1 /*- 2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $FreeBSD: head/sys/dev/virtio/block/virtio_blk.c 252707 2013-07-04 17:57:26Z bryanv $ 27 */ 28 29 /* Driver for VirtIO block devices. */ 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/kernel.h> 34 #include <sys/bio.h> 35 #include <sys/malloc.h> 36 #include <sys/module.h> 37 #include <sys/sglist.h> 38 #include <sys/sysctl.h> 39 #include <sys/lock.h> 40 #include <sys/queue.h> 41 #include <sys/serialize.h> 42 #include <sys/buf2.h> 43 #include <sys/rman.h> 44 #include <sys/disk.h> 45 #include <sys/devicestat.h> 46 47 #include <dev/virtual/virtio/virtio/virtio.h> 48 #include <dev/virtual/virtio/virtio/virtqueue.h> 49 #include "virtio_blk.h" 50 #include "virtio_if.h" 51 52 struct vtblk_request { 53 struct virtio_blk_outhdr vbr_hdr __aligned(16); 54 struct bio *vbr_bp; 55 uint8_t vbr_ack; 56 uint8_t vbr_barrier; 57 58 TAILQ_ENTRY(vtblk_request) vbr_link; 59 }; 60 61 enum vtblk_cache_mode { 62 VTBLK_CACHE_WRITETHROUGH, 63 VTBLK_CACHE_WRITEBACK, 64 VTBLK_CACHE_MAX 65 }; 66 67 struct vtblk_softc { 68 device_t vtblk_dev; 69 struct lwkt_serialize vtblk_slz; 70 uint64_t vtblk_features; 71 uint32_t vtblk_flags; 72 #define VTBLK_FLAG_INDIRECT 0x0001 73 #define VTBLK_FLAG_READONLY 0x0002 74 #define VTBLK_FLAG_DETACH 0x0004 75 #define VTBLK_FLAG_SUSPEND 0x0008 76 #define VTBLK_FLAG_DUMPING 0x0010 77 #define VTBLK_FLAG_BARRIER 0x0020 78 #define VTBLK_FLAG_WC_CONFIG 0x0040 79 80 struct virtqueue *vtblk_vq; 81 struct sglist *vtblk_sglist; 82 struct disk vtblk_disk; 83 cdev_t cdev; 84 struct devstat stats; 85 86 struct bio_queue_head vtblk_bioq; 87 TAILQ_HEAD(, vtblk_request) 88 vtblk_req_free; 89 TAILQ_HEAD(, vtblk_request) 90 vtblk_req_ready; 91 struct vtblk_request *vtblk_req_ordered; 92 93 int vtblk_sector_size; 94 int vtblk_max_nsegs; 95 int vtblk_request_count; 96 enum vtblk_cache_mode vtblk_write_cache; 97 98 struct vtblk_request vtblk_dump_request; 99 }; 100 101 static struct virtio_feature_desc vtblk_feature_desc[] = { 102 { VIRTIO_BLK_F_BARRIER, "HostBarrier" }, 103 { VIRTIO_BLK_F_SIZE_MAX, "MaxSegSize" }, 104 { VIRTIO_BLK_F_SEG_MAX, "MaxNumSegs" }, 105 { VIRTIO_BLK_F_GEOMETRY, "DiskGeometry" }, 106 { VIRTIO_BLK_F_RO, "ReadOnly" }, 107 { VIRTIO_BLK_F_BLK_SIZE, "BlockSize" }, 108 { VIRTIO_BLK_F_SCSI, "SCSICmds" }, 109 { VIRTIO_BLK_F_WCE, "WriteCache" }, 110 { VIRTIO_BLK_F_TOPOLOGY, "Topology" }, 111 { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" }, 112 113 { 0, NULL } 114 }; 115 116 static int vtblk_modevent(module_t, int, void *); 117 118 static int vtblk_probe(device_t); 119 static int vtblk_attach(device_t); 120 static int vtblk_detach(device_t); 121 static int vtblk_suspend(device_t); 122 static int vtblk_resume(device_t); 123 static int vtblk_shutdown(device_t); 124 125 static void vtblk_negotiate_features(struct vtblk_softc *); 126 static int vtblk_maximum_segments(struct vtblk_softc *, 127 struct virtio_blk_config *); 128 static int vtblk_alloc_virtqueue(struct vtblk_softc *); 129 static void vtblk_set_write_cache(struct vtblk_softc *, int); 130 static int vtblk_write_cache_enabled(struct vtblk_softc *sc, 131 struct virtio_blk_config *); 132 static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS); 133 static void vtblk_alloc_disk(struct vtblk_softc *, 134 struct virtio_blk_config *); 135 /* 136 * Interface to the device switch. 