1052 last = 0; 1053 1054 while ((req = virtqueue_drain(vq, &last)) != NULL) { 1055 vtblk_bio_done(sc, req->vbr_bp, ENXIO); 1056 vtblk_request_enqueue(sc, req); 1057 } 1058 1059 sc->vtblk_req_ordered = NULL; 1060 KASSERT(virtqueue_empty(vq), ("virtqueue not empty")); 1061 } 1062 1063 static void 1064 vtblk_drain(struct vtblk_softc *sc) 1065 { 1066 struct bio_queue_head *bioq; 1067 struct vtblk_request *req; 1068 struct bio *bp; 1069 1070 bioq = &sc->vtblk_bioq; 1071 1072 if (sc->vtblk_vq != NULL) { 1073 struct bio_queue queue; 1074 1075 TAILQ_INIT(&queue); 1076 vtblk_queue_completed(sc, &queue); 1077 vtblk_done_completed(sc, &queue); 1078 1079 vtblk_drain_vq(sc); 1080 } 1081 1082 while ((req = vtblk_request_next_ready(sc)) != NULL) { 1083 vtblk_bio_done(sc, req->vbr_bp, ENXIO); 1084 vtblk_request_enqueue(sc, req); 1085 } 1086 1087 while (bioq_first(bioq) != NULL) { 1088 bp = bioq_takefirst(bioq); 1089 vtblk_bio_done(sc, bp, ENXIO); 1090 } 1091 1092 vtblk_request_free(sc); 1093 } 1094 1095 static void 1096 vtblk_startio(struct vtblk_softc *sc) 1097 { 1098 struct virtqueue *vq; 1099 struct vtblk_request *req; 1100 int enq; 1101 1102 VTBLK_LOCK_ASSERT(sc); 1103 vq = sc->vtblk_vq; 1104 enq = 0; 1105 1106 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) 1107 return; 1108 1109 while (!virtqueue_full(vq)) { 1110 req = vtblk_request_next(sc); 1111 if (req == NULL) 1112 break; 1113 1114 if (vtblk_request_execute(sc, req) != 0) { 1115 vtblk_request_requeue_ready(sc, req); 1116 break; 1117 } 1118 1119 enq++; 1120 } 1121 1122 if (enq > 0) 1123 virtqueue_notify(vq); 1124 } 1125 1126 static void 1127 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error) 1128 { 1129 1130 /* Because of GEOM direct dispatch, we cannot hold any locks. */ 1131 if (sc != NULL) 1132 VTBLK_LOCK_ASSERT_NOTOWNED(sc); 1133 1134 if (error) { 1135 bp->bio_resid = bp->bio_bcount; 1136 bp->bio_error = error; 1137 bp->bio_flags |= BIO_ERROR; 1138 } 1139 1140 if (bp->bio_driver1 != NULL) { 1141 free(bp->bio_driver1, M_DEVBUF); 1142 bp->bio_driver1 = NULL; 1143 } 1144 1145 biodone(bp); 1146 } 1147 1148 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \ 1149 if (virtio_with_feature(_dev, _feature)) { \ 1150 virtio_read_device_config(_dev, \ 1151 offsetof(struct virtio_blk_config, _field), \ 1152 &(_cfg)->_field, sizeof((_cfg)->_field)); \ 1153 } 1154 1155 static void 1156 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) 1157 { 1158 device_t dev; 1159 1160 dev = sc->vtblk_dev; 1161 1162 bzero(blkcfg, sizeof(struct virtio_blk_config)); 1163 1164 /* The capacity is always available. */ 1165 virtio_read_device_config(dev, offsetof(struct virtio_blk_config, 1166 capacity), &blkcfg->capacity, sizeof(blkcfg->capacity)); 1167 1168 /* Read the configuration if the feature was negotiated. */ 1169 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg); 1170 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg); 1171 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, 1172 geometry.cylinders, blkcfg); 1173 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, 1174 geometry.heads, blkcfg); 1175 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, 1176 geometry.sectors, blkcfg); 1177 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg); 1178 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1179 topology.physical_block_exp, blkcfg); 1180 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1181 topology.alignment_offset, blkcfg); 1182 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1183 topology.min_io_size, blkcfg); 1184 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, 