137 */ 138 static d_open_t vtblk_open; 139 static d_strategy_t vtblk_strategy; 140 static d_dump_t vtblk_dump; 141 142 static struct dev_ops vbd_disk_ops = { 143 { "vbd", 200, D_DISK | D_MPSAFE }, 144 .d_open = vtblk_open, 145 .d_close = nullclose, 146 .d_read = physread, 147 .d_write = physwrite, 148 .d_strategy = vtblk_strategy, 149 .d_dump = vtblk_dump, 150 }; 151 152 static void vtblk_startio(struct vtblk_softc *); 153 static struct vtblk_request * vtblk_bio_request(struct vtblk_softc *); 154 static int vtblk_execute_request(struct vtblk_softc *, 155 struct vtblk_request *); 156 157 static int vtblk_vq_intr(void *); 158 static void vtblk_complete(void *); 159 160 static void vtblk_stop(struct vtblk_softc *); 161 162 static void vtblk_prepare_dump(struct vtblk_softc *); 163 static int vtblk_write_dump(struct vtblk_softc *, void *, off_t, size_t); 164 static int vtblk_flush_dump(struct vtblk_softc *); 165 static int vtblk_poll_request(struct vtblk_softc *, 166 struct vtblk_request *); 167 168 static void vtblk_drain_vq(struct vtblk_softc *, int); 169 static void vtblk_drain(struct vtblk_softc *); 170 171 static int vtblk_alloc_requests(struct vtblk_softc *); 172 static void vtblk_free_requests(struct vtblk_softc *); 173 static struct vtblk_request * vtblk_dequeue_request(struct vtblk_softc *); 174 static void vtblk_enqueue_request(struct vtblk_softc *, 175 struct vtblk_request *); 176 177 static struct vtblk_request * vtblk_dequeue_ready(struct vtblk_softc *); 178 static void vtblk_enqueue_ready(struct vtblk_softc *, 179 struct vtblk_request *); 180 181 static int vtblk_request_error(struct vtblk_request *); 182 static void vtblk_finish_bio(struct bio *, int); 183 184 static void vtblk_setup_sysctl(struct vtblk_softc *); 185 static int vtblk_tunable_int(struct vtblk_softc *, const char *, int); 186 187 /* Tunables. */ 188 static int vtblk_writecache_mode = -1; 189 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode); 190 191 /* Features desired/implemented by this driver. */ 192 #define VTBLK_FEATURES \ 193 (VIRTIO_BLK_F_BARRIER | \ 194 VIRTIO_BLK_F_SIZE_MAX | \ 195 VIRTIO_BLK_F_SEG_MAX | \ 196 VIRTIO_BLK_F_GEOMETRY | \ 197 VIRTIO_BLK_F_RO | \ 198 VIRTIO_BLK_F_BLK_SIZE | \ 199 VIRTIO_BLK_F_WCE | \ 200 VIRTIO_BLK_F_CONFIG_WCE) 201 202 /* 203 * Each block request uses at least two segments - one for the header 204 * and one for the status. 205 */ 206 #define VTBLK_MIN_SEGMENTS 2 207 208 static device_method_t vtblk_methods[] = { 209 /* Device methods. */ 210 DEVMETHOD(device_probe, vtblk_probe), 211 DEVMETHOD(device_attach, vtblk_attach), 212 DEVMETHOD(device_detach, vtblk_detach), 213 DEVMETHOD(device_suspend, vtblk_suspend), 214 DEVMETHOD(device_resume, vtblk_resume), 215 DEVMETHOD(device_shutdown, vtblk_shutdown), 216 217 DEVMETHOD_END 218 }; 219 220 static driver_t vtblk_driver = { 221 "vtblk", 222 vtblk_methods, 223 sizeof(struct vtblk_softc) 224 }; 225 static devclass_t vtblk_devclass; 226 227 DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass, 228 vtblk_modevent, NULL); 229 MODULE_VERSION(virtio_blk, 1); 230 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1); 231 232 static int 233 vtblk_modevent(module_t mod, int type, void *unused) 234 { 235 int error; 236 237 error = 0; 238 239 switch (type) { 240 case MOD_LOAD: 241 break; 242 case MOD_UNLOAD: 243 break; 244 case MOD_SHUTDOWN: 245 break; 246 default: 247 error = EOPNOTSUPP; 248 break; 249 } 250 251 return (error); 252 } 253 254 static int 255 vtblk_probe(device_t dev) 256 { 257 258 if (virtio_get_device_type(dev) != VIRTIO_ID_BLOCK) 259 return (ENXIO); 260 261 device_set_desc(dev, "VirtIO Block Adapter"); 262 263 return (BUS_PROBE_DEFAULT); 264 } 265 266 static int 267 vtblk_attach(device_t dev) 268 { 269 struct vtblk_softc *sc; 270 struct virtio_blk_config blkcfg; 271 int error; 272 273 sc = device_get_softc(dev); 274 sc->vtblk_dev = dev; 275 276 lwkt_serialize_init(&sc->vtblk_slz); 277 278 bioq_init(&sc->vtblk_bioq); 279 TAILQ_INIT(&sc->vtblk_req_free); 280 TAILQ_INIT(&sc->vtblk_req_ready); 281 282 virtio_set_feature_desc(dev, vtblk_feature_desc); 283 vtblk_negotiate_features(sc); 284 285 if (virtio_with_feature(dev, VIRTIO_BLK_F_RO)) 286 sc->vtblk_flags |= VTBLK_FLAG_READONLY; 287 if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER)) 288 sc->vtblk_flags |= VTBLK_FLAG_BARRIER; 289 if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE)) 290 sc->vtblk_flags |= VTBLK_FLAG_WC_CONFIG; 291 292 vtblk_setup_sysctl(sc); 293 294 /* Get local copy of config. */ 295 virtio_read_device_config(dev, 0, &blkcfg, 296 sizeof(struct virtio_blk_config)); 297 298 /* 299 * With the current sglist(9) implementation, it is not easy 300 * for us to support a maximum segment size as adjacent 301 * segments are coalesced. For now, just make sure it's larger 302 * than the maximum supported transfer size. 303 */ 304 if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) { 305 if (blkcfg.size_max < MAXPHYS) { 306 error = ENOTSUP; 307 device_printf(dev, "host requires unsupported " 308 "maximum segment size feature\n"); 309 goto fail; 310 } 311 } 312 313 sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg); 314 if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) { 315 error = EINVAL; 316 device_printf(dev, "fewer than minimum number of segments " 317 "allowed: %d\n", sc->vtblk_max_nsegs); 318 goto fail; 319 } 320 321 /* 322 * Allocate working sglist. The number of segments may be too 323 * large to safely store on the stack. 324 */ 325 sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT); 326 if (sc->vtblk_sglist == NULL) { 327 error = ENOMEM; 328 device_printf(dev, "cannot allocate sglist\n"); 329 goto fail; 330 } 331 332 error = vtblk_alloc_virtqueue(sc); 333 if (error) { 334 device_printf(dev, "cannot allocate virtqueue\n"); 335 goto fail; 336 } 337 338 error = vtblk_alloc_requests(sc); 339 if (error) { 340 device_printf(dev, "cannot preallocate requests\n"); 341 goto fail; 342 } 343 344 vtblk_alloc_disk(sc, &blkcfg); 345 346 error = virtio_setup_intr(dev, &sc->vtblk_slz); 347 if (error) { 348 device_printf(dev, "cannot setup virtqueue interrupt\n"); 349 goto fail; 350 } 351 352 virtqueue_enable_intr(sc->vtblk_vq); 353 354 fail: 355 if (error) 356 vtblk_detach(dev); 357 358 return (error); 359 } 360 361 static int 362 vtblk_detach(device_t dev) 363 { 364 struct vtblk_softc *sc; 365 366 sc = device_get_softc(dev); 367 368 lwkt_serialize_enter(&sc->vtblk_slz); 369 sc->vtblk_flags |= VTBLK_FLAG_DETACH; 370 if (device_is_attached(dev)) 371 vtblk_stop(sc); 372 lwkt_serialize_exit(&sc->vtblk_slz); 373 374 vtblk_drain(sc); 375 376 if (sc->cdev != NULL) { 377 disk_destroy(&sc->vtblk_disk); 378 sc->cdev = NULL; 379 } 380 381 if (sc->vtblk_sglist != NULL) { 382 sglist_free(sc->vtblk_sglist); 383 sc->vtblk_sglist = NULL; 384 } 385 386 return (0); 387 } 388 389 static int 390 vtblk_suspend(device_t dev) 391 { 392 struct vtblk_softc *sc; 393 394 sc = device_get_softc(dev); 395 396 lwkt_serialize_enter(&sc->vtblk_slz); 397 sc->vtblk_flags |= VTBLK_FLAG_SUSPEND; 398 /* XXX BMV: virtio_stop(), etc needed here? */ 399 lwkt_serialize_exit(&sc->vtblk_slz); 400 401 return (0); 402 } 403 404 static int 405 vtblk_resume(device_t dev) 406 { 407 struct vtblk_softc *sc; 408 409 sc = device_get_softc(dev); 410 411 lwkt_serialize_enter(&sc->vtblk_slz); 412 /* XXX BMV: virtio_reinit(), etc needed here? */ 413 sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; 414 #if 0 /* XXX Resume IO? */ 415 vtblk_startio(sc); 416 #endif 417 lwkt_serialize_exit(&sc->vtblk_slz); 418 419 return (0); 420 } 421 422 static int 423 vtblk_shutdown(device_t dev) 424 { 425 