1185 topology.opt_io_size, blkcfg); 1186 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, wce, blkcfg); 1187 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_sectors, 1188 blkcfg); 1189 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_seg, blkcfg); 1190 VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, discard_sector_alignment, 1191 blkcfg); 1192 } 1193 1194 #undef VTBLK_GET_CONFIG 1195 1196 static void 1197 vtblk_ident(struct vtblk_softc *sc) 1198 { 1199 struct bio buf; 1200 struct disk *dp; 1201 struct vtblk_request *req; 1202 int len, error; 1203 1204 dp = sc->vtblk_disk; 1205 len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE); 1206 1207 if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0) 1208 return; 1209 1210 req = vtblk_request_dequeue(sc); 1211 if (req == NULL) 1212 return; 1213 1214 req->vbr_ack = -1; 1215 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_GET_ID); 1216 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 1217 req->vbr_hdr.sector = 0; 1218 1219 req->vbr_bp = &buf; 1220 g_reset_bio(&buf); 1221 1222 buf.bio_cmd = BIO_READ; 1223 buf.bio_data = dp->d_ident; 1224 buf.bio_bcount = len; 1225 1226 VTBLK_LOCK(sc); 1227 error = vtblk_poll_request(sc, req); 1228 VTBLK_UNLOCK(sc); 1229 1230 vtblk_request_enqueue(sc, req); 1231 1232 if (error) { 1233 device_printf(sc->vtblk_dev, 1234 "error getting device identifier: %d\n", error); 1235 } 1236 } 1237 1238 static int 1239 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req) 1240 { 1241 struct virtqueue *vq; 1242 int error; 1243 1244 vq = sc->vtblk_vq; 1245 1246 if (!virtqueue_empty(vq)) 1247 return (EBUSY); 1248 1249 error = vtblk_request_execute(sc, req); 1250 if (error) 1251 return (error); 1252 1253 virtqueue_notify(vq); 1254 virtqueue_poll(vq, NULL); 1255 1256 error = vtblk_request_error(req); 1257 if (error && bootverbose) { 1258 device_printf(sc->vtblk_dev, 1259 "%s: IO error: %d\n", __func__, error); 1260 } 1261 1262 return (error); 1263 } 1264 1265 static int 1266 vtblk_quiesce(struct vtblk_softc *sc) 1267 { 1268 int error; 1269 1270 VTBLK_LOCK_ASSERT(sc); 1271 error = 0; 1272 1273 while (!virtqueue_empty(sc->vtblk_vq)) { 1274 if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq", 1275 VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) { 1276 error = EBUSY; 1277 break; 1278 } 1279 } 1280 1281 return (error); 1282 } 1283 1284 static void 1285 vtblk_vq_intr(void *xsc) 1286 { 1287 struct vtblk_softc *sc; 1288 struct virtqueue *vq; 1289 struct bio_queue queue; 1290 1291 sc = xsc; 1292 vq = sc->vtblk_vq; 1293 TAILQ_INIT(&queue); 1294 1295 VTBLK_LOCK(sc); 1296 1297 again: 1298 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) 1299 goto out; 1300 1301 vtblk_queue_completed(sc, &queue); 1302 vtblk_startio(sc); 1303 1304 if (virtqueue_enable_intr(vq) != 0) { 1305 virtqueue_disable_intr(vq); 1306 goto again; 1307 } 1308 1309 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) 1310 wakeup(&sc->vtblk_vq); 1311 1312 out: 1313 VTBLK_UNLOCK(sc); 1314 vtblk_done_completed(sc, &queue); 1315 } 1316 1317 static void 1318 vtblk_stop(struct vtblk_softc *sc) 1319 { 1320 1321 virtqueue_disable_intr(sc->vtblk_vq); 1322 virtio_stop(sc->vtblk_dev); 1323 } 1324 1325 static void 1326 vtblk_dump_quiesce(struct vtblk_softc *sc) 1327 { 1328 1329 /* 1330 * Spin here until all the requests in-flight at the time of the 1331 * dump are completed and queued. The queued requests will be 1332 * biodone'd once the dump is finished. 