426 return (0); 427 } 428 429 static int 430 vtblk_open(struct dev_open_args *ap) 431 { 432 struct vtblk_softc *sc; 433 cdev_t dev = ap->a_head.a_dev; 434 sc = dev->si_drv1; 435 if (sc == NULL) 436 return (ENXIO); 437 438 return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0); 439 } 440 441 static int 442 vtblk_dump(struct dev_dump_args *ap) 443 { 444 struct vtblk_softc *sc; 445 cdev_t dev = ap->a_head.a_dev; 446 uint64_t buf_start, buf_len; 447 int error; 448 449 sc = dev->si_drv1; 450 if (sc == NULL) 451 return (ENXIO); 452 453 buf_start = ap->a_offset; 454 buf_len = ap->a_length; 455 456 // lwkt_serialize_enter(&sc->vtblk_slz); 457 458 if ((sc->vtblk_flags & VTBLK_FLAG_DUMPING) == 0) { 459 vtblk_prepare_dump(sc); 460 sc->vtblk_flags |= VTBLK_FLAG_DUMPING; 461 } 462 463 if (buf_len > 0) 464 error = vtblk_write_dump(sc, ap->a_virtual, buf_start, 465 buf_len); 466 else if (buf_len == 0) 467 error = vtblk_flush_dump(sc); 468 else { 469 error = EINVAL; 470 sc->vtblk_flags &= ~VTBLK_FLAG_DUMPING; 471 } 472 473 // lwkt_serialize_exit(&sc->vtblk_slz); 474 475 return (error); 476 } 477 478 static int 479 vtblk_strategy(struct dev_strategy_args *ap) 480 { 481 struct vtblk_softc *sc; 482 cdev_t dev = ap->a_head.a_dev; 483 sc = dev->si_drv1; 484 struct bio *bio = ap->a_bio; 485 struct buf *bp = bio->bio_buf; 486 487 if (sc == NULL) { 488 vtblk_finish_bio(bio, EINVAL); 489 return EINVAL; 490 } 491 492 /* 493 * Fail any write if RO. Unfortunately, there does not seem to 494 * be a better way to report our readonly'ness to GEOM above. 495 * 496 * XXX: Is that true in DFly? 497 */ 498 if (sc->vtblk_flags & VTBLK_FLAG_READONLY && 499 (bp->b_cmd == BUF_CMD_READ || bp->b_cmd == BUF_CMD_FLUSH)) { 500 vtblk_finish_bio(bio, EROFS); 501 return (EINVAL); 502 } 503 504 lwkt_serialize_enter(&sc->vtblk_slz); 505 if ((sc->vtblk_flags & VTBLK_FLAG_DETACH) == 0) { 506 devstat_start_transaction(&sc->stats); 507 bioqdisksort(&sc->vtblk_bioq, bio); 508 vtblk_startio(sc); 509 } else { 510 vtblk_finish_bio(bio, ENXIO); 511 } 512 lwkt_serialize_exit(&sc->vtblk_slz); 513 return 0; 514 } 515 516 static void 517 vtblk_negotiate_features(struct vtblk_softc *sc) 518 { 519 device_t dev; 520 uint64_t features; 521 522 dev = sc->vtblk_dev; 523 features = VTBLK_FEATURES; 524 525 sc->vtblk_features = virtio_negotiate_features(dev, features); 526 } 527 528 static int 529 vtblk_maximum_segments(struct vtblk_softc *sc, 530 struct virtio_blk_config *blkcfg) 531 { 532 device_t dev; 533 int nsegs; 534 535 dev = sc->vtblk_dev; 536 nsegs = VTBLK_MIN_SEGMENTS; 537 538 if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) { 539 nsegs += MIN(blkcfg->seg_max, MAXPHYS / PAGE_SIZE + 1); 540 } else 541 nsegs += 1; 542 543 return (nsegs); 544 } 545 546 static int 547 vtblk_alloc_virtqueue(struct vtblk_softc *sc) 548 { 549 device_t dev; 550 struct vq_alloc_info vq_info; 551 552 dev = sc->vtblk_dev; 553 554 VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs, 555 vtblk_vq_intr, sc, &sc->vtblk_vq, 556 "%s request", device_get_nameunit(dev)); 557 558 return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info)); 559 } 560 561 static void 562 vtblk_set_write_cache(struct vtblk_softc *sc, int wc) 563 { 564 565 /* Set either writeback (1) or writethrough (0) mode. */ 566 virtio_write_dev_config_1(sc->vtblk_dev, 567 offsetof(struct virtio_blk_config, writeback), wc); 568 } 569 570 static int 571 vtblk_write_cache_enabled(struct vtblk_softc *sc, 572 struct virtio_blk_config *blkcfg) 573 { 574 int wc; 575 576 if (sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) { 577 wc = vtblk_tunable_int(sc, "writecache_mode", 578 vtblk_writecache_mode); 579 if (wc >= 0 && wc < VTBLK_CACHE_MAX) 580 vtblk_set_write_cache(sc, wc); 581 else 582 wc = blkcfg->writeback; 583 } else 584 wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_WCE); 585 586 return (wc); 587 } 588 589 static int 590 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS) 591 { 592 struct vtblk_softc *sc; 593 int wc, error; 594 595 sc = oidp->oid_arg1; 596 wc = sc->vtblk_write_cache; 597 598 error = sysctl_handle_int(oidp, &wc, 0, req); 599 if (error || req->newptr == NULL) 600 return (error); 601 if ((sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) == 0) 602 return (EPERM); 603 if (wc < 0 || wc >= VTBLK_CACHE_MAX) 604 return (EINVAL); 605 606 lwkt_serialize_enter(&sc->vtblk_slz); 607 sc->vtblk_write_cache = wc; 608 vtblk_set_write_cache(sc, sc->vtblk_write_cache); 609 lwkt_serialize_exit(&sc->vtblk_slz); 610 611 return (0); 612 } 613 614 static void 615 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) 616 { 617 618 struct disk_info info; 619 620 /* construct the disk_info */ 621 bzero(&info, sizeof(info)); 622 623 if (virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_BLK_SIZE)) 624 sc->vtblk_sector_size = blkcfg->blk_size; 625 else 626 sc->vtblk_sector_size = DEV_BSIZE; 627 628 info.d_media_blksize = sc->vtblk_sector_size; 629 info.d_media_blocks = blkcfg->capacity; 630 631 info.d_ncylinders = blkcfg->geometry.cylinders; 632 info.d_nheads = blkcfg->geometry.heads; 633 info.d_secpertrack = blkcfg->geometry.sectors; 634 635 info.d_secpercyl = info.d_secpertrack * info.d_nheads; 636 637 if (vtblk_write_cache_enabled(sc, blkcfg) != 0) 638 sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK; 639 else 640 sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH; 641 642 devstat_add_entry(&sc->stats, "vbd", device_get_unit(sc->vtblk_dev), 643 DEV_BSIZE, DEVSTAT_ALL_SUPPORTED, 644 DEVSTAT_TYPE_DIRECT | DEVSTAT_TYPE_IF_OTHER, 645 DEVSTAT_PRIORITY_DISK); 646 647 /* attach a generic disk device to ourselves */ 648 sc->cdev = disk_create(device_get_unit(sc->vtblk_dev), &sc->vtblk_disk, 649 &vbd_disk_ops); 650 651 sc->cdev->si_drv1 = sc; 652 sc->cdev->si_iosize_max = MAXPHYS; 653 disk_setdiskinfo(&sc->vtblk_disk, &info); 654 } 655 656 static void 657 vtblk_startio(struct vtblk_softc *sc) 658 { 659 struct virtqueue *vq; 660 struct vtblk_request *req; 661 int enq; 662 663 vq = sc->vtblk_vq; 664 enq = 0; 665 666 ASSERT_SERIALIZED(&sc->vtblk_slz); 667 668 if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) 669 return; 670 671 while (!virtqueue_full(vq)) { 672 if ((req = vtblk_dequeue_ready(sc)) == NULL) 673 req = vtblk_bio_request(sc); 674 if (req == NULL) 675 break; 676 677 if (vtblk_execute_request(sc, req) != 0) { 678 vtblk_enqueue_ready(sc, req); 679 break; 680 } 681 682 enq++; 683 } 684 685 if (enq > 0) 686 virtqueue_notify(vq, &sc->vtblk_slz); 687 } 688 689 static struct vtblk_request * 690 vtblk_bio_request(struct vtblk_softc *sc) 691 { 692 struct bio_queue_head *bioq; 693 struct vtblk_request *req; 694 struct bio *bio; 695 struct buf *bp; 696 697 bioq = &sc->vtblk_bioq; 698 699 if (bioq_first(bioq) == NULL) 700 return (NULL); 701 702 req = vtblk_dequeue_request(sc); 703 if (req == NULL) 704 return (NULL); 705 706 bio = bioq_takefirst(bioq); 707 req->vbr_bp = bio; 708 req->vbr_ack = -1; 709 req->vbr_barrier = 0; 710 req->vbr_hdr.ioprio = 1; 711 bp = bio->bio_buf; 712 713 switch (bp->b_cmd) { 714 case BUF_CMD_FLUSH: 715 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH; 716 break; 717 case BUF_CMD_READ: 718 req->vbr_hdr.type = VIRTIO_BLK_T_IN; 719 req->vbr_hdr.sector = bio->bio_offset / DEV_BSIZE; 720 break; 721 case BUF_CMD_WRITE: 722 req->vbr_hdr.type = VIRTIO_BLK_T_OUT; 