1333 */ 1334 while (!virtqueue_empty(sc->vtblk_vq)) 1335 vtblk_queue_completed(sc, &sc->vtblk_dump_queue); 1336 } 1337 1338 static int 1339 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset, 1340 size_t length) 1341 { 1342 struct bio buf; 1343 struct vtblk_request *req; 1344 1345 req = &sc->vtblk_dump_request; 1346 req->vbr_ack = -1; 1347 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT); 1348 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 1349 req->vbr_hdr.sector = vtblk_gtoh64(sc, offset / VTBLK_BSIZE); 1350 1351 req->vbr_bp = &buf; 1352 g_reset_bio(&buf); 1353 1354 buf.bio_cmd = BIO_WRITE; 1355 buf.bio_data = virtual; 1356 buf.bio_bcount = length; 1357 1358 return (vtblk_poll_request(sc, req)); 1359 } 1360 1361 static int 1362 vtblk_dump_flush(struct vtblk_softc *sc) 1363 { 1364 struct bio buf; 1365 struct vtblk_request *req; 1366 1367 req = &sc->vtblk_dump_request; 1368 req->vbr_ack = -1; 1369 req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH); 1370 req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1); 1371 req->vbr_hdr.sector = 0; 1372 1373 req->vbr_bp = &buf; 1374 g_reset_bio(&buf); 1375 1376 buf.bio_cmd = BIO_FLUSH; 1377 1378 return (vtblk_poll_request(sc, req)); 1379 } 1380 1381 static void 1382 vtblk_dump_complete(struct vtblk_softc *sc) 1383 { 1384 1385 vtblk_dump_flush(sc); 1386 1387 VTBLK_UNLOCK(sc); 1388 vtblk_done_completed(sc, &sc->vtblk_dump_queue); 1389 VTBLK_LOCK(sc); 1390 } 1391 1392 static void 1393 vtblk_set_write_cache(struct vtblk_softc *sc, int wc) 1394 { 1395 1396 /* Set either writeback (1) or writethrough (0) mode. */ 1397 virtio_write_dev_config_1(sc->vtblk_dev, 1398 offsetof(struct virtio_blk_config, wce), wc); 1399 } 1400 1401 static int 1402 vtblk_write_cache_enabled(struct vtblk_softc *sc, 1403 struct virtio_blk_config *blkcfg) 1404 { 1405 int wc; 1406 1407 if (sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) { 1408 wc = vtblk_tunable_int(sc, "writecache_mode", 1409 vtblk_writecache_mode); 1410 if (wc >= 0 && wc < VTBLK_CACHE_MAX) 1411 vtblk_set_write_cache(sc, wc); 1412 else 1413 wc = blkcfg->wce; 1414 } else 1415 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_FLUSH); 1416 1417 return (wc); 1418 } 1419 1420 static int 1421 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS) 1422 { 1423 struct vtblk_softc *sc; 1424 int wc, error; 1425 1426 sc = oidp->oid_arg1; 1427 wc = sc->vtblk_write_cache; 1428 1429 error = sysctl_handle_int(oidp, &wc, 0, req); 1430 if (error || req->newptr == NULL) 1431 return (error); 1432 if ((sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) == 0) 1433 return (EPERM); 1434 if (wc < 0 || wc >= VTBLK_CACHE_MAX) 1435 return (EINVAL); 1436 1437 VTBLK_LOCK(sc); 1438 sc->vtblk_write_cache = wc; 1439 vtblk_set_write_cache(sc, sc->vtblk_write_cache); 1440 VTBLK_UNLOCK(sc); 1441 1442 return (0); 1443 } 1444 1445 static void 1446 vtblk_setup_sysctl(struct vtblk_softc *sc) 1447 { 1448 device_t dev; 1449 struct sysctl_ctx_list *ctx; 1450 struct sysctl_oid *tree; 1451 struct sysctl_oid_list *child; 1452 1453 dev = sc->vtblk_dev; 1454 ctx = device_get_sysctl_ctx(dev); 1455 tree = device_get_sysctl_tree(dev); 1456 child = SYSCTL_CHILDREN(tree); 1457 1458 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode", 1459 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0, 1460 vtblk_write_cache_sysctl, "I", 1461 "Write cache mode (writethrough (0) or writeback (1))"); 1462 } 1463 1464 static int 1465 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def) 1466 { 1467 char path[64]; 1468 1469 snprintf(path, sizeof(path), 1470 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob); 1471 TUNABLE_INT_FETCH(path, &def); 1472 1473 return (def); 1474 } 1475