723 req->vbr_hdr.sector = bio->bio_offset / DEV_BSIZE; 724 break; 725 default: 726 KASSERT(0, ("bio with unhandled cmd: %d", bp->b_cmd)); 727 req->vbr_hdr.type = -1; 728 break; 729 } 730 731 if (bp->b_flags & B_ORDERED) { 732 if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) 733 req->vbr_barrier = 1; 734 else 735 req->vbr_hdr.type |= VIRTIO_BLK_T_BARRIER; 736 } 737 738 return (req); 739 } 740 741 static int 742 vtblk_execute_request(struct vtblk_softc *sc, struct vtblk_request *req) 743 { 744 struct sglist *sg; 745 struct bio *bio; 746 struct buf *bp; 747 int ordered, writable, error; 748 749 sg = sc->vtblk_sglist; 750 bio = req->vbr_bp; 751 bp = bio->bio_buf; 752 ordered = 0; 753 writable = 0; 754 755 if (sc->vtblk_req_ordered != NULL) 756 return (EBUSY); 757 758 if (req->vbr_barrier) { 759 /* 760 * This request will be executed once all 761 * the in-flight requests are completed. 762 */ 763 if (!virtqueue_empty(sc->vtblk_vq)) 764 return (EBUSY); 765 ordered = 1; 766 } 767 768 /* 769 * sglist is live throughout this subroutine. 770 */ 771 sglist_reset(sg); 772 773 error = sglist_append(sg, &req->vbr_hdr, 774 sizeof(struct virtio_blk_outhdr)); 775 KASSERT(error == 0, ("error adding header to sglist")); 776 KASSERT(sg->sg_nseg == 1, 777 ("header spanned multiple segments: %d", sg->sg_nseg)); 778 779 if (bp->b_cmd == BUF_CMD_READ || bp->b_cmd == BUF_CMD_WRITE) { 780 error = sglist_append(sg, bp->b_data, bp->b_bcount); 781 KASSERT(error == 0, ("error adding buffer to sglist")); 782 783 /* BUF_CMD_READ means the host writes into our buffer. */ 784 if (bp->b_cmd == BUF_CMD_READ) 785 writable += sg->sg_nseg - 1; 786 } 787 788 error = sglist_append(sg, &req->vbr_ack, sizeof(uint8_t)); 789 KASSERT(error == 0, ("error adding ack to sglist")); 790 writable++; 791 792 KASSERT(sg->sg_nseg >= VTBLK_MIN_SEGMENTS, 793 ("fewer than min segments: %d", sg->sg_nseg)); 794 795 error = virtqueue_enqueue(sc->vtblk_vq, req, sg, 796 sg->sg_nseg - writable, writable); 797 if (error == 0 && ordered) 798 sc->vtblk_req_ordered = req; 799 800 sglist_reset(sg); 801 802 return (error); 803 } 804 805 static int 806 vtblk_vq_intr(void *xsc) 807 { 808 vtblk_complete(xsc); 809 810 return (1); 811 } 812 813 static void 814 vtblk_complete(void *arg) 815 { 816 struct vtblk_softc *sc; 817 struct vtblk_request *req; 818 struct virtqueue *vq; 819 struct bio *bio; 820 struct buf *bp; 821 822 sc = arg; 823 vq = sc->vtblk_vq; 824 825 lwkt_serialize_handler_disable(&sc->vtblk_slz); 826 virtqueue_disable_intr(sc->vtblk_vq); 827 ASSERT_SERIALIZED(&sc->vtblk_slz); 828 829 retry: 830 if (sc->vtblk_flags & VTBLK_FLAG_DETACH) 831 return; 832 833 while ((req = virtqueue_dequeue(vq, NULL)) != NULL) { 834 bio = req->vbr_bp; 835 bp = bio->bio_buf; 836 837 if (sc->vtblk_req_ordered != NULL) { 838 /* This should be the only outstanding request. */ 839 KKASSERT(sc->vtblk_req_ordered == req); 840 sc->vtblk_req_ordered = NULL; 841 } 842 843 if (req->vbr_ack == VIRTIO_BLK_S_OK) 844 bp->b_resid = 0; 845 else { 846 bp->b_flags |= B_ERROR; 847 if (req->vbr_ack == VIRTIO_BLK_S_UNSUPP) { 848 bp->b_error = ENOTSUP; 849 } else { 850 bp->b_error = EIO; 851 } 852 } 853 854 devstat_end_transaction_buf(&sc->stats, bio->bio_buf); 855 856 lwkt_serialize_exit(&sc->vtblk_slz); 857 /* 858 * Unlocking the controller around biodone() does not allow 859 * processing further device interrupts; when we queued 860 * vtblk_complete, we disabled interrupts. It will allow 861 * concurrent vtblk_strategy/_startio command dispatches. 862 */ 863 biodone(bio); 864 lwkt_serialize_enter(&sc->vtblk_slz); 865 866 vtblk_enqueue_request(sc, req); 867 } 868 869 vtblk_startio(sc); 870 871 if (virtqueue_enable_intr(vq) != 0) { 872 /* 873 * If new virtqueue entries appeared immediately after 874 * enabling interrupts, process them now. Release and 875 * retake softcontroller lock to try to avoid blocking 876 * I/O dispatch for too long. 877 */ 878 virtqueue_disable_intr(vq); 879 goto retry; 880 } 881 lwkt_serialize_handler_enable(&sc->vtblk_slz); 882 } 883 884 static void 885 vtblk_stop(struct vtblk_softc *sc) 886 { 887 888 virtqueue_disable_intr(sc->vtblk_vq); 889 virtio_stop(sc->vtblk_dev); 890 } 891 892 static void 893 vtblk_prepare_dump(struct vtblk_softc *sc) 894 { 895 device_t dev; 896 struct virtqueue *vq; 897 898 dev = sc->vtblk_dev; 899 vq = sc->vtblk_vq; 900 901 vtblk_stop(sc); 902 903 /* 904 * Drain all requests caught in-flight in the virtqueue, 905 * skipping biodone(). When dumping, only one request is 906 * outstanding at a time, and we just poll the virtqueue 907 * for the response. 908 */ 909 vtblk_drain_vq(sc, 1); 910 911 if (virtio_reinit(dev, sc->vtblk_features) != 0) { 912 panic("%s: cannot reinit VirtIO block device during dump", 913 device_get_nameunit(dev)); 914 } 915 916 virtqueue_disable_intr(vq); 917 virtio_reinit_complete(dev); 918 } 919 920 static int 921 vtblk_write_dump(struct vtblk_softc *sc, void *virtual, off_t offset, 922 size_t length) 923 { 924 struct bio bio; 925 struct buf bp; 926 struct vtblk_request *req; 927 928 req = &sc->vtblk_dump_request; 929 req->vbr_ack = -1; 930 req->vbr_hdr.type = VIRTIO_BLK_T_OUT; 931 req->vbr_hdr.ioprio = 1; 932 req->vbr_hdr.sector = offset / 512; 933 934 req->vbr_bp = &bio; 935 bzero(&bio, sizeof(struct bio)); 936 bzero(&buf, sizeof(struct buf)); 937 938 bio.bio_buf = &bp; 939 bp.b_cmd = BUF_CMD_WRITE; 940 bp.b_data = virtual; 941 bp.b_bcount = length; 942 943 return (vtblk_poll_request(sc, req)); 944 } 945 946 static int 947 vtblk_flush_dump(struct vtblk_softc *sc) 948 { 949 struct bio bio; 950 struct buf bp; 951 struct vtblk_request *req; 952 953 req = &sc->vtblk_dump_request; 954 req->vbr_ack = -1; 955 req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH; 956 req->vbr_hdr.ioprio = 1; 957 req->vbr_hdr.sector = 0; 958 959 req->vbr_bp = &bio; 960 bzero(&bio, sizeof(struct bio)); 961 bzero(&bp, sizeof(struct buf)); 962 963 bio.bio_buf = &bp; 964 bp.b_cmd = BUF_CMD_FLUSH; 965 966 return (vtblk_poll_request(sc, req)); 967 } 968 969 static int 970 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req) 971 { 972 struct virtqueue *vq; 973 int error; 974 975 vq = sc->vtblk_vq; 976 977 if (!virtqueue_empty(vq)) 978 return (EBUSY); 979 980 error = vtblk_execute_request(sc, req); 981 if (error) 982 return (error); 983 984 virtqueue_notify(vq, NULL); 985 virtqueue_poll(vq, NULL); 986 987 error = vtblk_request_error(req); 988 if (error && bootverbose) { 989 device_printf(sc->vtblk_dev, 990 "%s: IO error: %d\n", __func__, error); 991 } 992 993 return (error); 994 } 995 996 static void 997 vtblk_drain_vq(struct vtblk_softc *sc, int skip_done) 998 { 999 struct virtqueue *vq; 1000 struct vtblk_request *req; 1001 int last; 1002 1003 vq = sc->vtblk_vq; 1004 last = 0; 1005 1006 while ((req = virtqueue_drain(vq, &last)) != NULL) { 1007 if (!skip_done) 1008 vtblk_finish_bio(req->vbr_bp, ENXIO); 1009 1010 vtblk_enqueue_request(sc, req); 1011 } 1012 1013 sc->vtblk_req_ordered = NULL; 1014 KASSERT(virtqueue_empty(vq), ("virtqueue not empty")); 1015 } 1016 1017 static void 1018 vtblk_drain(struct vtblk_softc *sc) 1019 { 1020 struct bio_queue_head *bioq; 1021 struct vtblk_request *req; 1022 struct bio *bp; 1023 1024 bioq = &sc->vtblk_bioq; 1025 1026 if (sc->vtblk_vq != NULL) 1027 vtblk_drain_vq(sc, 0); 1028 1029 while ((req = vtblk_dequeue_ready(sc)) != NULL) { 1030 vtblk_finish_bio(req->vbr_bp, ENXIO); 1031 vtblk_enqueue_request(sc, req); 1032 } 1033 1034 while (bioq_first(bioq) != NULL) { 1035 bp = bioq_takefirst(bioq); 1036 vtblk_finish_bio(bp, ENXIO); 1037 } 1038 1039 vtblk_free_requests(sc); 1040 } 1041 1042 static int 1043 vtblk_alloc_requests(struct vtblk_softc *sc) 1044 { 1045 struct vtblk_request *req; 1046 int i, nreqs; 1047 1048 nreqs = virtqueue_size(sc->vtblk_vq); 1049 1050 /* 1051 * Preallocate sufficient requests to keep the virtqueue full. Each 1052 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce 1053 * the number allocated when indirect descriptors are not available. 1054 */ 1055 nreqs /= VTBLK_MIN_SEGMENTS; 1056 1057 for (i = 0; i < nreqs; i++) { 1058 req = contigmalloc(sizeof(struct vtblk_request), M_DEVBUF, 1059 M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0); 1060 if (req == NULL) 1061 return (ENOMEM); 1062 1063 sc->vtblk_request_count++; 1064 vtblk_enqueue_request(sc, req); 1065 } 1066 1067 return (0); 1068 } 1069 1070 static void 1071 vtblk_free_requests(struct vtblk_softc *sc) 1072 { 1073 struct vtblk_request *req; 1074 1075 while ((req = vtblk_dequeue_request(sc)) != NULL) { 1076 sc->vtblk_request_count--; 1077 contigfree(req, sizeof(struct vtblk_request), M_DEVBUF); 1078 } 1079 1080 KASSERT(sc->vtblk_request_count == 0, ("leaked requests")); 1081 } 1082 1083 static struct vtblk_request * 1084 vtblk_dequeue_request(struct vtblk_softc *sc) 1085 { 1086 struct vtblk_request *req; 1087 1088 req = TAILQ_FIRST(&sc->vtblk_req_free); 1089 if (req != NULL) 1090 TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link); 1091 1092 return (req); 1093 } 1094 1095 static void 1096 vtblk_enqueue_request(struct vtblk_softc *sc, struct vtblk_request *req) 1097 { 1098 1099 bzero(req, sizeof(struct vtblk_request)); 1100 TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link); 1101 } 1102 1103 static struct vtblk_request * 1104 vtblk_dequeue_ready(struct vtblk_softc *sc) 1105 { 1106 struct vtblk_request *req; 1107 1108 req = TAILQ_FIRST(&sc->vtblk_req_ready); 1109 if (req != NULL) 1110 TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link); 1111 1112 return (req); 1113 } 1114 1115 static void 1116 vtblk_enqueue_ready(struct vtblk_softc *sc, struct vtblk_request *req) 1117 { 1118 1119 TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link); 1120 } 1121 1122 static int 1123 vtblk_request_error(struct vtblk_request *req) 1124 { 1125 int error; 1126 1127 switch (req->vbr_ack) { 1128 case VIRTIO_BLK_S_OK: 1129 error = 0; 1130 break; 1131 case VIRTIO_BLK_S_UNSUPP: 1132 error = ENOTSUP; 1133 break; 1134 default: 1135 error = EIO; 1136 break; 1137 } 1138 1139 return (error); 1140 } 1141 1142 static void 1143 vtblk_finish_bio(struct bio *bp, int error) 1144 { 1145 1146 biodone(bp); 1147 } 1148 1149 static void 1150 vtblk_setup_sysctl(struct vtblk_softc *sc) 1151 { 1152 device_t dev; 1153 struct sysctl_ctx_list *ctx; 1154 struct sysctl_oid *tree; 1155 struct sysctl_oid_list *child; 1156 1157 dev = sc->vtblk_dev; 1158 ctx = device_get_sysctl_ctx(dev); 1159 tree = device_get_sysctl_tree(dev); 1160 child = SYSCTL_CHILDREN(tree); 1161 1162 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode", 1163 CTLTYPE_INT | CTLFLAG_RW, sc, 0, vtblk_write_cache_sysctl, 1164 "I", "Write cache mode (writethrough (0) or writeback (1))"); 1165 } 1166 1167 static int 1168 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def) 1169 { 1170 char path[64]; 1171 1172 ksnprintf(path, sizeof(path), 1173 "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob); 1174 TUNABLE_INT_FETCH(path, &def); 1175 1176 return (def); 1177 } 1178